More than Words: A Festschrift for Dieter Wunderlich [Reprint 2015 ed.] 9783050081274, 9783050037592

Citation preview

Kaufmann, Stiebeis (eds.) M o r e than W o r d s

studia grammatica Herausgegeben von Manfred Bierwisch unter Mitwirkung von Hubert Haider, Stuttgart Paul Kiparsky, Stanford Angelika Kratzer, Amherst Jürgen Kunze, Berlin David Pesetsky, Cambridge (Massachusetts) Dieter Wunderlich, Düsseldorf

studia grammatica 53

Ingrid Kaufmann and Barbara Stiebels (eds.)

Ν\θΤβ f ^ p

V V O r d S

A Festschrift for Dieter Wunderlich

Akademie Verlag

Die Deutsche Bibliothek - CIP-Einheitsaufnahme More than words: a festschrift for Dieter Wunderlich / Ingrid Kaufmann; Barbara Stiebeis (eds.) - Berlin : Akad. Verl., 2002 (Studia grammatica ; 53) ISBN 3-05-003759-8

ISSN 0081-6469 © Akademie Verlag GmbH, Berlin 2002 Das eingesetzte Papier ist alterungsbeständig nach DIN/ISO 9706. Alle Rechte, insbesondere die der Übersetzung in andere Sprachen, vorbehalten. Kein Teil des Buches darf ohne Genehmigung des Verlages in irgendeiner Form - durch Photokopie, Mikroverfilmung oder irgendein anderes Verfahren - reproduziert oder in eine von Maschinen, insbesondere von Datenverarbeitungsmaschinen, verwendbare Sprache übertragen oder übersetzt werden. All rights reserved (including those of translation into another languages). No part of this book may be reproduced in any form - by photoprinting, microfilm, or any other means - nor transmitted or translated into a machine language without written permission from the publishers. Druck und Bindung: GAM Media GmbH, Berlin Printed in the Federal Republic of Germany

Contents

Ingrid Kaufmann and Barbara Stiebeis Introduction

9

Janet Grijzenhout Stress at the phonology-morphology interface

15

Richard Wiese A model of conversion in German

47

Veronika Ehrich On the verbal nature of certain nominal entities

69

Harald Clahsen, Peter Priifert, Sonja Eisenbeiss, and Joana Cholin Strong stems in the German mental lexicon: Evidence from child language acquisition and adult processing

91

Gereon Müller Remarks on nominal inflection in German

113

Albert Ortmann Economy-based splits, constraints, and lexical representations

147

Paul Kiparsky Disjoint reference and the typology of pronouns

179

Gisbert Fanselow Quirky "subjects" and other specifiers

227

Wolfgang Klein Why case marking?

251

Ilse Zimmermann Structural cases in Russian

275

Gerhard Jäger Gradient constraints in finite state OT: The unidirectional and the bidirectional case

299

Manfred Bierwisch A case for CAUSE

327

8

Contents

Heinz Vater Sein + participle constructions in German

355

Sebastian Löbner Is the German Perfekt a Perfect Perfect

369

Arnim von Stechow German seit 'since' and the ambiguity of the German Perfect

393

List of Dieter Wunderlich's publications

433

List of dissertations

444

Addresses of contributors

447

Introduction

Ingrid Kaufmann and Barbara Stiebeis

This volume is dedicated to Dieter Wunderlich on the occasion of his 65th birthday (June 14, 2002). The volume is entitled 'More than Words' for a number of reasons: First of all, Dieter Wunderlich has always strongly defended the view that the lexicon, which has been his main research topic for quite a few years, is more than simply a list of words. His work can be characterized as a constant plea for the lexicon as an autonomous module of grammar. Secondly, Dieter Wunderlich's research extends to various aspects of grammar beyond the domain of the lexicon (see below). Thirdly, he has always been open to approaches which are different from his own. He has had a constant exchange with proponents of other approaches, some of whom have contributed papers to this volume. Finally, Dieter has been a committed, encouraging, influential and devoted linguist, colleague and advisor, and the gratitude that many of those who have had a chance to work with him feel can only be partially reflected in the papers of a Festschrift. Dieter Wunderlich did not intend to become a linguist in the first place. After obtaining his Abitur in Seebad Heringsdorf (Usedom), he studied Physics in Jena, Leipzig and Hamburg. In 1964 he obtained a diploma in Nuclear Physics at the University of Hamburg. He worked as physicist for one year, but he soon became interested in formal analyses of style and went to Berlin to study German literature and language at the Technical University (TU). His interest in Linguistics developed from his contact with the Linguistics group of the Academy of Sciences in East Berlin, which at the time was one of the places where modern Generative Grammar was practised for the first time in Germany. In 1969 Dieter Wunderlich finished his doctoral dissertation on tense and time reference in German at the TU Berlin. From 1970-73 he was professor for German Linguistics at the Free University of Berlin, and since 1973 he has occupied the chair in General Linguistics at the Heinrich-Heine University in Düsseldorf, where he built up the Institute of Linguistics. Dieter Wunderlich is one of the few 'Generalists' among the German linguists. His research areas include temporal semantics, speech act theory, intonation, language and space, questions and answers, coordination, modals, comparatives, complex verbs, participles, syntax and semantics of PPs. In the last 12 years he has focused on various

10

Ingrid Kaufmann and Barbara Stiebeis

aspects of the lexicon (e.g., role of paradigms, stem allomorphy, lexical representation of verbs, lexical and functional categories) and its relation to syntax (e.g. agreement, argument linking, and adjunction). He has always advocated a lexical analysis of agreement, word formation, semantic decomposition of verbs and of complex predicates. Dieter Wunderlich's research is documented in a number of books written or edited by him and in numerous articles. A list of publications is given in the appendix. Dieter Wunderlich has run several research projects that were funded by the German Science Foundation (DFG). From 1978-80 he directed a project on the function of modalities in discourse and co-directed a project on comparative studies of routine formulas in German and Japanese (with Florian Coulmas). From 1983-86 he codirected a project on quantifiers in German (with Sebastian Löbner). Projects on sentence intonation and focus structure of wh-questions in German (1985-87) and on the semantics and conceptual structure of spatial localization (1988-90) followed later. Within the project cluster of the Sonderforschungsbereich 282 'Theory of the lexicon', whose speaker and stimulating and integrating spirit he has been since 1991, he has run projects on the lexical foundation of agreement (1991-96), on derivation and lexical semantics (1991-93; with Jürgen Lenerz), on verb meanings (1991-93; with Sebastian Löbner), on verb structures and complex predicates (1994-2002; since 1997 with Barbara Stiebels), and on non-concatenative morphology (2000-2002; with Janet Grijzenhout). From these projects, three lexically based theories emerged: a lexically based theory of agreement, which - like Head-driven Phrase Structure Grammar (HPSG) - treats agreement as a phenomenon of the syntax-semantics interface. Wunderlich's theory of agreement differs from HPSG in that it takes the morphology into account more thoroughly. Parallel and independent of Chomsky's Minimalist Program, Dieter Wunderlich developed a theory of inflectional morphology (Minimalist Morphology, MM), which is affix-based, makes extensive use of underspecification, builds on the notion of paradigms, and avoids arbitrary class features and zero morphology. Inspired by the two-level approach to meaning by Bierwisch and his co-workers and Kiparsky's linking theory, Dieter Wunderlich developed the Lexical Decomposition Grammar (LDG), a theory of the representation of semantic form and argument structure of lexical entries. LDG was designed in particular to deal with diatheses and argument linking. Both MM and LDG have been reformulated within Correspondence Theory (CT). With respect to inflection, CT-based MM provides an elegant account of gaps and substitutions in paradigms, as Dieter Wunderlich has shown for Yimas and Dalabon, whereas CT-based LDG accounts for the typological range of linking patterns in the languages of the world. Dieter Wunderlich's involvement in research projects documents his own interest in current problems of Linguistics, as well as his concern for the academic development of his students. He has always provided his students with a stimulating and encouraging scientific environment and has been willing to share his time and knowledge with them. The various M.A. and Ph.D. theses which he has supervised reflect his commit-

Introduction

11

ment to the advancement of young academics.1 His comments on papers by students and colleagues, often written in the very same evening after Dieter was given the paper, generally covered several pages and contained ideas and suggestions for alternative analyses Whereas in the Seventies and Eighties, Dieter Wunderlich focussed on various aspects of German, his later research is characterized by an increasing interest in typologically and genetically diverse languages. Starting with the analysis of verbal and nominal paradigms in Russian, Latin, and Macedonian (among others), he has carried out case studies on Basque, Japanese, Icelandic, Quechua, Georgian, Potawatomi, Hindi, Yimas, Dalabon, Hungarian, Finnish, Bulgarian, and Yucatec Maya - partly in collaboration with his Ph.D. students and his colleagues. Recently, Dieter Wunderlich has developed an interest in the evolution of language, and especially in the role that rich morphology has in the historical development of languages. He assumes that rich morphology is characteristic for the early phase of a morphological system, when syntax and morphology start to develop into distinct, separate modules of grammar. He is also interested in the typology of argument linking. For him, inverse systems and prominence/salience-based systems represent an early system of argument linking, whereas case and agreement linking are recent and more elaborate devices. Besides his research within Theoretical Linguistics, Dieter Wunderlich has shown a strong concern for the application of modern Linguistics in the teaching of German Grammar in secondary schools. He co-edited and contributed chapters of school books for German. In addition, he initiated and co-edited the journal Studium Linguistik (1976-88), which aimed at providing students of linguistics with information about the state of the art in the various linguistic subdisciplines, and which, in some sense, was a German forerunner of Glot. Dieter Wunderlich was one of the founding members of the German Linguistics Society (Deutsche Gesellschaft für Sprachwissenschaft; DGfS) and its president from 1978-80. From 1977-92 he belonged to the Advisory Board of the Max Planck Institute for Psycholinguistics in Nijmegen. From 1991-92 he was a fellow of the Centre for Advanced Studies (Wissenschaftskolleg) in Berlin, where he formed a research group with Manfred Bierwisch and Paul Kiparsky, both of whom have greatly influenced his work. The papers in this volume relate to various aspects of Dieter Wunderlich's work, either indirectly by reflecting on the role of lexical specifications, or directly by commenting on analyses proposed by Dieter Wunderlich. Janet Grijzenhout considers regular and irregular stress patterns in different languages and argues in favour of lexical specification of prosodie structure in stems and affixes that display irregular behaviour with respect to stress assignment. Her approach is cast in an Optimality Theoretical framework. She shows that the assumption of

1

The list of doctoral dissertations supervised or reviewed by Dieter Wunderlich is given in the appendix.

12

Ingrid Kaufmann and Barbara Stiebeis

Base-Output correspondences does not adequately account for deviating stress patterns. Instead, she argues that stress assignment in morphologically complex words is best captured by means of lexical specifications and concomitant alignment constraints. Richard Wiese is concerned with the question of how the categorial restriction on conversion in German can be explained. Building on a proposal made in Kiparsky (1982), he argues that three types of conversion exist: root conversion, stem conversion, and word conversion. At each level, conversion is restricted to a certain type: root conversion and word conversion derive nouns, while stem conversion derives verbs. Veronika Ehrich discusses the verbal and nominal properties of two types of event nominalizations in German, namely Mng-nominalizations and nominalized infinitives. Her starting point is Wunderlich's (1996c) classification of lexical categories by the features [±art(iculated)] and [±(dep)endent]. Ehrich argues that different degrees of "nouniness" can be connected to the degree of articulatedness of the nominalizations. The grade of articulatedness depends on how many of the grammatical properties that define the feature [±art] are instantiated by the nominalization. Harald Clahsen et. al. present psycholinguistic evidence for two hypotheses on the morphological representation of strong verb stems, namely that verb forms consisting of strong stems and inflectional affixes have decomposed representations, and that strong stems are represented as subnodes of hierarchically structured entries with underspecified feature content, as proposed in Wunderlich (1996d). Gereon Müller presents an OT analysis of nominal inflection in German, in which the form and distribution of inflection markers is syntactically determined. He argues for a distinction of case and agreement markers in order to cope with strong and weak inflection respectively. The distribution of markers is determined by constraints that require overt morphological case/agreement markers in certain syntactic domains. The form of the markers is determined by the interaction of the sonority hierarchy with a set of constraints which state that certain sets of morphosyntactic feature combinations are incompatible with certain phonological features. Albert Ortmann discusses morphosyntactic asymmetries ('splits') in object linking, subject number agreement, and noun-phrase internal agreement. Drawing on the general outline of Aissen (2000), he assumes a number of micro-constraints that are universally ordered by harmonic alignment. He provides an account of a conspiracy of two splits in Hungarian possessor agreement in terms of the interaction of general faithfulness constraints and markedness constraints derived from the alignment of the definiteness scale with a plurality scale. In his analysis, he pursues the idea that crosslinguistic variation in to the specification of inflectional features is partly determined by constraint ranking. Paul Kiparsky shows that the distribution of pronominals and reflexives is determined by a universal Obviation constraint, which requires coarguments to have disjoint reference, and the economy constraint of Blocking. These constraints interact with a hierarchy of binding domain constraints, thus generating a typology of pronouns. The constraint system predicts the category of obviative reflexives and the novel category of referentially dependent nonreflexive pronouns. It also explains the relationship be-

Introduction

13

tween the blocking of coargument reflexives, in the non-coargument case, and the ambiguity of reflexives between the bound anaphora and coreferential readings. Gisbert Fanselow discusses the notion of "quirky subjects" and argues that there is no compelling evidence for such a category in Icelandic, as well as in German. By reconsidering the criteria proposed for quirky subjects (coordination reduction, raising, control, agreement, and islandhood), he shows that the relevant data can simply be explained be reference to the behaviour of case and the role of the argument hierarchy in these languages. Wolfgang Klein pursues the question of what the function of inflectional morphology, in particular case marking, is. Departing from the observation that languages tend to reduce morphology, and that impoverished morphology is characteristic for the 'Basic Variety' of second language learners, he raises the question of why languages display inflectional morphology at all. With respect to case, Klein's answer is that morphological case-marking serves to identify the argument-time structure of the verb: Default rules determine the morphological case of an argument, depending on its occurrence(s) in the cluster of argument-time pairs that make up the lexical representation of verbs. Ilse Zimmermann argues that morphological case has to be distinguished from abstract case, and presents an OT analysis of structural case in Russian in terms of Jakobson's case features (to account for morphological case in the output), on the one hand, and the linking theory of LDG (to account for abstract case in the input) on the other. The mapping of these two classes of features is regulated by specific correspondence constraints. Manfred Bierwisch's paper deals with the question of whether semantic representations can do without the predicate CAUSE, as assumed in LDG. By discussing causative verbs, resultative constructions, and lexical items like cause, because, as, he comes to the conclusion that CAUSE is needed in addition to non-causal conjunction. Gerhard Jäger discusses issues of computational complexity of Optimality Theory (OT) in its standard and its bidirectional variety. Jäger integrates results on finite state implementation of OT systems by Frank & Satta (1998) and Gerdemann & van Noord (2000). He generalizes the latter construction and shows that it is always applicable if (a) the OT system in question only uses markedness constraints, and (b) the notions of global and local optimality defined by this system coincide. Moreover, he shows that a similar extrapolation of Gerdemann & van Noord's construction to bidirectionality is not possible. Since, in contrast to unidirectional optimization, bidirectional optimization that uses gradient (markedness) constraints goes beyond the bounds of finite state techniques, it is intrinsically more complex than its unidirectional counterpart. Heinz Voter's paper discusses the various readings of the sein 'be' + participle constructions in German, building on the classification proposed in Wunderlich (1997). Vater comes to the conclusion that five types of se/n+participle constructions have to be distinguished, two of which are non-perfective while the others have a perfective reading. The constructions differ with respect to the status of sein (auxiliary or copula) and the category of the participle (verb or adjective).

14

Ingrid Kaufmann and Barbara Stiebeis

Two papers in this volume are concerned with the semantics of the German perfect. Sebastian Löbner presents evidence for the ambiguity of the German perfect between an aspectual reading ("non-past perfect") and a tense reading ("past non-perfect"), by discussing the interpretation of the perfect in narrative contexts and in temporal clause constructions, as well as the combination of perfect forms with schon 'already* and noch 'yet'. He thus argues against approaches that assume a uniform semantic analysis of the German perfect. Arnim von Stechow examines the interaction of the German durative preposition seit 'since' with tense, in particular with the perfect. Like Löbner, von Stechow assumes that the German perfect is ambiguous. He provides evidence for a distinction of four perfect readings and proposes an analysis in which the interpretation of the perfect form is determined by the interplay of several combinations of semantic tense and aspect. Von Stechow furthermore shows that seit a is ambiguous too: one variant modifies an 'extended now' introduced by the perfect, while the other variant introduces an 'extended now'. We would like to thank Jennifer Austin, Miriam Butt, Ray Fabri, Thomas Gamerschlag, Janet Grijzenhout, Wolfgang Kehrein, Ekkehard König, Sebastian Löbner, Ralf Naumann, Albert Ortmann, Martina Penke, Wiebke Petersen, Chris Piñón, Carsten Steins, and Arnim von Stechow for reviewing papers of this volume. We would ALSO like to thank Manfred Bierwisch and the Akademie Verlag for including this volume in the series Studia Grammatica. Finally, we would like to thank Dieter Wunderlich himself for discreetly pointing out where the list of publications and supervised dissertations can be found on his computer. It has been both funny and strenuous to listen to his comments on remarkable and awful Festschriften. Hopefully, he will consider this Festschrift to be one of the former kind.

Düsseldorf, March 2002

Stress at the Phonology-Morphology Interface Janet

1.

Grijzenhout

Introduction*

Main stress alternations have become an important issue in the last few years; they are interesting for the study of the phonology-morphology interface especially in the light of recent developments within Optimality Theory (OT; Prince & Smolensky 1993). OT is a non-derivational constraint-based framework and current debates center around the question how to account for so-called 'transderivational phenomena' that used to be explained in generative linguistics by means of rule-ordering and level-ordering (see e.g. Kiparsky 1982, 1985). This paper contributes to the debate by investigating phonological and morphological properties of main stress assignment. Patterns of stress assignment depend on both grammatical and phonological factors. A grammatical factor determining the position of main stress is the morphological boundary (usually the word boundary). In some languages, this is the only relevant factor, so that the placement of stress is completely predictable (e.g. main stress is assigned to the first syllable of a word in Icelandic and to the penultimate syllable in Polish). The placement of stress may also depend on grammatical categories. For example, English main stress is on the first syllable in a bisyllabic noun (a récord, an insult) and on the second syllable in some bisyllabic verbs (to record, to insult). A phonological factor that often plays a role in stress assignment is syllable weight (e.g. stress is normally assigned to the penultimate syllable in Latin, except when that syllable is light; see Allen 1973). For languages where the grammatical and phonological factors mentioned above are the only aspects that determine stress placement, elegant accounts have been proposed

* The work presented here was supported by the German Science Foundation (DFG) through the SFB-282 project 'Lexical Phenomena in Correspondence Theory' at the Heinrich-Heine-Universität Düsseldorf. I would like to thank Diana Apoussidou, Dafna Graf, Franc Grijzenhout, René Kager, Martin Krämer and Alexandra Popescu for many discussions on the topic of main stress alternations in different languages. Dafna Graf, Wolfgang Kehrein, Andy Wedel and the editors of this volume read an earlier version of this paper and I thank them for valuable comments. Last but not least, I would like to thank Dieter Wunderlich. It has been a great pleasure to work, wonder, and - on occasion - wander with him in linguistic and other fields.

16

Janet Grijzenhout

in the literature (most notably by Hayes 1995; the account he provides can easily be transferred into the framework of OT, see Prince & Smolensky 1993: 38-66). More problematic for non-derivational frameworks are (i) cases where stress determines the selection of affixes and (ii) cases where different morphological categories have different demands on stress placement. The role of stress in the selection of affixes can be illustrated by the following examples from German and English. The German prefixes be- and ge- do not associate to an unaccented syllable: studier - studiert!*gestudiert 'study - studied (participle)' vs. régn - gerégnet 'rain - rained (participle)' (Kiparsky 1966: 70-75) and the English nominalizing suffix -al attaches to verb stems that have an unstressed-stressed syllable sequence: arrive - arrival vs. visit - *visital (Raffelsiefen 1998). These particular affixes in German and English seem to require a specific prosodie host and this can be accounted for in OT by means of so-called 'alignment constraints'.1 The demands of morphological categories on stress placement have been the focus of attention in recent OT-studies, for instance, by McCarthy (1995) for Rotuman, Benua (1997) for English and Kager (2000) for Dutch, as well as in studies carried out in Düsseldorf by Canclini (1999) for Italian, Graf (2000) for Modern Hebrew, Popescu (2000) for Romanian and Apoussidou (2001, 2002) for Modern Greek. Following Prince & Smolensky (1993) among others, these authors all assume that regular stress assignment (or 'default stress assignment') is most adequately accounted for by the interaction of wellformedness and faithfulness constraints; i.e. regular stress placement is attributed to the grammar rather than to the lexicon. Patterns of stress assignment that deviate from the regular pattern (in that they depend on demands of stems and/or affixes) receive different treatments in recent OT-analyses. McCarthy, Benua and Kager account for them by means of correspondence constraints that refer to morphologically related forms, whereas Inkelas (1994) and other authors suggest that they are best analyzed by means of exceptional markings in the lexicon together with concurrent constraints that require faithfulness to prosodie structure in the input. In this paper I will first show that both approaches have shortcomings, i.e. the correspondence constraints alone cannot account for stress in morphologically complex words and neither does it suffice to have underlying metrical structure as the only mechanism to account for irregular stress placement. The most important finding of this paper will be that we do not need correspondence constraints that refer to morphologically related forms to account for stress placement and I will argue that the only devices that a constraintbased theory needs to account for irregular stress patterns are prosodie specifications in the lexicon and morpheme-specific constraints. The issue what lexical markings for stress look like is also relevant in this respect. The paper is structured as follows. Section 2 discusses different proposals from the literature to account for main stress assignment in morphologically complex words within the framework of Optimality Theory. I will discuss the proposals put forward by McCarthy (1995), Benua (1997) and Kager (2000), respectively, which are all characterized by the use of constraint interaction for stress placement to the exclusion of lexi1. Consider as an example the alignment constraint ALIGN (ge- Right, à Left) for German which says that the right edge of the suffix ge- should coincide with the left edge of a stressed syllable.

Stress at the Phonology-Morphology

Interface

17

cal stress markings; the interaction between phonology and morphology is captured by correspondences between segments in simplex forms and segments in morphologically complex forms. These accounts have some problems and section 3 will consider Inkelas's (1994) alternative proposal to specify main stress in stems and/or suffixes in the lexicon. Her account works for Turkish data, but fails to capture important generalizations in other languages. In section 4 it will be argued on the basis of a case study of stress placement in Dutch that we need both prosodie specifications in the lexicon (i.e. underlying metrical structure) as well as morpheme-specific constraints.

2.

Stress in the grammar

In early OT literature, stress placement was explained by means of constraints and constraint rankings and no underlying metrical structure was assumed. Sections 2.1 and 2.2 discuss two accounts for stress alternations in morphologically complex words which use special constraints for stress alternations as well as the assumption of outputto-output relations (McCarthy 1995, Benua 1997). Section 2.3 illustrates Kager's (2000) OT-account which is based on the suggestion that the location of main stress in related words is the result of correspondences between a stress peak in a simplex base and a stress peak in a related morphologically more complex form.

2.1

McCarthy

(1995):

Output-Output Correspondence and

HEAD-MATCH

The two patterns of morphologically determined stress that we find in the world's languages are the following: (i) main stress is either placed on the same syllable in morphologically simplex words and in more complex forms (cf. párent - parenthood), or (ii) stress placement depends on the demands of roots, stems and affixes (cf. párent paréntal). With respect to the first possibility, the fact that morphologically related output forms tend to resemble each other was recognized early within Optimality Theory (e.g. Burzio 1994, Kenstowicz 1994, 1996) and gave rise to the notion of 'correspondence' between morphologically related forms; i.e. the assumption is that a certain output form stands in a correspondence relation with other output forms to which it is morphologically related. Within this framework, McCarthy (1995) maintains that main stress in one output form shows up at the same location in related morphologically more complex output forms due to a highly ranked faithfulness constraint which says that a segment which is a prosodie head (i.e. a segment which is marked for main stress) in one output form (a) must surface as a prosodie head in the optimal candidate of a morphologically related form (β): (1)

HEAD-MATCH ( M c C a r t h y 1995: 23)

If α is the prosodie head of the word and α 9Î β, then β is the prosodie head of the word.

18

Janet Grijzenhout

According to McCarthy (1995), correspondence is a relation (91) from segment to segment and there is no prosodie structure in the input; thus, HEAD-MATCH can only have an effect in cases where an output X stands in a correspondence relation with another output form Y and it is satisfied only if the segment which is marked for main stress in output X corresponds to a segment which is the prosodie head of output Y. McCarthy (1995) illustrates the effect of this constraint by an example from the central Oceanic language Rotuman (spoken on an island about 300 miles north of Fiji, see McCarthy 1995: 2 and references cited there). In this language, words have two forms depending on syntactico-semantic properties: one form for the so-called 'complete phase' and one for the so-called 'incomplete phase'. These forms have different phonological manifestations, but stress is usually assigned to the same vowel in both forms (cf. the complete phases tokiri 'roll' and ráko 'imitate' versus the corresponding incomplete phases tokir and rák). According to McCarthy (1995), stress preservation on the vowel is the result of the fact that HEAD-MATCH is ranked relatively highly (e.g. higher than the constraint which says that a segment in the input should have a correspondent in the output: MAX). McCarthy (1995) stipulates that the incomplete phase in Rotuman must end in a monosyllabic foot (INC-PHASE).2 For the complete phase form [rdko] 'imitate' with stress on the first vowel, this means that the candidate for the incomplete phase with main stress on the same vowel as in the complete phase and without the second vowel is the optimal output. In (2) and subsequent examples, square brackets mark prosodie word edges, round brackets indicate foot boundaries, and a dot indicates a syllable boundary; in accordance with the usual OT-practice, constraint violations are marked by an asterisk and fatal violations have an exclamation mark. Candidates (2a) and (2b) below end in a monosyllabic foot and thus satisfy the constraint INC-PHASE; candidate (2c) consists of a bisyllabic foot and this constitutes a fatal violation of this constraint. In (2b) main stress is not assigned to the same segment as in the complete phase ráko, which constitutes a violation of HEAD-MATCH. Even though (2a) has one segment less than the corresponding form ráko (and thus violates MAX), it is the winning candidate, because MAX is a lowly ranked constraint: (2)

OT-account of Rotuman main stress (McCarthy 1995) Complete phase: ra.ko INC-PHRASE

a. b. c.

cy

[(rdk)] [ra(kó:)] [(ra.ko)]

HEAD-MATCH

MAX *

*! *!

Liberman & Prince (1977) pointed out that stress is not a property of a segment, but rather of an entire syllable. In McCarthy's analysis, stress as a property of the syllable is accounted for in an indirect way: the vocoid in the nucleus position of a syllable is 2. McCarthy (1995) claims that the Rotuman main-stress foot is a moraic trochee, consisting of a heavy syllable or two lights, aligned at the right edge of the prosodie word. Since there is no report of secondary stress, McCarthy assumes that all syllables except those in the main-stress foot are unfooted and parsed directly by the prosodie word (PrWd).

Stress at the Phonology-Morphology

Interface

19

the 'head' of that syllable and if it is marked for stress, the entire syllable bears stress and is consequently the head of a foot which is the head of a prosodie word. In other words, HEAD-MATCH requires that if a vocoid α is marked for stress in some output, there should be a vocoid β which is stressed in a corresponding output, but other members of the syllable need not be identical (or present even). The implication is that for McCarthy, HEAD-MATCH is an IDENTITY relation rather than another kind of faithfulness relation like MAX (the requirement that an element in a certain string has an overt correspondent); the vocoid in one output should be identical with the vocoid in another corresponding output in one respect only: it must be a prosodie head. McCarthy's analysis has a number of problems. First, a crucial assumption is that stress is not present in a simplex input form, because the input does not have prosodie structure (prosodie structure is assigned by the grammar, i.e. the constraint ranking) and there are compelling arguments showing that this assumption cannot be maintained (see e.g. Jacobs 1994, Kiparsky 1998 and sections 3 and 4 below). Second, another important assumption is that main stress can be a property of a segment in an output form that functions as the input of a semantically related form, but it is not clear which words can have an output-to-output relationship and whether we actually need such 'output-to-output' relations in the grammar. Third, the assumption that stress is marked on a vocoid is problematic, because in this way stress is only an indirect property of a syllable. The latter problem was also recognized by Benua (1997) and we will consider her analysis of stress placement in morphologically complex forms next.

2.2

Benua (1997): Output-Output Correspondence and recursive constraint evaluation

Instead of using a constraint that refers to a segment which is the head of a prosodie constituent, Benua (1997) employs constraints that refer to the position of segments within prosodie constituents. Before discussing the details of her proposal, we will first look at her explanation of English main stress in morphologically simplex words. Her account primarily involves the two constraints below; the first constraint says that the syllable with main stress should be the last syllable in the word and the second constraint says that the final syllable cannot be part of a foot (this implies that it cannot be stressed, because only syllables that are the head of a foot can bear stress): (3)

ALIGN-HEAD-RIGHT (McCarthy ALIGN (Ó Right, PrWd Right)

& Prince 1993; Benua 1997: 172)

The head of the prosodie word is aligned at the right edge of the word (the main stressed syllable is at the right). (4)

NONFINALITYSYLLABLE (NONFLN-σ) ( B e n u a 1997: 1 7 2 )

Word-final syllables are not parsed into a foot. According to Benua, the English foot is a moraic trochee (i.e.it consists of a heavy syllable or two light ones) and constraints which demand moraic trochees are undomi-

20

Janet Grijzenhout

nated in English.3 For ease of exposition, I do not consider candidates with feet other than moraic trochees. Ranking NONRN-σ higher than ALIGN-HEAD-RIGHT results in optimal candidates with main stress on a penultimate heavy syllable (5) or - if the penultimate syllable is light - on the antepenultimate syllable (6):4 (5)

Evaluation of words with a heavy penultimate syllable in English NONFlN-σ

a. b. c.

«s*

[wis.con.(sin)] [wis.(cón).sin] [(wis).con.sin]

ALIGN-HEAD-RIGHT

*! *

**!

Evaluation of words with a light penultimate syllable in English NONFÌN-σ

a. b.

«s"

[(ò.ri).(gin)] [(ó.ri).gin]

ALIGN-HEAD-RIGHT

*! **

In English, main stress in morphologically complex words is either placed on a different syllable as in the related simplex form (cf. original vs. origin) or on the same syllable (cf. wónderfulness vs. wónder). Complex words with affixes like -al, -ate, -ic, -ity, -ous, in- etc. (class 1 affixes; see Siegel 1974) may deviate more from the simplex form with respect to stress assignment than complex forms with affixes like -able, -er, -ful, -ist, -ness, un-, etc. (class 2 affixes). In other words, forms with class 1 affixes are 'less faithful' to simplex forms than forms with class 2 affixes; words with class 2 affixes are fully faithful to stress placement in the stem. Following McCarthy (1997), Benua (1997) claims that simplex forms and complex forms have a so-called 'output-to-output correspondence relationship'. She attributes the phenomenon of non-alternating stress to a positional output-output faithfulness constraint for class 2 affixes which says that if a segment α is the leftmost segment in the foot of output X (the simplex form) and if there is a correspondence relation between segment α and segment β, then β is the leftmost segment in the foot of output Y (the complex form): 5 (7)

OUTPUTOUTPUTdass 2 Affixes ANCHOR-LEFT (Foot): 0 0 2 - A N C H O R If segment α is initial in a foot and α 91 β, then segment β is initial in a foot.

3. This presentation is necessarily a simplification in that it ignores exceptions; for a more extensive OT-account of English stress alternations see Burzio (1994) or Hammond (1999). 4. Violations of ALIGN-HEAD-RIGHT are counted by the number of syllables from the right word edge. Potential candidates in (6) like [o.(ri).gin] or [o.(ri.gin)] with a foot containing one light syllable and a light syllable plus a heavy syllable, respectively, violate the highly ranked constraints that demand moraic trochees. 5. In Correspondence Theory (e.g. McCarthy 1997), faithfulness to prosodie structure is enforced by anchor constraints which demand faithfulness to the edgemost position of correspondent segments within a morphological or prosodie category (see (7) which demands faithfulness to the leftmost position within the prosodie category 'foot').

Stress at the Phonology-Morphology

21

Interface

Consider as an example the output pair wonder and wonderfulness. According to Benua (1997), paradigms are evaluated asymmetrically against a recursive constraint hierarchy such that each recursion evaluates one member of the paradigm and violation in recursion (A) outranks violation in recursion (B). Highly ranked NONFlN-σ ensures that main stress is on the penultimate syllable in the word wonder and, consequently, that the segment which is the onset of the stressed syllable is foot-initial (see candidates (8b,c)). In recursion B, candidate (8a') cannot be a winner because (8a) is a loser in recursion A. In recursion B, OO2-ANCHOR is violated in (8b'), because the segment /w/ is foot-initial in (8b), but not in the corresponding output form with a class 2 suffix. The candidate pair (8c-c') with the forms (wón)der and (wón)derful does not violate this constraint (because /w/ is foot initial in both forms) and is thus selected as the optimal pair: (8)

Recursive evaluation for English stress in simplex forms and forms with class 2 suffixes according to Benua (1997) Recursion (A) Input: wonder NONFiN-σ 002ALIGN-HEAD» ANCHOR

a. b. c.

"S"

won.(dér) (wón).der (wón).der

RIGHT

*! * *

Recursion (B) » wonder + ful + ness

NONFlN-σ

002ANCHOR

a'. won.(dér).ful.ness b'. won.der.(fúl).ness c'. «3* (wón).der.ful.ness

ALIGN-HEADRIGHT **

*!

* ***

In the tableau below, candidate set (9a-a') violates the highly ranked constraint NON FlN-σ. In contrast to the constraint 002-ANCHOR, the constraint OOPANCHOR (for class 1 suffixes) is ranked relatively lowly and, for this reason, the pair origin and original (9c-c') is less optimal than the pair origin and original (9b-b') that has fewer violations of ALIGN-HEAD-RIGHT:

(9)

Recursive evaluation for English stress in simplex forms and forms with class 1 suffixes according to Benua (1997) Recursion (A) NON FIN-

σ a. o.ri.(gin) b. «3" (ó.ri).gin c. (ó.ri).gin

ALIGN-HEADOO2ANCHOR RIGHT

*! ** **

OOiANCHOR

»

22

Janet Grijzenhout Recursion (B)

» a'. b'. «a· c'.

NON FIN-

00 2 -

σ

ANCHOR

o.ri.(g0.nal o.(rí.gi).nal (ó.ri).gi.nal

ALIGN-HEADRIGHT

OOiANCHOR

* **

*

***!

This account is an improvement to McCarthy's (1995) account of stress placement in that it recognizes that stress is not a property of a single segment, but rather a relative property that is best captured by means of prosodie constituents (like the syllable and the foot). The analysis involves (i) recursive constraint evaluation, (ii) output-to-output positional faithfulness constraints for different classes of suffixes and (iii) the fact that information with respect to class membership is available for constraint evaluation (e.g. the constraint OO2-ANCHOR 'knows' that it is vacuously satisfied for every word that does not involve a class 2 suffix and a suffix 'knows' to which class it belongs; i.e. some indication as to class-membership has to be available in the input). A theory that can do without this relatively rich technical apparatus would be superior. The next subsection considers an alternative popular proposal to capture stress placement.

2.3

Kager (2000): Base-Output Correspondence

Another option to account for stress alternations that is widely discussed in the literature is the following. If we can recognize a simpler form in a morphologically complex form, the complex form may have the same stress placement as the simplex form due to the wish to resemble the simplex form in a certain aspect. Kager's (2000) analysis of Dutch word stress crucially relies on the assumption of a correspondence relationship between main stressed segments in morphologically related words (in so far as main stress in complex forms corresponds to the main stress of the simplex form). Before we consider Kager's proposal, it is useful to consider Dutch main stress placement in simplex words first. Main stress in Dutch morphologically simplex words is located on one of the three final syllables (see e.g. van der Hülst 1984, 1985 and Lahiri & Koreman 1988). In the majority of cases, main stress in nouns and verbs is on the antepenultimate syllable if the penultimate syllable is open and the ultimate one is closed (see (10a)). If the final syllable is superheavy, i.e. if it ends in a long vowel plus a consonant (-VVC) or a short vowel plus two consonants (-VCC), main stress is usually placed on the ultimate syllable (see 10b,c). 6

6. In Dutch orthography, the long vowels /o:/ and /a:/ are spelled as '00' and 'aa\ respectively in closed syllables. In this paper I put a stress mark on both letters (which represent one sound).

