Syntactic Structures and Morphological Information [Reprint 2011 ed.] 9783110904758, 9783110178241

Table of contents :
Introduction
Metagrammar of systematic relations: a study with special reference to Slavic morphosyntax
On-line morphology : The morphosyntax of Hungarian verbal inflection
Verbal morphology and agreement in Urdu
Particles and sentence structure: a historical perspective
Subject Case in Turkish nominalized clauses
On the licensing of null subjects in Old French
Periphrastic paradigms in Bulgarian
Transparent, restricted and opaque affix orders
Direction marking as agreement
On the semantics of cases
Index

Citation preview

Syntactic Structures and Morphological Information

W G DE

Interface Explorations 7

Editors

Artemis Alexiadou T. Alan Hall

Mouton de Gruyter Berlin · New York

Syntactic Structures and Morphological Information

edited by

Uwe Junghanns Luka Szucsich

Mouton de Gruyter Berlin · New York

2003

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

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

Library of Congress Cataloging-in-Publication Data Syntactic structures and morphological information / edited by Uwe Junghanns, Luka Szucsich. p. cm. - (Interface explorations ; 7) Includes bibliographical references and index. ISBN 3-11-017824-9 (alk. paper) 1. Grammar, Comparative and general — Morphosyntax. I. Junghanns, Uwe. II. Szucsich, Luka, 1968— III. Series. P290.S93 2003 415.dc22 2003019338

ISBN 3-11-017824-9 Bibliographic information published by Die Deutsche Bibliothek Die Deutsche Bibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data is available in the Internet at . © Copyright 2003 by Walter de Gruyter GmbH & Co. KG, 10785 Berlin All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Cover design: Christopher Schneider, Berlin. Printed in Germany.

Contents

Introduction Luka Szucsich and Uwe Junghanns Metagrammar of systematic relations: a study with special reference to Slavic morphosyntax Tania Avgustinova On-line morphology: The morphosyntax of Hungarian verbal inflection Huba Bartos Verbal morphology and agreement in Urdu Miriam Butt and Louisa Sadler

vii

1

25

57

Particles and sentence structure: a historical perspective Gisella Ferraresi and Maria Goldbach

101

Subject Case in Turkish nominalized clauses Jaklin Kornfilt

129

On the licensing of null subjects in Old French Esther Rinke

217

Periphrastic paradigms in Bulgarian Andrew Spencer

249

Transparent, restricted and opaque affix orders Barbara Stiebels

283

Direction marking as agreement Jochen Trommer

317

vi

Contents

On the semantics of cases Ilse Zimmermann

341

Index

381

Introduction Luka Szucsich and Uwe Junghanns

The present volume contains a selection of papers originating in the workshop entitled "Clause Structure and Models of Grammar from the Perspective of Languages with Rich Morphology" at the 23 rd annual meeting of the German Linguistic Society / 23. Jahrestagung der Deutschen Gesellschaft fur Sprachwissenschaft (DGfS-23) which took place in February 2001 at the University of Leipzig. The relation between morphology and syntax has been one of the most debated topics in linguistics. In most cases "words" do not appear in one single form (phonological matrix) irrespective of the syntactic environment in which they are embedded, but rather exhibit additional morphological markers depending on the "slot" they occupy in a complex syntactic structure. It is commonly assumed that these morphological markers, as a rule, do not belong to the substantial lexical meaning of the respective word form. Apart from that, in many languages there exists a set of morphological words (some of them are not necessarily phonological words) which merely represent grammatical/functional meanings of substantial lexical entries (auxiliaries and other free functional/synsemantic words). Morphological markers (bound or free) serve different purposes. They "identify" grammatical categories like tense, aspect, mood, sentence type, number, person, gender, case, definiteness, etc. which are connected to certain interpretations of complex linguistic expressions and they are correlated with the combinatorial potential of the respective word form, i.e. they "embed" a lexical entry into its syntactic environment. Here are just two examples: (i) Certain verbal morphology (e.g., active/passive or tense/finiteness morphology) influences the realization of the argument structure of the verbal lexical entry; (ii) a case marker indicates the "place" of a nominal expression within a more complex linguistic expression (in a minimalist approach it indicates the target the nominal constituent

viii

Luka Szucsich & Uwe Junghanns

merges with; in a functionalist approach it determines the syntactic function of a nominal expression). In recent theories, morphology has been treated rather diversely, either directly determining syntactic derivations or having no immediate relation to syntactic structures. In some models, morphological information appears as a specification or sub-matrix of nodes (e.g., Lexical-Functional Grammar, cf. Bresnan (ed.) 1982; Head-Driven Phrase-Structure Grammar, cf. Pollard and Sag 1994; Functional Generative Description, cf. Sgall, Hajicovä, and Panevovä 1986). In recent generative Principle and Parameter models, alongside the morphological characterization of word forms and the syntactic structures induced by them a varying number of special syntactic categories—so-called functional categories—are assumed that represent morphological and/or semantic information or that provide potential syntactic positions for argument expressions and adverbial phrases (cf. Pollock 1989; Chomsky 1995, 2000, 2001; Rizzi 1997; Cinque 1999). In view of these developments it is necessary to discuss the place of morphology in models of grammar and determine in a principled way which possible correlates morphological information has in the syntactic representation of the clause. The papers in this volume do not focus on languages with relatively impoverished morphology like, e.g., modern English, but rather on languages with rich morphology (inflecting, agglutinating, polysynthetic languages). Many of the contributors consider in their papers universal as well as language-specific aspects of the relation between morphology and clausal syntax. This volume was designed to assemble papers by researchers from different linguistic backgrounds and, thus, to mirror the on-going discussion concerning the relation between the morphological and the syntactic component of the grammar, taking rather different perspectives. The central issue of the workshop and the present volume was/is to come closer to a principled linguistic treatment of clause structure in morphologically rich languages, i.e. to discuss whether there is a generalized relation between morphological and

Introduction

ix

syntactic operations and, if this is the case, how this relation could be modeled in the respective framework. In the following we will give short summaries of the papers contained in this volume. In her paper "Metagrammar of systematic relations: a study with special reference to Slavic morphosyntax" Tania Avgustinova develops a standardized (universal) taxonomy for systematic relations in grammar (the so-called metagrammar) with a hierarchy of relational types and a system of admissible cross-classifications of different relational types. Those cross-classifications amount to a broad array of grammatical relations including marginal ones which are instantiated in various ways in different natural languages. Thus, one aim of the proposed metagrammar is to provide this inventory of possible syntactic combinations and (morpho-)syntactic relations which should be interchangeable between different syntactic and morphological theories. The practical side of this work is to formulate a system of grammatical relatedness which could be implemented in language tools and which should determine the design of shared grammatical resources for Slavic languages. Although one of the tasks is to develop a metagrammar serving different syntactic and morphological theories, the design of the type hierarchy of grammatical relations is based on type hierarchies known from the Head Driven Phrase Structure Grammar (HPSG) with higher- and lower-level relational types and multi-dimensional inheritance of relational properties from higher-level to lower-level types. All constraints associated with a particular relational type are consequently also inherited to lower-level types. In the introductory section Avgustinova lays out the fundamental assumption concerning the metagrammar of systematic relations. The second and central section is devoted to a description of the systematic relations. Avgustinova above all discusses a subpart of systematic relations which she calls "observable syntagmatics". These relations are connected to the overt linguistic form, in contrast to covert linguistic function ("structural syntagmatics"). Moreover, within the observable syntagmatics she mainly concentrates on combinatorial relations which largely correspond to morphological or

χ

Luka Szucsich & Uwe Junghanns

morphosyntactic relations. Avgustinova claims that the alignment dimension of observable syntagmatics determining the linear distribution of syntactic items is less relevant for Slavic languages (which means that the ordering of constituents in Slavic languages is not rigidly fixed but determined by information-structural requirements). To exemplify the abovementioned relations, Avgustinova has chosen Russian and Bulgarian examples which she presents in section 3, employing so-called "relational charts". These charts (in the shape of diagrams) consist of all lexical items of the actual sentence and additional cells for labeling the systematic relations between the individual lexical items or complexes of items (in section 1 the author explains the format of the charts—a certain relational chart comprises all metagrammatical relations of the respective sentence). Languages with a considerably rich morphology like Russian and Bulgarian call for a detailed elaboration of combinatorial relations. This type of metagrammar, of course, does not explain the gaps within the inventory of syntactic combinations for a particular language or from a cross-linguistic perspective. The author herself states that constraints that block certain cross-classifications of relational types are to be developed. These constraints, of course, cannot be theory-independent. Therefore, the aim of the paper is rather to provide tools for systematically classifying grammatical relations not least against the background of an application to automatic language processing. In the context of a derivational theory of morphology Huba Bartos investigates phenomena of Hungarian verbal morphology which at first sight seem to violate the Mirror Principle developed by Baker (1985). The aim of the paper is to save this principle without resorting to dubious mechanisms like morphological reordering or theoretically undesirable concepts like covert movement in morphology. One of the main theoretical claims of the paper is that morphology "shadows" syntax, but certain principles in morphology may cause orderings of morphemes that look like deviations from

Introduction

xi

strict mirroring in light of the scope properties of the respective morphemes. In the second section, following the introduction, Bartos presents the relevant Hungarian data. In Hungarian, inflectional morphology obeys a strict ordering V-Mod(ality)-T(ense)-M(ood). However, certain affix orderings may be associated with different scope orders (e.g. if only Mod- and T-morphemes are present, both readings, Mod > Τ and Τ > Mod, are possible). Assuming that semantic scope is determined syntactically via c-command, these data pose a problem for a strict interpretation of the Mirror principle. In the third section Bartos lays out the fundamentals of his account. Following recent syntactic theories he assumes that syntactic structures are built up derivationally and cyclically (obeying strict locality conditions). The morphological component has access to the syntactic structure at any point of the derivation, i.e. morphology strictly parallels syntax. If a derivational step has a consequence for morphology (word structure), the respective morphological operations have to be carried out obligatorily, e.g. what is known as head movement is always relevant for morphology. The relevant binary operation is called Morphosyntactic Merger. A central assumption is that as soon as a morphological operation is carried out, the newly built word-domain is opaque for further morphological operations, i.e. further operations always involve just the edges of word-domains. There is no such thing as restructuring within word-domains. In the following section, Bartos presents his account of the Hungarian data. He assumes three functional categories potentially filled with "genuine" morphemes (viz. Mod, T, and M). These categories, however, also provide templatic slots, if the respective functional category is "contentless" (i.e., a merely categorical frame which Bartos calls proxy). These proxies are available for features of lower functional categories to get scope over intermediate projections. The ordering of the morphemes, however, does not change, because the "lower" morpheme has already been morphosyntactically merged with the verbal root forming an opaque word-domain. This phenomenon produces the alleged violation of

xii

Luka Szucsich & Uwe Junghanns

the Mirror Principle. Bartos' account predicts that there should be no scope inversion where all functional categories are "contentfull", i.e. where Mod, T, and Μ contain interpretable formal features which in most cases are morphologically spelled out. This prediction is indeed borne out by the Hungarian data. In the fifth section, as a "by-product", Bartos provides an account for so-called verbal complexes in Hungarian by employing the concept of morphosyntactic merger. He derives both the ordering patterns within the verbal complexes (the so-called "roll-up" and the non-"roll-up" order) and the fact that these complexes display certain properties of words, e.g., they serve as inputs to derivational processes like other word-level units. The paper by Miriam Butt and Louisa Sadler is concerned with the general issue of the status of morphology within the theory of grammar, and in particular its relation to syntax. The points at issue are the interaction between morphology and syntax, on the one hand, and, on the other, the mechanisms and data structures that should be assumed as appropriate for the description of morphological systems. More specifically, the paper explores aspects of the morphology of case and agreement in Urdu within the framework of LexicalFunctional Grammar (LFG). The lexicalist hypothesis is taken to hold. Accordingly, syntactic and morphological processes belong to differing modules of grammar. However, quite a complex interaction between syntax and morphology is permitted due to the specific form in which the lexicalist hypothesis is embodied in LFG. The authors assume a finite-state morphological analyzer as used in computational work in LFG. While it preserves the separation of (external) syntax and morphology, it does not exclude contact between morphology and syntax entirely. Interaction comes about at an interface allowing syntactic functional information to filter through by means of tags or features which abstract away from the surface realization of the morphemes. Butt and Sadler claim that their model of the morphology-syntax interface is superior to a morpheme-based word syntax approach that suffers from a number of insufficiencies as they show.

Introduction

xiii

The paper consists of seven sections. The introductory section is followed by a section concerned with Lexical-Functional Grammar in which Butt and Sadler provide a brief sketch of the basic design principles of LFG. They introduce and exemplify an aspect of the formalism (known as constructive morphology, Nordlinger 1998) which permits a natural and straightforward approach to the ability of morphological elements (such as case markers) to define and project the relational structures which contain them. In section 3, Butt and Sadler briefly introduce case in Urdu and its relation to verbal agreement patterns. They sketch out a treatment of Urdu case marking in LFG. Section 4 deals with agreement. Specifically, Butt and Sadler present the facts of verbal agreement in Urdu. They formulate a relatively simple generalization concerning verbal agreement and show how constraints associated with verb forms will capture this generalization. Sections 5 and 6 are devoted to the details of the morphological analysis. In section 5, Butt and Sadler explore a word-syntax, or morpheme-based, implementation presenting several unwanted side-effects and drawbacks of this approach. In section 6, they examine an encoding of the same set of agreement data using a finite-state morphological analyzer interfaced to the syntax, showing how the difficulties encountered in the word syntax approach are resolved. Section 7 concludes the paper. Gisella Ferraresi and Maria Goldbach examine the syntactic and phonological status of the Old French sentence particle si 'thus' and its loss in the 17th century. The basic hypothesis of the paper is that a syntactic change emerges, if conditions of either the interface to the C-I system or the interface to the A-P system undergo changes which affect semantic and phonological properties of lexical items. The authors take functional categories as combinations of subsets of the set of formal features selected by a specific language. In different languages, functional categories may differ with respect to which subset of formal features is assembled in a particular functional category. Syntactic phenomena in a given language correspond to particular representations of functional categories at the semantic and phonological interface. Ferraresi and Goldbach show how the disappearance of the sentence particle si is related to its phonological

xiv

Luka Szucsich & Uwe Junghanns

status and to changes concerning the system of the syllabic structure of the phonological word and the clitic group in Late Latin and Old French. In section 1 the authors discuss the morphophonological shape of Old French si. By extensively examining iambic, decasyllabic Old French metrical poetry which strictly obeys the rhythmical structure of the involved lexemes, they reach the conclusion that si is a phonologically weak element. It may be part of clitic clusters, but unlike object and adverbial clitics it is not an obligatory clitic. In the second section Ferraresi and Goldbach provide evidence against a prevailing analysis according to which si occupies a Specposition within the C-domain like sentential adverbs. In contrast to the latter, si can occupy a position between a subject DP and the main verb. This fact cannot be explained for a V2-language like Old French, if si were an XP located in a Spec-position higher than the main verb (if preverbal sentential adverbs are present, subject DPs obligatorily occur postverbally). In addition, the distribution of si largely patterns with that of Welsh particles (e.g. only clitics may intervene between a particle and the main verb). These facts lead the authors, who adopt Rizzi's (1997) split CP analysis, to the conclusion that si is spelled out in a functional category, viz. Fin0. The prosodic facts and the fact that si may be split from the main verb only by object clitics are captured by the claim that the verbal root together with the clitics moves through the T°-node to the Fin°node where the complex merges with the particle si forming a word level unit which also constitutes a prosodic unit, viz. a clitic group (which in Old French is identical to a phonological word). Analyzing si as a functional category within the C-domain allows for an explanation of the fact that it does not occur in subordinate clauses where the same position is occupied by a different element (que). Section 3 is devoted to explaining the loss of the particle si in Modern French. From Old French to Early Modern French the initial syllable of a clitic group lost its secondary stress similar to the process in phonological words from Late Latin to Old French (the former, thus, represents a similar systematic change). This development went hand in hand with the apocope of the word-final

Introduction

xv

schwa making the right edge of phonological words and clitic groups prosodically strong. Since si was the initial syllable in a clitic group subject to the described prosodic changes, it became prosodically weak. This phonological reduction process is also attested for other initial clitics in Modern French (e.g., [il y a] > y'a 'there is'). This fact eventually led to the total loss of si in the 17th century. It is likely that the disappearance of the sentence particle si was facilitated by the existence of five homonymous particles in Old French, especially those who had a similar surface distribution in Old French, viz. the adverb of manner si 'so' (Modern French ainsi) and the subordinating conjunction si 'if which is also an element of the C-domain, although the latter does not directly compete with the sentence particle si. In her paper "Subject Case in Turkish Nominalized Clauses", Jaklin Kornfilt investigates the asymmetry between adjuncts and arguments, claiming that the argument-adjunct distinction can also play a role in determining the Case on the subject of a particular syntactic domain. She assumes that it is a clause's status as an adjunct versus as an argument which can determine the type of subject Case. The paper is also a case study in the interactions of morphology and syntax, as it claims that overt ^gT-(eement) determines subject Case (but only where Agr is licensed itself in this capacity). She shows another aspect of the morphology-syntax interaction, viz. the absence of a one-to-one relationship between syntactic and morphological Case: while morphological Genitive indeed reflects licensed nominal subject Case, morphological Nominative (possibly by virtue of being phonologically null) reflects both licensed verbal subject Case and default Case. Kornfilt makes the following specific proposals: (i) Turkish has three types of overt subjects: Those that bear genuine subject Case, those that bear default Case, and those which are Case-less. "Genuine subject Case" is licensed by a designated Case licenser; for Turkish, this is the overt ^gr(eement) marker. Such subject Case

xvi

Luka Szucsich & Uwe Junghanns

can be Nominative or Genitive in Turkish, depending on the categorial features of Agr. Default Case is possible as a last resort strategy, when subject Case is not licensed for an overt subject, and when no other licenser can license another appropriate Case (e.g. an ECM verb licensing Accusative). Case-less subjects are non-specific, and they are less mobile than the other two types of subjects. (ii) The proposed interaction between the argument-adjunct asymmetry and the designated subject Case licenser, i.e. overt Agr, is implemented as follows: Agr needs to be licensed itself in order to function as a subject Case licenser. This can happen in three ways: A. Categorially, i.e. via matching category features: A verbal Agr is licensed in a fully verbal extended projection, and a nominal Agr is licensed in a fully nominal extended projection. B. However, where there is a categorial mismatch, Agr must be licensed differently. This is when the argument-adjunct asymmetry comes into play: An argument domain bears a thematic index (cf. the proposal in Rizzi 1994 that arguments bear a "referential" index, while adjuncts don't); this index is inherited by the Agr (if there is one) that heads the argument domain in question. Kornfilt assumes that it is such indexation which licenses a categorially unlicensed Agr as a subject Case licenser. Thus, if Agr does not match its clause categorially, it is only where that clause is an argument that Agr will be able to license subject Case; where the domain is an adjunct, a categorially mismatched Agr cannot license subject Case. Thus, Kornfilt correctly predicts the existence of argumentadjunct asymmetries with respect to subject Case in categorially hybrid clauses, as well as the absence of such asymmetries in categorially homogeneous clauses. C. There is another way for a categorially mismatched Agr to receive an index and thus to get licensed as a subject Case licenser: via predication with an external head, i.e. when the domain headed by that Agr receives an index via predication (in headed operator-

Introduction

xvii

variable constructions like relative clauses and comparatives), and when, once again, the Agr head inherits the index of the clause in question. (iii) In all other instances (i.e. where there is no Agr, or where an existing, but categorially unlicensed Agr cannot receive an index by either "referential" theta-marking or under predication), no genuine subject Case is possible. The clause will have either a PRO subject or, if it has an overt subject, that subject will be in a default Case rather than in a genuine subject Case. Kornfilt discusses the issue of default Case and proposes criteria determining when default Case is possible and when it is not. She further proposes that the morphological realization of default Case may differ across languages; e.g. it is Accusative in English, while it is Nominative in Turkish. (iv) Subject Case is licensed locally within the extended functional projection of the clause; no clause-external nominal element is involved in this licensing—at least not directly, as the licenser of subject Case. (v) The account is compatible with approaches where AgrΡ is an independent projection (Pollock 1989, Kornfilt 1984), but also with approaches where Agr is positioned within the head of another functional projection, e.g. of the head of a Fm(iteness)jP (cf. Rizzi 1997), as long as Agr is housed in a projection separate from ΤΑΜ (i.e. Tense, Aspect, Mood). (vi) Kornfilt's paper is, at the same time, a case study concerning the two most widely used nominalization types in Turkish, with respect to genuine subject Case. The argument-adjunct asymmetry mentioned in (ii) is observed in one type of nominalization only (i.e. the indicative type) and not the other (i.e. the subjunctive type). The account proposed claims that, while both types of subordinate domains are DPs, only indicatives are also CPs. This explains the sensitivity of indicatives to "CP-level" phenomena and to thetamarking, and the lack of such sensitivity in non-indicative subordination. Kornfilt's paper consists of nine sections. The introductory section is followed by section 2 in which Kornfilt presents the two main

xviii

Luka Szucsich & Uwe Junghanns

asymmetries and establishes the relevance of Agr for subject Case. Section 3 offers a basic account of subject Case. Section 4 extends that account to predication. Section 5 draws preliminary conclusions. Section 6 discusses the nature of default Case. Section 7 proposes an explanation for when default Case may or may not be allowed. Section 8 discusses two rival approaches to the first asymmetry (i.e. the asymmetry between arguments and adjuncts in nominalized factive clauses) and presents counterarguments. Section 9 summarizes the conclusions and mentions some speculations. The paper is written in a general Principles and Parameters framework. In her paper, Esther Rinke discusses the licensing conditions for the omission of referential subjects in Old French (13th c.). She observes that the distribution of null subjects is constrained in the following way: Null subjects occur predominantly in main clauses with an initial non subject constituent and in conjunctional subordinate clauses which contain a preverbal topic. This has to be explained. To account for the facts Rinke draws on work by Mary Kato and Luigi Rizzi. Kato (1999) assumes that a [+pronominal] agreement system is the prerequisite for the licensing of null subjects. The agreement morphemes have the same grammatical status as a pronominal subject. As a result, they are able to check the EPP-feature of a given sentence via verb movement to AGRS°. This account has two important consequences: (i) pro can be eliminated, (ii) The projection of a specifier of AGRSP for EPP checking is excluded in null subject languages for reasons of economy. Rinke adopts Rizzi's (1997) split-CP hyphothesis. Thus, she assumes a number of functional layers in the C-domain, viz. ForceP, TopP, FocP, and FinP. The functional node Fin has an impact on the IP-system in that it selects the IP. Since the specification of the IPsystem in turn is crucial for the licensing of null subjects, Rinke proposes that the licensing of null subjects is contingent on the realization of the functional category Fin. This category is only available in main clauses and in conjunctional subordinate clauses

Introduction

xix

with a preverbal topic, but not in other types of subordinate clauses. This explains the distribution of null subjects in Old French. Rinke's crucial claim is that C is split into Force and Fin only in some types of clauses, e.g., in main clauses and conjunctional subordinate clauses with a preverbal adverbial topic, whereas in other types of clauses, e.g., in other subordinate clauses, Force and Fin collapse into one node. From this she goes further to claim that the licensing of null subjects in Old French is contingent on the realization of Fin. She finds additional evidence in the fact that null subjects in main clauses tend to co-occur with the sentence particle si. Following Ferraresi and Goldbach (2001) she assumes that si is placed in Fin. Rinke's account supports Rizzi's claim that Fin is the functional layer within the CP-system which selects the IP system and which may be endowed with agreement features. However, the nature of Fin and its precise feature composition are left for future research. In his paper, Andrew Spencer explores the notion of "paradigm" against the background of periphrastic or analytic constructions, i.e. multi-word combinations that express grammatical properties such as tense or aspect. He suggests treating them as reflexes of paradigmatic organization within a framework of paradigm-based morphosyntax (following Ackerman and Webelhuth 1998, Sadler and Spencer 2001, Spencer 2001), claiming that it is only by adopting the paradigm-based approach that the correct factoring of functions can be achieved and only a paradigm-based approach will eventually lead to an insightful account of these constructions. According to the paradigmatic view on periphrastic constructions, individual function words of multi-word combinations are not considered as lexical entries projecting their own set of features. Instead, they represent simply formatives which bear at most syntactic category features. Spencer presents two sorts of evidence for the paradigm-based view of periphrases, based mainly on the unusually rich periphrastic system of Bulgarian: (i) There are the kinds of gaps in these constructions that often show up in inflectional paradigms but which should not occur in genuinely compositional syntax, (ii) There are

xx

Luka Szucsich & Uwe Junghanns

instances of superexhaustivity, in which the paradigm extends beyond what would be expected from the normal combinatorial syntax. The author argues that Bulgarian periphrastic constructions (e.g. those containing the /-participle of 'be') would be extremely difficult to describe if we insisted on listing featural properties in lexical entries for function words. Spencer points out as an important problem to be tackled in the course of future investigation the question of how paradigm-driven mapping rules relate to other aspects of morphosyntax, such as linearization, clitic placement, agreement, ellipsis, etc. The paper consists of eight sections. Section 1 is the introduction. In section 2, Spencer introduces the idea that periphrastic or analytic constructions are best regarded as a kind of idiom, in which neither of the component parts has a meaning ("constructional idioms"). Section 3 discusses the fact that morphological paradigms are sometimes incomplete ("underexhaustivity"). In section 4, Spencer sketches the distinction between morphological and syntactic features (m-features vs. s-features). Section 5 introduces Ackerman's and Webelhuth's (1998) conception of "expanded predicate". In section 6, Spencer applies some of the discussion of section three to periphrastic paradigms in Bulgarian and develops a new concept of "superexhaustivity", under which periphrastic paradigms produce more forms than would be expected from the basic combinatorics of the syntax. Section 7 deals with the grammaticalization of clause structure in Bulgarian. Spencer demonstrates that future tense constructions in Bulgarian consist of constructional idioms that form paradigms. These paradigmatic periphrases contain subordinate clauses. Section 8 presents the conclusions. Barbara Stiebels defines as the goal of her paper to provide a programmatic and semantically based overview of possible affix orders within the domain of diathesis morphology. Two main questions need to be answered: (i) Which diathesis markers may be combined in principle? (ii) To what extent is the resulting morphological structure compositional, i.e. reflects the semantic composition and structural generation of forms?

Introduction

xxi

Included in the discussion are examples from the following languages: Chichewa, Quechua, Kinyarwanda (a Bantu language), Classical Nahuatl (an Uto-Aztecan language), Yucatec Maya, West Greenlandic, Chamorro, and Tukang Besi. Stiebels shows that Baker (1988) makes false predictions concerning possible diathesis combinations. According to her, the Mirror Principle is a violable constraint. She discusses the various versions of the Mirror Principle and its correlation with syntactic configurations as well as scope (Baker 1985,1988, Muysken 1986). Stiebels introduces the proposal made by Rice (2000) who puts emphasis on the availability of affix combinations that may receive different scope readings. Rice distinguishes three cases of affix combination: (i) Two affixes A and Β do not exhibit a scope relation; therefore, no affix order concerning A and Β is preferred. Both affix orders may be possible, or a language may arbitrarily choose one option, (ii) Each of the two affixes may take the other one into its scope. Therefore, both affix orders are relevant because they differ in their scopal interpretations, (iii) The scope relation is fixed such that only affix A may take affix Β into its scope; thus, only the order with A being the outer morpheme is possible. It is case (ii), the availability of two affix orders, that Siebels is most interested in. She assumes that differences in affix orders may result from semantic or syntactic properties. The paper consists of seven sections. After the introductory section, Stiebels discusses in section 2 the compositionality of affix orders and introduces the notion of transparent, restricted and opaque affix orders. Given that a particular combination of two morphemes A and Β has the universal potential for free order of application, and, hence, for the two affix orders A-B and B-Α, one must distinguish three subcases with respect to the resulting structures: (i)

Transparent affix order Both affix orders occur and transparently reflect the underlying scope relations.

xxii

Luka Szucsich & Uwe Junghanns

(ii)

Restricted affix order If, due to a language-specific constraint, only one affix order occurs which receives a surface-true, i.e. compositional interpretation, this affix combination is restricted. Restricted affix orders show a complete gap for a certain morpheme combination. (iii) Opaque affix order A given affix order has both the compositional and the noncompositional interpretation. The latter violates the Mirror Principle. These affix orders are opaque. Opaque affix orders only lack a distinct PF for one of the two readings. An even stronger case of opacity occurs if only one of the potential affix orders is allowed and if this has the interpretation of the inverse affix order, hence violates the Mirror Principle. Concerning the question of why non-transparent affix orders arise, Stiebels says: "One may speculate that different types of constraints are responsible for non-transparent affix orders: restricted affix orders presumably result from semantic and syntactic constraints, whereas opaque affix orders result from phonological and morphological surface constraints that dominate a constraint such as the Mirror Principle, or have to be explained in terms of languagespecific conditions on grammaticalization." Section 3 presents Baker's (1988) predictions for possible diathesis combinations and Stiebels' own analysis of diathesis operations. She develops a new version of the Mirror Principle·. "The affix order must mirror semantic composition." According to this principle, the semantics of a morphologically complex word is built up step by step from the innermost component to the outermost component. Unlike Baker, Stiebels assumes that the Mirror Principle is a violable constraint: opaque affix orders violate it due to some higher-ranked constraint. Following Wunderlich (1997b) and Dixon and Aikhenvald (2000), Stiebels distinguishes three types of

Introduction

xxiii

diathesis: (a) argument extension, (b) argument reduction, and (c) diatheses that bring about alternative argument realizations. Section 4 is concerned with diathesis combinations that yield an identical semantic output. Section 5 deals with diathesis combinations that differ in semantic terms. Section 6 treats diathesis combinations in which one of the possible orders subsumes the inverse one. Section 7 concludes the paper. Diathesis markers are treated as operations that change the Semantic Form and/or theta-grid of the base verb. The analysis does not invoke syntactic head-to-head movement. Stiebels argues in favour of an autonomous morphology in its interplay with semantics. She concludes that, in principle, diathesis markers can be combined in any order. Restricted affix orders mostly result from language-specific constraints on linking. The only invariant diathesis combination (causative and assistive or iteration of causatives) exhibits the predicted transparency. Only two cases of opaque affix orders have been attested: the combination of causative and applicative and multiple applicatives. In both cases, argumentextending diatheses are combined, which poses a challenge for structural linking in most languages. Stiebels points out that opacity and subsumptive affix orders are problematic. It needs to be shown whether opacity is strictly local, involving only adjacent morphemes, in order to define the way semantic processing works. A possible solution to the problem of subsumptive affix orders might be provided within the framework of Bidirectional Optimality Theory (Blutner 2000). Her final conclusion is that a typology of possible affix orders is not easily available within Optimality Theory because diathesis combinations interface with different modules of the grammar (syntax, semantics, morphology, discourse factors), which might not be evaluated in parallel. Jochen Trommer presents an analysis of cross-linguistic person agreement marking with transitive verbs and of direction markers in languages like Menominee and Turkana employing a constraintbased morphological theory, the so-called Distributed Optimality

xxiv

Luka Szucsich & Uwe Junghanns

(DO). Direction markers do not correspond directly to animacy hierarchies but are determined by universal constraints which correspond to prominence hierarchies. The constraints are defined in an OT-style way (they are violable and ranked). In the second and third section, Trommer develops the basic assumptions concerning the DO framework where he formulates the following principles: Morphology operating at the word-level spells out the output of syntax (locality); morphological morphemes have to be syntactically licensed (inclusiveness); all rankings of constraints are possible (free ranking). In DO word forms are defined as bundles of morphosyntactic features. The basic constraints which bring morphology to life are the so-called PARSE constraints requiring that certain feature combinations have to be realized by lexical items. On the other hand, there are constraints competing with PARSE constraints, first of all BLOCK constraints which suppress the spell out of morphemes. Additionally, PARSE constraints are linked to prominence scales determining which affix has to be realized with respect to a hierarchy of person features (e.g. [-3]/[+3], requiring that a 1st or 2nd person feature has to be spelled out by an affix if adjacent to a 3rd person head) and with respect to a syntactic (case) hierarchy ([+Nom]/[+Acc] where subject agreement is more prominent than object agreement). Subsequently, the author shows how particular rankings of universal constraints determine different patterns of person agreement markings in different languages like Menominee, Turkana, Dumi and Quechua. In section 4, Trommer turns to direction markers with a direct form where the subject of a transitive verb is higher on a prominence scale and an inverse form where the object is more prominent on that scale. He proposes that direction markers are in fact agreement markers. Trommer assumes a PARSE [Case] constraint as the trigger for the morphological spell out of direction marker, since these morphemes occur in contexts where both [+Nom] and [+Acc] features are present in a verb form. In Menominee the distribution of the direction markers is determined by further features (e.g. [ianimate], [±specified], [±obviative], i.e. these features are part of the feature specification of direction markers. Similar to the situation

Introduction

xxv

with person markers, the respective PARSE constraint is linked to a prominence hierarchy of the abovementioned features. There are, however, languages where particular transitive predications are not marked by direction markers. This zero marking is asymmetrical: If there are direction markers at all, there have to be inverse markers, but not direct markers; there are no languages with direct markers only. In section 5, the author captures this fact by introducing an IMPOVERISH constraint which again corresponds to a prominence scale. Depending on the ranking of this IMPOVERISH constraint and the PARSE [Case] constraint with respect to each other the desirable outcome is either a language with inverse markers only or a language with inverse and direct markers. In the 6th section Trommer discusses alternative analyses. In contrast to functional approaches he does not have to assume language-specific or even construction-specific feature hierarchies to capture problematic data where different person hierarchies seem to be involved (cf. the selection of prefixes and direction markers in Blackfoot). In his approach, these data are accounted for by assuming freely ranked constraints which are linked to a prominence scale. The correspondence relation to a prominence scale also accounts for the fact that certain constraints, at first glance, do not seem to be freely rankable (as assumed in other OT-style approaches). The paper by Ilse Zimmermann is concerned with the semantics of cases in Modern Standard Russian. Zimmermann argues that structural cases of complements can be characterized by abstract semantico-syntactic features which correspond to the semantic hierarchy of argument expressions and which are systematically interrelated with their morpho-syntactic realizations. Out of the various adjuncts to be found in Russian, she chooses the ones with instrumental case to illustrate her ideas. In particular, Zimmermann shows how morpho-syntactic case features of adjuncts are semantically interpreted and how one can account for the polysemy of the instrumental case by assuming context-dependent specifications of semantic parameters.

xxvi

Luka Szucsich & Uwe Junghanns

Zimmermann puts special emphasis on a strict differentiation between universal semantico-syntactic and language-specific morpho-syntactic case features. She assumes regular correspondences between the two types of case features. The paper consists of six sections. In section 1, Zimmermann outlines her objectives. The main aim of her paper is to bring together recent developments of the two-level semantics that has been developed by Manfred Bierwisch, the Linking Theory of Lexical Decomposition Grammar (Wunderlich, Stiebeis), and Roman Jakobson's case characterizations. She addresses the following questions: • How are case forms of complements and adjuncts interrelated with the semantics of these constituents? • Which types of cases must be distinguished? • Which types of configurations and of case features are involved? • Which complements of lexical categories count as structural complements? • Which rules guarantee the correct case realizations of argument expressions? • How do adjuncts get their case and how are they interpreted semantically? • How can one cope with the polysemy of adjunct cases? In section 2, Zimmermann characterizes the theoretical framework, in particular the division of labour between morphology, syntax, and semantics and the interface role played by the argument structure of lexical entries of the functor expressions. The theoretical background that Zimmermann adopts comprises a minimalist framework of sound-meaning correlation (Chomsky 1995), a lexicalist conception of morphology (Stiebels and Wunderlich 1994; Wunderlich and Fabri 1995; Wunderlich 1997a), and the differentiation between Semantic Form and Conceptual Structure (two-level semantics; cf. Bierwisch 1983; 1987, 1997; Bierwisch and Schreuder 1992; Lang 1987,1990,1994; Dölling 1997).

Introduction

xxvii

In section 3, Zimmermann demonstrates the far-reaching parallelism of verbal constructions and their nominalizations, with systematic case variation of structural arguments. Section 4 is concerned with case licensing. It presents a system of rules correlating abstract semantico-syntactic case features in the argument structure of lexical governors and the morpho-syntactic case features of corresponding argument expressions. Section 5 deals with the semantics of the instrumental as adjunct case. In subsection 5.1., Zimmermann discusses instrumental phrases as adverbial modifiers. Sub-section 5.2. is devoted to instrumental phrases as secondary predicates. Section 6 summarizes the paper. Zimmermann claims that cases of noun phrases can have semantic import. No case is a structural or semantic case per se. DPs with structural case are systematically related to the semantics of their governor. DPs as adnominal or adverbial modifiers and NPs as secondary predicates are associated with very abstract case semantics. In her analysis, structural cases are those predictable case realizations of complements which are correlated with the abstract case features +/-h(igher) r(ole), +/-l(ower) r(ole) in the argument structure of lexical governors and mirror the semantic hierarchy of DPs as argument expressions. In order to demonstrate DPs as adverbial modifiers, Zimmermann analyses Russian instrumental phrases. The semantic interpretation of the instrumental expression draws from a variety of possible meanings. Zimmermann invokes a semantic template that provides the morpho-syntactic case features at the level of S(emantic) F(orm) with a semantic parameter that is specified only at the level of C(onceptual) Structure). Semantic templates apply in the case of another type of instrumental expressions too—instrumental NPs that function as secondary predicates. In this case, the template brings in a semantic parameter to be specified at CS, depending on the context. Zimmermann concludes her paper by stating that the differentiation between SF and CS as two levels of semantic

xxviii

Luka Szucsich & Uwe Junghanns

interpretation turns out to be fruitful with respect to the semantics of cases.

Notes * All papers except of those by Tania Avgustinova and Esther Rinke derive from presentations at the abovementioned workshop organized by the editors. We take this opportunity to thank Tibor Kiss, Jürgen Pafel and Anita Steube for encouraging us to organize the workshop as well as all participants of the workshop for fruitful discussion of current issues in syntax and morphology (the volume's contributers, as well as David Adger, Artemis Alexiadou, John Bailyn, Ursula Bredel, Gisbert Fanselow, Daniel Harbour, Andrej Kibrik, Horst Lohnstein, Rosemarie Lühr, Gereon Müller, Jamal Ouhalla, Jean-Yves Pollock, Adam Przepiörkowski and Dieter Wunderlich). We are indebted to Gerhild Zybatow for her support as the main organizer of the DGfS-23 meeting and to Artemis Alexiadou and Tracy A. Hall for including this volume in the Interface Explorations series. Special thanks to David Adger, Piotr Banski, Martin Haspelmath, Christian Huber, Tracy Holloway King, Iliyana Krapova, Andrd Meinunger, Eduardo Raposo, Marga Reis, Maaike Schoorlemmer, Michal Starke, Anita Steube, Misha Yadroff and Ilse Zimmermann for reviewing the papers submitted for this volume. We are grateful to Sigrid Lipka, Andrew Mclntyre and Evan Mellander for helping us with the proofreading.

References Ackerman, Farrell and Gert Webelhuth 1998 A Theory of Predicates. Stanford University: Center for the Study of Language and Information. Baker, Mark 1985 The Mirror Principle and morphosyntactic explanation. Linguistic Inquiry 16: 373-415. 1988 Incorporation: a theory of grammatical function changing. Chicago: The University of Chicago Press. Bierwisch, Manfred 1983 Semantische und konzeptuelle Repräsentation lexikalischer Einheiten. In: Rü2i6ka, Rudolf and Wolfgang Mötsch (eds.), Untersuchungen zur Semantik, 61-99. (Studia grammatica 22.) Berlin: Akademie-Verlag

Introduction 1987

xxix

Semantik der Graduierung. In: Bierwisch, Manfred and Ewald Lang (eds.), 91-286. 1997 Lexical information from a minimalist point of view. In: Wilder, Chris, Hans-Martin Gärtner and Manfred Bierwisch (eds.), 227-266. Bierwisch, Manfred and Ewald Lang (eds.) 1987 Grammatische und konzeptuelle Aspekte von Dimensionsadjektiven. (Studia grammatica 26/27.) Berlin: Akademie-Verlag. Bierwisch, Manfred and Robert Schreuder 1992 From concepts to lexical items. Cognition 42: 23-60. Blutner, Reinhard 2000 Some aspects of optimality in natural language interpretation. Journal of Semantics 17: 189-216. Bresnan, Joan (ed.) 1982 The Mental Representation of Grammatical Relations. Cambridge (MA): The MIT Press. Chomsky, Noam 1995 The Minimalist Program. Cambridge (MA): The MIT Press. 2000 Minimalist Inquiries: The Framework. In: Roger Martin, David Michaels and Juan Uriagereka (eds.) Step by Step, 89-155. Cambridge (MA): The MIT Press. 2001 Derivation by Phase. In: Michael Kenstowicz (ed.) Ken Hale: a Life in Language, 1-52. Cambridge (MA): The MIT Press. Cinque, Guglielmo 1999 Adverbs and Functional Heads: A Cross-linguistic Perspective. New York/Oxford: Oxford University Press. Dixon, Robert M.W. and Alexandra Y. Aikhenvald 2000 Introduction. In: Robert M. W. Dixon and Alexandra Y. Aikhenvald (eds.), Changing valency: case studies in transitivity, 1-29. Cambridge: Cambridge University Press. Dölling, Johannes 1997 Semantic form and abductive fixation of parameters. In: Sandt, Rob van der, Reinhard Blutner and Manfred Bierwisch (eds.), From Underspecification to Interpretation, 113-139. (Working Papers of the Institute of Logic and Linguistics.) Heidelberg: IBM. Ferraresi, Gisella and Maria Goldbach 2001 Particles and sentence structure. Paper presented at DGfS-23, Leipzig, Feb 28 - March 2, 2001. [The written version of Ferraresi and Goldbach's paper is contained in this volume.] Kato, Mary 1999 Strong and weak pronominals in the null subject parameter. Probus 11: 1-37.

xxx

Luka Szucsich & Uwe Junghanns

Komfilt, Jaklin 1984 Case marking, agreement, and empty categories in Turkish. Ph.D. dissertation, Harvard University. Lang, Ewald 1987 Semantik der Dimensionsauszeichnung räumlicher Objekte. In: Bierwisch, Manfred and Ewald Lang (eds.), 287-458. 1990 Sprachkenntnis, Objektwissen und räumliches Schließen. Zeitschrift für Literaturwissenschaft und Linguistik 78: 59-97. 1994 Semantische und konzeptuelle Struktur: Unterscheidung und Überschneidung. In: Schwarz, Monika (ed.), Kognitive Semantik: Ergebnisse, Probleme, Perspektiven, 25-40. (Tübinger Beiträge zur Linguistik 395.) Tübingen: Narr. Muysken, Pieter 1986 Approaches to affix order. Linguistics 24: 629-643. Nordlinger, Rachel 1998 Constructive Case: Evidence from Australia. Stanford, California: CSLI Publications. Pollard, Carl and Ivan A. Sag 1994 Head-driven Phrase Structure Grammar. Chicago: University of Chicago Press. Pollock, Jean-Yves 1989 Verb movement, Universal Grammar, and the structure of IP. Linguistic Inquiry 20: 365-424. Rice, Keren 2000 Morpheme order and semantic scope. Cambridge: Cambridge University Press. Rizzi, Luigi 1994 Argument/Adjunct (a)symmetries. In: Guglielmo Cinque, Jan Köster, Jean-Yves Pollock, Luigi Rizzi, and Raffaella Zanuttini (eds.), Paths Towards Universal Grammar: Studies in Honor of Richard S. Kayne, 361-376. Washington, D.C.: Georgetown University Press. 1997 The fine structure of the left periphery. In: Liliane Haegeman (ed.), Elements of Grammar, 281-337. Dordrecht: Kluwer Academic Publishers. Sadler, Louisa and Andrew Spencer 2001 Syntax as exponent of morphological features. In: Gert Booij and Jaap van Marie (eds.), Yearbook of Morphology 2000, 71-96. Dordrecht: Kluwer Academic Publishers. Sgall, Petr, Eva HajiCovä and Jarmila Panevovä 1986 The Meaning of the Sentence in its Semantic and Pragmatic Aspects. Dordrecht: Reidel.

Introduction

xxxi

Spencer, Andrew 2001 Agreement morphology in Chukotkan. In: Wolfgang U. Dressier, Oskar E. Pfeiffer, Markus Α. Pöchtrager and John R. Rennison (eds.), Morphological Analysis in Comparison, 191—222. Amsterdam: John Benjamins Publishing Co. Stiebels, Barbara and Dieter Wunderlich 1994 Morphology feeds syntax: The case of particle verbs. Linguistics 32: 913-968. Wilder, Chris, Hans-Martin Gärtner and Manfred Bierwisch (eds.) 1997 The Role of Economy Principles in Linguistic Theory. (Studia grammatica 40.) Berlin: Akademie Verlag. Wunderlich, Dieter 1997a The minimalist model of inflectional morphology. In: Wilder, Chris, Hans-Martin Gärtner and Manfred Bierwisch (eds.), 267-298. 1997b CAUSE and the structure of verbs. Linguistic Inquiry 28: 27-68. Wunderlich, Dieter and Ray Fabri 1995 Minimalist morphology: An approach to inflection. Zeitschrift für Sprachwissenschaft 14.2: 236-294.

Metagrammar of systematic relations: a study with special reference to Slavic morphosyntax Tania Avgustinova

1.

Introduction

A motivation for the present study is the assumption that shared grammatical resources for Slavic languages should, by design, take grammatical relatedness seriously.1 While the categorisation of primitive linguistic entities tends to be language-specific or even construction-specific, the relationships observable in linguistic constructions between these entities allow various degrees of abstraction corresponding to important metagrammatical generalisations. The notion of construction is commonly assumed to refer to the syntactic arrangement of patterning within a grammatical unit. Cross-linguistically, the key role in building syntactic constructions is played by what is interpreted as structural syntactic dependencies. This essential generalisation tends to be implicit both in modern and in traditional linguistic research, and certainly deserves attention as a feasible linguistic universal. But structural syntactic dependencies are linguistic abstractions. Being not directly observable, they are externalised by various presumably general mechanisms which compete or complement each other. Abstract relations are locally distributed as partial information across syntactic items, i.e. words and larger fragments ("constituents") - consider the following statement from (Comrie and Corbett 1993): Syntactic relationships deal with the interdependence of words in sentences or in segments of sentences (syntactic constructions). These relationships may be purely semantic (for example, agent, patient, beneficiary) or they may represent different levels of linguistic structure: syntagmatic (subject, direct or indirect object, predicate, complement) or paradigmatic (case, gender, person). The latter enter into larger classes of morphosyntactic relationships, known as grammatical categories ... (p. 107)

2

Tania Avgustinova

While paradigmatic relations are functional contrasts and involve differentiation, syntagmatic relations are possibilities of combination of interacting items chosen within a framework of rules and conventions (both explicit and implicit), e.g., on the basis of a grammar system. They link together elementary constituent segments or minimal significant units which themselves belong to various paradigms. Mature linguistic theories develop nowadays powerful ontologies for linguistic objects and categories (like words, phrases, sentences) as well as rich inventories of relations among properties of these objects (e.g., agreement, subcategorisation, long-distance dependencies, etc.). Yet, no consistent account of grammatically relevant relations can be found. The key hypothesis put forward here is that systematic relations motivate shared patterns of variation cross-linguistically as well as across constructions. To adequately encode the metagrammar of systematic relations, we need an ontological level of grammatical abstraction. This ontological level can be encoded as compatible multidimensional hierarchies of types of linguistic entities, with constraint inheritance from more general to more specific types. In short, the hierarchy allows for the expression of (possibly cross-cutting) generalisations and a specification of (in)compatibility between types. The generalisations are captured by factoring them out as constraints on super-types, as well as allowing types to have more than one super-type. It is worth noting, however, that it is rare in such multiple inheritance hierarchies to find all possible combinations of the super-types instantiated via mutual sub-types. And this is one sense in which the type hierarchy represents linguistically relevant (sub)generalisations. Attempts to derive the range of possible syntactic combinations of any two constituents from a multidimensional inheritance network of basic relational grammatical types open new ways of classifying relations that are non-standard or marginal in most theories. A number of these relations fall out from the proposed cross-classification in natural and theoretically rewarding ways. The classification presented here is originally designed to systematise the inventory of syntactic relationships found across Slavic languages. A far-reaching outcome of this study is promoting the view that systematic relations holding

Metagrammar of systematic relations: Slavic morphosyntax

3

between the components of syntactic constructions must be treated as research objects in their own right. As soon as the relations holding in syntactic constructions are organised in an HPSG-style type hierarchy, the type subsumption can be interpreted as modelling a continuum from general - and presumably universal - systematic relations to more and still more specific instances of these relations resulting from admissible crossclassifications. Every type in such a multidimensional hierarchy is associated with a set of constraints which is appropriate for this particular type and is inherited by all its immediate and non-immediate sub-types. Thus, linguistic generalisations of different nature can be captured in a theoretically elegant way, allowing immediate formalisation (e.g., in HPSG) and straightforward implementation. For the actual linguistic material, we need a representation that would be close to the overt perceptible expressions of language (syntactic items) and to the abstract relational information that they directly express (arrays of systematic relations). Therefore, in visualising the systematic relations in syntactic constructions, relational charts will be used, a representation of linguistic examples originally employed by (Avgustinova and Uszkoreit 2000). A relational chart is a diagram of the type illustrated in (Figure 1) for a sentence consisting of the items α, β, γ, and δ (exactly in this order). The array of systematic relations holding between any two selected items occupies, thus, the respective "crossing" cell. Figure 1. Relational chart diagram | item a

systematic relations (α β) item β

systematic relations (α γ) systematic relations (β γ) item γ

systematic relations (α δ) systematic relations (β δ) systematic relations (γ δ) item δ

Information of various types of constituency with respect to nonminimal syntactic items, i.e. multiword phrases, can be encoded in the relational chart too. This requires merging the respective cells as sketched in (Figure 2) for the systematic relations holding between

4

Tania Avgustinova

the item α and the complex item βγ, or between the complex item βγ and the item δ. Figure 2.

Relational chart with merged cells (α δ)

(a( 3 γ))

item a

(α β) item β

(α γ) (ßr) item γ

(β δ) (γδ)

((ß?) δ) item δ

Yet, merging cells with respect to particular relation types does not block the encoding of other relation types that require the same cells to be separated - i.e. relations holding between α and β, between α and γ, between β and δ, and between γ and δ. In order to better illustrate the approach developed here and for convenient reference throughout the main discussion in Section 2, all relational charts of actually considered linguistic examples will be given in Section 3.

2.

Systematic relations

The major dimensions of classification introduced for (the arrays of) systematic relations discernible in syntactic constructions are sketched in (Figure 3).2 The focus of our attention will be on segmental systematic relations in terms of syntagmatics, as they play a constitutive role in syntax. In accord with the traditional "form-function" perspective in theoretical linguistics, it is important to distinguish dimensions of observable syntagmatics (which is concerned with the overt linguistic form) and structural syntagmatics (which is concerned with the covert linguistic function). Structural syntagmatics is crucial in interpreting the observable syntagmatic relations which, in turn, can be classified as combinatorial (i.e. morphosyntactic) and alignment (i.e. configurational). The intonational and prosodic aspect is certainly important too in establishing instant connections between linguistic entities in various

Metagrammar of systematic relations: Slavic morphosyntax

5

constructions, but due to the morphosyntactic orientation of the present work I will not consider it here. Nevertheless, the integration of supra-segmental systematic relations into the ontology, as well as the accommodation of any further relevant dimension, is immediately possible and can be performed in the same principled way. Figure 3. Systematic relations: dimensions of classification systematic relation 'segmental'

'supra-segmentar

intonation / prosody

syntagmatics Mod az udvaron vdrhattak the yard-on wait-poss-past-3pl

b. Az eliteltek csak the convicts only α lätogatökra. the visitors-for 'The convicts could / were allowed to wait for the visitors only in the yard.' (4)

a.

b.

It is POSSIBLE [that they waitED].

POSS [PAST [...]]

Mod > Τ k talän α mäsik kapunäl vdrhattak, they perhaps the other gate-at wait-poss-past-3pl ezert kerültük el egymäst. so miss-past-1 pi asp each-other 'They may perhaps have waited at the other gate, that's why we missed each other.'

Clearly, the form in (2) has two scope interpretation possibilities. If, as generally assumed, the morpheme order is the mirror image of the order of syntactic projections, then the Τ > Mod scope relation is the straight order, while the Mod > Τ scope relation is the inverse, i.e. unexpected one.

On-line Morphology: Hungarian Verbal Inflection

29

These data pose an immediate problem for the combined validity of (la, b, c). A single morpheme order presupposes a single, invariant projectional hierarchy by (lc), in accordance with (lb), but variable scope order necessitates variable syntactic hierarchy by (la), not reflected here in morpheme order, in defiance of (lc). In the particular case of (2), for example, the V-poss-past morpheme order suggests that the projectional hierarchy is "T [ Mod [ V ]]", but this warrants only the Τ > Mod scope relation, while the existence of the Mod>T scope order suggests some syntactic configuration where Mod c-commands Τ at some level. Since (la) enjoys the widest currency of the three, and it is often taken to be axiomatic (assumed as a principle, cf. Aoun and Li's (1993) Scope Principle), and (lb) is empirically well-motivated,3 we must loosen up (lc), or more precisely, since (lc) is "merely" a generalization (which may or may not be universally valid), we must derive (lc) in such a way as to make room for the observed variability. The alternatives would be (i) to resort to covert rearrangement, dispreferred on theoretical grounds (see above), or (ii) to seek a semantic solution, relegating the issue of scope inversion to semantics. But the inspection of further data suggests that such an approach may be even more problematic. Next, consider the case of cond and past cooccurring. (5) shows the verb form, and (6, 7) the two possible scope orders: (5)

Vdr-t-ak vol-na. wait-past-3pl Vdum-cond 'They would have waited.' or 'They wished to wait.'

(6)

a.

It WOULD be the case [that they waitED]. COND [PAST [...]]

M>T

b. Az rök vdrtak volna, ha mondtad the guards wait-past-3pl Vdum-cond if tell-past-2sg volna nekik. Vdum-cond to-them 'The guards would have waited if you had told them to.' (7)

a.

It WAS the case [that they WOULD wait]. PAST [COND [...]]

Τ> Μ

30

Huba Bartos

b. A vendegek igazän värtak volna meg, the guests really wait-past-3pl Vdum-cond still de tul kes volt. but too late was-3sg 'The guests really would have waited / wished to wait still, but it was too late.' This, again, looks very much like the case of Τ and Mod: both the "straight" and the "inverted" scope order is available, for the same morphemic sequence. However, as it turns out, while this too is an example of scope variance, it is still unlike the situation we have just seen for Τ and Mod — this is not a simple case of scope inversion, in the sense that while (6) presents a clear case of Μ scoping over T, (7) is a rather more complicated matter, where what we perceive at first sight as narrow-scope mood is in fact some sort of modality. Nevertheless, I wish to extend my general analysis to this type, as well. More details about the data will be presented, along with the proposed treatment, in section 4.3.2. Consider the third possible pair: conditional mood and possibility combined, as in (8). (9) gives the straight scope reading, but (10) shows that scope inversion is unavailable here, unlike in the former two cases.4 (8)

Vär-hat-nä-nak. wait-poss-cond-3pl 'They could wait.' / *'They possibly wish to wait.'

(9)

a. b.

(10)

a.

It WOULD be

[for them to wait]]. COND [POSS [...]] M>Mod Ha sziikseges volna, värhatnänak idebent. if necessary would-be-3sg wait-poss-cond-3pl in-here 'They could wait in here if it was necessary.' [POSSIBLE

*It is POSSIBLE [that they WOULD wait]. *POSS [COND [...]]

*Mod > Μ

On-line Morphology: Hungarian Verbal Inflection

b.

31

#Nem ertem, miert nem mennek haza; not understand-lsg why not go-3pl home talän vdrhatndnak? perhaps wait-poss-cond-3pl Ί don't see why they don't go home; maybe they would wait?'

If Mod can invert with T, and Τ can invert with M, why can Mod not invert with M? One might find some semantic factor, though, to rule out the inversion on semantic grounds. However, what is the simplest and most straightforward case for a syntactic solution (as will be shown presently) is the worst case for the semanticist. When all the three contentful inflectional categories are marked on a verb form, only one scope order is attested, the full straight order (Μ > Τ > Mod), even though in principle there could be as many as six scopal variants, and even if we take into account the lack of scope inversion between Μ and Mod, in the light of the attested inversions in (4) and (7), we would still expect four different readings: (11)

Vdr-hat-t-ak vol-na. wait-poss-past-3pl Vdum-cond 'They could have waited.'

(12)

a.

It WOULD be the case [that it WAS [POSSIBLE [for them to wait]]]. Μ > Τ > Mod b. A firik vdrhattak volna, ha valaki the boys wait-poss-past-3pl Vdum-cond if someone szolt volna nekik. tell-past-3sg Vdum-cond to-them 'The boys could have waited if someone/anyone had told them to.'

(13)

a. c. e.

Μ > Τ > Mod * T > M > Mod *Mod > Μ > Τ

b. d. f.

*M > Mod > Τ *T > Mod > Μ *Mod > Τ > Μ

32

Huba Bartos

It is obviously not an easy task to find a semantic explanation for the loss of scope variance in this case, as compared to the cases in (2) and (5). In the model presented and defended here, however, this case will simply fall out as a natural outcome. For this to be seen, the next sections will introduce the model of morphosyntax, and provide the analyses for the above data. As a final note to this section, let me dwell briefly on the nature of modality represented by -hAt. Kiefer (1981) gave a detailed inventory of the types of modality associated with this suffix. They can nevertheless be classified under the two widely used terms: root vs. epistemic modality (the former comprising deontic, circumstantial, dispositional, etc. modalities). Now, as the few illustrative examples above also show, generally speaking, the epistemic reading of -hAt and wide scope Mod (i.e. Mod > T) go hand in hand, while the root reading corresponds to the narrow scope case. This might lead us to posit two distinct Mod domains, one above Τ (call it Modepist), and another one below Τ ( M o d ^ . However, on the one hand this would still beg the question of why/how the same affix order arises with both types of Mod, and on the other hand, the generalization is not fully true. While root modality never scopes over T, epistemic possibility may have narrow scope with respect to T, as e.g. in (14): (14)

St. Germainben tegnap hovihar volt, tehdt Pdrizsban St. Germain-in yesterday blizzard was so Paris-in is havaz-hat-ott. also snow-poss-past-3sg 'There was a blizzard yesterday in St. Germain, so it could be snowing in Paris, too.'

Weather predicates cannot normally be interpreted with root modality, since natural phenomena do not conform to obligations/ permissions, do not have wishes or dispositions, etc., so any sort of root reading is out. Then this must be epistemic modality, but it is embedded under past tense: the primary interpretation of the second clause is not "It is POSSIBLE [that it snowED in Paris]" but rather: "it WAS a case of [being POSSIBLE [for it to snow in Paris]]". Although the distinction is subtle, the two cases can still be kept

On-line Morphology: Hungarian Verbal Inflection

33

apart, and havazhatott in (14) is interpreted with Τ > Mod.3 Therefore I will not pursue here the line of splitting Mod apart in syntax.

3.

On-line morphology

The model of syntax and morphology advocated here is essentially the same as the one presented in Bartos (2000)6 — in essence, it is the seemingly somewhat strange marriage of a derivational, minimalist system of structure building (Chomsky 1995, 1999, 2001) and the antilexicalist, late-insertion model of Distributed Morphology (Halle and Marantz 1993, Marantz 1997). I assume, in line with mainstream minimalism, that syntactic structure building proceeds bottom-up, cyclically, by the operation Merge, and displacement of elements (Move) is triggered by attraction, the aim of which is the satisfaction (though not necessarily the checking) of some feature. But I follow Distributed Morphology in assuming that words are assembled in the course of derivation, rather than lexically, and that the actual soundshapes are inserted in/after morphology, so they are not present in syntax. Moreover, I propose that the morphological derivation runs on-line with syntax: each step of the syntactic derivation is immediately scanned by morphology, and if that step has any consequence for word-structure, the corresponding morphological operation is carried out promptly, before the next step is taken in syntax. So for instance if two heads are combined in syntax, e.g. by adjunction, a word-domain comprising the morphological exponents of the two heads is immediately established in morphology. Feedback from morphology to syntax, and concomitant morphologically driven syntactic repair is possible, too. Beyond the general minimalist operations, I introduce an operation called morphosyntactic merger, which resembles (and replaces) Halle and Marantz's (1993) morphological merger. The crucial difference is that morphosyntactic merger includes chain formation between the participant items, so it also has syntactic consequences, beside the obvious word-formational ones, and therefore it takes place in syntax, rather than at the level of morphology. The introduction of this operation should not be seen as extra cost, further

34

Huba Bartos

complication of syntax, however, since it replaces head movement in all those cases where it would be carried out to attach affixes to a stem (or, equivalently, where the checking of an affixal feature would be the only motivation of head movement). In fact, this should be seen as an improvement on frameworks applying adjunctive headmovement, because head-movement has certain problematic properties, excluding this type of movement from the general family of movement operations — for a discussion see Brody (1997). The exact definition of morphosyntactic merger is as follows: (15)

MORPHOSYNTACTIC MERGER

A [+affix] category X can morphosyntactically merge with a potential stem Y under structural adjacency. This yields a word domain {x, y} (where χ realizes X, and y realizes Y) at the level of morphology, and a chain in syntax. (16)

(cf. also Frampton and Gutmann 1998, 1999) X and Y are structurally adjacent iff (i) X c-commands Y, and (ii) there is no Z, such that the projectional status of Ζ is identical to that of X, and X c-commands Z, and Ζ ccommands Y STRUCTURAL ADJACENCY

As can be seen from these definitions, morphosyntactic merger is strictly local, thereby incorporating the effects of the head movement constraint even in the absence of movement as such. The reference to "projectional status" in the definition of structural adjacency concerns the contextually determined maximality/minimality of projections, in the sense of Chomsky (1995): a projection is minimal, if it contains only word-level terms, and it is maximal if it is immediately dominated by a categorially distinct projection. The status identity requirement in (16ii) ensures that items which are both minimal and maximal projections at the same time (such as certain adverbs), do not count as blockers of structural adjacency for purely minimal or purely maximal projections.7

On-line Morphology: Hungarian Verbal Inflection

35

Once morphosyntactic merger (or in fact any other word-forming operation) joins two elements, the order of their morphological exponents will be determined by their lexical properties: if for instance in a word-domain {x, y} Λ: is lexically specified as a suffix, it will follow y (the stem) in the surface phonological form and if χ is a prefix, it will precede y. Furthermore, since morphosyntactic merger is a binary operation, items are joined pairwise, and each step is a new cycle, so in the word structure every new item can only be added to the edge" of the word-form created by the previous step. In other words, bracket erasure obtains after every step in the morphological derivation (a standard assumption about morphology in general), so once a word-domain w, is embedded into a larger word-domain w2, the inner structure of wl becomes unavailable for any later morphological operation.' Finally, the spell-out locus of the morphosyntactically merged word-form is determined by the position of the root, i.e. the whole inflected word will be pronounced in the position of the root category. That is, the position of the morphophonologically independent element is what counts, which is natural, since this element is the one that "picks up" the dependent elements: the affixes.

4.

The analyses

On the basis of the scope relations in (2-13), and the affix order facts (assuming that the Mirror Principle is largely valid), we can establish that the likely order of the functional projections under examination is (Agrs > Agr0 >) Μ > Τ > Mod.10 Agr suffixes always appear after other inflectional suffixes, and Agr-categories do not enter into scope relations at all, so we have every reason to place them to the periphery in syntax, too. Let us now treat the combinations of the contentful inflectional categories one by one, starting with cases where only two of them are marked.

36 Huba Bartos 4.1. Tense and Modality The figure in (17) depicts the syntactic structure created up to the insertion and projection of some Agr category; the relevant feature content of the inflectional heads is given on the right-hand side: (17)

AgrP

Agr [+aff, +V,...] Τ [+aff, +V, +past] Mod

VP

Mod [+aff, +V, +poss]

Here the morphosyntactic derivation is obvious and straightforward: the word-domain {{{V, Mod}, T}, Agr}word and the chain are created by a sequence of applications of morphosyntactic merger, and the order of the affixes follows from the order of assembly plus their lexical property of being all suffixes. As regards the LF interpretation, the Τ > Mod scope relation is selfevident: Τ asymmetrically c-commands Mod. The more interesting question concerns the derivation of inverse scope between Τ and Mod. For this to obtain, the derivation must reach a point where Mod gets to c-command T, but it must be ensured that the affix order remains intact. Suppose that at the point when the projection of TP is complete, we have an option of merging in an entirely contentless Μ — the V-form has not been specified for this category, so it is potentially available, but if inserted, it will be inert (thus V will be interpreted with unmarked, indicative mood). This Μ is a mere categorial frame, totally void of feature content, including categorial underspecification — let us call it a proxy (in a sense remotely reminiscent of, though in fact markedly different from, the notion of proxy in Nash and Rouveret 1998). Alternatively, we may look at this proxy "M" as a templatic slot, assuming a

On-line Morphology: Hungarian Verbal Inflection

37

(possibly universal, or universally grounded) template of functional projections, much like that presented in Cinque (1999). It is a general property of Hungarian clause structure (as a particular instantiation of the universal template) that there are exactly three contentful inflectional categories available: Μ, T, and Mod, thus there are exactly three templatic slots. And whenever one of these is not specified by a particular item drawn from the (narrow) lexicon, it can be made use of as a proxy, if merged into the structure, which is an option in all such cases. As a matter of fact, though, in Hungarian only the templatic slot for Μ is ever used as a proxy, since Τ is always specified (there are no tenseless clauses — infinitivals being marked, too, for null tense), and while it is true that Mod can be left unspecified, it is the lowest slot of this domain, so there is no other inflectional category further down in the clause structure that could access it. So, coming back to the particular case under scrutiny, the proxy "M" will act as a hosting site for some other verbal inflectional category to fill it, i.e. to flesh it out featurally, and the movement due to the "vacuum-effect" of the unspecified proxy will have to be substitution, rather than adjunction, for it to gain proper specification. Otherwise it will cause a crash at LF, being uninterpretable. In principle, there are two categories that may fill it: Τ and Mod. But there are at least two factors that single out Mod as the proper filler of the proxy. For one thing, mood and modality have generally been assumed to be closely related notions. And while any detailed discussion of even just the Hungarian facts would go far beyond the scope of this paper, it is worth mentioning that there are obvious links of compatibility between the two categories. For example, while deontic modality is compatible with either realis or irrealis mood (cf. 0KJenny had to eat the pasta, and she did / but she didn V), epistemic modality is not compatible with either — it obviously only tolerates unspecified mood (cf. "(For all we know,) Jenny may have eaten the pasta, and in fact she did / but in fact she didn 7). And the modality of volition is compatible only with irrealis mood; see 4.3.2 below. In view of this close relation between Μ and Mod, it is likely that Mod is the expected filler of a proxy M.

38

Huba Bartos

Moreover, although at first sight Τ seems to be closer to the proxy Μ than Mod, recall that Τ and Mod are links of the same chain now, so they plausibly count as equidistant for the attractor. This is "backwards equidistance", the reversal of Chomsky's (1995, chapter 3) notion of equidistance, in order to fit the attraction-based conception of movement, as in Chomsky (1995, chapter 4)." So we have two options to satisfy the attracting needs of M: raising T, or raising Mod. The former yields an output that is not distinct in any significant way from the output of the alternative derivation in which no proxy Μ is accessed: neither the phonological string, nor the semantic interpretation is affected. Therefore inserting a proxy M, then raising Τ to it constitute redundant steps, so this derivation is barred on grounds of economy (cf. Fox 1995, Reinhart 1997, Chomsky 1998). We are then left with the second option: raising Mod to M. This is a case of substitution, so the category label of the proxy will be provided by the raised element, Mod. This yields (18): (18)

ModP Mod, A

TP

Μ []

Mod [+V, +poss]

Τ [+aff, +V, +past] Mod [+aff, +V, +poss]

Now Mod c-commands T, which has two welcome consequences for us. First: this makes the inverse scope reading Mod > Τ available, second: this leads to an output to semantics that is distinct from the one arrived at by the simplest derivation, i.e., without inserting the proxy M, so economy will not block this more complex derivation. Let us turn our attention now to what happens in morphology. The word-form up to Τ is built as in the derivation of the straight scope order: {{V, Mod}, T}word. Next comes the insertion of the empty M, which has no morphological reflex whatsoever. Then we raise Mod to M. But now morphology will do nothing: specifically, it cannot lift

On-line Morphology: Hungarian Verbal Inflection

39

the Mod-affix (-hAt) out of its place between the V-stem and the Taffix, because of strict cyclicity / bracket erasure. The internals of the deeper domain, {V, Mod} have become inaccessible by now. In other words: once Τ is added to the word-form, the Mod-affix is frozen in its place, in the innermost domain. Morphology attempts to follow syntax, but it has its own principles, hence its own limits. This is why and how the two distinct syntactic structures, corresponding to the two distinct scope readings, have the same morphological correlate: [V-Mod-T-Agr ...]word. So the moral of this subsection is the following: Phrase structure can be built in accordance with the invariant functional projectional hierarchy hypothesis (lb), yet scope relations can subsequently be changed, provided there is space (a proxy) for such moves; and identical morphological sequences can be arrived at via different derivations — thus the Mirror Principle is maintained derivationally, up to the separate limitations of the grammatical modules affected: morphology follows syntax to the extent that it can. Also, the mirroring generalization is understood as unidirectional, rather than bidirectional (as in Baker (1985)): morphology strives to mirror (i.e. follow) syntax, but not the other way round.12

4.2. Conditional Mood and Modality Recall that we found no scope inversion effect for this pair: Mood always scopes over Modality in the verb forms investigated, see (810). On the assumptions laid out in the previous sections, this follows naturally: the ordering of projections is as before (Μ > Τ > Mod), so the emergence of the attested [word V-Mod-M-Agr] morphological string is evident, but the only category potentially available as a proxy is T, which however is not higher than M, so Mod cannot get to c-command Μ in any way (remember that Agr is not available: it is always filled with some content, it is never unspecified). Raising Mod to Τ is interpretationally vacuous, thus blocked. This way the only derivable scope order is the "default" Μ > Mod.

40

Huba Bartos

4.3. Conditional Mood and Tense Apart from our prime concern, the emergence of inverse scope, there is yet another issue here: Why and how is the dummy V-stem inserted to carry the M-affix, as seen above in (6) and (11)? Let us begin with this latter question. 4.3.1. The insertion of the dummy V The dummy V-root vol- is clearly an extra item, with no semantic contribution, so its occurrence must be constrained by economy, i.e. it can only appear when it must, to ensure convergence. It is the appropriate root form of the verb van 'be', this bound allomorph appears immediately before past tense or conditional mood marking. In the more optimal case, its insertion would be a purely morphological matter: it has to be inserted to repair an otherwise illformed string, at the point where (post-syntactic) vocabulary insertion takes place. However, there are reasons to attribute syntactic status to it, so its insertion must be seen as morphologically induced syntactic repair." The particular motivation for its insertion is the following. Recall that since Τ is lower in the projectional hierarchy than M, it gets inserted sooner. Then morphosyntactic merger associates Τ with V. In the next step, Μ is introduced into the structure, so a second application of morphosyntactic merger puts together Μ and [V+T], both syntactically, extending the chain, and morphologically, placing Μ into the same word-domain: {{V, T}, M}word. But this leads to ill-formedness at vocabulary insertion: both the past tense and the conditional mood affixes are analytic (in the sense of Kaye (1995)), as regards their morphophonological properties. And there is a templatic filter in Hungarian morphophonology prohibiting any sequence of adjacent analytic affixes (Rebrus 2000). This means that the past and the conditional suffixes cannot be adjacent. This is what necessitates repair, in the form of inserting the dummy V-root, to host the mood affix.14 But notice now that [M vol-na] is now in an appropriate position to host the next higher affix: Agr, via morphosyntactic merger, without

On-line Morphology: Hungarian Verbal Inflection

41

any further movement — contrary to fact, since the Agr-affixes appear on the [V-T] bloc: (19)

a.

b-

[Agr [V+T]+Agr] [MP [M Vdum+M]] var-t-unk vol-na wait-past-1 pi Vdum-cond 'We would have waited.' * [[Vdum+M]+Agr] [V+T]] * vol-nä-nk vär-t Vdum-cond-lpl wait-past

The derivation leading to (19b), by attaching Agr after M, though, should block the one leading to (19a), which is longer since it has to raise V over M, to pick up Agr. (Morphosyntactic merger will not work: it must be strictly local.) But there is an additional factor here: the Case assignment/checking of the subject DP, to be performed by the contentful verb. The cheaper derivation, with [M Vdum+M] forming a chain with Agr, inevitably leads to a crash: the dummy V lacks both theta- and Case-features, so it is not a proper Case-assigner/ checker for the subject. Agr therefore must form a chain with the bloc including T, which is usually assumed to carry the nominative Case-feature in minimalist models (e.g. Chomsky 1995). This is possible only by raising V to Agr, since Τ is in the morphosyntactic bloc (i.e., chain and word-domain) built around V, so it is not independent, not free to move on its own any more.15 The raising of V all the way up would violate the Head Movement Constraint, were it a valid restriction on movement. But given the general problems of adjunctive head-movement, alluded to in 3 above, it is quite conceivable that the usual effects of the HMC can be put down to the strictly local nature of morphosyntactic merger, as long as it can be taken to replace head-movement in a large number of cases, while genuine head-movement may fall under somewhat looser locality constraints (cf. also the cases of long head-movement, as e.g. in Rivero 1991, 1994). More specifically, if we follow Chomsky (1995, chapter 4) in building the notion of minimality into the definition of Attraction" (itself a supplant for the definition of Move), and assume that head-movement obeys the same kind of minimality as phrasal

42

Huba Bartos

movement, then we can say that Agr will attract the closest item that can check some of its features, and this will presumably be T, which, however, is part of the morphosyntactic V-bloc, and thus the raising of V itself will be induced. This way the Agr-affix will also be picked up by the [V+T] bloc. Observe that the raising of V will place it to the left of volna in the linearized structure, and for spellout it "carries with itself' its morphosyntactically merged affix (past), since they are one unbreakable word-domain. This is why we get the string in (19a), rather than *volna värtunk, or anything else.17 Choosing this more costly derivation over the cheaper but crashing one with Agr morphosyntactically merged to the [M Vdum+M] bloc is validated by a basic tenet of derivational economy: crashing derivations, however cheap they may be, may never block costlier but convergent derivations (Chomsky 1995) — unless they compete in minimality, which is not the case here (see fn. 16)." 4.3.2. Scope variance between Μ and T? By now it must be obvious that the account of scope inversion between Τ and Mod, and its absence between Μ and Mod will not help explain scope inversion between Μ and T. There is no proxy for Τ to raise to, so as to c-command M. And on closer scrutiny, as was already hinted at in section 2, this scope inversion turns out to be only apparent. Consider again the two readings of (4): (20)

Vdr-t-unk vol-na. a. Μ > Τ — 'We would have waited / Had we waited b. Τ > Μ — 'We wished / were willing to wait.'

Clearly, in (20b) Μ is not interpreted as conditional mood proper — instead, it represents one kind of modality·, that of volition or disposition. But if so, then (20b) is not a genuine case of scope inversion: it is not the case that Τ and Μ invert relative scope. What Τ has scope over is some modality, conceivably associated with the syntactic category Mod. Nevertheless, both the fact that this modality is encoded by a mood affix, and the ensuing mismatch between syntactic and morphological structure need an explanation. It seems

On-line Morphology: Hungarian Verbal Inflection

43

that the simplest way of accounting for these facts is to assume, on the one hand, that in (7) and (20b) there is a lexical item VOLITION, instantiating the category Mod. (Note that no other Mod is, or even could be, present in these structures.) However, this VOLITION exists merely as a syntactic-semantic feature bundle (a morpheme in the narrow lexicon, in the sense of Marantz (1997)), with no matching sound-shape in the vocabulary of Hungarian, which could be inserted in morphology. On the other hand, there seems to be a strong coocurrence restriction on this modal item: it may only occur in the scope of [+irrealis] mood (i.e. it is like an "irrealis polarity item"), which probably has to do with its property of being incompatible with realis mood: if someone has the volition/disposition of making an event happen, that event must belong to the realm of not (yet) realized desires, i.e. to a world not identical to the real one. So once again, the category of mood is called for, and hence its phonological exponent, the suffix -nA will become part of the structure. An alternative, "brute force" solution might be to list -nA in the vocabulary as an exponent for some inflectional element marked as [+V, +affix, +VOLITION], apart from being the regular exponent of conditional mood. But it would then be rather difficult to explain why this does not surface between the V-stem and the tense suffix (resulting in a complex form [V-M-Agr] + [Vdum-T], something like *vär-nä-nk vol-t). Thus this kind of account should not be seriously entertained.

4.4. Mood, Tense, and Modality all together As was shown in (11-13), when all these inflectional categories are simultaneously marked on a verb form, there is absolutely no scope variability: the affix order, as well as the only attested scope order, correspond directly to the hierarchy of the functional projections involved. And whereas in a semantic account something special should be said about the unexpected loss of scope inversion vis-ä-vis the cases in (2) and (5), on our morphosyntactic account the unique scope order simply falls out from the analyses of the other cases. There is just no room for scope variation. Since all three inflectional

44

Huba Bartos

categories are marked, no proxy is available. And the overtly visible presence of a (root) POSSIBILITY Mod excludes the occurrence of the VOLITION Mod, which might be linked to the conditional mood affix in the way suggested in the previous subsection. So it comes as no surprise that nothing is variable in this case. This elegantly concludes the analyses of the "scope variance vs. unique affix order" dichotomy: I have shown that given the proper semantic classification, and the right model of the syntaxmorphology interface, everything falls into its place, and even the Mirror Principle is maintained, as a unidirectional (syntax -» morphology), derivational generalization. The next section is devoted to showing how morphosyntactic merger can be put to use in accounting for certain verbal complexes in Hungarian.

5.

Extended words by morphosyntactic merger

Hungarian displays two basic orders (and their combinations) of verbs in stacked infinitival constructions, i.e. where several infinitival clauses are serially embedded one under the other (Kenesei 1989, Brody 1997, E. Kiss 1999, Koopman and Szabolcsi 2000). The straight order corresponds to the order of embedding from left to right: the main predicate is the leftmost item, and the most deeply embedded clause is the rightmost one, as in (21a). The inverted, socalled "roll-up" order is the exact opposite: the most deeply embedded verb begins the string, and the last item to be rolled up (sometimes the highest verb) is at the end: (21b). The name "roll up" is suggestive of how the lowest V raises to the next higher one, picking it up, then the complex of those two raises to the next higher V again, and so each of the verbs is picked up in this ever-growing "roll up". (21)

a.

Most fogok akarni kezdeni enekelni. now will-lsg want-inf begin-inf sing-inf 'Now will I want to begin to sing.' b. Most fogok [enekelni kezdeni akarni].

On-line Morphology: Hungarian Verbal Inflection

45

There have been two major types of approaches to the analysis of roll-up structures. The first one is in terms of XP-movement, as proposed by Koopman and Szabolcsi (2000): remnant VPs are moved successive cyclically to create the roll-up sequence." The second one assumes that roll-up structures result from some headtype relation: they are built by recursive V-incorporation, as in E. Kiss (1999), or are parts of a single extended word, as in Brody (1997). I will now propose that roll-ups can be created by morphosyntactic merger - thus my analysis falls in the second type, and is closest to Brody's model - , and that this is the cheapest solution, since almost every important property of the roll-up structure follows automatically from such an analysis, unlike in the other models. The key assumption underlying this analysis is that the rolled-up infinitives behave like suffixes on the most deeply embedded verb. As has been observed in the relevant literature, the lowest verb is the only fully thematic verb, while the higher ones are thematically weak or deficient to a certain extent. Although a clear thematic characterization is still not available, most authors agree that the rolled-up verbs are light in some sense. Let us take this to be reflected in the assumption that they can optionally be assigned a [+suffix] feature on entering the derivation. Morphosyntactic merger will therefore apply to them, as if they were "real" affixes: they will form a chain, and a single word-domain with the lowest verb, which acts as their "hosting stem". Morphosyntactic merger picks up these infinitives one by one, in a strictly local fashion, as follows from the definition of this operation, given in (15-16) above. This strict locality property immediately explains why inekelni kezdeni akarni in (21b) only means 'to want to begin to sing', and not 'to begin to want to sing', as observed by Brody (1997): the roll-up cannot skip any V, as would be needed to join 'sing' with 'want' across 'begin' in (21b). The process is schematized in (22):

46

(22)

Huba Bartos

a.

... akarni [ kezdeni [enekelni]]] [+suff] [+suff] morphosyntactic merger 1 morphosyntactic merger 2

b.

[enekelni-kezdeni-akarni]w Mod scope interpretation of the example) can be highlighted by adding a clause creating semantic anomaly in an alleged Mod > Τ reading, while leaving the valid Τ > Mod reading intact. Compare: (i)

Tegnap havazhatott Parizsban, es tudjuk, hogy yesterday snow-poss-past-3sg Paris-in and know-pres-lpl that havazott is. snow-past-3sg too Τ > Mod 'Yesterday it could be snowing in Paris, and we know that it did, in fact.' Mod > Τ #'It may have snowed in Paris yesterday, and we know that it did, in fact.'

50

Huba Bartos (ii) St. Germainben tegnap hovihar volt, tehät Parizsban is St. Germain-in yesterday blizzard was, so Paris-in also havazhatott, es tudjuk, hogy havazott is. snow-poss-past-3sg and know-pres-lpl that snow-past-3sg too 'There was a blizzard in St. Germain yesterday, so it could be snowing in Paris, too, and we know that it did, in fact.'

6.

7.

8.

9.

(i) is anomalous on the Mod > Τ reading: we do not normally speak of mere possibility when we have factual knowledge, in the present, of the past state of affairs in question. On the other hand, (ii) is not anomalous on the scope reading preferred for (14), i.e., Τ > Mod, exactly because we are not considering a possible past (Mod > T) here, but a possibility that was available at a past time (T > Mod). Note that the model was devised to handle various aspects of Hungarian morphosyntax, including nominal inflection (Bartos 1999), so it is not specific to the particular analyses presented here. Some requirement to this effect is necessary for all models capitalizing on ways of morphologically associating items other than movement, cf. the notions of structural adjacency in Distributed Morphology, and linear adjacency in Frampton & Gutmann's (1998) model. Frampton & Gutmann (1998, in. 5) allude to the possible application of some multi-tier structure, where e.g. adverbs are represented on a separate tier, so that they will not interfere with head-head relations on the main projection line of what are sometimes referred to as extended projections. Another possibility, not explored here in detail, would be the countercyclic insertion of adverbs into the clause structure, after the morphological associations discussed in this paper have been established. Unless it is an infix, but then again this property is lexically specified. There are no infixes in Hungarian, though, so the exact details need not be our concern here. A reviewer raises the question of how this model would account for the wellknown bracketing paradoxes discussed in the literature, especially those treated in Pesetsky (1985). The essential property of these cases is that the interpretationally and selectionally relevant word-structure is different from their appropriate morphophonological composition. E.g., unhappier should be [[un + happy] + er] as far as its semantics ('more unhappy', rather than 'not happier') is concerned, but should be [un + [happy + er]] morphophonologically, since the comparative suffix -er does not attach to words longer than two syllables. To overcome this problem, Pesetsky proposes that the morphophonologically relevant structure can be altered by a covert process he terms 'affix QR', which raises an affix to a higher position (in the particular case of unhappier, it raises -er to a position c-commanding the whole of [un + happy]), on condition that this transformation is string-vacuous. This

On-line Morphology: Hungarian Verbal Inflection

51

way, the semantically interpreted structure will diverge from the phonologically interpreted one, exactly as needed. The model advocated here achieves the same effects with ease, being rather similar in spirit to Pesetsky's proposal. Although I would not call this transformation 'QR\ and my analyses only concern the inflectional domain, my model could allow the raising of the relevant affixal categories (NB unlike in Pesetsky's framework, what move here are affixal feature-bundles, not actual affixes with sound-shapes) in syntax, but these rearrangements would not be replicated in morphology for exactly the same reason as in the Hungarian inflectional cases: morphology is strictly stepwise cyclic (see also Frampton & Gutmann (1999: 3)), so once unhas been added to happy+er, yielding {un, {happy, er}}word, the constituents of the inner domain, happy and er, become frozen in their places, and cannot be dislocated any more. I.e., the syntacic (LF-oriented) displacement of the lower affixal category will not be reflected in the phonological output. And although the string-vacuity constraint would preclude cases of linear affix reordering anyway, rebracketing of the same linear string is not possible either. 10. A note is in order here about how this hierarchy relates to Cinque's (1999) universal hierarchy of functional projections, a template which was assembled relying heavily on the Mirror Principle. There the order of these three particular categories is just the reverse: Mod^^, > Tpast > Mood^,^, while Mod„,ot is below each of them. But Hungarian affix order provides prima facie evidence that if the Mirror Principle is a valid generalization, then this partial hierarchy is not accurate, especially as far as the placement of irrealis Mood below past Tense is concerned. For details, see the analyses in section 4. 11. The head movement constraint is not an independent principle in Chomskyan minimalism, its effects have been built into the definition of Move/Attract instead (Chomsky 1995), so skipping Τ when raising Mod to Μ does not violate locality. 12. There is another set of data in the domain of combining Τ and Mod, which militates against a semantic account, but fits neatly into our analysis. Future tense is often expressed in Hungarian by an auxiliary: fog, which, unlike English will, is entirely void of any modal content. So e.g. a simple future Vform would look like (i): (i) fog

olvas-ni

FUT read-infinitive 'will read'

For some reason, it is impossible to combine this future form with modality represented by the modal affix -hAt, i.e. *fog-hat olvas-ni and *fog olvas-hat-ni are both ill-formed. The purported meaning can only be expressed periphrastically, by a biclausal structure: (iia), or by a simple present form plus a future time adverbial: (iib).

52

Huba Bartos (ii) a. Lehet, [ hogy fog olvasni]. be-poss that FUT read-inf 'It may be that he will read' b. Majd olvas-hat later read-poss-3sg 'Later he can/will be allowed to read.'