Stress at the Phonology-Morphology

Interface

23

(10) Antepenultimate and final stress in Dutch monomorphemic nouns a. álmanak 'almanac' b. kameráád 'comrade' c. experimént 'experiment' In cases where these conditions are not met, simplex words have main stress on the penultimate syllable: (11) Penultimate stress in Dutch monomorphemic nouns a. ceremónie 'ceremony' b. agènda 'diary; agenda' c. detèctor 'detector' d. hértog 'duke' The regular stress pattern in nouns and verbs is summarized in words as follows: (12) Dutch Word Stress a. Main stress is assigned to the final syllable in a PrWd if the final syllable is heavy (-VVC, -VCC); b. main stress is assigned to the antepenultimate syllable in a PrWd if the final syllable is closed and the penultimate syllable is open or contains schwa; c. elsewhere, main stress is assigned to the penultimate syllable in a PrWd. Constraints that account for regular stress placement in Dutch simplex words and their respective ranking are as follows: the constraint that requires main stress on a superheavy syllable (referred to as WEIGHT-TO-STRESS-PRINCIPLE 'WSP' in OT-literature) outranks the constraints FOOTBINARITY (which says that each foot consists of two syllables) and TROCHAIC (which says that the leftmost syllable in the foot is strong).7 These constraints outrank NONFINHEAD (the constraint that bans main stress on a final syllable) which, in its turn, outranks ALIGN-HEAD-RIGHT. (13) Dutch constraint ranking for regular main stress placement in morphologically simplex nouns and verbs W S P » FOOTBINARITY, TROCHAIC » NONFINHEAD » ALIGN-HEAD-RIGHT

Basing themselves on different criteria, different authors have made interesting proposals to classify Dutch suffixes (e.g. Booij 1995, Trommelen & Zonneveld 1989). 7. A Dutch foot usually contains one superheavy syllable, a heavy syllable, or two light syllables. Such feet are preferred to a foot which contains a heavy syllable followed by a light syllable which in turn is preferred to a foot consisting of two heavy syllables. In nouns and verbs, a final foot consisting of one heavy syllable does not bear main stress in the majority of cases (i.e. *(al.ma)(nák) and *(her)(tóg) are ruled out by NONFINHEAD), but there are exceptions (see section 4). Feet containing a single light syllable or a light syllable followed by a heavy one are ruled out by FOOTBINARITY and TROCHAIC, respectively. An OT-account of Dutch regular stress involves constraints which are not yet introduced in this paper and to save space I will refrain from presenting a detailed OT-account of Dutch regular stress patterns in simplex words (suffice it to say that the conditions of foot formation mentioned here and the constraint ranking proposed in the text make the correct generalisation).

24

Janet Grijzenhout

With respect to stress assignment, four types of suffixes can be distinguished in Dutch: (14) Four types of suffixes in Dutch a. type I: stress changing affixes change the stress pattern of the base and conform to regular patterns of stress assignment (-ief, -iteit, -ieel, etc.) e.g., product 'product' - productief 'productive' b. type II: prestressing suffixes require main stress on the preceding syllable (-(e)lijk, -ig) e.g., hértog 'duke' - hertógelijk 'ducal' nóódlot 'fate' - noodlóttig 'fatal' c. type III: stress neutral suffixes do not change the stress pattern of the base (-baar, -dom, -schap, etc.) e.g., hértog 'duke' - hértogdom 'duchy' kameráád 'comrade' - kameráádschap 'companionschip' d. type IV: stress attracting suffixes attract stress to themselves (-in, -egge, -es, etc.) e.g., hértog 'duke' - hertogin 'duchess' barón 'baron' - baronés 'baroness' Kager (2000) suggests that morphemes that occur as independent lexical words with main stress may function as a base for Base-Output correspondences between segments. Affixes do not occur as independent lexical items and they are therefore not subject to Base-Output correspondences. Kager's treatment of type I suffixes is based on the stipulation these suffixes attach to a root and that roots are not eligible baseforms. According to Kager, this implies that even though the Dutch words presidént 'president' and product 'product' may occur in isolation (with main stress on the final syllable), these words do not function as bases for presidentiéél 'presidential' and productiéf 'productive', respectively, because the suffixes -ieel and -ief select a root and not a stem as their host. Stress placement in words consisting of a stem plus one of the adjectival suffixes -elijk [a.lak] or -ig [ax] is never on the suffix. For suffixes of type Π, Kager (2000: 139) introduces the following morpheme-specific constraint (which he calls 'SUFFIXTO-PEAK'):

-(e)lijk) Left, stress peak Right): ALIGN {-ig, -(e)lijk} The left edge of the affixes {-ig, -(e)lijk} coincides with the right edge of the stress peak.

( 1 5 ) ALIGN({-I'G,

Kager (2000) is mainly concerned with suffixes of type HI. Consider as an example of the adjective beklemtoonbaar 'accentable', which consists of the verbal prefix be-, the nominal stem klemtoon 'accent' and the adjectival type ΙΠ suffix -baar. The stem may 8. Other suffixes that have the same properties and behave in the same way (in that they begin with a full vowel, associate to the root, and attract stress) are e.g. -aal, -age, -ant, -eel, -eer, -ent, -ide -iek, -ist, -uur. For a more extensive list of these so-called 'stress changing' (or in Booij's terminology 'stress bearing') non-native suffixes the reader is referred to Booij (1995: 75-76).

Stress at the Phonology-Morphology

Interface

25

occur in isolation and has main stress (or a 'stress peak') on its first syllable {klémtoon). The stress peak is on the same syllable in beklémtoonbaar and this leads Kager to the assumption that related forms have stress on the same syllabic nucleus as smaller forms that can occur in isolation (the bases).9 To capture this effect, he proposes the following constraints which say that a segment with the main stress peak in the base should have a correspondent with main stress in the output and vice versa: ( 1 6 ) a.

PEAK-MAX (BASE/OUTPUT) ( K a g e r 2 0 0 0 : 1 2 7 )

Let α be a segment in the Base and β be its correspondent in the Output. If α is the stress peak of the Base, then β is the stress peak of the Output. b.

PEAK-DEP (BASE/OUTPUT) (Kager 2000: 127)

Let α be a segment in the Output and β be its correspondent in the Base. If α is the stress peak of the Output, then β is the stress peak of the Base. In adjectives in which we cannot recognize a smaller unit that may occur in isolation, main stress is assigned to the rightmost full syllable (i.e. any syllable that does not contain schwa): oranje [o. Vary a] 'orange', violet [vi.o.'let] 'violet'. Moreover, in adjectives that contain two words that may both occur in isolation, stress is also assigned to the rightmost full syllable: water.dícht 'waterproof, kleur.écht 'colour-fast'. Kager attributes the location of the stress peak in adjectives to the following constraint which says that main stress is rightmost in adjectives: (17) ADJ-PEAK: ALIGN (Adjective Right, stress peak Right) (Kager 2000: 131) Main stress is on the rightmost syllable in adjectives. A conflict arises when a morphologically complex adjective has a base in which the peak is not rightmost: PEAK-MAX (B/O) requires main stress on a segment which also has stress in the base form and ADJ-PEAK requires main stress on the last syllable. To avoid that a word has more than one stress peak, Kager proposes a highly ranked constraint which requires that each word has one syllable with main stress: (18) UNI-PEAK (Kager 2000: 130) Words must have a unique stress peak. In the following tableau for the adjective beklemtoonbaar 'accentable', PEAK-DEP (B/O), PEAK-MAX (B/O) and UNI-PEAK outrank the constraint which demands that adjectives have their stress peak on the final syllable. PEAK-DEP is violated when main stress in the output candidate does not correspond to a stress peak in one of the bases. PEAK-MAX is violated each time when main stress in one of the four bases is not realized in the output:

9. More precisely, in beklemtoonbaar we recognise the verb stem 'beklemtoon' plus the adjectival suffix -baar. The verb stem in its turn is related to the nominal stem klémtoon and Kager (2000: 130) maintains that the latter is a compound consisting of the two members klém and tóón. Thus, according to the logic of Kager's argument, beklemtoonbaar has in fact four bases: klém, tóón, klémtoon and beklémtoon.

26

Janet Grijzenhout (19) Kager's (2000) OT-account for a Dutch stem plus an adjectival suffix Base: [klém], [tóón], [klémtoon], [beklémtoon] PEAK-DEP

a. b. c. d.

beklémtoonbáár beklemtoonbáár beklémtoonbaar beklemtóónbaar

*

UNI-PEAK

PEAK-MAX

*!

*

ADJ-PEAK

* *

*

**Ι*

*

Note that the burden of explanation for stress assignment in words with a suffix like -baar is not on the output per se, but rather on the specification of the bases. The fact that stress placement is exceptional in beklémtoonbaar in the sense that the final superheavy syllable is not the main stressed one (as in regular kannibáál 'cannibal') is not attributed to the suffix; instead, the stress-determining property is transferred to one of the bases (outputs have to be faithful to one of their bases and suffixes like -baar have no effect on stress placement).10 PEAK-DEP ( B / O ) and PEAK-MAX ( B / O ) guarantee that in nouns and verbs which consist of a stem plus an affix, main stress is placed on the syllable that corresponds to the stressed syllable of the stem in isolation. Compounds consist of two or more stems. Kager introduces the constraint LEFTMOST to ensure that in nominal and verbal compounds the candidate with main stress on the syllable that corresponds to the stressed syllable of the leftmost stem wins: (20) LEFTMOST: ALIGN (PrWd Left, stress peak Left) (Kager 2000: 130) The left edge of the word is aligned with the left edge of a stressed syllable (i.e. main stress is on the leftmost syllable in a prosodie word). Consider as an example the verb rangschik 'to arrange, to rank' which has two bases: rang 'to rank' and schik 'to order'. Each base may occur as a stressed independent lexical item. The tableau in (21) is from Kager (2000: 131); candidate (21c) has no violations of PEAK-MAX ( B / O ) because both the peak of ráng and the one of schik has a correspondent. However, this candidate has a fatal violation of highly ranked UNIPEAK. Candidates (21a) and (21b) each have one violation of PEAK-MAX ( B / O ) and the candidate that has main stress on the leftmost syllable is selected as the optimal output:

10. Kager (2000) proposes that some bases may affect the selection of suffixes; the suffix -ig seems to select a base with final stress, so that moerás 'swamp' is an acceptable base for moerásig 'swampy', whereas pías 'clown' with prefinal stress is not a possible base for *piassig. However, some native speakers accept piássig and we also find forms with a stress shift when the base does not have final stress (e.g. misdaad 'crime' - misdádig 'criminal'). Moreover, many nouns with final stress do not accept the suffix -ig (e.g. soldáát 'soldier', *soldáátig). Hence, the fact that pias does not take -ig may be an accidental gap (similar to the fact that mist 'fog' is a base for mistig 'foggy', whereas the noun sneeuw 'snow' never takes the suffix -ig (*sneeuwig)).

Stress at the Phonology-Morphology

Interface

27

(21) OT-account for Dutch compound verbs according to Kager (2000) Base: [rang], [schik] PEAK-DEP : UNI-PEAK PEAK-MAX ADJ-PEAK: LEFTMOST

rang.schik a. b. «S" rang.schik c. rang.schik

;

*

;

*

I

:

*!

*!

Since PEAK-DEP (B/O) penalizes any output with main stress on the syllable of a nonbase, suffixes of type ΙΠ cannot change the stress pattern of the base. For words which consist of a stem plus a single suffix, this is indeed the case: main stress is on the stem syllable in adjectives like beklémtoonbaar and slápeloos 'sleepless' (from slááp 'sleep' and -eloos). However, combinations of one suffix of this type followed by the type ΙΠ suffix -heid give rise to a pattern that is unaccounted for in Kager (2000): main stress is realized on the leftmost suffix in werkelóósheid 'unemployment' and slapelóósheid 'sleeplessness'. These cases pose a serious problem to Kager's analysis, because they violate highly ranked PEAK-DEP (B/O) (a peak is realized on an element that does not have a peak in a base form) as well as PEAK-MAX (B/O) (the peak in the base is not realized as a peak in the form ending in -heid).11 Another drawback of the account sketched in this section is that it does not work for the second example that Kager (2000: 124) introduces in his article and which he often quotes, but never explains. Under the assumption that each word that can occur in isolation is a potential base, the word badhanddoek 'bath towel' has four bases: bád 'bath', hánddoek 'towel', hánd 'hand', and dóék 'cloth' and, hence, four potential peaks to be faithful to.12 Assuming the constraint ranking assumed by Kager (2000), the candidate that incurs the fewest violations of PEAK-MAX (22c) is predicted to be the winner. The correct form, however, is the form with stress on the first syllable (22d). I mark a wrong output selected as optimal with a bomb (é*) and the actual (nonselected) output with a sad smiley (@). (22) Account according to Kager (2000) with wrong result for Dutch compounds Base: [bád], [hánddoek], [hánd], [dóek] a. b. c. é * d. ®

bád.hand.dóek bad.hand.dóek bad.hánd.doek bád.hand.dóek

UNI-PEAK *!

PEAK-MAX

ADJ-PEAK

LEFTMOST

**

***!

*

**

*

***!

11. Kager's account would work for German where main stress is assigned to the first syllable in Arbeitslosigkeit (from Arbeit 'work' and -los + -ig + -keit) and Schlaflosigkeit (from Schláf 'sleep' plus -los + -ig + -keit). 12. The assumption that each word that can occur in isolation is a potential base is made explicit in Kager (2000: 134,141) for the adjectives rangschikbaar and noodlottig which have three bases: [rángschik], [ráng] and [schik] for rangschikbaar and [nóodlot], [nóod] and [lót] for noodlóttig.

28

Janet Grijzenhout

With respect to suffixes of type Π, it is easy to see that the constraint ALIGN {-ig, -(e)lijk) in (15) must be ranked higher than PEAK-MAX (B/O) to ensure that main stress is on a syllable preceding these suffixes (cf. noodlóttig/*nóódlottig 'fatal' with the bases nóódlot 'destiny', nóód 'need' and lót 'fate' plus the suffix -ig). Kager ranks a constraint which says that every morpheme in the input has a correspondent in the output (MOR-MAX) relatively low, so that a hypothetical form *waarheidig (from wáár 'true' plus the suffix -Heid plus -ig) is correctly ruled out, because stress in a hypothetical output *wáárhéidig violates UNI-PEAK, *wáárheidig violates ALIGN {-ig, -(e)lijk} and *waarhéidig violates PEAK-DEP, whereas the null-form only violates lowly ranked MOR-MAX. However, as Kiparsky (1998: 20-24) points out, Kager's analysis is inadequate for three cases where -elijk and -ig actually cause a stress shift. First, combinations of one suffix of type ΙΠ followed by a suffix of type Π give rise to an unexpected pattern: in those cases, main stress is not assigned to one of the bases, but rather to the suffix (cf. vriendscháppelijk 'friendly' from vríéndschap 'friendship' + -elijk). Second, compounds with bound second members exhibit an unexpected stress shift as well (e.g. aandáchtig 'attentively' from áándacht 'attention' + -ig). Third, we find a stress shift in simple words (as hertógelijk 'ducal' from hértog 'duke' + -elijk). Highly ranked PEAK-DEP (B/O) incorrectly rules out forms with main stress on the syllable preceding -ig and -(e)lijk in the three cases above. Consider as an example the adjective wonderbáárlijk 'amazingly' (which has as its base-form the stem wonder and involves the suffixes -baar and -lijk). The next tableau shows how the constraints assumed by Kager give the wrong result. (23) Account according to Kager (2000) with wrong results for Dutch adjectives Input: {wonder, -baar, -lijk} Base: [wonder] PEAK ALIGN {-ig, UNI- MOR- PEAK ADJ- LEFT -DEP PEAK MAX -MAX PEAK MOST -(e)lijk) * * wón.der.báár.lijk *! * * * wón.der.báár.lijk *! * *! wón.der.báár.lijk

a. b. ® c. d. «* 0

***

Kager's (2000) analysis is based on the following stipulations: (i) Base-Output correspondences are crucial for stress placement in morphologically complex words and only stems that may occur in isolation qualify as base forms for more complex structures (i.e. structures with suffixes of type ΠΙ and compounds), (ii) PEAK-MAX is vacuously satisfied in all forms involving roots plus suffixes of type I and (iii) the analysis of stress in words involving the suffixes -ig and -(e)lijk requires a morphemespecific constraint (see (15)). Despite its merits, Kager's (2000) analysis of Dutch word stress suffers from some empirical flaws. It has been pointed out above that he cannot explain words like slapelóósheid with main stress on the leftmost stress neutral suffix, nor words like wonderbáárlijk and aandáchtig with prestressing suffixes and main stress on a syllable that does not have a corresponding stressed base. Moreover, neither can he explain

Stress at the Phonology-Morphology

Interface

29

words like hertogin 'duchess' (from the stem hértog 'duke' plus the feminine suffix -in) with main stress on the final syllable.13 Even though Base-Output correspondence is the most widely used theoretical device to explain the transderivational property of stress, I do not adopt it here because of its empirical flaws. In section 4 I will suggest an OT-solution to the problem of main stress alternations in Dutch which does not rely on the concepts of output-to-output faithfulness or base-forms and which is nevertheless able to account for stress changing suffixes of type I like -ieel and -ief, the prestressing adjectival suffixes -elijk and -ig, stress neutral suffixes like -baar and -schap and stress attracting suffixes like -in. Kager uses morpheme-specific constraints only to save a few exceptions. In section 4 I will argue that if we accept one more morpheme-specific alignment constraint, there is no need at all to assume Base-Output correspondences and concurrent faithfulness constraints to account for main stress in morphologically complex words. Morpheme-specific alignment constraints are a controversial issue in OT. Basing herself on Turkish data, Inkelas (1994) argues against morpheme-specific alignment constraints and claims that irregular stress patterns should be accounted for instead by marking stress (or prosodie structure) on morphemes in the input. We will consider her arguments against morpheme-specific alignment constraint next and in section 4 I will show that her claims cannot be maintained because some grammars actually need both input specifications for stress as well as morpheme-specific alignment constraints.

3.

Stress in the lexicon: input to output correspondence

Inkelas (1994; published in 1999) compares two possible analyses to account for stress patterns in Turkish: a grammatical account and a lexical one. The grammatical account does not assume underlying specifications and relies on morpheme-specific constraints only; the lexical account can do without morpheme-specific constraints but assumes prespecified metrical structure in the lexicon. Inkelas' analysis of Turkish regular stress assignment and stress assignment involving neutral suffixes (i.e. suffixes that conform to regular stress assignment in that main stress is on the final syllable) involves the constraint FtNALSTRESS :1 13. In a footnote, Kager (2000: 129) suggests that these suffixes 'require brute-force accentuation, presumably by input specification plus top-ranking peak faithfulness'. If Kager accepts 'bruteforce accentuation' for some cases, the question is why he doesn't accept input specifications for other cases as well. The theoretical apparatus thus involves (i) Base-Output correspondences, (ii) a morpheme-specific alignment constraint (i.e. (15)) and (iii) input specifications. In section 4 I will return to this point and argue that only two of these three aspects are needed and I will provide an analysis for cases that Kager cannot explain. 14. Note the fact that in the definitions of ALIGN-HEAD-RIGHT (3) and FINALSTRESS (24), the arguments are reversed: ALIGN-HEAD-RIGHT says that every stressed syllable is final in some PrWd, whereas FINALSTRESS demands that every PrWd ends in some stressed syllable. When a prosodie word does not end in a stressed syllable, this counts as one violation of the latter constraint.

Janet Grijzenhout

30

(PrWd Right, σ Right) (Inkelas 1 9 9 4 / 1 9 9 9 : 1 5 0 ) The right edge of the word is aligned with the right edge of a stressed syllable (i.e. each prosodie word ends in a stressed syllable).

( 2 4 ) FÏNALSTRESS: ALIGN

More interesting for our present purpose is her analysis of stress patterns that deviate from the regular pattern. Such cases involve exceptional roots and/or exceptional suffixes. Roots can be exceptional in that they display penultimate stress (e.g. ab.lú.ka 'blockade', penjé.re 'window') or antepenultimate stress (pé.nal.tt 'penalty kick'). There are two classes of suffixes with irregular stress: prestressing suffixes like -mE (NEG) and -ml (INTERROG) that require main stress on an immediately preceding syllable and stressed suffixes like -Iyor (PROG) that require main stress on their initial syllable. Before we discuss the two possible accounts, let us first consider some examples. In cases where a prestressing suffix follows a stressed root, stress is assigned to the root syllable (25a). When a prestressing suffix or an initially stressed suffix follows a neutral root, the stress specification of the suffix wins (25b,c). When two suffixes follow a neutral root, the leftmost suffix determines the stress of the word (25d,e): (25) Turkish word stress (Inkelas 1999: 161) a. b. c. d. e.

pen.jé.re + -ml a.ra.ba + -ml bi.rak + -Iyor bi.rak + -mE + -Iyor gel + -Iyor + -ml

pen.Jé.re.mi a.ra.bá.mi bi.rak.iyor bi.rák.miyor gelíyormu

'window-lNTERROG' ' car-lNTERROG ' 'leave-PROG' 'leave-NEG-PROG' ' COme-PROG-lNTERROG '

Inkelas proposes that in a purely grammatical account, irregular stress placement as in the examples above involves constraints which align one edge of a morpheme with an edge of a prosodie constituent (stressed syllable or foot): (26) Alignment constraints for Turkish roots and suffixes (Inkelas 1999: 164) a. ALIGN (penjere Right, Foot Right): The right edge of the root penjere coincides with the right edge of a foot. b. ALIGN (-ml Left, σ Right): The left edge of the suffix -ml coincides with the right edge of a stressed syllable. c. ALIGN (-Iyor Left, Foot Left): The left edge of the suffix -Iyor coincides with the left edge of a foot, etc. To account for the fact that the leftmost morpheme determines main stress in the word, we are forced to rank alignment constraints for roots higher than those for suffixes. Moreover, alignment constraints for suffixes like -mE that are relatively close to the root should outrank those for suffixes like -Iyor that are closer to the right word edge:15

15. In (27) and (31), I abstract away from the fact that the final syllable of the suffix -mE and the initial syllable of the suffix -Iyor are subject to vowel harmony and 'merge', so that the actual output form is [birákrmyor] 'leave-NEG-PROG'.

Stress at the Phonology-Morphology

Interface

31

(27) Evaluation of Turkish root + prestressing + initially stressed suffix Input: bi.rak + -mE + -Iyor ALIGN (-mE Left, ALIGN (-Iyor Left,

σ Right) a. bi.rak.ma.(iyor) b. «S* bi.(rák.ma).iyor c. (bi.rak).ma.iyor

Foot Left)

RNAL STRESS

*

*

*

*

*

*! *!

Inkelas (1999: 167) mentions as a disadvantage of this approach the fact that it is an accident that the morpheme whose alignment constraint is ranked highest occurs to the left of the morpheme whose alignment constraint is ranked lower. In her lexical account of main stress placement, lexical entries are specified for foot structure. Instead of the morpheme-specific alignment constraints in (26), a lexical account relies on the input specifications in (28). In accordance with the general pattern in Turkish words, all prespecified feet are bisyllabic trochees (the trochaic foot is represented by rounded brackets; the first syllable in the foot is stressed and the second syllable is unstressed): (28) Lexical entries for Turkish roots and suffixes (Inkelas 1999: 169) a. roots specified for trochaic foot: pen.(jére); (pé.nal).ti b. prestressing suffixes specified for trochaic foot: (a.ml); (a.mE) c. stressed suffix specified for trochaic foot: (I.yor) To capture the fact that a stressed foot in the input is preserved in the output, Inkelas invokes a faithfulness constraint localized to the foot: 16 (29) PARSE-FOOT (referred to as 'MAX-FTs' below) (Inkelas 1999: 152) Perserve, in the output, any stress feet that are in the input. In Turkish, MAX-FTs and UNI-PEAK outrank RNALSTRESS (it is more important to

parse an input foot than to build a foot whose head is word-final). When two feet are specified in the input, the leftmost one is parsed. Inkelas proposes the following constraint to capture this fact (called STRESS INITIAL in Inkelas 1999 and rephrased below as ALIGN-HEAD-LEFT in accordance with McCarthy's 1995 and Benua's 1997 proposals discussed in section 2.2): (30) ALIGN-HEAD-LEFT: ALIGN (Ó Left, Domain Left) (Inkelas 1999: 170) The left edge of a stressed syllable is aligned with the left edge of a domain (each stressed syllable is initial in a prosodie word). When RNALSTRESS outranks ALIGN-HEAD-LEFT, main stress is final when no foot is prespecified in the lexicon. By ranking MAX-FR S higher than RNALSTRESS, Inkelas 16. MAX-FTs merely says that a stressed foot in the input should surface in the output, but nothing is said about the location of that foot in the output. This problem can easily be circumvented by adopting an anchor constraint à la Benua (1997) and McCarthy (1997) which says that a segment that is final in a main stress foot in the input has an identical correspondent that is final in a main stress foot in the output (INPUT-OUTPUT ANCHOR-RIGHT (Foot)).

32

Janet Grijzenhout

captures the fact that the foot structure of the leftmost foot is preserved when more than one foot is prespecified: (31) Lexical specifications of prestressing + initially stressed suffix in Turkish Input: bi.rak +( Χ""1 (for η = 0 , - 1 ) b. X m - > Y m X m (for m < 0)

(level shift) (categorial combination)

No distinction between complements and adjuncts is represented here. This is motivated by the fact that compounds allow both an argument-head relationship and a modifier-head relationship between the two parts. A further difference to phrase structures in syntax consists in the fact that the non-head is not a maximal projection, but a projection of the same level. But the basic analogy to syntax still exists. It consists in the crucial distinction between heads and non-heads, in the fact that recursion is allowed, and in the role of different levels of projections in these recursive structures. In earlier work, I have used the categories root, stem and word as organizing units of morphology, and have argued that a number of generalizations systematically relate to these categories. The domains in which root, stem and word have been found to be useful, if not necessary, units include not only the formation of compounds, as seen above, but also the layering of derivational and inflectional morphology (Wiese 2000:115-149), the formulation of phonotactic generalizations (size constraints and edge marking (Golston & Wiese 1998, Wiese 2001a, Wiese 2001c). Partly on the basis of this work, it is possible at this point to summarize some of the properties of roots, stems and words, as these will be immediately relevant for the following analysis of conversion. A list of category-specific phonological and morphological diagnostics for roots, stems and words without any claim completeness is given in (18). (18) Properties of categories: a. Roots can be: - bound, - marked phonotactically, (see Arabic ktb 'write' or German ordn- 'order'), - category-neutral. b. Stems are: - phonologically possible forms, - categorically specified (Ν, V, A, P). c. Words can be: - free forms in syntax, - complete with respect to all morphosyntactic features required for syntax.

4.

An analysis of conversion

4.1

Core proposal

The scene is now set for the central proposal on conversion in this paper: just as compounding is possible for roots, stems and words, so is conversion. The same of course holds for explicit derivation and inflection by means of affixation. In Wiese (2000:115-

59

A model of conversion in German

149), I argue that plural inflection in German is distributed over roots, stems and words, and that there are at least two types of derivational affixes. For similar proposals with respect to English and German, see also Giegerich (1999). In other words, the three central types of conversion as introduced in (2) to (4) are to be analyzed as conversion of roots, stems and words, respectively. Thus, the verbal root schau 'look', e.g., is converted into a nominal root Schau 'show N '. The structure of this derived root conforms to the pattern given in (8), with the level information (i.e., root) added. Note that this analysis presupposes that roots can be complex just as other units can. There cannot be more types of conversion, because there are neither more nor fewer basic morphological categories (in German). The proposal is summarized here in (19). The following section will present various sorts of evidence for the claims expressed here, again in part based on observations on conversion put forward by Kiparsky (1982a: 140-43). (19) Types of conversion: a. V/A —» N-conversion: b. N/A—» V-conversion: c. V —¥ N-conversion:

root conversion stem conversion word conversion

It would perhaps be possible for German to have fewer than three types of conversion, but this would require an additional explicit principle blocking one of the three categories with respect to conversion. The claim that there are three basic classes of conversion, postulated in (19), and exemplified at the outset of the paper in (2) to (4) can be substantiated with the generalizations discussed in the following section. In analogy to the structures used for compounds above, the word-syntactic rules for conversion then must be as in (20), instantiating the schema put forward in (17). As always, the right-hand head (phonologically empty) determines the category of the whole. If adjectives are included as left-hand members, the categories Ν and V need to be generalized into appropriate features including the adjectives. (20) Rules for conversion: a. root conversion: b. stem conversion: c. word conversion:

[Ν"2] —> [V"2] [V"1] - » [Ν 1 ] [N°] —» [Vo]

[N"2] [V"1] [N°]

The structures defined in this way are completely identical to those assumed for other complex words. The difference is exclusively one of phonological form, consisting in the emptiness of the right-hand head. To speak of "rules for conversion" is misleading insofar as there are no special rules for this domain.

11. Kiparsky's discussion of conversion concerns English, and does not include the third type of conversion, word conversion.

60

Richard Wiese

4.2

Three arguments for three classes

First, root conversion as exemplified in (2) is least productive (see also Olsen 1990:195-6, Fleischer & Barz 1995:209-11, Eisenberg 1998:284). There is no principled way of saying which verbal root can undergo conversion, and which cannot. Furthermore, the process seems to be fossilized. To the extent that converted roots seem to be coined recently at all, they may be regarded as loan formations. Second, the meaning of converted nouns is lexicalized, wide-ranging, and only idiosyncratically related to that of the verb. Third, as also shown in (2a), the deverbal nouns appear with all genders. Fourth, the plural allomorph is also not predictable. The typical affixes for native nouns, -e, -er, -en, all occur in plural forms of root conversion.12 In summary, converted roots have all the property of other native nominal roots of German. Conversions on the basis of ablauted verb forms (see also section 2 above) are examples of root conversion as well. The idiosyncratic patterning includes the behaviour of inseparable prefix verbs. Such verbs often (though not always) allow for conversion of their roots. However, as shown in (21), each verb idiosyncratically selects a subset of prefixes, and an idiosyncratic subset of such complex verbs enters conversion into nouns. (21) Conversion of prefix verbs13 verb root converted noun prefix verb befall fall Fall gefall verfall Tritt betret tret vertret Schluss beschließ schließ verschließ entschließ erschließ — beweis weis erweis verweis

converted noun 'fall' Befall —

Verfall —

'kick'

—

Beschluss Verschluss Entschluss

'close'

—

Beweis

'demonstrate'

—

Verweis

The converted verbs, products of stem conversion, display very different properties. First, this type of conversion is generally productive, as many earlier analyses have noted; see Olsen (1990:196-98), Neef (1998), Eschenlohr (1999), Eisenberg (1998). However, productivity is restricted in various ways, mostly by the existence of competing patterns of explicit word formation. This is particularly relevant for A-V12. The few recent formations of this type ( T r e f f - s 'meetings') all show the plural suffix -s. These forms might belong to a completely different class, possibly loan items which are not yet assimilated to the German system. 13. Examples in (21) are all drawn from the verbs with irregular inflection. Here, the verb roots have an irregular inflection in the prefix verbs identical to the simplex verb, demonstrating that the prefix verb derives from the respective simplex verb.

A model of conversion in German

61

conversion, with the result that this type of conversion is not productive in present-day conversion. There are also prosodie constraints, but these are not specific for conversions, but hold for verbs in general. Second, stem conversion is largely restricted to simplex bases, but it is also found with some noun compounds and with some derived nouns, in particular with those compounds called stem compounds above, and with nouns bearing the nominal suffix -er, see schneidern 'tailor v '. 14 In general, an adequate description of the conditions under which derived nouns can be converted into verbs is still missing; see Eschenlohr (1999) for extensive discussion of phonological, structural and semantic conditions on stem conversion. Whatever the adequate description will be, this set of constraints is particular to this type of conversion. Third, the semantics of stem conversion is, if viewed sufficiently general, rather regular, with the verb denoting an activity in which the referent of the noun is involved as a crucial participant (although with different roles); see Clark & Clark (1979), Stiebels (1996). Word conversion, finally, seems to bear no restrictions at all. Any infinitival verb can be converted into a noun bearing neuter gender. No formal or semantic restrictions are known. It is probably for this reason that Eichinger (2000:22) and others see this type of conversion into nouns, i.e. word conversion, as the central case of conversion.15 This is only true to the extent that more restricted types of conversion (see above) are excluded from analysis for the very reason that they follow restrictions. However, the existence of various types of restrictions is a hallmark of morphological forms in general, and should not exclude a pattern from analysis. The meaning of word conversion is uniform; it is always a reanalysis of the verb's meaning as a referential object. The first line of evidence for the existence of three types of conversion therefore derives from the fact that the formal and the semantic properties of the three types are rather distinct. The second piece of evidence is that the properties found are in accordance with those listed in (18) for roots, stems and words in general. This is also obvious in relation to the fact that the restrictions and exceptions on conversion increase considerably when comparing words with stems, and stems with roots. The prediction (often formulated in the theory of Lexical Phonology) is that morphology becomes less and less exceptional, if moving from the root domain (level I) via the stem domain (level Π) to the word domain (level ΙΠ). Thirdly, there is a particular input-output-relationship between the three types of conversion, a point also addressed in Kiparsky's analysis (Kiparsky (1982b: 142)). It appears that a verbal root can be input to root conversion (22a), just as the infinitive of this root can be input to word conversion (22c), resulting in two types of converted nouns. Both roots (unchanged, as stems) and converted stems can serve as input to word conversion (22b,c). This leads to a particular feeding relationships for conversion, illustrated here by Bagger 'excavator'.

14. Kiparsky (1982a: 141) makes the same observation for English noun-to-verb conversion. 15. Eschenlohr (1999:47ff.), on the other hand, uses the unrestricted productivity and the semantic transparency of deinfinitival nominalizations as an argument to exclude this type from the domain of conversion.

Richard Wiese

62 (22) Input-output-relations: category category shift examples a. root V —» Ν schauv 'looky' BaggerN b. stem Ν —> V 'excavator' c. word V —» Ν schau(en)v 'looky' bagger(n)v 'excavatey'

—> SchauN 'show N ' - » bagger(n)y 'excavate' —> (das) SchauenN 'looking N ' -> (das) Baggern N 'excavating N '

Of all possible feeding relationships between relevant words, only the two illustrated in (22) (underived and converted verbs into nouns) are actually attested. Others, in particular those which are impossible given the specific distribution of conversion rules over the categories, do not occur. In the present proposal, no stipulation is necessary to derive this result. The ultimate interaction of conversion with nearly all available formal mechanisms in word grammar occurs in the derivation of some complex verbs. As demonstrated in (23), the verb root halty 'hold v ' may undergo a series of derivations, finally yielding the complex noun Haushalten 'house holding N '. Note that composition in this example exemplifies stem composition in the sense introduced above. (23) Complex verbs and conversion: root V-N conversion composition halt —> Halt —» Haus+halt 'holdv' 'hold N ' 'household^

N-V conversion V-N conversion —> haus+halt(en) —> (das) Haus+halten 'householdy' 'house holding N '

All three types of conversion are part of this derivation; furthermore, only the order of conversions exemplified here seems possible. This order (from root to stem to word, V —» Ν —> V —» Ν) would need to be stipulated under any different approach; in the present framework it follows from the a priori dominance between the three categories in question. An additional aspect of these complex verb formations is that irregular inflection of the base verb does not transfer to the complex verb. There are some more examples of the sort illustrated in (23), as shown in (24). In all of these cases presented here with their three irregular root forms in the paradigm, an irregular verb is inflected regularly once it is turned into a complex verb. (24) Conversion and inflection: a. Simplex irregular verbs INF lassen 'let' halten 'hold'

PAST ließ

PART gelassen

hielt

gehalten

b. Complex regular verbs INF PAST PART veranlassen veranlaßte veranlaßt 'cause' haushalten haushaltete gehaushaltet 'household'

63

A model of conversion in German haben hatte gehabt 'have' schlagen schlug geschlagen 'hit'

handhaben handhabte gehandhabt 'use' ratschlagen ratschlagte geratschlagt 'discuss'

The verbal bases all display irregular inflections, as witnessed by the preterite and past tense forms. But once they have undergone the layers of derivations given in (22), inflection is regular; see the preterite and past tense forms on the right. Thus, there is no reason to assume that the behaviour of haushalten as in (22) is an idiosyncratic property of this verb. There are other important morphophonological differences between regular and irregular verbs, supporting the view that they belong to different domains of grammar, identified as the root and the word in the present analysis: first, schwa epenthesis in the paradigm of verbal inflection is restricted to the word level (except for liquid-final roots (segel+n 'sailv', ruder-n 'row v '). Second, umlaut and vowel change (mostly raising) are restricted to the root level; degemination is found here as well. (25) illustrates these generalizations. The 3rd person sg. suffix /t/ merges (degeminates) with the preceding /d/ or Ixl in the umlauting and ablauting verbs given in (25a), while the same suffix leads to schwa epenthesis in the case of the regular verbs in (25b). (25) Morphophonological alternations in regular and irregular inflection: a. irregular inflection b. regular inflection lädt [let] leid+et [lajdat] 'load.3SG' 'suffer.3sG' rät [reí] reit+et [Raitat ] 'advise.3SG' 'ride.3SG' tritt [tRit] töt+et [tetat] 'kick.3SG' kill.3SG' bat [bat] bitt+et [bitat] 'bid.PAST.3SG'

5.