This is a fact, again, which is not likely to have a semantic explanation, but follows from the analyses presented here. Since fog carries subject agreement (e.g. fog-ok 'will-lsg'), it must originate below Agr. Its most likely status is that of T, since it obviously expresses future tense. Thus fog may be seen as the exponent of T, and what appears as the infinitival V-form after fog (as in fog olvas-ni, see above) is in fact a bare V-form (monomorphemic, but bimorphic, by fission occurring in morphology, splitting the single bare V into a V-root and a -ni marker — which is presumably related to the fact that what appears on the surface as a bare V-root, like olvas 'read' is in fact an inflected 3sg form, the marker of 3sg agreement being null, and this is also the dictionary form of Vs in Hungarian, so the infinitive is fissioned to be overtly marked as distinct from the 3sg, non-bare, form). Fog olvasni is clearly grammatical: the verb olvas 'read' picks up the infinitive marker, and fog occupies T. *Fog olvas-hat-ni is totally out if olvas-ni spells out a bare V, because this whole form is inserted under V° as a monolithic form, so -hAt in Mod cannot get in between the V-root and -ni. Finally, "'foghat olvas-ni is ruled out, too: if olvas-ni is a bare V spellout, then -hAt could only get onto fog if it raised over Τ (cf. the definition of morphosyntactic merger in (15)), to the proxy/empty Μ slot, but this time it cannot do so, because the proxy "M" only attracts fog in Τ as a filler — Mod is not equidistant with T, because now they are not in the same head-chain, since fog in Τ is a separate word, blocking chain-formation between Mod and M. This also yields the fact that Mod cannot scope over future T, either. 13. For a detailed discussion of the nature of vo/-insertion, see Bartos (2000). The essential argument for regarding it as syntactic, rather than morphological, is that it participates in / is visible for syntactic processes, such as right-node raising, and that it can be separated (though not moved away) from the real Vbloc by certain particles (e.g. the yes-no question particle -e, or is 'too'). 14. This further indicates that irrealis (conditional) Mood is higher than (past) Τ in Hungarian sentence structure (contra Cinque (1999)), or else the conditional affix could attach first to the real V-stem, and the dummy V-root would host the past affix. 15. The nominative assigning/checking Τ and the subject DP cannot meet in any lower domain, e.g. TP, either, on the assumption that the specifiers of the contentful inflections are retained exclusively for adverbials (Cinque (1999)).

On-line Morphology: Hungarian Verbal Inflection

53

16. Κ attracts F if F is the closest feature that can enter into a checking relation with a sublabel of Κ (Chomsky 1995: 297). Note that Chomsky takes minimality as inviolable this way (the very definition of movement makes reference to minimality), but since the [Vduinmy+M] bloc has no checking to offer for Agr, it is not a competitor for [V+T] in this respect, so these two alternative derivations are legitimate competitors on economy, and the crashing nature of one renders the other grammatical. 17. It is interesting to draw a parallel with Polish conditionals here: they are also composite forms, consisting of the thematic V-stem carrying some sort of participial marking plus another stem, the "conditional auxiliary", bearing subject agreement (Borsley and Rivero 1994), as shown in (i). (Irrelevantly, the conditional auxiliary can surface in any other position to the left of the main verb, too.) (i)

Ty jego widzia by. you him see-prt cond-2sg 'You would see him.'

At first sight, it seems to be the equivalent of the Hungarian (19b), modulo the order of the two verb forms: as if in Polish the conditional auxiliary could assume subject agreement by virtue of being capable of assigning/checking nominative to/with the subject, thereby making use of the shortest, most economical derivation. This may well be true: proper auxiliaries often assign/check nominative, so the difference between Polish and Hungarian boils down to the fact that in Hungarian volna is not an auxiliary, just a dummy verb. In fact, in earlier stages of Hungarian volna was probably a genuine auxiliary, taking a participial complement, and the participial verb form coincided with the past tense form, with the agreement marker analyzable as some sort of "possessive" agreement (characteristic of present-day Hungarian infinitives, too) on the participle: (iia). This is confirmed by the fact that in old Hungarian the auxiliary could also bear its own past tense, forming a complex past form of the scheme "V-participle-Agr + Aux-past": (iib). (ii) a. % vdr-t-unk vol-na wait-part-lpl beaux-cond 'We would have waited.' (lit.: '(There) would be our (having) waited.') b. % var-t-unk vol-t wait-part-lpl beaux-past 'We (had) waited.' (lit.: '(There) was our (having) waited.') But in present-day Hungarian there is no good reason to attribute auxiliary status to volna — it is a mere dummy, a minimal root inserted to carry the mood affix.

54

Huba Bartos

18. Alternatively, the fronting of V might be analysed as a case of phrasal movement, and then the issue of HMC and minimality would not even arise. We do not pursue this line here, though. 19. Here I do not argue against the XP-movement approach: for counterarguments with respect to the verbal complexes, see Brody (1997) and E. Kiss (1999), while Bartos (forthcoming) shows that Koopman and Szabolcsi's XPmovement-based model of affixation is unable to capture the scope phenomena in sections 2-4. 20.1 am grateful to Ildiko Toth for pointing me to these data.

References Aoun, Joseph and Yen-hui Audrey Li 1993 Syntax of Scope. Cambridge, MA: The MIT Press. Baker, Mark 1985 The Mirror Principle and morphosyntactic explanation. Linguistic Inquiry 16: 373-415. Bartos, Huba 1999 Morfoszintaxis es interpretacio: A magyar inflexiös jelens6gek szintaktikai härtere. [Morphosyntax and interpretation: The syntactic aspects of Hungarian inflectional phenomena.] Doctoral dissertation, ELTE, Budapest. Bartos, Huba 2000 Affix order in Hungarian, and the Mirror Principle. In: Gabor Alberti & Istvan Kenesei (eds.), Approaches to Hungarian vol.7. (JATEPress, Szeged), 54-70. Bartos, Huba forthc. Verbal complexes and morphosyntactic merger. Ms., Research Institute for Linguistics, Budapest. Borsley, Robert D. and Maria Luisa Rivero 1994 Clitic auxiliaries and incorporation in Polish. Natural Language and Linguistic Theory 12: 373-422. Brody, Michael 1997 Mirror theory. Ms., University College, London & Research Institute for Linguistics, Budapest. Chomsky, Noam 1995 The Minimalist Program. Cambridge, MA: The MIT Press. Chomsky, Noam 1998 Minimalist inquiries: The framework. In: Juan Uriagereka et al. (eds.), Step by Step: Essays in Honor of Howard Lasnik, Cambridge, MA: The MIT Press. Chomsky, Noam 1999 Derivation by Phase. Ms., MIT.

On-line Morphology: Hungarian Verbal Inflection

55

Chomsky, Noam 2001 Beyond explanatory adequacy. Ms., MIT. Cinque, Guglielmo 1999 Adverbs and Functional Heads: A Cross-Linguistic Perspective. Oxford University Press, Oxford. E. Kiss, Katalin 1999 Strategies of complex predicate formation and the Hungarian verbal complex. In: Istvan Kenesei (ed.), Crossing Boundaries, 91-114 Amsterdam & Philadelphia: John Benjamins. Fox, Danny 1995 Economy and scope. Natural Language Semantics 3: 283-341. Frampton, John and Sam Gutmann 1998 Distributed morphological interface and V-to-I raising. Ms., Northeastern University. Frampton, John and Sam Gutmann 1999 Cyclic computation, a computationally efficient minimalist syntax. Syntax 2: 1-27. Halle, Morris and Alec Marantz 1993 Distributed Morphology and the Pieces of Inflection. In: Samuel Jay Keyser & Ken Hale (eds.), The View from Building 20, 111-176, Cambridge, MA: The MIT Press. Kaye, Jonathan D. 1995 Derivations and interfaces. In: Jacques Durand and Francis Katamba (eds.), Frontiers of Phonology: Atoms, Structures, Derivations, 289332, London: Longman. Kayne, Richard S. 1998 Overt vs. covert movement. Syntax 1: 128-191 Kenesei, Istvan 1989 Logikus-e a magyar szorend? [Is Hungarian word order logical?] Altalänos Nyelveszeti Tanulmänyok 17: 105-152. Kiefer, Ferenc 1981 What is possible in Hungarian? Acta Linguistica Academiae Scientiarium Hungaricae, 31: 147-185. Koopman, Hilda and Anna Szabolcsi 2000 Verbal Complexes. Cambridge, MA: The MIT Press. Marantz, Alec 1997 No escape from syntax: Don't try morphological analysis in the privacy of your own lexicon. UPenn Working Papers in Linguistics 4.2: 201-226. Nash, Lea and Alain Rouveret 1998 Feature fission and the syntax of argument DPs and clitics. Paper presented at GLOW '98, Tilburg. Pesetsky, David 1985 Morphology and Logical Form. Linguistic Inquiry 16: 193-246.

56

Huba Bartos

Rebrus Pöter 2000 Morfofonolögiai jelensögek. [Morphophonological phenomena]. In: Ferenc Kiefer (ed.), Strukturälis magyar nyelvtan 3. — Morfolögia [A structural grammar of Hungarian vol.3. — Morphology], 763947, Budapest: Akadömiai Kiadö. Reinhart, Tanya 1997 Wh-in-situ in the framework of the Minimalist Program. Natural Language Semantics 6: 29—56. Rivero, Maria Luisa 1991 Long Head Movement and Negation: Serbo-Croatian vs. Slovak and Czech. The Linguistic Review 8: 319-351. Rivero, Maria Luisa 1994 Clause structure and V-movement in the Languages of the Balkans. Natural Language and Linguistic Theory 12: 63-120.

Verbal morphology and agreement in Urdu Miriam Butt and Louisa Sadler

1.

Introduction

The status of morphology within the theory of grammar, and in particular its relation to syntax, remains somewhat controversial.1 A number of important issues can be distinguished, concerning the interaction between morphology and syntax and the mechanisms and data structures appropriate for the description of morphological systems. On the relation between syntax and morphology, one major line of thinking, first proposed by Chomsky (1970), is often referred to as 'lexicalist' and holds that syntactic and morphological processes belong to differing modules of grammar. In the tradition of derivational syntax (e.g., DiSciullo and Williams 1987), these modules only interact at the point at which lexical insertion of a morphologically fully formed word into the syntax takes place. A radically different view blurs the distinction between syntax and morphology in that inflectional morphology is taken to result essentially from head-to-head movement, a syntactic operation. As such, inflectional morphology forms a subtheory of syntax and morphological processes are predicted to be subject to the same constraints as syntactic processes (e.g., Baker 1985, 1988). The most recent formulation of this position is Distributive Morphology (Halle and Marantz 1993). The issue of mechanisms and data structures for morphological description is hardly less controversial. In broad terms, we may distinguish inferential-realizational approaches, including word-and-paradigm models, in which forms are viewed as exponents of sets of features or paradigmatic cells (e.g., Anderson 1992; Aronoff 1994; Stump 1991, 2001) from 'lexical' or morpheme-based models in which the mapping between meaning and form is given in the lexical entry for a morpheme. On this view, an inflectional affix is in many ways granted a status similar to that of a lexeme, and it is natural to regard word

58

Miriam Butt and Louisa Sadler

structures as very similar to syntactic structures (as, for example, in Selkirk's 1982 word syntax approach). This paper explores aspects of the representation of morphology in Lexical-Functional Grammar (LFG) by examining the treatment of verbal morphology and agreement phenomena in the South Asian language Urdu.2 The theory of Lexical-Functional Grammar (Bresnan 1982, 2001; Dalrymple 2001) is lexicalist in the former sense, that of the lexicalist hypothesis. The specific form in which the lexicalist hypothesis is embodied in LFG does, however, permit quite a complex interaction between syntax and morphology, as we shall see. The precise nature of the morphological component itself has received relatively little attention in LFG. While most theoretical work in LFG has assumed a morpheme-based approach, often for expository convenience, there is no reason in principle why LFG should espouse a lexical morphology of this sort, and indeed several papers propose the integration of a realizational or constructional morphology with LFG. Computational work in LFG has generally used finite-state morphological analyzers (Karttunen, Kaplan and Zaenen 1992, Kaplan and Kay 1994), with an interface defined between the morphological and syntactic analysis (e.g., Butt, King, Nino and Segond 1999). Finite state technology is not committed to a morphemic or sign-based view of morphological structure, and preserves the required separation of (external) syntax and morphology. From the point of view of the syntactic module, the morphological analyzer is a 'black box'. The processes by which words are formed are entirely opaque to the syntax and the only point of contact is the syntactic functional information which filters through the interface between morphology and syntax. In this paper, we explore some complex morphosyntactic phenomena in Urdu, showing how LFG permits a natural treatment of the interaction between syntax and morphology. We then explore some deficiencies of the morpheme-based view on the basis of a computational implementation of the grammar fragment under discussion and go on to examine an alternative model of the morphology-syntax interface. The rest of this paper is structured as follows. In section 2 we first provide a brief sketch of the basic design principles of LFG for those

Verbal Morphology and Agreement in Urdu

59

readers who are unfamiliar with the theory. Section 2.2 introduces and exemplifies an aspect of the formalism (known as constructive morphology, Nordlinger 1998) which permits a natural and straightforward approach to the ability of morphological elements (such as case markers) to define and project the relational structures which contain them. Section 3 briefly introduces case in Urdu and its relation to verbal agreement patterns and sketches out a treatment of Urdu case marking in LFG. In section 4 we move on to verbal agreement in Urdu. We formulate a relatively simple generalization concerning verbal agreement and show how constraints associated with verb forms will capture this generalization. In the next two sections we turn to the details of the morphological analysis, exploring first a word-syntax, or morphemebased, implementation in section 5 and presenting several unwanted side-effects and drawbacks of this approach. The final section examines an encoding of the same set of agreement data using a finite-state morphological analyzer interfaced to the syntax and shows how the difficulties encountered in the word syntax approach are resolved.

2.

Lexical Functional Grammar

LFG is a non-derivational syntactic theory which posits two levels of syntactic representation, each with its own distinct vocabulary and data structures. External syntax, or constituent structure (c-structure), is modelled in terms of (phrase structure) trees of the familiar sort, based on categories decomposable as simple feature bundles. These structures are directly motivated by considerations of surface variability: that is, where the forms of exponence differ, so do the external (cstructure) forms. Thus while a fully elaborated X' theory with fully projective lexical and functional categories is appropriate for the description of highly configurational, endocentric constructions in languages such as English, alternative, non-configurational modes of organization for external syntactic structure are appropriate in other languages, and still others may exhibit a mix of configurational and nonconfigurational external structure.

60

Miriam Butt and Louisa Sadler

LFG uses a second syntactic level of representation for internal, relational structure. This level of representation distinguishes predicates and their syntactic arguments, with relations such as SUBJECT, OBJECT, ADJUNCT and OBLIQUE being primitive notions. Internal syntax is represented at the level of f(unctional)-structure. Functional structures are represented as AVMS (attribute-value matrices) and are finite functions from attributes to values, which may themselves be functions. While languages differ quite radically in their external (surface) structure, f-structures are intended to be 'largely invariant' across languages (Butt, King, Nino and Segond 1999; Bresnan 2001). C-structure and f-structure represent different dimensions of syntactic structure and are related by a mapping function φ, from nodes in the c-structure to f-structures. Thus for example, the annotations to the nodes in (1) relate these nodes to their image in the domain of fstructures: t abbreviates the expression φ(Μ(*)) and may be read as "the f-structure of my mother node" and I abbreviates the expression φ(*) and may be read as "the f-structure of the present node". Thus the annotation on the DP node in (1), (fsUBJ) = I, states that the value of the SUBJ attribute in the f-structure of the IP node is the f-structure associated with the DP node itself, and the annotation t = I states an identity between the f-structures of the mother and daughter nodes in question.

(1)

IP (tSUBJ) = 4, DP

t=l I'

he

t=; VP is

t=; V leaving

Verbal Morphology and Agreement in Urdu

61

(2) " PRED SUBJ

'pro'

1

PERS

THIRD

NUM

SING

GEN

MASC J

PRED

'leave

TENSE

PRES

ASP

PROG

< SUBJ > '

Further equations specifying f-structure information are associated with lexical items in the lexicon. Finally it should be noted that the set of annotations appropriate for tree fragements are largely defined by general principles, but discussion of these principles would take us too far afield (see Bresnan 2001 for extensive discussion). 2.1. LFG morphosyntax The strict separation of external syntactic structure and word-internal structure is encapsulated in LFG in the principle of Lexical Integrity, a recent statement of which is that: "morphologically complete words are leaves of the c-structure tree and each leaf corresponds to one and only on c-structure node" (Bresnan 2001). This ensures that there is no syntacticization of morphological phenomena—in particular, affixes do not enter into syntactic structural relations with elements of the external syntax. Note that the principle of lexical integrity is stated in such a way as to preserve the invisibility of morphological structure to c-structure while permitting what we might think of as the internal f-structure of words to be visible to f-structure. In fact, as is clear from morphologically rich languages, (external) syntax and morphology are equal, interacting and competing contributors in the functional domain. The direct contribution of morphology to the functional domain can be easily seen in the analysis of phenomena such as head marking and pronominal incorporation in LFG. For example, Bresnan and

62

Miriam Butt and Louisa Sadler

Mchombo (1987) show that both subject and object markers in Chichewa are word internal elements, using standard tests for lack of separability and the occurrence of allomorphic variation. They show that the optional object marker always fulfills the argument function, optionally doubled by a full noun phrase topic. On the other hand the obligatory subject marker is sometimes an agreement marker (doubled by an overtly expressed subject noun phrase) and sometimes fulfills the argument function. These properties are expressed by associating the appropriate functional information with the subject and object inflections of the verb (denoted here by SM- and OM-). Adopting a morpheme-based morphology, Bresnan and Mchombo provide the following entries for the affixes, as input to the morphological word building rules: notice that the equation which associates a pronominal PRED value with the subject marker is optional, accounting for the observed alternation between subject coding morphology as agreement and as incorporated pronominal. The verbal affixes thus (partially) define the f-structures of the subject and object—the contribution to f-structure by the verb zi-nä-wä-lüm-a in (7) is shown in (9) below: (3) OM-:V i n / / :(tOBJ)=| (4-agr) = a (4.PRED) = 'pro' (4) SM-:V i M / z :(tsUBJ)=; (4,agr) = a ((|pred) = 'pro') (5) Njuchi zi-nd-lum-a

a-lenje

lO.bee 10.S-PST-bite-FV 2-hunter 'The bees bit the hunters.' (Chichewa, Bresnan 2001: 150) (6) zi-nä-lüm-a

a-lenje

10-S-PST-bite-FV 2-hunter 'They bit the hunters.' (Chichewa, Bresnan 2001: 150)

Verbal Morphology and Agreement in Urdu (7) Njüchi zi-nä-wä-lüm-a

63

a-lenje

lO.bee 10.S-PST-2.O-bite-FV 2-hunter 'The bees bit them, the hunters.' (Chichewa, Bresnan 2001: 150) ( 8 ) Njüchi

zi-nä-wä-lu.m-a

lO.bee 10.s-PST-2.o-bite-FV 'The bees bit them.' (Chichewa, Bresnan 2001: 150) (9) OBJ

" PRED

'PRO'

PERS

THIRD

NUM

PLUR

1 J

PRED

'bite < (SUBJ), (OBJ) > '

TENSE

PAST

SUBJ

' PERS NUM

THIRD 1 PLUR

J

Although the formal framework of LFG naturally accommodates the direct contribution of morphology to the definition of f-structres, a number of open issues remain. F-structure is defined quite independently of exponence, but many analyses involve the appearance of exponence related features at f-structure. This may be problematic, for when f-structure is too closely dependent on details of exponence, the invariant part may be 'submerged' in a wealth of language-dependent morphosyntactic features. In some cases the presence of morphosyntactic features (like VFORM in periphrastic constructions, e.g., Böqars, Vincent and Chapman 1997) in f-structures will cause (otherwise correct) analyses to fail (Frank 1996), and there is evidence that both morphological and syntactic features sets are required in some domains (Sadler and Spencer 2001; Sells 2001). As more work is done on a range of languages, and especially on those with richer morphology, it is becoming clear that the interface between syntax and morphology is often more complex than a simple form of the lexical integrity

64

Miriam Butt and Louisa Sadler

hypothesis might at first suggest. A number of architectural developments and/or extensions to the formalism have been suggested partly in response to these issues, including the postulation of a separate projection for morphosyntactic features (Butt et al. 1996; Frank and Zaenen 2000) and the positive restriction architecture (Andrews and Manning 1999), but space precludes detailed discussion of these proposals.

2.2.

Constructive morphology

As we have seen, LFG can easily capture the direct contribution of head marking morphology in projecting the f-structure for a verbal head and its syntactic arguments. In recent work, Nordlinger ( 1 9 9 8 ) has explored the contribution of dependent marking morphology to the projection of f-structure. She shows how the formalism can directly capture the role of case markers in identifying the grammatical functions borne by the noun phrases which they mark, in languages in which such grammatical functions are not configurationally defined. In the approach known as constructive case, she associates inside-out constraints (Halvorsen and Kaplan 1988; Dalrymple 1993, see also Andrews 1 9 9 6 : 4 1 - 4 3 ) with morphological elements, allowing case morphology on the noun or noun phrase to define the larger syntactic (f-structure) context in which they are embedded. Consider for example, the sentence in (10) from the radically non-configurational language Wambaya. Here the subject big dog is a discontinous constituent in which both parts are marked with ergative case.

(10) galalarrinyi-ni gini-ng-a

dawu bugayini-ni dog.I-ERG 3 S G . M A S C . A - 1 . 0 - N F U T bite big.I-ERG "The big dog bit me.' (Wambaya, Nordlinger 1998: 96)

Nordlinger assigns the following morphological structure to galalarrinyi-ni (the functional specifications are associated with the stem and affix in (sub-)lexical entries):

Verbal Morphology and Agreement in Urdu

(11)

65

Ν

t= I

t= I

Ν

Aff

galalarrinyi (tPRED) = dog

ni (tCASE) = ERG

((SUBJ t) OBJ) As shown above, it is the ergative case which specifies that the nominal element is a SUBJ in a larger f-structure. (12) specifies (a) that the attribute:value pair CASE:ERG is defined for the f-structure of the case marked nominal (that is, the f-structure denoted by and (b) that this f-structure (that of the case-marked nominal) is the value of the SUBJ attribute in a containing f-structure (i.e., the f-structure of the sentence—denoted by (SUBJ t)X which also contains an OBJ: (12) ni:

ERG

(FCASE) = ERG ((SUBJ T) OBJ)

Together the affix and the stem galalarrinyi, 'dog' then define the following f-structure: (13) SUBJ

PRED

'dog'

CASE

ERG

The treatment of the (discontinuous) adjunct is more complicated. The stem bugayini 'big', specifies that it is the value of an ADJUNCT attribute in a containing f-structure (see annotations in (14)), while the case affix specified that the f-structure containing the ADJ is an ergative case SUBJ. The relevant morphological tree structure is as follows:

66

Miriam Butt and Louisa Sadler

(14)

bugayini (tPRED) = big (ADJ t)

((ADJ t) CASE) = ERG ((SUBJ ADJ t) OBJ)

The f-description associated with the ergative case affix in (14) differs from that in (11) in that it specifies (adJ t) where the latter specified t · This local substitution of designators is due to the Principle of Morphological Composition which (roughly) embeds the functional designator of the stem under that of the affix. As we are not concerned with this composition here, we have nothing more to say about it.3 The f-structure of bugayini-ni is as in (15). The f-structures (13) and (15) combine gracefully as (16) to give the f-structure of the discontinuous subject (f 1) within the f-structure of the sentence (JO). (15) SUBJ

Γ CASE ADJUNCT

ERG r [ PRED

lι . « ι big J

1

(16)

/o

SUBJ

fx

CASE

ERG

PRED

'dog'

ADJUNCT

[ PRED

'big' ]

Nordlinger (1998) shows how inside out statements associated with case morphology can successfully describe a range of phenomena including case stacking, case marking on discontinuous nominal constituents (as shown here) and so-called modal case, introducing sentential tense and aspect features. Although Nordlinger (1998) uses a

Verbal Morphology and Agreement in Urdu

67

morpheme-based morphology in her formulation of constructive case, she explicitly states that the use of inside-out statements is independent of a morphological model: "while I use a morpheme-based morphology here, the overall analysis does not depend on this view of morphology and could equally be adapted into a rule-driven analysis" (Nordlinger 1998: 63).

3. Exploring the syntax-morphology interface: Urdu case In the following sections, we attempt to give the preceding material a more concrete form by dicussing some aspects of two different but interrelated phenomena in Urdu morphosyntax: case and agreement. 3.1.

Case, agreement and pro-drop

South Asian languages have in common a number of areal characteristics which are also found in Urdu. These include the widespread use of non-nominative subjects (see section 3.3 for some Urdu examples) and the existence of relatively free word order among nominal arguments of a clause, together with the ability to pro-drop quite freely. On the other hand, the agreement patterns of these languages differ significantly, irrespective of these shared areal characteristics, something which is rather unexpected on the view (which we do not espouse) which posits a very direct structural relationship between patterns of agreement and case marking. The work of Mahajan (1990, 1992) on Hindi is fairly representative of approaches which see structural Case assignment as involving functional agreement heads. The pretheoretical generalization with respect to Urdu verbal agreement is as follows (cf. Mohanan 1994). The verb agrees only with an unmarked (i.e. nominative) direct argument (subject or object). Verbal agreement in Urdu involves person, number and gender. If the subject is unmarked, agreement is with the subject (see (17a)). If the subject is marked, but the object is unmarked, the verb agrees with the object (see (17b)). If both subject and object are unavailable for agreement

68

Miriam Butt and Louisa Sadler

(that is, are overtly case-marked), the verbal complex reverts to third person, masculine singular morphology ((17c)). In summary, one can say that overt case repels verbal agreement. (17) a. nadya gari cala-t-i h-εϊ Nadya.F.Nom car.F.Nom drive-Impf-F.Sg be-Pres.3.Sg 'Nadya drives a car.' b. nadya=ne/adnan=ne gari cala-yi h-εί N.F=Erg/A.M=Erg car.F.Nom drive-i'erf.F.Sg be-Pres.3.Sg 'Nadya/Adnan has driven a car.' c. nadya=ne gari=ko cala-ya h-εί Nadya.F=Erg car.F=Acc drive-Perf.M.Sg be-Pres.3.Sg 'Nadya has driven the car.' Unlike Urdu, some other South Asian languages do exhibit (verbal) agreement with overtly case marked nominale. Nepali and Gujarati have free word order among nominale, the ability to pro-drop and also allow ergative and dative (psych predicates) subjects, and are thus similar to Urdu in these respects. However, Nepali differs from Urdu in that the verb agrees with the ergative subject, shown in (18). (18) mai-le mero lugä dho-en I-Erg my clothes.Nomwash-Past.l.Sg Ί washed my clothes.' (from Deo and Sharma 2002) In Gujarati, on the other hand, as shown in (19), the overt presence of an accusative has no effect on verb agreement with the object (Butt and Deo 2001). Again, this is unlike Urdu/Hindi, where overt case blocks agreement. (19) a. mai-le mero lugga dho-en I-Erg I.M.Sg.Gen clothes.M.Pl.Nom wash-Perf.3.Pl Ί washed my clothes.' b. ram-e gadi-ne jo-yi Ram.M.Sg-Erg car.F.Sg-Acc see-Perf.F.Sg 'Ram has seen a/the car.'

Verbal Morphology and Agreement in Urdu

69

In a recent typological study of agreement in South Asian languages Subbarao (1999) finds what amounts to a singular lack of correlation between the distribution of case and agreement. That is, in some languages the presence of a lexical case marker blocks agreement (e.g., Urdu/Hindi), while in others the presence of a lexical case marker is obligatory for agreement (e.g., Maithili), and in still others the presence or absence of a case marker is of no consequence for agreement (e.g., Mizo, Hmar, Paite). Subbarao postulates null agreement checking when there are no strong agreement features. In our view, this essentially divests the postulated connection between case and agreement of any empirical consequences, and therefore amounts to giving up on the strong connection between case and agreement postulated in much derivational generative syntax (cf. Bhatt 2002 who comes to essentially the same conclusion). Another issue that arises in connection to the relation between case and agreement is pro-drop; Urdu, like all South Asian languages, allows rampant pro-drop. An example is shown in (20). (20) a. tum=ne nadya=ko Ifana di-yal you=Erg Nadya.F=Dat food.M.Sg.Nom give-Perf.M.Sg 'Did you give Nadya (some) food?' b. ji, di-ya yes.Polite give-Perf.M.Sg 'Yes, gave.' One standard view is that pro-drop is correlated with rich verb agreement (e.g., Rizzi 1986). As we have seen, Urdu does have rich verb agreement and also permits pro-drop. However, as the monologue in (21) from the Hindi movie Dilwale Dulhania Le Jayenge shows, both agreement and case are orthogonal to the possibility of pro-drop. Note in particular that since Urdu does not have indirect object agreement, the permitted absence of the indirect object in (20) cannot be explained in terms of licensing by agreement. The first sentence in (21a) begins the monologue by referring to some pigeons who are seen pecking at seeds outside. There is no pro-

70

Miriam Butt and Louisa Sadler

drop here. The topic of the utterance is nominative 'they'(=some pigeons). In (21b) and (21c) the 'they'(=pigeons) has been pro-dropped. In (21b) the overt realization of 'they'(=pigeons) would be ergative. This means that the verb is not showing agreemeent with the prodropped element. Rather, as indicated, it agrees with the nominative (unmarked) object. In (21c), on the other hand, the overt realization of 'they'(=pigeons) would be nominative. Consequently there is agreement with the pro-dropped element here.4 (21) a. \ye\T bbi mer-i=ki tdräh-εϊ Pron.3.Sg also I.Gen-F.Sg=Gen.F.Sg like be-Pres.Pl 'They fopic are also like me.' (Dilwale Dulhania Le Jayenge) b. jahä dana delf-a where seed.M.Sg.Nom see-Perf.M.Sg 'where (theycon, .topic) s e e a s e e d ' c. udar ga-ye or pet bar kar there go-Perf.M.Pl and stomach.M.Sg.Nom fill having ur ga-ye rise go-Perf.M.Pl 'there (they contJopic ) go and having filled (their) stomach cont.topic^

fl

y

awa

y·'

We thus conclude that case and agreement appear to function relatively independently in South Asian languages. In the following subsection we explore the nature of the syntax-morphology interface by taking a closer look at case and agreement patterns. 3.2.

Case in Urdu

Case is marked in Urdu by a set of case clitics which derive diachronically from nouns or participles in an earlier stage of the language (see Butt and King 2002a). There are two cases which are not marked overtly by means of case clitics. One of these is the so-called bare

Verbal Morphology and Agreement in Urdu

71

locative, which is nevertheless often identifiable by the occurence of the head noun in an oblique form. The other one is the 'direct' case found on subjects and direct objects. In this paper, we consistently gloss this case as the nominative. See Butt and King (2002a, 2002b) for an extensive discussion. The distribution of case markers in Urdu is relatively complex. Urdu is generally described as a split ergative language, with clausal tense and aspect features (partly) determining the case marking pattern. The ergative must appear on the subjects of transitives when the verb is marked with perfect morphology (-a/-i/-e) as in (22). Subjects of unaccusative intransitives are always nominative, whereas subjects of unergative intransitives are optionally ergative (Davison 1999, see section 3.3). Furthermore, the ergative alternates with a dative subject in a be+infinitive construction (see section 3.3). (22) a. ram gari cala-t-a (h-εί) Ram.M.Sg.Nom car.F.Sg.Nom drive-Impf-M.Sg be-Pres.3.Sg 'Ram drives a car.' b. ram=ne gari cala-yi (h-εί) Ram.M.Sg=Erg car.F.Sg.Nom drive-Perf.F.Sg be-Pres.3.Sg 'Ram has driven a/the car.' Note that (22) also illustrates the relation between verbal agreement and the lack of case marking discussed above.

3.3.

Case Alternations

The patterns of case alternations in the language provide evidence that Urdu cases are not mere morphological reflexes of structural position or agreement relations, but are 'constructive' in the sense of Nordlinger (1998), serving to define the f-structure grammatical functions born by the nominals they mark, and directly contributing to, and interact with, clause level functional and semantic information.

72

Miriam Butt and Louisa Sadler

Consider, for example, the clitic ko. This clitic expresses both dative and accusative case in Urdu. As a dative it is associated with goals (indirect objects, subject of pysch-verbs, etc.). As an accusative it marks specificity of objects, much as in Turkish (Εης 1991), and alternates with a nominative or unmarked object, as shown in (23). (23) a. ram=ne jiraf defd-i Ram=Erg giraffe.F.Nom see-Perf.F.Sg 'Ram saw a/some giraffe.' b. ram=ne jiraf=ko delf-a Ram=Erg giraffe.F=Acc see-Perf.M.Sg 'Ram saw the (particular) giraffe.' Because the specificity of an object has an effect on the clausal semantic interpretation in terms of further properties such as telicity or aspectual interpretation (e.g., Krifka 1992; Tenny 1994), the case clitic must be seen as interacting with the clausal syntax and semantics. Mahajan (1992) attempts to correlate specificity with object agreement, but his analysis makes exactly the wrong prediction with regard to the interpretation of specificity. Under his analysis, the unmarked object moves to an agreement position to receive Case. In this position, the unmarked (nominative in our terms) object is also associated with specificity features. However, as seen above, it is precisely the unmarked agreeing object which allows for non-specific readings. The ko (accusative) marked non-agreeing object, on the other hand, must be interpreted as specific (see Butt 1993). Another example of a productive case alternation is found with intransitive unergative clauses. As shown in the minimal pair in (24), the ergative alternates with the unmarked nominative to produce a meaning difference in terms of intentionality/volitionality. Again, the clear meaning difference is marked solely by the case alternation on the (subject) with this class of verbs. (24) a. ram l&äs-a Ram.M.Nom cough-Perf.M.Sg 'Ram coughed.'

Verbal Morphology and Agreement in Urdu

73

b. ram=ne l&äs-a Ram.M=Erg cough-Perf.M.Sg 'Ram coughed (purposefully).' Similarly, with the be+infinitive in (25) the choice of ergative or dative case determines sentential features. While the ergative introduces a desiderative modality, the dative functions more as a default marker which could be interpreted either in terms of obligational force or with the desiderative modality, depending on the particular context of the utterance. Note that the verbal complex itself is once again invariant across this pair of sentences. (25) a. nadya=ne zu ja-n-a h-εϊ Nadya.F=Erg zoo.M.Loc go-Inf-M.Sg be-Pres.3.Sg 'Nadya wants to go to the zoo.' b. nadya-ko zu ja-n-a h-ei Nadya.F=Dat zoo.M.Loc go-Inf-M.Sg be-Pres.3.Sg 'Nadya has/wants to go to the zoo.' 3.4.

Constructive case in Urdu

In the above sentences, the semantic contrasts are directly related to the choice of nominal case. This is consonant with the constructive view of case (and other 'functional' features) taken in Nordlinger (1998), but is at odds with the standard view in derivational approaches, in which case is seen as a mere spell-out of functional features. Furthermore, as word order is largely free in Urdu (e.g., Mahajan 1990; Butt 1995; Kidwai 1997), and neither case nor grammatical functions can be associated with any particular structural position (Butt 1995), it is evident that the case markers themselves play an important role in determining the grammatical functions of the noun phrases they are attached to. This entails an essentially constructive view of case clitics and can be stated in a simple and intuitive fashion by means of the lexical entry for the dative/accusative ko shown in (26).

74

Miriam Butt and Louisa Sadler

(26) ko:Κ [ (tCASE) = ACC

(OBJt) (fSPECIFIC) = + ]

V [ (tCASE) = DAT

( ° B W t ) V (SUBJt) ] The entry for ko allows for 3 possibilities. As an accusative marker, it 'constructs' an object: the statement (OBJ f ) has existential force and specifies that the f-structure of the accusative case marked nominal (t) is the value of the OBJ attribute in an immediately containing f-structure. The first disjunct also states that the object will have accusative case and that it is to be interpreted as specific. The second disjunct covers the dative uses of ko. Datives can be either indirect objects (OBJg 0 ) or subjects, as in (25b) above. Again, the inside-out statements (OBJ go f ) and (SUBJ ΐ ) specify that the f-structure of the case marked nominal is the OBJ go or the SUBJ in the f-structure corresponding to the clause. This entry captures the distribution and semantic effect of the accusative/dative ko efficiently and accurately without further recourse to syntactic rules. We stated above that Urdu case markers are syntactic clitics. As such, the natural treatment of them in LFG is as co-heads of the NP they mark, possibly as members of a (functional) category K, as shown in (28) (see Butt and King 2002a for a discussion of the use of KP with respect to Urdu). The nominative is phonological null in Urdu and therefore cannot receive an entry on a par with the other (phonologically substantial) case markers.5 We assume that in the absence of overt case particles, nominative case (i.e., (fCASE) = NOM) and the associated constructive identification of the grammatical function is assigned via default rules. For a detailed analysis of this and the case alternations presented above see Butt and King (2002a, 2002b). (27) larke=ko boy.M.Obl=Dat

Verbal Morphology and Agreement in Urdu

75

KP

(28) t=4 NP

t=l Κ

t=; Ν

ko

larke 3.5.

Case clitics vs. case inflection

A fundamental notion in LFG is that while morphology and syntax may make the same type of functional contribution to the analysis of a sentence (in different languages or, sometimes within the same language, e.g., synthetic and analytic expression of temporal and aspectual clausal features in Latin, Sanskrit, Italian, French, Welsh and many other languages), this does not imply that syntax and morphology share an expression structure. As we have noted above, the principle of strict lexicalism itself requires that morphological structures are opaque to the syntax. If syntax and morphology are, in this sense, equal partners, it is expected that the two means of expression may co-occur and even compete within the same language. A particular instance of syntax and morphology as alternative means of expression occurs in Urdu where pronominals may bear inflectional case marking, while lexical noun phrases, as we have seen, are generally marked by case clitics, which are syntactic elements. The inflectional case marking on pronominals is a vestige of the earlier Sanskrit case system. In some cases, pronouns permit either a case clitic or an inflectional case marker, as shown in (29b). (29) a. nadya=ko dakxane ja-n-a Nadya.F=Dat/I.Obl-Dat post offic.M.Obl.Loc go-Inf-M.Sg h-ei be-Pres.3.Sg 'Nadya has/wants to go to the zoo.'

76

Miriam Butt and Louisa Sadler b. muf=ko/muf-e

dakxane

ja-n-a

I.Obl=Dat/I.Obl-Dat post offic.M.Obl.Loc go-Inf-M.Sg h-ei

be-Pres.3.Sg Ί have/want to go to the zoo.' While the dative noun in (29a) can only be marked with a case clitic, the pronoun in (29b) is more permissive. The inflectional affix -e contributes exactly the same information to the syntactic and semantic analysis as the ko case clitic in (26). Again, see Butt and King (2002a) for a more in-depth discussion of the pronominal case marking paradigm in Urdu.6

3.6.

Summary

The brief discussion of case marking in this section has illustrated how the role of case can be captured simply and straightforwardly in LFG. Case markers are treated as syntactic elements introducing a set of constraints over f-structure (and contributing semantic information). Case-inflected pronominals are associated with an identical set of (fstructure) constraints. Urdu case is 'constructive' in the sense that the cases themselves project the grammatical functions. In Urdu, the relation between case marking and grammatical function is complex and there is no simple structural relation between grammatical relation and position. Furthermore, case alternations are almost exclusively semantically motivated. All these aspects of the role of case in Urdu can be captured straightforwardly in LFG. In the following section, we turn to verbal agreement. The verbal agreement pattern in Urdu is wholy inflectional and involves person number and gender agreement distributed over parts of periphrastic expressions. We do not deal with the details of the formation of periphrasis here. While both case clitics and case inflections can be analysed straightforwardly in LFG, the analysis of verbal agreement uncovers some challenges.

Verbal Morphology and Agreement in Urdu

4.

77

Agreement

4.1.

The verbal tense/aspect system in Urdu

The modern South Asian languages display various differing agreement systems, most of which involve number, person and gender (but not case) agreement in some distribution over auxiliaries and former participles. These different systems are descended from the Sanskrit, which had person and number agreement in the verbal domain and number, gender and case agreement on adjectives and participles. In this section, we confine our discussion to verbal agreement. With respect to verbal agreement, South Asian languages differ considerably. The modern Indo-Aryan language Bengali, for example, has dispensed entirely with gender and number agreement in the verb, while other languages like Marathi have re-introduced person agreement. Despite these differences, the Urdu/Hindi verbal paradigm is fairly representative of South Asian (Indo-Aryan) languages. Urdu has a mixed inflectional and periphrastic tense/aspect system. The indicative pattern summarized in (30) is illustrative for the verb mar- 'hit' (the equivalent of the English simple present is rendered periphrastically in Urdu as shown in several of the examples below).

(30)

The Basic Modern Urdu Tense/Aspect System Impf Perf Prog Pres Past Fut Pres/Past Pres/Past Pres/Past mara marega marta mara mar raha + Aux (be) + Aux (be) + Aux (be) mar- 'hit' — 3.Sg.M

Amongst these forms, the future is the only form which inflects for number and person.

78

Miriam Butt and Louisa Sadler

(31) Singular M/F 1st mar-ü-g-a/i 2nd mar-e-g-a/i 3rd mar-e-g-a/i mar-' hit'

Urdu Future Paradigm Plural Respect (ap) Familiar (tum) M/F M/F M/F mar-e-g-e/i mar-e-g-e/i mar-o-g-e/i mar-e-g-e/i

The vowel encoding number and person agreement (immediately following the stem) might be a vestige of the auxiliary h-εi 'be' (McGregor 1968:161), or may derive historically from the original present inflections (Ashwini Deo, p.c., August 2000). Compare the paradigm for the present tense of the modern Urdu verb ho- 'be'. (32) Singular 1st h-ü 2nd h-εί 3rd h-ei ho- 'be'

Present of Urdu be Plural Respect (ap) Familiar (tum) het he! h-o he!

This verb is the only verb that has a present tense. When this same morphology (u, e/ε, ο) appears on the other verbs in the language, it is interpreted as subjunctive or imperative. These tenses are thus the only other forms which show person and number agreement. They are also, like the forms of the verb ho 'be', remnants of the original Sanskrit system. Present tense readings for other verbs are expressed via the progressive and imperfect periphrastic constructions, as illustrated in (37) and (39). The other parts of the verbal morphological paradigm involve only number and gender agreement. Table (33) shows the 'imperfect' (habitual) forms for mar 'hit'. (33) Singular Μ mar-t-a F mar-t-i

Urdu Imperfect Plural Respect (ap) mar-t-e mar-t-e mar-t-i mar-t-i

Familiar (tum) mar-t-e mar-t-i

Verbal Morphology and Agreement in Urdu

79

All the forms which show number and gender agreement derive from participles. Table (34) shows the "perfect" forms for mar 'hit', which also inflects for number and gender. (34) Singular Μ mar-a F mar-i mar- 'hit'

Urdu Perfect Plural Respect (ap) mar-e mar-e mar-i mar-i

Familiar (tum) mar-e mar-i

The folllowing table summarizes the distribution of number, person and gender agreement across the verbal forms in Urdu. Verb Form Past, Perfect, Imperfect, Progressive, Past 'be' Imperative, Subjunctive, Non-past 'be' Future

Number and Gender

Number and Person —

—

y/

y/ y/

The core periphrastic tenses in Urdu are the present and past imperfects, present and past perfects and present and past progressives. These analytic tenses take the form of a non-finite (aspectually marked) form of the verb (inflected for gender and number) combined with a form of the auxiliary be. Hence the present imperfect arises from a combination of the forms in (32) and (33). The verb ho 'be' lacks past morphology (though it can appear in the present (32), future, imperfect and perfect). It therefore forms a suppletive paradigm with another 'be' verb: f - (based on a participle form of Sanskrit sthä 'stand'). This verb only has a past form, as shown in (36). The past imperfect thus combines the forms in the table (33) with those in (36), which are inflected for number and gender.

80

Miriam Butt and Louisa Sadler

1st 2nd 3rd

Past of Urdu be Singular Plural Respect (ap) Familiar (tum) M/F M/F M/F M/F h h t -e/i t -a/i h th-e/r t -a/i th-e/i th-a/i th-e/i

An example of the periphrastic imperfect is given in (37). (37) a. nadya gari cala-t-i h-ei Nadya.F.Nom car.F.Nom drive-Impf-F.Sg be-Pres.3.Sg 'Nadya drives a car.' b. nadya gari cala-t-i P-i Nadya.F.Nom car.F.Nom drive-Impf-F.Sg be.Past.F.Sg 'Nadya used to drive a car.' Similarly the present perfect combines the forms in (34) with (32) and the past perfect the forms in (34) with (36).7 (38) a. nadya=ne gari cala-yi h-ei Nadya.F=Erg car.F.Nom drive-Perf.F.Sg be-Pres.3.Sg 'Nadya has driven a car.' b. nadya=ne gari cala-yi f-i Nadya.F=Erg car.F.Nom drive-Perf.F.Sg be.Past.F.Sg 'Nadya had driven a car.' The progressive combines forms of rah 'stay' with the above forms. This verb is inflected just like mar 'hit' in tables (31), (33) and (34) above. As the progressive does not concern us any further here, we simply provide some examples in (39). (39) a. nadya gari cala rah-i h-ei Nadya.F.Nom car.F.Nom drive stay-Perf.F.Sg be-Pres.3.Sg 'Nadya is driving a car.'

Verbal Morphology and Agreement in Urdu

81

b. nadya gari cala rah-i Nadya.F.Nom car.F.Nom drive-Perf.F.Sg stay-Perf.F.Sg f-i be.Past.F.Sg 'Nadya was driving a car.' c. nadya gari cala-ti rah-t-i Nadya.F.Nom car.F.Nom drive-Impf.F.Sg stay-Impf-F.Sg h-ei be-Pres.3.Sg 'Nadya keeps driving a car.' d. nadya gari cala-ti rah-t-i Nadya.F.Nom car.F.Nom drive-Impf.F.Sg stay-Impf-F.Sg f-i be.Past.F.Sg 'Nadya used to keep driving a car.' As is clear from the above, the tense/aspect system of Urdu is quite complex and a full description would take us beyond the scope of this paper. For example, in addition to the forms illustrated above, morphological reduplication can be used to expression repetitive action.

4.2.

Formalizing the agreement generalizations

To illustrate how verbal agreement is analysed, we deal only with the number and gender agreement exemplified above—it should be clear that the person/number agreement (found with the present of 'be' and in the (synthetic) future tense) can be expressed along the same lines. A fully inflected word such as calayi 'drive-Perf.F.Sg' will be associated with the following functional information in the syntax (the predicateargument structure has been left out for purposes of illustration):

82

Miriam Butt and Louisa Sadler

(40) cala-yi ( t A S P ) = PERF [

(tSUBJ CASE) = c NOM (tSUBJ GEN) = FEM (tSUBJ NUM) = SING

V

[

]

(tSUBJ CASE) φ NOM (tOBJ CASE) = c NOM flOBJ GEN) = FEM (tOBJ NUM) = SING

]

The agreement information in the entry in (40) is within the disjunction (notated by means of [ ] V [ ]). Recall that the choice of this (non-default) verb form is only appropriate so long as either the subject or the object is a feminine singular nominative nominal. The first disjunct covers the case where the subject is nominative: it may be read as assigning the attribute value pairs GEN:FEM and NUM:SG to a subject on condition that the subject is nominative. The constraining equation, notated by means of the constrained equality =c, does not assign CASE:NOM to the subject but checks for the presence of this attribute value pair (this is thus a filter rather than an information defining statement). In similar fashion, the second disjunct assigns the attribute value pairs GEN:FEM and NUM:SG to the object provided that the subject is not nominative and on condition that the object is nominative.8 As is evident, agreement (a) is seen essentially as a lexical constraint and (b) involves the verbal element directly constraining values of the index features of the f-structure of the nominal, a view which is quite standard in LFG. By way of illustration, consider the syntactic analysis of the sentence in (41), in which the verb agrees with the (nominative) object. (41) gari adnan=ne cala-yi h-ei car.F.Nom Adnan.F=Erg drive-Perf.F.Sg be-Pres.3.Sg 'Adnan has driven a car.' As noted above, the order of nominal constituents within the clause is syntactically rather free in Urdu (see Mahajan 1990; Butt 1995; Kid-

Verbal Morphology and Agreement in Urdu

83

wai 1997). This is modelled in LFG by means of a nonprojecting endocentric category S which directly dominates nominal constituents mapped to clause internal grammatical functions (cf. King 1995; Bresnan 2001). The precise determination of the grammatical function (GF) associated with each KP is rendered via an interaction of information from the case clitics and the verbal predicate.9

(42)

S

—•

(43)

KP*, (tGF)=+

V' t=l

S (TGF) = ; KP

(|GF) = ^

gari

adnan-ne

Τ

I

cala-yi

I

h-εϊ

For the example at hand, these jointly determine that adnan ne is the SUBJ and gari 'car' is the OBJ. Consider now the lexical information associated with calayi (40). Since the SUBJ is not nominative and the OBJ is nominative, the first disjunct does not apply, but the second does, requiring that the agreement features born by the OBJ itself are consistent with the values contributed by the verb, namely, FEM and SG. The corresponding f-structure analysis is shown in (44). As can be seen in (44), agreement is dealt with in LFG at the level of f-structure by permitting the verbal head to directly constrain the agremeent features of the argument, constraints which are not mediated by the level of constituent-structure representation.

84

Miriam Butt and Louisa Sadler

(44)

PRED

SUBJ

OBJ

5.

4

d r i v e < (tSUBJ ) (tOBJ) > ' PRED

'Adrian

CASE

ERG

NUM

SG

GEND

MASC

PERS

3

PRED

'car' "

CASE

NOM

NUM

SG

GEND

FEM

PERS

3

TENSE

PRES

ASP

PERF

Α simple morpheme-based morphology

As a syntactic theory which incorporates a strong lexicalist hypothesis, LFG is in principle compatible with any theory of morphology from lexical to inferential realizational (e.g., Stump 2001). However, where the main focus has been on primarily syntactic phenomena, work in LFG has generally assumed, largely for expository convenience, a version of a word syntax approach, in which morphological structures are phrase structure trees of the familiar sort, with stems and affixes as leaves, associated with their own separate lexical entries. For our examples, this involves defining sub-lexical entries for affixes yi (ASP:PERF, GENDIFEM, NUM:SG), ei (TENSE:PRES, NUM:SG, PERS:3), and so on, and for the verb stems, and a set of sublexical phrase structure rules to combine these elements. We explore a number of ways in which this is less that ideal.

Verbal Morphology and Agreement in Urdu

85

5.1. Default agreement The disjunctions (number of possibilities) needed for the treatment of masculine singular verb forms, such as calaya 'drive', are more complex than the example shown with feminine singular agreement in the previous section in (40). The reader will recall from earlier discussion that verb agrees with the highest nominative argument in the clause (where SUBJ is higher than OBJ) and in the absence of such an argument defaults to the masculine singular. Thus masculine singular forms have both an agreement and a non-agreement (that is, default) usage. A partial sample entry is shown in (45). (45)

- y a , V ^ C T A S P ) = PERF

[

( | S U B J CASE) = C NOM (tSUBJ GEN) = MASC (tSUBJ NUM) = SING

V

[

]

(tSUBJ CASE) φ NOM (fOBJ CASE) = c NOM (tOBJ GEN) = MASC

V

[

(|OBJ NUM) = SING

]

(tSUBJ CASE) φ NOM (tOBJ CASE) φ NOM

]

This (very standard) approach certainly guarantees the correct syntactic behavior—the constraints expressed in sub-lexical entries for the verbal agreement suffixes (such as (45) and (40) ensure that feminine singular verbs co-occur with feminine singular nominative arguments, masculine plural verbs with masculine plural nominative arguments, and so on, as well as ensuring that the (default) masculine singular is the only form available in the absence of a nominative argument. However what this fails to express is the fact that forms suffixed by -ya such as calaya, diya, and so on, are morphologically masculine singular forms. The obvious generalization, namely: if there is no appropriate argument to agree with, use the masculine singular form, is not directly expressed. This cannot be expressed in terms of f-structure features, for adding agreement information to the third (default case) disjunct as in (46) above would simply and inappropriately specify the clause itself as masculine singular.

86

Miriam Butt and Louisa Sadler

(46) (fSUBJ CASE) φ NOM (tOBJ CASE) φ NOM (tGEND) = MASC (|NUM) = SG

What seems necessary is to recognise a distinction between morphological and syntactic features, which is not captured in this most simple approach in which affixes are directly associated with f-structure features. Several proposals exist for an additional, morphological or morphosyntactic projection (the μ projection) which permits a clearer separation of purely morphological from syntactic features. The postulation of an additional projection is, of course, independent of the choice between a morpheme-based or a realizational model of morphology. The μ projection or level of representation was first proposed by Butt, Nino and Segond (1996) in order to deal with morphological wellformedness dependencies in German and English verbal complexes. The subsequent proposal by Frank and Zaenen (2000) assumes a morpheme-based morphology, while Sadler and Spencer (2001) combine the use of an m-structure for morphological features with a realizational model of morphology. The proposals by Frank and Zaenen (2000) and Sadler and Spencer (2001) use an architecture in which m-structure is projected from f-structure as shown in (47). (47) c-str —> f-str

m-str

This is in many ways counterintuitive, and in section 6 we discuss another possible model of the morphology-syntax interface—one that is instantiated and exemplified by the use of finite-state morphological analyzers in conjunction with syntactic representations. But first, we return to a further discussion of the word-syntax approach. 5.2.

Controlling combinations—the role of paradigms

In a word syntax approach, there is no role for the notion of paradigm. Affixes themselves are treated as meaning/form pairs. One of the problems that this entails is that of controlling adequately the combination of stems and affixes. This is often possible only at the cost of the

Verbal Morphology and Agreement in Urdu

87

introduction of additional, otherwise unmotivated, features (in the fstructure) and/or the introduction of otherwise unnecessary categorial distinctions. A case in point is provided by the inflectional possibilities for the two verbs 'be': one of which occurs only in past tense forms (inflecting for number and gender) and the other of which occurs in the other forms (inflecting for person and number). (48) a. tum

yahäh-o

you.Nom here be-Imp.2.Sg 'You be here!' a. tum

yahä

f-i

you.Nom here be.Past-F.Sg 'You were here.' It is clear that these two 'be' verbs form part of the same paradigm: both verbs function as auxiliaries in exactly the same way and both verbs have the same copula uses. The f - is restricted to the past and thus forms a suppletive paradigm with the ho form. As such, one would like to be able to deal with both the 'be' verbs by means of a single unifying sublexical rule such as in (49). This rule produces tensed auxiliaries by taking a base form of an auxiliary and affixing a tense morpheme such as -εί or -i above in (48a) and (48b), respectively (a sample sub-lexical tree produced by this rule is shown in (50)). (49) COPtns

—»

COPtns-BASE

(50)

V-TNS-AFF

COPtns t=l COPtns-BASE

t=4 V-TNS-AFF

However, if no further specifications are added, this rule will also produce the incorrect sentence in (51), where the past 'be' auxiliary is invested with person and number agreement morphology, rather than number and gender morphology.

88

Miriam Butt and Louisa Sadler

(51) *tum yahäf-o you.Nom here be.Past-2.Sg 'You were here.' On the word syntax view, the person-number affix is a sub-lexical leaf of the same kind as the number-gender affix, and this makes it difficult to control the combinatorial possibilities. The possibility of (51) must be ruled out via cumbersome constraints in the (sublexical) lexion, whereby the person-number affix for the 'be' verb is prohibited from appearing with the past tense and the number-gender affix is restricted to the past tense.10 Explicit recognition of a suppletive 'be' paradigm in which the past tense stem is and all other cells involve the stem ho- would permit us to capture this restriction directly (rather than by doing what amounts to engineering an 'accidental' distributional gap) for both copula and auxiliary uses of 'be'. Furthermore, in directly pairing content and form in lexical entries for inflectional affixes, the word syntax approach can easily fail to make the distinctions which we wish to make on morphological grounds from those which have syntactico-semantic content. Thus on main verbs the use of present tense inflections signals subjunctive mood (recall that present tense does not occur otherwise on lexical verbs— present time reference is signalled periphrastically). (52) nadya a-yel Nadya.F.Nom come-Pres.3.Sg 'Should Nadya come? On the other hand, the present morphology on 'be' in both its copula and auxiliary use does indeed signal present tense. This would seem to be a case where the verbs share a set of morphological forms, but these map differently to f-structure information (or syntactic featural content). This could be stated simply if we separated morphology (and morphological features) cleanly from the syntax (cf. Sadler and Spencer 2001), but again, in the present approach, cumbersome stipulation is required by means of an explicit rule which refers only to the non-past copula.

Verbal Morphology and Agreement in Urdu

89

5.3. Allomorphy and homophony Finally, the (word syntax) grammar is considerably complicated by the existence of a high degree of allomorphy and homophony in Urdu. Allomorphy causes a duplication of lexical entries—indicating clearly that a more adequate approach would need at least to factor out predictable morphophonological alternations. The degree of homophony in the Urdu inflectional system is striking, with the inflection -e (a mid vowel which many originally differing morphemes have developed into in the course of language change), for example corresponding to at least the following functions: (i) number and gender with some verbal forms (perfect, imperfect, future); (ii) person and number with other verbal forms (future); (iii) oblique inflection on infinitives and nouns; (iv) adjective agreement; (v) nominal agreement. Treating this by means of a single sub-lexical entry for the affix requires a complicated disjunction of conditional statements with several substatements (to deal with the suppletive 'be' forms). This is difficult to write, error prone, and more importantly, does not capture any useful linguistic generalization. Note that this type of homophony is crosslinguistically extremely common. It is therefore imperative to adopt an approach to the morphology-syntax interface which can deal with these kinds of phenomena elegantly and correctly. 6. Finite-state morphology In this section we provide a very brief illustration of how some of these issues can be addressed within a computational environment using finite state technology (Karttunen, Kaplan and Zaenen 1992; Kaplan and Kay 1994) to encode morphological descriptions, interfaced appropriately to the syntax (e.g., see Butt, King, Nino and Segond 1999). The use of finite state technology does not, of course, constitute a theoretical model of morphology, but it sheds an interesting (practical) light on concerns which interest theoretical and descriptive morphologists. The standard application of finite-state technology to morphological analysis and generation involves transduction between two levels or

90

Miriam Butt and Louisa Sadler

strings—a surface form (that is, the actual word form itself) and what is known as a lexical form, which is typically the canonical dictionary form (lemma) and a string of features or morphological subcategories. These levels are related by transducers which directly encode morphological alternations, and relate all inflected forms of the same word to the same canonical dictionary form (lemma), accompanied by different features. Clearly, moving from a lemma or root and set of features to a surface form is very much like looking up a surface form in a paradigm listing, a point of similarity between finite state morphology and theoretical realizational models. The examples below show some surface/lexical pairings for simple words, illustrating the sorts of morphological features found in lexical forms. (53) dogs 1. dog+Noun+Pl 2. dog+Verb+Pres+3sg (54) kutte 'dogs' 1. kutt+Noun+Masc+Pl 2. kutt+Noun++Masc+Sg+Obl As can be seen, an English form such as dogs is analyzed as being one of two possible forms: a plural noun or a present singular verb. The base form (lemma) 'dog' is returned in the morphological analysis, but the morpheme -s finds no representation beyond the list of features. The Urdu example illustrates the same thing, except that Urdu nouns are often marked for gender. The form kutte in Urdu can either be analyzed as the simple unmarked plural form, or as a singular form with oblique inflection. This oblique inflection is a prerequisite for case marking, as in kutte=ko 'to the dog'. The relation between a surface (realized) string and an abstract analysis into a lemma and one or more tags may be thought of as the morphology proper—in analysis mode, the output is a lemma and string of (morphological) features, in generation mode, the output is a surface string. The finite state morphological component is entirely separate from the syntactic component. However, the abstract tags (string

Verbal Morphology and Agreement in Urdu

91

of morphological features) can be used to define an interface to the syntax. This interface consists of two parts: 1) a sub-lexical grammar which parses the tags; 2) an association of syntactically relevant information with the tags. The sub-lexical grammar which parses the tags takes the form of sub-lexical rules as in (55), which takes care of tensed copulas. (55) COPtns — • COPtns-S

V-T V-F*

At first glance, this rule seems identical to the rule presented in the previous section in (49). However, whereas the rule in (49) was designed to parse morphemes, this rule is designed to parse the abstract features or tags provided by the morphological analyzer. Consider the analyses of the 'be' verbs hei and te, respectively. (56) a. hei 1. be+Verb+Pres+3P+Sg 2. be+Verb+Pres+2P+Sg b. the 1. be+Verb+Past+Masc+Pl The rule in (55) parses the tags in (56): it expects a base form (be) followed by a verbal tag (V-T) in the rule, (+Verb is the tag), followed by any or no number of verbal features. This allows for the tense, gender, number and person features such as +Past, +P1, +Masc. Due to the fact that the morphological analyzer itself only ever allows the +Past tag in association with the t - forms, the problem of having to constrain the grammar so that only the right kinds of morphemes appear with the right kind of stem does not arise. The well-formedness of sequences of tags is guaranteed by the definition of continuation classes in the finite state morphology, which specify what tags can be consumed/output in transition from one internal state to the next. In (57) the first state is 'be', the next state is '+Verb\ etc. These continuation classes can be defined in simple and general terms. So, for example, it should have been clear from section 4 on agreement that there are a number of verb forms which take number and

92

Miriam Butt and Louisa Sadler

gender affixes of the same kind. The verb f - is one of these. In fact, this verb only allows this morphology and is inherently past. This basic property can be taken care of by defining the tag +Past as part of the entry for f - , as shown in (57). The 'GendV' defines the continuation class for this verb as containing form: content pairs signaling number and gender, as shown in (58). These morphs have no further continuation class (indicated by the '#'), and thus no other affixal processes. (57) be+Verb+Past:th (58)

GendV +Masc+Sg:a +Fem+Sg:i +Masc+Pl:e +Fem+Pl:i

GendV;

#; #; #; #;

The 'be' verb ho, on the other hand, is defined with a different continuation class because this verb, unlike P-, can occur in a variety of tenses and aspects. One of its possible continuation classes is to take person and number morphology, as we have seen in the examples above. This continuation class, called 'PN' is defined as in (59). (59)

PN +Pres+lP+Sg:ü +Pres+2P+Sg:ei +Pres+3P+Sg:ei +Pres+lP+Pl:gT +Pres+2P+Resp: εϊ +Pres+3P+Pl:£i +Pres+2P+Fam:o

# # # # # # #

Note that there is no one-to-one correspondence between tags such as +Past or +Pres and forms. The stem f - is itself associated with the Past feature, while in the case of ho, on the other hand, the tense may differ. If the verb is followed by only 'person and number' morphology, the interpretation has to be present.

Verbal Morphology and Agreement in Urdu

93

In effect them, whereas the word-syntax approach of the previous section dealt with the combinatorial possibilities of stem and affixes with sub-lexical rules which also provided the f-structure information, the current approach separates out the morphology proper from the morphology-syntax interface, and the combinatorial possibilities are dealt with within the morphological component which relates surface forms to an abstract system of tags/features. The morphological features are integrated into the LFG analysis via a mapping between the sets of morphological tags and f-structure attribute value pairs. In the simplest case, the tag +Pres, for example, gives rise to the f-structure information TENSE:PRES, as shown in (60). (60)

+Pres:

V-F

( | T E N S E ) = PRES

It is this mapping which introduces the necessary separation between morphological and syntactic features to provide an intuitive treatment of both default and agreeing uses of masculine singular verb forms, as discussed in the previous section: morphologically, all uses are masculine singular, but syntactically, f-structure constraints over the SUBJ or OBJ are introduced only in the non-default usage. The finite-state analysis we have briefly sketched is not a morphemebased analysis of the type explored in the first implementation. The mapping between surface forms and functional (f-structure) or wellformedness (m-structure) information is not modeled via a simple relation between surface morphology and abstract information. Rather, several levels of analysis are presupposed. The first is a mapping from a surface string to a more abstract representation into a lemma and tags. This level of analysis includes a treatment of morphophonologically predictable allomorphy via the integration of phonological rules into the morphological analysis. All this is invisible to the morphologysyntax interface. The morphologically driven information becomes visible to the syntax only through the association of the abstract tags with f-structure information, as shown in (60). This more complex view of the morphology-syntax interface furthermore has the advantage that paradigms can be identified within the morphological component, rather than being the accidental by-product

94

Miriam Butt and Louisa Sadler

of sublexical rules within the syntactic component (cf. the two 'be' verbs in (58) and (59), which can now be considered part of the same paradigm, though allowing for differing continuation classes). Within this architecture, the initial morphological analysis is arrived at within an independent module whose syntax is of a very different kind than that of a grammar (e.g., continuation classes and morphophonological rules rather than phrase structure trees). The architecture permits a separation of strictly morphological information from syntactic information and preserves lexical integrity: at the point of lexical insertion the abstract morphological information encoded in the tags is mapped into information relevant for the f-structural analysis of the sentence. Although this mapping between tags and LFG features sets will often be quite trivial, the many-to-one nature does permit a sophisticated and clearly defined morphology-syntax interface. 7.

Conclusion

This paper has presented some aspects of LFG morphosyntax, discussing in particular case and agreement in Urdu, and showing how both can be given a straightforward analysis in LFG. We then turned to the nature of the morphological component itself, looking more closely at issues of morphological representation. After sketching a morpheme-based word syntax approach to the generalizations previously formulated, we discussed a number of insufficiencies in this approach and showed how they can be largely overcome under the assumption of a different model of the morphology-syntax interface. LFG'S stance on morphology can perhaps be compared to its stance on semantics: in principle, several differing theories are compatible with the fundamental architectural assumptions of LFG. For example, Discourse Representation Theory (Kamp and Reyle 1993) is in principle just as compatible with LFG as a semantic analysis based on linear logic (Dalrymple 2001). Similarly, morpheme-based analyses are in principle just as compatible with the fundamental tenets of LFG as are realizational models or finite state morphologies. The basic requirements that LFG does impose on the morphological component is that it

Verbal Morphology and Agreement in Urdu

95

respect lexical integrity (as such Distributive Morphology is not compatible with LFG) and that the functional information contributed by morphological elements be able to play a role in the syntactic and semantic analysis of a clause on a par with phrasal information. After exploring two differing approaches to the morpho-syntax interface, we concluded that the method traditionally used by theoretical linguists (often more as a matter of convenience), namely, morphemeor word-based morphology, has several undesirable side-effects. The architecture assumed forfinite-statemorphological analysis (employed primarily within computational circles), on the other hand, would seem to yield the desired results. For one, the morphological component operates according to rules and principles which are distinct from that of the syntax. For another, the interface to the syntactic and semantic analysis abstracts away from the surface realization of the morphemes, but still allow functional information contributed by the morphology to enter the syntactic analysis in a well-defined and regular manner. This is precisely the kind of syntax-morphology interface required for an analysis of languages with rich morphology.

Notes 1. This paper was first presented as part of the Workshop on Clause Structure and Models of Grammar from the Perspective of Languages with Rich Morphology at the DGfS in Leipzig. We would like to thank Uwe Junghans and Luka Szucsich for organizing the workshop. The issues discussed in this paper arose partly out of a workshop organized by Louisa Sadler and Andrew Spencer on Morphology in LFG during the LFG conference held at Berkeley in 2000. We would like to thank an anonymous reviewer, Mary Dalrymple, Anette Frank, Ron Kaplan, Lauri Karttunen, Rachel Nordlinger, Andrew Spencer and Annie Zaenen for some very stimulating discussion. Miriam Butt's contribution to this paper was made possible by financial support from the DFG (the German Science Foundation) via the SFB 471 at the University of Konstanz. 2. The South Asian languages Urdu and Hindi are closely related. Both are among the official languages of India and are spoken primarily in the north of India. Urdu is the national language of Pakistan. 3. The Principle of Morphological Composition is given as (i), where χ is a string of attributes:

96

Miriam Butt and Louisa Sadler i.

4.

5.

6. 7.

8. 9. 10.

Stem Äff ( ( G F ^ t j ) ((GFm ( t ) ) x)

=ί>

Stem (GF"t)

Äff ((GFm (GFn f))*)

This principle is defined in terms of annotated trees in the morphology. In a realizational model, it would seem to require an operation over f-descriptors to be associated with the application of certain morphological operations. A reviewer notes that a similar combination of lack of verb agreement and extremely rampant pro-drop also characterises a number of other South Asian languages, thus reinforcing our point. Bresnan (2001) and Falk (2001) make limited use of empty categories in unbounded dependency constructions in some coniigurational languages. Some arguments in favour of this analysis are adduced from weak crossover—however Dalrymple, Kaplan and King (2001) argue that no empty categories are needed in LFG even in view of the weak crossover data. The form *muf-e=ko is blocked due to syntactic status of the pronoun muf-e as a Κ heading a KP: the case clitics take NPs as sisters. The auxiliary is required when past (that is, in the past imperfect, past perfect and the past progressive), but is optional when present (in the present imperfect, perfect and progressive). Of course, there is no relevance to the order in which the disjuncts appear in the lexical entry—they could as well be in the opposite order. For more details see Butt (1995), Butt and King (2001,2002a, 2002b). This is a simple illustration of a much larger problem of controlling combinatorial possibilities. The issues much more tricky in languages with more complex morphology, for example where there are dependencies between affixes.

References Anderson, Stephen R. 1992 A-morphous Morphology. Cambridge: Cambridge University Press. Andrews, Avery 1996 Semantic case-stacking and inside-out unification. Australian Journal of Linguistics 16(1): 1-55. Andrews, Avery and Christoper Manning 1999 Complex Predicates and Information Spreading in LFG. Stanford, California: CSLI Publications. Aronoff, Mark 1994 Morphology by itself: Stems and inflectional classes. Cambridge, Massachusetts: The ΜΓΓ Press.

Verbal Morphology and Agreement in Urdu 97 Börjars, Kersti, Nigel Vincent and Carol Chapman 1997 Paradigms, pronominal inflection and periphrasis: a feature-based account. In Geert Booij and Jaap van Marie (eds.) Yearbook of Morphology, 155-180. Dordrecht: Kluwer Academic Publishers. Baker, Mark 1985 The Mirror Principle and Morphosyntactic Explanation. Linguistic Inquiry 16: 373-416. 1988 Incorporation: A Theory of Grammatical Function Changing. Chicago, Illinois: University of Chicago Press. Bresnan, Joan (ed) 1982 The Mental Representation of Grammatical Relations. Cambridge, Massachusetts: The ΜΓΓ Press. Bresnan, Joan 2001 Lexical-Functional Syntax. Oxford: Blackwell. Bresnan, Joan and Sam Mchombo 1987 Topics, Pronoun and Agreement in Chichewa. Language 63:741-782. Bhatt, Rajesh 2002 Long Distance Agreement in Hindi-Urdu. Manuscript, The University of Texas, Austin. Butt, Miriam 1993 Object Specificity and Agreement in Hindi/Urdu. In Papers from the 29th Regional Meeting of the Chicgo Linguistic Society, 80-103. 1995 The Structure of Complex Predicates. Stanford, California: CSLI Publications. Butt, Miriam and Ashwini Deo 2001 Ergativity in Indo-Aryan. In KURDICA Newsletter for Kurdish Language and Studies. http://www.cogsci.ed.ac.ukTsiamakr/kurdica.html. Butt, Miriam, Maria-Eugenia Nino and Frederique Segond 1996 Multilingual Processing of Auxiliaries in LFG. In D. Gibbon (ed.), Natural Language Processing and Speech Technology: Results of the 3rd KONVENS Conference, Bielefeld, October, 111-122. Berlin: Mouton de Gruyter. Butt, Miriam, Tracy Holloway King, Maria-Eugenia Nino and Frederique Segond 1999 A Grammar Writer's Cookbook. Stanford, California: CSLI Publications. Butt, Miriam and Tracy Holloway King 2002a The Status of Case. In Veneeta Dayal and Anoop Mahajan (eds.), Clause Structure in South Asian Languages. To Appear. Dordrecht: Kluwer Academic Publishers. 2002b Case Systems: Beyond Structural Distinctions. In Ellen Brandner and Heike Zinsmeister (eds.), New Perspectives on Case Theory. To Appear. Stanford, California: CSLI Publications.

98 Miriam Butt and Louisa Sadler 2001

Non-Nominative Subjects in Urdu: A Computational Analysis. In Proceedings of the International Symposium on Non-nominative Subjects, 525-548. Tokyo: ILCAA. Chomsky, Noam 1970 Remarks on Nominalization. In R. A. Jacobs and RS. Rosenbaum (eds.), Readings in English Transformational Grammar. Waltham, Massachuset Ginn. Also in N. Chomsky. 1972. Studies on Semantics in Generative Grammar. The Hague: Mouton. Dalrymple, Mary 1993 The Syntax ofAnaphoric Binding. Stanford, California: CSLI Publications. 2001 Lexical Functional Grammar. Syntax And Semantics Volume 34. New York: The Academic Press. Dalrymple, Mary, Ronald M. Kaplan and Tracy Holloway King 2001 Weak Crossover and the Absence of Traces. In Miriam Butt and Tracy Holloway King (eds.), The Proceedings of the LFG01 Conference. Hong Kong. http://cslipublications.stanford.edu/LFG/ Davison, Alice 1999 Ergativity: Functional and Formal Issues. In Michael Darnell, Edith Moravcsik, Frederick Newmeyer, Michael Noonan and Kathleen Wheatley (eds.), Functionalism and Formalism in Linguistics, Volume I: General Papers, 177-208. Amsterdam: John Benjamins. Deo, Ashwini and Devyani Sharma 2002 Typological Variation in the Ergative Morphology of Indo-Aryan Languages. Manuscript, Stanford University. DiSciullo, Anne-Marie and Edwin Williams 1987 On the Definition of Word. Cambridge, Massachusetts: The MIT Press. Εης, Mürvet 1991 The Semantics of Specificity. Linguistic Inquiry 22(1): 1-25. Falk, Yehuda 2001 Lexical-Functional Grammar: An Introduction to Parallel ConstraintBased Syntax. Stanford, California: CSLI Publications. Frank, Anette 1996 A Note on Complex Predicate Formation: Evidence from Auxiliary Selection, Reflexivization, and Past Participle Agreement in French and Italian. In Miriam Butt and Tracy Holloway King (eds.), The Proceedings of the LFG '96 Conference. Rank Xerox, Grenoble. http://cslipublications.stanford.edu/LFG/ Frank, Anette and Annie Zaenen 2000 Tense in LFG: Syntax and Morphology. In Hans Kamp and Uwe Reyle (eds.), How we say WHEN it happens: Contributions to the theory of temporal reference in natural language.To Appear. Niemeyer, Tübingen.

Verbal Morphology and Agreement in Urdu

99

Halle, Morris and Alec Marantz 1993 Distributed Morphology and Pieces of Inflection. In K. Hale and S.J. Keyser (eds.), The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger, 111-176. Cambridge, Massachusetts: The MIT Press. Halvorsen, Per-Kristian and Ronald M. Kaplan 1988 Projections and Semantic Description in Lexical-Functional Grammar. Proceedings of the International Conference on Fifth Generation Computer Systems, 1116-1122. Tokyo. Kaplan, Ronald M. and Martin Kay 1994 Regular Models of Phonological Rule Systems. Computational Linguistics 20(3): 331-378. Kamp, Hans and Uwe Reyle 1993 From Discourse to Logic. Dordrecht: Kluwer Academic Publishers. Karttunen, Lauri, Ronald M. Kaplan and Annie Zaenen 1992 Two-level Morphology with Composition. In Proceedings of the 14th International Conference on Computational Linguistics (COLING-92), 141-148. Kidwai, Aeysha 1997 Scrambling and Binding in Hindi-Urdu PhD dissertation, Jawaharlal Nehru University, New Delhi. King, Tracy Holloway 1995 Configuring Topic and Focus in Russian. Stanford, California: CSLI Publications. Krifka, Manfred 1992 Thematic relations as links between nominal reference and temporal constitution. In Ivan Sag and Anna Szabolcsi (eds.), Lexical Matters, 29-53. Stanford, California: CSLI Publications. Mahajan, Anoop 1990 The A/A-Bar Distinction and Movement Theory PhD dissertation, MIT. 1992 The Specificity Condition and the CED. Linguistic Inquiry 23(3): 510516. McGregor, R.S. 1968 The Language oflndrajit ofOrchä. Cambridge: Cambridge University Press. Mohanan, Tara 1994 Argument Structure in Hindi. Stanford, California: CSLI Publications. Nordlinger, Rachel 1998 Constructive Case: Evidence from Australia. Stanford, California: CSLI Publications.

100 Miriam Butt and Louisa Sadler Rizzi, Luigi 1986 Null Objects in Italian and Theory of pro. Linguistic Inquiry 17(3): 501-557. Sadler, Louisa and Andrew Spencer 2001 Syntax as an exponent of morphological features. In Geert Booij (ed.), Yearbook of Morphology 1999. Sells, Peter 2001 Syntactic Information and its Morphological Expression. In Louisa Sadler and Andrew Spencer (eds.), Relating Morphological and Syntactic Structure: Morphology in LFG. To Appear. Stanford, California: CSLI Publications. Selkirk, Elisabeth O. 1982 The Syntax of Words. Cambridge: Massachusetts: The MIT Press. Subbarao, Κ. V. 1999 Agreement in South Asian Languages. In the Proceedings of the South Asian Language Analysis Roundtable. To Appear. Chicago, Illinois: University of Illinois at Urbana-Champaign. Stump, Gregory, T. 1991 A Paradigm-theory of morphosemantic mismataches. Language 67(4): 675-725. Stump, Gregory, T. 2001 Inflectional Morphology: A Theory of Paradigm Structure. Cambridge: Cambridge University Press. Tenny, Carol 1994 Aspectual Role and the Syntax-Semantics Interface. Dordrect: Klu wer Academic Publishers.

Particles and sentence structure: a historical perspective* Gisella Ferraresi and Maria Goldbach

1.

Preliminaries

We sketch here a small fragment of the syntax of Old French which shows that syntactic change should be viewed as the interplay of different micro factors, phonology and lexical semantics. As far as methodology is concerned, we follow Keenan (2001) and Longobardi (2001), who base themselves on the hypothesis that syntactic change occurs only as a consequence of changes of the interface conditions, in Keenan's terminology as the consequence of semantic and/or phonological 'decay' (where semantic erosion often co-occurs with semantic decay). Our results support this hypothesis. That means that certain syntactic phenomena depend on the semantic and phonological representation of corresponding functional elements. We believe the analysis of syntactic parameters in generative terms to be a new kind of approach. Our research is a first try at Old French Verb-second (V2) phenomena, while, however, we do not look any further into the nature of a V2 parameter. In Minimalism it is assumed that a system comprising a set of formal features is available and that a language makes a selection from these features. Parameterisation is the way the different features are realised in individual lexical items. For example, in one language Τ and Agr may be realised in one and the same lexical item, whereas in another language Τ and Agr may have distinct lexical realisations. So, the task of the child is to distinguish what the subset of these features is and how these features are combined. In other words, which features are assembled into a given category. However, some strong restrictions are imposed by UG. For example, Tense will never select a C, only the contrary will be possible. Some restrictions however could simply be a question of implementation, e.g. in the phonology

102

Gisella Ferraresi and Maria Goldbach

the sequence lateral-rhotic (*lr) is excluded, but this may be a question of articulation rather than of Universal Grammar. So we take parameters to be the way in which formal features are realised in different categories. Drawing on proposals for the syntax from Lightfoot (1999) and for the phonology from Dresher (1998), we assume that a child uses 'grammar fragments' (in Lightfoot's terminology) from the input for inferring the nature of the features in her grammar. These fragments are the cues from which the child identifies and assembles the syntactic/formal features of her particular grammar. Thus we consider that parameters are not located in the grammar as individual properties of a language, rather they are a consequence of the particular feature make-up in the language. Clustering effects, such as those properties associated with the so-called prodrop parameter, are merely an epiphenomenon. The case study we present here is the loss of the sentence particle si in French. We believe that macro changes which in the past have been explained through reanalysis, are not falsifiable; the aim, as Longobardi (2001:278) points out, is to find 'deep explanations for a portion of even small fragments of grammatical history of a language, rather than ... general principles immediately applicable to the totality of superficially observable changes'. In the Old French period the sentence particle si is in an intermediate stage: we will show that it is in the process of grammaticalisation, on its way to becoming a bound morpheme. But instead of becoming morphologized it disappeared in the 17th century. We will illustrate which factors contributed to the vanishing of si in Early Modern French. In addition, we want to show how the presence of particles in a language can throw light on sentence structure. In much of the generative literature on Old French si has been neglected, and in those works which do mention it, si has been treated as an adverb occupying a specifier position (of CP or of AgrP). In the following we will demonstrate that this view has to be abandoned and that si instead is a functional head.

Particles and sentence structure: a historical perspective

1.

103

The morphophonological shape of Old French si

Old French (OF) had several sentence particles (puis 'then', or 'now', ainz 'but') the most wide-spread of which is si 'thus'. It occurs in direct speech, in literary texts (poetry), in charters and in chronicles (narration). The other particles are less frequent and they are nearly absent from certain text types (e.g. ainz occurs only very rarely in charters, cf. Reenen and Schesler 2001). We focus therefore in this article on si. Si has different spellings: it appears as si, se, ci or ce in OF texts. Concerning its lexical accent, we observe that si functions as the (phonological) host of OF object clitics (me 'me', te 'you', le 'him', les 'them', etc.) and adverbial clitics (en 'of it', i 'there') as in (la), but there are examples where si is contextually reduced and cliticizes onto another element as in (lb). (1)

a.

Sin [< si en] vois vedeir alques de sun semblant (Roland 19, 270). Si of it will (I) see something of his appearance '(I) will see something of his appearance.' b. Car lavez, s'alez asseoir (Charrete, 1028, quoted in Buridant 2000:507). Thus wash (you), Si come sit down 'Thus wash (your hands), come sit down.'

On the other hand, si can stand alone, i.e. cliticization is not obligatory, and is even permitted in sentence initial position. (2)

Si prisent le palais ä force, si y menerent I'empereeur; ... (Clari 33,40-41). Si took (they) the palace by force, SI there led (they) the emperor. '(They) took the palace by force, (they) led the emperor there.'

In this respect si differs from true clitics (OF object and adverbial clitics) which (in Old French) are excluded from sentence initial

104

Gisella Ferraresi and Maria Goldbach

position and which are always phonologically dependent on a host. Hence, si takes part in clitic clusters where it functions as a host of right adjoined clitics. But unlike true clitics it may occur in absolute first position and it is not obligatorily subject to external sandhi rules (e.g. by apocope of its final '/'). This means that unlike object and adverbial clitics si has a lexical accent which may be reduced in certain contexts. In this respect si resembles exponents of functional categories such as auxiliaries, complementizers and determiners which can appear stressless depending on their position in the sentence (cf. Selkirk 1996 and references cited there). In order to obtain more evidence about the prosodic character of si we investigated its status in OF poetry, in particular in the Chanson de Roland (11th century). OF poetry had a fixed rhythmical frame into which the lexemes were inserted in accordance with their inherent rhythmical structure. Along these lines the canonical syntax could be altered in verse, i.e. a lexical item could surface in a syntactically inappropriate position, whereas poetry avoided its insertion into a rhythmically inappropriate position. Therefore we can deduce from metrical poetry little (if nothing) about OF syntax, instead we can infer the metrical character of OF lexical items (cf. Klausenburger 1970).

(3)

a.

b.

χ χ χ χ χ χ χ Si vait fe- rir || Ge- rin par sa grant force '(He) was about to attack Gerin with his enormous force.' (Roland 122,1618). χ χ X X X X X Ven- get Ii reis || si nus pur- rat ven- ger 'The king will come and (he) will be able to avenge us.' (Roland 132,1744).

The Chanson de Roland is a decasyllable rhyme, that is each line contains ten syllables, hence ten metrically relevant positions. By a rhythmical caesura (marked by the double slash || in (3)) a line is divided in two hemistichs. In early poetry the caesura in the decasyllable appears after the fourth syllable. The fourth and the

Particles and sentence structure: a historical perspective

105

tenth syllables always bear major stress (they receive the primary ictuses). The basic rhythmical pattern of the decasyllable is iambic, that is, a weak syllable is followed by a strong one (in example (3) above marked by superscribed x). Metrical rules - such as inversion - may change the iambic rhythm, but these rules never operate across the caesura nor do they involve the tenth syllable (see Nespor and Vogel 1986 and references cited there). Thus the fourth and the tenth syllables never lose their primary ictuses. As far as si is concerned, we never found it to occur in the fourth or in the tenth position. This cannot be due to a syntactic constraint since syntax in verse is much less rigid than in prose. From this we conclude that si cannot bear major stress. Moreover, in the great majority of verses we found si in the first position of a hemistich, that is, line-initially or in postcaesura position. The only other option where si can be hosted is the second position. Apart from the fourth and the tenth positions, si never surfaces in the third or in the sixth to ninth positions. Considering that the basic rhythm is iambic, the first and the fifth position - the main locations of si - are weak positions. We deduce from this that in principle si occurs in a metrically unstressed position. Interestingly, this is not a peculiarity of the Chanson de Roland, rather our results coincide with the findings of Marchello-Nizia (1985) who investigated the St. Alexis poem, also a decasyllabus rhyme from the 11th century. Summing up so far, the phonologically and metrically weak character of si suggests that it is a functional category and not a lexical one. In the next section we shall look at the syntactic behavior of OF si.

2. The syntax of si Like the Germanic languages, Old French has been analysed as a V2language in the pre-generative as well as in the generative literature (cf. Meyer-Lübke 1899, Adams 1987, Roberts 1993, Lemieux and Dupuis 1995, Vance 1997). Clearly, in such an analysis, the element si is viewed as an ordinary adverb, i.e. an XP in a specifier position; Lemieux and Dupuis (1995), for example, arguing against V-to-C

106

Gisella Ferraresi and Maria Goldbach

movement for Middle French, propose that si is generated in SpecAgrP. In Vance (1997) si occupies SpecCP. Thus, si is considered to be a lexical category. These analyses, however, fail to account for sentences like the following. (4)

[La damoisele a qui tu as parle] si est li anemis (Queste 113,1; from Vance 1997). 'The maiden to whom you have spoken si is the enemy.'

If si is in SpecCP, what is the position of the subject-DP? Furthermore, we do not find other (VP) adverbials in the position occupied by si nor can si occur in the position that they occupy. (5)

a. * [La damoisele a qui tu as parle] vraiement est li anemis. 'The maiden to whom you have spoken surely is the enemy.' b. [La damoisele a qui tu as parle] est vraiement li anemis. c. * [La damoisele a qui tu as parle] est si li anemis.

Sentential adverbs or adverbial expressions like apres gou 'after that' or adonc 'then' do not obligatorily trigger the postposition of the subject (5'a,b), although they appear adjacent to the verb, too (5'c,d). (5')

a.

Adonc li messageprisent congie ... (Clari 19,42). 'Then the messengers said good-by ...' b. Apres il sait que vos avez mis le vostre (Villeh. 113,6-7). 'After all he knows that you have given (all) yours (/all your goods)' c. Adonc atira li marquis son oirre ... (Clari 20,3) 'Then took the marquis his road d. Apres cele quinzaine vint li marquis Bonifaces de Monferrat (Villeh. 112,30-31). 'Behind these fifteen came the marquis Boniface de Monferrat'

These findings suggest that si occupies a position of its own not shared by other adverbials. We propose that si does not occupy a

Particles and sentence structure: a historical perspective

107

specifier position but heads its own phrase. To ascertain what the appropriate syntactical position of si is we compared the OF data with Celtic particles. Welsh has two different sets of particles, fe and mi which introduce affirmative main clauses and a and y which belong to a general focus strategy and are in complementary distribution with felmi (cf. Roberts 2000: 39). (6)

a.

b.