'bid.3SG'

Functional considerations

A further question to ask is why conversion is found at all in many languages. The question is relevant as affixless derivation is often felt to be at odds with the preference of languages to be explicit in their marking of categories. In theories of Natural Morphology, in particular, a derived category should be encoded by a derived form, for reasons of iconicity, uniformity and transparency. Conversion clearly violates such principles or preferences, and is therefore less natural than affixal derivation. A partial answer is, first, to note that conversion fulfills the basic task of changing word-class. In other words, conversion serves the basic operations of nominalization and verbalization as elementary categorial shifts between verbalization of objects, acts,

64

Richard Wiese

and events. With respect to the category change, it is just as functional as any other form of derivational morphology. Second, of course, conversion is defined by the lack of any formal marker, and is therefore felt to be "ill-behaved" or marked. Indeed, Kühnhold & Wellmann (1973) and Olsen (1990) emphasize that converted verbs in German tend to be made unambiguous by a verbal prefix. Note that the present analysis somewhat attenuates this view of conversion. If conversion produces nouns for roots, verbs for stems, and nouns again for words, as depicted once again in (26), there is a very systematic alternation or switch between levels or domains. This type of switch has been observed to hold with respect to other properties as well, in particular for stress directionality and for the preferred marking of final edges of roots, stems and words by consonants or vowels, as discussed in Wiese (2001c). This systematic switch removes some more of the arbitrariness from conversion as a morphological oddity because of its lack of formal marking. (26) Alternating pattern of word class change Word _> Ν I Stem I Root

6.

V -»

Ν

A featural analysis of lexical categories

The categories of root, stem and word are crucial elements, but they are not necessarily atomic categories in the grammar of words. The following featural analysis of these categories adopts an idea from the theory of X-bar syntax, namely that bar projections have both a lower and an upper bound. These bounds are expressed by features of minimality and maximality. With these two features, the categories can be analyzed as in (27). (27) Featural break-down for root, stem and word: stem

features

root

[minimal]

+

word

-

-

[maximal]

-

-

+

Under this approach, a natural breakdown of the categories in question into natural classes then is as in (28). Stems can form a group with either roots or words. An interesting consequence of this analysis is that it makes the stem a recursive category: only the stem can recursively dominate itself. On the other hand, there can only be one maximal category, the word, and one minimal category, the root. This interpretation is

A model of conversion in German

65

at odds, however, with the view, expressed above in section 3, that roots can be complex as well as being simple: a root suffixed by a root suffix (class I suffix in the traditional terminology of Lexical Phonology) is a root as well. I leave this matter open. (28) Natural classes:

[-min]

Word [+max] Stem [-min,-max] Root [+min]

[-max]

The present paper is part of a more general inquiry into "ill-behaved morphs" (Anderson 1988). Other studies into such morphology involve areas such as subtractive morphology (Golston & Wiese 1996), reduplication (Wiese 1990), truncation (Wiese 2001b). In each of the cases, the ideal of Hnear-concatenative morphology in which free and bound morphemes mark grammatical categories is seemingly violated. But in all of these cases, closer analysis reveals that there is no convincing reason to assume a real difference in morphological structures. Rather, the seemingly deviant morphology is caused by a simple lack of phonological material (conversion, truncation) or by specific demands on phonological wellformedness.

References Anderson, Stephen R. (1988). Morphological theory. In Linguistics: the Cambridge survey, Vol. 1. Frederick J. Newmeyer (ed.), 146-191. Cambridge: Cambridge University Press. Beard, Robert (1998). Derivation. In The handbook of morphology, Andrew Spencer; and Arnold Zwicky (eds.), 44-66. Oxford: Blackwell Publishers. Chomsky, Noam A. (1970). Remarks on nominalization. In Readings in English Transformational Grammar, R. Jacobs and P. S. Rosenbaum (eds.). Waltham, Mass.: Ginn and Co. Clark, Eve V. and Herbert H. Clark (1979). When nouns surface as verbs. Language 55,767-811. Don, Jan (1993). Morphological conversion. Unpublished Ph. D. dissertation, Utrecht University. Eichinger, Ludwig M. (2000). Deutsche Wortbildung. Eine Einführung. Tübingen: Narr. Eisenberg, Peter (1998). Grundriss der deutschen Grammatik., Vol 1: Das Wort. Stuttgart; Weimar: Verlag J.B. Metzler. Eschenlohr, Stefanie (1999). Vom Nomen zum Verb: Konversion, Präfigierung und Rückbildung im Deutschen. Hildesheim: Olms.

66

Richard Wiese

Fleischer, Wolfgang and Irmhild Barz (19952). Wortbildung der deutschen Gegenwartssprache. Tübingen: Max Niemeyer Verlag. Giegerich, Heinz J. (1999). Lexical Strata in English. Morphological causes, phonological effects. Cambridge: Cambridge University Press. Golston, Chris and Richard Wiese (1996). Zero morphology and constraint interaction: subtraction and epenthesis in German dialects. In Yearbook of Morphology 1995, Geert E. Booij and Jaap van Marie (eds.), 143-159. Dordrecht: Kluwer. —;— (1998). The structure of the German root. In Phonology and Morphology of the Germanic languages, Wolfgang Kehrein and Richard Wiese (eds.), 165-185. Tübingen: Max Niemeyer Verlag. Inkelas, Sharon (1989). Prosodie constituency in the lexicon. Ph.D. thesis, Stanford University. Kiparsky, Paul (1982a). From Cyclic Phonology to Lexical Phonology. In The Structure of Phonological Representations, Harry van der Hülst and Norval Smith (eds.), 131-175. Dordrecht: Foris Publications. —(1982b). Lexical Morphology and Phonology. In Linguistics in the Morning Calm, I.-S. Yang (ed.), 3-91. Seoul: Hanshin Publishing Co. —(1985). Some consequences of Lexical Phonology. Phonology Yearbook 2,,85-138. Kühnhold, Ingeborg and Hans Wellmann (1973). Deutsche Wortbildung: Typen und Tendenzen in der Gegenwartssprache. Band 1: Das Verb. Düsseldorf: Schwann. Lieber, Rochelle (1981). Morphological conversion within a restrictive theory of the lexicon. In The scope of lexical rules, Michael Moortgat, Harry van der Hülst, and Teun Hoekstra (eds.), 161-200. Dordrecht: Foris Publications. Marchand, Hans (1969). The categories and types of present-day English wordformation. A synchronic-diachronic approach. München: Beck. Neef, Martin (1998). A declarative approach to conversion into verbs in German. Yearbook of Morphology 1998, 199-224. Olsen, Susan (1990). Konversion als ein kombinatorischer Wortbildungsprozeß. Linguistische Berichte 127, 185-216. Packard, Jerome L. (2000). The Morphology of Chinese: A Linguistic and Cognitive Approach. Cambridge: Cambridge University Press. Selkirk, Elisabeth O. (1982). The syntax of words. Cambridge, ΜΑ: ΜΓΓ Press. Spencer, Andrew (1991). Morphological Theory. Cambridge: Cambridge University Press. Stiebels, Barbara (1996). Lexikalische Argumente und Adjunkte. Zum semantischen Beitrag von verbalen Präfixen und Partikeln. Berlin: Akademie Verlag. —; and Dieter Wunderlich (1994). Morphology feeds syntax: the case of particle verbs. Linguistics 32, 913-968. Vogel, Petra Maria (1996). Wortarten und Wortartenwechsel. Berlin: de Gruyter. Wiese, Richard (1990). Über die Interaktion von Morphologie und Phonologie: Reduplikation im Deutschen. Zeitschrift für Phonetik, Sprachwissenschaft und Kommunikationsforschung 43, 603-624. —(1996). Phrasal compounds and the theory of word syntax. Linguistic Inquiry 27, 183-193.

A model of conversion in German

67

—(2000). The Phonology of German. Oxford: Oxford University Press. —(2001a). How prosody shapes German words and morphemes. International Journal of Germanic Linguistics and Semiotic Analysis 6, 155-184. —(2001b). Regular morphology vs. prosodie morphology? The case of truncations in German. Journal of Germanic Linguistics 13, 131-177. —(2001c). The structure of the German vocabulary: edge marking of categories and functional considerations. Linguistics 39, 95-115. Williams, Edwin (1981). On the notions "lexically related" and "head of the word". Linguistic Inquiry 12, 245-274. —(1986). Probleme der Wortstruktur. Zeitschrift für Sprachwissenschaft 5, 209-252. —(1987a). An investigation of lexical composition: the case of German be- verbs. Linguistics 25, 283-331. —(1987b). Partizipien im Deutschen. Linguistische Berichte 111, 345-366. —(1996). Lexical categories. Theoretical Linguistics 22, 1-48.

On the verbal nature of certain nominal entities Veronika Ehrich

1.

Introduction

*

Defining the major parts of speech (Ν, V, A, P) according to clear-cut criteria is basic to every grammatical theory. Wunderlich (1996) suggests two bundles of features, [+dependency] and [+articulateness], which characterize the relevant categorial distinctions. Nouns and verbs are referentially independent [-dep] in that they directly refer to entities in an extra-linguistic domain and thus have a referential argument, whereas adjectives and prepositions (having no referential argument of their own) need the support of an independent linguistic element and are considered [+dep], accordingly. Verbs and prepositions are lexically defined for (up to three) obligatory thematic arguments and are classified as [+art] because of that. Nouns and adjectives mostly lacking obligatory thematic arguments are classified as [-art]. Wunderlich discusses various properties defining the feature [+art]: nouns may undergo the possessive extension, they form a presupposition grid (i.e. nominal terms are used in reference to entities whose existence is taken for granted across different possible worlds), and they are the target of semantically empty conversions (p. 15-26). Verbs, while sharing none of these features, have a possible world component, conveyed by morphological markers for mood (such as optative or subjunctive) and status (such as imperative, interrogative, jussive or evidential). The feature [±art] thus covers a bundle of mutually related grammatical properties. The complex nature of the noun-verb distinction implies that a * Many years ago, I happened to attend a seminar quite unrelated to my field of study. It introduced into a subject I had never heard of before: Transformational Grammar. This seminar, given by Dieter Wunderlich, was the first in my student career that gave me an idea of what science, in general, might be about. The experience made me graduate later than I had been supposed to - and with a thesis on nominalizations inspired by Wunderlich (1971). I have always remembered the seminar with gratitude. The research reported here was supported by a grant from DFG (Eh 180/1-3). Suggestions by the editors, Ingrid Kaufmann and Barbara Stiebels, led to useful revisions of an earlier draft. I am also indebted to Ewald Lang, who generously offered helpful comments, while hosting me at ZAS Berlin. Irene Rapp and I have been sharing our views on lexical semantics and nominalizations for several years. The relation to her work is obvious. The remaining shortcomings are, of course, in my own responsibility.

70

Veronika Ehrich

given item can be [±art] to different degrees, depending on how many of the defining properties it instantiates. The present paper discusses event nominalizations as instantiations of hybrid categories. Different types of nominalizations are more or less 'verbal' viz. 'nominal', which is to say that they instantiate the [+art]-properties to different extents. Verbal infinitives (V-Inf) like (1) are nominal in virtue of their grammatical function (subjecthood) and verbal in that they require adverbial modifiers and accusative complements. Nominal infinitives (N-Inf) like (2) or ««^-dérivations (NOM-wng) like (3) with adjectival modifiers and genitive complements are more nominal. (1)

Das Kosten schnell senkenv ist ein vorrangiges Ziel. 'Lowering the costs fast is a prior aim.'

(2)

Das schnelle SenkenN der Kosten ist ein vorrangiges Ziel. 'Fast lowering of the costs is a prior aim.'

(3)

Die schnelle Senk-ungN der Kosten ist ein vorrangiges Ziel. 'The fast lowering of the costs is a prior aim.'

This is common knowledge in linguistics, drawing back to Chomsky (1970) and, as far as German is concerned, to Wunderlich (1971). English, similarly, distinguishes between verbal gerunds (4), nominal gerunds (5) and derivations (6). (4)

John's refusing the offer was a surprise.

(5)

John's refusing of the offer was reluctant.

(6)

John's refusal of the offer was a surprise/was reluctant.

Verbal gerunds or 'imperfect nouns', as they are called by Vendler (1967a), can be used in reference to facts, nominal gerunds in reference to events (see Zucchi 1993 and Hamm 1999 for further discussion). The semantic split between these two types of English gerunds is different from that between verbal and nominal infinitives in German. Infinitive nominalizations of type V-Inf refer neither to facts, nor to particular events, but to generic events in the sense of Montague (1974), cf. Ehrich (1991). (7)

*(Das) die Spende ablehnen war eine Überraschung. 'Refusing the donation was a surprise.'

(8)

*(Das) die Spende ablehnen war zögerlich. 'Refusing the donation was slow.'

(9)

(Das) Spenden ablehnen fällt manchem schwer. 'Refusing donations is difficult to some people.'

Nominalizations instantiate most of the properties characterizing nouns, but due to their verbal inheritance they also share certain [+art] properties of verbs. Depending on the amount of [+art] properties instantiated, nominalizations of morphologically distinct types can be ranked with respect to articulateness in the following way:

71

On the verbal nature of certain nominal entities (10) [-art]

[+art]

< Eventive < Eventive < Eventive Eventive zero conversion NOM-wngN-Inf conversion V-Inf conversion Nominalizations of the V-Inf type rank highest, zero conversions rank lowest on the articulateness scale. Verbal infinitive nominalizations behave like verbs in that they don't admit the omission of obligatory arguments, not even where the argument in question is inferable from contextual or factual knowledge. (11) a. Die Stadtreinigung entfernt *(den Müll) von den Straßen. 'The collection department removes (the garbage) from the streets.' b. Der Briefträger überbringt *(die Post). 'The post man delivers (the mail).' (12) a.

*[...] Entfernen von den Straßen ist Aufgabe der Stadtreinigung. 'Removing [...] from the streets is the task of the collection department.' b. *[...] Überbringen fällt dem Postboten zunehmend schwerer. 'Delivering [...] is continuously getting harder for the post man.'

Implicit derivations/zero-conversions rank lowest on the articulateness scale, as is indicated by the fact that they do not admit the THEME argument of the verb in the structural position of the internal argument. (13) a. Jonathan tritt den Sack. 'Jonathan kicks the sack.' b. Er stößt den Dieb von der Mauer. 'He pushes the burglar off the wall.' (14) a.

*der Tritt des Sacks 'the kick of the sack' b. *der Stoß des Diebs von der Mauer 'the push of the burglar off the wall'

The present paper is devoted to the intermediate values on the scale, wng-nominalizations and nominal infinitive nominalizations. In the remaining part of the paper, I will discuss their articulateness properties with respect to thematic structure (section 2.1) as well as event structure (section 2.2) and offer a more formal account for the semantic representation of nominalizations (section 3).

72

Veronika Ehrich

2.

Articulateness properties

2.1

N o m i n a l i z a t i o n a n d T h e m a t i c Structure

2.1.1

Structural properties of German nominalizations

Events usually involve at least one (possibly more than one) participant. Nominalizations referring to events are therefore deeply relational. They are transparent for thematic role information, even where the relata in question are left implicit ((15), (16)). (15) a.

Die Behandlung [...] nahm viel Zeit in Anspruch. 'The treatment [...] took a lot of time.' b. Die Vernehmung [...] hat schon begonnen. 'The interrogation [...] has already started.'

(16) a.

Die Verschönerung [...] ist gelungen. 'The remodelling [...] was successful.' b. Die Befreiung [...] kam überraschend. 'The liberation [...] came as a surprise.'

t/ng-nominalizations differ from verbs in that obligatory verb arguments (17) become optional ((18), see also Bierwisch 1989). (17) a.

Wickert befragte *(den Kanzler). 'Wickert interviewed (the chancellor).' b. Der Mediziner befruchtete *(das Ei) in vitro. 'The medic fertilized (the ovum) in vitro.' c. Ein Erdbeben zerstörte *(die Stadt). 'An earthquake destroyed (the city).'

(18) a.

Die Befragung (des Kanzlers) wurde unterbrochen. 'The interview (of the chancellor) was interrupted.' b. Die Befruchtung (eines Eis) in vitro ist heute nicht mehr schwierig. 'The fertilization (of an ovum) in vitro isn't difficult anymore.' c. Nach der Zerstörung [ i ] wurde die Stadtj bald wieder aufgebaut. 'After the destruction [of i] the city¡ got soon rebuilt.'

Omission of the internal argument is possible in nominal infinitives, as well, but requires contexts coercing a generic-event reading (19) instead of a particular-event reading (20). N-Inf constructions behave like the corresponding V-Inf-constructions in this respect. (19) a.

Kluges Befragen will gelernt sein. 'Sensible interviewing will have to be learned.' b. Künstliches Befruchten ist in der Landwirtschaft üblich. 'Artificial inseminating is common in rural economy.'

On the verbal nature of certain nominal entities c. (20) a. b. c.

73

Jedes Zerstören freut den Vandalen. 'Any destroying pleases the vandal.' ??Das Befragen wurde unterbrochen. 'The interviewing got interrupted.' ??Das künstliche Befruchten verhalf ihnen zu einem Kind. 'The artifical inseminating helped them to a child.' ??Dieses Zerstören freute den Vandalen. 'This destroying pleased the vandal.'

Grimshaw (1990) distinguishes between complex event nomináis (CEN) and result nomináis (RN). CEN (21) always preserve the argument structure of their base verb, RN, denoting objects (22) or simple events (23), have no argument structure. (21) a. The instructor's examination of the papers took a long time, b. The enemy's destruction of the city was awful to watch. (22) a. The examination/exam for the students is on the table. b. The book's translation/the translation of the book is on the table. (23) a. Reagan's defeat was a surprise. b. The politician's nomination was a surprise. CEN as opposed to RN behave like verbs in that they permit event control (24) (cf. Roeper 1993) and carry aspectual meanings, as is indicated by combinations with frequency predicates (25). (24) a. the translation of the book in order to make it available to a wide readership b. the assassination of the dictator, in order to preserve peace c. *the book's translation, in order to make it available to a wider readership d. *the murder in order to fight terrorism (25) a. b. c. d.

The frequent examination of one's feelings is desirable. The constant assignment of unsolvable problems is to be avoided. *The frequent expression is desirable. *The politician's frequent/constant nomination.

The prenominai genitive is almost restricted to proper nouns (26a) in modern German (see Olsen 1991). Fully articulate structures like (26b,c) are marginally acceptable. (26) a.

Caesars Zerstörung der Stadt 'Cesar's destruction of the city' b. ??des römischen Kaisers Zerstörung der Stadt 'the roman emperor's destruction of the city' c. ??des römischen Kaisers Zerstören der Stadt 'the roman emperor's destroying of the city'

This raises the question, whether the structural distinction between CEN and RN can be drawn in the same way for German and English. The prototypical construction of a

74

Veronika Ehrich

German img-nominalization such as (27) even suggests that German doesn't have CEN. (27) die Zerstörung der Stadt 'the destruction of the city' [DP X D° Υ [NP Ν DPoen]] This assumption is supported by the fact that German ung-nominalizations don't permit event control, not even in fully articulated structures like (28a). (28) a.

b.

*Caesars Zerstörung der Stadt, um Gallien zu erobern, kam überraschend. 'Cesar's destruction of the city, in order to capture Gaulle, came as a surprise' *Die Zerstörung der Stadt, um den Terrorismus zu bekämpfen, brachte vielen den Tod. 'The destruction of the city, in order to fight terrorism, brought death to many people'

If German nominalizations are not of the CEN type, they might not have argument structure at all. The genitives accompanying them should then be analyzed as argument adjuncts (AAs), which are conceptually linked to the arguments of a verb, but behave like adjuncts in terms of their syntax (see Grimshaw 1990). Indeed, the fact, that German «ng-nominalìzations never have obligatory complements, supports the assumption that the head adjacent genitive DPs are AAs instead of arguments. However, obligatoriness is not criterial for argument status: for some verbs, arguments may be optional, others have obligatory adjuncts (Höhle 1978, Vater 1981).

2.1.2

Thematic asymmetries in nominalizations

Due to the loss of a fully articulate structure in German NPs (cf. Demske 2000), there is only one structural position available for hosting the arguments of the base verb. This position filled by the post-nominal genitive is head-adjacent by rule.1 The adherence constraint has the effect that only one of the arguments of a given verb can be spelled out as a post-nominal genitive, such that both arguments of a transitive verb have to compete for being linked to the NP-internal argument position. If nominalizations shared the linking behavior of verbs, the element surviving in the NP-internal argument position would always be identical to the internal argument of the verb. The wng-nominalizations in (29) adhere to the verbal pattern, indeed: The internal THEME argument (TH) of the verb shows up as post-nominal-genitive in (29) and the AGENT argument is excluded from this position. ( 2 9 ) a.

die A u f l ö s u n g des VertragestH

/*des ArbeitgebersAG

'the cancellation of the contract'

/'of the employer'

1. Exceptions to the rule are discussed in Löbel (1990).

75

On the verbal nature of certain nominal entities b. die Abschaffung des Magisterstudiengangs™ 'the abolition of the MA curriculum' c. die Entkernung der Zelle™ 'the denucleation of the cell' d. die Verteilung der Beute™ 'the distribution of the haul' ... kam überraschend. '... came as a surprise.'

/*des DekansAG /'of the dean' /*des ForschersAG /'of the researcher' /*des RäubersAG /'of the robber'

Nominalizations seem to adhere to the same linking constraints as verbs. However, the this regular linking behavior is restricted to nominalizations of accomplishments and achievements, ««^-nominalizations of activities deviate from the standard pattern in that they admit AGENTs (or THEMES) in NP-internal position (30). (30) a.

die Vernehmung des Zeugen™ 'the examination of the witness' b. die Behandlung des Patienten™ 'the treatment of the patient' c. die Beobachtung des Verdächtigen™ 'the surveillance of the suspect' d. die Messung der Emission™ 'the measuring of the emission' ... wird fortgesetzt ('is continued')

/des RichtersAG /'of the judge' /des ArztesAG /'of the doctor' /der PolizeiAG /'of the police' /des Ingenieurs A o /'of the engineer'

As pointed out in section 1, implicit derivations (zero-conversions) even restrict this position to the AGENT argument (31). (31) a

b. c.

*der Tritt des Sacks·™

der Tritt des PferdesAG

'the kick of the bag' *der Kuss des Rings™ 'the kiss of the *der Dreh des Knopfs ΤΗ 'the turn of the button'

'the kick of the horse' der Kuss der Muse A o 'the kiss of the muse' der Dreh des Tänzers 'the turn of the dancer'

ring'

We have thus three different patterns for representing verb arguments in nominalizations: a verb-like pattern restricting the critical position to the THEME argument, a nominal pattern in implicit derivations restricting it to the AGENT and the split pattern giving equal priority to AGENT and THEME. Infinitive nominalizations of the N-Inf type adhere to the verbal pattern, irrespective of whether they have accomplishment and achievement or activity bases (32). (32) a. b. c.

das Vernehmen des Zeugen ΤΗ 'the interrogating of the witness' das Entkernen der Zelle™ 'the denucleating of the cell' das Überwachen des Verdächtigen™ 'the surveying of the suspect'

/*des Richters AG /'of the judge' /*des ForschersAG /'of the researcher' /*der PolizeiAG /'of the police'

76

Veronika Ehrich d. das aas Messen messen der aer Emissioncmission-m 'the measuring of the emission'

/*des IngenieursAG /'of the engineer'

The asymmetries between (29) and (30) suggest a conceptual explanation: processes (activities), as opposed to events (accomplishments and achievements), highlight the course of action and therefore give equal attention to its object and its instigator. Accomplishments and achievements put the focus on the outcome of the action and give higher priority to participants undergoing a change in virtue of the outcome. The conceptual explanation, however, applies only to img-nominalizations. The nominal infinitives in (32) denote processes, too, (as is indicated by combinations with time span predicates (33)), but exclude the AGENT from the internal argument position, nonetheless. (33) a.

Das stundenlange Vernehmen des Zeugen hat den Richter ermüdet. 'The interrogating of the witness going on for hours has tired the judge.' b. Das tagelange Überwachen des Verdächtigen war ergebnislos. 'The surveying of the suspect going on for days didn't lead to a result.' c. Das wochenlange Messen der Emissionen war ermüdend. 'The measuring of the emissions going for week was tiring.'

If the asymmetries discussed above were to be explained in purely conceptual terms, the formal construction type of the nominalization shouldn't make a difference. 2 The fact that the thematic interpretation interacts with the morphological type of the nominalization indicates that the asymmetries discussed are rooted in grammar.

2.2

Event Structure

Events take place at certain locations and limited times. Nominalizations denoting events show a reflex of the temporal structure inherent to events, though to different degrees - depending on the morphological construction type. 2.2.1

Tense and mood

Verbs are marked for finiteness/infiniteness, nouns for definiteness/indefiniteness. The functional categories (I and D), specifying these features, both link a linguistic expression to a domain of reference. Events as denoted by verbs are situated in time via aspect and tense; objects denoted by nouns are essentially non-temporal, they receive an indirect time specification by reference to the events they participate in. Nominalizations, in virtue of being nouns, are untensed by default. Verbal infinitives (like verbal gerunds in English) do occur in perfect form (34b). The perfect form is possible, albeit fairly uncommon, in nominal infinitives (35) as well. 2. Things get more complicated, once we consider pluralized ung-nominals, cf. die Ausgrabung Schliemanns 'the excavation of Schliemann' as opposed to die Ausgrabungen Schliemanns 'the excavations by Schliemann'. See Ehrich (2002) for details.

On the verbal nature of certain nominal entities

11

(34) a.

(Das) die Marseillaise singen macht Spaß. 'Singing the Marseillaise is fun.' b. (Das) die Marseillaise gesungen haben ist ein erhebendes Gefühl. 'To have been singing the Marseillaise is an uplifting feeling.'

(35) a. b.

?Das Gesungenhaben der Marseillaise ist ein erhebendes Gefühl. 'The having been singing the Marseillaise is an uplifting feeling.' ?Das Gerettetsein der Bergsteiger ist eine Überraschung. 'The having been rescued of the mountaineers is a surprise.'

Derived nomináis are necessarily untensed, (36a) doesn't give us any linguistic indication as to whether the opening of the world championship is a past, present or future event. (36) a.

die Eröffnung der Weltmeisterschaft 'the opening of the world championship' b. die Eröffnung der Weltmeisterschaft im Jahr 2000 'the opening of the world championship in 2000' c. Die Eröffnung der Weltmeisterschaft war schwach besucht. 'The opening of the world championship was poorly attended.' d. Die Eröffnung der Weltmeisterschaft wird eine Sensation. 'The opening of the world championship will be a sensation.'

Temporal information about the event denoted by an «ng-nominalization has to be derived from encyclopedic knowledge (36a), is explicitly added by use of a time adverbial (36b), or may be inferred from the tense of the matrix verb (36c,d). Event splitting in the sense of Reichenbach (1947) is, therefore, defined as an operation over untensed propositions (37a). A nominal, then, receives its temporal reference from a wide scope tense operator as in (37b). (37) a.

the opening of the world championship tv [OPEN (x, the world championship)]* (v) b. The opening of the world championship was fun. PAST (FUN (tv [OPEN (x, the world championship)]* (v))

Modal information (distinguishing factual from hypothetical and counterfactual statements) is inexplicable in nouns, even verbal infinitives exclude modal markings. The modal status of an event noun has to be inferred from the logical status of the matrix predicate (extensional vs. intensional, (38)) and/or from the sentence mode (39). (38) a. Die Zerstörung der Stadt war verheerend, (extensional, factual) 'The destruction of the city was devastating.' b. Die Zerstörung der Stadt kommt nicht in Frage, (intensional, non-factual) 'The destruction of the city is out of the question.' (39) a. Die Zerstörung der Stadt wäre verheerend, (irrealis, hypothetical) 'The destruction of the city would be devastating.'

Veronika Ehrich

78

b. Die Zerstörung der Stadt wäre nicht in Frage gekommen, (irrealis, counterfactual) 'The destruction of the city would have been out of the question.' c. Niemand wollte die Zerstörung der Stadt. 'Nobody wanted the destruction of the city.' Extensional predicates (38a) presuppose that the event mentioned in the nominalization did, indeed, occur. Intensional predicates (38b) and hypotheticals (39a) leave this open. Counterfactuals (39b) indicate that the event in question did not occur. Wollen is underspecified in this respect, (39c) can be truthfully asserted about a situation where either the city was destroyed against everybody's will, or, where it was not destroyed, because everybody's will was opposed to a destruction. The non-factual interpretation of the nomináis in (38b), (39) can be highlighted by the use of the indefinite article (40). (40) a. Eine Zerstörung der Stadt kommt nicht in Frage. Ά destruction of the city is out of the question.' b. Eine Zerstörung der Stadt wäre verheerend. Ά destruction of the city would be devastating.' c. Eine Zerstörung der Stadt wäre nicht in Frage gekommen. Ά destruction of the city would have been out of the question.' d. Niemand wollte eine Zerstörung der Stadt. 'Nobody wanted a destruction of the city.' e. Niemand wollte eine Zerstörung der Stadt, sie geschah durch ein Versehen. 'Nobody wanted a destruction of the city, it happened by mistake.' However, indefiniteness as such does not guarantee a non-factual reading, (40e) should otherwise be inconsistent. 2.2.2

Situation type (aktionsart)

Eventive «rag-nominalizations always preserve the situation type (aktionsart) of their verbal base and, hence, behave like verbs when combined with temporal predicates. Time span predicates apply to nominalizations of activities (41), temporal frame predicates to nominalizations of accomplishments (42). (41) Time span predicates a. die einstündige Überwachung/Vernehmung/Verfolgung des Verdächtigen 'the surveillance/interrogation/chasing of the suspect for an hour' b. *die einstündige Füllung des Zahns/Ausgrabung der Mumie/Überquerung des Flusses 'the filling of the tooth/excavating of the mummy/crossing of the river for an hour'

On the verbal nature of certain nominal entities c.

79

*die einstündige Erreichung des Gipfels/Entdeckung des Diebs/Tötung des Feindes 'the reaching of the summit/discovering of the burglar/killing of the enemy for an hour'

(42) Temporal frame predicates a. *die Überwachung/Vernehmung/Verfolgung des Verdächtigen innerhalb einer Stunde 'the surveillance/interrogation/chasing of the suspect within an hour' b. die Füllung des Zahns/Ausgrabung der Mumie/Überquerung des Flusses innerhalb einer Stunde 'the filling of the tooth/the excavating of the mummy/the crossing of the river within an hour' c. die Erreichung des Gipfels/Entdeckung des Dieb/Tötung des Feindes innerhalb einer Stunde3 'the reaching of the summit/the discovering of the burglar/the killing of the enemy within an hour' In other words, ««g-nominalizations, though unmarked for tense and mood, are still sufficiently articulate for preserving the aktionsart of their base verb.4 Due to their inherently temporal nature, they occur in frequency contexts and iterative contexts, which is impossible for object nouns (43). (43) a. Die erneute Veröffentlichung dieser alten Theorie kam überraschend. 'Publishing this old theory once more was a surprise.' a'. *Das erneute Buch kam überraschend. 'The book once more was a surprise' b. die Wiederaufführung der Oper 'performing the opera again' b \ *das Wiederlibretto der Oper 'the libretto of the opera again' Nominal infinitives unlike ««^-derivations always denote processes, irrespective of the underlying verb. This implies that they preserve the aktionsart, when applied to activity verbs, but alter it, when applied to accomplishments. (44) a. das stundenlange Überwachen/Vernehmen/Verfolgen des Verdächtigen 'the surveying/examining/chasing of the suspect going on for hours'

3. Temporal frame predicates combining with nominalizations of achievements (42c) don't specify the time of the event denoted by the nominalization, but the time of a preceding event leading to the achievement in question (see Zybatow 2001). 4. The fact that aktionsart survives in ««^-derivations is, by the way, a strong indicator for its essentially lexical nature.

80

Veronika Ehrich b. das stundenlange Füllen des Zahns/Ausgraben der Mumie/Überqueren des Flusses 'the filling of the tooth/excavating of the mummy/crossing of the river going on for hours' (45) a.

*das Überwachen/Vemehmen/Verfolgen des Verdächtigen innerhalb einer Stunde 'the surveying/examining/chasing of the suspect within an hour' b. ??das Füllen des Zahns/Ausgraben der Mumie/Überqueren der Straße innerhalb einer Stunde 'the filling of the tooth/excavating of the mummy/crossing of the river within an hour'

The fact that ung-derivations are sensitive to aktionsart is a reflex of their verbal origin and shows that they are more articulate than ordinary nouns. Alternations with respect to aktionsart form part of the grammar of verbs. The alternation brought about by the infinitive nominalization reflects its high rank on the articulateness scale. There are instances of event nominalizations which seem to be underspecified with respect to their event structure properties. Two cases must be distinguished here: either the verb itself allows for different event structure interpretations (behandeln 'treat' in (46) is a case in point), or it has (like verschmutzen 'pollute' in (47)) two different diatheses. In both cases, the eventive w/ig-nominalization is interpretable in each of the readings permitted by the verb. (46) a. Der Arzt behandelte die Krankheit drei Wochen lang, (activity) 'The doctor treated the illness for three weeks.' b. Der Arzt behandelte die Krankheit innerhalb von drei Wochen, (accomplishment) 'The doctor treated the illness within three weeks.' c. die dreiwöchige Behandlung der Krankheit 'the treating of the illness for three weeks' c'. die Behandlung der Krankheit innerhalb von drei Wochen 'the treating of the illness within three weeks' (47) a. Der Nachbar verschmutzte das Wasser mit Motoröl. (causative/ accomplishment) 'The neighbor polluted the water with engine oil.' b. Motoröl verschmutzt das Wasser, (anticausative/state) 'Engine oil pollutes the water.' c. die Verschmutzung des Wassers innerhalb eines Tages, (causative/ accomplishment) 'the polluting of the water within a day' c'. die bestehende Verschmutzung des Wassers (anticausative/state) 'the existing pollution of the water' Underived event nouns like Fest 'festivity', Gewitter 'thunderstorm', Sturm 'storm' are never underspecified with respect to event structure properties (48), meteorological

On the verbal nature of certain nominal entities

81

zero-conversions must be interpreted as semelfactives (49a,b), nominal infinitives, as pointed out above, are restricted to process interpretations (49c). (48) a.

das tagelange Fest/?das Fest innerhalb einer Woche 'the party lasting for days/the party within a week' b. das drei Stunden dauernde Gewitter/*das Gewitter innerhalb von drei Stunden 'the thunderstorm that lasts three hours/the thunderstorm within three hours'

(49) a.

*der minutenlange Donner, der sekundenlange Blitz 'several minutes of thunder, lightning of a few seconds' b. der Donner Schlag Mitternacht, der Blitz um Punkt zwölf 'the thunder at midnight precisely, the lightening precisely at twelve' c. das minutenlange Donnern, das stundenlange Blitzen 'several minutes of thundering, lightening of several hours'

2.2.3

Nominalizations in temporal contexts

Eventive img-nominalizations occur as complements to temporal prepositions (50a) in virtue of the temporal nature of events. Base nouns are possible in temporal contexts as well (50b), but have to be reinterpreted as event-denoting expressions in these cases; the reinterpretation can be made explicit by an additional verb (50c). (50) a.

Vor der Entdeckung des Penicillins war Lungenentzündung sehr gefährlich. 'Before the discovery of penicillin, pneumonia was very dangerous.' b. Vor dem Penicillin war Lungenentzündung sehr gefährlich. 'Before the penicillin, pneumonia was very dangerous.' c. Bevor das Penicillin entdeckt wurde, war Lungenentzündung sehr gefährlich. 'Before penicillin was discovered, pneumonia was very dangerous.'

Temporal prepositions denote a relation between the time of the event under discussion, the thematic event e, and the time of a RELATUM event e' (see Bierwisch 1988, Herweg 1989, Klein 1990, Wunderlich 1990). (50a) says that the time, when pneumonia was dangerous (= time of the thematic event ), is anterior to the time of the discovery of the penicillin (= the time of the RELATUM event). Nominalizations of different situation types differ with respect to the RELATUM time they select (cf. Ehrich & Rapp 2001). Vor 'before' in combination with nominalizations of activities selects the beginning of the relatum event (51), nach 'after' selects its (contingent) termination (52). (51) a.

Vor der Belagerung der Stadt versorgten sich die Einwohner mit Vorräten. 'Before the siege of the city, the inhabitants provided themselves with stocks.'

82

Veronika Ehrich b. Vor der Überwachung des Verdächtigen suchte der Detektiv eine Bar auf. 'Before the surveillance of the suspect, the sleuth visited a bar.' (52) a.

Nach der Belagerung bauten die Einwohner die Stadt wieder auf. 'After the siege, the inhabitants rebuilt the city.' b. Nach der Überwachung des Verdächtigen alarmierte der Detektiv die Polizei. 'After the surveillance of the suspect, the sleuth alerted the police.'

Nominalizations of accomplishments provide two different RELATUM times for vor 'before' : the beginning of the relatum event or its completion. (53) a.

Vor der Renovierung der Wohnung legt der Maler einen Kostenvoranschlag vor. 'Before the redecoration of the place, the painter presented his costs estimate.' b. Vor der Renovierung der Wohnung kann sie nicht einziehen. 'Before the redecoration of the place, she cannot move in.' c. Vor der Verlesung des Urteils begann der Richter zu husten. 'Before the reading out of the verdict, the judge started to cough.' d. Vor der Verlesung des Urteils darf der Angeklagte den Saal nicht verlassen. 'Before the reading out of the verdict, the defendant is not allowed to leave the courtroom.'

(53a) says that the painter presents his estimate before he starts redecorating the place, (53b) says that she can't move in, before the redecoration of the place has been accomplished. In other words, (53a) relates the thematic event to the beginning, (53b) relates it to the completion of the RELATUM event. Nach 'after', on the other hand, provides only a single RELATUM time, which is given by the completion of the RELATUM event in question. (54) a.

Nach der Renovierung der Wohnung legt der Maler die Rechnung vor. 'After the redecoration of the flat, the painter presented the invoice.' b. Nach der Verlesung des Urteils brach der Angeklagte zusammen. 'After the reading out of the verdict, the defendant broke down.'