Bore'ma, mi glywes i' rnewyddion ary radio. Morning this PRT heard I the news on the radio 'This morning, I heard the news on the radio.' Ydynion a werthodd yd. the men PRT sold the dog. 'It's the men who have sold the dog.'

Celtic particles and the finite verb are always adjacent to each other and adverbs can only precede both. The only elements intervening between the particles and the finite verb are so called "infixed pronouns" (Roberts 2000:38). (7)

Mi' ch gwelais i. PRT you (pi.) saw I Ί saw you' (Roberts 2000:38).

In sentence (6a) the adverbial phrase bore'ma precedes the particle mi and the finite verb glywes, the subject i is postverbal and is followed by the object-DP r newyddion. This pattern is parallel to the one we found in OF particle sentences, such as (8). (8)

a.

Adonc si

manda li Dux tous les haus conseils de la vile. adv. PRT Vfm subj. obj. 'Thus the duke summoned all the municipal councils' (Clari 21,17).

108

Gisella Ferraresi and Maria Goldbach

b.

Et si vous metrons pro cinquante galies ä no coust. obj.pron. Vfm obj. 'And we will assign you 50 galleys at our cost' (Clari 21, 25-26).

In (8b) the clitic pronoun vous is inserted between the particle si and the finite verb metrons. No other type of object DP nor a subject pronoun can occur in this position. This mirrors the picture of the Welsh structure in (7) where we observe that the phonological host of the pronoun ch is the particle mi. In Old French it is generally the finite verb that is the phonological and the syntactic partner of the clitic, however, si may also fulfil this function. (9)

Sil [< si le] saluerentpar amur epar bien (Roland 121). '(They) welcome him amicably and seemly.'

The structural similarities between Old French and Welsh observed so far are represented in (10). (10)

XP - si/fe, mi - Clitics - Verbfin - Subj .-Pron. - Obj .-DP

We notice that the relation between the particle, the object clitic and finite verb in Old French is close to such a degree that no nonclitical element can intervene and that the whole complex is subject to specific phonological rules (below we will show that these items constitute one single prosodic unit). Turning now to the elements preceding si, we see that the initial XP can be the subject as in (11) or an adverb as in (12).' (11)

[La damoisele a qui tu as parle] si est li anemis (Queste 113,1; from Vance). 'The maiden to whom you have spoken si is the enemy.'

(12)

Adonc si manda li Dux tous les haus conseils de la vile (Clari 21,17). 'Thus the duke summoned all the municipal councils.'

Particles and sentence structure: a historical perspective

109

Firstly, subjects preceding si must be non-pronominal and must have definite, specific reference, as in (13). (13)

a.

b.

c.

[Cil de la ville de Jadres] ... si eurent molt grant peur (Clari 26,2-4). 'Those from Jadres had great fear.' [Cil qui en Blakie estoit fuis] si y fu si povres ... (Clari 31,20). 'The one who took refuge in Blakie (he) was so poor there...' [Jqffrois Ii mareschaus de Champaigne et Alarz Makeriaus] si s 'en alerent droit en France (Villeh. 102, 10-11) 'Joffrois li mareschaus de Champaigne and Alarz Makeriaus went straight to France'

Though subjects of definite specific reference also occur postverbally, indefinite subjects are excluded from the position preceding si. Instead, they surface in postverbal position. (14)

Apres si avint un jour que ... (Clari 36, 36). 'Then there came one day that....'

Subject pronouns are excluded from occurring in front of si (15a). Neither in Villehardouin nor in Clari we found them in this position. This means that despite of their definite specific reference they cannot be ahead of si.2 Nevertheless they may be topicalised as evidenced by (15b). (15)

a.

et si n'avons nous mie tous ceus nommes qui portoient banieres (Clari 18, 31-32) 'And we did not mention all those who bore banners.' b. Et il, par son sens et par son engin, ..., les mist en ce que ... (Villeh. 100, 36-37) 'And he because of his spirit and his ability manipulated them that...'

110

Gisella Ferraresi and Maria Goldbach

Secondly, the XP preceding si cannot be the direct or the indirect object. There seems to be a restriction against fronting of internal arguments. Again, if si is missing object DPs may be topicalised as in (16).

(16)

a.

b.

[Des paroles que li dux dist bones et belles] ne vos puis tout raconter (Villeh. 101,35-36) 'Of the good and nice words that the duke said I cannot tell you all' [Une autre partie] commanda Ii cuens de son avoir a retenir... (Villeh. 102, 8-9) 'The lord ordered to hold back another part of his goods'

To summarize: if si is present only subject DPs of definite specific reference and sentence adverbs may go before it. If si is not realised, also subject pronouns and direct and indirect object DPs may be topicalised. From this we conclude that there is a topic position in front of si where adverbs such as apres and subjects are placed. This position is not available for fronted internal arguments and for subject pronouns. Going back now to the comparison with Welsh, we see that here also the position preceding the particle is subject to some restriction. But in this case it is a restriction on focussed elements. Only one XP may be fronted over the verb and the corresponding particles are a (associated with fronted subjects, direct objects and VPs) and y (associated with all other XPs) (see Roberts 2000:39). (17)

a. b.

Y dynion a werthodd y ci. '(It's) the men PRT (who) have sold the dog.' Ym Mangory siaradais i llynedd. '(It was) in Bangor PRT I spoke last year.'

These findings suggest that the quantity and quality of the position preceding the particle is restricted in some way, presumably in a language-specific way.3 In the following we will propose a descriptive model of OF sentence structure. Parallel to Roberts' analysis for Celtic languages,

Particles and sentence structure: a historical perspective

111

we adopt for Old French the proposal from Rizzi 1997 where on the basis of data from Italian and English he proposes to split the CP into several distinct syntactic positions. Taking into consideration the distributional properties of si so far discussed we take the structure in (18) as a basis for future investigation. We hope eventually to motivate our description by specifying which kind of features are grouped into the respective categories and which kind of operations they perform. (18)

The Split-CP analysis (Rizzi 1997) ForceP Force TopP Top

FocP Foe

TopP Top

FinP Fin

TP

In Roberts' analysis, which follows Rizzi's 1997 system, Welsh particles are merged in Fin, and verbs do not move into the C-system but stay in AgrS. Now, let us see where OF subjects are. In contrast to Roberts 2000 we suppose that the finite verb has moved to Fin where si is merged. Si, object clitics, and the finite Verb form a complex head as a consequence of head movement. This way we model their intimate relationship since together these items constitute - as we will illustrate in the next section - a single prosodic constituent. The following structure summarises our proposal.4

112

Gisella Ferraresi and Maria Goldbach

(19)

si

TopP

ne

obj-cl root tense t

...

V'

V°

obj-DP

OF adverbs and subjects preceding si are in some kind of TopP, seemingly in different ones as subjects are governed by a reference constraint. What is it, then, that justifies si to be the head of Fin? We have noted that si occurs in neither finite nor infinite subordination (cf. ftn 3) If si really is the head of Fin - as we think it to be - how is Fin then realised in Old French subordinates, where there is no si? Let us have a look at an Old French subordinate construction. For its descriptive analysis we adopt Rizzi's (1997) proposal, according to which there are two positions for complementisers, Force and Fin. The following Old French construction hints at Force and Fin in the embedded clause not being realised syncretically but separately and

Particles and sentence structure: a historical perspective

113

at the same time. Examples like (20) can be found regularly in Old French texts. (20)

a.

b.

... et si leur disons [que] [s'il nous veulent rendre ces trente six mile marcs] ... [que] nous les metrons outre mer. (Clari 24,11-15) 'and si them tell we that if they us want to-give-back these 36000 marks that we them take to overseas (to North Africa).' Si avoient pourchacie unes lettres de Rome, [que] [trestout cil qui les guerroieroient ou qui leur feroient nul domage] [qu] 'il fuessent escommunie. (Clari 26, 13-15) 'Si have (they) letters sent to Rome that all those who would fight them or would cause them any damage that they would be excommunicated.'

In Rizzi's sense we assume the second occurrence of que in the sentence in (20) is in Fin, the first one, on the other hand, in Force. This construction gives a hint at why si does not occur in an embedded context: the embedded Fin head is reserved for que. As a further argument for the assumption that si is in the head of Fin we consider the fact that si regularly appears adjacent to other preceding particles (like lors 'now', or 'now', ensi 'so') and to other preceding (mostly temporal) scene-setting adverbial phrase. (21)

a.

[Quant Theodore Lascaris οϊ la novele] [/orc] si manda plus efforciement quanque il pot de gent, ... (Villeh. 194, 11-15). 'When Theodore Lascaris heard the news then (she) send by supreme effort as many people she could send' b. [Au matin] si fu li parlemenz en un vergier ä I 'abate madame Sainte Marie de Soissons. (Villeh. 103,36-38) 'In-the morning si there-was an assembly in a garden of the abbey of Our Lady St. Mary of Soissons'

We find, among others, the combinations puis si 'then si' (Clari 30, 34), lors si 'then si' (cf. (21a)), apres si 'after that si' (Clari 30,

114

Gisella Ferraresi and Maria Goldbach

35; 25, 36), lors apres si 'now after that si' (Villeh. 161,25-27), or si 'now si' (Villeh. 100,29), adonc si 'now si' (Clari 31,28-29; 35, 22), done si 'now si' (Clari 29, 39), puis apres si 'now after that si' (Clari 43, 38), apres quant-CP si 'after that when-CP Si' (Clari 48, 14-15) etc. (cf. Reenen and Schesler 2001 for a comprehensive list). We now assume that the scene-setting adverbial phrases and the particle-like adverbials (i.e. those that always trigger V2) puis, or, lors, ensi are distributed between two positions, Top and SpecFin. For a detailed semantic reason for scene-setting adverbials to be in a TopP we refer to Maienborn 2001. Thus we arrive at the following description. (22)

TopP FinP

scenesetting adverbials

___ Spec or puis lors ensi

Fin' Fin0 si ainz

We can, at the present state of our survey, only notice desciptively that SpecFin can also be occupied by quant-CPs, but that quant-CPs can also be in TopP. Now that we have briefly outlined the OF sentence structure we will come to the question of V2-effects in Old French. Given that si disappeared in the 17th century along with the loss of the V2phenomena we believe that these two incidents are underlyingly connected. We assume that si can be superficially absent in two different ways. In the first case si (respectively, any particle) is missing because Fin is not projected, hence V3-structures arise. In the second case si is realised in syntax, i.e. Fin is projected, but it disappears by a postsyntactic reduction process which belongs to PF (we deal with this PF-process in section 3). In the latter case the featural effects of Fin result - quasi epiphenomenal - in V2-structures. So we see that the featural material in Fin, of which si and or are

Particles and sentence structure: a historical perspective

115

exponents, restricts the realisation of the topic position: if Fin is realised by si a particular topic position (to be defined in future research) is not available. On the other hand if Fin is not projected this topic position is accessible for subject pronouns as well as for object DPs. The benefit of our approach is that we need not invoke CP-recursion in order to account for V3-structures in Old French. Accordingly we ascribe the following structure to V3-sentences. (23)

a.

b.

[TopP A Vendemain, auques matin], [Spec Lanselos] [T se lieve]] (...) (Tristan 63,17-18). 'The day after, in the morning, Lancelot gets up ...' [TopP Apres] [χρ [Spec la gent de la vile] [x alerent] au palais (...)] (Clari 33, 39-40). 'Then the people of the city went to the palace ...'

If Fin is not projected the head movement stops in Τ and V3configurations occur. (24) Top quant-CP se- CP maintenant apres gou

TP vP nominal or pronominal subjects

However, in our approach we cannot capture the fact that fronted direct and indirect objects produce apparent V2-effects. Remember the observation that topicalisation of internal arguments is bared if Fin is realised (mainly by si). Our account predicts that topicalised direct and indirect objects do only occur in V3-sentence since we argue that V2-phenomena result from PF-reduction of si and that the syntactic consequences of Fin-realisation take place prior to PF-

116

Gisella Ferraresi and Maria Goldbach

reduction. Clearly, this prediction is not empirically borne out in Old French. (25)

grant chose nos ont requise ... (Villeh. 100, 2). 'They asked us for big things ...'

For the time being we have no solution to this problem. But note that our proposal reduces the explanatory task to explaining the mechanisms of topicalisation. Our model does not need to reconcile OF V3-sentences with a V2-syntax. Beside V3 order also VI challenges the analysis of Old French as a V2-language (although in most cases the verb is not in absolute initial position but preceded by et 'and'). (26)

a.

Et sejournerent li pelerin en I'isle de Corfu ... (Clari 37,43-38,1). 'And the pilgrims stayed on the island of Corfu ...' b. Et chevauchierentpar lor jornees ... (Villeh. 18,4-6). 'And (they) rode in daylong marches ...' c. Conseillierent soi et parlerent ensemble cele nuit (Villeh. 100, 32-33) 'They discussed with each other and they talked together in this night' d. Querons lor qu'il le nos aident a conquerre (Villeh. 107, 33) 'We ask them that they help us to conquer him'

Our account treats these sentences as analogous to the V2structures: the particle is removed by PF-reduction and no Top-node is projected. Now that we have analysed the syntactic behaviour of OF si we discuss the aforementioned PF-reduction which caused its loss in early Modern French.

Particles and sentence structure: a historical perspective

3.

117

The loss of the particle si in Modern French

3.1. Prosodic Factors The particle si was lost during the 17th century and we can trace its demise back to prosodic and semantic factors. First, we consider the prosodic aspects. The theoretical basis of our proposal rests on Nespor and Vogel's (1986) prosodic hierarchy illustrated in (27). (27)

Syllable > Foot > Phonological Word (ω) > Clitic Group (C) > Phonological Phrase (Φ) > Intonational Phrase (I) > Phonological Utterance

In Old French the main word accent falls on the final syllable unless it contains a schwa. In this case the penultimate syllable gets main stress.5 Home (1990) has shown that in this period initial syllables of phonological words were weak. In open initial syllables ο and ο raised to u and ε and e reduced to schwa (cf. Home 1990: 6): (28)

ο, ο > u; ε, e > a Gallo-Roman > nepotem > debere > dolorem > moriri

Old French neveu devoir duleur murir

'nephew' 'to have to' 'sorrow' 'to die'

Despite the fact that the orthographic shape of the initial vowels in neveu and devoir is the letter e, their phonetic realisation is schwa. This is confirmed by transliterations of OF words in Hebrew characters, e.g. in a Vatican elegy from the end of the 13th century (cf. Darmesteter 1874). In this text, the Latin letter e in the initial unstressed syllable is transcribed as Hebrew 'sheva' [.], subscribed under the preceding consonant, while e in open stressed syllable appears in the Hebrew text as 'tsere' [ ] (Home 1990:6). The phonetic value of sheva is schwa [a], that of tsere is closed [e].

118

(29)

Gisella Ferraresi and Maria Goldbach

*sheva' {.} =: schwa [β], 'tsere' { } =: [e] I'apelet 'called him' -

, DN.1?

perir 'perish' - Τ . Ί . D Thus, the Hebrew manuscript makes a systematic distinction between the closed [e] and the initial syllable [a] which is not possible in the Latin spelling system since it does not have distinct characters to mark the different pronunciations. However, the Hebrew text sufficiently identifies the OF pronunciation of the first unstressed e. In contrast to OF phonological words, the initial element of a clitic group (into which phonological words are organized) had a secondary stress, hence it was not weak. This is indicated by the fact that clitics intervening between the initial member and the main stress exponent optionally undergo syncope. (30)

de le cor del cor 'of the horn' ne me vidrent nem vidrent 'they did not see me' jo le pert -> jolpert Ί lose it'

If there are two pretonic clitics, only the first clitic is reduced. (31)

ne le te dit nel te dit Ί do not tell it to you' se les te donet ses te donet 'if he gives them to you'

In this respect the OF clitic group mirrors the rule of syncope operative in the Late Latin phonological word (4th - 7th century) (cf. Home 1990:4). (32)

a.

One pretonic syllable Latin OF bönitäte > honte liberdre > livrer mänducäre > mangier lepordriu > levrier

'kindness' 'deliver' 'to eat' 'greyhound'

Particles and sentence structure: a historical perspective

b.

Two pretonic syllables OF Latin > sotement subitamente ancessor äntecessore > > auctoricäre > otreiier arbaleste ärcuballista

119

'suddenly' 'ancestor' 'to concede' 'crossbow'

In (32a) the syllable between the first one, which bears secondary stress, and that which bears main stress is reduced by syncope. If there are two syllables between the initial and the accented syllable as in (32b) only the first pretonic syllable is deleted. Thus by the OF period, Late Latin syncope operating on phonological words applied to the clitic group. Home draws attention to a further parallel between the Late Latin phonological word and the OF clitic group: as we have illustrated in (28) and (29) above, the secondary stress on the initial syllable extant in Late Latin phonological words had disappeared in OF phonological words. Similarly, the OF clitic group lost its initial secondary stress in Early Modern French. The reason for this development is the apocope of wordfinal schwa during the 16th century, which caused general oxytony not only at the level of the phonological word but also at the level of the clitic group and the phonological phrase (cf. Klausenburger 1970). That is, the right edge of these prosodic categories became strong and they became strictly right-headed.6 We think of this process in terms of scales which lose their equilibrium when more weight is thrown onto one side. Turning now to OF si, we conclude that it is organized into a clitic group with adjacent clitics. Evidence for this can be gleaned from the following examples. (33)

a. b.

[Sil saluerent c] par amur epar bien (Roland 121). '(They) welcome him amicably and seemly.' [Si I 'en dunat c] s 'espee e s 'escarbuncle (Roland 1531). '(He) gave him his sword and his carbuncle.'

Compare (32) with the clitic groups in (33). Clitics intervening between si and the main stress exponent (i.e. the finite verb) are reduced. If there are two clitics in between, the first one is reduced.

120

Gisella Ferraresi and Maria Goldbach

Since the initial element of the OF clitic group weakened by apocope of word final schwa in the 16th century, it lost its secondary accent and was subsequently subject to reduction processes. In this way, we can explain the following reduction processes showing up in Modern French. (34)

il [ne le veut c] pas > il le veutpas 'he doesn't want it' [ilya c ] . . . >y'a ... 'there is ...' [tu as vu c] ... > t'as vu ... 'you have seen ...'

(35)

Reduction Rule [[CL]w(eak) - [CL]W - ... - [

TENSE FUT

HAVE [CATRAUX]

left +

V [VFORM:EN]

TENSE FUT

will have

left

Given this analysis, neither have nor the ending -t of left 'means' ASPECT PERFECT in has left. Notice in particular that there is no mfeature [Aspect:Perfect] in the first place. A further point to note is that the perfect is expressed by a construction which has its own morphosyntactic properties. In particular, the auxiliary verb can express its own tense forms, and in the case of the future tense this is itself done periphrastically. For this reason, when we realize a given tense form of the perfect or progressive aspect or the passive voice we have to treat the auxiliary like any other finite verb and realize the appropriate TENSE s-feature on that auxiliary. This is represented directly in (6).

5.

Periphrastic paradigms

Implicit in what we've said so far is the claim that s-features such as as well as TENSE PAST are properties of individual (verb) lexemes. In other words, has left is the perfect form of the verb 'leave', in much the same way that left is the past tense form ASPECT PERFECT

Periphrastic paradigms in Bulgarian

257

and leaves is the 3sg form. Nonetheless, expressions such as has left have their own syntactic properties, too, which they share with other auxiliary constructions. Ackerman and Webelhuth (1998: 143) describe expressions such as has left as 'expanded predicates'. Their characterization of this notion is given in (8): (8)

Form and function of (expanded) predicate:

Function the contentive aspect of the predicate, i.e., its meaning and its function inventory

Form its categorial core, the auxiliaries and particles needed to express the predicate in the syntax

Thus, a phrase such as will have been making up (a story) can be thought of as the Future Perfect Progressive form of a lexeme 'invent, concoct' whose categorial core is the verb form make, whose full lexical form is a multi-word combination (lexical root + particle, make up) and whose contentive aspect is additionally expressed by the three auxiliaries. The state of affairs shown in (8) is represented directly in pedagogic and traditional grammars. However, the status of periphrastic constructions is less clear in many recent models of generative grammar, because there is generally no level at which to state the fact that the expression types in (8) form a paradigm. The paradigmatic organization of English auxiliaries is rather complex so to illustrate the problem I'll consider the slightly less controversial case of Russian. In the Russian verb system it is common to distinguish a three tense system (Past, Present, Future) with two aspects (Perfective, Imperfective). These are expressed as in (9): (9)

Basic paradigm system for RASPISAT' 'to write out'

Past Present Future

Perfective raspisal — raspiset

Imperfective raspisyval raspisyvaet budet raspisyvat'

258

Andrew Spencer

In the Future Tense forms the single word form in Perfective aspect, raspiset, is in opposition to the Imperfective Future, a periphrastic construction formed from the Future Tense form of the verb 'to be' and the (Imperfective) infinitive. The Imperfective Future likewise stands in paradigmatic opposition to the single word form Present and Past tense forms of the Imperfective. Notice that there is no Present tense form of the Perfective aspect in Russian. Given this, a definition of the set of s-features for Russian will therefore contain the declarations given in (10): (10) Partial statement of s-feature paradigm for Russian a.

TENSE {PAST, PRESENT, FUTURE} ASPECT {PERFECTIVE, IMPERFECTIVE}

b.

TENSE PRESENT => ASPECT IMPERFECTIVE

The feature cooccurrence restriction given in (10b) ensures that the paradigm is not complete, in that the 'virtual' cell corresponding to TENSE PRESENT ASPECT PERFECTIVE can never be filled. In addition to definining the s-feature paradigm space, a grammar must specify a mapping which defines how each cell is filled. Where the paradigms are realized by words formed by an agglutinating morphology or by compositional syntactic constructions with function words the mapping will be fairly trivial. However, in the general case we will find deviations from a one:one mapping (Ackerman and Webelhuth 1998, Sells 2000, Sadler and Spencer 2001). The point of paradigm-based models of morphosyntax is to develop formal machinery for expressing such mappings and integrating them into the rest of the morphology and syntax. The paradigm-based perspective on periphrastic constructions leaves open an important question: (11)

How do we distinguish paradigmatic periphrasis (expanded predicates) from 'ordinary' syntactic constructions involving two separate lexemes?

Periphrastic paradigms in Bulgarian

259

In other words, how do we know that the English expression wants to leave is an infinitival complement of LEAVE subordinated to the main verb WANT and isn't the desiderative mood form of the lexeme LEAVE? This amounts to the problem of grammaticalization: how do we know when grammaticalization has taken place (and been completed)? Equivalently, from the synchronic point of view, (11) is equivalent to (12): (12)

How do we individuate grammatical features (as opposed to lexical meanings)?

In other words how do we know that we are dealing with, say, a feature such as TENSE FUTURE or MOOD INTENTIONAL rather than a verb lexeme DESIRE(x, y)? Now, (12) is a question which any theory of grammar, in any theoretical framework, needs to provide an explicit answer to. However, as far as I know this is a question which is almost uniformly ignored. Clearly, however, this question is crucial to all current accounts of morphosyntax, and certainly isn't peculiar to paradigm-driven models such as that which I am proposing. However, the paradigm-based approach does have the virtue of throwing the problem into relief and making it less easy to sweep it under the carpet. In the rest of this paper I address the questions in (11, 12) by proposing a set of diagnostics for determining when we have a paradigmatic system. The thrust of the discussion will be the observation that periphrastic paradigms tend to have certain properties which make them look more like morphological paradigms than syntactic constructions.

260

6.

Andrew Spencer

Properties of periphrastic paradigms

6.1. Quasi-morphological behaviour In Spencer (2001) I outline a variety of ways in which periphrastic constructions in Slavic languages resemble morphological systems, in that they exhibit cumulation, zero exponence, meaningless morphs and so on. I argue there that we can capture the regularities adequately only if we make use of sets of rules or constraints mapping sfeature complexes to syntactic constructions organized in the manner of realization rules advocated by morphologists such as Stump (2001). In this section I extend those observations by highlighting two important properties of periphrastic paradigms which distinguish them from the 'syntactic ideal' of common-or-garden compositional and regular constructions.

6.2. Incomplete paradigms An underexhaustive periphrastic paradigm is the most obvious deviation from the 'syntactic ideal', as illustrated by the gap in the Russian verb paradigm shown in (9). There is no Present Perfective form in Russian, yet there is no particular reason, morphological, syntactic or semantic why Russian couldn't have developed the pseudoparadigm in (13): (13)

Exhaustive pseudo-paradigm system for RASPISAT' 'to write out'

Past Present Future

Perfective raspisal raspiset budet raspisat'

Imperfective raspisyval raspisyvaet budet raspisyvat'

It might be thought that there are good semantic reasons for disallowing the exhaustive paradigm, in that the semantics of perfectivity seem to be incompatible with Present tense meaning. However,

Periphrastic paradigms in Bulgarian

261

this would be a mistake (borne in part of taking traditional semantically based morphological labels too seriously). Bulgarian, a Slavic language with a similar system of verbal aspect, is quite happy to allow Present tense and Perfective aspect to cooccur (and even permits an Imperfect tense of the Perfective aspect and an Aorist tense of the Imperfective aspect). Thus, the reason why Russian has the underexhaustive system in (9) and not the exhaustive system in (13) is effectively an accident of linguistic history.

6.3. Superexhaustive paradigms The second type of deviation is one which seems to have escaped theoretical notice hitherto, but which provides strong motivation for distinguishing between compositional, productive syntax and grammaticalized paradigmatic periphrasis. Just as there are 'underexhaustive' paradigms, so, I shall argue, there are cases in which a paradigm has more forms than would be expected from the normal principles of compositional syntax. I shall call such paradigms 'superexhaustive'. They are systems in which the paradigm develops forms which are mandated neither by existing paradigmatic oppositions nor by the 'inherent' syntax of the periphrasis. Such phenomena pose an interesting puzzle for grammaticalization theory, but they clearly pose serious problems for a syntactic, non-paradigmatic account of periphrasis. Superexhaustive paradigms are not particularly common, but Bulgarian provides a particularly interesting example of the phenomenon, the Emphatic Renarrated mood. To understand how this works we need to look at Bulgarian morphosyntax in more detail. Bulgarian distinguishes three tenses morphologically, the Present, Imperfect and Aorist. In addition, it has a set of compound Perfect tenses, formed by combining tensed forms of the verb 'be' with the /participle. The /-participle is formed by adding - I to the verb stem, thus from the (Aorist) stem pisa 'write' we obtain pisal. This participle agrees with its subject in number and gender but not in person. The Perfect can appear in various tense forms, including the Present

262

Andrew Spencer

Perfect and Past Perfect or Pluperfect: säm pisal Ί have written', bjax pisal Ί had written'. The latter is formed by putting the auxiliary 'be' into the Imperfect or the Aorist tense (these tenses of 'be' only differ in the 2/3 sg forms). Bulgarian has developed a special Renarrated mood by reinterpreting the Perfect tense series. This is a type of evidential mood indicating that the speaker didn't witness the event described but learnt of it by hearsay ('They say that', Ί gather that', 'Apparently'). In (14) I give the basic forms for the Renarrated mood in 3sg forms (ignoring Future tense based forms, which we will return to). (14)

Bulgarian Renarrated mood (preizkazno naklonenie), nonfuture forms (Scatton 1984: 331), Pi§E 'write' Indicative Renarrated

Present

pise 'writes' pisel

Imperfect

pisese 'was writing'

Aorist

pisa 'wrote'

Present Perfect

pisal e 'has written'

pisal

bil pisal Past Perfect

be(se) pisal 'had written'

First notice that the Present/Past distinction is neutralized in the Renarrated mood, except that the Aorist has its own unique Renarrated form. This form is identical to the Present Perfect Indicative in lst/2nd person forms (an instance of the common paradigm property of syncretism) and consists of the Present tense of the auxiliary 'be' and the /-participle form of the verb, pisal. However, the Perfect Indicative retains the auxiliary in 3rd person forms, while in the 3rd person Renarrated forms the auxiliary is dropped (an intriguing example of zero exponence discussed in more detail in Spencer 2001: 294-95). The same /-participle form, pisal, is the basis of the Pres-

Periphrastic paradigms in Bulgarian

263

ent/Past Perfect Renarrated. However, the /-participle for the Present/Imperfect Renarrated is slightly different, pis el. This is the Imperfect /-participle (minalo nesvarSeno dejatelno pricastie), based on the Present/Imperfect stem and is found only in the Renarrated forms, not in Perfect Indicative forms. This morphological innovation is necessary to ensure that a complete paradigm of Renarrated forms is possible, otherwise there would only be a single form for the Present, Imperfect and Aorist tenses. The innovation represents an extension to the normal morphological paradigm of verb forms, and is an instance of what I shall call 'morphological superexhaustivity'. As a result the verb paradigm is now skewed, because it includes a form which can't be used for any Indicative tense/aspect constructions. Of greater interest is an extension of the Renarrated mood system. Bulgarian has a paradigm of Emphatic Renarrated forms. The basic scheme is shown in (15): (15)

Bulgarian Emphatic Renarrated forms, lsg (Scatton 1984: 33If), PISE 'write' Renarrated

Emphatic Renarrated

Present/Imperfect

säm pisel

bil säm pisel

Aorist

säm pisal

bil säm pisal

Present Perfect

bil pisal

—

The Emphatic Renarrated forms are derived systematically from the Renarrated forms by creating an additional Renarrated form for the Present tense auxiliary forms (shown here in the lsg form). This creates a construction which is otherwise unlicensed. In some Slavic languages this would be the way to create a Pluperfect from a Present Perfect (because the expression bil säm would be the Past tense of the auxiliary). However, the Pluperfect in Bulgarian is formed by taking a synthetic past tense of the auxiliary (Imperfect or Aorist).

264

Andrew Spencer

Outside of the system of Renarration, a form such as bil säm pisal could only be the Perfect of a Perfect, and this is no more motivated for Bulgarian than the corresponding English expression would be: *has had written (the letter). As it is we seem to have a Renarrated form of a Renarrated form, but this too is semantic nonsense. In effect, we have here a non-compositional extension of a construction which is already pretty non-compositional. The pressure to express further grammatical meanings has led to an extension of the periphrastic paradigm in a way that is very difficult to reconcile with a picture on which the auxiliary verb has its own lexically defined set of grammatical features. There seems to be no alternative but to treat this as a constructional idiom which has been extended in the way that synthetic morphological paradigms are occasionally extended. I shall call this type of phenomenon 'periphrastic superexhaustivity'. Some examples are given below (where 'L' in morphemic glosses indicates 7-participle'): (16) a.

Aorist Renarrated: Ti si napisala pismoto you are write.L the.letter

'You wrote the letter (reportedly)' b.

(17) a.

Emphatic Aorist Renarrated: Ti si bila napisala pismoto you are be.L write.L the.letter 'You wrote the letter (reportedly, emphatic)' Present/Imperfect Renarrated: Ti si pisela pismoto you are write.iMPF.L the.letter

'You are/were writing the letter (reportedly)' b.

Present/Imperfect Emphatic Renarrated: Ti si bila pisela pismoto you are be.L write.iMPF.L the.letter 'You are/were writing the letter (reportedly, emphatic)'

Periphrastic paradigms in Bulgarian

265

Bulgarian linguists often distinguish a further modal category called variously the 'Presumptive' (predpolozitelna forma), 'Inferential' (umozakljucitelna forma), 'Conclusive' (konkluziv) and other terms (Bojadziev, Kucarov and Pencev 1998: 401, 410—413). This is identical to the Renarrated mood, except that the 3rd person auxiliary is always present. This means that the forms of the Aorist Conclusive coincide with those of the Present Perfect. The one form which is therefore unique to this mood is the 3rd person forms with the Imperfect /-participle, such as e pis el 'was writing (it seems)' (see Kucarov 1994 for a detailed discussion of this mood). According to Bojadziev et al. (1998: 423f) the Emphatic or Doubled Renarrated form is in fact the Conclusive of the Renarrated form (konkluzivna preizkazna forma). Thus we can set up the following correspondences, in which the Emphatic Renarrated is derived by 'renarrating' the Conclusive forms, that is, by replacing the 'indicative form' auxiliary of the Conclusive with a 'renarrated' form. Compare their table shown in (18) with the table in (15) above: (18) Conclusive and Emphatic Renarrated forms Conclusive

Emphatic Renarrated

pisel e bil e pisal stjal e da pise stjal e da e pisal pisal e

bil pisel no renarrated form stjal bil da pise stjal bil da e pisal bil pisal

If this is true then we may ask (with an anonymous referee) whether the Emphatic Renarrated paradigm really does constitute an instance of superexhaustivity, as opposed to just the operation of regular formation rules. The answer remains 'yes'. Now, the meaning of the Emphatic Renarrated is itself a matter of controversy, but what it cannot possibly mean is the Renarration of the Conclusive or the Conclusive of the Renarrated. In other words, bil pisel can't mean 'it is said that the speaker infers that he wrote' or 'the speaker infers

266

Andrew Spencer

that it is said that he wrote'. But the fact that it's possible to relate the Emphatic Renarrated to two existing morphosyntactic processes at the purely formal level without preservation of the conventional semantics for those two processes is precisely what gives rise to superexhaustivity. The Emphatic Renarrated form, whatever its meaning and however it is formed, is non-compositional. The discussion in Bojadziev et al. (1998) highlights this aspect of the construction very clearly. The Emphatic Renarrated paradigm, therefore, is mandated purely by morphological form, not by the meaning normally associated with its components. Superexhaustive paradigms provide no less evidence for grammaticalization and paradigmatic organization than underexhaustive paradigms. In each case we are dealing with a deviation from syntactic compositionality which is entirely unmotivated from the point of view of syntactic representations. In the next section we look at Future tense forms in Bulgarian, to examine the full extent to which syntax can be commandeered in the service of functional categories.

7.

Grammaticalization of clause structure in Bulgarian

7.1. Bulgarian subordinate clauses Before we can consider Bulgarian Future forms we need to know something about Bulgarian complement clauses. Bulgarian lacks infinitive forms. Where many other Slavic languages have infinitival complements, Bulgarian has a variety of modal and other verbs which take finite subordinate clause complements introduced by DA. An interesting peculiarity of these clauses is that they permit fronting of arguments and adjuncts before the subordinator. (19) a.

Ivan iska pisma da Ivan wants letters DA 'Ivan wants to write letters'

pise writes

Periphrastic paradigms in Bulgarian

Ivan iska Marija da pise Ivan wants Marija DA writes 'Ivan wants Marija to write letters' (20) a.

Ivan iska Ivan wants otgovarja answers

267

pisma letters

da DA

pravilno correctly

na vaprosite to the.questions

pravilno da Ivan iska Ivan wants correctly DA na vaprosite otgovarja to the.questions answers 'Ivan wants to answer the questions correctly' The DA-clause starts its own clitic domain (clitics are shown in bold): (21) Ne sie Ii vie iskali da mu gi pokaza? NEG AUX Q you want DA to.him them I.show 'Don't you (reportedly) want me to show them to him?' 'Haven't you been wanting me to show them to him?' Example (21) also illustrates the basic way in which negation is effected in Bulgarian, by means of a negative particle, NE, which then heads the clitic cluster. If there are no clitics, NE attaches to the first full verb form (the lexical verb or non-clitic forms of the 'be' auxiliary). There is controversy in the literature as to how best to describe the DA-element. In particular, it isn't clear whether we should regard DA as a complementizer in all its uses or whether it is really a modal particle of some kind in some of its uses (see Rudin 1986: 54f for a survey of views). This question is irrelevant to our present concerns. What is crucial is that the DA-element has a wide variety of morphosyntactic uses and some, at least, of its morphosyntactic properties are preserved across those uses.

268

Andrew Spencer

7.2. Grammaticalization of OA-clauses - Imperatives and Futures The DA-clause is used to realize certain types of imperative (Scatton 1984: 339). In examples (22b) and (23) we see that the subject can come before the DA-element, as we saw in (19b): (22) a.

Da dojde Ivan DA come Ivan

b.

Ivan da dojde Ivan DA come 'Ivan should come!'

(23) a.

Neka Ivan da let Ivan DA 'Ivan should come'

b.

dojde come

Neka Ivan da ne let Ivan DA NEG 'Ivan shouldn't come'

idva come.iMPFV

These examples show that, in this type of imperative at least, the DA-clause construction (viewed as a piece of 'pure' syntax) can sometimes serve as an exponent (or even the principal exponent) of a feature MOOD IMPERATIVE. The Imperative is not the only category to use the DA-clause. This is also the structure of two of the four compound tenses based on the Future (see Scatton 1984: 320 and Bojadziev et al. 1998: 382 for a basic description of the tense/aspect system). The basic Future tense construction (bädeäte) is relatively simple. It is formed by taking an uninflecting particle ste and combining this (as a clitic) with the Present tense form of the verb, either in the simple aspect or in the Perfect (bädeste predvaritelno). Thus, we have (24,25): (24) a.

Az ste pisa I FUT write. lSG Ί will write the letter'

pismoto the.letter

Periphrastic paradigms in Bulgarian

(25) a.

Te ste pisat they FUT write.3PL 'They will write the letter'

pismoto the.letter

Az

ste

pismoto

I

FUT be.LSG write.L.SG

säm

napisal

269

the.letter

Ί will have written the letter' b.

Te ste sa napisali they FUT be.3PL write.L.PL 'They will have written the letter'

pismoto the.letter

The prefixed forms napisal/napisali represent the Perfective aspect, not to be confused with the Perfect tense/aspect. Perfective aspect is more natural than Imperfective aspect in Perfect tense forms such as these. The verb 'be' has two equivalent Future tense forms, one using the ste + the Present tense of 'be' and the other with a special form of 'be' based on the root bäd-, which was historically a Future tense form. The Perfect aspect forms in (25) can be based on either type of Future, e.g. Az ste bäda napisal pismoto, Te ste bädat napisali pismoto. A number of other tense/mood/aspect expressions are possible based on this design, but all of them involve a DA-clause. Traditionally, these are all taken as part of the verbal paradigm. First, in the Indicative mood there are the Past Future (bädeste ν minaloto) and the Past Future Perfect (bädeste predvaritelno ν minaloto). In the Past Future the future auxiliary takes the Person/Number inflections of the Imperfect (past) tense, and which I shall gloss as §TA (the traditional lsg citation form). This then combines with a DA-clause whose lexical verb is either in the Present tense (Past Future, (26)) or Present Perfect (Past Future Perfect, (27)): (26) a.

Az stjax da I STA.ISG DA Ί would write the letter'

pisa write. ISG

pismoto the.letter

270

Andrew Spencer

b.

(27) a.

b.

Te stjaxa da pisat they STA.3PL DA write.3PL 'They would write the letter' Az

stjax

I

STA.ISG

da

säm

DA

be.LSG

napisal

pismoto

write.L.SG

the.letter

pismoto the.letter

Ί would have written the letter' Te stjaxa da sa they

§TA.3PL

DA

napisali

pismoto

write.L.PL

the.letter

be.3PL

'They would have written the letter' The commonest interpretation of the Past Future is, perhaps, that of a Conditional (as seen in the glosses for (26,27)). Next I consider the Negative Future forms. These are constructed with the verb IMA 'have'. Now, as a lexical verb IMA has a special fused negated form as seen in examples (28,29): (28) a.

Toj

ima

pari

he

have.3SG.PRES

money

'He has money' b.

(29) a.

b.

Toj

njama

pari

he NEG.have.3sG.PRES ' H e doesn't have any money'

money

Te imaxa they have.3PL.IMPF 'They had/used to have money'

pari money

Te njamaxa pari they NEG.have.3PL.LMPF money 'They didn't (used to) have money'

Periphrastic paradigms in Bulgarian

271

The verb IMA is also used impersonally (in 3sg forms) in existential sentences of the kind 'there is/are X': (30) a.

b.

Tuk ima here have.3sG 'There's money here'

pari money.PL

Tuk

njama

pari

here

NEG.have.3SG

money.PL

'There's no money here' Notice that PARI 'money' is a plurale tantum noun, but it doesn't trigger agreement on the verb. The examples in (28-30) show that the fused root njam- is used uniformly instead of the expected ne + ima (which is ungrammatical). Exactly the same idiosyncratic allomorphy is found when IMA is used to form the negated future tenses. In the Present tense, the impersonal 3sg form, njama (glossed as NJAMA), is used for all persons followed by a DA-clause in the Present or Present Perfect: (31) a.

Az

njama

da

I

NJAMA

DA

pisa write. ISG

pismoto the.letter

Te njama da pisat they NJAMA DA write.3PL 'They will not write the letter'

pismoto the.letter

Ί will not write the letter'

(32) a.

Az

njama

da

säm

I

NJAMA

DA

be.LSG

napisal

pismoto

write.L.SG

the.letter

Ί will not have written the letter'

272

Andrew Spencer

b.

Te they

njama

da

sa

NJAMA

DA

be.3PL

napisali

pismoto

write.L.PL

the.letter

'They will not have written the letter' For the Past Future we use the 3sg Negative Imperfect form of njamase (glossed NJAMASE), with the DA-clause:

IMA,

(33) a.

pisa write. ISG

pismoto the.letter

Te njamase da pisat they NJAMASE DA write.3PL 'They would not write the letter'

pismoto the.letter

Az

njamase

da

I

NJAMASE

DA

Ί would not write the letter'

(34) a.

Az

njamase

I

NJAMASE

da

säm

DA

b e . ISG

napisal

pismoto

write.L.SG

the.letter

Ί would not have written the letter' Te

njamase

da

sa

they

NJAMASE

DA

be.3PL

napisali

pismoto write.L.PL the.letter 'They would not have written the letter' Finally, we can form Renarrated and Emphatic Renarrated Futures, based on the /-participles of §TA 'Future', NJAMA 'Negative Future' and SÄM 'be', stjal, njamal, and bil: (35) a.

Tja

stjala

she

STJAL.FEM.SG

da

e

napisala

pismoto

DA be.3sG write the.letter 'She will/would have written the letter (reportedly)'

Periphrastic paradigms in Bulgarian

Tja

bila

stjala

she

BIL.FEM.SG

STJAL.FEM.SG

273

da e napisala pismoto DA be.3sG write the.letter 'She will/would have written the letter (reportedly, emphatic)' (36) a.

Tja

njamalo

She

NJAMAL.NEUT.SG

da e napisala pismoto DA be.3SG write the.letter 'She will/would not have written the letter (reportedly)' Tja

njamalo

bilo

She

NJAMAL.NEUT.SG

BIL.NEUT.SG

da e napisala pismoto DA be.3sG write the.letter 'She will/would not have written the letter (reportedly, emphatic)' According to Scatton (1984: 332) and Bojadziev et al. (1998: 460—462) there are also forms with ne instead of njamalo, which would give Tja ne stjala da e napisala pismoto for (36a) and Tja ne bila stjala da e napisala pismoto for (36b). In those alternative forms it appears that we are negating the 'be' auxiliary and not the ste form, though this is only visible in non-3rd person forms: Az ne säm stjal da säm napisal pismoto Ί would not have written the letter (reportedly)'. The DA-clause in these compound Future tenses is syntactically still a DA-clause. Thus, it creates its own clitic domain, so that those clitics which relate to the lexical verb (pronominal object clitics and the perfect auxiliary) are found in that clause, while 'main clause' clitics, that is the interrogative LI and perfect auxiliaries relating to Renarrated forms are found outside the DA-clause, just as though the Future auxiliary were a matrix verb selecting a subordinate clause complement. This is seen in examples such as the following (Avgustinova 1997: 74):

274

Andrew Spencer

(37) Stjaxte

Ii

mi PAST.FUT.2PL Q DA to.me 'Would you give them to me?'