Nominalizations, though necessarily untensed, thus convey temporal readings similar to those encoded by tensed verbs. (53a) has a present tense paraphrase, whereas (53b) is to be paraphrased by a present perfect. Similarly, (53c) has a simple past paraphrase and (53d) requires a past perfect paraphrase. (53') a.

Bevor er die Wohnung renoviert, legt der Maler einen Kostenvoranschlag vor. 'Before he redecorates the flat, the painter presents a costs estimate.' b. Bevor die Wohnung renoviert worden ist, kann sie nicht einziehen. 'Before the flat has been redecorated, she cannot move in.'

On the verbal nature of certain nominal entities

83

c. Bevor er das Urteil verlas, hustete der Richter nervös. 'Before he read out the verdict, the judge was coughing nervously.' d. Bevor das Urteil verlesen worden war, hatte niemand mit der Verurteilung des Angeklagten gerechnet. 'Before the verdict had been read out, nobody had expected the defendant's conviction.' The temporally unambiguous sentences in (54) permit only the (present or past) perfect paraphrases (54'). (54') a. Nachdem er die Wohnung renoviert hat, legt der Maler die Rechnung vor. 'After he has redecorated the flat, the painter presents the invoice.' b. Nachdem das Urteil verlesen worden war, brach der Angeklagte zusammen. 'After the verdict had been read out, the defendant broke down.' Vor 'before' in combination with infinitive nominalizations always relates the thematic event to the beginning of the relatum event, never to its completion. (55a) says that the painter presented his estimate before he started redecorating the place. (55) a. Vor dem Renovieren der Wohnung legt der Maler einen Kostenvoranschlag vor. 'Before the redecorating of the place, the painter presents a cost estimate.' b. ?Vor dem Renovieren der Wohnung kann sie nicht einziehen. 'Before the redecorating of the place, she cannot move in.' c. Vor dem Verlesen des Urteils begann der Richter zu husten. 'Before the reading of the verdict, the judge began to cough.' d. ?Vor dem Verlesen des Urteils darf der Angeklagte den Saal nicht verlassen. 'Before the reading out of the verdict, the defendant may not leave the hall.' In other words, the culmination point is not accessible as relatum time for infinitive nominalizations. Nach 'after' is therefore just marginally acceptable in combinations with nominal infinitives (56). (56) a.

?Nach dem Renovieren der Wohnung legt der Maler die Rechnung vor. 'After redecorating the place the painter presents the invoice.' b. ?Nach dem Verlesen des Urteils brach der Angeklagte zusammen 'After the reading out of the verdict, the defendant broke down.'

This is evidence in favor of the assumption defended above that nominal infinitives always denote processes. They preserve the aktionsart if the base verb itself is an activity, but alter it if it is an accomplishment. The examples in (56) illustrate another difference between nominal infinitives and img-derivations. The former, but not the latter, seem to be obligatorily controlled by the subject of the matrix verb (57).

84

Veronika Ehrich (57) a.

Vor dem [i] Renovieren legt der Maler¡ einen Kostenvoranschlag vor. Vor dem [i] Verlesen des Urteils begann der Richte^ zu husten.

The low acceptability of (55b,d) results from the fact that subject control is incompatible with the semantic content of the sentences in question. Obligatory control thus provides another piece of evidence for the high degree of articulateness characterizing infinitive nominalizations.

3.

Semantic representations

As pointed out in Wunderlich (1996), verbs as well as nouns have at least one argument: the referential argument. Relational nouns like kinship terms are defined for a RELATUM argument (y in (58a)) as well as for a silent referential argument (x printed bold face and bound by the innermost lambda in (58a)). Non-relational nouns have the referential argument as their single argument (58b). (58) a.

Vater 'father' Xy λ χ VATER (χ,y) b. Mann 'man' λ χ [MANN (χ)]

Nominalizations denoting situations (processes, states, events) are defined for a referential situation argument inherited from the corresponding verb (see Davidson 1967). Vernehmen 'interrogate' as a verb is defined for an AGENT x, a THEME y and (since it is an activity verb) for a process argument r as its referential argument (printed bold face in (59)). Each of these arguments is projected to syntax and thus bound by (a bold face) lambda in (59a): The representation for the corresponding nominalization (59b) is only slightly different: Each of the thematic arguments is a candidate for the single argument position provided by the nominal construction, but may also be left implicit, which is why the lambdas binding these arguments are put into parentheses. (59) a.

vernehmen 'interrogate' Xy λ χ Xr [DO ((x,y), r)] b. Vernehmung 'interrogation' (λ Υ ) (λχ) λ Γ [DO ((x,y), r)]

Things get more complicated for accomplishments or achievements and their nominalizations. The LSS of these verbs consists of three sublexical predicates DO, BECOME/DEVELOP and BE (60), (61); each is defined for its own situation argument: DO for a process argument r, BE for a state argument s, BEOME/DEVELOP for an event variable e (see Pustejovsky 1991, 1995 for a similar approach). 5 5. Accomplishments and achievements both denote events (= changes of state), which is why both are assigned an event variable e. They differ with respect to the temporal characteristics of the event in question (protracted vs. non-protracted event). This is why they are represented by differ-

On the verbal nature of certain nominal entities

85

reinigen 'clean' 6 λy λ χ λβ [CAUSE (, )] 7 b. Reinigung 'cleaning' (Xy) λβ [CAUSE (, )]

(60) a.

Of course, a given verb like reinigen (60) has only one referential argument; therefore a selection has to be made between the situation type variables (r, s, e) of the different sublexical predicates. This selection is based on a situation type hierarchy (STH), comparable to the well-known thematic hierarchy (see Ehrich & Rapp 2000). (61) Situation Type Hierarchy (STH) a. events e > processes r > states s b. For any decomposition of a verb, the situation argument ranking highest in STH is the referential argument As has been pointed out in section 2.1, nominalizations of accomplishments or achievements differ from nominalizations of activities in that only the THEME is permitted in the structural argument position (of the post-nominal genitive) and can thus be bound by lambda. The AGENT argument χ is obligatorily implicit and not bound by lambda in (60), accordingly. Some wng-nominalizations can be used in reference to events or objects. Lieferung 'delivery' is a case in point. The situation argument of the verb (62a) is preserved as referential argument of the event nominalization Lieferung (62b). The possession y (the thing to be delivered) is made the referential argument of the corresponding object nominalization, and bound by lambda in (62c); the situation argument is left unbound. (62) a.

liefern λζ λy λχ λβ [ CAUSE (, )] b. Lieferungενεη( (λζ) Xe [ CAUSE (, )] c. Liefe rung object Xy [ CAUSE (, )]

There has been some debate with respect to the linking asymmetries between ungnominalizations of change of possession verbs like liefern/Lieferung 'deliver/delivery' and ««^-nominalizations of applicative verbs like beliefern/Belieferung 'supply/supplying' While the possession y becomes the internal argument of liefern/Lieferung (63), the possessor figures as internal argument of beliefern/Lieferung (64).

ent sublexical predicates (DEVELOP for accomplishments (60), BECOME for achievements (62), see Rapp (1997) for details. 6. Different printing types for parentheses have no special status here, but may serve to improve the legibility of the formulas. 7. Unlike Wunderlich (1997), I assume that CAUSE is a relation between eventualities, see also Dowty (1979).

86

Veronika Ehrich (63) a. Die Firma liefert dem Hotelz das Gemüsey. 'The company delivers the vegetables to the hotel.' b. die Lieferung des Gemüsesy an das Hotelz 'the delivery of the vegetables to the hotel' c. *die Lieferung des Hotelsz mit Gemüsey 'the delivery of the hotel with the vegetables' (64) a. Die Firma beliefert das Hotelz mit Gemüsey. 'The company supplies the hotel with vegetables.' b. die Belieferung des Hotelsz mit Gemüsey 'the supplying of the hotel with vegetables' c. *Die Belieferung des Gemüsesy an das Hotelz 'the supplying of the vegetables to the hotel'

The standard explanation for these asymmetries recurs to the fact that the post-nominal genitive is a structural counterpart the accusative governed by a given verb (Toman 1983). This assumption, however, does not explain the different linking patterns for the verbs liefern and beliefern, in the first place. Adopting a semantically based approach to argument structure, one can show that, contra to the first impression, the thematic structures of liefern and beliefern are distinct. Liefern is a change of possession verb, beliefern is an applicative verb. The 'hotel' ζ is POSSESSOR (= first argument of POSS) in (60) and GOAL (= second argument of APPL) in (65). (65) a.

beliefern λζ Xy λχ λβ [CAUSE ( , )] b. Belieferungeveat (λζ) λε [CAUSE ( , )]

Following Rapp (1997), I assume that each sublexical predicate defines its own thematic hierarchy, the first argument of a sublexical predicate Ρ being always higher than the second. The POSSESSOR ζ as first argument of POSS is higher than the POSSESSUM y in (62). The APPLICATUM y as first argument of APPLY is higher than the GOAL ζ as second argument of APPLY in (65). (62) and (65) both link the lowest argument of the inchoative component (DEVELOP/BECOME-component) to the structural position of the NP-intemal argument. Lfag-nominalizations of accomplishments and achievements behave like V-Inf in this respect, but show different behaviors with respect to the situation argument. Eventive ung-nominalizations preserve the situation type of the corresponding verb, whereas infinitive nominalizations of the N-Inf-type always denote processes and thus alter the situation type, when applied to accomplishment verbs like reinigen. The process argument of the DO component becomes referential in infinitive nominalizations (66), accordingly. STH (62) gets reverted by the same token. (66) (das) Reinigen (λγ) Xr [CAUSE (, )]

On the verbal nature of certain nominal entities

4.

87

Conclusion

Nominalizations, as being derived or converted from verbs, inherit articulateness features, such as thematic structure and event structure, from their base. Different morphological construction types are particulate] to different degrees, t/ng-derivations are less articulate than the corresponding nominal-infinitives. Both types of nominalizations are sensitive to the lexical aspect (aktionsart) of their respective verbal counterparts and transparent for the thematic roles of their arguments. The relevant findings are summarized in (67). (67) Patterns and Principles of Articulateness in Nominalizations Situation-Type-Principles al. ί/ng-derivations preserve the situation type of the corresponding verb. a2. Infinitive conversions always denote processes and thus convert the situation type of the base verb, when applied to accomplishment or achievement bases) Thematic-Linking-Principles b l . Nominalizations link only one argument to an NP-internal argument position b2. Infinitive conversions and wng-derivations of accomplishments and achievements give priority to the lowest argument of the inchoative (DEVELOP/BECOME) component. b3. f/ng-derivations of activities give equal priority to the thematic arguments of the process component (DO-component). There is an ongoing debate about functional as opposed to formal accounts for part of speech distinctions. Functional accounts (e.g. Croft 1991) define lexical categories in terms of ontological distinctions, such as the opposition between events (denoted by verbs) and objects (denoted by nouns) Formal accounts define lexical categories in terms of grammatical features, such as tense marking (for verbs) and case marking (for nouns). Purely formal approaches neglecting the sortal meaning of event nominalizations offer no explanation for the fact that nominalizations are sensitive to event structure properties. Functional approaches recurring exclusively to ontological concepts cannot explain why event nominalizations of different morphological types behave differently with respect to the thematic interpretation of their arguments. Wunderlich's approach may help to overcome the shortcomings of either account. On the one hand, his [±art] features can be interpreted as having a root in ontology: events prototypically involve participants and the expressions referring to them are therefore defined for arguments representing them. The ontological root, on the other hand, doesn't define the grammatical category exhaustively. The properties making up the [±art] bundle are essentially formal in nature. A given noun may instantiate each of the [-art] features (defining nouns) and none of the [+art] features (defining verbs) and thus perfectly match the nominal prototype. But there are expressions instantiating some of the [-art]

88

Veronika Ehrich

properties and some of the [+art] properties at the same time. Nominalizations of different types are a case in point, they deviate from the prototype to a higher or lower degree. Category membership is thus a matter of weight in terms of the [±art] features instantiated by a given expression. Wunderlich (1996) develops the relevant weighing measures. A grammar of nominalizations without any measures of this kind would remain incomplete.

References Bierwisch, Manfred (1988). On the Grammar of Local Prepositions. In Syntax, Semantik und Lexikon, Manfred Bierwisch, Wolfgang Mötsch, and Ilse Zimmermann (eds.), 1-65. Berlin: Akademie-Verlag. —(1989). Event Nominalizations: Proposals and Problems. In Grammatische Studien 194, 1-73. Berlin: Akademie der Wissenschaften der DDR. Chomsky, Noam (1970). Remarks on Nominalizations. In Readings in English Transformational Grammar, Roderick Jacobs and Peter Rosenbaum (eds.), 184-221. Waltham, Mass.: Ginn and Company. Croft, William (1991). Syntactic Categories and Grammatical Relations. Chicago: University Press. Davidson, Donald (1967). The Logical Form of Action Sentences. In The Logic of Decision and Action, Nicholas Rescher (ed.), 81-95. Pittsburgh: University Press. Demske, Ulrike (2000). Zur Geschichte der wwg-Nominalisierung im Deutschen: Ein Wandel morphologischer Produktivität. Beiträge zur Geschichte der deutschen Sprache und Literatur 122, 365—411. Dowty, David (1979). Word Meaning and Montague Grammar. The Semantics of Verbs and Times in Generative Semantics and in Montague's PTQ. Dordrecht: Reidel. Ehrich, Veronika (1991). Nominalisierungen. In Semantik. Ein Handbuch der internationalen Forschung, Arnim von Stechow and Dieter Wunderlich (eds.), 441-458. Berlin, New York: de Gruyter. —(2002). The Thematic Interpretation of Plural Nominalizations. To appear in The Syntax-Semantics-Interface. Linguistic Structures and Processes, Holden Härtl and Susan Olsen (eds.). —; and Irene Rapp (2000). Sortale Bedeutung und Argumentstruktur: img-Nominalisierungen im Deutschen. Zeitschrift für Sprachwissenschaft 19, 245-303. —; and Irene Rapp (2002). Nominalizations and Temporal Prepositions. To appear in ZAS papers in Linguistics. Grimshaw, Jane (1990). Argument Structure. Cambridge, Mass.: MIT Press. Herweg, Michael (1989). Ansätze zu einer Beschreibung topologischer Präpositionen. In Raumkonzepte in Verstehensprozessen. Interdisziplinäre Beiträge zu Sprache und Raum, Christopher Habel, Michael Herweg, and Klaus Rehkämper (eds.), 99-127. Tübingen: Niemeyer.

On the verbal nature of certain nominal entities

89

Hamm, Fritz (1999). Modelltheoretische Untersuchungen zur Semantik von Nominalisierungen. Habilitation thesis, University of Tübingen. Höhle, Tilman N. (1978). Lexikalistische Syntax. Die Aktiv-Passiv-Relation und andere Infinitkonstruktionen im Deutschen. Tübingen: Niemeyer. Klein, Wolfgang (1990). Überall und nirgendwo. Subjektive und objektive Momente in der Raumreferenz. Zeitschrift für Literaturwissenschaft und Linguistik 78, 9-42. Löbel, Elisabeth (1990). Apposition und das Problem der Kasuszuweisung und Adjazenzbedingung in der Nominalphrase des Deutschen. In Strukturen und Merkmale syntaktischer Kategorien, Gisbert Fanselow and Sascha. W. Felix (eds.), 1-32. Tübingen: Narr. Montague, Richard (1974). On the Nature of Certain Philosophical Entities. In Formal Philosophy. Selected Papers of Richard Montague. Edited and with an introduction by Richmond H. Thomason, 149-187. New Haven: Yale University Press. Olsen, Susan (1991). Die deutsche Nominalphrase als "Determinansphrase". In DET, COMP und INFL. Zur Syntax funktionaler Kategorien und grammatischer Funktionen, Susan Olsen and Gisbert Fanselow (eds.), 35-56. Tübingen: Niemeyer. Pustejovsky, James (1991). The Syntax of Event Structure. Cognition 41, 47-81. —(1995). The Generative Lexicon. Cambridge/Mass.: ΜΓΓ Press. Rapp, Irene (1997). Partizipien und semantische Struktur. Zu passivischen Konstruktionen mit dem 3. Status. Tübingen: Stauffenburg. Reichenbach, Hans (1947). Elements of Symbolic Logic. New York: The Free Press. Roeper, Thomas (1993). The Explicit Syntax in the Lexicon: The Representation of Nominalizations. In Semantics and the Lexicon, James Pustejovsky (ed.), 185-220. Dordrecht: Kluwer. Toman, Jindrich (1983). Wortsyntax: Eine Diskussion ausgewählter Probleme deutscher Wortbildung. Tübingen: Niemeyer. Vater, Heinz (1981). Valenz. In Kasusgrammatik und Fremdsprachenunterricht, Günther Radden and Rene Dirven (eds.), 217-235. Trier: Wissenschaftlicher Verlag. Vendler, Zeno (1967a). Facts and Events. In Linguistics in Philosophy, Zeno Vendler (ed.), 122-146. Ithaca: Cornell University Press. —(1967b). Verbs and Times. In Linguistics in Philosophy, Zeno Vendler (ed.), 97121. Ithaca: Cornell University Press. Wunderlich, Dieter (1971). Warum die Darstellung von Nominalisierungen problematisch bleibt. In Probleme und Fortschritte der Transformationsgrammatik. Referate des 4. Linguistischen Kolloquiums, Dieter Wunderlich (ed.), 189-218. München: Hueber. —(1990). Ort und Ortswechsel. Zeitschrift für Literaturwissenschaft und Linguistik 78, 43-58. —(1996). Lexical Categories. Theoretical Linguistics 22, 1-48. —(1997). CAUSE and the Structure of Verbs. Linguistic Inquiry 27, 27-68. Zucchi, Alessandro (1993). The Language of Propositions and Events. Issues in the Syntax and the Semantics of Nominalization. Dordrecht: Kluwer. Zybatow, Tatjana (2000). Grammatische Determinatoren von Zeit- und Sachverhaltsverlauf im Deutschen. Doctoral Dissertation, University of Leipzig.

Strong stems in the German mental lexicon: Evidence from child language acquisition and adult processing Harald Clahsen, Peter Priifert, Sonja Eisenbeiss, and Joana Cholin

1.

Introduction*

This study addresses the question of how marked stem forms are represented in the mental lexicon focusing on the marked stems of the so-called strong verbs of German. Results from error analyses of children's spontaneous speech and elicited production experiments with children and adults will be presented. In addition, we will briefly summarize previous findings from word-recognition experiments on marked stem forms. The findings from these studies suggest that inflected word forms that consist of regular affixes and strong stems have decomposed representations and that strong stems constitute subnodes of hierarchically structured entries with underspecified grammatical feature content. Consider inflected word forms such as those in (1): (1)

(ich) werf-e

(du) wirf-st

(sie) warf-en

'(I) throw.PRES-lSG'

'(you) throw.PRES-2SG'

'(they) throw.PRET-3PL'

Morphologists have posited (different kinds of) morphological rules and representations to analyze such forms. The forms in (1), for example, are derivable from combinatorial rules that combine stems or roots with regular affixes. Affixes (or exponents) may be represented in inflectional paradigms defined by their morpho-syntactic features. The relationships between the various stem alternants in (1) may either be captured by (morphologically conditioned) phonological rules (see e.g. Beedham 1994), or they may be directly encoded in the lexicon, through lexical redundancy rules (Chomsky 1970, Jackendoff 1975), default inheritance representations with underspecified entries (Corbett & Fraser 1993, Wunderlich 1996), or equivalent mechanisms. * The research reported in this paper is supported by the Deutsche Forschungsgemeinschaft (DFG) 'German Research Council' (grant SFB 282/C7). We thank the members of our research group, Ingrid Sonnenstuhl, Meike Hadler, Helga Weyens, Kerstin Mauth, and Axel Huth for assistance in designing experimental materials and for many useful ideas. We are also grateful to Wolfgang Dressier, Martin Durrell, Claudia Felser, Dirk Janssen, Martina Penke, Monika Rothweiler, and the editors of this book for comments and helpful suggestions.

92

Harald Clahsen et al.

Psycholinguists have asked the additional question of which of these morphological operations and representations are employed by the speaker/hearer in language comprehension and production, and by children in language acquisition. After all, one possibility would be that even though affixation, paradigms, structured lexical entries and the like provide useful tools for the description of patterns among inflected word forms, they do not play any role for mental representation and acquisition. This latter view is indeed held by many psycholinguists working within connectionist approaches to language (see Rumelhart & McClelland 1986, Elman et al. 1996, Bybee 1995, among many others). In these accounts, inflected words are all stored in associative networks which through repeated exposure to multiple sets of inflected words will create connections among the various forms. In this way, the network forms patterns that range over sets of connections, and morphological rules and representations come out as secondary, derivable from associations between words. In English, for example, there are thousands of past tense forms ending in -ed, and in a connectionist network this leads to the creation of a strong pattern that makes the network behave as if it had a separate -ed rule, even though the network itself does not utilize any kind of morphological rules; see Marcus (2001) for a critical discussion of these models. On the other hand, many psycholinguists have argued that morphological structure and representation is important for understanding the acquisition and processing of inflected and derived word forms. Several psycholinguistic studies have focused, for example, on the contrast between regular and irregular inflection (see Pinker 1999, Clahsen 1999), and there is now evidence from different languages for a dualmechanism account according to which regulars are segmented or parsed into their constituent morphemes while irregular forms are stored as whole forms in memory (see also Marslen-Wilson & Tyler 1998). However, inflected word forms do not always fall neatly into one of these two clusters. Consider the forms in (1) again and note that the person and number endings are all highly regular; they are easily segmentable and can be attached to almost any verb. On the other hand, the forms in (1) differ in terms of their stems: werf- is the unmarked (or so-called 'weak') stem/root form, which is identical to that of the infinitive (werf-en), whereas wirf- and warf- are marked (alternated) stems which are different from that of the infinitive. In the present study, we will examine how different stem forms of the same lexeme are represented in the mental lexicon. One question we will address is whether strong stems such as wirf- and warf- are rulegoverned, e.g. derived from werf-, or whether they are lexically stored. Another question concerns the role of morphological features for the representation of stem forms. The phonological string wirf-, for example, appears as the stem in both imperatives and 2/3SG forms. Does that mean that there are two stem forms wirf-, which are distinguished in terms of morphological features, or is one stem form based on the other and how are these relationships represented? Psycholinguistic evidence from different sources will be presented that bears on these questions. Before turning to the empirical results, we will provide a brief description of strong verb inflection in German.

93

Strong Stems in the German Mental Lexicon

2.

Strong verb inflection in German

There are about 160 simplex verbs in German that belong to the strong class. These verbs have marked stems in present tense, preterite, or participle forms. Most of them (= 155) fall into three minor classes, illustrated in (2); see Wunderlich & Fabri (1995) for a detailed classification. While A-B-A and A-B-C verbs have differently marked stem forms in preterite and participle forms, A-B-B verbs exhibit the same vowel change for both preterites and participles. Moreover, a large number of strong verbs have subjunctive forms with umlauted preterite stems, as shown in (2) for gab-lgäb'gave' ,flog-/flög- 'flew' and sang-lsäng- 'sang'. There is also a small number of strong verbs in which 2SG and 3SG present tense forms as well as imperatives have fronted vowels (e.g. werfen vs. er wirft 'throw.iNF' vs. 'he throws'). A-B-A A-B-C A-B-B

Infinitive geben singen fliegen

Preterite gabsangflog-

Subjunctive gäbsängflög-

Participle gegeben gesungen geflogen

'give' 'sing' 'fly'

Except for suppletive forms such as those of sein 'be' and the forms of modals like dürfen 'be allowed to', the marked stems of strong verbs are regularly inflected for person and number in the same way as the unmarked stems of weak verbs. The linguistic representation of stem alternations is controversial. One suggestion is to derive stem alternants from a single underlying form by (morphologically conditioned) phonological rules. Several attempts have been made to demonstrate this in respect to strong verb formation in German, see Beedham (1994, 1995/1996), Bittner (1996), Barbour (1982), among others. Beedham, for example, observes that particular sequences of onsets and rhymes of root syllables coincide with whether or not a verb has strong stem alternants in German. For instance, verbs whose root syllables have the rhyme -ing- (e.g. singen 'sing') tend to be strong, whilst verbs with root syllables that have the onset [va] (e.g. wagen 'dare') are predominantly weak. Beedham posits corresponding phonological rules that specify 50 relevant sequences of onsets and/or rhymes to establish the phonological sequences which, in his view, distinguish weak from strong verb roots. One problem of any account that attempts to handle stem alternations by phonological rules is the large number of exceptions and counter-examples; see Durrell (1980, 2001), Wiese (1996). An alternative solution is to represent stem alternants in the lexicon. One suggestion has been to directly list stem forms such as wirf-, warf-, wiirf- etc. in the permanent lexicon in much the same way in which suppletion of the go - went type is listed (see e.g. Lieber 1981). This, however, does not capture the fact that a number of the vowel patterns are recurrent and seem to form family resemblance clusters. Bybee & Newman (1994) have therefore argued that stem changes (as well as inflectional affixes) are represented in the lexicon via associative pattern networks or schémas; see Kopeke (1998) for an attempt to classify some of the strong verbs of German in terms of sehe-

94

Harald Clahsen et cd.

mas. A third variant of a lexicalist analysis of stem alternants comes from Wunderlich (1996). In this account, the morphological relationships between the various stem variants of strong verbs are represented in terms of non-monotonic default inheritance hierarchies in which stem variants constitute subnodes of hierarchically structured lexical entries. Each subnode is defined in terms of a phonological string and a morphological feature. Consider, for illustration, the lexical entry for the German verb werfen 'throw' from Wunderlich (1996: 96): (3)

[νειΑ+ν

[..i..]-i

[-a.-l+pRET

[·—]+iMP

[..y..X] + suBJ

["Ο..ηΐ+PART

Each node in a structured lexical entry represents a pair (), e.g. [..i..] vs. [..a..] and [-1] vs. [+PRET] for wirf- vs. warf-, and each subnode inherits all information from its mother, except for the features it replaces or adds; for example, the subnode [..a..]+PRET inherits the onset w-, the coda -rf, and the categorial feature [+V] from the higher node. The vowels in the subnodes replace the corresponding elements of the higher nodes, /n/ in the participle and X (for schwa) in the subjunctive forms represent additional segments. As illustrated in (3), most stem variants have impoverished entries (to avoid redundancies), and the various stem forms are hierarchically structured. The base or default stem form is werf-, the form that occurs in most present tense forms and the infinitive. The other stem variants occur under specific circumstances, e.g. wirf- for 2SG and 3SG present tense forms and in imperatives, warf- in preterite forms, (ge)worfen in participles and wiirf- in subjunctives. The fact that the [..i..] stem does not occur in plural forms follows from a general constraint on affix order: no affix can be attached if it expresses a category that is lower ranked than any of those already instantiated (Wunderlich & Fabri 1995: 249). Thus, [+pl] cannot be attached to a form that is already specified for PERSON, because PERSON is higher in the hierarchy of functional categories than NUMBER. Structured lexical entries such as those in (3) may form lexical templates, based on shared subnodes and shared structure. The subnodes of the entry in (3), for example, are shared by several other strong verbs (sterben 'die', verderben 'spoil', helfen 'help'). In this way, structured lexical entries are maximally underspecified, and, at the same time, account for the overall similarity of stem alternants to their base forms and for the family resemblance structure of strong verbs. In the following two sections, we will determine the generalization properties of different stem forms by examining the kinds of stem overgeneralization errors produced by children learning German as their native language and by testing experimentally which stem forms adult native speakers apply to nonce words. Both studies provide

Strong Stems in the German Mental Lexicon

95

support for a lexicalist treatment of strong verb stems along the lines of Wunderlich (1996).

3.

Stem formation errors in German child language

One important psycholinguistic source for determining the generalization properties of a linguistic pattern or rule comes from overgeneralization errors produced by children. Consider, for example, past tense formation in English child language where only one type of inflectional error was found to be productive: overapplications of the regular past tense affix -ed to irregular stems (*go-ed, etc.). Overapplications of irregular patterns (*brang, *wope), on the other hand, were reported to be extremely rare (see e.g. Marcus et al. 1992, Xu & Pinker 1995). It was also found that irregular verbs are sensitive to frequency and similarity: children make overgeneralization errors more often with low-frequency irregular verbs, and they produce fewer of them with irregular verbs that fall into families with more numerous and higher-frequency members. The same pattern of errors was found in two inflectional systems of German, past participles and noun plurals (Clahsen et al. 1992, Clahsen & Rothweiler 1993, Weyerts & Clahsen 1994, Bartke et al. 1996). These findings have been taken to indicate that regular and irregular inflectional processes are dissociated in children's grammars in basically the same way as in the adult grammar. From a dual-mechanism perspective, the observed differences in the generalization properties of regular and irregular inflection are claimed to follow from different representations, rules (or equivalent operations) with unrestricted productivity for regular inflection, and memory-based representations that only allow for restricted (similarity-based) generalizations for irregulars. In this account, past-tense overregularization errors such as *bring-ed are due to the child's applying a regular (-ed) rule in cases in which the lexical entry for the irregular word form brought is not available, and they disappear once the child can reliably retrieve the correct irregular word form. For the present paper, we investigated stem forms of strong verbs in longitudinal data from 7 children and in an additional elicited production experiment with 26 children. We examined two families of strong verbs which both require stem changes in (particular kinds of) present tense forms, namely verbs of the geben type ('give'), which have an -i- stem, and verbs of the schlafen type ('sleep'), which have an umlauted stem, both occurring in 2/3SG present tense forms (e.g. gib-st 'give-2SG.PRES', schläf-st 'sleep-2SG.PRES'); for verbs of the geben type, the -i- stem also occurs in imperatives. Note that all participants of the studies reported in the present article come from Northwest Germany and that in these varieties, non-umlauted present-tense forms are ungrammatical; in spoken Southern varieties of German, however, such forms (e.g. er lauft 'he runs', er schlaft 'he sleeps') are quite common. Other families of strong verbs require stem changes in the past participle, the preterite and/or the subjunctive. Participle formation in German children was examined in two earlier studies (Clahsen &

96

Harald Clahsen et al.

Rothweiler 1993, Weyerts & Clahsen 1994), and the relevant results on stem formation will be summarized below. Simple preterite and subjunctive forms of main lexical verbs, on the other hand, are hardly ever used in spoken German, which is why instances of such verb forms are basically absent, or at least extremely rare, in the speech of young children. Method The longitudinal data we have investigated consist of 73 samples of spontaneous speech from7 children covering the age period of 1 ; 11 to 3;8 (see table 1). The corpora were collected in the LEXLERN project at the University of Düsseldorf (see Clahsen et al. 1993, 1994, 1996). In these data, we extracted by hand all correct and incorrect stem forms of strong verbs of the geben type produced by the children in contexts that require an -ι- stem in the adult language and of the schlafen type in contexts that require an umlauted stem in the adult language. We also searched for cases in which the children (incorrectly) produced -i- or umlauted stems for weak verbs of German. Child Annelie Hannah Katrin Leonie Philipp Sabrina Svenja Totals

Age 2;4-2;9 2;0 - 2;7 2;1 - 2 ; 6 1;11 - 2;11 3;1 - 3 ; 8 1;11 - 2;2 2;9-3;3 1 ; 11 - 3;8

Number of recordings 6 8 15 15 8 6 15 73

Table 1: Data To investigate stem formation in older children, we performed an auditory elicited production task with 26 children covering the age range of 6;2 to 10;5. The materials were recorded and presented to the children from a laptop computer; their responses were recorded on a digital tape. The children first listened to a sentence containing an infinitive form of a verb as the final word, e.g. sehen in (4a). They then heard a second sentence in which the appropriate finite verb form of the previously presented verb was replaced by a beep, and they were asked to produce the correct verb form, e.g. sieht in (4b). In all cases, the finite verb to be filled in was the 3SG present tense. (4)

a.

Martin will unbedingt den neuen Pokemon-Film sehen. 'M. definitely wants to see the new pokemon movie', b. Also gibt ihm seine Mutter Geld und Martin beep den Film. ^ sieht 'Hence his mother gives him some money, and Martin the movie'

There were 49 pairs of sentences such as those in (4). Five pairs preceded the actual experiment to familiarize children with the task. Of the remaining 44 sentence pairs, 22

97

Strong Stems in the German Mental Lexicon

contained the experimental items, and 22 were used as fillers. The filler sentences contained strong or weak verbs (e.g. schwimmen 'swim', schenken 'give (presents)', fangen 'catch', etc.), and they were syntactically similar to the experimental sentences. Sentences containing experimental items were pseudo-randomized with sentences containing fillers. The experimental items were 12 strong verbs of the geben type, and 10 strong verbs of the schlafen type. To examine effects of frequency, each of these two types of strong verbs was further divided into two conditions, a high-frequency and a low-frequency one, each according to the frequencies of the marked stems (i.e. the -ior the umlauted stem) in the CELEX database of spoken German (Baayen et al. 1993). For verbs with -i- stems, the high-frequency condition consisted of 6 verbs with a mean (-/-) stem frequency of 37 per million (geben 'give', sehen 'see', nehmen 'take', sprechen 'speak', lesen 'read', essen 'eat'), while in the low-frequency condition there were 6 verbs with a mean (-/-) stem frequency of 0.1 per million (stehlen 'steal', befehlen 'order', werben 'advertise', stechen 'sting', schmelzen 'melt', flechten 'weave'); for verbs that require umlauted stems, the high-frequency condition consisted of 5 verbs with a mean (umlaut) stem frequency of 6.1 per million (halten 'hold\ fahren 'drive', tragen 'carry', laufen 'run\ fangen 'catch'), while in the low-frequency condition there were 5 verbs with a mean (umlaut) stem frequency of 0.1 per million (laden 'load', raten 'guess', blasen 'blow', graben 'dig', braten 'roast'). Results Consider first the results from the longitudinal data of the younger children. Here we found two types of error. The first one is overapplication of the unmarked (nonalternated) stem in cases in which strong stem forms are required in the adult language. Some examples of such stem overregularizations are given in (5). (5) a. b.

die lauft (= läuft) du lest (= liest)

'she runs' Katrin (2;4) 'you read' Svenja (3;2)

There are 84 stem errors of this kind in the longitudinal data against 122 strong stem forms with correct -i- stems for verbs of the geben type or correct umlauted stems for verbs of the schlafen type. Note that (as illustrated by the examples in (5)) most stem errors co-occur with the correct person and number suffix; there are only 15 cases that might be analysed as bare uninflected forms or infinitives, e.g. der fang 'he catch' (Svenja, 3;2) or du nehmen 'you take' (Sabrina, 2;0). Not counting these 15 cases, we are left with 47 stem overregularizations for verb forms that are clearly finite (in terms of their suffixes) and in which a strong stem form is required in the adult language. The second type of error is incorrect occurrences of marked -i- or umlauted forms such as those illustrated in (6): (6)

a.

alle fäll da ranter 'everybody fall down there' b. ich gib dir das Ί give you that'

(correct: fall-en)

Katrin (2;5)

(correct: geb(e))

Philipp (3;4)

98

Harald Clahsen et al. c.

ich sieh Ί see'

(correct: seh(e))

Svenja(3;l)

This error type is much less common than overapplications of the unmarked stems. There were 11 cases such as those in (6) in the entire longitudinal data set. Interestingly, even though these 11 forms are incorrect in the syntactic contexts in which the children use them, the stem forms produced by the children are not incorrect. For example, the forms fäll-, gib- and sieh- are the correct adult stem forms for 2SG and 3SG present tense indicative forms of these verbs, i .e. fallen, geben and sehen. The errors in the children's utterances arise because the children apply a marked stem form to a LSG form and (in one case) to a 3PL form. That is, a particular morpho-syntactic constraint on the use of marked stem forms, namely that -i- stems and umlauted stems are restricted to 2/3SG (and to imperatives) is violated. Note, however, that there was no single instance of an irregularization error, i.e. a case in which a marked (alternated) stem would be overapplied to a weak verb. Thus, the distribution of stem errors in the longitudinal data shows a clear contrast between unmarked and strong stems; the former are frequently overregularized, in 26,6% (i.e. 47 out of 177 cases), whereas stem irregularizations (i.e. strong stems overapplied to weak verbs) are non-existent. Another finding concerns the development of the stem overregularization errors over time. To investigate developmental changes, the various recordings were assigned to stages of development defined in terms of the mean length of utterance (MLU); see Brown (1973). Table 2 shows a breakdown of the stem overregularization errors against the number of correct marked stem forms across these stages. Stages I (MLU < 1.75) Π (1.75 < MLU 3.5) Totals

Number of stem errors 1 18 5 23 47

Number of correct stems 11 44 17 58 130

Error percentages 8.3% 29,0% 22,7% 28,4% 26,6%

Table 2: Development of stem overregularization errors Table 2 shows that the period from stage Π onwards in which stem overregularization errors frequently occur is preceded by a period (= Stage I) in which almost all of the required marked stem forms are correctly produced by the children. Moreover we can see from table 2 that in the age period represented in the longitudinal data, there are no signs of stem overregularization errors disappearing from the speech of the children or decreasing over time. Consider now the experimental results from the older children. We found that the children did not show any difficulty performing the task: 97% of their responses were appropriate in that they produced a finite form of the prompted lexeme, e.g. a finite form of sehen in (4). There are, however, many stem errors, even in these older chil-

99

Strong Stems in the German Mental Lexicon

dren, all of which were overregularizations of the unmarked (non-alternated) stem in cases in which strong stem forms are required in the adult language, i.e. errors such as those in (5) above. Table 3 presents a breakdown for the experimental conditions. Required stem form and stem frequency -i- / high frequency -i- / low frequency -ä- / high frequency -ä- / low frequency Totals

Number of stem errors 19 89 9 51 168

Number of correct Error percentages stems 12.2% 136 59 37.3% 7.0% 118 74 40.8% 387 30.2%

Table 3: Elicited stems forms (Target: 3SG present tense) Table 3 shows a clear frequency effect in the stem overregularizations for both classes of strong verbs: low-frequency stems elicit significantly more stem errors than highfrequency ones (i(25) = 10.399, ¿X.001). Finally, in order to examine developmental changes, we plotted in Figure 1 the percentages of overregularizations against age. Figure 1 shows that overall overregularization errors decrease with age, but that the 8-to-9-year olds produce more errors than the 7-to-8-year olds. Consequently, a regression analysis revealed that even though the correlation is not strong, overall stem overregularizations are significantly linked to age

Age (in years) Figure 1 : Stem overregularizations in relation to age

100

Harald Clahsen et al.

Discussion The results on children's stem formations obtained in the present study are similar to those of earlier studies on participle forms of strong verbs (Clahsen & Rothweiler 1993, Weyerts & Clahsen 1994). In these studies, we examined longitudinal data from 9 children between the ages of 1;4 and 3;9 and elicited production data from 70 children between ages 3;6 and 8; 10; there were 222 stem errors in participle forms, more than 90% of which were cases of unmarked (non-alternated) stems in cases in which strong stem forms are required in the adult language, e.g. * gehelft instead of the correct form geholfen 'helped'. We also found that the period in which children produce stem overregularizations is preceded by a stage without stem errors in their participle forms, and that stem overregularizations do not disappear during the age period under study; in the elicited data, even the 7- to 8-year olds produced such errors. These results are parallel to what we found for the two families of strong verbs examined in the present study. One way of accounting for the kinds of stem errors produced by children is in terms of structured lexical entries such as those proposed by Wunderlich (1996) for strong verbs of German (see (3) above). Recall that in such representations, the various stem alternants are not listed separately and completely, but that strong stems may have impoverished (i.e. underspecified) entries and that when these items are used, their base entry is filled in to obtain the full interpretation and form. Given this format, stem errors arise when subnode information containing the correct pairing of phonological strings and morphological feature values for a strong stem form is not available or not accessible to children. In such cases, they fall back on the base entry, i.e. the highest form of a structured lexical entry, producing errors such as *helf-st instead of the correct hilf-st. In this way, the unmarked base stem serves as a default form in circumstances in which the required specific forms are not retrieved. The results also suggest that stem alternants are stored in the children's lexicon. There are two indications for that in the current data set, (i) the frequency effect for stem overregularizations, and (ii) the overall decrease of the number of stem errors with age found in the child experiment. Memory storage and retrieval are dependent on frequency of exposure, and hence low-frequency forms are likely to yield more stem errors than high-frequency ones. Moreover when children get older, memory traces for the correct irregulars are likely to become stronger and more reliable, and hence the observed decrease in the development of stem errors. We will further discuss these results in the next section after presenting an experiment on stem formation in adult native speakers.

Strong Stems in the German Mental Lexicon

4.

101

Generalizing stem forms in adult native speakers

Another method for determining the generalization properties of inflectional patterns or rules are elicited production experiments with nonce words. In such experiments, participants are presented with one or two forms of a nonce word, they are then asked to repeat the nonce form, and finally they have to produce a different inflectional form of that nonce word which they have not seen before. Such experiments have been used to study the generalizability of different inflectional phenomena (see e.g. Bybee & Pardo 1981, Clahsen 1997). For the present study, we examined strong verbs such as werfen 'throw' which have an -e- stem as the unmarked base and an -i- stem for imperatives and 2/3SG present tense forms (see (3) above). Two observations are relevant for the distribution of these stem forms. The first one is that strong verbs with an -e- stem in the infinitive and an -istem in the imperative always exhibit the -i- stem in 2/3SG present tense forms; compare geben - gib - gibst (*gebst), 'to give - give.IMP - give.2SG'. The second observation is that most verbs with -e- stems in the infinitive and an -i- stem in 2/3SG present tense forms also have an -i- stem in imperative forms; there are, however, some exceptions, e.g. werden - werd(e) (*wird) - wirst 'to become - become.lMP - become.2SG'. These two generalizations can be captured in different ways. One possibility would be to postulate rules which derive -/- forms from -e- stems, for both imperative and 2/3SG present tense forms of strong verbs. As verbs like werden have an irregular -istem in 2/3SG present tense forms, but not in imperatives, the rule generating imperatives will have more exceptions than the rule for the formation of 2/3SG present tense forms. Such exceptions would then block the application of the rule. A second possibility would be to assume that both -e- and -/- stems are stored separately and that the patterns mentioned above follow from associations between them. These associations would have to capture the fact that since all verbs with an -i- stem in the imperative also have an -i- stem for 2/3SG present tense forms, but not vice versa. This might be achieved by positing strong associative links from -i- imperative forms to -/- 2/3SG present tense forms, but relatively weaker links for the reverse connection. Consequently, we would expect that when speakers are presented with an imperative form containing an -i- stem they are more likely to produce an -i- stem for 2/3SG present tense forms than vice versa. A third possibility would be to account for the distribution of irregular stem forms in terms of lexical templates (Wunderlich 1996). Consider, for example, the lexical template in (7) which is derived from the structured lexical entries of strong verbs such as werfen 'throw', sterben 'die', verderben 'spoil'. Here, the subnodes for imperatives and 2/3SG present tense forms are asymmetrically connected to each other, such that the imperative node fully inherits its stem form from the 2/3SG present tense forms, whereas the onsets and the codas of the stems of 2/3SG present tense forms are inherited from the unmarked base form and the -i- stem vowel change is introduced at this level.

102

Harald Clahsen et al.

(7)

[.·ε..]+ν

[..ι..]-ι

[..a..]+pRET

[••••]+IMP

[..y..X]+suBj

[..o..n]+PART

From this account, we would expect to find differences in generalizability between -eand -i- stems. The former represents the unmarked base stems and should therefore be used under default circumstances, i.e. when it is not blocked by a more specific entry; -i- stems should be generalized more restrictively, i.e. only to strong verbs which have subnodes with marked stem forms. Moreover, the hierarchical structure in (7) is meant to capture the differences between 2/3SG present tense forms and imperatives with respect to -i- stems. For example, lower subnodes may be lost in the course of language change for some verbs, yielding irregular verbs that have an -e- stem in the infinitive and an -i- stem in 2/3SG present tense forms, but no -i- in imperative forms (e.g. werden - werd(e) (*wird) - wirst 'to become - become.lMP - become.2SG')· In contrast to that, an -i- stem in an imperative form presupposes the existence of an -i- stem in the 2/3SG present tense. Thus, strong verbs with an -e- stem in the infinitive and an -i- stem in the imperative always exhibit the -i- stem in 2/3SG present tense forms; compare geben - gib - gibst (*gebst), 'to give - give.lMP - give.2SG'. This is accounted for in (7) by the fact that -i- imperative stems are inherited from the higher node representing 2/3SG present tense forms. The view that the formation of stem forms is based on default inheritance representations such as (7), in which the imperative stem inherits its vowel from the 2/3SG present tense stem but not vice versa, leads to the following experimental predictions. When speakers are presented with an irregular 2/3SG present tense -i- stem of a nonce verb, inheritance makes it likely for them to use the same stem for the formation of the corresponding imperative. By contrast, a 2/3SG present tense form cannot inherit the irregular -/- stem from an imperative form. Hence, we would expect that when speakers are presented with an irregular -i- imperative form of a nonce verb, they are less likely to use this stem for the formation of the corresponding 2/3SG present tense form. We tested these predictions in an elicited production experiment. Method 36 adult native speakers of German (mean age: 33) participated in a paper and pencil task in which they were asked to complete sentences in a booklet by filling in blanks. The booklet consisted of 60 different short stories containing phonotactically wellformed nonce verbs and nouns. The crucial experimental items were the nonce verbs; nonce nouns were added to reduce uncontrolled semantic associations. The experiment consisted of three steps. Subjects were first given each nonce verb in the infinitive and the imperative or 2/3SG present tense form ("step 1"). They were then asked to use the

Strong Stems in the German Mental Lexicon

103

previously presented forms to fill two blanks, one for the infinitive and one for imperative or 2/3SG present tense form ("step 2"). Finally, they had to fill in a third blank, this time by providing a form of the nonce verb that was not presented to them before ("step 3"); see (8) for illustration: (8)

Step 1: presentation of nonce verb in infinitive and imperative or 2/3SG present tense a. Peter hat meistens keine Lust, den Prüm zu quelmen. 'Most of the time, Peter is not keen to quelm the prum.' b. Wenn er ihn doch einmal quelmt/quilmt, hat er noch drei Tage später schlechte Laune. 'If he quelms/quilms it anyway, he is still in a bad mood even three days later.' Step 2: repetition of verb forms presented c. Wenn es wieder an der Zeit ist, den Prum zu 4 quelmen, 'If it is time to quelm the prum again' d. muß Susi dafür sorgen, daß Peter ihn quelmt/quilmt. 'Susi must take care that Peter quelms/quilms it.' Step 3: production of novel verb form e. Meistens muß sie Peter anschreien: "Verdammt noch mal ^ quelm/quüm endlich mal den Prum!" 'Most of the time, she has to yell at Peter: "Damn it, quelm/quilm the prum!'"

There were three sets of short stories. In one set of 20 stories, the nonce verbs were introduced in their infinitive and a 2/3SG present tense form. In step 2, participants had to supply both of these verb forms, and in the third (crucial) step, they had to fill in the imperative form of the nonce verb. In 1 0 of these stories, the 2/3SG present tense form had the unmarked base stem -e- (e.g. quelmen-quelmt); in the other 10 stories, the 2/3SG form was presented with the strong -i- stem (e.g. quelmen-quilmt). In the second set of 20 stories, participants were given the infinitive and an imperative form of a nonce verb, and they had to fill in the 2/3SG present tense form in step 3. In 1 0 of these stories, the imperative had the unmarked -e- stem (e.g. quelmen-quelm), in the remaining 10 stories the imperative was presented with the strong -i- stem (e.g. quelmenquilm). In addition to the 40 experimental items mentioned above, participants were given 20 other short stories in which they were presented with sentences containing infinitives and regular or strong preterite forms of nonce verbs; they were asked to repeat these forms and to produce the corresponding past subjunctive form. Filler items were added to obscure similarities between experimental items and to prevent subjects from developing any particular task-specific response strategies. All experimental and the filler items were pseudo-randomized.

104

Harald Clahsen et al.

Results For the following analysis, we only included the participants' responses of a given verb in step 3 if they had correctly reproduced the infinitive and the imperative or 2/3SG forms of that verb in step 2; this was the case in 94% of the participants' responses, i.e. in 1352 out of 1440 cases. Table 4 presents a breakdown of the relative percentages of the participants' responses in step 3 (imperatives or 2/3SG present tense responses), shown separately for verbs that were presented with an -e- stem in their imperative or 2/3SG forms in step 1, and for verbs that were presented with an -i- stem in their imperative or 2/3SG forms in step 1; absolute numbers of responses for each category are shown in brackets.

Imp. or 2/3SG

pre-

sented in step 1 with -e- stem with -ι- stem

89% (303) 15% (51)

Participants ' responses in step 3 imperative 2/3SG present tense other other other other -i-ivowel form vowel form 1% 9% 0% 88% 6% 0% 6% (32) (300) (21) (21) (0) (4) (0) 83% 0.3% 2% 22% 1% 71% 7% (280) (73) (235) (23) (6) (2) (1)

Table 4: Stem forms of imperative and 2/3SG present tense forms of nonce verbs The first two columns within each of the two response types in table 4 shows that in more than 90% of the newly created verb forms subjects made use of -e- or -i- stems. There were only three instances with correct inflectional endings, but with the stem vowel -a-; see the third column for the two response types. There was also a small number of cases in which 'other forms', e.g. past tense forms were produced. Table 4 also shows that participants preferred the vowel with which the novel verb was introduced to them in step 1. However, this preference was stronger for -e- than for -istems. In more than 88%/89% of all stories, the unmarked -e- stem presented in step 1 was also employed by the participants in their own word formations in step 3. In contrast, the strong -i- stems were used in 83% of the imperative contexts and 71% of the 2/3SG present tense contexts. Instead of maintaining the -i- stem from step 1, participants relied on the -e- stem (= 15%, 22%). Finally, we note that the -i- stem is more often used for imperatives (83%, 9%) than for 2/3sg. present tense forms (71%, 6%). The observed differences are also confirmed statistically. We determined the number of cases in which in step 3 participants made use of the stem form that was presented to them in step 1. For these responses, we carried out a repeated measures Analysis of Variance with two factors, Stem Type (-e- vs. -/-) and Verb Form (imperative vs. 2/3SG). This analysis revealed the following effects: first, a significant main effect of Stem Type which is due to the overall preference for -e- stems (Fi(l,35)=12.048, p —IAM:[+dorsal] (*/e/, */R/) b. GMCOR (GM Correspondence): [+masc,+gov] items in the same minimal residue have identical markers.

*DAM is a FCR relativized with respect to agreement markers. It states that all categories which are not [-obi] and [-P1] are incompatible with a dorsal agreement marker, i.e., with /R/ and /e/.28 GMCOR is a constraint that is of a different type from those discussed so far (the marker triggers CASE and AGR, the FCRs, and SONHIER). This constraint is reminiscent of correspondence constraints in optimality theory because it requires two categories α and β to have identical markers. The three new constraints are integrated into the overall ranking as follows. AGR outranks CASE (although this will only become relevant in subsection 4.2 below). *DAM is ranked higher than SONHIER. Finally, GMCOR must intervene in the domain of the SONHIER constraints: It is 30 ranked lower than *m, but ranked higher than *n. The following tableaux give illus27. More generally, the distinction between Case markers and agreement markers in the present approach parallels the standard distinction between strong and weak inflection. These latter notions play no independent role in the present analysis. 28. Note that [-P1] is an abbreviation for -i[-masc,-fem]. In this case, a straightforward disjunctive formulation of the constraint might arguably be easier to grasp: *DAM prohibits dorsal agreement markers on categories that are plural or oblique. 29. However, it is worth pointing out that GMCOR does not belong to any of the subclasses of correspondence constraints envisaged in McCarthy and Prince (1995). It is not an input/output correspondence constraint because (Case or agreement) markers are not yet part of the input; and it is not an output/output correspondence constraint because GMCOR requires only those markers to be identical in form that occur within the same output, not on some other output that has independently been optimized. The input/output vs. output/output correspondence distinction can be viewed as similar to the distinction between rules of exponence and rules of referral (see Stump 2001). Both output/output constraints and rules of referral may initially look attractive in inflectional morphology. However, they are extremely powerful means that the present analysis does not resort to (see also Wunderlich 1996:107-112). 30. In fact, *DAM only has to be ranked higher than *n; thus, it could (but, unlike, GMCOR, does not have to) have a ranking between *m and *n.

131

Remarks on nominal inflection in German

trative examples of how weak adjective declension arises, i.e., how agreement markers are distributed on adjectives. Since an agreement marker presupposes a Case marker in the same minimal residue, more structure than before must be taken into account. For present purposes, it suffices to consider the adjective together with a preceding determiner (most of the optimal outputs in what follows will have to violate CASE once, but these violations incurred by caseless Ns can safely be ignored throughout). Assuming, as before, that there are five a priori conceivable marker forms plus one form without a marker, and two relevant NP-intemal categories on which markers may occur, this gives us thirty-six relevant outputs to consider. Of these, I will generally confine myself to those candidates that have a chance to become optimal. Let us begin with a 31 NOM.FEM ([-mase,+fem,-obi,-gov]) specification on both D and A. Tableau T9 shows why /e/ is optimal both as a Case marker on dies, and as an agreement marker on gut in this context.32 T9: diese gute (Milch) [Input: /dies, gut/; NOM.F] AGR CASE *DAM

045: 051: 052: 053: 054: »s· 055: 056: 066:

dieser gute diese gutes diese gutem diese guten diese guter diese gute diese gut dies gut

*XCM

*S

*m

GMCOR

*n *R *e *!

* *

*!

*

*!

*

*! *!

* ** *

*! *!

Outputs like 0 4 5 that do not have /e/ as the Case marker for the determiner are illformed for the reasons discussed above (see Tl). Among the outputs 0 5 1 - 0 5 6 that have a Case marker /e/ on dies, the SONHIER constraints straightforwardly predict 0 5 5 to be optimal. Note that whereas 0 5 6 fatally violates AGR, 0 6 6 does not: AGR does not per se require an agreement marker on an adjective below a determiner; it does so only if the determiner is Case marked. Next consider an input with a GEN.MASC ([+masc,-fem,+obl,-gov]) specification. Here, a Case marker /s/ and an agreement marker /n/ emerge as optimal; see tableau T10. 31. Recall that I assume that it follows from syntactic agreement constraints that these features must be identical on DP and AP specifiers of N. These constraints are not to be confused with the constraint AGR introduced above: They govern the distribution of morpho-syntactic features in minimal residues; AGR (like CASE) is exclusively concerned with the morphological realization of a given morpho-syntactic specification. 32. Two remarks on notation. First, I systematically leave out *NCM in the tableaux of this subsection, and I group the remaining four FCRs for Case markers together under the label *XCM. Second, the output candidates are numbered according to a self-explanatory system in which "1" stands for /s/, "2" for Imi, "3" for /n/, "4" for ÍRJ, "5" for /e/, and "6" for the absence of a (Case or agreement) marker.

132

Gereon Müller

TIO: dieses guten (Tees) [Input: /dies, gut/; GEN.M] AGR CASE *DAM

Oil: 012: •s· 013: 014: 015: 016:

dieses dieses dieses dieses dieses dieses

gutes gutem guten guter gute gut

*XCM

*s *m GMCOR *n *R *e **i *

*!

*

*!

*

*!

*

* * *

*

*!

Any output without /s/ as the Case marker will invariably be suboptimal; see T5. This time, however, /e/ is blocked as an agreement marker because of *DAM, which prohibits dorsals in contexts that include GEN.MASC (see 015). Since the same goes for /R/ (see 014), /η/ is chosen as the optimal agreement marker by the SONHŒR constraints (see 0 1 3 vs. 012, O i l ) . So far, GMCOR has not been active in competitions. This changes when [+masc,+gov] specifications are considered. In particular, consider the competition in ACC.MASC ([+masc,-fem,-obl,+gov]) contexts, which is illustrated in tableau T i l . Til: diesen guten (Tee) [Input: /dies, gut/; ACC.M] AGR CASE *DAM

031: 032: «s· 033: 034: 035: 036: 044: 055:

diesen gutes diesen gutem diesen guten diesen guter diesen gute diesen gut dieser guter diese gute

*XCM

*S

*m GMCOR *n *R *e *

*! *!

*

* * **

*! *! *!

*!

*

*!

*

*

*

* *

** **

Without GMCOR, we would expect 0 3 5 to be optimal, because of the ranking *n » *e in SONHŒR (note that *DAM does not block 034, 035 here). But GMCOR requires marker identity for dies and gut (which 0 3 4 and 0 3 5 do not exhibit), and since GMCOR outranks *n, 033, which has identical markers, has a better constraint profile and is correctly classified as optimal. Other candidates (like 044, 055) may also respect GMCOR, but they fatally violate higher-ranked *XCM constraints. Whereas tableau T i l has shown that the ranking GMCOR » *n is important, tableau T12 shows that the ranking *m » GMCOR is also indispensable. In DAT.MASC ([+masc, -fem,+obl,+gov]) contexts, 0 2 3 (without matching Case markers) blocks 0 2 2 (with /m/ as both a Case and an agreement marker). The reason is that GMCOR has to be fulfilled only if no higher-ranked SONHŒR constraint is violated; but *m, by assumption, is ranked higher. There are other outputs (like 033, 044, and 055) that satisfy GMCOR

133

Remarks on nominal inflection in German

without violating *m, but these fatally violate higher-ranked constraints in *XCM (in s o m e cases, also *DAM). 33

T12: diesem guten {Tee) [Input: /dies, gut/; DAT.M, DAT.N] AGR CASE *DAM

021: 022: e? 023: 024: 025: 026: 033: 044: 055:

diesem gutes diesem gutem diesem guten diesem guter diesem gute diesem gut diesen guten dieser guter diese gute

*XCM *

*s *m GMCOR *n *R *e *!

*

*

**i

* *

*

*

*!

*

*

*

*!

*

*

*

*

*

*

*!

* * *

**! *!

**

*!

**

** ** **

As a final example of how weak adjective agreement is created by syntactic optimization, consider ACC.NEUT ([+masc,+fem,-obi,+gov]) contexts; see tableau T13. T13: dieses gute (Bier) [Input: /dies, gut/; ACC.N] AGR CASE *DAM *XCM

011: 012: 013: 014: •s* 015: 016: 022: 033: 044: 055:

dieses gutes dieses gutem dieses guten dieses guter dieses gute dieses gut diesem gutem diesen guten dieser guter diese gute

*s

*m GMCOR *n *R *e

**! *

*! *! *! *! *!

*

*!

*

*

*

*

*

*

*

*

*

*! *! *

** ** ** **

*DAM does not discriminate between the competitors here. Outputs that satisfy GMCOR fatally violate either *XCM constraints that regulate the optimal Case marker shape (see 022, 033, 044, and 055), or the higher-ranked SONHIER constraint *s (Oil). Consequently, GMCOR will have to be violated, and the decision is handed over to the SONHIER constraints, which select /e/, as in 015. 34

33. Recall from T4 that /m/ violates *SONCM as a Case marker in this context, but that all competitors violate higher-ranked constraints in *XCM. 34. The question arises of whether a more general formulation of GMCOR is feasible that dispenses with a restriction to [+masc,+gov] contexts. This would raise certain problems. To name just one, the Case marker fSJ in NOM.MASC contexts should also be imposed on the agreement marker, as in *dieser guter (Tee) ('this good teaNOM') instead of dieser gute (Tee): a NOM.MASC agreement marker does not have to obey *DAM.

Gereon Müller

134

To sum up so far, agreement marking (weak inflection) is triggered on adjectives by a Case-marked determiner. However, AGR does not specifically mention Case-marked determiners. It is more general since it requires an agreement marker on an adjective whenever there is a non-adjectival head in the same minimal residue that bears a Case marker. This makes interesting predictions for syntactic environments where there is no determiner present, but a Case-marked noun. This leads to the issue of strong adjective inflection.

4.2

Strong inflection

The paradigm P8 for strong adjective inflection is almost identical to the determiner inflection paradigm PI. The only difference is that GEN.MASC and GEN.NEUT specifications employ Ini rather than /s/. P8: Strong adjective inflection gut NOM ACC DAT GEN

M.SG N.SG F.SG PL

er en em en

es es em en

e e er er

e e en er

Strong adjective inflection is used in an NP whenever there is no preceding determiner; compare (11) with (8). (11) NOM guter Tee, gutes Bier, gute Milch, gute Gläser ACC guten Tee, gutes Bier, gute Milch, gute Gläser DAT gutem Tee, gutem Bier, guter Milch, guten Gläsern GEN guten Tees, guten Bieres, guter Milch, guter Gläser 'good tea (MASC), good beer (NEUT), good milk (FEM), good glasses (PL)' This basic distributional pattern follows directly from the approach developed so far: CASE requires a Case marker at the left edge of thç minimal residue of an NP, and if there is no determiner that can provide a Case marker, the leftmost item that can in principle bear a Case marker must do so. Hence, an adjective in this syntactic environment requires a Case marker, with AGR being vacuously satisfied if there is no other Case marker present.35

35. Strong adjective inflection is also chosen in contexts where the adjective is preceded only by an NP, as in Karls guter Tee ('Karl's good tea'). Given that the prenominai NP bears a possessive (-like) marker here and cannot participate in NP-internal Case inflection for principled reasons, the present analysis implies that CASE must be violated by a well-formed output in this context (since the leftmost item in SR cannot receive a Case marker), but that this violation must be kept minimal (so that the leftmost item in SR that can receive a Case marker - viz., the adjective - is indeed Case marked).

135

Remarks on nominal inflection in German

The problem that remains to be solved, then, is the fact that a [+masc,+obl,-gov] specification requires /n/, not /s/ as a Case marker; recall from tableau T5 that /n/ should be blocked in this context as a Case marker because of a fatal *SONCM violation that (of all the potential Case markers) only /s/ manages to avoid. Closer inspection reveals that this dilemma persists only as long as we take paradigms to be independently existing objects that are formed extra-syntactically. If we abandon this hypothesis, as I do here, there is nothing that would demand a homogeneous status of the markers in P8. Thus, I would like to suggest that whereas most of the markers in P8 are indeed Case markers, some are agreement markers whose presence is required by a Case marker - not a Case marker on the determiner (as in all the cases considered so far), but a Case marker on the noun.36 If this view is essentially correct, we are dealing with a clear case of contextually determined inflection, which supports the general perspective adopted here. To execute the idea, it turns out that one additional assumption has to be made: The partitioning of CASE into two subconstraints (la) and (lb) has not played a role so far, but it does now: CASERIGHT (CR, (lb)) must be ranked higher than C A S E L E F T (CL, (la)). With this in mind, consider first an example where strong adjective inflection and determiner inflection converge. Given, e.g., a NOM.MASC ([+masc,-fem,-obl,-gov]) specification on an adjective and a noun in a determinerless NP input, the optimal output has a Case marker /R/ on the adjective (compare T2), and no Case marker on the noun (compare T6, which has another input specification but illustrates the relevance of * N C M ) . This is shown in tableau Τ14 (which does not consider outputs that fatally violate *NCM by having a Case marker on N). Τ14 : guter Tee [Input: /gut, Tee/; ΝΟΜ.M] AGR CR CL *DAM

016: 026: 036: us" 046: 056: 066:

gutes Tee gutem Tee guten Tee guter Tee gute Tee gut Tee

*XCM

*

*m

GMCOR

*n *R *e

*!

*

*!

*

*!

*

*

* *

*S

*!

*

*!

Note that higher-ranked AGR is vacuously fulfilled because there is no adjective that shares a minimal residue with a Case marker on a non-adjectival head. Consider next a [+masc,+obl,-gov] specification. What is interesting about this context is that optimization leads to a Case marker /s/ on a noun with this specification because *NCM is not violated (see T7 above). Now, a conflict may arise between AGR and CASELEFT: By virtue of being the leftmost head that can receive a Case marker, the adjective should bear a Case marker to satisfy CASELEFT; and by virtue of being an adjective that co36. This approach builds on an insight in Eisenberg (2000:173), who concludes that the "fixed grammaticalization" of en in GEN.MASC and GEN.NEUT contexts of strong adjective inflection can be traced back to strong inflection on the noun itself.

136

Gereon Müller

occurs with a non-adjective (i.e., a noun) with a Case marker in the same minimal residue, the adjective should bear an agreement marker. Given that a marker cannot serve both Case and agreement, the conflict must be resolved by ranking, and given the ranking AGR » CASELEFT, agreement prevails. Therefore, the optimal marker in this context is /n/, and not /s/. This reasoning is illustrated in tableau T15 on the basis of the specification [+masc,-fem,+obl,-gov]. T15 : guten Tees [Input: /gut, Tee/; AGR

Oll: 016: es· 031: 036: 066:

gutes Tees (*!C) gutes Tee guten Tees (*!C) guten Tee gut Tee

CR

GEN.M]

CL

*!

(*A) (*A) (*A) (*A)

*!

*

*!

*D

*X

AM

CM

*s

*m

GM COR

*n *R

*e

*(*!A) *

(*C) (*C)

*

* *

For each marker on the adjective, a decision must be made as to whether it acts as a Case marker or as an agreement marker; this is expressed in the tableau by attributing each violation where the decision matters to the interpretation as a Case marker (*C) or as an agreement marker (*A). 0 3 1 is the optimal output: If the marker /n/ on the adjective is interpreted as an agreement marker, CASELEFT is violated, but the higher-ranked constraints AGR and CASERIGHT are respected. All other choices of agreement marker form (e.g., /s/ in O i l ) incur fatal violations of the constraints that regulate agreement markers (*s, in the case at hand); see T10. In all outputs in which the marker on the adjective is interpreted as a Case marker, we end up with a fatal violation of either AGR (if there is a Case marker on the noun; in particular, this holds for O i l if /s/ is interpreted as a Case marker) or CASERIGHT (if there is no Case marker on the noun, and an agreement marker is not required). This approach has a further consequence that is theoretically interesting, but empirically vacuous. Recall from T8 that there is a second context in which a noun is Case marked in German, viz., DAT.PL. Hence, the analysis given for GEN.MASC, GEN.NEUT /n/ vs. /s/ in the case of strong adjective inflection automatically predicts that in DAT.PL forms like guten Büchern ('good books') (though not in DAT.PL forms like guten Opas 'good grandfathers'), /n/ on the adjective is not a Case marker, but an agreement marker (or, using the standard terminology, the adjective in guten Büchern shows weak inflection even though a determiner is absent).37 I would like to contend that the system developed so far captures the gist of the weak/strong distinction with adjective inflection. Of course, in a comprehensive account, more will ultimately have to be said about certain exceptions, idiosyncracies, 37. Note in passing that this argument reinforces the decision to exclude archaic DAT.MASC, DAT.NEUT Id on nouns from the productive system of nominal inflection in the present analysis. If /e/ on a noun were to act as a Case marker triggered by CASERIGHT, we should wrongly expect agreement marking to prevail on the adjective, yielding ill-formed *aus harten Holze ('of solid wood') instead of aus hartem Holzt.

Remarks on nominal inflection in German

137

and, perhaps most importantly, cases of optionality. However, it seems to me that it will not be difficult to integrate most of these complications (e.g., by invoking faithfulness constraints that refer to lexically marked properties, by resorting to variable ranking or constraint ties, by motivating recategorization procedures, etc.), and leave the core of the present proposal unaffected. 38 Still, there is one kind of systematic exception that has received a lot of attention in the literature, viz., the phenomenon of "mixed inflection".

4.3

Mixed inflection

P9 gives the mixed adjective inflection paradigm as it can be found in, e.g., Eisenberg (2000:172). The standard generalization is that an adjective inflects according to this pattern if it follows a determiner belonging to a certain class that includes ein 'a', kein 'no', and mein 'my', among others. P9: Mixed adjective inflection gut

M.SG N.SG F.SG PL

NOM

er

es

e

en

ACC

en

DAT

en

es

e

en

en

en

GEN

en

en

en

en

en

P9 is identical to the weak inflection paradigm, except for the "strong" forms in NOM.MASC, NOM.NEUT, and ACC.NEUT contexts. Accordingly, this paradigm is often not accorded a status of its own, but is argued to arise on the basis of the other two inflection paradigms (see, e.g., Zwicky 1986). The mixed paradigm provides yet another clear case of contextually determined inflection: The syntactic environment that triggers the choice of a strong form in a post-determiner position has another property that evidently acts as the trigger: In NOM.MASC, NOM.NEUT, and ACC.NEUT contexts, determiners like kein are incompatible with the expected strong inflection themselves; hence, the adjective takes over the task of expressing strong inflection in these contexts (and only in these); see (12). (12) NOM kein guter Tee, kein gutes Bier, keine gute Milch, keine guten Gläser ACC keinen guten Tee, kein gutes Bier, keine gute Milch, keine guten Gläser DAT keinem guten Tee, keinem guten Bier, keiner guten Milch, keinen guten Gläsern 38. A particularly well-known case of optionality concerns the tendency of certain determiners to take on markers of the strong adjective inflection paradigm, as in (Januar) diesen Jahres ('January this yearGEN') alongside the expected (Januar) dieses Jahres. This effect might be captured by optional recategorization procedures (D —» A), by genuine output/output constraints, or some related means; but I will leave this question open here. See Zifonun et al. (1997:1928-1950) for further remarks, and for an overview of the variation encountered in this domain with different types of determiners.

Gereon Müller

138

GEN keines guten Tees, keines guten Bieres, keiner guten Milch, keiner guten Gläser 'no good tea (MASC), no good beer (NEUT), no good milk (FEM), no good glasses (PL)' This explanation can be integrated into the present analysis as follows. Suppose that there is a constraint *DECM outranking CASELEFT which yields the effect that determiners with certain features (morpho-phonological features, semantic features, or a combination thereof) are incompatible with a Case marker in NOM.MASC, NOM.NEUT, and ACC.NEUT contexts. Then, CASELEFT will have to be violated in these contexts, but given that this constraint is gradient, its violation can and, therefore, must be kept minimal by placing a Case marker on an adjective that follows the Case marker-less determiner.3 This is shown for an ACC.NEUT specification in the input in tableau T16, which lists only three relevant candidates - 015, which we would expect to be optimal without the overriding effect of *DECM (compare Τ13), 061, which is optimal, and 065, which fatally violates either CASELEFT (by having an uncalled-for agreement marker) or *VCM (by choosing a suboptimal Case marker form). (Irrelevant constraints like CR and *DAM are left out in T16.) T16: kein gutes (Bier) [Input: /kein, gut/; ACC.N] 015: keines gute ι®" 061: kein gutes 065: kein gute

*DECM AGR *!

CL

*XCM

*

*(*!A) (*!C)

*s *m GMCOR *n *R *e *

*

*

* *

*

*

In contrast, in all those environments where *DECM permits a Case marker on a determiner like kein, the CASELEFT violation incurred by candidates of the 0 6 1 type will be fatal, and inflection proceeds exactly as shown in subsection 4.1.

4.4

Adjective sequences

I have not yet addressed NPs with more than one adjective, i.e., more than one AP specifier. Given AGR, all adjectives in (APs in) SpecN positions require an agreement marker if there is a Case marker on either a determiner or a noun in the same minimal residue, and this prediction is correct. The more interesting case is provided by multiple adjective enviroments without a Case marker on a non-adjective nearby. As we have seen, the first adjective must then have a Case marker, because of CASELEFT. But what about adjectives that follow the first one? CASELEFT does not require a Case marker here, and AGR does not require an agreement marker (because the Case marker is on an adjective, i.e., the same type of head). Thus, if nothing else is said, the prediction is that non-first adjectives in these environments have no marker at all. This pre39. If there is no adjective in the input, a violation of CASELEFT cannot be minimized - kein Tee ('no teaN0M') and Karls Tee ('Karl's tea') violate CASELEFT and CASERIGHT in exactly the same way.

139

Remarks on nominal inflection in German

diction is not borne out, though. With one exception (DAT.MASC, DAT.NEUT), these adjectives must have the same marker that the first adjective has; see (13). (13) NOM guter alter/*e Tee, gutes altes/*e Bier, gute alte Milch, gute alte/*n Gläser ACC guten alten Tee, gutes altes/*e Bier, gute alte Milch, gute alte/*n Gläser DAT gutem altem/?n Tee, gutem altem/?n Bier, guter alter/*n Milch, guten alten Gläsern GEN guten alten Tees, guten alten Bieres, guter alter/*n Milch, guter alter/*n Gläser 'good old tea (MASC), good old beer (NEUT), good old milk (FEM), good old glasses (PL)' Thus, the following constraint suggests itself. (Adjective Correspondence): Adjectives in the same minimal residue have identical markers.

( 1 4 ) ADJCOR

Assuming that ADJCOR dominates the SONHIER constraints, the uniformity effect visible in German adjective sequences follows. This is shown for a NOM.MASC input specification in tableau T17. Note in particular that ADJCOR successfully blocks a (superfluous) agreement marker and non-marking on the non-first adjective (see 0456, 0466), by forcing marker identity. 0446 then has a better constraint profile than, e.g., 0556 for the same reason that 0 4 6 has a better constraint profile than 0 5 6 in T14. T17: guter alter Tee [Input: /gut, alt, Tee/; ADJ

NOM.M]

AGR CR CL

*XCM

*s *m

COR

ι®· 0446: 0456: 0466: 0556:

guter alter Tee guter alte Tee guter alt Tee gute alte Tee

GM

*n *R *e

COR *

**

*!

*

*

*!

*

*

*

*

**

*!

Consider now the DAT.MASC, DAT.NEUT context in (13). As shown in tableau T18, the present analysis predicts 0226, with identical Case markers, to be the sole optimal output. However, whereas 0226 is indeed well formed for all speakers, many speakers also accept 0236, without matching markers. Thus, for a [+masc,+obl,+gov] specification, and for this specifcation only, marker identity is not necessarily imposed on adjective sequences, and optionality may arise. (As with most instances of optionality, speakers that permit both options may then associate functional or minor semantic differences with the two options.) Τ18: gutem altem Tee [Input: /gut, alt, Tee/; ADJ

AGR CR CL

COR

er 0226: gutem altem Tee 0236: gutem alten Tee

*!