(38) Njamase NJAMASE

da

gi them

li

da

ste

Q

DA

AUX.2PL

dadete? you. give

dali? mi gi give.L them to.me 'Would you not have given them to me?' (39) Steli STEL.PL

li

ste

(bili)

Q

AUX.2PL

(AUX.L)

da

mi gi badete prestavili? to.me them FUT.2PL introduced 'Were you (reportedly) going to introduce them to me?'

DA

(40) Njamalo NJAMAL.NEUT

da

ste

DA

AUX.2PL

li

(bilo)

Q

(AUX.L)

mi bili predstaveni? to.me AUX.L introduced 'Weren't you (reportedly) going to be introduced to me?'

Similarly, elements such as objects and adverbials can climb out of the DA-clause and appear in a preposed position between the auxiliary complex and the DA complementizer (Avgustinova 1997: 51): (41) Njamase NJAMASE

li

pravilno

da

e

Q

correctly

DA

be.3SG

otgovoril

na väprosite?

reply

to the.questions

'Would he not have answered the questions correctly?'

Periphrastic paradigms in Bulgarian

(42) Ti you da

stjal

li

275

si

predvaritelno STJAL Q AUX.2SG in.advance si ni ja pokazal? DA AUX.2SG to.us her show 'Will you (reportedly) have shown her to us in advance?'

7.3. Theoretical implications Clearly, the data in the previous section illustrate periphrastic paradigms in the sense that I have been using this term. A reasonably full range of constructions is found (even if some of the more baroque ones are very seldom used) and the forms of the more complex structures are largely predictable from those of the simpler structures from which they are built up. Thus, we have a situation in which a DA-clause construction is used as part of a verbal paradigm, while retaining most of the syntax of the original syntactic construction. This is theoretically significant. The Future constructions with DA illustrate in a particularly graphic fashion the way in which syntax can lose its compositionality and can serve as the (partial) exponent of morphosyntactic features organized paradigmatically. I propose to account for these paradigms by setting up a family of s-feature mappings similar to those proposed in (6) for the English Perfect. These are shown in (43) (with obvious abbreviations): (43) Where ' V' stands for any lexical verb, and where SÄM stands for the auxiliary verb ('be'): A.

[V, ASP PERF, TENSE PRES] SÄM TENSE PRES

B.

+

=> V [VFORM:L]

[V, ASP PERF, TENSE PAST] SÄM TENSE IMPF

+

=> V

[Vform:l]

276

Andrew Spencer C.

[V, TENSE FUT]

Ste

+

V TENSE PRES

d.

[V, ASP PERF, TENSE PAST ]

ste

+

V ASP PERF TENSE PRES

e.

[V, TENSE PAST_FUT]

sta + da

+

TENSE IMPF f.

V TENSE PRES

[V, ASP PERF, TENSE PAST_FUT]

sta + da

+

TENSE IMPF

V ASP PERF TENSE PRES

(44) Where Ύ ' stands for any lexical verb: a.

[V, TENSE FUT, POL NEG]

njama + da

+

V TENSE PRES

b.

[V, ASP PERF, TENSE FUT, POL NEG]

njama + da

+

V ASP PERF TENSE PRES

C.

[V, TENSE PAST_FUT, POL NEG]

njamase + da

+

V TENSE PRES

Periphrastic paradigms in Bulgarian

d.

[ V , ASP PERF, TENSE PAST_FUT, POL NEG] njamase

+ da

+

277

=t>

V ASP PERF TENSE PRES

Obviously, these rules would have to be supplemented by further principles regulating agreement and so on. This is an interesting exercise in itself, since agreement for person, number, gender can be spread across the entire construction. The four rules for negated forms will override the default negation rule which just adds NE to the left edge of the verb or clitic cluster. There are various notation conventions one could imagine to collapse these two sets of rules into a single rule schema. For example, we may wish to write (44) as (45): (45)

V TENSE ASP

=>

POL NEG njama

+

da

+

V

TENSE

ASP

PER/NUM 3 SG

TENSE PRES

Similarly, one can expand these rules in obvious ways to accommodate the Renarrated and Emphatic Renarrated forms. Crucial to this way of looking at things is the assumption that formatives such as sta and njama are grammatical elements serving as exponents of mapping rules, and not fully fledged lexical entries. The syntactic idiom chunks illustrated in (43, 44) are self-contained morphosyntactic units which serve to realize morphosyntactic properties, they are not inflected forms of lexical entries. Thus, none of the forms of §TE or STA is treated as a lexical entry bearing the feature TENSE FUTURE (much less [Tense:Future]!) Even njama isn't treated as a listed negation element. Rather, we have a variety of grammatical formatives realizing morphosyntactic properties. It's

278

Andrew Spencer

just that some of those formatives happen to be inflecting words in their own right. To be sure, one could imagine an analysis in which sta, njama and the like were treated as verbs in the lexicon bearing a set of features (e.g. [Tense:Future], [Polarity:Negation]) and having selectional features (e.g. \+da\). The syntax of the compound Futures would then be essentially that of any lexical verb which selects a DA-clause. Indeed, Krapova (1999) has recently argued for precisely this position. She argues that STA, when conjugated, is a lexical modal verb like ISKAM 'want'. However, her argument rests on a faulty characterization of the semantics of STA. She claims that STA always assigns its subject a volitional/intentional semantic role, much like ISKAM (Krapova 1999: 82). This is simply false. Casual inspection of these constructions shows that the traditional characterization as Future-inthe-Past (shifting to Conditional) is correct. Krapova explicitly states that it is more 'principled' to relate all tokens of a grammatical formative such as English HAVE or Bulgarian IMAM 'have' to a single underlying meaning. In other words, she denies the possibility of homonymy or polysemy. Given Krapova's assumptions we would have to concede that the lexemes EAR (for hearing) and EAR (of corn) were one and the same (underspecified) lexeme. A particularly telling observation is the fact that the negative form of the conjugated STA constructions is formed with NJAMA, just like the straightforward Future Indicative with the particle STE (which Krapova agrees is simply a functional element and not a lexical verb). But by Krapova's 'No homonymy' principle njama must be a form of IMA and not of STA (or STE). Thus, she would be forced to argue that there simply is no Negative Future form in Bulgarian. Another important and related point which is overlooked by Krapova is the fact that the grammatical formatives do not have full paradigms of their own. Since STA even when conjugated is itself the exponent of TENSE PAST FUTURE it cannot have a Future or a Past Future form of its own. If sta, njama and so on really did represent forms of autonomous lexical entries it would be a complete mystery why they lack precisely the inflections which they themselves are exponents of. Thus, why is there no Future form of either sta or

Periphrastic paradigms in Bulgarian

279

njama, along the lines of *ste sta or *ste njamal It is no more possible to construct such strings than it is to have a Perfect aspect form of the Perfect auxiliary HAVE: *I have had left early. Those that believe that auxiliaries such as English HAVE or Bulgarian §TA are fully fledged lexical entries have to explain this otherwise mysterious complementary distribution. Notice that homophonous verbs do have Perfect forms, e.g. I have had an idea or I have had to leave early. It is difficult to see how such contrasts can be explained without voiding the 'No homophony' principle of any content. Clearly we must reject Krapova's bizarre 'No homonymy' assumption out of hand, but her discussion is useful as a reductio ad absurdum of the strongest version of the view that functional words have their own lexical entries. Is there a satisfactory weakening of the lexical entry thesis? No doubt it is possible to set up special lexical entries, with a plethora of selectional features guaranteeing that just the right collocations are generated. It is easy to see that such entries would still retain the mysterious property of complementary distribution pointed out above: these would be lexical entries which themselves lacked those forms which correspond to the features of which they are exponents.

8.

Conclusions

I have argued for a paradigmatic perspective on periphrastic constructions. On this approach when we consider multi-word combinations such as English has been writing or Bulgarian njamase da e pisal '(he) would not have written' we do not regard individual function words such as has been or njamase da e as lexical entries projecting their own set of features. Instead, we regard them as simply formatives which bear at most syntactic category features. I presented two sorts of evidence for this paradigm-based view of periphrases, based mainly on the unusually rich periphrastic system of Bulgarian: first, we find the kinds of gaps in these constructions that we often see in inflectional paradigms but which should not occur in genuinely compositional syntax; second, we find instances of super-

280

Andrew Spencer

exhaustivity, in which the paradigm takes on a life of its own, so to speak, and extends beyond what would be expected from the normal combinatoric syntax. Constructions such as the Bulgarian Emphatic Renarrated are extremely difficult to describe if we insist on listing featural properties in lexical entries for function words. What conceivable lexical entry could we posit for bil, the /-participle of 'be', that would account for its use as an exponent of the Present/Aorist Emphatic Renarrated form bil säm cetjal Ί am reading/read (reportedly, emphatic)', while simultaneously accounting for its appearance as the marker of Perfect Renarrated in bil säm napisal Ί have written (reportedly)' or Present Perfect Indicative and Present/Aorist Renarrated of bil säm Ί have been, I am (reportedly), I was (reportedly)'? There remain many interesting questions. In particular, it is necessary to explain exactly how paradigm-driven mapping rules such as those of (43, 44) relate to other aspects of morphosyntax, such as linearization, clitic placement, agreement, ellipsis and so on. These, of course, are problems for all current theories of morphosyntax. I claim that it is only by adopting the paradigm-based approach that the correct factoring of functions can be achieved and only a paradigm-based approach will eventually lead to an insightful account of these constructions.

Notes 1. Parts of this paper have been presented to audiences at the Workshop on Constructions, Linguistics Association of Great Britain, 5 April 2001, University of Leeds, and the Workshop on Historical Morphosyntax, 6 June 2001, Universität Konstanz, as well as to the Arbeitsgruppe 12 of the DGfS23 meeting, Universität Leipzig, 1 March 2001.1 am grateful to Guergana Popova, the editors, and an anonymous reviewer for helpful comments. 2. Except where it would be fussy to do so, I write forms of lexemes in italics and the name of the lexeme itself in SMALL CAPITALS. 3. Naturally, this account presupposes a theory of constructions within the grammatical architecture. For preliminary discussion of this see Ackerman and Webelhuth 1998, Sells 2000.

Periphrastic paradigms in Bulgarian

281

References Ackerman, Farrell and Webelhuth, Gert 1998 A Theory of Predicates. Stanford University: Center for the Study of Language and Information. Aronoff, Mark 1994 Morphology By Itself. Cambridge, MA: MIT Press. Avgustinova, Tania 1997 Word Order and Clitics in Bulgarian. (Saarbrücken Dissertations in Computational Linguistics and Language Technology, Volume 5.) Saarbrücken: University of the Saarland. Bojadziev, Todor, Ivan Kucarov and Jordan Penöev 1998 Sävremenen bälgarski ezik [Modern Bulgarian], Sofija: Izdatelska kääta 'Petar Beron'. Chomsky, Noam 1981 Lectures on Government and Binding. Dordrecht: Foris. Goldberg, Adele 1995 Constructions. A Construction Grammar Approach to Argument Structure. Chicago: Chicago University Press. Kibrik, Aleksandr E., Sandro V. Kodzasov, and Irina A Muravöva 2000 Jazyk ifol'klor aljutorcev [The Language and Folklore of the Aljutor]. Moscow: IM1I RAN, 'Nasledie'. Krapova, Iliyana 1999 The system of auxiliaries in Bulgarian. In: Mila Dimitrova-Vulchanova and Lars Hellan (eds.), Topics in South Slavic Syntax and Semantics, 59—89. Amsterdam: John Benjamins. Kucarov, Ivan 1994 Edno ekzoticno naklonenie na bälgarskija glagol [An Exotic Bulgarian Verb Mood]. Sofija: Universitetsko izdatelstvo "Sv. Kliment Oxridski". Mal'ceva, Alia A. 1998 Morfologija glagola ν aljutorskom jazyke [Verb Morphology in Aljutor], Novosibirsk: Sibirskij Xronograf. Rudin, Catherine 1986 Aspects of Bulgarian Syntax: Complementizers and Wh-Constructions. Columbus, OH: Slavica Publishers, Inc. Sadler, Louisa and Andrew Spencer 2001 Syntax as an exponent of morphological features. In: Geert Booij and Jaap van Marie (eds.), Yearbook of Morphology 2000, 71—96. Dordrecht: Kluwer Academic Publishers.

282

Andrew Spencer

Scatton, Ernest 1984 A Reference Grammar of Modern Bulgarian. Columbus: Slavica Publishers. Sells, Peter 2000 The morphological expression of syntactic information in Optimality Theoretic syntax. Linguistics Association of Great Britain Henry Sweet 2000 Lecture, University of Durham, 7 September 2000. To appear in Louisa Sadler and Andrew Spencer (eds.), Projecting Morphology. Stanford University: Center for the Study of Language and Information. Spencer, Andrew 2000 Agreement morphology in Chukotkan. In: Wolfgang U. Dressier, Oskar E. Pfeiffer, Markus A. Poechtrager and John R. Rennison (eds.), Morphological Analysis in Comparison, 191—222. Amsterdam: John Benjamins Publishing Co. Spencer, Andrew 2001 The paradigm-based model of morphosyntax. Transactions of the Philological Society 99: 279-313. Stump, Gregory T. 1993 On rules of referral. Language 69,449-479. Stump, Gregory T. 2001 Inflectional Morphology. A Theory of Paradigm Structure. Cambridge: Cambridge University Press. 2ukova, Aleftina N. 1972 Grammatika korjakskogo jazyka [A Grammar of Koryak]. Leningrad: Nauka. 2ukova, Aleftina N. 1980 Jazyk palanskix korjakov [The Language of the Palan Koryaks]. Leningrad: Nauka.

Transparent, restricted and opaque affix orders Barbara Stiebels

1.

Introduction*

Cross-linguistic research on the morphological structure of words has revealed two tendencies for possible affix orders: whereas functional categories (e.g. tense-aspect-mood systems) show a strong tendency for fixed affix orders (see Bybee 1985 and Wunderlich 1993), which only exhibit a small range of cross-linguistic variation, adverbial affixes and diathesis markers surface in variable orders that correlate with systematic differences in meaning. The behavior of both classes of morphemes can be motivated semantically; the current literature on affix order, though, is mainly dominated by syntactic approaches (e.g. Baker 1985, Pesetsky 1985, Muysken 1986, Speas 1991, Alsina 1999). The research on affix order has been stimulated by Baker's (1985) Mirror Principle, which states that affix orders should mirror syntactic derivations: (1)

Mirror Principle (Baker 1985: 375) Morphological derivations must directly reflect syntactic derivations (and vice versa).

Whereas in the original paper, Baker's proposals concerning the nature of the relevant operations are quite vague, Baker (1988) proposes a system where the affix order results from underlying syntactic configurations by head movement. In most cases, a given affix order can only receive a unique interpretation. Gaps in potential affix orders result from violations of syntactic principles (e.g. Case Filter, Empty Category Principle).

284

Barbara Stiebeis

Muysken (1986) interprets the Mirror Principle in terms of scope: if an affix A has scope over affix B, it must be external with respect to B, which may be illustrated as follows: (2)

a. Affix order:

V-AFFI-AFF 2 -...

VS.

V-AFF2-AFF1-...

b. Semantic scope:

AFF2(AFFI(V))

VS.

AFFI(AFF2(V))

The representations in (2) are meant to also include the mirror image, where all affixes are realized as prefixes. In case where the relevant affixes do not attach at the same side of the verbal stem, affix orders by themselves normally do not indicate their order of application. Therefore, the following structures are possible: (3)

[AFF Ι - [Verb-AFF 2 ] ]

vs.

[ [ AFF Ι - Verb] - AFF 2 ]

However, in some languages, such affix orders can be distinguished due to structural properties (e.g. linking patterns such as case distributions) or due to certain allomorphies. I will provide evidence for this in the following sections. A recent proposal by Rice (2000) puts emphasis on the availability of affix combinations that may receive different scope readings. According to Rice, three cases of affix combination have to be distinguished: first, two affixes A and Β do not exhibit a scope relation; therefore, no affix order concerning A and Β is preferred. Both affix orders may be possible, or a language may arbitrarily choose one option. The combination of the Chichewa (Bantu) intensifier (INT) -its 'do V well, intensively' with various diathesis markers is a case in question: (4)

Position of the intensifier morpheme in Chichewa (Hyman and Mchombo 1992) a. V-INT-APPL b . V-INT-PASS C. V-REC-INT

*V-APPL-INT *V-PASS-INT V-INT-REC

Transparent, restricted and opaque affix orders

285

With the applicative (APPL) -Ir and the passive (PASS) -Idw, the intensifier may only occur as inner morpheme; thus, the affix order is arbitrarily fixed. However, with the reciprocal (REC) -an, it may show up in both orders, yielding no interpretational difference. Secondly, each of the two affixes may take the other one into its scope. Therefore, both affix orders are relevant because they differ in their scopal interpretations. Thirdly, the scope relation is fixed such that only affix A may take affix Β into its scope; thus, only the order with A being the outer morpheme is possible. The first two cases are instances of local variability, i.e., there may be language-internal or cross-linguistic variation regarding the actual affix orders, whereas the third case is predicted to show global uniformity, i.e., all languages should display the relevant affix order. The second case is the one I am most interested in: the availability of two affix orders. The notion of scope, proposed by Muysken and Rice, will be clarified by considering explicit semantic representations. Differences in affix orders may result from semantic or syntactic properties. If, for instance, a causative affix (CAUSE) is combined with an adverbial affix (MOD), the readings in (5a/b) obtain: In (5a) the (outer) adverbial affix modifies the complex situation of causation, whereas in (5b), it only modifies the subevent expressed by the base verb.1 (5c-e) show the simplified representations for a transitive base verb, the verb extended by an adverbial affix and the causativized variant of the verb. Following the tradition of Lexical Decomposition Grammar (Joppen and Wunderlich 1995, Wunderlich 1997b, Stiebeis 1999), I represent the argument structure of a lexical item as a sequence of λ-abstractors (abstracting over the argument variables in Semantic Form [SF]): the referential argument of the verb, i.e. the situational variable s, is considered to be the highest argument and written as right-most argument on the theta-grid. The other arguments are written to its left according to their depth of embedding in SF and, thus, to their rank on the argument hierarchy.

286

Barbara Stiebeis

(5)

Combination of causative and adverbial morpheme a.

V-CAUSE-MOD

λy λχ λιι Xs' 3s [[ACT(u) & V(x,y)(s)](s') & MOD(s')] b. V-MOD-CAUSE λy λχ Xu λδ" 3s [ACT(U) & [V(x,y)(s) & MOD(S)]](S')

c. V d. V-MOD e. V-CAUSE

Xy λχ λβ V(x,y)(s) Xy λχ Xs [V(x,y)(s) & MOD(S)] λy λχ Xu Xs' 3s [ACT(U) & V(x,y)(s)](s')

Structural differences of affix orders often depend on the accessibility of arguments. Certain adverbial affixes, for instance, if combined with an applicative, may access the applied argument only as the outer morpheme. Wechsler (1989) has shown that adverbial affixes such as 'again' can only take direct arguments into their scope, which requires the applicative to apply before the affixation of the adverbial morpheme. In the following examples from Chichewa, the clitic nso 'again' can take the instrumental phrase into its scope only if the latter has been integrated as structural argument via applicativization as in (6b); in (6a), the instrumental phrase is realized as oblique adjunct. (6)

Repetitive in Chichewa (Wechsler 1989: 429) a. mu-lembe=nso chimangirizo ndi nthenga 2SG-write=again essay with feather 'you write the essay again, with a quill (this time)' b. mu-lembe-re-nso nthenga chimangirizo 2SG-write-APPL=again

feather

essay

'you write the essay with a quill again' It is the goal of this paper to provide a programmatic and semantically based overview of possible affix orders within the domain of diathesis morphology: which diathesis markers may be combined in principle and to which extent is the resulting morphological structure compositional, i.e. reflects the semantic composition and structural generation of forms? I will show that Baker (1988) makes wrong

Transparent, restricted and opaque affix orders

287

claims concerning possible diathesis combinations and that the Mirror Principle is a violable constraint. In the following section, I will discuss the compositionality of affix orders and introduce the notion of transparent, restricted and opaque affix orders. Section 3 briefly presents Baker's (1988) predictions for possible diathesis combinations and my analysis of diathesis operations. Section 4 is concerned with diathesis combinations that yield an identical semantic output, whereas section 5 is concerned with those that differ in semantic terms. Section 6 finally treats diathesis combinations in which one of the possible orders subsumes the inverse one.

2.

Compositionality of affix orders

Given that a particular combination of two morphemes A and Β has the universal potential for free order of application, and, hence, for the two affix orders Α-B and B-Α, one must distinguish three subcases with respect to the resulting structures: The most unproblematic case is the one in which both affix orders occur and transparently reflect the underlying scope relations. I will call these cases transparent affix orders. The following example from Bolivian Quechua shows the transparency of the combination of hortative and assistive. The assistive adds an assister argument to the base verb, which is realized as subject. The hortative, some kind of intensifier, expresses that the action denoted by the verb is executed with a certain amount of energy. (7)

Assistive/hortative in Quechua (van de Kerke 1996: 198) a. p'acha-ta t'aqsa-ysi-rqu-wa-rqa cloth-ACC wash-ASS-HORT-1.A-3 SG.PAST

'she helped me wash the clothes energetically' b. p'acha-ta t'aqsa-rqu-ysi-wa-rqa cloth-ACC wash-HORT-ASS-1. A-3 SG.PAST 'she helped me energetically wash the clothes'

288

Barbara Stiebeis

(7a) has the expected interpretation that the assisting action is executed energetically, whereas (7b) denotes the situation of energetic washing. If due to a language-specific constraint, only one affix order occurs, which receives a surface-true, i.e. compositional interpretation, this affix combination is restricted. Quechua, for instance, allows the repetitive affix -kipa 'again' only to be internal to the causative affix. The inverse order is not possible. The interpretation is compositionally fixed to the repetition of the situation expressed by the base verb. (8)

Causative/repetitive in Quechua (van de Kerke 1996: 176) mama-y p'acha-ta t'aqsa-kipa-chi-wa-rqa mother-ISG.P cloth-ACC wash-REP-CAUSE-l.A-3SG.PAST

'my mother made me rewash the clothes' #'again my mother made me wash the clothes' The most problematic case regarding the realization of a particular morpheme combination is found in languages in which a given affix order has both the compositional and the non-compositional interpretation. The latter violates the Mirror Principle. These affix orders are opaque. Whereas restricted affix orders show a complete gap for a certain morpheme combination, opaque affix orders only lack a distinct PF for one of the two readings. The combination of hortative and causative in Quechua is an example for an opaque affix order: the surface order HORT-CAUSE has the additional non-compositional interpretation that the causing event is executed energetically. (9)

Hortative/causative in Quechua (van de Kerke 1996: 177) Maria-wan p'acha-ta t'aqsa-rqu-chi-na-yki tiya-n Maria-COM cloth-ACC wash-HORT-CAUSE-NOML-2sG be-3SG

a. 'you should make Maria wash the clothes with energy' b. 'you must energetically make Maria wash the clothes' An even stronger case of opacity occurs if only one of the potential affix orders is allowed and if this has the interpretation of the

Transparent, restricted and opaque affix orders

289

inverse affix order, hence violates the Mirror Principle. This case is illustrated in (lOd): the first line shows the morphological orders (with V being the verbal stem), the second line the underlying scopal relations. (10) Schema of attested affix orders in multiscopal contexts a. transparent order V-A-B V-B-A scope

B(A(V)) A(B(V)) c. opaque\ V-A-B

*V-B-A

B(A(V)) A(B(V))

b. restricted V-A-B *V-B-A B(A(V)) *A(B(V)) d. opaquej V-A-B *B(A(V))

*V-B-A A(B(V))

These few examples from Quechua have already illustrated that a language may display transparent, restricted and opaque affix orders within the same domain of morphology, and that some affixes may even surface in both transparent and opaque affix orders (e.g. the hortative). One may speculate that different types of constraints are responsible for non-transparent affix orders: restricted affix orders presumably result from semantic and syntactic constraints, whereas opaque affix orders result from phonological and morphological surface constraints that dominate a constraint such as the Mirror Principle, or have to be explained in terms of language-specific conditions on grammaticalization.

3.

Order of diathesis markers

The most elaborate proposal concerning possible diathesis combinations has been made by Baker (1988). He analyzes diathesis markers as affixal heads that need to be incorporated into a governing head. Baker distinguishes three types of complex incorporation: whereas

290

Barbara Stiebeis

cyclic incorporation involves consistent movement of affixal heads into governing heads, acyclic incorporation means that an intermediate head is skipped and incorporated separately. Separate incorporation consists of parallel head movement of the heads of sister categories into the governing head. Acyclic incorporation is excluded in principle by the Empty Category Principle. Among the diathesis combinations that are based on cyclic or separate incorporation, some are excluded by the Stray Affix Filter (affixes should be attached to a stem) and the Case Filter. According to this analysis, the possible diathesis combinations should pattern as follows: (11) Possible diathesis combinations according to Baker (1988) Diathesis markers derivation cyclic CAUSE/ANTIPASS acyclic separate CAUSE/APPL separate cyclic CAUSE/PASS cyclic separate ANTIPASS/APPL acyclic cyclic APPL/PASS acyclic

affix order ANTIPASS-CAUSE * CAUSE-ANTIPASS CAUSE-APPL * APPL-CAUSE

(type 1/2) PASS-CAUSE (type 2) CAUSE-PASS

*ANTIPASS-APPL * APPL-ANTIPASS APPL-PASS *PASS-APPL

According to Baker, only causative and passive may be combined in both orders - at least in type 2 languages, whereas in type 1 languages, PASS-CAUSE violates the Stray Affix Filter. ANTIPASS-APPL and APPL-CAUSE violate the Case Filter under Baker's assumptions. Moreover, antipassive and applicative should not combine in any case. However, as cross-linguistic studies reveal, Baker's approach is far too restrictive. I will provide the relevant counter-evidence in the following sections. Within the framework I would like to propose, all diathesis markers can be combined in principle in both orders but may be restricted due to language-specific constraints on linking, i.e. the morphosyn-

Transparent, restricted and opaque affix orders

291

tactic realization of arguments. I assume that the Mirror Principle should be formulated in semantic terms (see also Muysken 1986): (12) Mirror Principle (own version) 'The affix order must mirror semantic composition.' This version of the Mirror Principle requires that the order of semantic integration of morphemes corresponds to their position in morphological structure, i.e. their relative distance to the stem. Unlike Baker, I assume that the Mirror Principle is a violable constraint: opaque affix orders violate it due to some higher-ranked constraint. In the following I will discuss to what extent the various combinations of diathesis markers yield affix orders that need to be distinguished in syntactic or semantic terms and to what extent transparent, restricted, and opaque affix orders occur. Following Wunderlich (1997b) and Dixon and Aikhenvald (2000) I distinguish three types of diathesis: (a) argument extension such as causative, assistive or applicative, (b) argument reduction as found with agentless passive, 'patientless' antipassive and reflexivization, and (c) diatheses that bring about alternative argument realizations such as agentive passive, antipassive with oblique realization of the internal argument, dative shift and locative alternation. I will not consider dative shift and locative alternation in the following. The representation of the causative has already been given in footnote 1. The assistive also introduces a highest argument but must be represented as an object control verb: it takes a verbal predicate, adds an assister argument and identifies the 'assisted' with the highest argument of the base verb; since there is no evidence in van de Kerke's data that these verbs may express indirect assistence, I do not assume that a new situational variable is introduced: (13) Representation of assistive (Quechua) ASS

λ Ρ λ χ λη Xs ASSIST(U,X,P(X))(S)

292

Barbara Stiebeis

In contrast to causative and assistive, the applicative introduces a lowest (or second-to-lowest) argument, namely the applied argument, which is realized as direct object. The following example from the Bantu language Kinyarwanda shows a benefactive applicative, in which a beneficiary ('boy') is added. (14) Applicative in Kinyarwanda (Kimenyi 1980: 32) umukoöbwa a-ra-som-er-a umuhuungu girl 3SG.N-PRES-read-APPL-ASP boy 'the girl is reading a book for the boy'

igitabo book

The argument extension found in the applicative is triggered by the integration of a semantic predicate, which I will simplify as APP(s,u), a place-holder for more specific predicates that integrate a beneficiary, instrument and so on (see (15a)). The applicative cannot be represented as a functor on verbs because this would yield inconsistencies between the argument hierarchy predicted from the process of semantic composition via Functional Composition and the arguments' depth of embedding in SF (Stiebels 1996, Wunderlich 1997a). I assume that the base verb undergoes argument extension as in (15b), i.e. it is extended by a predicative argument, and that the applicative is integrated via Functional Composition as shown in (15c) so that the arguments of the applicative are inherited to the base verb. (15) Representation and derivation of applicative a. APPL

Xu Xs APP(S,U) with APP G {INSTR(S,Z), LOC(S,Z), POSS(U,V), ...}

b. V

λy λχ hi V(x,y)(s) λΡ Xy λχ Xs [V(x,y)(s) & P(s)]

c. V-APPL

Xz Xy λ χ λβ [V(x,y)(s) & APP(S,Z)]

I assume that the agentless passive is represented as a functor that existentially binds the highest argument of the base verb (see (16a)). With agentive passive, the highest argument is marked as oblique (see (16b)).

Transparent, restricted and opaque affix orders

293

(16) Representation of passive a. λΡ λβ 3x P(x)(s) [agentless passive] b. λΡ λχ XsP(x)(s) [agentive passive] +obl Antipassive functions as the mirror image of passive. It either existentially binds the lowest argument of the base verb as shown in (17a) for a transitive verb, or marks this argument as oblique as in (17b). (17) Representation of antipassive a. λΡ λχ Xs 3y P(x,y)(s) b. λΡ λy λχ Xs P(x,y)(s) +obl

[patientless antipassive] [oblique antipassive]

Finally, reflexivization involves either co-indexation of θ-roles if it takes place in syntax (see (18a)), or multiple λ-abstraction if it is encoded morphologically (see (18b)).2 (18) Representation of reflexivization (transitive base verb) a. λ ^ λχϊ λβ V(x,y)(s) [syntax] b. λχ V(x,x)(s) [morphology] In this paper, I am concerned with morphological reflexives/reciprocals. The various combinations of diathesis markers show a varying tendency toward transparent, restricted and opaque affix orders, as I will show in the following. In principle, combinations of diathesis markers may be restricted due to semantic/conceptual factors (e.g. the role of specified agent arguments, the potential ambiguity of forms) and structural factors such as the maximal number of structural linkers and structural arguments in the particular language, the linker inventory, the symmetry or asymmetry of objects (Bresnan and Moshi 1993) and the obligatoriness of morphological marking of argument saturation (e.g. by means of pronoun or noun incorporation); these parameters constitute the linking profile of the language.

294

Barbara Stiebeis

Recall that languages with symmetric objects allow both internal arguments to be alternatively realized as the subject of a passive verb - besides other symmetries. Further restrictions are attested: in many languages, diathesis operations that follow argument extensions must not affect the structural realization of arguments that have been introduced into the base verb, whereas diathesis operations that follow argument reductions may be affected by the lack of structurally accessible arguments. In the following, I will first discuss diathesis combinations that yield an identical semantic output; then I will discuss those combinations that differ in their semantic output. Finally I will show to what extent affix orders may be in a subsumption relation. Apart from one exception (see section 5.1.), all diathesis combinations are affected by the language-specific linking profile and thus expected to show cross-linguistic variation (see also Alsina 1999).

4.

Diathesis combinations with identical semantic output

Diathesis combinations that have an identical semantic output, i.e. have an identical SF, may still differ in their θ-grid. Therefore I will distinguish two cases: diathesis combinations with identical SF and identical θ-grid and diathesis combinations with identical SF but distinct θ-grid. Only the first type is predicted to be either realized by a single affix order or to show free variation.

4.1. Diathesis combinations with identical θ-grid A diathesis combination that yields an identical output both for SF and θ-grid in any order is the combination of passive and reflexive, as shown for a transitive base verb: (19) Combination of passive and reflexive (transitive base verb) a. V-PASS-REFL Xy Xs Ξχ V(x,y)(s) ->• Xs Ξχ V(x,x)(s) b. V-REFL-PASS λχ Xs V(x,x)(s) -» Xs Ξχ V(x,x)(s)

Transparent, restricted and opaque affix orders

295

As with all similar cases, the two affix orders differ, however, in their intermediate step. In (19a) the possible antecedent of the reflexive is bound prior to reflexivization, whereas in (19b) it is bound after reflexivization, which might lead to a slight preference for (19b). Alsina (1999) claims that (19a) is universally excluded. The order V-PASS-REFL could be impossible in languages that require antecedents to be structurally realized. In principle, both combinations of passive and reflexive are ungrammatical with 2-place verbs in languages that do not allow impersonal passives. Moreover, with 3- and 4-place verbs, V-REFL-PASS is only possible in languages with symmetric objects (Alsina 1999) because only then can one of the remaining internal arguments be promoted to subject position. In Classical Nahuatl, an Uto-Aztecan language, the order of passivization and reflexivization can be determined on the basis of the actual reflexive allomorphs. In general, a 'specific reflexive' (with person and number agreement) is used if the argument in question is bound by the highest argument as in (20a). If the antecedent is not realized structurally as highest argument, the 'unspecific' reflexive ne- is used as in (20b): here, the highest argument is existentially bound and thus not accessible. (20) Passive/reflexive in Classical Nahuatl (Launey 1979: 61) a. ni-no-tlätia 1 SG.N-1 SG.REFL-hide Ί hide myself b. ne-tläti-lo usp.REFL-hide-PASS 'People hide' In order to account for the reflexive allomorphy, one must assume that reflexivization applies after passivization, which contradicts Alsina's (1999) claim. The order V-REFL-PASS is not attested in Nahuatl. Identical SFs are also generated with the combination of passive and antipassive: (21) Combination of passive and antipassive V-PASS-ANTIPASS/V-ANTIPASS-PASS

Xs By 3x V(x,y)(s)

296

Barbara Stiebeis

It is, however, dubious whether languages should make use of both argument reductions; this only seems plausible if multiple argument extensions apply.

4.2. Diathesis combinations with different θ-grids There are two cases in which diathesis combinations result in the same SF, but differ in their θ-grid: the combination of causative and passive on the one hand and the combination of antipassive and applicative on the other hand. Concerning the combination of causative and passive, the causer is existentially bound in the order V-CAUSEPASS, whereas the causee is bound in the inverse order: (22)

Combination of causative and passive a. b.

V-CAUSE-PASS λy λ χ Xs' 3u 3s [ACT(u) V-PASS-CAUSE

& V(x,y)(s)](s')

Xy λιι Xs' 3x 3s [ACT(U) & V(x,y)(s)](s')

The combination of causative and passive depends on constraints on structural linking and the requirement for morphologically encoded binding of arguments. V-PASS-CAUSE is superfluous in languages with optional (oblique) causees because there is no need to bind the causee. This affix order, however, is highly relevant in languages with obligatory (morphological) argument saturation or in languages with restrictions on structural linking: with the latter, causativization may be restricted to intransitive or transitive verbs. In Yucatec Maya, only two structural arguments are allowed (Krämer and Wunderlich 1999); therefore, causativization is restricted to intransitive verbs. In order to causativize an underlyingly transitive verb, argument reduction must take place.

Transparent, restricted and opaque affix orders

297

(23) Causative/passive in Yucatec Maya (Bricker 1978: 22) a. k=u kd?an-s-ik INCOMP=3

learn.PASS-CAUSE-IMPF

'he is teaching him' b. k=u kä?an-s-ä?al INC0MP=3 learn.PASS-CAUSE-PASS.IMPF 'he is being taught'

In (23a), the verb 'learn' is passivized before its argument structure is extended by a causer argument. (23b) shows that a causativized verb may undergo passivization. Therefore, both orders are attested in Yucatec Maya. The order V-CAUSE-PASS may be ungrammatical in languages that do not allow new arguments to be existentially bound or realized obliquely. Depending on the order of application of antipassive and applicative, different arguments are existentially bound or realized obliquely. In this respect, the combination of antipassive and applicative is a mirror image of the combination of causative and passive. If antipassive precedes the applicative, the base object is existentially bound or realized obliquely as in (24a). Such an order of application is often used if the language exhibits restrictions on structural linking: the antipassive reduces the number of structural arguments thus allowing subsequent argument extension. If the antipassive follows the applicative, the applied argument is existentially bound or realized obliquely as in (24b). (24) Combination of antipassive and applicative a. V-ANTIPASS-APPL b. V-APPL-ANTIPASS

λ χ Xs λζ λ ζ Xy - » λy

3y λχ λχ λχ

V(x,y)(s) Xs By [V(x,y)(s) & APP(s,z)] Xs [V(x,y)(s) & APP(S,Z)] Xs Bz [V(x,y)(s) & APP(S,Z)]

Languages that do not allow the existential binding of new arguments, should not display affix orders such as (24b); therefore, the combination of antipassive and applicative may be restricted. In

298

Barbara Stiebeis

West Greenlandic, applicative and antipassive may be iterated, thus transparently showing both affix orders: (25) Antipassive/applicative in West Greenlandic (Fortescue 1984: 270) 'he went out' (V-3SG) a. am-vuq 'he went out with it' (V-APPL-3SG/3SG) b. anni-p-paa c. anni-s-si-vuq 'he went out with something' (V-APPL-ANTIPASS-3 SG)

d. anni-s-si-vig-aa

'he went out with something to him' (V-APPL-ANTIPASS-APPL-3SG/3SG)

Note that West Greenlandic also exhibits several applicative variants and that the surface form is subject to many morphophonological processes. The resulting verb forms of the combination of causative and antipassive differ in their linking patterns - at least in languages with asymmetric objects. Depending on the linking conditions in causativized transitive verbs (oblique causee vs. oblique base object), the antipassive existentially binds the structural internal argument of the causativized verb (compare (26a/b)); therefore, only the causer argument remains structural (str). (26) Combination of causative and antipassive a. V-CAUSE-ANTIPAS S

λχ obi b. Xy obi

Χα str Xu str

Xs' 3y 3s [ACT(U) & V(x,y)(s)](s') [obi. causee] Xs' 3x 3s [ACT(U) & V(x,y)(s)](s') [obi. base obj.]

C. V-ANTIPASS-CAUSE λχ Χα Xs' 3y 3s [ACT(u)

str

& V(x,y)(s)](s')

str

If, however, the antipassive applies first, the base object must be existentially bound; therefore, the causee argument can be realized structurally. Note that Baker (1988) predicts both orders to be un-

Transparent, restricted and opaque affix orders 299

grammatical. The following examples from Chamorro provide counter-evidence to his claim (West Greenlandic would also be a case in question). In (27a) the antipassive applies prior to causativization; as expected, the causee häm 'us' is realized structurally (as NOM-marked pronoun), whereas the base object is oblique. In (27b) the antipassive follows causativization (umlauting the causative morpheme); here, both causee and base object are oblique.3 (27) Causative/antipassive in Chamorro (Gibson 1992: 175/150) a. ha=na '-fan-aitai häm / ma'estrak-ku 3SG.E=CAUS-ANTIPASS-read lPL.EX.N the teacher-ISG.P

ni

esti na

lebblu

OBL this LINK book 'my teacher made us read the book'

b. man-nä'-eksamina

häm

i

doktu

PL-ANTIPASS.CAUS-examine IPL.EX.N the doctor

as

nana-n-mami

OBL mother-n-LPL.EX.P 'we had the doctor examine our mother'

5.

Diathesis combinations that differ semantically

Since argument extensions are triggered by the integration of further predicates into the SF of the base verb, combinations of argument extensions yield outputs that differ according to their order of application. In addition, the combination of diathesis markers with reflexives or reciprocals may yield outputs that differ in their binding relations. I will begin with the discussion of the diathesis combinations that yield different SFs and different θ-grids.

5.1. Diathesis combinations with different θ-grids Since some structures and processes universally single out the highest argument of verbs ('logical subject'), the order of application of

300

Barbara Stiebeis

diathesis markers that introduce a highest argument is highly relevant. It is the combination of such diathesis markers that exhibits the strongest requirement and tendency for transparent affix orders. (28) represents the differences between the orders of application of assistive and causative. If the assistive applies first as in (28a), the causer u is the highest argument and, hence, realized as subject. If the causative applies first, the assister ν is the highest argument, and the causer is identified with the assisted. (28) Combination of assistive and causative a. V-ASS-CAUSE λ y λ χ λ ν λιι Is' b. V-CAUSE-ASS

3s [ACT(u) & ASSlST(v,x,V(x,y))(s)](s')

λy λχ Xu λν λβ1 3s [ASSIST(V,U,[ACT(U) & V(x,y)(s)]](s') The following example from Quechua shows the predicted transparency. (29a) represents the order V-ASS-CAUSE, (29b) the order V CAUSE-ASS.

(29) Causative/assistive in Quechua (van de Kerke 1996: 179) a. Maria-wan wawa-s-ta maylla-ysi-chi-wa-n Maria-COM child-PL-ACC wash-ASS-CAUSE-l.A-3SG

'she makes Maria help me wash the children' b. Maria-wan wawa-s-ta maylla-chi-ysi-wa-n Maria-COM child-PL-ACC wash-CAUSE-ASS-l.A-3SG

'she helps me to make Maria wash the children' Independent of the order of application, the causee is realized by the comitative and the assisted by object agreement, which indicates a certain asymmetry between the causative and the assistive, requiring further elaboration. Additional examples from Quechua and other languages (e.g. the iteration of causatives) confirm the prediction that the combination of diathesis markers introducing a highest argument should always be transparent.

Transparent, restricted and opaque affix orders

301

The combination of causative and assistive is partly mirrored by the combination of applicatives. Depending on the order of application, the resulting verbs differ in their SF and their θ-grid. (30) Combination of Applicatives a. V-APP1-APP2

λν λιι λy λχ Xs [V(x,y)(s) & APPI(s,u) & APP2(S,V)] b. V-APP2-APP1

λιι λν λy λχ Xs [V(x,y)(s) & APP2(S,V) & APPI(s,u)] In contrast to the combination of diathesis markers that introduce a highest argument, the combination of applicatives underlies language-specific linking constraints and, thus, does not exhibit global uniformity. In languages with asymmetric objects, the applied argument introduced last is predicted to be realized as structural object, whereas the internal argument introduced by the first applicative is oblique. With these languages, affix orders should be clearly distinguished due to their structural effects. In languages with symmetric objects, the order of application does not play a role: both internal arguments are structural and may be accessed likewise. Despite its structural relevance one hardly finds examples for multiple applicatives in which both orders are attested. Many languages - even those with asymmetric objects - have a strong preference for one of the possible orders of applicatives. In Tukang Besi, for instance, the combination of locative and comitative applicative is restricted: only the order LOC-COM is possible and the comitative argument 'with my younger sister' is realized structurally (i.e. NOM), as expected.4 (31) Locative/Comitative applicative in Tukang Besi (Donohue 1999: 249) ku-wil(a)-isi-ngkene-'e na iai-su 1 SG-g0-L0C-C0M-3 .A

(di

NOM

younger.sister-lSG.P

ompu-su)

OBL grandparent-lSG.P

Ί visited my grandmother with my younger sister'

302

Barbara Stiebeis

In contrast, the combination of comitative and benefactive applicative is opaque: the morphological order is restricted to COM-BEN (compare (32a/b)); however, only the comitative argument 'with her friend' can be realized structurally (compare (32a/c)), suggesting a scope relation COM over BEN. (32) Benefactive/comitative applicative in Tukang Besi (Donohue 1999: 248/252) a. no-homoru-ngkene-ako-'e te iaku na kene-no 3.REAL-weave-COM-BEN-3.A CORE

te

wurai

te

CORE

sarong

CORE

LSG

NOM

friend-3.P

ompu-su grandparent-lSG.P

'my grandmother w o v e a sarong for me with her friend'

b. * no-homoru-ako-ngkene c. * no-homoru-ngkene-ako-aku 3.REAL-weave-C0M-BEN-LSG.A

te CORE

kene-no friend-3.p

te

wurai na ompu-su sarong NOM grandparent-1 SG.P 'my grandmother wove a sarong for me with her friend'

CORE

5.2. Diathesis combinations with identical θ-grids Among the combinations of diathesis markers that differ in their semantic output are some that still have an identical θ-grid. This is due to the fact that the two orders only differ in the predicate's argument variables. A clear case is given by the combination of causative and reflexive/reciprocal. The order V-CAUSE-REFL is predicted to allow two readings in principle: one in which the causer binds the causee as in (33a.i) and one in which the causer binds the base object as in (33a.ii). Note, however, that this binding may violate locality constraints of particular languages because the causee is a potential interfering binder. The affix order V-REFL-CAUSE only allows the reading in which the causee binds the base object as in (33b).