DAT.M] *X

*s *m

CM

GM

*n *R *e

COR

*

*

**

*

*

*

*

*

Gereon Müller

140

It is unclear to me whether one should account for the optional availability of 0236 in the same way in which the optimality of 0226 is derived. At least for present purposes, it seems preferable to classify 0236 as not resulting from the core system of nominal inflection in present-day German. Arguably, its existence can successfully be traced back to the influence of analogy (the Iml-Ini sequence of Case and agreement markers is both formally conspicuous and frequent); like other instances of analogy, this could be formulated in terms of output/output constraints.40

4.5

Weak noun inflection

Let me finally turn to the inflection of weak masculine nouns, which was alluded to, but not addressed, in section 3. What follows is a brief sketch of how weak masculine nouns might fit into the overall approach pursued here. Paradigm PIO lists the markers for weak masculine nouns like Dirigent ('conductor') and Bote ('messenger'). Presence or absence of e depends on the nature of the coda of the stem (consonantal or vocalic); we can thus assume In/ as the sole marker.41 PIO: Weak masculine noun inflection Dirigent

M.SG

ACC DAT GEN

Bote

M.SG

NOM

NOM

en en en

ACC DAT GEN

η η η

Clearly, the present analysis does not support the idea that we are dealing with a second type of Case marking on masculine nouns here; this would lead directly to the postulation of declension class features, an assumption that is otherwise not necessary in German, and perhaps dispensable even for other languages (see Wunderlich 1996 and Fraser and Corbett 1995 for opposing views). In fact, we have theory-internal evidence that the markers in PIO are not Case markers at all. First, the interaction of FCRs for Case markers and the SONHIER constraints preclude /n/ as a Case marker in DAT.MASC and GEN.MASC contexts. Second, in contrast to what we have seen with true Case markers in subsection 4.2, the GEN marker /n/ of a weak masculine noun fails to trigger the (AGR-induced) change from Isl to InJ: In the marginal cases in which a masculine noun that is [+animate] (which is the prototypical feature of weak masculine nouns) can occur without a determiner in GEN contexts at all, a regular masculine noun 40. Note in passing that it would not be possible to simply assume that ADJCOR can be tied with *m. Under this assumption, *gutem alt Tee ('good old tea') would optionally become optimal, rather than the intended gutem alten Tee (recall that there is no trigger for an agreement marker here). Moreover, *s would now invariably outrank ADJCOR, and a well-formed NP like gutes altes Bier ('good old beer') should not be possible anymore. 41. The plural marker for weak masculine nouns is uniformly /n/; hence, Case marking in the plural can be ignored.

Remarks on nominal inflection in German

141

induces agreement marking on a preceding adjective; see (15a) vs. (15b). However, a weak masculine noun fails to induce Ini on the preceding adjective; in addition, in this environment, presence of Is/ is actually somewhat more acceptable than with regular masculine nouns; see (15c) vs. (15d). (15) (Aber das muss er doch tun! Das ist doch ...) ('But he has to do this! This is ...') a. ?guten Mannes Pflicht b. *gutes Mannes Pflicht good man0EN duty' c. *guten Jungen Pflicht d. ??gutes Jungen Pflicht 'goodboyoEN duty' If the /n/ markers on weak masculine nouns are not Case markers, the question arises of what else they might be. Here I would like to suggest that they are agreement markers. Thus, suppose that weak masculine nouns, as a lexical property, underlie AGR, and that there is a higher-ranked constraint which precludes their bearing a NOM marker. We have seen that agreement markers can only arise in the presence of Case markers. Hence, if a Case marker is present in the SR, a weak masculine noun in the HR of the same minimal residue will receive an agreement marker /n/ in ACC, DAT, and GEN contexts, as shown in subsection 4.1 (and no marker at all in NOM contexts). If, on the other hand, there is no Case marker present in the same NP (because there is no determiner or adjective available in the input), the weak masculine noun will not receive a marker. This way, Gallmann's (1996, 1998) observation is derived that (ACC or DAT) /n/ markers on weak masculine nouns require an inflecting adjective or determiner in the same domain: (16a) blocks (16b) as suboptimal because the latter fatally violates AGR, whereas (16c) is blocked by (16d) because the former fatally violates *n (in the absence of a trigger for an agreement marker). (16) a. b. c. d.

(ein Orchester mit) [np eigenem Dirigenten] *(ein Orchester mit) [np eigenem Dirigent] *(ein Orchester mit) [ NP Dirigenten] (ein Orchester mit) [np Dirigent] 'an orchestra with (own) conductor'

The ranking AGR » CASERIGHT ensures that in the case of conflict, a weak masculine noun will choose agreement marking over Case marking. However, given the highranked constraint *NCM, there is no potential conflict, except for GEN environments. Here, agreement marking prevails if a pre-nominal determiner (or, marginally, adjective - see above) is present; compare des Dirigenten ('the conductorGEN') with *des Dirigente. Still, there is a single context in which the present approach could wrongly predict a Case marker to occur on a weak masculine noun, viz., GEN environments without a preceding determiner. Thus, it seems that the optimal form should now be *Dirigents instead of Dirigent. However, as noted by Gallmann (1996, 1998) something needs to be said for these cases anyway: Irrespective of the question of whether a

142

Gereon Müller

suitable sentence could be construed into which an NP with a bare weak masculine noun could fit, other masculine or neuter nouns turn out to be impossible as bare GEN NPs as well; see, e.g., Gallmann's examples (die Verarbeitung) *Holzes ('the manufacturing WOOCIGEN') and (die Herstellung) *Betons ('the production concreteGEN'). Thus, whatever is responsible for this ban on GEN NPs headed by bare nouns in general is likely to also cover the subcase of bare masculine nouns.

5.

Conclusion

To conclude, I have developed an optimality-theoretic approach to nominal inflection in German that determines the form and the distribution of inflection markers in one and the same syntactic component (and, in doing so, dispenses with the hypothesis that nominal inflection paradigms are linguistic objects in their own right, as well as with the standard assumption that German nominal inflection markers are morphemes); and I have tried to argue that such an approach is both viable and independently supported (given the importance of the syntactic environment for the choice of inflection marker, and the partial, sonority-based motivation for the shape of nominal inflection markers). I have argued that there are two kinds of inflection markers in the NP domain: Case markers and agreement markers, the latter dependent on the presence of the former. The analysis relies on four types of constraints. CASE (decomposable into CASELEFT and CASERIGHT) and AGR are the main triggers for Case and agreement markers, respectively. The form of inflection markers is determined by the interaction of low-ranked SONHIER constraints (i.e., the sonority hierarchy: *s » *m » *n » *R » *e) and higher-ranked feature co-occurrence restrictions (FCRs) which correlate morpho-syntactic features and phonological features, relativized with respect to Case and agreement markers (*X/CM, * x / A M ) . Finally, in addition to inflection triggers, the sonority hierarchy, and FCRs, the approach employs a fourth type of constraint, viz., correspondence constraints on markers (GMCOR, ADJCOR); these constraints require marker identity in certain domains and can thereby also act as triggers for inflection marking. Needless to say, the analysis given here has left open many questions, both empirically and theoretically. On the empirical side, it is clear that a fully comprehensive account would have to say more in various domains: e.g., in the case of lexically determined variation and partial optionality with some lexical items, particularly determiners, or in the case of certain additional restrictions on inflection marking in GEN NPs. On the theoretical side, an obvious question concerns the status of the constraints invoked here. A standard assumption in optimall y theory is that the ranked constraints used for harmony evaluation are universal. Viewed from this perspective, CASE and AGR, the SONHIER subhierarchy, and the marker identity constraints are arguably acceptable. The relativized FCRs certainly have a highly specific shape; still, they exhibit similar patterns. One might speculate that these constraints are not universal as such, but can be generated in a language by (i) taking universal morpho-syntactic and phonological features as basic, (ii) combining

Remarks on nominal inflection in German

143

the features by Boolean operations, and (iii) correlating the specifications of morphosyntactic features and phonological features achieved this way by logical implication. On this view, what would be universal is not a FCR per se, but the vocabulary that it uses, and the scheme that it employs.42 For the moment, I have to leave these and other questions open.

References Aissen, Judith (2000). Differential object marking: Iconicity vs. economy. Unpublished manuscript, University of California, Santa Cruz. Anderson, Stephen (1992). A-Morphous Morphology. Cambridge: Cambridge University Press. Bierwisch, Manfred (1967). Syntactic features in morphology: general problems of socalled pronominal inflection in German. In To Honour Roman Jakobson, 239-270. Mouton: The Hague/Paris. Blevins, James (1995). Syncretism and paradigmatic opposition. Linguistics and Philosophy 18, 113-152. Chomsky, Noam (1995). The Minimalist Program. Cambridge, Mass.: ΜΓΓ Press. —(2001). Derivation by phase. In Ken Hale. A Life in Language, Michael Kenstowicz (ed.), 1-52. Cambridge, Mass.: MIT Press. Eisenberg, Peter (2000). Grundriß der deutschen Grammatik. Band 1: Das Wort. Stuttgart: Metzler. Elgersma, Diana and Paul Houseman (1999). Optimality theory and natural morphology: An analysis of German plural formation. Folia Linguistica 33, 334-353. Fraser, Norman and Greville Corbett (1995). Gender, animacy, and declension class assignment: A unified account for Russian. In Yearbook of Morphology 1994, Geert Booij and Jaap van Marie (eds.), 123-150. Dordrecht: Kluwer. Gallmann, Peter (1996). Die Steuerung der Flexion in der DP. Linguistische Berichte 164, 283-314. —(1998). Case underspecification in morphology, syntax and the lexi-con. In Possessors, Predicates and Movement in the Determiner Phrase, Artemis Alexiadou and Chris Wilder (eds.), 141-175. Amsterdam: Benjamins. Gazdar, Gerald, Ewan Klein, Geoffrey Pullum and Ivan Sag (1985). Generalized Phrase Structure Grammar. Oxford: Blackwell. Gelhaus, Hermann (1998). Die Wortarten. In Duden: Die Grammatik. Mannheim: Dudenverlag.

42. Similar considerations apply in the case of the two constraint generation mechanisms harmonic alignment and constraint conjunction that have been suggested in Prince and Smolensky (1993) and Smolensky (1995), respectively.

144

Gereon Müller

Golston, Chris and Richard Wiese (1996). Zero morphology and constraint interaction: subtraction and epenthesis in German dialects. In Yearbook of Morphology 1995, Geert Booij and Jaap van Marie (eds.), 143-159. Dordrecht: Kluwer. Grimshaw, Jane (1997). Projection, heads, and optimality. Linguistic Inquiry 28, 373422. —(2001). Optimal clitic positions and the lexicon in Romance clitic systems. In Optimality-Theoretic Syntax, Géraldine Legendre, Jane Grimshaw and Sten Vikner (eds.), 205-240. Cambridge, Mass.: ΜΓΓ Press. Halle, Morris (1994). The Russian declension: An illustration of the theory of distributed morphology. In Perspectives in Phonology, Jennifer Cole and Charles Kisseberth (eds.), 29-60. Stanford: CSLI Publications. Heck, Fabian and Gereon Müller (2000). Repair-driven movement and the local optimization of derivations. Unpublished manuscript, Universität Stuttgart and IDS Mannheim. Hoberg, Ursula (2001). Das Genus des Nomens. Unpublished manuscript, IDS Mannheim. Ito, Junko, and Armin Mester (2001). Structure Preservation and Stratal Opacity in German. In Segmental Phonology in Optimality Theory, Linda Lombardi (ed.), 261295. Cambridge: Cambridge University Press. Kiparsky, Paul (1982). From cyclic phonology to lexical phonology. In The Structure of Phonological Representations, vol. 1, Harry van der Hülst and Neil Smith (eds.), 131-175. Dordrecht: Foris. Legendre, Géraldine, Paul Smolensky and Colin Wilson (1998). When is less more? Faithfulness and minimal links in wh-chains. In Is the Best Good Enough?, Pilar Barbosa et al. (eds.), 249-289. Cambridge, Mass.: ΜΓΓ Press and MITWPL. McCarthy, John, and Alan Prince (1995). Faithfulness and reduplicative identity. In Papers in Optimality Theory, Jill Beckman, Laura Walsh-Dickie and Suzanne Urbanczyk (eds.), 249-384. Amherst, Massachussetts: UMass Occasional Papers in Linguistics 18. Müller, Gereon (2001). Free word order, morphological case, and sympathy theory. Unpublished manuscript, IDS Mannheim. To appear in Resolving Conflicts in Grammar: More Papers on OT, Gisbert Fanselow and Caroline Féry (eds.). (Special Issue of Linguistische Berichte). Neef, Martin. 1998. A case study in Declarative Morphology: German case inflection. In Phonology and Morphology of the Germanic Languages, Wolfgang Kehrein and Richard Wiese (eds.), 219-240. Tübingen: Niemeyer. Prince, Alan and Paul Smolensky (1993). Optimality Theory. Constraint Interaction in Generative Grammar. Unpublished manuscript, Rutgers University. To appear: Cambridge, Mass.: ΜΓΓ Press. Ross, John (1980). Ikonismus in der Phraseologie. Zeitschrift für Semiotik 2, 39-56. Samek-Lodovici, Vieri (2001). Agreement impoverishment under subject inversion. A crosslinguistic analysis. Unpublished manuscript, UC London. To appear in Resolving Conflicts in Grammar: More Papers on OT, Gisbert Fanselow and Caroline Féry (eds.). (Special issue of Linguistische Berichte).

Remarks on nominal inflection in German

145

Smolensky, Paul (1995). On the internal structure of Con, the constraint component of UG. Unpublished manuscript, Johns Hopkins University. Stump, Gregory (2001). Inflectional Morphology. Cambridge: Cambridge University Press. Tesar, Bruce and Paul Smolensky (2000). Learnability in Optimality Theory. Cambridge, Mass.: ΜΓΓ Press. Wegener, Heide (1999). Die Pluralbildung im Deutschen - ein Versuch im Rahmen der Optimalitätstheorie. Linguistik Online 4, 3/99. Wiese, Bernd (1996). Iconicity and syncretism. On pronominal inflection in Modern German. In Theoretical Linguistics and Grammatical Description, Robin Sackmann (ed.), 323-344. Amsterdam: Benjamins. —(1999). Unterspezifizierte Paradigmen. Form und Funktion in der pronominalen Deklination. Linguistik Online 4, 3/99. Wiese, Bernd (2001). Pronominale Deklination. Handout of a talk at IDS Mannheim. Williams, Edwin (1997). Blocking and anaphora. Linguistic Inquiry 28, 577-628. Wunderlich, Dieter (1996). Minimalist morphology: The role of paradigms. In Yearbook of Morphology 7995, Geert Booij and Jaap van Marie (eds.), 93-114. Dordrecht: Kluwer. —(1997a). A minimalist model of inflectional morphology. In The Role of Economy Principles in Linguistic Theory, Chris Wilder, Hans-Martin Gärtner and Manfred Bierwisch (eds.), 267-298. Berlin: Akademie Verlag. —(1997b). Der unterspezifizierte Artikel. In Sprache im Fokus, Christa Dürscheid, Karl Heinz Ramers and Monika Schwarz (eds.), 47-55 Tübingen: Niemeyer. —(1999). German noun plural reconsidered. Unpublished manuscript, Universität Düsseldorf. Wurzel, Wolfgang Ullrich (1990). The mechanisms of inflection: Lexicon representations, rules, and irregularities. In Contemporary Morphology, Wolfgang Dressler et al. (eds.), 203-216. Berlin: Mouton de Gruyter. —(1998). Drei Ebenen der Struktur von Flexionsparadigmen. In Models of Inflection, Ray Fabri, Albert Ortmann and Teresa Parodi (eds), 225-241. Tübingen: Niemeyer. Zifonun, Gisela (2001). Grammatik des Deutschen im europäischen Vergleich: Das Pronomen, Teil 1: Überblick und Personalpronomina. Mannheim: IDS working paper amades 4/01. —; Ludger Hoffmann, and Bruno Strecker (1997). Grammatik der deutschen Sprache. Berlin: de Gruyter. Zwicky, Arnold (1986). German adjective agreement in GPSG. Linguistics 24, 957990.

Economy-based splits, constraints and lexical representations Albert Ortmann

1.

Introduction*

This paper is concerned with morphosyntactic asymmetries - commonly known as splits - that are found with argument linking and phrase-internal agreement. What has been dealt with in the recent literature (Aissen 1999, Stiebels 2000a,b) is the question which out of more than one conceivable linker is chosen; for example, which set of agreement affixes, or which morphological case - thus, basically, the issue of split ergativity. I am concerned with a related, though different problem, namely the split between a more specific morphological form in one grammatical environment and a less specific form in another, where by "less specific" I understand either a less expressive default marking, or no inflection at all. One major type of such splits, namely that of object marking asymmetries, has recently been addressed by Aissen (2000). I follow the general outline of the Differential Object Marking account proposed there, although I propose some modifications along the lines in Stiebels (2000b). With this theoretical background, I provide analyses for plurality splits in the subject agreement of e.g. Arabic, as well as for a particularly intricate split in the possessor agreement of Hungarian. The first goal I pursue is to establish a common denominator for the various types of asymmetries that is based on functional economy. The second goal concerns the question whether such asymmetries should be stated in the constraint inventory, or in the lexical representations of the involved morphemes. I argue for a specific inter-dependence of a number of micro-constraints that are (following in essential respects Aissen's proposal) ordered by harmonic alignment, and explicit lexical specifications of the

* The work reported here was conducted in the project "Categories and Systems of Inflection" within the Research Programme SFB 282 "Theory of the Lexicon", financed by the German Research Foundation (DFG). I hope that Dieter Wunderlich's inspiration is visible in these pages, at least in so far as the ideas presented here were sharpened through his perspective on morphology and on linguistics in general. I would like to thank Thomas Gamerschlag, Birgit Gerlach, Chris Pifión, Peter Sells, and Carsten Steins for comments and discussion, as well as the audiences of oral presentations at Stanford, Cologne, and at the conference "The Lexicon in Linguistic Theory", Düsseldorf, August 2001. Finally, thanks go to the editors of this volume, Ingrid Kaufmann and Barbara Stiebels, as well as to an anonymous reviewer for their helpful criticism.

Albert Ortmann

148

involved word forms. I propose that positive specifications of inflectional features contrast with both unspecified values and with negative values that are inserted per default, where the decision between the latter two is determined by the constraint ranking. The paper is organised as follows: section 2 is concerned with salience splits found in object linking. I discuss the harmonic alignment account of object marking by Aissen (2000) as well as the version proposed by Stiebels (2000b). Section 3 looks at salience splits in subject agreement, for which a further markedness hierarchy is proposed, the plurality scale. In section 4, I point out the relevance of economy in plural noun phrases as analysed in Ortmann (2000). Section 5 shows how the inventory developed thus far accounts for Hungarian possessor agreement, which involves a combination of two individual splits. Section 6 concludes with the major findings.

2.

Object linking splits governed by salience

Languages with morphological object linking normally display asymmetries in that agreement or case morphology is only observed under specific conditions. It is particularly typical of languages with object agreement that the realisation of object markers is restricted to noun phrases with either human (or animate) referents, or to those with a definite (or specific) interpretation.

2.1

Data, salience hierarchies, lexical entries and constraints

One language where object agreement is restricted in the above sense is Swahili. Indefinite inanimate objects like the one in (la) are not reflected in verbal agreement. 1 (1) Swahili (Givon 1976: 159) a. ni-li-soma ki-tabu lSG-PAST-read CL7-book Ί read a book.' c.

b. ni-li-ki-soma ki-tabu lSG-PAST-CL7-read CL7-book Ί read the book.'

ni-li-mw-ona m-tu m-moja lSG-PAST-CLl-see CLl-person CLl-one Ί saw one person.'

1. Throughout the paper, I use the following abbreviations in glosses: ANIM 'animate', CL 'noun class', COP 'copula', DET 'determiner', DF 'definite article', F 'feminine', M 'masculine', MOD 'licenser of modifier', NOM 'nominative', OBJ 'object', PAST 'past tense', PERF 'perfective', PL 'plural', POR 'possessor', PRÊT 'preterite', PRON 'personal pronoun', PRT 'particle', SG 'singular', SUBJ 'subject'.

Economy-based splits, constraints and lexical d. ni-li-mw-ona

yula

representations

149

m-tu

lSG-PAST-CLl-see DET.CLl C L l - p e r s o n

Ί saw the person' While the presence of an agreement marker for an inanimate object correlates with the interpretation of the object as [+definite], as can be seen from (la) vs. (lb), for animate objects agreement is specified on the verb regardless of the value for definiteness, as in (lc,d). We are therefore dealing with a split that is based on the combination of definiteness and humanness of the referent of the noun phrase, more precisely a disjunction of the features [+definite ν +animate]. Similarly, in Palauan, an Austronesian language of Micronesia, the conditions for direct object agreement in the perfective aspect involve a combination of specificity and humanness. According to Woolford (1995: 659), "[w]hen the object is human, it triggers object agreement regardless of what other features it has." Non-human objects, by contrast, are only cross-referenced if they are specific. Consider the examples in (2); the particle a marks a nominal phrase, while el licenses all kinds of modifiers of the noun. (2)

Palauan (Woolford 1995: 658ff) a. Te-'illebed a bilis a rengalek 3.SUBJ-PERF.hit PRT dog PRT children 'The kids hit a dog / the dogs / some dog(s)' b. Te-'illebed-ii a bilis a rengalek 3 .SUBJ-PERF.hit-3SG PRT dog PRT children 'The kids hit the dog.' c.

Ak mils-terir a retede el sensei. I saw-3PL.OBJ PRT three MOD teacher Ί saw three teachers.'

In (2a), the verb fails to show object agreement, the object being [±specific,-human]. There is, however, agreement with objects that are either specific or human, as (2b,c) show. The split can therefore be characterised in terms of the feature disjunction [+specific ν +human]. 2 As for object linking by case marking, consider the Semitic language Maltese, where direct objects are marked by means of lil only if they refer to [+human] individuals: (3)

Maltese (Fabri 1993: 117f) a. Raj-t *(lil) Pawlu. see-lSG CASE Paul Ί saw Paul.'

b.

Xtraj-t

(*lil-)il-ktieb.

buy-lSG CASE-DF-book

Ί bought the book.'

An overview of animacy-based linking splits is found in Ortmann (1998), where the phenomenon is seen in the context of numerous other asymmetries governed by ani-

2. Note that, strictly speaking, the category of number is also involved in that non-human plural objects are not respected in agreement even if they are specific. Hence, the exact specification would obviously be [(+specific,-pl) ν +human].

150

Albert Ortmann

macy, including those in gender systems, and the cross-linguistic relevance of the [±animate] (or [±human], respectively) distinction for inflection is pointed out. Generalising from these examples, one can state for object agreement and object case marking that the more explicitly individuated the referent of a noun phrase is, the more likely is a morphological specification. In addition to animacy, the rather vague notion of explicit individuation involves categories such as specificity and definiteness, person and "pronounness", all of which will have to be considered in the analysis. Pretheoretically, one can think of referential-semantic explicitness as salience, hence this notion can be seen as the overall concept between the subdimensions that govern the splits. In the typological literature, various "relevance" or "salience" hierarchies have been put forward, among them the "personal hierarchy" in (4a) (Siewierska 1988), the "semantic hierarchy" in (4b) (Silverstein 1976), and the "plurality splits" hierarchy (for more on which see section 3) in (4c) by Smith-Stark (1974). (4)

a.

"personal hierarchy" (Siewierska 1988): 1st > 2nd > 3rd human > higher animals > other organisms > inorganic matters > abstracts b. "semantic hierarchy" (Silverstein 1976): 1st > 2nd > 3rd human > 3rd animal > 3rd inanimate c. "plurality splits" hierarchy by Smith-Stark (1974): speaker > addressee > kin > rational > human > animate > inanimate

The fact that object linking splits like the above ones follow these hierarchies, in the sense that languages choose some step on them as the threshold for the split, can be functionally explained by the requirement for an economic and efficient linking system. Plank (1980: 293f) characterises the overall strategy so as "to distinctively mark direct objects only if the risk of actual ambiguity is relatively high, i.e. if they have some of the semantic or pragmatic properties (such as animacy, topicality, definiteness) normally characteristic of subjects". The distribution of splits is therefore economic in the best sense, since the morphological markers are avoided in cases of little concrete individuation: object linking on the verb is avoided when the object does not have prototypical properties of subjects. This is also pointed out by Aissen (2000: 3), who speaks of those splits as differential object marking (henceforth DOM), and uses the notion of iconicity: "Functionally, the overt marking of atypical objects facilitates comprehension where it is most needed, but not elsewhere. DOM systems are thus relatively economical." As to the representation of this type of functional economy, there are two conceivable ways in which they can be implemented in a theoretical analysis. One possibility would be to assign the disambiguating function of overt object marking to the involved morphemes themselves, that is, to the lexical inventory. For the object markers involved in the above examples, we might therefore want to state input restrictions like the ones in (5):

Economy-based splits, constraints and lexical representations (5)

a. Swahili ki-: b. Palauan -ii: c. Maltese lil:

151

[+hr,]/ [+definite ν +animate]3 [+hr,] / [+specifíc ν +human] [+hr] / [+human]

The lack of object markers with inanimate and/or unspecific referents would thus be captured by the failure to match the input conditions of these markers. Such a solution, however, does not seem desirable. First, it would assign unnecessary burden to the lexical items. In particular, each one of the more than fifteen object class markers of Swahili would have to bear the same input restriction without any statement in the grammar that this common property is not accidental. Second, it would have little explanatory force, given that the functional motivation that was characterised as being due to economy, in that it restricts the markers only to potentially ambiguous contexts, would not be reflected by the analysis. Therefore, another solution suggests itself according to which the observed asymmetries are implemented by an economy constraint that requires that agreement morphology be not realised if it is of little functional load; in this case, if it does not help to disambiguate the role of subject and object. This constraint will interact with a conflicting constraint that requires explicit overt licensing of argument. A proposal in terms of such fairly general constraints is found in Ortmann (2002). One constraint penalises the lack of agreement marking. More generally, it requires morphological material to be overtly realised and thus belongs to the family of MAX constraints as introduced in Correspondence Theory (McCarthy & Prince 1995). As regards markedness constraints that require economy, in Ortmann (2002) it is assumed that AVOID constraints require morphological effort to be "saved". The specific instantiation that is relevant for the splits at issue is AVOIDAGR(NON-SALIENTOBJ). (6)

MAX(F)/ARG:

Every specification of an argument in the input has a correspondent in the output. ("Arguments are morphosyntactically licensed.") (7)

AVOIDAGR(NON-SALIENTOBJ):

a.

General version: Avoid object agreement morphology where the object has no prototypical subject-properties. b. Specific micro-constraints: Avoid object agreement morphology for [-animate]/[-specific]/ .... objects. With the ranking in (8), where the economy constraint dominates the explicitness constraint, a still very rough account of the lack of object agreement in the above examples can be sketched:

3. For the sake of the argument, in (5a) I ignore the fact that in Bantu languages such as Swahili, the default assignment for animates is to class 1, the "human" class. As to the output feature [+hr] in (5a)-(c), for the moment it can be understood as "object case" or AgrObj; more background will be provided in 2.3.

152 (8)

Albert Ortmann Ranking for Swahili, Palauan, Maltese,... : AVOIDAGR(NON-SALIENTOBJ) » MAX(F)/ARG

The ranking ensures that an object marker is only realised with a [+animate] and/or [+specific] object, for otherwise the higher-ranked constraint A V 0 I D A G R ( N 0 N SALIENTOBJ) would be violated. Obviously, however, this constraint is still rather vague: the language-specific subtleties of object splits in terms of the above salience scales, that is, what exactly in a given language counts as salient, should be stated more explicitly. Formally, one would either have to assume conjunctions or disjunctions of feature specifications in it, or a cluster of micro-constraints, with language-specific local conjunctions. A more principled analysis that chooses the latter alternative is offered in a recent paper by Aissen (2000), which further pursues the analytical framework outlined in Aissen (1999).

2.2

Aissen's Differential Object Marking approach

Aissen's (2000) analysis of object-linking splits phenomena (which she refers to as DOM) draws on the formal concept of harmonic alignment as originally introduced by Prince & Smolensky (1993). Harmonic alignment is a formal procedure that takes two scales and associates each element of one scale with each element of the other. This gives rise to a set of markedness constraints in an order which is universally fixed, that is, not subject to language-specific ranking. (9)

Suppose a scale X > Y, next to a scale a > b > ... > z. Then harmonic alignment of both dimensions is the pair of the following scales: H x : X / a D X / b D ... >X/z H y : Y/z 3 ... > Y/b 3 Y/a The constraint alignment is the pair of constraint hierarchies: C x : *X/z » ... » *X/b » *X/a C y : *Y/a » *Y/b ... » *Y/z

Given the general scheme in (9), Aissen assumes one of the two scales that are to be aligned to be instantiated by the relational scale "subject higher than object", and the other by either the animacy scale or the definiteness scale (recall that the latter two specify different subdimensions of salience). The harmonic alignment of the two is as in (11), which is to be interpreted as a markedness hierarchy: (10) Relational Scale: Animacy Scale: Definiteness Scale:

Su(bject) > Ob(ject) Hum(an) > Anim(ate) > Inan(imate) Pronoun > Name > Definite > Indefinite Specific > Nonspecific

(11) Harmonic alignment:

Su/Pro 3 Su/PN 3 Su/Def 3 Su/Spec 3 Su/Nspec ObNSpec 3 Ob/Spec 3 Ob/Def 3 Ob/PN 3 Ob/Pro

Economy-based splits, constraints and lexical representations

153

According to this markedness hierarchy, pronominal subjects are more harmonic than other subjects, and likewise, non-specific objects are more harmonic than other objects. Inverting the harmonic alignment hierarchy yields the following hierarchy of markedness (or "avoid") constraints: (12) Avoid constraints:

*Su/NSpec » *Su/Spec » *Su/Def » *Su/PN » *Su/Pro *Ob/Pro » *Ob/PN » *Ob/Def » *Ob/Spec » *Ob/Nspec

The order in (12) renders a non-specific subject "worse" (in the sense that its realisation constitutes a higher violation) than that of a specific one, and this in turn than that of a definite one, etc., while a pronominal object is worse than a proper name object, etc. Aissen furthermore assumes local conjunctions with two general constraints that are responsible for explicitness and economy, respectively, namely * 0 and *STRUC. Both are used in their instantiations for case, noted by sub-indices * 0 c and *STRUCC. (Basically, the former thus corresponds to MAX(arg), while the latter corresponds to AVOIDAGR(NON-SALIENTOBJ), as used in the preceding subsection.) This gives rise to the following constraint hierarchies, which are responsible for DOM: (13) a. b.

*Ob/Hum & *0 C » *Ob/Anim & *0 C » *Ob/Inan & *0 C *Ob/Pro & *0 C » *Ob/Name & *0 C » *Ob/Def & *0 C » *Ob/Spec & *0 C » *Ob/NSpec & *0 C

According to the hierarchy in (13a), a human object without case universally constitutes a higher violation than a non-human object without case. It is in this sense that universal hierarchies formulated by such conjunctions account for the observed markedness findings. For example, if a language does not mark animate objects, it will not mark inanimate objects either. In other words, object marking is most "forceful" with the most marked objects, that is with those that have typical subject properties. The attractive thing about dealing with markedness splits in terms of harmonic alignment hence is that although we have a considerable proliferation of constraints, the ranking of these constraints does not impose any learnability effort, but rather is universally fixed. The language-specific part of the asymmetry, that is, the threshold for object marking in an individual language, is stated by interpolating the economy constraint *STRUCc- The interpretation of this interpolation is the tension of markedness and economy, in the sense that a given language saves overt marking below a certain cutoff point of the universal constraint hierarchy. For the straightforward cases such as Turkish, where the split follows just one sub-dimension (in this case, definiteness), Aissen notes rankings like the following: (14) ... » *Ob/Def & *0 C » *Ob/Spec & *0 C » *STRUCC » *Ob/NSpec & *0 C This ranking will ensure that specific (and consequently, definite, pronominal, etc.) objects are case-marked in Turkish, while non-specific ones are not. The question is how to account for the more complex cases, that is, languages in which object marking involves certain combinations of animacy and definiteness specifications. Aissen dis-

154

Albert Ortmann

cusses Persian, Hindi and Spanish, which in their complexity closely resemble Swahili and Palauan as exemplified above. Some languages involve conjunctions of features, such as [+specific,+animate] for case marking of objects by the preposition a in Modern Spanish and [+specific,+human] for object agreement in Romanian, while others involve disjunctions of the Swahili and Palauan type. Aissen proposes to employ the cross-product of the animacy scale and the definiteness scale, which gives rise to constraints that involve composite properties, such as *Ob/Hum-Pron, *Ob/Hum-Name, ... *Ob/Inan-NSpec, ordered in a lattice. Thus, case marking in Spanish is accounted for by a ranking in which *Ob/Anim-Spec & * 0 and everything "further up" in the lattice (e.g., *Ob/Hum-Spec & * 0 , *Ob/Anim-Def & * 0 ...) dominates *STRUCc. The scheme of harmonic alignment, then, seems to be most suitable to account for linking splits, since it provides us with a tool for accounting for the micro-variation one finds among the different languages. Although it brings about a considerable proliferation of constraints, all of which must be conceived of as universal, this does not imply much effort for the language learner, because the order among these constraints is conceptually fixed and largely follows from the logical relation of entailment. It is therefore universal, and all the learner has to detect is where *STRUCc is interpolated. Note that as far as the role of case markers in underlying representations is concerned, Aissen (2000: 9) mentions that she considers case morphology not to be present in the input. This implies that on her account, not realising case morphology does not give rise to a MAX violation - a consequence which raises the question of which other condition would require the realisation of linking material. And although Aissen does not touch upon agreement morphology, the same question applies to object agreement, given that its distribution cross-linguistically involves the same kind of splits as object case. The constraint Aissen holds responsible for realising case morphology at all is * 0 C ("star zero case"). To motivate this constraint, she speculates that overt morphological case is due to a "listener-oriented functional principle" (ibid.). Since, however, case and agreement are the designated morphological means of argument linking, I take the information as to which case or which agreement specification is expected to be part of the input. I shall, however, argue that the conceptual advantages of the DOM approach can still be maintained with the slight modifications just mentioned. I shall next address the version developed by Stiebels (2000a,b).

2.3

Stiebels' Markedness Approach

Stiebels (2000a,b) provides an account of argument linking splits which also draws on harmonic alignment, but suggests some revisions of Aissen's (1999) approach. Stiebels' account is coached in the theory of Lexical Decomposition Grammar (henceforth Τ DG); see Wunderlich ( 1997a,b) for the general outline of this approach to argument linking. In LDG, argument positions are defined in terms of the features [±hr] and [±lr]. These features are conceived of as relational, with the interpretation "there is a/no higher argument role", and "there is a/no lower argument role", respectively, and thus

Economy-based splits, constraints and lexical representations

155

refer to the relative depth of embedding of an argument in the decomposition structure of the Semantic Form. This gives rise to the representations in (15). ( 1 5 ) a.

b.

c.

give:

eat:

sleep:

Xz

Xy

λχ

+hr -lr

+hr +lr

-hr +lr

λy

λχ

Xs

+hr -lr

-hr +lr

λχ

Xs

Xs

[ACT(x) & BECOME(POSS(y,z))] (s)

[EAT(x,y)] (s)

[SLEEP(x)] (s)

-hr -lr The highest argument of a (di)transitive verb will be [-hr,+lr], and the lowest [+hr,-lr], whereas the only argument of an intransitive will have no positive specification. Since structural cases are characterised by the same two features, argument linking is treated as feature unification. The specifications that account for the cross-linguistic linking inventory are as in (16). (16) nominative: accusative:

[] [+hr]

ergative: dative:

[+lr] [+hr,+lr]

genitive:

[+hr]/nouns

Every argument role will be linked by the most specific compatible linker of the inventory of the language in question. Thus, the argument of an intransitive verb is linked by the unmarked case, the nominative (in ergative languages often referred to as the absolutive).4 Stiebels (2000b: 113ff) follows Aissen in using the mechanism of harmonic alignment, and incorporates it into LOG. Her account departs, however, from Aissen's in that it uses no longer the relational scale subject > object, the assumption of which entails that the status of these categories as basic linguistic notions is taken for granted, whereas theories on syntax and argument linking other than LFG consider these notions to be cover terms of clusters of structural properties rather than grammatical primitives. Stiebels points out that as long as no reference to the linker itself is made, a restriction such as *Su/3 & * 0 c will be observed by any linker; thus, a subject being marked by accusative case would turn out to be no worse than one being marked by ergative case. Stiebels therefore proposes to align the salience scale instead with a preference hierarchy defined over the linkers, namely [+hr] > [+lr]: (17) a. b.

*[+hr]/Inan » *[+hr]/Anim » *[+hr]/Hum ... *[+lr]/Hum » *[+lr]/Anim » *[+lr]/Inan

What is above the cut-off point of a given language is labelled for short as "D". D is representative of "D-prominent", which means high referential potential, that is, pro4. For more details and exemplary LDG analyses see, for example, the works by Joppen & Wunderlich (1995), Krämer & Wunderlich (1999), Stiebels (1999, 2000b), Wunderlich (1997a,b), and Wunderlich & Lakämper (2000).