Transparent, restricted and opaque affix orders

303

(33) Combination of causative and reflexive/reciprocal5 a. V-CAUSE-REFL (i) Xy Xu Xs' Bs [ACT(U) & V(u,y)(s)](s') (ii) λχ Xu Xs' 3s [ACT(U) & V(x,u)(s)](s') b. V-REFL-CAUSE λχ Χα Xs' Bs [ACT(u) & V(x,x)(s)](s') Some dialects of Quechua show the predicted affix orders and their corresponding interpretations: (34) Combination of causative and reflexive in Quechua (van de Kerke 1996: 180) a. maylla-chi-ku-n wash-c AUSE-REFL-3 SG (i) 'he lets himself be washed' (ii) 'he causes himself to wash someone' b. maylla-ku-chi-n wash-REFL-CAUSE-3 SG

'he causes someone to wash himself Other Quechuan dialects (van de Kerke 1996) display a restriction disallowing the affix order V-REFL-CAUSE, which might be explained by the fact that these dialects require the antecedent to be the highest argument of the verb. Again, Classical Nahuatl exhibits the expected reflexive allomorphy. In (35b), the reflexive verb form is causativized so that the causee binds the internal argument, whereas in (35c) reflexivization operates on the causative verb form so that the causer binds the reflexive. In the first case, the unspecific reflexive is used, in the second case the specific reflexive. (35) Reflexive/causative in Classical Nahuatl (Launey 1979: 186) a. mo-tlaso'tla-' 3.REFL-love-PL b.

'they love one another' ni-kin-ne-tlaso'tlal-tia 1 SG.N-3PL.A-USP.REFL-love-CAUSE

Ί cause them to love one another'

[REFL-CAUSE]

304

Barbara Stiebeis

c. ni-k-no-tti-tia

[CAUSE-REFL]

1 SG.N-3SG.A-1 SG.REFL-see-CAUSE

(i) Ί show myself to him' (ii) Ί make him see me' There are also languages that show restricted affix orders for Kinyarwanda (Kimenyi 1980) exhibits only the morphological structure [REFL-V-CAUSE], which is structurally ambiguous. However, the interpretation based on the order V-REFL-CAUSE is blocked; the reflexive must be bound by the highest argument. In contrast, Tukang Besi (Donohue 1999) does not allow the order VCAUSE-REC in the combination of causative and reciprocal. The combination of causative and applicative also yields two affix orders that differ in semantic terms. With the order V-CAUSE-APPL, the applied argument is expected to be related to the complex situation of causation as in (36a), whereas with the order V-APPL-CAUSE, the applied argument should be related to the subevent denoted by the base verb, as shown in (36b). CAUSE/REFL:

(36) Combination of causative and applicative a.

V-CAUSE-APPL

λζ λy λχ λ\ι Xs* 3s [[ACT(u) & V(x,y)(s)](s') & APP(s\z)] b.

V-APPL-CAUSE

λζ λy λχ Xu Is' 3s [ACT(u) & [V(x,y)(s) & APP(s,z)]](s')

The interpretational differences become evident with instrumental and locative phrases: is the instrument part of the causing event or part of the subevent denoted by the base verb? Likewise, does the locative refer to the place of the causing event or to the place where the action denoted by the base verb is situated? However, the order V-APPL-CAUSE is rarely attested, which led Baker to conclude that this order is ungrammatical in any case. Evidence for such an order is found, for instance, in Chamorro (and with certain verbs in Tukang Besi, see Donohue 1999). Overtly, the two affix orders cannot be distinguished because the causative is realized as prefix and the applicative as suffix. The linking patterns indicate

Transparent, restricted and opaque affix orders

305

the underlying derivation: (37a) corresponds to the order V-CAUSEAPPL because the applied argument is realized by the nominative (NOM). Every derivation subsequent to the applicative would render the applied argument oblique, which is not the case in (37a). In (37b), the causee is realized by the nominative, which can only be explained with respect to the order V-APPL-CAUSE. Note that the applied argument 'Joaquin' is related to the subevent of telling a story. (37) Causative/applicative in Chamorro (Gibson 1992: 110/122) a. hu=na'-puni'-i yu' nu i bäbui as Juan 1 SG.E=CAUS-kill-APPL ISG.N OBL the p i g

Ί made Juan kill me the pig' [CAUS-APPL] b. si tata-hu ha=na'-sasngan-i NOM father-ISG.P 3sG.E=CAUs-tell-APPL

as

Joaquin nu

OBL J

i

OBL the

OBL Juan

yu' ISG.N

estoria-mu story-2sG.P

'my father made me tell Joaquin your story' [APPL-CAUS] As overwhelming tendency, the combination of causative and applicative is realized by means of the opaque affix order V-CAUSEAPPL. This is true, for instance, for Quechua, as the following example illustrates: (38) Causative/applicative in Quechua (van de Kerke 1996: 192) mama-y Ana-wan chompa-ta ruwa-chi-pu-wa-n mother-1SG.P Ana-COM sweater-ACC make-CAUSE-APPL-l-3SG

a. 'in my place my mother made Ana make a sweater' b. 'my mother made Ana make a sweater in my place' c. 'my mother made Ana make me a sweater' Reading (38a) is compositional, whereas the other two readings in which the beneficiary is related to the subevent denoted by the base verb are not. Similarly, Chichewa and Nahuatl only exhibit opaque affix orders with CAUS/APPL. The Chichewa sequence lir-its-ir (cryCAUSE-APPL) occurs for an instrumental applicative in which the instrument is used in the causing event as well as for a benefactive ap-

306

Barbara Stiebeis

plicative in which the beneficiary is related to the crying event (Hyman and Mchombo 1992). The fact that the combination of causative and applicative shows the strongest tendency for opaque affix orders among all diathesis combinations suggests that the difference in meaning is not very crucial and, hence, is not reflected in morphology. The sortal properties of the applied argument determine to which situation argument it is related.

6.

Diathesis combinations with a potential subsumption relation

There are also cases in which one of the two affix orders may be semantically or structurally ambiguous such that it subsumes the interpretation or the linking pattern of the inverse order.

6.1. Potential semantic subsumption Potential semantic subsumption is found with the combination of applicative and reflexive/reciprocal. The order V-APPL-REFL has two possible interpretations, namely those indicated in (39a): the highest argument binds the applied argument as in (39a.i) or the base object as in (39a.ii).6 The inverse order can only have an interpretation that is identical to (39a.ii). (39) Combination of applicative and reflexive/reciprocal a. V-APPL-REFL

(i) Xy λ χ Xs [V(x,y)(s) & APP(s,x)]

(ii) λ ζ λ χ Xs [V(x,x)(s) & APP(S,Z)] b . V-REFL-APPL

λ ζ λ χ Xs [V(x,x)(s) & APP(s,z)]

Therefore, V-APPL-REFL subsumes V-REFL-APPL in principle. Such a subsumption may result in two compensation strategies: V-REFLAPPL may either block interpretation (ii), perhaps due to a highranked ambiguity constraint, or it may be blocked morphologically

Transparent, restricted and opaque affix orders

307

by the order V-APPL-REFL because some kind of economy constraint rules out superfluous morphological structure: the affix order with the wider extension is preferred. In Chichewa both affix orders occur and show the full range of possible interpretations: the order V-APPL-REC is ambiguous, as (40a/c) show, and does not blockη the inverse order (see (40b)), a fact that needs further investigation. If the reciprocal is the inner morpheme, a wellformedness constraint requires that a copy of the reciprocal is added to the following diathesis marker. 8 (40) Applicative/reciprocal in Chichewa (Hyman and Mchombo 1992, Alsina 1999: 12) a. mang-ir-antie-APPL-REC b. mang-an-ir-antie-REC-APPL-AN

'tie for each other' 'tie each other for/with/at'

c. alenje a-na-meny-er-an-ά mikondo hunters.2 CL.2-PAST-hit-APPL-REC-FV spears.4 'the hunters hit each other with spears' Classical Nahuatl distinguishes the two readings due to the selection of the reflexive allomorph: if the applied argument is bound, the specific reflexive is used as in (41a), if, however, the base object is bound, the unspecific reflexive is used as in (41b). (41) Applicative/reflexive in Classical Nahuatl (Launey 1979: 196) a. ni-k-no-kwi-tl-s 1 SG.N-3 SG. A-1 SG.REFL-take-APPL-FUT Ί will take it for m y s e l f

b.

ni-k-ne-tläfi-lia 1 SG.N-3 SG. A-USP.REFL-hide-APPL

Ί hide myself from him' Morphological structures such as those of Nahuatl do not allow any conclusion with respect to the underlying affix order in case of subsumptive relations.

308

Barbara Stiebeis

6.2. Potential structural subsumption Depending on the symmetry of objects, the orders of combining applicative and passive must be distinguished. Generally, the order VPASS-APPL only allows the internal argument of the base verb to be realized as the subject of a passive verb because the applied argument is not locally accessible (see (42a)). In languages with asymmetrical objects, V-APPL-PASS only displays a structure with the applied argument being the subject of the passive verb (see (42b)). In languages with symmetrical objects both internal arguments may be alternatively realized as subject. In this case, the order V-APPL-PASS subsumes the inverse order with respect to its linking potential. (42) Combination of applicative and passive a.

V-PASS-APPL λζ Xy Xs 3x [V(x,y)(s) & APP(s,z)]

NOM b. V-APPL-PASS λζ λy Xs 3 x [V(x,y)(s) & APP(s,z)]

NOM (NOM) Note, however, that V-PASS-APPL is excluded in languages in which applicatives require the presence of a specified structural highest argument; this is, for instance, relevant in Tukang Besi (Donohue 1999: 297). In Chichewa, instrumental and benefactive applicatives may only occur in the order V-APPL-PASS. AS in Tukang Besi, these applicatives require a specified structural agent argument. With the locative applicative, both orders occur:

Transparent, restricted and opaque affix orders

309

(43) Passive/locative applicative in Chichewa (Alsina 1999: 10/11, Alsina and Mchombo 1993: 42) a. ukönde u-ku-lük-ir-idw-ά ρά-mchenga net. 14 CL.14-PRES-weave-APPL-PASS-FV

CL.16-sand.3

(ndi äsödzi) by fishermen.2 'the net is being woven on the sand (by fishermen)' b. ukönde u-ku-luk-idw-ir-ά ρά-mchenga net. 14 CL.14-PRES-Weave-PASS-APPL-Fv

CL.16-sand.3

(ndi äsödzi) by fishermen.2 'the net is being woven on the sand (by fishermen)' c. pa-mchenga pa-ku-lvk-ir-idw-ά mikeka CL.16-sand.3 CL.16-PRES-weave-APPL-PASS-FV mats.4 'the beach is being woven mats on' The order V-APPL-PASS is structurally ambiguous (see (43a/c)) because this applicative licences object symmetry; hence, both the applied argument and the base object may be promoted to subject position. The fact that both (43 a) and (43b) are acceptable although a blocking effect is expected needs to be clarified.

7.

Conclusions

The preceding discussion has shown that, in principle, diathesis markers can be combined in any order. In the case of passive and reflexive and passive and antipassive, an identical output is generated. It is the intermediate step that might favor one affix order over the other, depending on the linking constraints of the relevant language. The following table summarizes the findings along the following dimensions: (a) whether an identical SF is generated, (b) whether the two orders yield an identical θ-grid, (c) whether there may still be differences in the resulting linking patterns despite an identical θ-grid and (d) whether the diathesis combination is influenced by the specific linking conditions of the language in question.

310

Barbara Stiebeis

The last column indicates the tendency with respect to the actual affix orders: t - transparent, r - restricted, ο - opaque. However, further typological studies are necessary to validate the observed patterns. (44) Properties of diathesis combinations same same ΘSF grid

same linking pattern

parameterized according to linking profile

affix order

+

+ +

+ +

r ? t/r t/r r t/r t/r t/r o/t t r/o

PASS/REFL

+

PASS/ANTIPASS

+

+ +

CAUSE/ANTIPASS

+

-/+

APPL/REFL

+/-

PASS/APPL

+

+ +

CAUSE/PASS

+

-

-

ANTIPASS/APPL

+

-

-

CAUSE/REFL

-

+

CAUSE/APPL

-

+

+ +

+ + +

CAUSE/ASS

-

—

-

—

APPL/APPL

-

-

+

+/-

-/+

+ + +

+

+

Restricted affix orders mostly result from language-specific constraints on linking. Since almost all diathesis combinations interact with the linking profile of the language, restrictions are expected. The only invariant diathesis combination (causative and assistive or iteration of causatives) exhibits the predicted transparency. Up to now, only two cases of opaque affix orders have been attested: the combination of causative and applicative and multiple applicatives. In both cases, argument-extending diatheses are combined, which pose a challenge for structural linking in most languages. Apart from the fact that the factors that trigger opacity need to be determined, opacity in itself is a serious problem for morpheme-based approaches. If one does not want to make use of lateinsertion models (Distributed Morphology, Halle and Marantz 1993), post-syntactic filters, morphological circumscriptions (Hyman and

Transparent, restricted and opaque affix orders

311

Mchombo 1992) or covert LF-movements, which are all very powerful mechanisms, the question arises as to which alternatives are available. Moreover, one must ask whether the semantics is processed at each step of morphological concatenation, which is desirable from isomorphism, or whether semantic processing may be postponed. The latter alternative may be plausible if it can be constrained. Therefore, further studies must show whether opacity is strictly local, involving only adjacent morphemes. Another challenge is given by subsumptive affix orders. To what extent do the predicted blocking effects occur? How may they be modelled? A possible solution might be provided within the framework of Bidirectional Optimality Theory (Blutner 2000). Finally, a typology of possible affix orders is not easily available within Optimality Theory because diathesis combinations interface with different modules of the grammar (syntax, semantics, morphology, discourse factors), which might not be evaluated parallel in one step.

Notes * This paper is based on research that has been conducted within the Sonderforschungsbereich 'Theory of the Lexicon', funded by the German Science Foundation (DFG). I would like to thank Dieter Wunderlich, the audience in Leipzig and the anonymous reviewer for helpful comments. Throughout the paper, I will make use of the following abbreviations: '=': clitic boundaiy, '#': deviant semantic interpretation; A: object agreement, ACC: accusative, ANTIPASS: antipassive, APPL: applicative, ASP: aspect, ASS: assistive, BEN: benefactive applicative, CAUSE: causative, CL: class marker, COM: comitative (case/ applicative), CORE: core case, E: ergative agreement, EX: exclusive, FUT: future tense, FV: final vowel, HORT: hortative, IMPF: imperfective, INCOMP: incompletive aspect, INT: intensifier, LINK: linker, LOC: locative applicative, MOD: modifier, N: subject agreement, ΝΟΜ: nominative, NOML: nominalization, OBL: oblique, P: possessor agreement, PASS: passive, PAST: past tense, PL: plural, PRES: present tense, REC: reciprocal, REFL: reflexive, REP: repetitive, SG: singular, USP: unspecified 1. I assume that the causative morpheme is a functor on the verb with the following Semantic Form: λΡ Xu Xs' 3s [ACT(u) & P(s)](s')

312

2.

3.

4.

5. 6.

7.

8.

Barbara Stiebeis

The causative integrates a verbal predicate Ρ via functional composition, binds its situational variable and adds the causer argument u and the complex situational variable s' (Wunderlich 1997b). ACT denotes an unspecified activity. The causal relation is inferred from conceptual coherence constraints (Kaufmann 1995). Modifiers that do not add arguments can also be represented as functors on verbs. Given that ditransitives should be included in the discussion of reflexive binding, I do not see a possibility to represent the reflexive morpheme as a functor on the verb. Therefore, I assume that it might be represented as a template that operates on the base verb's SF (with consequences for λ-abstraction). Although the causee 'the doctor' does not receive an oblique marker, its position (following the subject) renders it oblique in (27b). In (27a) the causee precedes the subject. Tukang Besi has an unusual linking system. The subject of intransitive verbs is marked by ΝΟΜ. With transitive verbs, the object is marked by NOM if the verb exhibits object agreement; otherwise the subject is marked by NOM. Structural arguments that are not realized by NOM are marked by the 'core marker' te. Note that (33a.i) is preferred with reciprocals and (33a.ii) with reflexives. Interpretation (39a.ii) may be blocked if a language requires the reflexive to correspond to a structural argument; in languages with asymmetric objects, the base object is often not structural and, hence, possibly not accessible to reflexivization. One might speculate that the order V-REC-APPL is chosen in cases in which base object and applied object both qualify as target for anaphoric binding, i.e. with the benefactive applicative, and in which the binding of the base object is to be ensured. V-APPL-REC would then be used for the binding of the applied argument. With the other applicatives, such an ambiguity is less likely and VREC-APPL is avoided as a superfluous form. The numbers in (40c) indicate noun class.

References Alsina, Alex 1999 Where's the mirror principle. The Linguistic Review 16: 1-42. Alsina, Alex and Sam Mchombo 1993 Object asymmetries and the Chichewa applicative construction. In: Sam Mchombo (ed.), 17-45. Baker, Mark 1985 The Mirror Principle and morphosyntactic explanation. Linguistic Inquiry 16: 373-415.

Transparent, restricted and opaque affix orders Baker, Mark 1988

313

Incorporation: a theory of grammatical function changing. Chicago: The University of Chicago Press. Blutner, Reinhard 2000 Some aspects of optimality in natural language interpretation. Journal of Semantics 17: 189-216. [Special volume edited by Petra Hendriks, Helen de Hoop and Henriette de Swart] Bresnan, Joan and Lioba Moshi 1993 Object asymmetries in comparative Bantu syntax. In: Sam Mchombo (ed.), 47-91. Bricker, Victoria R. 1978 Antipassive constructions in Yucatec Maya. In: Nora C. England, Collette C. Craig and Louanna Furbee-Losee (eds.), Papers in Mayan linguistics, 3-23. Columbia: Museum of Anthropology, University of Missouri. Bybee, Joan 1985 Morphology: A study of the relation between meaning and form. Amsterdam: Benjamins. Dixon, R. M. W. and Alexandra Y. Aikhenvald 2000 Introduction. In: R. M. W. Dixon and Alexandra Y. Aikhenvald (eds.), Changing valency: case studies in transitivity, 1-29. Cambridge: Cambridge University Press. Donohue, Mark 1999 A grammar ofTukang Besi. Berlin: Mouton de Gruyter. Fortescue, Michael 1984 West Greenlandic. London: Croom Helm. Gibson, Jeanne D. 1992 Clause union in Chamorro and in Universal Grammar. New York: Garland. Halle, Morris and Alec Marantz 1993 Distributed Morphology and the pieces of inflection. In: Kenneth Hale and Samuel J. Keyser (eds.), The view from building 20: Essays in linguistics in honor of Sylvain Bromberger, 111-176. Cambridge, Mass.: MIT Press. Hyman, Larry M. and Sam Mchombo 1992 Morphotactic constraints in the Chichewa verb stem. Berkeley Linguistic Society 18, 350-364. Joppen, Sandra and Dieter Wunderlich 1995 Argument linking in Basque. Lingua 97: 123-169.

314

Barbara Stiebeis

Kaufmann, Ingrid 1995 Konzeptuelle Grundlagen semantischer Dekompositionsstrukturen: Die Kombinatorik lokaler Verben und prädikativer Komplemente. Tübingen: Niemeyer. Kerke, Simon van de 1996 Affix order and interpretation in Bolivian Quechua. Ph.D. Dissertation, University of Amsterdam. Kimenyi, Alexandre 1980 Α relational grammar of Kinyarwanda. Berkeley: University of California Press. Krämer, Martin and Dieter Wunderlich 1999 Transitivity alternations in Yucatec, and the correlation between aspect and argument roles. Linguistics 37: 431-479. Launey, Michel 1979 Introduction a la langue et a la litterature Azteques. Vol. 1, Grammaire. Paris: L'Harmattan. Mchombo, Sam (ed.) 1993 Theoretical aspects of Bantu grammar. Stanford: CSLI publications. Muysken, Pieter 1986 Approaches to affix order. Linguistics 24: 629-643. Pesetsky, David 1985 Morphology and logical form. Linguistic Inquiry 16: 193-246. Rice, Keren 2000 Morpheme order and semantic scope. Cambridge: Cambridge University Press. Speas, Margaret. 1991 Functional Heads and Inflectional Morphemes. Linguistic Review 8: 389-417. Stiebels, Barbara 1996 Lexikalische Argumente und Adjunkte: Zum semantischen Beitrag von verbalen Präfixen und Partikeln. (Studia grammatica 39.) Berlin: Akademie Verlag. Stiebels, Barbara 1999 Noun-verb symmetries in Nahuatl nominalizations. Natural Language and Linguistic Theory 17: 783-836. Wechsler, Stephen 1989 Accomplishments and the prefix re-. Proceedings of the NorthEastern Linguistic Society 19,419-438.

Transparent, restricted and opaque affix orders

315

Wunderlich, Dieter 1993 Funktionale Kategorien im Lexikon. In: Frank Beckmann and Gerhard Heyer (eds.), Theorie und Praxis des Lexikons, 54-73. Berlin: Walter de Gruyter. Wunderlich, Dieter 1997a Argument extension by lexical adjunction. Journal of Semantics 14: 95-142. Wunderlich, Dieter 1997b Cause and the structure of verbs. Linguistic Inquiry 28: 27-68.

Direction marking as agreement Jochen Trommer

1.

Introduction

The typological literature (e.g. Croft 1990) assumes that certain languages mark the (un-)naturalness of predication types with respect to animacy hierarchies by special affixes. For example, in the Algonquian language Menominee (Bloomfield 1962), verbs mark predications which involve 1st or 2nd person subjects and third person objects by the affix -a- and predications with 3rd person subjects and lst/2nd person objects by -eko! (1)

a.

b.

ke-nan-a-w-aw (kenanawaw) 2-fetch-D- [+3 ]- [-1 +pl] 'you (pi.) fetch him' (Bloomfield 1962: 153) ke-nan-eko-w-aw (kenanekowaw) 2-fetch-D-[+3]-[-l+pl] 'he fetch you (pi.)' (Bloomfield 1962: 154)

In terms of the Algonquianist literature (Hockett 1966), -a marks a "direct" situation since a speech act participant is supposed to be a more "natural" subject than a 3rd person argument while -eko, which appears in the inverse case, marks an "inverse" constellation. Both types of marking are referred to by the term "direction marking". A statement from Comrie (1980: 62) describes how direction marking is related to animacy hierarchies: "Languages which have an opposition between direct and inverse verb forms build directly upon the animacy hierarchy: the direct forms are used when the subject of the transitive verb is higher on the scale of animacy than the direct object ... The inverse form is used when the subject is lower in animacy than the object..." The animacy hierarchy typically has the form in (2a) or (2b):

318

(2)

Jochen Trommer

The Animacy Hierarchy: a. 2 > 1 > 3 > inanimate, or b. 1 > 2 > 3 > inanimate

In this article, I propose to analyze direction marking in a much simpler way, namely as transitive agreement. Under this account, direction markers do not refer to feature hierarchies as in (2), but are governed by universal markedness constraints which correspond to prominence hierarchies in a well-defined way. Constraints are taken to be violable and ranked in the sense of Optimality Theory (OT, Prince and Smolensky 1993). Crosslinguistic variation in direction marking, which has been interpreted as evidence for languagespecific prominence hierarchies (e.g. Croft 1990), follows from different ranking of these constraints. The claim that direction marking is agreement receives additional support by the fact that another aspect of agreement in the discussed languages is also regulated by constraints referring to feature hierarchies: the competition for feature realization by case-less agreement affixes. This phenomenon will be analyzed in section 3 using the version of OT-morphology introduced in section 2. The account is then extended to direction markers themselves (section 4). Section 5 deals with the distribution of zero direction marking. The article concludes with a comparison of the analysis with alternative analyses of direction marking (section 6) and a short summary (section 7).

2.

The Framework

The formal framework I will adopt in this paper is Distributed Optimality (DO, Trommer 2001), a constraint-based morphological framework based on Optimality Theory (OT, Prince and Smolensky 1993). DO assumes a "minimalist" conception of OT-morphology embracing the restrictive assumptions in (3):

Direction Marking as Agreement

(3)

a.

b.

c.

319

Locality: Morphology interprets the output of syntax. Morphological constraints can only refer to small wordlike units, not to syntactic phrases. Inclusiven ess: All morphemes in the morphological output have to be licensed by a syntactic item whose features they subsume. There is no insertion of features. Free Ranking: All rankings of the assumed constraints yield a possible grammar.

All these assumptions are potentially problematic for an adequate formalization of direction marking. If direction markers express the relation of subject and object, they must have access to at least the clause level. If direction markers evaluate the relation of subject and object with respect to a feature hierarchy, the morphosyntactic features of these markers cannot simply subsume the features of syntactic structures which do not contain feature hierarchies in any sense. Free Ranking is a standard claim in OT, but it is systematically violated by Harmonic Alignment (see section 6.) which plays a prominent role in OT-formalizations of markedness hierarchies. Since feature hierarchies seem crucial to direction marking, it is a special challenge to test whether fixed constraints hierarchies can be dispensed with and Free Ranking can be maintained. In DO, word forms are characterized by bundles of morphosyntactic features which derive from syntactic derivations. For example, for (la) we can assume the following representation2:

(4)

[+V]

"+NOM" +2-1 +pl

"+ACC " +3 +sg

Vocabulary items, such as na n:[+Y], or Ae:[+2] associate morphosyntactic features with pieces of sound and are used to spell out these feature bundles. PARSE constraints require that certain feature combinations are realized by vocabulary items of the output form. If a feature is not realized in the output, the corresponding constraint is violated.

320

Jochen Trommer

(5)

OT-tableau for the input in (4) I PARSE PARSE PARSE [NUM] 1 [CAT] [P]

a. «y b. c.

ke[+2] na n r+vi ke[+2]

na n -a· r+Vl D -a- -w d [+3] na n -a· [+V] D

-w -a w [+31 [-1+pl] -a w r-i+pii -w [+3]

*

*!

*

*!

**

Thus, PARSE [NUM] in (5) is violated by each number feature from the input (+sg, +pl) that is not realized in the output, and PARSE [P] correspondingly for person features (+2,+3, -1). Each violation is depicted by a star in the tableau. Following the principles of OT, that candidate is optimal (indicated by ts·) which induces the least serious constraint violations. In (5), this is kena na wa w because the only constraint it violates once (PARSE [NUM] for +sg) is also violated once or twice by the other candidates which violate additional constraints. Crucially, even the optimal candidate is not perfect: it violates PARSE [NUM]. The reason is that there is no vocabulary item in Menominee that expresses [+sg]. Feature realization can also be prevented by other constraints. Such constraints will be introduced in the following sections.

3.

Hierarchy-Based Competition

The term "hierarchy-based competition" is intended to cover cases where person features compete for realization and asymmetries between the features determine which one surfaces. In Turkana (Dimmendaal 1983), finite verbs agree with subjects and objects, and the same person markers are used for subject and object agreement. However, object agreement is suppressed if both arguments are participants (6a), or both are non-participants, and subject agreement is suppressed if the subject is 3rd person and the object [-3] (6b):

Direction Marking as Agreement (6)

a.

k-a-ram-i

D-l-beat-ASP Ί will beat you' b.

321

(Dimmendaal 1983: 122)

k-ä-mn-ä

D-l-love-ASP 'he loves me'

(Dimmendaal 1983: 123)

In effect, person agreement is always marked by one single affix. Thus Turkana person agreement exhibits a phenomenon that has been one of the major motivations for OT: a rule conspiracy. Different processes (suppression of subject and object agreement) "conspire" to reach the same goal: the restriction of agreement to a single person affix. I propose to capture this fact by the constraint in (7), which is violated by any word form that contains more than one nonportmanteau affix marked for person. (7)

BLOCK [PERSON]

As we will see in the following, many languages exhibit the same or related constraints. Thus, BLOCK [PERSON] is not a languageparticular stipulation, but seems to be part of the universal constraint set of Universal Grammar. Note that the restriction of (7) to person affixes is necessary, since other inflectional affixes like aspect, direction and number marking can coocur with the person affixes3. What remains to be done is to explicate the choice for which affix is actually suppressed to satisfy (7). Given the prominence scales in (8a,b), there is a simple principle behind the affix choice in these cases (8c): (8)

a. b. c.

Subject > Object 1/2 > 3 Choose the affix that corresponds to the higher scale position

However, (8c) cannot be maintained in its most general form since it leads to a contradiction for 3 1 predications, where (8b) seems to

322

Jochen Trommer

outrank (8a). Thus, I propose to replace (8c) by (9a), which gives us the constraints in (9b,c). (9)

a. b. c.

If there is a prominence scale A > Β there is a PARSE constraint PARSE [ P ] ^ PARSE [P]["3W+31 PARSE [p][+nom]/[+acc]

PARSE [P]*® is to be read as follows: Realize the person features of a syntactic head containing A if this is adjacent to a head containing B. Thus, PARSE [P][_3J/[+3] requires that the person features of a 1st or 2nd person head are spelled out by an affix, if it is neighboured by a 3rd person head. Now it is crucial how these constraints are ranked, since for the evaluation procedure of OT optimization for higher constraints is always more important than optimization for lower constraints. For Turkana, I will assume that the ranking is BLOCK [P] » PARSE [Ρ]ι"3]/[+3] » PARSE P>][+nom]/[+acc] I f o n e a r g u m e n t i s [ + 3 ] and the other [-3], we get the following tableau: (10)

Mixed:[+Nom +3], [+Acc +1]2 BLOCK [PI

a. tsb. c.

r+1], Γ+31, Γ+1Ί, Γ+31,

PARSE [pjHM+3]

PARSE

|pj[+Nom]/[+Acc] *

*!

PARSE [P] *

*

*!

Spell-out of both heads would violate BLOCKING, therefore (10c) is discarded (depicted by "!" after the relevant violation mark). Suppression of the [-3] head (10b) would violate PARSE [P]["3J/[+3], which is also discarded. The only remaining and hence optimal candidate is (10a). If both arguments are [-3], PARSE [Ρ][_3Μ+3] becomes irrelevant, and PARSE [P][+N0MW+ACC] favours the appearance of the nominative head:

Direction Marking as Agreement

(11)

Only SAP Arguments: [+Nom +2], [+Acc +1]2 BLOCK

[PI

a. isb. c.

323

PARSE

rp][-3]/[+3]

PARSE

j-pj[+Nom]/[+Acc]

PARSE —

[PI

.. „

[+21,

r+n,

[+21, [+11,

*l

siwifn

*!

The same is true if both agreement heads are [+3]. While subject and object agreement do not differ in morphological expression, the account predicts that the surfacing marker is coindexed with the subject. According to the principles of OT, all possible rankings of constraints should yield an attested or at least plausible language type. In the following, I will show that this indeed holds for the proposed constraints. If PARSE [P] is ranked above BLOCK [P], both Agrs + Agr 0 are realized (12): (12)

PARSE [P] » BLOCK [P]

Otherwise, there are three possibilities: If PARSE [P][+NOMJ/[+ACC] and BLOCK [P] are ranked above PARSE [Ρ]Η]/[+3] (13a), only subject agreement is realized. This can be observed in a standard Indo-European language such as English. If PARSE [Ρ][-3Μ+3] and BLOCK [P] are above PARSE [P][+N0MW+ACC], we get the distribution of Turkana (13b). The third possibility is that PARSE [P]h3]/[+3] and PARSE [pfNOMM+Acc] b o t h dominate BLOCK [P] (13c) (PARSE constraints are abbreviated by the respective superscripts, the {} brackets enclose constraints whose ranking with respect to each other is irrelevant, "&" is used to combine different subrankings, i.e., each of the rankings in a., b. and c. must be combined separately with BLOCK [P] » PARSE [P]>:

324

(13)

Jochen Trommer

[BLOCK [P] » PARSE [Ρ]]

a·

-

b

&

·

c.

[ {

[

;

n

l ä

[{ BLOCK m

c

1

}

}

"

[^«""I'I+H

- [ { Ä L ] } »"««in]

In languages of the type of (13c), subject agreement should always be realized, but object agreement should be suppressed, unless the object is higher on the person hierarchy than the subject. This corresponds closely to the analysis of Quechua proposed in Lakämper and Wunderlich (1989: 127): (14)

a.

b.

Object-Subject Constraint (OSC): The object may be marked separately from the subject only if it refers to person that is higher on the hierarchy of person than the person to which the subject refers Hierarchy of person: 1 > 2 > 3

What distinguishes Quechua from (14c) is then only the relevant hierarchy (1 > 2 > 3 instead of 1/2 > 3). Since the latter type of hierarchy is also well-documented, (14c) is also a plausible language.

3.1. Capturing different Hierarchies

Indeed, there are languages which exhibit similar blocking phenomena as Turkana, but according to slightly different hierarchies: (15)

a. b. c.

Turkana: 1/2 > 3 | Nom > Acc Nocte: 1 > 2 > 3 Menominee: 2 > 1 > 3

Direction Marking as Agreement

325

In contrast to Turkana, in these languages, competition under blocking is resolved exclusively with reference to person features. However, in some cases 2nd person agreement wins over 1st person, and in others it is the other way around. This can be integrated in the proposed account by assuming the more elaborated hierarchy in (16a) and replacing (9a) by (16b): r Μ

(16)

ι

l [+2] J > [+3]

b.

If A is distinct from B, and A Β on a prominence scale S then there is a PARSE constraint PARSE [P]*®

This licenses the PARSE constraints in (17): (17)

a. b. c. d.

PARSE PARSE PARSE PARSE

PER[+l]/[+3] PER[+2W+3] PER[+1]/t+2] PER[+2]/[+1]

Assuming that PARSE PER[+Noml/[+Acc] is dominated by BLOCK [P], we can now account for all the patterns in (15). Turkana (18a), Dumi (18b), Menominee (18c) and Quechua (18d): (18)

[BLOCK [P] » PARSE [P]] &

Γ [+1M+3] 1 [+2]/[+3] [ BLOCK [P] r [+i]/[+3] ι [+2]/[+3] b. [+l]/[+2] I BLOCK [P] f [+1M+3] 1 c. < [+2]/[+3] [+2]/[+l] BLOCK [P] a.

» [+Nom]/[+Acc] »

» i\ [+Nom]/[+Acc] J1

»J 1 \ [+Nom]/[+Acc] j

{ [ ™ }

326

Jochen Trommer

[+l]/[+3] [+2]/[+3] [+l]/[+2] [+Nom]/[+Acc]

» BLOCK [P] »

[+2]/[+l]

3.2. Further parameters of competition For implementing the effect of the person hierarchy in Turkana, I have assumed constraints such as (19a), but equally well we could take the slightly different (19b): (19)

a.

b.

PARSE [P]™+2> Realize agreement of a [+1] head in the context of a [+3] head PARSE [+l]/[+2] Realize [+1] agreement of a head in the context of a [+3] head

The outcome is identical in both cases, but it is not in others. Thus, to account for the fact that person agreement is with the nominative argument, we have to choose (20a) not (20b) since the agreement markers itself can realize person features of subjects and objects, and hence cannot be marked for a case feature: (20)

a.

b.

PARSE [P]t+Nom]/[+Accl Realize person agreement of a [+Nom] head in the context of a [+Acc] head PARSE [+Nom]/[+Acc] Realize [+Nom] agreement of a head in the context of a [+Acc] head

The null assumption is now that constraints for person as in (19) work in the same way, i.e., the correct formulation is (19a) not (19b). However, there is strong evidence against this assumption, as can be seen if we look at the Menominee person prefixes ne-:[+1], ke-\[+2] and o-:[+3]. If there is a [+2] argument (and no [+1] argument), ke-

Direction Marking as Agreement

327

appears: In a parallel fashion, ne- appears if one of the arguments is [+1] (and none [+2]). Now, there are two situations where both items would be licensed. In transitive forms where one argument is 2nd and the other 1st person and in forms with an inclusive ([+1 +2]) plural. In both cases, ke- appears: (21)

ke-pose-q 2-embark-lpl 'we (inc.) embark'

(Bloomfield 1962: 150)

However, (22b) in this case will not lead to the correct results since for [+1 +2] ke- as well as ne- realize agreement with a [+1] head. The requirement that [+1] (and hence ne-) appears is only captured by (22a): (22)

a. b.

PARSE [+2]/[+l] PARSE PER(+2W+I1

Thus I conclude that different features of prominence hierarchies involve slightly different types of PARSE constraints and (16b) has to be formulated more liberally: (23)

If A is distinct from B, and A Β on a prominence scale S then there is a PARSE constraint PARSE [ P ] ^ or PARSE A/B

Also BLOCK constraints seem not to be uniform. Thus, in Turkana, only person affixes seem to be subject to blocking while in Warlpiri only number affixes are involved. In Menominee, apart from the person prefixes, there are other affix types (person suffixes and number suffixes), which also involve blocking, but crucially there is no blocking between different affix types. Thus, there seem to be different BLOCK constraints referring to different affix classes. Ideally, all types of blocking should be reduced to a single constraint, but I leave this open for future research.4

328

Jochen Trommer

3.3. Advantages of the account Hierarchy-driven competition has only scarcely been treated in the literature. Other formal accounts such as Wunderlich (1996) for the Menominee prefixes, rely heavily on lexical stipulation. The typological literature only notes the phenomenon in passing. For example Croft (1990: 113) writes: "In a number of languages found scattered around the world, the transitive verb agrees not with the subject (A), or the absolutive (P), but whichever of A and Ρ is higher on the person hierarchy."5 The account proposed in this section improves in several respects over this rough characterization: First, it seems not to be true that blocking in general involves agreement as a whole.6 As has been shown for Turkana, it often targets only specific types of agreement. Second, the constraint-based account relates hierarchy-based competition to more common systems. Both agreement types simply emerge from different constraint rankings. Finally, systems that show partial violation of BLOCKING, such as Quechua can also be accounted for by constraint ranking.

4.

Competition among Direction Markers

In this section, I take Menominee to show how the distribution of direct and inverse markers in a single language also follows from hierarchy-based competition. For reasons of space, I will only treat two of the five direction markers in Menominee, but the analysis extends straightforwardly to the missing markers (see Trommer 2001 for details). Recall from (1) in section 1. that -a- is used if the subject is 1st or 2nd person, and the object is third person, while -eko is used in the converse constellations. The first question is now how to represent direction markers. Recall the assumption from section 2. that affixes encode a subset of the syntactic features they interpret. Since the typical distribution of direction markers is in transitive verb forms with person/number affixes that are not specified for case, it is natural to assume that

Direction Marking as Agreement

329

direction markers express just the case features left unexpressed by other affixes and have roughly the form in (24): (24)

[+Nom ...][+Acc ...]

Assuming the constraint PARSE [Case], this explains why direction affixes must appear. The presence of [+Acc] also ensures that they only appear in transitive contexts. But additional feature specifications are needed to characterize different direction markers in a given language such as -a- and -eko in Menominee. Note first that the distribution of these markers is actually much more complex than stated above, as shown in (25). Both markers appear in combinations with an "unspecified actor'" ([3 -spec +an]), and in combinations of inanimates ([3 -an]) with other 3rd person arguments. Further, if both arguments of the verb are 3rd person animate, direction marking is sensitive to the contrast between proximate ([3 -obv +an]) and obviative ([3 +obv +an]) NPs, where "proximate" corresponds roughly to NPs referring to topic information and "obviative" to NPs introducing new discourse referents. In transitive predications, either the subject or the object (but not both) are obviative. Apart from the unspecified actor case, which has no corresponding patient category, -a- represents the mirror image of -eko. (25)

Distribution of -a- and -eko. -a-

[1/2 +an] [3 -spec +an] [3 -obv +an] [3 -obv +an] [3 +obv +an]

-> -> -> -> ->·

-eko

[3] [3 +spec] [3 +obv +an] [3 -an] [3 -an]

[3] [3 -spec +an] [3 +obv +an] [3-an] [3 -an]

- > - > — >

— >

[1/2 +an] [1/+2 +an] [3 +obv +an] [3 -obv +an] [3 +obv +an]

A crucial generalization emerges from (25): Whenever -a- is used, the subject is [+an]; if -eko appears, the object is [+an]. Since this feature is not realized by any other agreement affix in Menominee, it

330

Jochen Trommer

is plausible that it is also part of the specification of the direction markers as in (26): (26)

-a-eko

[y b.

If A is distinct from B, and A Β on a prominence scale S then there is a PARSE constraint PARSE f+an]^

This leads among others to the PARSE constraints PARSE [+an][+3 -specM+obv] a n d Ρ A R S E [+an][+1M+3 "specl, which ensure that -a- is chosen in unspecified actor constructions with another 3rd person argument, but -eko, if the other argument is 1st person. In a similar way (31a) ensures the correct distribution of both markers in all relevant cases."

5.

Zero Direction Marking

Even direction-marking languages do not necessarily mark all transitive predications by direction markers, i.e. for many constellations the direction markers are zero. Interestingly, zero marking shows an asymmetry between direct and inverse markers: There are languages with inverse markers and without direct markers but no languages with direct markers only (cf. Croft 1990: 137). Thus, in Turkana no direction marking obtains for 1/2 3 predications:

332

(32)

Jochen Trommer

a.

ä-mm-ä

1-love-ASP Ί love her' b.

(Dimmendaal 1983: 69)

k-ä-mn-ä

D-l-love-ASP 'she loves me'

(Dimmendaal 1983: 123)

Now, assuming that k- is specified maximally simply, as [+Nom] [+Acc], and that its appearance is favoured in all transitive forms by PARSE constraints, it is natural to assume a counter-constraint that prevents its appearance in direct contexts. This cannot be a BLOCKING constraint, since all affixes that could block k- in (31a) should also block it in (31b). Hence I propose to use a different constraint type, which I call IMPOVERISH according to the impoverishment rules from Halle and Marantz (1993): (33)

IMPOVERISH [+Nom][+Acc] / [+Nom -3][+Acc +3]

This means that, no [+Nom][+Acc] affix should occur in a form which corresponds to the underlying feature structures [+Nom -3] and [+Acc +3]. If all relevant IMPOVERISHMENT constraints for [+Nom] [+Acc] do as (33) and refer to direct constellations, this account predicts the observation that there are direction-marking languages with only inverse markers but none without. To see this, assume for the moment that (33) is the only relevant IMPOVERISHMENT constraint, abbreviated in the following as "IMPOVERISH CASE", and look at the two possible rankings for this and PARSE CASE. If IMPOVERISH CASE is ranked higher than PARSE CASE, we get an inverse language (only inverse markers): (34)

Inverse language/inverse: [+Nom +3][+Acc -3] IMPOVERISH CASE / [-3] [+31

a. «y I-V b. V

PARSE CASE *!

Direction Marking as Agreement

(35)

Inverse language/direct: [+Nom -3][+Acc +3] IMPOVERISH CASE / [-3][+3]

a. b.

333

D-V V

PARSE CASE

*!

Ä

*

If the ranking is reversed, a full direction marking language emerges (inverse and direct markers): (36)

Direction language/inverse: [+Nom +3][+Acc -3] PARSE CASE

a. t+3i If A is distinct from B, and A Β on a prominence scale S then there is a IMPOVERISH constraint IMPOVERISH [+Nom][+Acc] / [+Nom A][+Acc B]

If all resulting constraints are ranked below PARSE [CASE] we get a language where all transitive predicates have a direction marker (such as Menominee, see section 4). If single IMPOVERISHMENT constraints are ranked higher, only the corresponding inverse configurations will be marked. This predicts i.a. languages where

334

Jochen Trommer

direction is marked in 3 -> 1/2 and 2 -» 1 predications, but not in any other. This seems to be the case in Nocte (Gupta 1971). Other languages make reference to different prominence scales. Thus, in Kutenai (Dryer 1994), direction marking appears only for different 3rd person arguments (obviative and proximative, see section 4. for discussion of these features). Note that also markers not traditionally thought of as direction markers fall under this characterization. Thus in Ancash Quechua (Lakämper and Wunderlich 1989) there is one portmanteau marker for 2 -> 1. In my account there is no formal difference between portmanteau affixes for subject and object agreement and direction markers. Thus, this instantiates the case where IMPOVERISH CASE/[+l][+2] and IMPOVERISH CASE/[+3][+l/+2] are ranked above while IMPOVERISH CASE/[+2][+l] is ranked below PARSE [CASE], There is ample evidence that direction markers mark agreement. In the most extreme form this leads to large paradigms of direction markers as in Arizona Tiwa (Klaiman 1993), with different markers for specific number and person features. In other languages (e.g. many Algonquian languages) there are special direction markers for 2 -> 1 and 1 -> 2 predications bleeding the direction markers used elsewhere in these languages. Thus it is obvious that direction markers encode agreement features. The question is, whether they additionally refer to feature hierarchies. I suggest that they do not and that the hierarchy effects are captured by IMPOVERISHMENT constraints, which are independently necessary to account for other aspects of agreement morphology Thus in Menominee, the [-3] marker -m which marks the presence of 1st or 2nd person arguments is suppressed in lsg -> 2sg forms: (39)

a.

b.

ke-natom-enene-m-enaw 2-call-D-[-3]-lpl 'we call you (sg./pl.)' (Bloomfield 1962: 157) ke-na tom-enene (kena tomen) 2-call-D Ί call you (sg.)' (Bloomfield 1962: 157)

Direction Marking as Agreement

335

Again, this cannot be the effect of a surface filter since the context of (39b) is a subset of the contexts of (39a); thus, everything that is blocked in a. should also be blocked in b. Assuming IMPOVERISHMENT, the data can be accounted for by a high-ranked IMPOVERISH [-3]/[+l +sg][+2+sg].

6.

Alternative Analyses

Functional approaches to direction marking assume that direction markers in some way encode the "well-behavedness" of the alignment between Subject/Object and animacy features. As Aissen (1999) notes, this implies - given the cross-linguistic diversity of direction markers - the problematic assumption of language-specific hierarchies. What is worse, hierarchy-based competition is often driven by slightly different hierarchies than direction marking. Thus in Blackfoot (DeLancey 1985: 643), 2nd person is ranked higher than 1st person for prefix selection, while 2 1 verbs carry the inverse marker. This suggests the hierarchy ... 2 > 1 ... for prefixes and ... 1 > 2 ... for direction markers. Hence, in an account that relies on direct reference to feature hierarchies in the grammar of single languages, the hierarchies must in principle be not only languagespecific, but also construction-specific. A constraint-based account avoids this un-desirable consequence, while incorporating the insight that prominence hierarchies govern the distribution of direction markers.9 Aissen's OT-analysis of direction marking (Aissen 1999) is much more limited in empirical domain than the one proposed here and problematic in several ways. To see this, I briefly discuss, how she would treat a language similar to Turkana which marks 3 -> 1,2 predicates through an inverse marker, and leaves all other predications unmarked. At the heart of an Aissen-style analysis of such a language is the fixed constraint ranking in (40): (40)

*0D & *Subj/3 & *Obj/l,2 » *0 D & *Subj/l,2 & *Obj/3

336

Jochen Trommer

*0 D marks non-realization of the direction category, *Subj/3 a 3rd person subject and *Obj/l,2 a 1st or 2nd person object. Thus the first constraint in (40) (*0 D & *Subj/3 & *Obj/l,2) which is formed from these three constraints by local conjunction (indicated by "&") marks the situation, where the subject is 3rd person, the object lst/2nd and there is no direction marking. (40) can hence be paraphrased as in (41): (41)

Mark Direction for 3

1,2 » Mark Direction for 1,2 —> 3

These two constraints interact with the economy constraint *StructD which marks direction marking in general. There are three possible situations: (42)

a. b. c.

*StructD » 3:1,2 » 1,2:3 => no direction marking 3:1,2 » *StructD » 1,2:3 => Inverse, but no direct marking 3:1,2 » 1,2:3 » *StructD => Inverse and direct marking

As long as 3 -> 1,2 and 1,2 -> 3 are not reranked, this derives the same typology as the one proposed in section 5. The ranking of the constraints in (41) is systematically related to hierarchies by a technique called harmonic alignment, which derives fixed constraint rankings from prominence hierarchies. Thus the role of harmonic alignment is roughly the same as for the statements in (16) and (38) in my approach. While this account partially derives the same results as my analysis it comprises several limitations and problems. First, as Aissen notes herself (Aissen 1999: fri. 21), her account does not extend to languages with inverse and direct marking, such as Menominee. Second, in her account, it is completely unclear what a direction marker formally is. In the functionalist literature a direction marker is supposed to be a marker that encodes inverse or direct configurations with respect to language specific feature hierarchy, but if this would be correct the constraints on its distribution would be unnecessary. Third, Aissen does not capture the systematic relation between direction markers and the caseless nature of other agreement morphology

Direction Marking as Agreement

337

in direction marking languages. This of course follows from the uncertain formal nature of direction markers in her analysis. Finally, Aissen has to stipulate a fixed order of constraints. While this is common practice in the OT literature, it goes against the spirit of OT, where constraints are supposed to be freely rankable. Of course, it remains to be seen, if other analyses stated in terms of harmonic alignment can be rephrased without fixed constraint orders.

7.

Summary

In this article, I have analyzed direction markers as agreement affixes. The distribution of direction markers crosslinguistically was accounted for by different rankings of universal but violable constraints, which were linked to feature hierarchies. All assumed constraint types were independently motivated by their usefulness to account for the behaviour of simple agreement markers in direction marking languages. The proposed approach has been shown to be superior to other accounts of direction marking. Note that the proposed account obeys all the restrictive assumptions outlined in section 2: Thus, direction marking is formalized locally referring only to adjacent agreement heads, (locality), all assumed affixes encode subsets of the relevant syntactic features (inclusiveness), and all assumed constraints can be reranked leading to other possible grammars. The analysis hence supports a restricted version of OTmorphology.

Notes 1. The following abbreviations are used: Agr = agreement, Acc = accusative case, an = animate, D = direction marker, I = inverse marker, inc. = inclusive (plural), Nom = nominative case, Num = number, obv = obviative, Ρ = person, pi = plural, sg = singular, spec = specified (actor). 2. Agreement with subjects of transitive clauses will be represented in the following by [+Nom] and object agreement by [+Acc]. Obviously, this notation has to be refined to extend to ergative languages.

338

Jochen Trommer

3. A similar constraint type ("the monosuffix constraint") is assumed by Aronoff and Fuhrhop (2002) to account for coocurrence restrictions in English derivation and inflection. 4. Note that in standard Indoeuropean languages, where person and number agreement is fused, and there is no direction markings, different blocking constraint will conspire to reduce agreement to one single affix. 5. With reference to Tangut, which functions roughly along the lines of Turkana. 6. But see footnote 4. 7. In "unspecified actor forms", the subject is unspecified in a passive likemanner. Bloomfield indeed calls these forms passives, and I will follow him here in the translations. 8. Note that 1 —> 2 and 2 -> 1 predicates have different direction markers (cf. Trommer 2001 for discussion). 9. An anonymous reviewer notes that the functional approach predicts correctly that inverse marker are phonologically longer than direct markers, while the constraint-based account does not. While this observation seems to be correct in many cases, there are some problems. Thus the shortest direction marker in Menominee is -e which is used in forms with 1st person objects. Since at least part of these forms (3 1) are clearly inverse, and also 2 -> 1 forms are at least ambiguous between direct and inverse (DeLancey 1985: 644), -e should be at least as long or longer as the direct marker -a . But obviously it is the other way around.

References Aissen, Judith 1999 Markedness and subject choice in optimality theory. Natural Language and Linguistic Theory 17: 673-711. Aronoff, Marc and Nanna Fuhrhop 2002 Restricting suffix combinations in German and English. To appear in: Natural Language and Linguistic Theory. Bloomfield, Leonard 1962 The Menomini Language. New Haven: Yale University Press. Comrie, Bernard 1980 Inverse verb forms in Siberia. Folia Linguistica Historica 1: 6174. Croft, William 1990 Typology and Universals. Cambridge: Cambridge University Press.

Direction Marking as Agreement

339

DeLancey, Scott 1985 An interpretation of split ergativity and related patterns. Language 51: 626-657. Dimmendaal, Gerrit Jan 1983 The Turkana Language. Dordrecht: Foris Publications. Dryer, Matthew S. 1994 The discourse function of the Kutenai inverse. In: Givon, Talmy (ed.), Voice and Inversion, 65-100. Amsterdam: John Benjamins. Gupta, Das 1971 An Introduction to the Node Language. Shillong: North-East Frontier Agency. Halle, Morris and Alec Marantz 1993 Distributed Morphology and the pieces of inflection. In: Hale, Kenneth and Keyser, Samuel Jay (eds.), The View from Building 20,111-176. Cambridge, M.A.: MIT Press. Hockett, Charles F. 1966 What Algonquian is really like. International Journal of American Linguistics 321: 59-73. Klaiman, Miriam H. 1993 The relationship of inverse voice and head-marking in Arizona Tewa and other Tanoan languages. Studies in Language 17: 343370. Lakämper, Renate and Dieter Wunderlich 1989 Person marking in Quechua - a constraint-based minimalist analysis. Lingua 105: 113-148. Prince, Alan and Paul Smolensky 1993 Optimality theory: Constraint interaction in generative grammar. Ms. Technical reports of the Rutgers University Center of Cognitive Science, New Brunswick. Trommer, Jochen 2001 Distributed Optimality. Ph.D. dissertation, University of Potsdam. Wunderlich, Dieter 1996 A minimalist model of inflectional morphology. In: Chris Wilder, Manfred Bierwisch and Hans-Martin Gärtner (eds.), The Role of Economy Principles in Linguistic Theory, 267-298. Berlin: Akademie-Verlag.

On the semantics of cases* Ilse Zimmermann

The present study is concerned with the semantics of cases in Russian. Within a minimalist framework of sound-meaning correlation, it is argued that structural cases of complements can be characterized by abstract semantico-syntactic features which correspond to the semantic hierarchy of argument expressions and which are systematically interrelated with their morpho-syntactic case realizations. As to cases of adjuncts, the analysis focuses on the semantics of the instrumental. It is shown how morpho-syntactic case features of adjuncts get a semantic interpretation and how one can account for the polysemy of the instrumental by assuming context-dependent specifications of semantic parameters.

1.

Objectives

On the basis of data from contemporary noncolloquial Russian, this paper is concerned with the semantics of cases. It tries to bring together recent developments of the two-level semantics, the linking theory of Lexical Decomposition Grammar (LDG), as elaborated by Dieter Wunderlich and his colleagues (Wunderlich 1997a, Stiebels 1996, 2000a) and Roman Jakobson's case characterizations (Jakobson 1936, 1958). The main concern will be complements of verbs and of event nominalizations with structural cases and adjuncts in the instrumental. The particular questions to be raised are the following:

342

Ilse Zimmermann

- How are case forms of complements and adjuncts interrelated with the semantics of these constituents? - Which types of cases are to be discriminated? - Which types of configurations and of case features are involved? - Which complements of lexical categories count as structural arguments? - Which rules guarantee the correct case realizations of argument expressions? - How do adjuncts get their cases and how are they interpreted semantically? - How can one cope with the polysemy of adjunct cases? An argument will be made for a strict differentiation of universal semantico-syntactic and language-specific morpho-syntactic case features and for the existence of regular correspondences between them. As for the semantics of cases, it will be shown that structural cases of complements have a systematic semantic background and that for the morpho-syntactic cases of adjuncts we have to assume semantic parameters which are specified at the level of Conceptual Structure. The paper is organized as follows: Section 2 characterizes the theoretical framework, first of all the division of labour between morphology, syntax and semantics and the interface role played by the argument structure of lexical entries of functor expressions. Section 3 demonstrates the far-reaching parallelism of verbal constructions and their nominalizations, with systematic case variation of structural arguments. Section 4 is concerned with case licensing. It presents a system of rules correlating abstract semanticosyntactic case features in the argument structure of lexical governors and the morpho-syntactic case features of the corresponding argument expressions. Section 5 concentrates on the semantics of the instrumental as adjunct case. Section 6 is a summary. Sections 2-4 are a shortened version of Zimmermann (2002).

On the semantics of cases

2.

343

The framework

Within a minimalist framework of sound-meaning correlation (Chomsky 1995), the analysis follows a lexicalist conception of morphology (Stiebels and Wunderlich 1994, Wunderlich and Fabri 1995, Wunderlich 1997b) and the differentiation of Semantic Form and Conceptual Structure (Bierwisch 1983, 1987, 1997, Bierwisch and Schreuder 1992, Lang 1987, 1990, 1994, Dölling 1997). I assume Phonetic Form (PF), Logical Form (LF) and Semantic Form (SF) as relevant grammatically determined levels of representation. My conception of syntax is very restrictive (cf. Jacobs 1995). For sentences and DPs, I presuppose the structural layers in (1) and (2), respectively. (1)

CP MoodP TP NegP vP* VP

(2)

DP FP nP* NP

In the base structure, argument expressions with structural cases of verbs and of the corresponding deverbal nouns are placed in Spec VP, SpecvP or in SpecNP, SpecnP, respectively. DPs with nonstructural cases and PPs appear in the complement position. The verb raises to Mood or to C (Zimmermann 1999a) and—in parallel to sentence structures—the deverbal noun overtly moves to a high functional projection F (Alexiadou 1999), so that all argument expressions of Ν will be to its right (Haider 1992). I will not discuss the nature of the category F. Possibly, it is a further n. Adjuncts can be integrated at all projections (Maienborn 1996, 1997, 2000). The syntactic configurations at the level of LF are the input for semantic interpretation. This implies that syntactic movements of constituents can have an effect in SF (Zimmermann 1999a). For functor expressions like verbs and their nominalizations this means that they are combined with their arguments semantically on the basis of LF configurations where chains with traces of moved argument expressions must be taken into consideration (see (3)). In such derived structures, the head of the chain, the case bearing argument

344

Ilse Zimmermann

expression DPj occupies some derived position whereas the tail of the chain tj is in the complement or specifier position of V, ν, Ν or n. (3)

(DPi, ..., tO

The lexical entries for functor expressions like verbs and their nominalizations include in their argument structure grammatical requirements which must be fulfilled by the respective argument expressions. I call these requirements grammatical argument addresses Gj. They are associated with lambda operators λχί which represent the argument positions of the respective functor expression. (4)

λχ η ... λχί [ . . . χι ... x n ... ] Gn Gi argument structure predicate-argument structure

The argument positions λχί are ordered from right to left according to the relative depth of embeddedness of the arguments Xj in the predicate-argument structure. The highest argument xi of verbs and event nominalizations constitutes the referential argument (Williams 1981, Bierwisch 1989, Bischof 1991). For mnemonic reasons, I will represent it for verbs and event nominalizations as s (referring to situations).1 The other arguments are participant, propositional or predicate arguments. (5)

λχ η ... λχχ Xs [ . . . s ... χι ... x n ... ] Gn Gi with s e e, x, e {e, t }

λχί in (5) represents the argument position of the external argument, λχ η is the argument position of the lowest internal argument. For DP arguments, the grammatical features Gi are case requirements (Zimmermann 1967) which must be fulfilled by the corresponding DPs as heads in L F chains. So we have the following hierarchy of argument positions:

On the semantics of cases

(6)

345

internal < external < referential

This hierarchy corresponds to the relative adjacency of the argument expressions to the verb or its nominalization, from left to right. The lowest internal argument expression or its trace is the immediate neighbour of the functor expression in its base position. Higher argument expressions are not adjacent to the governor in the underived structure. The hierarchy of arguments is also valid for the classification of argument expressions with respect to their predictable or nonpredictable case forms. Typically, the lowest internal argument expression can have idiosyncratic (lexically determined) case marking.2 As regards the nature of case requirements of argument expressions, I assume the following: Cases are understood as analysable morpho-syntactic characteristics of inflected nouns, pronouns, adjectives, determiners and quantifiers. These characteristics are represented by a small number of morpho-syntactic case features, which allow morphological cross-classification and underspecification (see section 4). I assume that idiosyncratic case requirements of argument expressions are directly referred to by these case features as instantiations of Gj in the argument structure of lexical governors (see the examples in (7)-(ll)). In contrast, predictable case forms of argument expressions are understood as structural cases. They are conditioned by the semantic argument hierarchy. Wunderlich (1997a, 1999), Stiebels (1996, 1997, 2000a, 2000b, 2001) and Wunderlich and Lakämper (2000) posit configurational case features for complements of verbs and nouns in general, without discriminating between semantico-syntactic and morpho-syntactic feature systems. Their features are +/-hr (there is a/no higher role), +/-lr (there is a/no lower role), as proposed similarly by Kiparsky (1992). I will restrict the use of these features to the characterization of argument positions for structural arguments in the argument structure of lexical governors and correlate them by correspondence rules to the morpho-syntactic case features of the pertinent argument expressions. Also with respect to German, I regard morphological case classifications in the same way as, for instance, Bierwisch (1967) or Gallmann (1998),

346

Ilse Zimmermann

namely as autonomous morpho-syntactic qualifications which are not reducible to a system of semantico-syntactic case features. It seems necessary to distinguish the following case types: • Structural cases are predictable case forms of DPs as argument expressions. They correspond to semantically conditioned abstract case features +/-hr, +/-lr as argument addresses in the argument structure of functor expressions. • Lexical cases are unpredictable case forms of DPs idiosyncratically required for an argument expression by the respective governor. They are represented by language-specific morphologically conditioned features. • Semantic cases are morpho-syntactic case feature bundles of DPs which receive a semantic interpretation.

3.

Verbs and event nominalizations as governing heads and their structural arguments

In the following, verbs and event nominalizations are analysed with respect to the case form of their DP arguments. Of main concern are DPs with structural (i.e., predictable) cases, in short structural arguments. It will be shown which case requirements of argument expressions remain unchanged in nominalizations and which arguments receive alternative case realizations.

3.1. The argument structure of verbs and their nominalizations Words as syntactic atoms are fully inflected items. They enter syntactic representations with all affixes of word formation and inflection. With Bierwisch (1989) and Bischof (1991), I assume that nominalizations of verbs—at least in German and in Russian—are derived morphologically and do not constitute products of syntactic rules.3 Verbs and event nominalizations have the same semantic charac-

On the semantics of cases

347

terization. The respective nominalizing suffix simply converts the verb into a noun without changing the SF.4 In (7)-(ll), I give some lexical representations of verbs and event nominalizations with semantic and morpho-syntactic information which is relevant for the case realization of argument expressions. The predicate-argument structure is unanalysed. Positions for structural arguments are associated with abstract case features +/-hr, +/-lr, which predict the admissible systematic case forms of argument expressions, depending on the syntactic category of the respective governor. Idiosyncratic case requirements are represented by morpho-syntactic features, which will be analysed in section 4. The case forms of the corresponding argument expressions are indicated—for convenience—by traditional case names. In (7)-(9), case information is systematic, redundant and therefore omissible. In contrast, the internal argument of the lexical entries in (10)-(11) idiosyncratically shows up in the dative and in the instrumental, respectively. Here one has to do with unsystematic lexical case which must be learnt. (7)

vyzdorovet' 'recover' / vyzdorovlenie 'recovery' vozniknut' 'emerge' / vozniknovenie 'emergence' λχ λβ [ . . . s ... χ ... ] -hr -lr V: nom N: gen

(8)

usvoit' 'acquire' / usvoenie 'acquisition' znat 'know' / znanie 'knowledge' λy λχ λβ [ . . . s ... χ ... y ... ] +hr -hr -lr +lr V: acc nom N: gen instr

348

Ilse Zimmermann

(9)

soobscit' vrucit'

'inform' / soobscertie 'information' 'hand in' / vrucenie 'handing in' λζ λy λχ Xs [ . . . s ... χ ... y ... ζ ... ] +hr +hr -hr -lr +lr +lr V: acc dat nom N: gen dat instr

(10) izmenit' pomoc'

Xy +R +P V: dat N: dat

'betray' / izmena 'betrayal' 'help' / pomosc 'help' λχ λβ [ . . . s ... χ ... y ... ] -hr -lr nom gen

(11) obmenjat'sja

'exchange' / obmen 'exchange' 'be engaged' / zanjatie 'engagement' λy λχ λβ [ . . . s ... χ ... y ... ] +Ρ -hr -lr V: instr nom N: instr gen

zanimat'sja

In the last two examples, the external argument position λχ is associated with the abstract case features -hr, -lr although there is a lower argument. Due to its idiosyncratic case requirements, the argument position λy is invisible for the feature specifications +/-hr and +/-lr. So λχ is characterized as the lowest structural argument position. Furthermore, the two examples illustrate that idiosyncratic case requirements of argument expressions of verbs are inherited by event nominalizations. The same is true for the structural dative, as illustrated by (9). Bayer, Bader and Meng (2001) observe that dative complements behave like complements with idiosyncratic case marking insofar as

On the semantics of cases

349

they do not alternate with nominative or genitive phrases in passive, middle and nominalised constructions, in contrast to argument expressions in the accusative. This case resistance of dative complements shows up in Russian nominalizations of ditransitive verbs even more consistently than in German where the dative phrases alternate with PPs. In Russian, the dative is preserved in event nominalizations (see (9) and (20)). Bayer, Bader and Meng (2001) classify only the nominative and the accusative as true structural cases, which characteristically undergo alternations in function changing operations. The corresponding phrases are categorized as DPs (or as QPs) whereas dative and genitive phrases are analysed as KPs (case phrases) constituting barriers for various syntactic operations. I cannot see whether this categorial differentiation could serve as a possible explanation for the different case forms of argument expressions in function changing operations. I regard the dative of complements of ditransitive verbs and of their nominalization and also the genitivus subjectivus and the genitivus objectivus in nominalizations as structural cases and represent the respective phrases as DPs. Alternating case realizations of structural arguments are connected with the external argument position λχ in (7)-(ll) and with the lowest position for a structural argument, λy in (8) and λζ in (9). The accusative complement of transitive and ditransitive verbs corresponds to the genitive complement of the deverbal noun whereas the nominative argument of these verbs shows up in the instrumental of event nominalizations. The nominative argument of intransitive verbs as in (7), (10) and (11) appears in the genitive, in nominalizations. These regularities concerning alternating case forms of structural arguments must be captured by the rules which interrelate the abstract semantico-syntactic case features +/-hr, +/-lr of structural argument positions with the morpho-syntactic case features of the respective DPs as argument expressions (see section 4).

350

Ilse Zimmermann

3.2. Reflexive verbs and their nominalizations As can be seen from (11) and (12), Russian nominalizations do not allow the combination with the reflexive morpheme -sja, in contrast to Polish (cf.formowac (sie) / formowanie (sie) 'form / formation') and to Nahuatl (Stiebeis 1997).5 (12) a.

Vertolet prizemlilsja. helicopter-NOM landed 'The helicopter landed.'

b.

Vertolet prizemljalsja / byl prizemlen helicopter-NOM was landed (odnim passazirom). one passenger-instr 'The helicopter was landed (by one of the passengers).'

c.

prizemlenie vertoleta (odnim passazirom) landing helicopter-GEN one passenger-INSTR 'the landing of the helicopter (by one of the passengers)'

The deverbal noun prizemlenie 'landing' in (12c) is ambiguous. It corresponds to the reflexive verb prizemljat'sja / prizemlit'sja 'land' in (12a) and to the transitive verb prizemljat' / prizemlit\ which in (12b) is passivised, without any marking of reflexivity. I assume that the reflexive formative -sja is added at the right end of verb forms on the basis of a morpho-syntactic feature +refl. This feature corresponds to the reflexive pseudoargument in German which, too, is restricted to verbal constructions. What is crucial for the present investigation of structural arguments is the observation that reflexive elements are always correlated with the lowest structural argument position, i.e., with an absent true argument in the structural accusative, and that in passive, middle and anticausative constructions they are accompanied by the absence of the external argument.6 In addition to (12a) with an anticausative reflexive verb, (13) is a case of reflexivum tantum, which characteristically does not combine

On the semantics of cases

351

with an accusative argument. (14) illustrates the reflexive passive of an imperfective verb, and (15) the middle. (13) Mal'cik smeetsja. boy-NOM laughs 'The boy is laughing.' (14) Plan razrabatyvaetsja. plan-NOM is worked out 'The plan is being worked out.' (15) Kniga xoroso prodaetsja. book-NOM well sells 'The book sells well.' In all these constructions, the reflexive exponent is added on the basis of the morpho-syntactic feature +refl of the respective verb. In nominalizations, this marking does not apply. In (16)-(17), I give complex lexical representations for two causative/anticausative verb pairs and their event nominalizations.7 These lexical entries respect the far-reaching correspondence of reflexivity and transitivity of verbs. The presence of the reflexive formative -sja on the verb corresponds to the absence of a structural accusative argument. As in German, the deverbal noun of such pairs is systematically ambiguous. (16) prizemlit'(sja)-a 'land' /prizemlenie 'landing' Xy (λχ) α λβ [ . . . s ...(... χ ...) α ... y ... ] ahr -hr -lr +lr V: acc nom N: gen instr V: nom N: gen

352

Ilse Zimmermann

(17) obucit'(sja).a 'teach, learn' / obucenie 'teaching, learning' λζ λγ (λχ) α λβ [ ... s ... (... χ ...)α ··· Υ ··· ζ ··· ] +R ahr -hr +Ρ -lr +lr V: dat acc nom N: dat gen instr V: dat nom N: dat gen

3.3. Examples The following noun phrases with deverbal heads illustrate the case realizations of the pertinent argument expressions, in contrast to infinitival phrases. The examples are given with normal word order. Derived word order variations, I do not consider here. It is important to notice that Russian nominalizations preserve the order of the argument expressions relative to the lexical governor in its base position. In contrast to German, the genitival complement need not be adjacent to the noun. In the nominalizations, the de verbal noun precedes the highest structural argument expression. This results from head movement of Ν to F (see section 2, (2)). (18) a.

b.

vyzdorovlenie pacienta recovery patient-GEN 'the recovery of the patient' vyzdorovet' 'recover'

On the semantics of cases

(19) a.

b.

353

znanie rebenkom jazyka knowledge child-INSTR language-GEN 'the knowledge of the language by the child' znat' jazyk know language-ACC 'know the language'

(20) a.

nemedlennoe soobscenie institutami firme immediate informing institutes-INSTR firm-DAT svoix zakazov their orders-GEN 'the institutes' immediate informing the firm of their orders'

b.

nemedlenno soobscit' firme svoi zakazy immediately inform firm-DAT their orders-ACC 'inform the firm immediately about their orders'

(21) a.

b.

obmen tovariscej opytom exchange comrades-GEN experience-INSTR 'the exchange of experience by the comrades' obmenjat'sja opytom exchange experience-INSTR 'exchange experience'

354

Ilse Zimmermann

(22) a.

obucenie (mater'ju) rebenka teaching / learning mother-INSTR child-GEN cteniju reading-DAT 'the teaching of reading to the child by the mother' / 'the learning of reading by the child'

b.

obucit' rebenka cteniju teach child-ACC reading-DAT 'teach the child reading'

c.

obucit'sja cteniju learn reading-DAT 'learn reading'

3.4. The structural instrumental As is apparent from the examples, the instrumental of the external argument in nominalizations is accompanied by the argument in the genitivus objectivus. With Bischof (1991), I regard this instrumental as structural case. It should not be confused with the agentive instrumental phrase in passive constructions. (19a) illustrates a nominalization of a Stative verb which does not have any passive. Nevertheless, the external argument appears in the instrumental. Moreover, (20a) demonstrates that the argument in the instrumental binds the reflexive pronoun svoj-, which is not possible for instrumental phrases in passive constructions. Cf.: (23) Nemedlenno soobscalis' firme institutami immediately were informed firm-DAT institutes-INSTR * svoi zakazy. their orders-NOM 'The firm was informed immediately by the institutes about their orders.'

On the semantics of cases

355

Thus, I differentiate between the structural instrumental of the external argument in event nominalizations as in (19a), (20a) and (22a), the lexical instrumental of the internal argument as in (21a, b) and the semantic instrumental of modifiers including the so-called argument adjunct of passive constructions.8

3.5. The structural genitive There are various kinds of genitives. Argument expressions and adjuncts can be marked by the genitive. Here, I will concentrate on the structured adnominal genitive of event nominalizations.9 As is evident from the lexical entries in 3.1. and 3.2. and from the examples in 3.2. and 3.3., the structural accusative of transitive and ditransitive verbs and the structural nominative of intransitive verbs correspond to the genitive of the pertinent argument expressions of the event nominalizations. This systematic correspondence of structural cases shows up in many languages. I assume that the adnominal structural genitive is associated with the following types of argument structures of nouns as functor expressions: (24) a.

b.

(...) λχ Xs [ . . . s ... χ ... ] -hr -lr gen (...) ky +hr -lr gen

λχ Xs [ . . . s ... χ ... y ... ] -hr +lr

356

Ilse Zimmermann

c.

λζ λy λχ λβ [ . . . s ... x ... y ... z ... ] +hr +hr -hr -lr +lr +lr gen

d. λy λχ +hr -hr gen

[ . . . χ ... y ... ]

(24a) represents deverbal nouns of intransitive verbs like vyzdorovlenie 'recovery', pomosc' 'help' and deadjectival nouns like zavisimost' 'dependence'. (24b) and (24c) characterize deverbal nouns of transitive resp. ditransitive verbs like usvoenie 'acquisition', soobscenie 'information', and (24d) represents relational nouns like mat' 'mother' and semantically enriched sortal nouns with a 'possessor' argument as in dom vraca 'house of the doctor'. I depart from Stiebels' assumptions on the argument structure of nouns (Stiebels 1997, 2000a, 2001) in following Bierwisch (1989) and Bischof (1991) with regard to the treatment of event nominalizations. I do not accept Stiebels' proposal to treat relational, semantically enriched sortal nouns and event nominalizations alike. In Stiebels' system, all non-highest argument positions of nouns are characterized as +hr, without any further differentiation. Thus, event nominalizations of ditransitive verbs get the following representation: (25)

λζ λy λχ λβ [ . . . s ... χ ... y ... ζ ... ] +hr +hr +hr -hr

Since the adnominal structural genitive is classified by Stiebels as +hr linker, all non-highest argument expressions of nouns could be genitive phrases. Several additional principles of case realization must help to avoid this. I do not need such additions. I assume that the argument structure of verbs is preserved in nominalizations, including complete information on structural cases (see (24a-c)).

On the semantics of cases

4.

357

Case realizations and case licensing

DPs can emerge with morpho-syntactic case information of the language-specific case system.10 It must be guaranteed that the case forms of DP-argument expressions fulfil the case requirements of their respective governor. Some licensing must take place. I will adopt the devices of Optimality and Correspondence Theory to account for this (Stiebeis 2000a, Wunderlich and Lakämper 2000). It must be emphasized that my conception of syntax does not assume any movements of a DP in order to get its case licensed. Movements of DPs obey requirements of scope and of information structure. Nevertheless, some principles must be at work to check the admissibility and co-occurrence of case forms of DPs as they are prescribed in the argument structure of the governing lexical head of the construction. I assume that correspondence rules and very general principles of argument realization relate the language-specific case characterizations of DPs to the pertinent requirements in the argument structure of the governor.

4.1. Case systems

Without going into the details of Russian nominal inflection, I presuppose the case system of Jakobson (1936) by giving his case qualifications the status of +/-valued features and by adding the feature obl(ique), whose specifications I borrow from Franks (1995):11

358

Ilse Zimmermann

(26) Jakobson's (1936) enriched case system R

Ρ

u

G

obi

nom acc

+

dat

+

instr

+

+

+

+

genl loci

+

gen2 loc2

+

+

+

+

+

+

+

+

+

+

+

with the correlations12 R=

Bezug (directional: "signalizing the goal of the event")

Ρ=

Rand (marginal: "assigning the entity an accessory place in the message")

U = Umfang (quantified: "focusing upon the extent to which the entity takes part in the message") G = Gestaltung (cases of shaping)13 As the features of Jakobson's system serve morpho-syntactic and semantic generalizations, it is no coincidence that LDG's features +/-hr, +/-lr roughly coincide with Jakobson's features +/-R, +/-P, respectively. Cf.:

On the semantics of cases

359

(27) The case system of LDG hr

lr

GEN

dir

instr

nom acc

+

dat

+

gen

+

+ +

It is essential to keep in mind that Wunderlich and his colleagues do not assume a special system of features for the morpho-syntactic case realizations of DPs. In their system of case linkers, the semantico-syntactic features +/-hr, +/-lr figure together with further differentiations of diverse origin. Characteristically, the additional features do not serve cross-classification. This is in sharp contrast to the system proposed in (26). Whereas the features +/-hr, +/-lr correspond to the semantic hierarchy of complements and allow us to characterize the nominative, accusative, dative and the ergative, the genitive needs further qualifications in order to be differentiated from the accusative. I put GEN in (27) (cf. Wunderlich 2000) provisionally as an additional characterization of the genitive. In Stiebels (2000a, 2001), the contextual categorial features -articulated, -dependent (i.e. N) proposed by Wunderlich (1996) make the feature specification +hr of the structural genitive dependent on a nominal governor. But the genitive as lexical case cannot be characterized in this way. It occurs with V, A and Ρ as governor in German, Russian and in other languages. Furthermore, Russian genitive phrases systematically alternate with the accusative of the direct object and the nominative of the subject in negated sentences (Jakobson 1936) and should be regarded as structural arguments. In view of this and of the correspondence of the structural genitive in event nominalizations with the nominative and

360

Ilse Zimmermann

accusative of verb complements, it seems questionable to have the qualification +hr (or +R of the system in (26)) for the genitive. The additional features dir, instr and others in (27) characterize so-called semantic cases in richer case systems, as, for instance, Hungarian. In sum, I cannot see how such an enlarged system of case features could deliver homogeneous criteria for cross-classification and generalizations. It seems more enlightening to keep morpho-syntax and semantics apart and to try to capture the existing correlations between the various subdomains of grammar by special rules and principles. My high estimation of Jakobson's (1936, 1958) case features first of all concerns the general approach and less the details. Furthermore, his semantic generalizations nowadays can be captured more adequately by assumptions about the semantic decomposition of functor expressions and about the semantic hierarchy of complements. Nevertheless, there are very fundamental and subtle observations in Jakobson's semantic characterizations of differences expressed by cases which deserve attention and recognition (see Demjjanow and Strigin 2000a, 2000b). The same is true with respect to the morphological generalizations offered by Jakobson's system of case features. It takes into account case inflection of nouns, adjectives, pronouns and numerals of Russian. According to the case system (26), all cases—except for the nominative as the most unmarked case—are identifiable by positively specified features. Thus, the markedness constraint (28) for lexical representations of morphemes, which has been proposed by Stiebels (2000a, 2000b), can be observed. (28) *-:

Avoid the negative specification of a morpho-syntactic feature in the lexical representation of morphemes.

Furthermore, at least some syncretisms can be characterized more or less economically (Zimmermann 2002). The conditions are more or less complex and demand principled limitations. I must leave it to further investigation which case syncretisms are systematic and

On the semantics of cases

361

which ones should be regarded as homophonous morphemes and to what extent case syncretism is a reliable criterion for the choice of case features (cf. Stiebels 2000a, Franks 1995). It should be clear from the foregoing assumptions on the system of morpho-syntactic case features for Russian that nouns, adjectives, pronouns and numerals can enter syntax underspecified for case distinctions. Their combination with other case-inflected entities can add case information. The same is true for the combination of DPs with their governors, which require specific cases for their complements. Agreement of nouns, adjectives, numerals and determiners and fulfilment of case requirements of a governor by the complements are the syntactic means of further case specification.

4.2. Feature

matching

For lexical cases, the matching of the case requirements of governors with the case linkers of the pertinent argument expressions is simply unification of the involved morpho-syntactic case features. For structural cases, it is necessary to interrelate the abstract case features +/hr, +/-lr in the argument structure of functor expressions with the morpho-syntactic case characterizations of the complements. I assume that the following correspondence constraints are at work to determine admissible case realizations of structural arguments in Russian, i.e., the structural nominative, accusative, dative for complements of verbs, the structural genitive 1, dative, instrumental for complements of event nominalizations and the structural genitive 1 for the complement of relational or semantically enriched sortal nouns: (29) Correspondence constraints for structural arguments a. b.

CORR(ahr) = a R CORR(alr) = a P

362

Ilse Zimmermann

The constraints take into account the semantic background of the morpho-syntactic case features +/-R, +/-P and the far-reaching correspondence of the feature systems (26) and (27). Other principles of argument realization are borrowed from LDG's system of constraints and adapted to my assumptions on morpho-syntactic case features. (30) Constraints of argument realization a.

MAX(+hr): A +hr argument position must be specified by an argument expression with a corresponding +R linker.

b.

MAX(+lr): A +lr argument position must be specified by an argument expression with a corresponding +P linker.

c.

MAX(+hr+lr): A +hr+lr argument position must be specified by an argument expression with a corresponding +R+P linker.

d.

MAX(lex): Lexical case requirements must be respected.

e.

DEP(F): A feature in the output has a correspondent feature in the input.

f.

DEFault: One structural argument must be realized in a default case, the nominative for verbs, the genitive 1 for nouns as respective governors.14

g.

UNIqueness: The argument expressions must show up in different morpho-syntactic cases.

h.

*+F: A positively specified morpho-syntactic case feature must be avoided.15

i.

*+R-obl / aV+N: Avoid the accusative for complements of nouns and adjectives.

j.

*-R-obl / -V+N: Avoid the nominative for complements of nouns.

On the semantics of cases

363

These principles and the correspondence constraints in (29) constitute a system of faithfulness, expressivity and economy constraints, which are ranked. They determine the degree of acceptability of case co-occurrences for argument expressions of lexical governors. Tentatively, I assume the following constraint ranking (cf. Wunderlich 2000): (31) Constraint ranking I MAX(lex), II ΜAX(+hr+lr) » ///DEF, IV UNI, V *+R-obl / aV+N, VI *-R-obl / -V+N » VII CORR(ahr), VIII CORR(alr), IX MAX(+hr) » X MAX(+lr), XI DEP(F), XII *+F The tableaus in (32)-(34) show the optimal decisions for znat' 'know' / znanie 'knowledge' (see (8)), izmenit' 'betray' / izmena 'betrayal' (see (10)) and for prizemlit'sja 'land' / prizemlenie 'landing' (see (16)). I must leave it to the reader to add in the tableaus less optimal case arrays (see Zimmermann 2002). The input to the tableau representations are the syntactic category of the governor and its nonreferential argument positions with abstract case features and—if necessary—with idiosyncratic case requirements of the pertinent argument expressions. Output of the representations are arrays of exclusively positively specified morphosyntactic case characterizations abbreviated by traditional case names (see table (26)). The syntactic configurations of the corresponding DP complements are assumed to be given in the respective nonderived structure, i.e., in the tail position of DP chains (see section 2). It should be clear that each tableau is applicable to classes of lexical governors which belong to the pertinent syntactic category and have the specified argument positions. (32) shows the evaluation of the linking patterns of a transitive verb and its derived nominal. With verbs, DEF is violated if the linking pattern does not include NOM, with nominals, it is violated if the linking pattern does not include GEN1. Since NOM is excluded

364

Ilse

Zimmermann

for nouns, INSTR becomes the optimal linker for the external argument because it makes +lr visible by means of its +P specification. GEN1, which is less faithful for the internal argument than ACC, is preferred in nominale due to the high ranking of DEF. (32) znat' 'know' / znanie 'knowledge' aV

-ccN

λΥ

λχ

+hr

-hr

-lr

+lr

acc

nom

N: genl

instr

V:

III

IV

V

VI ν π

vni

IX

X

XI XII

*

* *

3*

4*

High-ranked MAX(lex) ensures that lexical markings are preserved both in verbs and nominals. (33) evaluates the linking patterns of a verb and its nominalization with dative object. (33) izmenit' 'betray' / izmena 'betrayal' aV

-aN

λγ

λχ

+R

-hr

+P

-lr

I

ΠΙ IV

V

VI VII VIII IX X

XI XII

V: dat

nom

*

3*

N: dat

genl

3*

5*

The anticausative variant of verbs with a causative and an anticausative form and its nominalization are evaluated in (34). Note that the causative variant behaves like the functor expressions discussed in (32). Since the anticausative form is an intransitive verb, DEF rules

On the semantics of cases

365

out any candidate with a linker other than the category-specific default linker. (34) prizemlit\sja).$ 'land' / prizemlenie 'landing' aV -aN -ßrefl Xy

(λχ)β

ßhr

-hr

-lr

+lr

ΠΙ

IV

V

VI VII

IX

X

XI

XII

2*

2*

nom genl

It remains to be emphasized that the foregoing considerations and decisions should be considered as a first attempt at using Optimality and Correspondence Theory for the aims of explaining optimal case configurations of Russian argument expressions. Much empirical and theoretic work must be done for a deeper understanding of the interplay of morphology, syntax and semantics. The proposed constraints and their ranking must be related to constraints on word order variations and on the omissiblity of argument expressions. I hope to have made clear that structural cases are deeply based on the semantics of the governors and that they are category-dependent systematic case realizations of argument expressions licensed by a small number of constraints and their relative weight.

5. Semantic cases of adjuncts So far, we have considered structural cases of complements. Now, we will turn to cases of adjuncts. The main concern will be the instrumental (cf. Demjjanow and Strigin 2000a, 2000b). Syntactically, DPs in the genitive, dative, accusative and instru-

366

Ilse Zimmermann

mental can show up as adjuncts. Semantically, these phrases are modifiers. I do not assume any governor, for instance K(ase), Preposition), Pred(icate) and/or Mod(ifier), for adjunct DPs.16 Instead I assume that semantic templates are at work which associate the adjunct cases with very abstract relational meanings giving the adjunct DP the semantic status of predicates. Thus, they can serve as modifiers, like other adjunct phrases. A further semantic template serves the unification of arguments of the respective modifier and of the modificandum (Zimmermann 1988, 1992, 1999b, 2000), without mediation by any special syntactic category. In a sense, such templates can be looked at as phonetically silent phrasal affixes serving semantic accommodations of the pertinent constituents.17 (35) shows the operations which are involved in the interpretation of adjunct DPs with semantic cases. (35) ( MOD ( «e,t>,,