156

Albert Ortmann

nominal rather than nominal, definite rather than indefinite, and so forth. Everything below the cut-off point is labelled N, for "N-prominent" (meaning weak referential potential). For objects splits, then, the following constraint hierarchy is relevant: (18) *[+hr]/N » MAX(+hr) » *[+hr]

The interpretation is that realising the linking feature of objects is a higher requirement than saving it, except where the linking feature is associated with a markedness condition - in the case of *[+hr]/N with low salience, for which morphological effort should be avoided. In contrast to Aissen's approach, then, under the markedness approach no reference is made, and in fact can be made, to the grammatical functions su(bject) and ob(ject), but only to the linkers themselves. To account for the conjunctive combination of salience-features in languages such as Spanish and Romanian, and the disjunctive combination in e.g. Palauan and Swahili, respectively, I propose that instead of composite properties as used in the DOM approach, the latter type involves a local conjunction of avoid-linker-constraints with the respective context information. For Swahili, for example, this will ensure agreement with objects which are either animate or definite. For the former type the ranking of both the involved individual constraints higher than MAX will ensure that case marking is saved unless the object is both animate (or human, respectively) and specific. (19) Swahili: Palauan: Spanish: Romanian:

*[+hr]/Inan & *[+hr]/Indef *[+hr]/Anim & *[+hr]/Nspec *[+hr]/Inan, *[+hr]/NSpec *[+hr]/Anim, *[+hr]/NSpec

» » » »

MAX(+hr) MAX(+hr) MAX(+hr) MAX(+hr)

» » » »

*[+hr] *[+hr] *[+hr] *[+hr]

Note that the use of MAX(+hr) instead of * 0 has an immediate consequence concerning the question of the input and of lexical entries (recall the discussion at the end of the previous sub-section). Since object agreement or object case markers have the specification [+hr], and are part of the input, this allows for computing output forms in terms of Correspondence Theory. To summarise thus far, (i) following Aissen (1999, 2000), I implement the role of economy in object linking by harmonic alignment, (ii) As for the scale that is aligned with the salience scale, I follow Stiebels' markedness approach where reference is made to the linkers themselves rather than to grammatical relations, (iii) The interpolation of MAX constraints rather than of * 0 constraints allows for the computation of candidates in terms of Correspondence Theory. In the remainder of the paper, I will address other types of economy-based splits that have not previously been analysed within the concept of harmonic alignment.

Economy-based splits, constraints and lexical

3.

representations

157

Subject agreement splits governed by salience: plurality agreement splits

In Persian, number agreement on the verb is only obligatory with [+animate] subjects, while it is optional with [-animate] subjects (Amin-Madani & Lutz 1972, Mahootian 1997: 135f). Consider (20a) and (20b): (20) Persian (Amin-Madani & Lutz 1972: 329) a. Baradar-an-am nazde-u bud-and. brother-PL-POR. lSG at-3SG.ANIM COP-3PL

b.

'My brothers were with him.' saxe-ha sekast-and / sekast branch-PL crack.PRET-3PL / crack.PRET.3SG

'The branches broke.' Very similar behaviour is displayed by Turkish (Lewis 1967: 246, Kornfilt 1997: 271), as well as by Georgian, where according to Harris (1981) for many speakers number agreement is also restricted to animate subjects, whereas inanimate ones are found with the singular. The singular thus serves as the grammatical default. Consider the contrast of (21a) and (21b): (21) Georgian (Harris 1981: 21) a. knut-eb-i gorav-en. kitten-PL-NOM roll-3PL

'The kittens are rolling.'

b. burt-eb-i

gorav-s.

ball-PL-NOM roll-3SG

'The balls are rolling.'

In this type of split, to which I will refer as "plurality agreement splits", 5 then, verbal subject agreement with respect to plural is only observed with [+animate] subjects. It is obvious that plurality agreement splits are also motivated by economy, in that morphological effort is saved in less explicitly individuated, hence less salient cases. Similar examples are discussed in Smith-Stark (1974), who proposes the "plurality splits" hierarchy in (4c) above. This hierarchy is meant to capture the typological observation that in many languages only nouns of high salience receive a plural marker, or trigger plural agreement (see also chapter 3 of Corbett 2000 on this issue). Unlike with the object linking splits of the previous section, however, the behaviour observed here cannot be traced back to the affinities or differences of prototypical objects and subjects, since here it is the more prototypical (that is, the more harmonic) subjects that are more explicit in their agreement specification. In this respect, the plural agreement splits apparently point into the opposite direction of what the DOM approach as discussed in 2.2 predicts, namely that subject linking splits should be the converse of object linking splits. This prediction follows from the fact that subject and

5. I use this rather complex term in order to avoid confusion with the more general meaning of the notion of "plurality split" as used by Smith-Stark (1974).

158

Albert Ortmann

object are basic notions, ordered with respect to each other by a hierarchy, and from the constraint hierarchy given by harmonic alignment: (22) Avoid constraints for differential subject marking: *Su/NSpec » *Su/Spec » *Su/Def » *Su/PN » *Su/Pro Consequently, Aissen (2000: 27) states that "[the] account predicts that there should be case marking systems in which some subjects are marked, but not all and it predicts that the factors that favor differential subject marking (DSM) will be the mirror image of those that favor DOM." Among the hierarchies that are predicted by the DSM approach is therefore the following: (23) *Su/Inan & * 0 C » *Su/Anim & * 0 C » *Su/Hum & * 0 C What will be expected, then, is that verbal subject agreement with inanimate subjects implies agreement with animate and human subjects. As far as split ergativity is concerned, this prediction is borne out: more articulated marking of subjects, that is ergative case, or ergative agreement morphology, is often found with, say, non-pronominal, or third person subjects, while more prototypical subjects, that is those further up the salience hierarchies, are in the nominative (or "absolutive"), hence in the morphologically unmarked, case. However, as for the plurality agreement splits of Persian, Turkish, Georgian, the more articulated agreement specification is only found with the more salient subjects. In this respect, these languages exhibit the reverse of what one might expect, even though their object marking system fits into the pattern of DOM. Note, however, that the splits observed here do not involve all kinds of morphological subject marking, but rather only agreement with respect to number. Therefore, in order to account for the markedness asymmetry of subject-verb agreement plurality splits, I propose the integration of a further scale, the plurality scale, into the harmonic alignment approach. This scale can be characterised as "a [+pl] agreement specification is more informative than underspecification", which I represent as in (24). 6 (24) AGR[+pl] > AGR[ ] The scale refers to the category feature AGR, which I assume to be inherent to verbs in languages with subject agreement; 7 its value is a set of attribute-value pairs that specify agreement (here: +pl). Harmonic alignment of the plurality scale and the animacy scale (or definiteness scale, respectively) gives rise to the following markedness hierarchies: 6. A question that comes to mind is whether there are similar scales for other agreement categories such as person and gender. I am, however, not aware of any agreement splits that affect these categories but leave number untouched. As to gender, this may have to do with the fact that verbal agreement with respect to this category is typologically much rarer than with respect to plural. As to person, splits appear to be excluded for logical reasons, since dimensions such as animacy and specificity are only relevant for third person. 7. More formally, in order to differentiate subject agreement from object agreement, one would have to specify AGR-S; thus AGR-S[+pl] > AGR-S[ ]. Since, however, there seems to be no risk of confusion in what follows, I shall simply write AGR in the scale and the constraints that refer to it.

Economy-based splits, constraints and lexical representations

159

(25) a. AGR[+pl]/Hum » AGR[+pl]/Anim » AGR[+pl]/Inan b. AGR[ ]/Inan » AGR[ ]/Anim » AGR[ ]/Hum According to the interpretation of the plurality scale, the hierarchies in (25) refer to a verb in an agreement relation with an inanimate, or human, etc., subject noun phrase, rather than to the noun phrase itself. Consequently, plural agreement with a human subject is more harmonic than with others, and lack of plural agreement is more harmonic with inanimate subjects than with others. Reversing the values of the hierarchy yields the following ranking of avoid constraints: (26) a. b.

*AGR[+pl]/Inan » *AGR[+pl]/Anim » *AGR[+pl]/Hum ]/Hum » *AGR[ ]/Anim » * A G R [ ]/Inan

*AGR[

The interpretation of these avoid constraints is that it is more marked to realise plural agreement for referents of low salience than for those of high salience. Likewise, it is more marked to leave out a plural specification of salient referents. Interpolating this constraint hierarchy with the relevant MAX constraint, and with the economy constraint 8 A V O I D A G R ( + P L ) , respectively, yields the rankings in (27). (27) a. b.

*AGR[+pl]/Inan » MAXAGR(+pl) »... » *AGR[+pl]/Hum *AGR[ ]/Hum »... » AV0IDAGR(+pl) » *AGR[ ]/Inan

Language-specific cut-off points of the ranking in (27a) are thus MAXAGR(+pl) being interpolated in the harmonic alignment of the plural and the salience scale. For Persian and Georgian, where plural agreement with inanimate subjects, MAXAGR(+pl) dominates *AGR[+pl]/Anim and further down the scale.

stated by agreement is "saved" everything

(28) Persian, Georgian: *AGR[+pl]/Inan » MAXAGR(+pl) » *AGR[+pl]/Anim » ... With the ranking in (28), verbs in combination with inanimate subjects will bear no plural specification: the optimal form violates only the lower ranked MAxAGR(+pl), rather than *AGR[+pl]/Inan; see tableau (29b). I use the predicate A G G R for an "aggregate" of individuals in order to represent the semantics of plural. To mediate the correlation between the presence of this predicate in the semantic representation and the corresponding morphosyntactic specification that holds for the unmarked case, I assume a default rule AGGR(x) —» x .

8. Note that with respect to plural, *AGR[ ] constraints can be understood as MAXAGR(+P1) constraints, in that the absence of an underlying agreement specification is starred. Note furthermore that owing to the complementary extension of [+pl] ("a plural agreement specification is present") and [ ] ("no plural agreement specification is present") the two ranking types in (27a) and (27b) lead to the same result. My choice in using (27a) in the following is only because it seems more intuitive.

160

Albert

Ortmann

(29) Evaluation of subject agreement in Georgian a. Input: λχ < + ρ 1 > [ROLL(x)] (Dy < + p l > KITTEN (y) & AGGR(y)) *AGR[+pl]/ Inan

MAXAGR(+PO

ι®" knut-eb-i gorav-en [+pl] knut-eb-i gorav-s [ ] b.

*AGR[+pl]/ Anim *

*!

Input: λχ < + ρ 1 > [ROLL(x)] (Dy < + p l > BALL(y) & AGGR(y)) *AGR[+pl]/ Inan

burt-eb-i gorav-en [+pl «s* burt-eb-i gorav-s [ ]

MAXAGR(+PO

*!

*AGR[+pl]/ Anim *

*

Note that Persian and Georgian differ in one crucial regard: whereas in the former inanimate subjects can occur with no agreement suffix at all, the latter uses the singular, or more precisely, the elsewhere form -s. This detail indicates the relevance of the proper specification of the suffix. For languages where agreement is not subject to splits, such as German, one can assume default rules that apply to those forms that paradigmatically contrast with positive specifications. The mechanism of default value assignment for number inflection in German as assumed in Minimalist Morphology (Wunderlich & Fabri 1995) is as follows: nouns and finite verbs that are projected into the syntax are fully specified for number. The base form, paradigmatically contrasting with a form marked for [+pl], has the specification [-pi], where the latter results per default at the lexicon-syntax-interface. This gives rise to the representations in (31) for the German nouns and verbs in example (30). (30) a.

Das Baby schnarch-te. DF.NEUT baby snore-PRET 'The baby snored.'

(31) a. b. c. d.

Baby: Babys: schnarchte·. schnarchten:

[+N,-V] [+N,-V] [-N,+V] [-N,+V];

b. Die Baby-s schnarch-te-n. DF.PL baby-PL snore-PRET-1/3PL 'The babies snored.' λχ < _ ρ 1 > [BABY(x)] λχ < + ρ 1 > [BABY(x) & AGGR(x)] λχ

λχ

[

S N 0 R E ( x )

]

[SNORE(x)]

If agreement is technically conceived of as unification of features, as outlined in such theories as H P S G (Pollard & S a g 1994, chapter 2) and Wunderlich (1994), the specifications given in (31) account for subject-verb agreement: the unification of < - p l > on the verb with < + p l > on the noun fails, hence the ungrammaticality of *Die Babys schnarchte ('the babies snored.SG'). Thus in German, a finite verb form without an explicit plural affix will be [-pi]. For Georgian, by contrast, a default rule that adds [-pi] to the third person singular form gorav-s would give the wrong result for the combination with inanimate plural subjects as in (21b). The implication is that we cannot generally assume negative values to be added per default, but rather have to differentiate between negative specifications and no specification at all, that is underspecifica-

Economy-based splits, constraints and lexical

representations

161

tion. In order to account for the third person verbal affixes of Georgian, I follow an unpublished proposal by Dieter Wunderlich in assuming the specifications for plural -en and for -s. The latter suffix is only specified for person, while the value of number is unspecified, rather than being [-pi] per default. (Note that the feature [±animate] need not be referred to in these specifications, since its effect is implemented by the harmonic alignment ranking in (28).) In contrast to Georgian, no suffix is available for inanimate subjects in Persian, since this language does not mark third person as such. As a consequence, the third person singular verb is a suffixed form in Georgian (gorav-s), whereas it is unsuffixed (sekast) in Persian. Abstracting away from the difference between Georgian and Persian, the general implication is that for a language to display a split as the ones discussed here means that MAX is dominated by at least the highest avoid constraint in the hierarchy, or equivalently, in terms of (27b), if AVOIDAGR dominates at least the lowest explicitness constraint. Such a ranking in combination with the requirement of feature unification has the corollary that the unmarked form (in this case, the singular) is unspecified rather than negatively specified for the morphological feature at issue. With these prerequisites, a simple account can be provided of a split that has often been observed but has not yet been treated in the light of other salience-based plurality splits, namely, the split in the subject agreement of Arabic. In Classical Arabic, plural agreement is restricted by the syntactic position of the argument, as shown in (32): (32) Classical Arabic (Aoun et al. 1994: 197) a. Naam-a /* naam-uu l-?awlaad-u. slept-3SG.M / slept-3PL.M DF-children-NOM

'The children slept.' b. ?al-?awlaad-u * naam-a DF-children-NOM

/ naam-uu.

slept-3SG.M / slept-3PL.M

'The children slept.' At first sight it appears that this split is merely governed by linear order: the pluralform in subject-verb-agreement is only found if the subject precedes the verb, as in (32b). For the reverse order, agreement is only with respect to gender, as in (32a). The economic aspect of the split is, unsurprisingly, to save agreement morphology where it is, so to speak, not considered necessary. It is, however, not immediately obvious why a plural specification is considered necessary in (32a), but not in (32b). That not only some principle of linearity is at work here is clear from the fact that we are also dealing with a second asymmetry. Unlike with the previous examples, the threshold for this asymmetry is neither animacy nor humanness, since even human subjects are not considered sufficiently salient. Rather, the line is drawn between pronominal and non-pronominal: only in the former case is number agreement obligatory with post-verbal subjects; see the contrast of (33) and (32a): (33) Classical Arabic (Aoun et al. 1994: 205) * Naam-a / Naam-uu hum. slept-3SG.M / slept-3PL.M PRON.PL.M

'They slept.'

162

Albert Ortmann

In a second approximation, then, the split can be described as "plural in subject-verb agreement is only with [+pronominal] subjects". Recall that "pronounness" is one of the steps on the definiteness scale proposed by Aissen (2000) (rendered in (10) above). But why is the omission of plural agreement only possible with post-verbal subjects; which other, higher-ranked requirement forces number agreement in clauses with a preverbal subject? The explanation of the two-fold split is to be looked for in the syntactic structure of the preverbal subject construction. According to the analysis of Aoun et al. (1994: 206), the order VSO in (32a) results from movement of the verb from Io to a higher functional projection, called F. The order SVO in (32b), by contrast, requires one further step of movement, namely - according to a speculation of the authors movement of the subject into the "Topic" position, hence into a designated specifier position (or possibly a left-peripheral position in the clause). In other words, the subject leaves its canonical position in Spec,IP and goes to a position that is not governed by the verb, as sketched in (34): (34) [fp [topic subjecti]

[ F [f0 Vj ] b t¡ tj]]].

If this structure is essentially correct, the rationale of the observed asymmetries lies in the requirement that the subject 0-role be saturated in the syntactic domain of the verb. This latter condition is considered axiomatic by most current theories of syntax. It is stated, for example, by the Extended Projection Principle of PPT or the Minimalist Program, as well as by the Completeness condition of LFG. In the theoretical approach that is adopted here, namely LDG-OT, the requirement can be formulated as a MAX constraint: (35) MAX(arg): Structural arguments of the verb are overtly realised in IP. I assume that the form with the full number specification on the verb, naam-uu, is capable of saturating the subject θ-role. In other words, the suffix -uu in (32b) and (33) is 'pronominal' in the sense of licensing a null subject (Arabic being a pro-drop language), while the less specific default form in (32a) is incapable of saturating an argument role. Since, moreover, the affix specification must unify with that of the subject, it must be assumed that in the case of -a we are not dealing with an explicit singular specification [-pi], but rather with number underspecification. I represent this in the lexical entries of the suffixes: (36) a. b.

-a: -uu:

[+pron, ]; λχ [χ]

According to (36b), the pronominal status of the suffix -uu is explicitly stated in its lexical representation, expressed by the identity mapping. This accounts for the fact that leaving out plural agreement is only allowed if the argument position is already saturated; in this case, by the subject-NP in its canonical position Spec,I. In (32b), the subject is moved outside of the domain of the inflected verb (into Spec,F). Therefore, it has to be recoverable within IP by pronominal agreement on the verb. For (32a), which does not involve such movement, it is economic to leave out the "number component" of verbal agreement - except if the subject is pronominal. Under this analysis of the

Economy-based splits, constraints and lexical representations

163

facts, it becomes obvious that we are dealing with a split similar to those observed so far: plural is respected in the subject-verb agreement of Arabic only for subject DPs that are either [+pronominal] or [+topic]. Translating Aoun et al.'s insights into OT, one must assume that MAX(arg) is ranked higher than *AGR[+pl]/NonPron. This is stated in (37): (37) Classical Arabic: MAX(arg) » *AGR[+pl]/NonPron » MAXAGR(+pl) » *AGR[+pl]/Pron The ranking gives rise to the tableau in (38). For VSO order - that is, with the subject in its canonical position - the optimal candidate in the more salient case with a pronominal subject is the one with number agreement. In the less salient case, that is, with a lexical subject as in (38b), it is the one without number agreement. For SVO order, by contrast, that is, the structure with the subject being moved out of the domain of the verb as in (38c), the candidate with number agreement wins even for lexical subjects, since it respects MAX(arg).9 (38) Evaluation of subject agreement in Classical Arabic MAX

(arg) a.

*AGR[+pl]/ M A X A G R *AGR[+pl]/ NonPron Pron (+pi)

[ipnaam-a hum]

* !

«sr [n> naam-uu hum] b.

*

"SF [n> naam-a l-?awlaad-u]

*

[n> naam-uu l-?awlaad-u] c.

«®· ?al-?awlaad-u [n· naam-a] ?al-?awlaad-u [¡ρ naam-uu]

* ! *

* ! *

If the present interpretation of the facts in (32) and (33) is correct, then, we are dealing with a two-fold split in that an economy constraint of the harmonic alignment scale, namely *AGR[+pl]/Pron (and everything further down the scale), is dominated by MAXAGR(+pl). The latter is in turn dominated by the constraints *AGR[+pl]/NonPron and MAX(arg). To summarise the results of this section, plurality agreement splits were shown to call for an addition to the DSM/DOM approach. Under the present proposal, the role of economy is implemented by a plurality scale, which is harmonically aligned with the specificity (or animacy) scale. A consequence of the constraint ranking for the lexical inventory is that the involved "singular" forms are unspecified rather than negatively specified.

9. For two related accounts of an agreement asymmetry of Romanian, which is similar to that of Arabic subject agreement in that it is based on the object's position and focus status, see Ortmann (2002: 150-52) and Gerlach (2001: 209-13).

164

Albert Ortmann

In the next section, I turn to plural splits that in contrast to the previous ones are governed by factors other than salience.

4.

Plural splits governed by redundancy avoidance

4.1

The lack of DP-internal agreement specifications in plural contexts

In languages with overt number inflection, there are two major strategies of nounphrase internal agreement in contexts of plural reference. In Ortmann (2000), these types are referred to as "Type English" and "Type Hungarian", respectively. Type English languages exhibit several overt realisations of plural in the noun phrase. For example, German displays three overt realisations of plural in addition to the numeral, as in die drei große-n Häus-er 'the.PL three large-PL house-PL'. Similarly, the Bantu concord prefixes are well known to involve number information in each token. By contrast, Type Hungarian languages (among others, Persian, Kurdish, Turkish, Quechua, Georgian, as well as the North-East Caucasian languages), allow for only one overt realisation of the concept of plurality within the noun phrase. Consequently, no plural marker is found on the noun if plurality is realised by a numeral or quantifier: (39) a. Hungarian: (egy) hajó (one) ship 'a/one ship' b. Georgian: knut-i kitten-NOM 'kitten'

hajó-k

öt

ship-PL 'ships'

five ship 'five ships'

hajó

sok

hajó

many ship 'many ships'

knut-eb-i

sami knut-i

kitten-PL-NOM 'kittens'

three kitten-NOM 'three kittens'

We are thus dealing with another plural split, since in general nouns are marked for plural, but they are not if plurality is already expressed by a numeral or quantifier. Moreover, in Hungarian, the attributive adjective as it occurs in (40a,b) does not exhibit number agreement although the adjective can in principle have a plural specification, as is clear from its predicative use in (40c). (40) Hungarian a. egy gyors hajó a fast ship 'a fast ship'

b. gyors hajó-k fast ship-PL 'fast ships'

c. A hajó-k gyors-ak. DF ship-PL fast-PL 'The ships are fast.'

Hungarian adjective agreement is thus another instance of the same type of split: in plurality contexts, adjectives are overtly marked for plural unless used attributively with a plural noun as the head of the NP.

Economy-based splits, constraints and lexical

4.2

representations

165

The Economic Number Agreement analysis

The analysis of the above data that is proposed in Ortmann (2000) implements the notion of morphological economy directly.1 Its basic assumptions are (i) unification of agreement features, as in theories of agreement such as HPSG (Pollard & Sag 1994) or Wunderlich (1994), and (ii) an economy constraint according to which number should not be realised redundantly for one and the same DP-referent within a syntactic domain such as noun phrase or clause, stated in (41):11 (41) EC0N0MICPLURAL(DP) : DP-referents receive only one realisation of "plurality" within the DP. The requirement of explicit plural marking, which is in conflict with ECONOMICPLURAL, is conceived of as a MAX constraint. In contrast to MAXAGR(+pl) as used in the previous section, a more general constraint MAX(+PL) is required here that stars any "missing" exponent of plural, also those where no agreement relation obtains, as in the case of, for example, bare nouns. (42) MAX(+PL):

An underlying specification of [+pl] has a correspondent in the output.

In using a constraint of the MAX family I slightly depart from an earlier analysis, and assume as the input a semantic representation that is associated with morphosyntactic feature specifications. 12 Output candidates are surface forms the specifications of which are checked with the input in terms of Correspondence Theory, that is, MAX, DEP and IDENT constraints. I assume ( 4 3 ) to be the ranking for Type Hungarian languages, where MAX(+PL) is dominated by ECONOMICPLURAL. (43) ranking for Type Hungarian: ECONOMICPLURAL » MAX(+PL) I take the requirement of unification of agreement features, like that of other grammatical features, to be part of the GEN function of the grammar. That is, candidates with

10. Note that an account of the data merely in terms of underspecification is not feasible. If the base form of the noun (the singular form) is formally unspecified, this will correctly render the Type Hungarian constructions grammatical. It will, however, fail to explain the ungrammatically of combinations of [+pl] and [+pl], such as plural noun and plural adjective, since unification is just as possible here as in the case of [+pl] and [ ]. 11. This rather specific constraint is actually a shorthand notation for the effect of two more general constraints that are defined in Ortmann (2002), namely EXPRESSPLURALITY ("The semantic concept of plurality must be visible in the output.") and the multiply violable *PL(DP) ("Avoid the realisation of [+pl] in the DP."). The ranking of these two constraints above MAX(+PL) ensures that there is at most one indication of plurality. 12. In Ortmann (2000) it is semantic representations without morphological features that qualify as inputs. The requirement of explicit plural marking is conceived of as a constraint MAP that stars any improper mapping of semantic properties to morphological specifications. This conception, however, does not allow for a computing of candidates in terms of Correspondence Theory and is therefore no longer taken into account.

166

Albert Ortmann

conflicting feature specifications such as [+pl] combined with [-pi] will not be generated, hence not be considered in the following. (Alternatively, one may think of feature compatibility as an undominated constraint UNIFICATION, as assumed in Ortmann 2000, 2002.) Capturing the noun plural marking split of Type Hungarian languages: All numerals >1 are, by virtue of their cardinality value, assumed to contribute the semantics of "plurality", expressed by the semantic predicate AGGR(egate). As a consequence, they induce violations of ECONOMICPLURAL when combined with a plural noun. According to the ranking in (43), the evaluation of candidates for the combination of noun and numeral is as shown in (44) for Hungarian: (44) Evaluation of number of nouns in combination with numerals in Hungarian: Input: λχ [SHIP(x) & AGGR(x) & CARD(x) = 5] ECONOMICPLURAL

a.

öt

hajó-k [+pl]

f i v e ship-PL

b.

«s" öt

hajó

MAX(+PL)

* !

*

f i v e ship

The winning candidate (44b) does not realise the morphological specification [+pl] on the noun, hence violates the MAX constraint, but is still optimal since it satisfies the higher-ranked constraint. Type Hungarian numerals are thus correctly predicted to combine with singular nouns. 3 Capturing the adjective plural marking split of Hungarian: As for the combination of plural nouns and adjectives as illustrated in (40b), the input is the semantic representation of the conjunction of the predicates SHIP, FAST, and AGGR associated with the morphological specification that is derived from the semantics. In the presence of the semantic predicate AGGR, the specification [+pl] will be part of the input for any element that involves a paradigm with the respective feature contrast. Consider the evaluation of the relevant output candidates in (45).

13. It cannot be derived from the ranking whether Type Hungarian numerals, as well as quantifiers such as sok 'many', bear a specification of the feature [iplural]; neither is it evident from the ranking whether singular nouns are negatively specified or underspecified. It seems, however, plausible to assume that due to the paradigmatic contrast with [+pl] noun forms, singular nouns are [-pi], since the ranking does not preclude this specification to come about by default. This assumption entails that numerals can at least bear no positive specification for reasons of feature unification. The specification [-pi] for noun phrases such as öt hajó is also motivated by the fact that they obligatorily trigger singular verbal agreement (Ortmann 2000, 2002).

Economy-based splits, constraints and lexical

167

representations

(45) Evaluation of attributive adjective-noun-agreement in Hungarian: Input: λχ [SHIP(x) & FAST(x) & AGGR(x)] ECONOMIC M A X ( + F ) / MAX(+PL) HEAD PLURAL

a. b. c. d.

gyors-ak[+pi]

hajó-k[+pi]

fast-PL

ship-PL

gyors-ak[+pi] is- gyors[ ] gyors[ ]

*!

hajó[ ] hajó-k[+pi] hajó[ ]

*!

*

*!

**

*

The presence of plural markers on both the adjective and the noun in (45a) violates the high-ranked ECONOMICPLURAL. In (45b) and (45c) the adjective and the noun, respectively, are left unspecified for plural in the output, hence these candidates violate the low-ranked MAX constraint. The choice between underspecified adjective and underspecified noun is made by the constraint MAX(+F)/HEAD ("Input specifications have a correspondent in the output on the head of the phrase")· Note that the analysis also captures the fact illustrated in (40e) above that unlike in its attributive use, the adjective is obligatorily inflected for plural in its predicative use. Since ECONOMICPLURAL is trivially satisfied in DP-external environments, a candidate with plural on the adjective will be preferred over a candidate without a plural suffix, which would invoke a fatal violation of MAX(+PL). That is, the ranking does not enforce different entries for adjectives in attributive and predicative uses, in spite of their different behaviour in plural agreement. Capturing a split in Kurdish subject-verb agreement: Besides the noun plural marking split of other Type Hungarian languages, in Kurdish the subject noun phrase is, according to Akrawy (1982), not marked for plural at all if the verb bears a plural specification of the subject. In other words, if the verb displays plural agreement as in (46d), redundant marking of plural on the noun is avoided. (46) Kurdish (Akrawy 1982) a.

mirov man '(the) man'

b.

c.

Mirov hat.

d. Mirov(*-an)

man

come.PAST.SG

'The man came'

mirov-an man-PL '(the) men' man(-PL)

hat-in. come.PAST-3PL

'The men came.'

This split can be characterised by stating that nouns are marked for plural, but only if they are not in an agreement relation with a verb that signals plural. In Kurdish, then, the domain of economic plural agreement is the clause rather than only DP. I assume EcONOMICPLURAL(Clause) to be the relevant instantiation of ECONOMICPLURAL. I furthermore assume the ranking MAXAGR(+pl) » *AGR[pl], which yields the following evaluation of candidates for verbal agreement:

168

Albert Ortmann

(47) Evaluation of subject-verb agreement in Kurdish:

a. b. c. d.

Mirov-an[+pi] hat-in[+pi]. Mirov-an[+p\] hat[ ]. US' Mirov[ ] hat-in[+p\]. Mirov[ ] hat[ ].

~

Ρ

MAX(+PL)

o

*AGR[+pl]

ECONOMIC PLURAL (CLAUSE)

I n p u t : λ χ < + ρ 1 > [PERF (COME(x))] ( D y < + p l > MAN(y) & AGGR(y))

*

*!

*

*! *

*!

* **

Plural morphology on both the noun and the verb violates ECONOMICPLURAL(Clause). (47b) and (47d) invoke a violation of MAXÄGR(+pl), which penalises lack of agreement, hence (47c) is the optimal candidate. A consequence of this evaluation for the representation of Kurdish nouns is that, unlike with nouns of other Type Hungarian languages, the base form is unspecified for number rather than being specified [-pi]: (48) a.

singular noun:

b. plural noun:

mirov:

[+N,-V];

λ χ < > [MAN(X)]

mirov-an:

[+N, -V];

λ χ < + ρ 1 > [MAN(x) & AGGR(x)]

In the account presented in this section, then, the role of economy in the distribution of plural markers is implemented by a constraint ECONOMICPLURAL. It was furthermore shown that articulated feature representations for lexical entries are indispensable in order to account for splits based on redundancy avoidance found in the DP of languages such as Hungarian, as well as in Kurdish verbal agreement. A theoretical implication of the present analysis is that properties of the lexicon, such as the underspecification of Hungarian adjectives, of Kurdish singular nouns, and of "differential" object markers like those of Swahili, are enforced by the grammar, that is, by the constraint ranking - a result that is in harmony with current works in OT since Grimshaw (1997), advocated most radically in Grimshaw (2001).

5.

A conspiracy of splits: Hungarian possessor agreement

With the prerequisites of the preceding sections, we can also account for a particularly intricate set of data, namely Hungarian possessor agreement. Similar to verbal agreement in Arabic, two different splits have to be disentangled which conspire.

169

Economy-based splits, constraints and lexical representations

5.1

The facts

In Hungarian, a noun can be suffixed by a possessor marker, thus exhibiting agreement with an overt or non-overt possessor with respect to person: (49) a.

c.

az én

kalap-om

b. a te

kalap-od

DF PRON.lSG hat-POR. lSG

DF PRON.2SG hat-POR.2SG

'my hat'

'your hat'

a fiú-k kalap-ja-i DF boy-PL hat-POR.3-PL 'the boys' hats'

As far as possessor agreement with respect to number is concerned, the facts are widely captured by the economic plural analysis of the preceding section. In particular, if the possessor refers to a noun phrase that is modified by a numeral, the possessor suffix on the noun is singular; for discussion and explicit analysis see Ortmann (2000). An aspect, however, that still calls for a more detailed analysis concerns the difference between lexical and pronominal possessors. Consider first the data in (50). (50) a.

A nagynénij iil DF aunt

a 0¡

ház-á¡-ban.

sit.3SG DF PRON house-Por.3SG-lNESSlVE

'The aunt is sitting in her house.' b. A nagynénij iil DF aunt

a 0¡

szék-é¡-n.

sit.3SG DF PRON chair-POR.3SG-SUPERESSIVE

'The aunt is sitting on her chair.' c.

A nagynéni-kj ül-nek a 0 ¡ DF aunt-PL

ház-uk.-ban.

sit-3PL DF PRON house-POR.3PL-INESSIVE

'The aunts are sitting in their house(s).' d. A nagynéni-kj ül-nek a 0 ¡ DF aunt-PL

szék-íikj-en.

sit-3PL DF PRON chair-POR.3PL-SUPERESSlVE

'The aunts are sitting on their chairs.' For better illustration, I note non-overt pronominal possessors that are coreferential, and hence coindexed, with the subject of the sentence in each of the examples in (50). (I shall later argue that it is in fact the possessor agreement suffixes themselves that can also be pronominal.) In (50c,d), the noun bears a possessor suffix that signals plural and thus accords with the number of the antecedent. However, in Hungarian, such a plural specification in possessor agreement is only found with pronominal, not with lexical possessors. 14 This can be shown by the following contrastive pairs, where (51a,b) involve pronominal possessors, and (51c,d) lexical possessors.

14. Thanks to Péter Siptár for pointing this out to me.

Albert Ortmann

170

(51) a.

az ö

b.

ház-a

'her house' c.

a nagynéni ház-a DF aunt

az ö

ház-uk

DF PRON.3SG house-POR.3PL 'their house'

DF PRON.3SG house-POR.3SG d.

house-POR.3SG

'the aunt's house'

a nagynéni-k ház-a DF aunt-PL house-POR.3SG 'the aunts' house'

That the plural of the possessor is not marked in combination with lexical possessors can be seen from the invariant use of the singular form in (51c) and (5 Id), where the form házuk would give rise to ungrammaticality. Plural is only marked in combination with pronominal possessors, as is clear from the different agreement specification of (51a) vs. (51b). The same difference is also displayed by the contrast of (50a,b) and (50c,d), showing that the identification of an antecedent outside of the DP follows the same agreement pattern as with pronominal antecedents within the DP. The former constellation is therefore often assumed to involve a zero pronoun, as indicated in the examples in (50) for the sake of better illustration. The point to be noted about splits here is that possessor agreement within the DP in Hungarian always observes the proper specification of person, whereas number is only specified in the presence of an antecedent to be identified outside the DP. We therefore encounter another plurality agreement split of the type discussed in section 3, in that only pronominal possessors are marked by number agreement. What complicates the matter further is that even in cases of plural reference like (5 lb), the pronoun we encounter is that of third person singular rather than of plural, the third person plural pronoun otherwise being ok in Hungarian: (52) a.

ö

sört

isz-ik

PRON.3SG beer.ACC drink-3SG

'S/he drinks beer.' c.

A nagynéni sört isz-ik DF aunt beer.ACC drink-3SG 'The aunt is drinking beer.'

b.

ök

sort

isz-nak.

PRON.3PL beer.ACC drink-3PL

'They drink beer.' d. A nagynéni-k sort isz-nak. DF aunt-PL beer.ACC drink-3PL 'The aunts are drinking beer.'

This raises the question of how to represent the pronoun δ as it occurs in plural reference: should it be treated as a [-pi] pronoun? This seems to make little sense for reasons of feature unification, since we must assume an agreement constellation with the lexical plural possessor. Note furthermore that the observed behaviour is confined to possessor agreement, rather than being a general property of agreement in Hungarian: in verbal agreement, a lexical subject does trigger plural agreement, as witnessed by (52d). Alternatively, if o is taken to be unspecified for number, the question must be answered what excludes the plural form 5k in (51b), and at the same time requires it in clausal contexts such as (52b).

Economy-based splits, constraints and lexical representations

5.2

171

Why an account solely relying on lexical entries fails

For an account of the Hungarian possessor agreement split that relies solely on the lexical representations of the involved categories, one might assume specifications of the involved suffixes like the ones in (53). (53) a. -a\ b. -uk:

[AGR-Por ] [AGR-Por ]

One would furthermore have to assume a separate entry for possessor contexts, to be distinguished from o in clausal contexts such as (52a). The pronoun in its use in (51a,b) would be underspecified for number as in (54a), as opposed to (54c), which represents the use of ο as a non-possessive personal pronoun: (54) a. b. c. d.

δ: 0: o: ôt.

[+pron,+poss, -anaph, < - l , - 2 > ] [+pron,+poss,+anaph, ] [+pron,-poss, ] [+pron,-poss, ]

According to these representations, a possessed noun like ház-a will be able to combine with either pronominal or lexical possessors. Lexical DPs, being specified as [-pron], will not be compatible with the specification of -uk as in (53b), hence can only combine with -a. An additional lexical entry for δ is assumed for the use of this pronoun in possessor constructions, hence the construction-specific feature [poss]; 0 as the "anaphoric" variant would account for the data in (50). Underspecification for number allows for the combination with either a singular or a plural possessor specification on the noun. Under such an account, the Hungarian possessor agreement split is stated in terms of complex lexical entries of both possessor suffixes and personal pronouns. The disadvantages of such a solution are obvious. First, it offers no principled account of why the asymmetry concerns only plural, and not person as well. No connection would be drawn to the economic plural of Hungarian as introduced in the previous section either. Moreover, the combination of additional personal pronouns for possessor contexts, an ad hoc agreement feature [pron] for -uk, as well as the constructionspecific [+poss] appear to be means of "brute force" rather than of a principled explanation. In the following, I propose to analyse the split in terms of the already established notion of plural economy in the noun phrase, in combination with explicit entries of the involved morphemes.

5.3

Analysis: constraints ranking determines lexical entries

Since plural in Hungarian possessor agreement is only found with [+pronominal], it seems natural to relate the facts to those of other plurality agreement splits. In the analysis to be developed now I explain the data under discussion as a combination of a salience-based plurality agreement split as discussed in section 3 and an economic

172

Albert Ortmann

plural split as discussed in the last section. The ingredients of my analysis are the following: (i) a constraint ranking of the plurality agreement split type, derived by harmonic alignment, and (ii) the redundancy avoidance constraint ECONOMICPLURAL(DP); note that the possessor phrase and its agreement specification on the possessum noun involve two plural specifications within one DP that refer to the same entity, hence the context of ECONOMICPLURAL(DP) is met. The facts fall out from the following partial ranking, where the faithfulness constraint MAXAGR(+pl) is interpolated in the economy constraints derived from the harmonic alignment of the plurality scale and the definiteness scale: *AGR[+pl]/NonPron » MAXAGR(+P1), ECONOMICPLURAL » MAX(+PL),

*AGR[+pl]/Pron. T h e

relevant

candidates, that is, the possible combinations of plural and unmarked for both pronominal and lexical possessors, are thus exclusively evaluated by the constraints that have been employed in the two preceding sections; consider the tableaux in (55). 15 (55) Evaluation of Hungarian possessor agreement: a. Input: Dx [HOUSE(x) & POSS(y,x)]

s+ )rs S » υβ2 l-J< o S ζ Β "è β O O Í + U < fc !__)

Λ.

Oh

*

az δ [ ]

ház-á [ ]

DF PRON.3

house-POR.3SG

az ok [+pl] ház-á [ ] DF PRON.3PLhouse-POR.3SG

«3" az δ [ ] DF PRON.3

*

* er ^ .S Su 05 2 + O O-i

< *

tì

*!

**

*!

*

ház-uk [+pl]

*

*

house-POR.3PL

az ôk [+pl] ház-uk [+pl] DF PRON.3PLhouse-POR.3PL

*!

*

15. Strictly speaking, the avoid constraints that refer to plural agreement still call for a context restriction to possessor agreement. This is because Hungarian subject agreement exhibits no comparable split, see (52d) above; hence we are dealing with a peculiarity of possessor agreement. Such a restriction may formally be noted as *AGR-PoR[+pl]/..., as opposed to *AGR-S[+pl]/... as used for the constraints in section 3.

Economy-based splits, constraints and lexical

173

representations

a nagynéni [ ] ház-á [ ] DF aunt.SG house-POR.3SG «τ a nagynéni-k [+pl] ház-á [ ] DF aunt-PL house-POR.3SG a nagynéni [ ] ház-uk [+pl] DF aunt.SG house-POR3PL a nagynéni-k [+pl] ház-uk[+pi] DF aunt-PL house-POR3.PL

Υ

Sî

υ ω

*

OH

S A

EU

*

**I

*

*

*

*!

*!

^

§ d χ

*AGR[+pl Pron

*AGR[+pl NonPron

b. Input: Dx [HOUSE(x) & POSS(DyAUNTS(y), x) & AGGR(y)]

*

Recall that the output consists of surface forms the morphological specifications of which are compared with those of the input forms. MAX(+PL) stars all missing plural specifications in the output, that is, those lacking on the possessor itself as well as in the agreement morphology; consequently, any candidate with no plural specification at all is excluded. For pronominal possessors as in (55a), the candidates that exhibit the unmarked form on the noun, hence fail to show overt agreement with a pronoun, are excluded by virtue of violating MAXAGR(+pl). As for the candidates that show a plural specification on the possessed noun, (55a) also explains the so far problematic fact that the plural form 5k of the pronoun is ungrammatical: since both the pronoun and the agreement suffix on the noun specify the same individual, plural on both gives rise to a violation of ECONOMICPLURAL(DP), which was independently shown to dominate MAX(+PL). Consequently, an unmarked pronoun is better than a plural pronoun; whereas the reverse pattern, the candidate with plural on the pronoun and the unmarked form on the noun, violates MAXAGR(+pl). The plurality agreement split analysis developed in section 3 thus makes the correct predictions for Hungarian, too: it is not the lack of plural on the pronoun itself that gives rise to a violation of MAXAGR(+pl), but rather the lack of plural agreement with it. It is in this sense that the plurality agreement split and the economic plural split conspire in Hungarian possessor agreement. The candidate evaluation of agreement with lexical possessors in tableau (55b) shows that plural agreement on the noun will violate *AGR[+pl]/NonPron, and, depending on the specification of the possessor, also ECONOMICPLURAL(DP). The ranking of the former constraint with respect to MAXAGR(+pl) and MAX(+PL) correctly predicts that the noun exhibits no plural agreement when combined with a lexical possessor. From the data at issue, no relative ranking of the constraints MAXAGR(+pl) and ECONOMICPLURAL is obvious, as well as of *AGR[+pl]/Pron and MAX(+PL). The constraint ranking that could be established is therefore the following:

174 (56)

Albert Ortmann *AGR[+pl]/NonPron » MAXAGR(+pl), *AGR[+pl]/Pron

ECONOMICPLURAL(DP) » M A X ( + P L ) ,

Note that under my proposal, the singular pronoun 5 is formally underspecified for number as a result of the specific ranking of Hungarian: both MAXAGR(+pl) and E C O N O M I C P L U R A L ( D P ) are ranked higher than M A X ( + P L ) , hence the non-plural form δ is favoured over the more specific ök. In order to be compatible with the plural agreement of the possessor, it is underspecified in spite of the paradigmatic contrast to [+pl] ôk.16 (57) a.

o:

[+pron]; λχ AGR[ ], which is harmonically aligned with the salience scale(s). The third type is that of anti-redundancy splits, under which underlying plural specifications of nouns or modifiers occur only if there is no other indication of plurality in a relevant domain (usually, the DP). It may happen that two asymmetries conspire, as was shown to be the case with a plurality agreement split and an anti-redundancy split in Hungarian possessor agreement. I have proposed to account for this complex case by the interaction of ECONOMICPLURAL and the constraints derived from the plurality scale by harmonic alignment. Similarly, Arabic subject-verb agreement could be explained by the interaction of the latter with the syntactic explicitness constraint MAX(arg). A crucial implication of the analyses is that for the lexical inventory, there is a threeway representational distinction of positive specifications of inflectional features, unspecified features and minus feature values. Some categories receive minus values for certain inflectional features (usually per default), while others (such as the third singular pronoun and possessor suffix of Hungarian) are stored as underspecified ([ ]) with respect to the same features. This result is in line with the by now widely held view that cross-linguistic variation with respect to the lexical inventory is enforced by the grammar. In the case under discussion, lexical properties are explained as a consequence of the language-specific ranking of economy with respect to faithfulness constraints.

176

Albert Ortmann

References Aissen, Judith (1999). Markedness and subject choice in Optimality Theory. Natural Language and Linguistic Theory 17, 673-711. —(2000). Differential Object Marking: Iconicity vs. Economy. Ms., UC Santa Cruz. Akrawy, F.R. (1982). Standard Kurdish Grammar. (Published by the author). Amin-Madani, Sadegh and Dorothea Lutz (1972). Persische Grammatik. Heidelberg: Groos. Aoun, Joseph, Elabbas Benmamoun, and Dominique Sportiche (1994). Agreement, Word Order, and Conjunction in Some Varieties of Arabic. Linguistic Inquiry 25, 195-220. Beckmann, Jill, Laura Walsh Dickey, and Suzanne Urbanczyk (eds.)(1995). Papers in Optimality Theory. Amherst: GSLA (= University of Massachusetts Occasional Papers in Linguistics 18). Corbett, Greville G. (2000). Number. Cambridge: Cambridge University Press. Fabri, Ray (1993). Kongruenz und die Grammatik des Maltesischen. Tübingen: Niemeyer. Gerlach, Birgit (2001). Clitics between Syntax and Lexicon. Doctoral Dissertation, Universität Düsseldorf. Grimshaw, Jane (1997). Projection, Heads and Optimality. Linguistic Inquiry 28, 373422. —(2001). Optimial Clitic Positions and the Lexicon in Romance Clitic Systems. In Optimality-Theoretic Syntax, Geraldine Legendre, Jane Grimshaw, and Sten Vikner (eds.), 205-240. Cambridge, ΜΑ: ΜΓΓ Press. Harris, Alice C. (1981). Georgian Syntax. A study in relational grammar. Cambridge: Cambridge University Press. Joppen, Sandra and Dieter Wunderlich (1995). Argument linking in Basque. Lingua 97, 123-169. Kibrik, Aleksandr E. (1994). Archi. In The indigenous languages of the Caucasus, Vol. 4: North East Caucasian languages 2, The three Nakh languages and six minor Lezgian languages. Rieks Smeets (ed.), 297-367. Delmar, NY: Caravan. Kornfilt, Jaklin (1997). Turkish. Lodon: Routledge Krämer, Martin and Dieter Wunderlich (1999). Transitivity alternations in Yucatec, and the correlation between aspect and argument roles. Linguistics 37, 4 3 1 ^ 7 9 . Lewis, G.L. (1967) Turkish Grammar. Oxford: Clarendon. Mahootian, Shahrzad (1997). Persian. London: Routledge McCarthy, John and Alan Prince (1995). Faithfulness and Reduplicative Identity. In Jill Beckmann, Laura Walsh Dickey, and Suzanne Urbanczyk (eds.), 249-384. Ortmann, Albert (1998). The Role of [±animate] in Inflection. In Models of Inflection. Ray Fabri, Albert Ortmann, and Teresa Parodi (eds.), 60-84. Tübingen: Niemeyer. —(2000). Where plural refuses to agree: feature unification and morphological economy. Acta Linguistica Hungarica 47, 249-288.

Economy-based splits, constraints and lexical representations

111

—(2002). Kategorien des Nomens: Schnittstellen und Ökonomie. Tübingen: Niemeyer. Plank, Frans (1980). Encoding grammatical relations: acceptable and unacceptable non-distinctions. In Historical Morphology. Fisiak, Jacek (ed.), 289-325. The Hague: Mouton. Pollard, Carl and Ivan A. Sag (1994). Head-driven Phrase Structure Grammar. Chicago: The University of Chicago Press. Prince, Alan and Paul Smolensky (1993). Optimality Theory: Constraint Interaction in Generative Grammar. Ms., Rutgers University, New Brunswick, and University of Colorado, Boulder. Siewierska, Anna (1988). Word Order Rules. London: Croom Helm. Silverstein, Michael (1976). Hierarchy of features and ergativity. In Grammatical Categories in Australian Languages. R.M.W. Dixon (ed.), 112-171. New Jersey: Humanities Press. Smith-Stark, T. Cedric (1974). The plurality split. Chicago Linguistics Society 10, 657-671. Stiebels, Barbara (1999). Noun-verb symmetries in Nahuatl nominalizations. Natural Language and Linguistic Theory 17, 783-836. —(2000a). Linker inventories, linking splits and lexical economy. In Lexicon in Focus. Barbara Stiebels and Dieter Wunderlich (eds.), 211-245. Berlin: Akademie-Verlag. —(2000b). Typologie des Argumentlinkings: Ökonomie und Expressivität. Habilitation thesis, Universität Düsseldorf. To appear in Studia Grammatica. Berlin: Akademie Verlag Woolford, Ellen (1995) Object Agreement in Paluan: Specificity, Humanness, Economy and Optimality. In Jill Beckman, Suzanne Urbanczyk and Laura Walsh (eds.), 655-700. Wunderlich, Dieter (1994). Towards a lexicon-based theory of agreement. Theoretical Linguistics 20, 1-35. —(1997a). Cause and the Structure of Verbs. Linguistic Inquiry 28, 27-68. —(1997b). Argument Extension by Lexical Adjunction. Journal of Semantics 14, 94142. —; and Ray Fabri (1995). Minimalist Morphology: An Approach to Inflection. Zeitschrift für Sprachwissenschaft 14, 236-294. —; and Renate Lakämper (2000). On the interaction of structural and semantic case. Lingua 111, 377-418. Special issue On the Effects of Morphological Case, ed. by Helen de Hoop et al.

Disjoint reference and the typology of pronouns Paul Kiparsky

1. 1.1

Issues in Binding Theory Accounting for di sj oint reference

Obviation versus Blocking. Two approaches to the distribution of anaphors and pronominale have been explored in Binding Theory. The OBVIATION approach, originating in Lasnik (1976) and extensively developed in the GB tradition, posits autonomous disjoint reference principles which directly filter out illicit coindexations in certain structural domains. The BLOCKING approach treats disjoint reference derivatively, by making anaphors obligatory under coreference in the binding domain, and invoking a syntactic or pragmatic principle that forces disjoint reference pronominals in the "elsewhere" case. Obviation. Chomsky (1981) equated the domains of both anaphoric binding (principle A) and of disjoint reference (principle B) and defined this domain in terms of government. If principles A and Β hold in the same domain, anaphors and pronominals 1. My interest in reflexive pronouns comes partly from historical syntax and partly from the Case theory that Dieter Wunderlich and I have been thinking about since 1991. The first version of this paper was written that year, and I was fortunate to be able to discuss it with Dieter at the time, who made insightful suggestions especially about the Swedish material. I then set it aside for some years, realizing that many of the things I was trying to do were being done in a more sophisticated way by Reinhart & Reuland and by Burzio. Still, my conclusions differed from theirs on some points, and so I returned to the paper in 1996, adding the OT analysis, which was presented at an OT syntax conference at Stanford, and revising it once more for this publication in honor of Dieter Wunderlich. I am grateful to Cleo Condoravdi for her detailed comments on several drafts of this paper and for her advice and encouragement over the years. Special thanks also to Annie Zaenen and Peter Sells for their guidance, and to Ekkehard König and Tomas Riad for the interest they have taken in this work. Many people have been generous with their time and linguistic expertise: Per-Kristian Halvorsen, Elisabet Engdahl, Tan Fu, K.P. Mohanan, Shuichi Yatabe, Hinrich Schuetze, Ivan Sag, K.S. Yadurajan, Smita Joshi, Kalpana Bharadwaj, Will Leben, Hadar Shem-Tov, Thora Árnadottir, Christopher Culy, Karl Zimmer, Giiven Giizeldere, Lauri Karttunen, Höskuldur Thráinsson, Young-mee Cho, Ki-Sun Hong, William Poser, and Makoto Kanazawa. The responsibility for all errors is mine.

180

Paul Kiparsky

must be in complementary distribution. This theory faces the conceptual problem that the complementarity lacks a deeper rationale, and the empirical problem that the predicted complementarity systematically fails to hold in certain structural configurations. Subsequent versions of Binding Theory either tried to exploit the partial complementarity so as to achieve a simpler and theoretically more attractive formulation of the theory, or to address the empirical problem of overlapping distribution by having principles A and Β apply in different domains. These two endeavors proved hard to reconcile. A first attempt to distinguish the domains of principle A and Β was to define "governing category" differently for anaphors and pronominals (Huang 1983, Chomsky 1986: 169). Chomsky's redefinition was couched in partly θ-theoretic terms. He introduced the notion of a COMPLETE FUNCTIONAL COMPLEX (CFC), defined as a maximal projection in which a predicate's complements and subject (in effect, all its θ-roles) are realized, and reformulates Principles A and Β to the following effect: (1)

An anaphor/pronominal must be bound/free in the least CFC containing a lexical governor in which it could be bound/free.

The idea behind (1) is that a pronominal can always satisfy Binding Theory (BT) by being free in the minimal NP that contains it, but an anaphor, because it must be bound, can satisfy Binding Theory only in the minimal CFC that contains a potential antecedent for it (one to which the anaphor could be bound). It follows that the distribution of pronominals and anaphors can overlap in the subject position of NPs. But this resolves just a small part of the overlapping distribution problem. That a domain defined in terms of a predicate's arguments (akin to the CFC) is specifically the domain of disjoint reference (Principle Β effects) was argued by Hellan (1983, 1988), Sells (1986), and Farmer & Harnish (1987), who each proposed some version of the principle stated in (2), sometimes called OBVIATION, a term borrowed from Algonquian grammar. (2)

OBVIATION:

Coarguments have disjoint reference.

A second key idea is that disjoint reference is a principle of semantic interpretation which requires certain types of arguments of multi-place predicates to be distinct.2 (Keenan 1974, Bach & Partee 1980, Pollard & Sag 1983, Sells 1991, Pollard & Sag 1992, Reinhart & Reuland 1993). Reinhart & Reuland explicitly argue that principle A is defined on syntactic predicates and principle Β on semantic predicates. The proposal developed below adopts both these ideas. Obviation will be argued to be a universal violable constraint interacting with other constraints in the ranked constraint systems that define the binding patterns of languages. Blocking. The second approach to disjoint reference replaces principle Β by a BLOCKING constraint that makes anaphors obligatory in their binding domain under the

2.

Semantic disjoint reference also plays a key role in the theory of reciprocals developed by Heim et al. (1991), though they do not unify it with the disjoint reference phenomena seen in pronouns and reflexives.

Disjoint reference and the typology of pronouns

181

appropriate conditions, superseding the coreferential interpretation of pronominale in that domain. Blocking can be seen either as a grammatical condition applying to syntactic representations (Bouchard 1983, Yadurajan 1987, Burzio 1996, 1998), or as an extragrammatical (pragmatic) condition involving generalized conversational implicature, informativeness, hearer strategies for recovering speaker meaning, etc. (Dowty 1980, Reinhart 1983a,b, Farmer & Harnish 1987, Levinson 1987, 1991, 2000).3 Combining Blocking and Obviation. The main thesis of my paper is that Blocking and Obviation are not competing accounts of disjoint reference, but that both - appropriately formulated - constitute universal constraints that interact with each other, and with other constraints, in accord with partly language-specific rankings. I establish their independence by showing that one applies to a syntactic representation, the other to a semantic interpretation, that they apply to different classes of elements, and that they apply in different domains. Conceptually, they differ as well. Blocking is related to the "Elsewhere" principle that governs morphological expression; following Burzio's work I treat it as an Economy principle. Obviation, on the other hand, is a constraint on the licensing of coreferential coarguments which is specific to binding theory.

1.2

The typology of pronouns

With the advent of OT, grammatical theory has become inseparable from typology. In a pioneering typology of reflexives and reciprocals, Faltz (1977) showed that their binding properties vary along at least two dimensions: the size of the domain within which they must be bound, and the nature of the antecedent in the clausal domain. He further noted that, since languages may have several anaphors that differ in these respects, the binding domain and the antecedent requirement is a lexical property of individual anaphors, rather than simply a syntactic property of the language at large. These conclusions were confirmed by Yang (1983), and their consequences were explored in the more general framework of the principles-and-parameters approach to syntax by Wexler & Manzini (1987). Independently, Everaert (1986) and Thráinsson (1991) proposed to specify the binding properties of anaphors by systems of crossclassifying features. Hellan (1983), working from Norwegian, arrived at still another featural parametrization (greatly elaborated in Hellan 1988), which was further developed by Bresnan et al. (1985), whose ideas were in turn extended and formalized by Dalrymple (1993). In the spirit of these studies I argue that the typology of pronouns emerges from two cross-classifying properties: whether they are obviative or not, i.e. whether they can license a coreferential coargument, and what their antecedent domain is. The antecedent domain is defined by a hierarchy of constraints which impose successively inclu3. Distant antecedents of this idea include Chomsky's "Avoid Pronoun" Principle, and the rules of "obligatory reflexivization" in earlier transformational treatments (Lees & Klima 1963, Kuno 1987).

182

Paul Kiparsky

sive requirements on the structural relation between the pronoun and its antecedent. The factorial typology predicts a class of pronominal elements which have hitherto not been recognized in Binding Theory. I show that this predicted class, referentially dependent pronominals, is cross-linguistically well attested, and in fact is instantiated in one pronoun of modern English. The constraint hierarchy defines five successively more inclusive antecedent domains, each characterizing a class of pronouns. These five types of pronouns are cross-classified for obviation. The resulting inventory of possible pronouns is richer than what traditional Binding Theory countenances, but offers a better approximation to the attested diversity of anaphoric systems. This simple typology appears to match the range of known pronouns quite well.4

1.3

Gaps and optionality

"One form, one meaning" may be the grammatical ideal, but deviations from it in both directions are ubiquitous in grammar, anaphora not excepted. Contextual neutralization of semantic distinctions is a normal part of pronominal expression, and so are multiple realizations of the same meaning, such as "optional" long-distance anaphora. Such complementarity failures are potentially problematic for any theory that relies on Blocking, but OT makes available interesting new approaches for accounting for them. To deal with gaps I will make use of the idea suggested by Legendre et al. (1996) that markedness constraints can enforce gaps, violating a lower-ranked faithfulness constraint that prohibits them. Optionality, in the form of variation between pronouns and reflexives in ostensibly identical contexts, is every bit as common in anaphora as gaps are. However, not all apparent variation can be taken at face value. In fact, I will argue that many apparent cases of "optional reflexivization" that have been cited again and again in the literature are not optional at all but reflect different meanings, i.e., they correspond to different inputs. Once these cases of fake optionality are sorted out, the multiple constraint rankings approach explored by Anttila (1997) and others can be seen to be eminently suited for the residue of genuine optionality in the domain of Binding.

2. 2.1

Coargument disjoint reference Four generalizations about disjoint reference

In English and other familiar languages, the distinction between obviative and proximate pronouns converges with that between pronominals and anaphors (reflexive/reciprocal elements). But many languages distinguish between obviative and proximate 4. The cross-linguistic data n o w assembled in Huang (2000) poses a number of prima facie difficulties for this claim, but deeper study of the problematic binding systems is required.

Disjoint reference and the typology of pronouns

183

reflexives (e.g. Swedish sig versus sig själv), and some distinguish between obviative and proximate pronominals (e.g. Algonquian). The obviation requirement and the binding domain, then, are in fact independent properties of pronouns. In addition to Principle Β effects, the Obviation constraint accounts for a class of disjoint reference phenomena that are not addressed by standard versions of Binding Theory. Four such phenomena are treated below. Two of them involve the absence of an expected reading in an anaphoric element. One such gap is that the "coreferential" as opposed to the "bound anaphora" interpretation of reflexives is lacking in certain contexts. I tentatively formulate the relevant descriptive generalization as follows: (3) An anaphor whose antecedent is a coargument has a bound variable but not a coreferential reading. The other gap concerns the distributive reading of plurals and conjoined DPs. (4) A plural or conjoined DP which overlaps in reference with a coargument has a collective reading but not a distributive reading. The other two generalizations are, on the face of it, syntactic, for they involve distributional constraints on the occurrence of certain anaphoric elements, and they are language-specific, or more precisely, overtly instantiated in a minority of languages. First, disjoint reference phenomena are not restricted to pronominals. Some languages, such as Norwegian (Hellan 1988) have OBVIATIVE REFLEXIVES, usually morphologically simple (rather than compounded). Such obviative reflexives are subject to a disjoint reference constraint: (5) An obviative and its coarguments have disjoint reference.5 Conversely, not all pronominals are obviative. Some languages (such as the Algonquian languages) have two series of pronominal endings, OBVIATIVE and PROXIMATE, whose distribution is grammatically constrained as follows: (6) At most one coargument may be proximate. The four generalizations (3)-(6) will be subsumed under a generalized account of Principle Β effects in which an Obviation constraint plays a key role. I take them up in turn below. 2.2

Bound variable vs. coreferential readings

VP anaphora. Generalization (3) is famously manifested in VP anaphora. Reflexive pronouns (e.g. English himself) get only the bound anaphora reading when their antecedent is a coargument, so that "sloppy identity" is obligatory. Elsewhere, there is an 5. Coarguments are arguments of the same predicate (Hellan 1988). For example, Mary and John are not coarguments in Mary considers John (to be) a fool, because John is not an argument of consider; rather, John is an argument of be a fool, and the lowest argument of consider is John (to) be a fool.

184

Paul Kiparsky

ambiguity between the bound variable and coreferential readings, as illustrated by (7) (Hestvik 1990). (7)

John considers himself competent, and so does Fred, a. Fred considers John competent, too. b. Fred considers himself competent, too.

(ambiguous) (strict) (sloppy)

If we assume that the second conjunct in do so VP anaphora gets its meaning from the first, then (7) has the strict reading (a) from (8a) and the sloppy reading (b) from (8b). (8)

a. λχ (χ considers (John competent)) (John) b. λχ [X considers (x competent )) (John)

(coreferential) (bound variable)

The point of interest is the contrast between (7) and the ostensibly parallel (9), where the "sloppy" reading is obligatory, (9)

John hates himself, and so does Fred. a. ¿ Fred hates John, too. b. Fred hates himself, too.

and where the reflexive, therefore, has only the bound variable reading of (10b). (10) a. λχ (χ hates y Λ y = John) (John) b. λχ (χ hates χ) (John) This observation has been expressed by saying that the bound variable reading is obligatory when the anaphors are "locally bound" (Lebeaux 1985, Bouchard 1983, Hestvik 1990). But what exactly is the relevant "local" domain? Contrary to what is often assumed, it is not configurationally defined. The right generalization rather seems to be that the bound variable reading is obligatory when the antecedent is a coargument. All anaphora, local or not, shows both readings as long as the antecedent is not a coargument: (11) a. John thought that Mary's parents would approve of someone like himself, and so did Fred. (ambiguous) = 'Fred thought that they would approve of someone like Fred' (sloppy) = 'Fred thought that they would approve of someone like John' (strict) b. John loves his wife, and so does Fred. (ambiguous) c. John has a picture of himself, and so does Fred. (ambiguous) d. Mary quoted everyone except herself, and so did Bill. (ambiguous) e. John will succeed in spite of himself, and so will Fred. (ambiguous) Reappearance of the ambiguity. Introducing another predication, whether explicit or covert, restores the strict reading. For example, (12) has both the strict reading (12a) ('... than Ted defended John') and the sloppy reading (12b) ('... than Ted defended himself) (Sells et al. 1987):

Disjoint reference and the typology of pronouns

185

(12) John defended himself better than Ted did. a. λχ ((* defended χ) better than ((λ>> (y defended χ)) (Ted))) (John) b. λχ ((χ defended χ) better than ((λγ (y defended y)) (Ted))) (John) Similarly, many speakers find (13) ambiguous between the readings '... nobody else hates John', and '... nobody else hates themselves' (Dahl 1973). On the assumption that only Ρ means 'P and nobody other than P', the two meanings would correspond to the following representations, both of which are consistent with Obviation. (13) Only John hates himself. a. λχ ((χ hates χ) Λ (V y Φ χ (y doesn't hate χ))) (John) b. λχ ((χ hates χ) Λ (V y # χ (y doesn't hate y))) (John) But if only is construed with the reflexive, the sentence remains unambiguous, the only translation being (14):6 (14) John hates only himself. λχ ((χ hates χ) Λ (V y ¿ x (χ doesn't hate >))) (John) Again, the local non-coargument case (15) is ambiguous. (15) Only John considers himself to be smart. a. 'Nobody else considers themselves to be smart.' b. 'Nobody else considers John to be smart.' What is the local domain? Lebeaux (1985) and Bouchard (1983) identify the local domain with the governing category which constitutes the Binding Domain for Principle A. But in fact, it does not coincide with the domain that the other locality properties pick out: namely, the domain within which the binding relation observes c-command (or its equivalent), where split antecedents are prohibited, and where free variation with pronouns is not allowed. For example, a picture noun anaphora notoriously has nonlocal properties, yet shows the ambiguity between the strict and sloppy reading, as in (11c). The point is that both antecedent-anaphor relations below are local (where the brackets demarcate an argument structure), but only in (16b) is the bound variable interpretation obligatory. (16) a. X, [ Y, . . . ] b. [ X, . . . Y, ] Hestvik (1990) interprets the contrast between (7) and (9) in almost the opposite way. According to him, it is Principle A which applies in the coargument domain, while Principle Β applies in the domain of the governing category, as a purely syntactic principle in the spirit of Chomsky (1981). The anaphors in the subject position of ECM constructions, as in (7), and presumably in the other non-coargument cases in (11) as well, are then dealt with by extending the core binding domain by the "accessible sub6. Note also that Everybody likes John does not entail that John likes himself (although it is compatible with it), since John need not be part of the domain of quantification. A person complaining Nobody likes me would not be consoled by being reminded that, after all, he does like himself.

186

Paul Kiparsky

ject" condition (as in Chomsky 1986). Hestvik considers this extended domain to be a non-local binding domain, taking the availability of the strict identity reading under do so ellipsis (as in (7)) as the essential criterion for non-local binding, notwithstanding the fact that the other locality diagnostics (c-command, obligatoriness of the anaphor) pick out a different kind of locality, as discussed in section 2.1 above. On that approach, it remains unexplained why obligatory bound variable interpretation should hold precisely in the "core binding domain" (a fact which for us is a direct consequence of Obviation). Actually the coargument domain has to my knowledge not been shown to constitute the binding domain for any anaphor in any language. Therefore, far from being the core binding domain, it is probably not a binding domain at all (although it is the domain of the universal Obviation principle); the real minimal binding domain is the so-called "extended" binding domain (the accessible subject domain). That assumption will prove to be a productive one, furnishing the key to explaining contrasts such as those in (43a) vs. (47a) in section 2.4 below. Disambiguation in Russian. In English, the distinction between the bound variable reading and the coreferential reading of pronouns normally has no direct formal reflex. But in languages with a suitable inventory of pronominal elements the distinction becomes overt. In Russian, the reflexives svoj (poss.) and sebja (acc.) can have not only third person antecedents, but also first and second person antecedents, in which case they compete with the regular first and second person possessive pronouns. Precisely in the coargument case, the reflexive is however obligatory, and has only the bound variable reading (Dahl 1973). (17) a. Ja ljublju odnogo menja. I like one me Ί like only me.' b. Ja ljublju odnogo sebja. I like one self Ί like only myself.'

(bound variable only)

When the pronoun is not a coargument of its antecedent, the contrast between pronominal and anaphor marks the coreferential versus bound variable reading: (18) a. Ja ljublju moju zenu, i Ivan toze. I love my wife, and Ivan also Ί love my wife, and Ivan also (loves my wife).' (coreferential) b. Ja ljublju svoju zenu, i Ivan toze. I love selfs wife, and Ivan also. Ί love my wife, and Ivan also (loves his wife).' (bound variable) Let's begin by formulating the constraints behind the Russian data. If all goes well, our constraints should account also for the generalizations illustrated in (7)—(11) about English, and for the Swedish and Algonquian data to follow. Towards an analysis. Given that arguments are associated with variables and that these variables are assigned to individuals, there are three possible relations between a pronoun and a (potential) antecedent: bound anaphora (they are coindexed, i.e. associ-

Disjoint reference and the typology of pronouns

187

ated with one variable), coreference (distinct variables, which are assigned to the same individual), and disjoint reference (distinct variables assigned to distinct individuals). Writing arguments as A, Β variables as x, y,..., and individuals as a, b,..., I will use the mnemonic notation in (19): (19) a. Bound anaphora: b. Coreference: c. Disjoint reference:

Ax — (χ —> a) Ax — (jt —* a, y —ï a) Ax — Βy(x a,y —>b, a

Therefore Binding Theory cannot be based simply on coindexing. We need not only constraints on anaphoric relations, marked by the coindexation of arguments in certain syntactic domains, but also constraints governing the interpretation of the coindexed arguments. Assume that the inputs are syntactic structures with variables and assignment to intended referents. The constraint system determines the most harmonic output corresponding to a given input. The input candidate set are logical forms, with any coindexation and with any assignment of individuals to variables. Suppose in particular the following constraints: (20) Constraints: a. BINDING DOMAIN (BD): A pronoun has a compatible antecedent in a designated domain D. b. OBVIATION (OBV): An obviative and its coarguments have disjoint reference. c. PROX: A proximate is a bound anaphor (i.e. it is indexed to the same variable as its antecedent). (The BINDING DOMAIN constraint will not play a role until section 3, but is included here for future reference.) I further assume a PARSE constraint which bars the null candidate. However, PARSE is outranked by the constraints in (20), with the effect that some inputs have no outputs. These constraints account for the gaps noted above, but do not yet suffice to select a unique output in all cases. The choice between remaining candidates is adjudicated by two markedness (economy) constraints, which respectively penalize featural (semantic) complexity and morphological complexity. Featural economy selects anaphors over pronominale (the assumption is that the latter have richer intrinsic feature content), and morphological economy selects simple pronouns over morphologically complex pronouns (as in Burzio 1998). ( 2 1 ) ECONOMY (ECON)

a. b.

FEATURAL ECONOMY:

Avoid pronominale. Avoid morphologically complex pronouns.

MORPHOLOGICAL ECONOMY:

Let us assume that the pronouns of Russian have the same properties as the corresponding pronouns of English, except that the reflexives svoj, sebja have no person features. As in English, then, the reflexives are proximate, and the pronominals, such as menja, moju, are obviative. The tableau in (22) indicates the relevant properties of the input logical forms using the notation in (19). Candidate sets 1 and 2 illustrate

188

Paul Kiparsky

the input logical forms using the notation in (19). Candidate sets 1 and 2 illustrate coargument reflexives, with the bound anaphora and coreferential readings, respectively, and candidate sets 3 and 4 illustrate non-coargument reflexives (possessives, in this case) with the same two readings. For each of these input readings, three output candidates are given: the candidate with a pronominal, the candidate with an anaphor, and the null candidate (the syntactic gap). (22) Russian reflexives and pronouns BD OBV PROX PARSE ECON

la. lb. lc. 2a. 2b. 2c. 3a. 3b. 3c. 4a. 4b. 4c.

χ —» a

Ja* ljublju menjax Bä ' Ja* ljublju sebja* 0 x->a,y-^a Ja* ljublju menja^ Ja* ljublju sebjav 0 χ —> a Jax ljublju moju* zenu ia * Ja, ljublju svoju* zenu 0 ra ' χ —> a, y —> a Ja* ljublju mojuy zenu Ja., ljublju svojuv zenu 0

*

*! *!

*

*! *! *

*! *! *

*! *!

Gaps in the output, which violate the PARSE constraint, arise when higher-ranked constraints exclude the alternative candidates. Thus, a coargument reflexive does not get a coreferential reading; contrast candidate set 2 with candidate set 4.

2.3

An obviative reflexive

Swedish sig. We come now to generalization (5). The existence of obviative reflexives has been demonstrated in several languages, notably Hellan's (1988) classic study of Norwegian anaphora. I will draw my examples from Swedish, whose sig is very similar to Norwegian seg (Diderichsen 1937) and Danish seg (Vikner 1985). Unlike morphologically complex reflexives such as English himself, bare sig cannot have a coargument as its antecedent.7 (23) a.

b.

* John föredrar sig. John prefers REFL 'John prefers himself.' * John, berättade för Per; om sig, ; . John told for Per about REFL 'John, told Per^ about himself,j.'

7. There is a class of verbs for which this statement does not hold (see 3.8 below).

Disjoint reference and the typology of pronouns

189

In the finite domain within which sig must be bound, it can however be freely anteceded by a c-commanding subject, as long as it is not a coargument. This is illustrated in (24) by ECM and small clause constructions and by possessives and complements of nouns. (24) a. John anser sig ha blivit bedragen. John considers REFL have become cheated 'John considers himself to have been cheated.' b. John betraktar sig som expert. John regards REFL as expert 'John regards himself as an expert.' c. John älskade sina barn. John loved REFL children 'John loved his (own) children.' d. John hittade en bild av sig i tidningen. John found a picture of REFL in newspaper.the 'John found a picture of himself in the newspaper.' The subsystem of Swedish pronouns of interest here is built from three classes of morphemes: (1) the pronominals (han, hon, honom, henne, hans, hennes 'he, she, him, her, his, her (poss.)' etc., (2) the obviative simple reflexive sig, with its possessive form sin, and (3) själv, which can occur on its own and be combined with either the reflexives or the pronominals, like German selbst, French même, Spanish mismo, Italian stesso (Giorgi 1983), Russian sam (Klenin 1980). Sig is subject-oriented and allows long-distance binding: (25) a.

Generalen, tvingade översten; att hjälpa sina,j soldater. general.the forced colonel.the to help self's soldiers 'The general forced the colonel to help his soldiers.' b. Han, hjälpte sin, dam, att fâ eld pâ sin,y /hans, »y /hennes»,,; cigarrett. he helped selfs lady to get fire on self's/his /her cigarette 'He helped his date to light his/her cigarette.'

In (25) the short binding is also permissible because the genitive and the PRO subject of the infinitive clause are not coarguments. But because of the disjoint reference requirement, (26) is unambiguous: (26) Generalen, tvingade oversten, att hjälpa sig, *,. 'The general forced the colonel to help himself.' ('himself = the general) Contrast (27), where lova 'promise' is a subject control verb, and neither reading is acceptable: (27) Generalen,· lovade översten; att undersöka sig*,· »y. 'The general promised the colonel to examine himself.' Here the Obviation constraint applies within the lower clause, blocking coreference of sig and PRO (generalen). The subject requirement on the antecedent of sig blocks it

190

Paul Kiparsky

from being coindexed with översten, so the sentence is ungrammatical. Thus, the coindexation in (27) is ruled out by (20).8 These data support the claim that Obviation is a semantic constraint. If Obviation were a syntactic constraint on coindexing, longdistance binding (acceptable in Swedish as (26) shows) should be able to make an "end run" around Obviation, referring to its coargument (PRO) indirectly by being bound to its antecedent (generateti, in this case). Some special stipulation would then have to exclude the unwanted coindexation in (28).9 (28) *Generalen, lovade översten, att PRO, undersöka sig,. Unsurprisingly, the various forms of the pronominal honom are also subject to Obviation, like their English counterparts. So the sentences in (29) are also ungrammatical; (29) a.

* John, älskar honom,. 'John, loves him,.' b. *John, berättade för Per; om honom,·,. 'John, told Pei}· about him,j. '

and (30) is unambiguous, and has the same meaning as the version with the anaphor sig in (26): (30) Generalen, tvingade översten; att undersöka honom, *j. 'The general forced the colonel to examine him.' The function of själv. Where sig and honom are excluded by Obviation, as in (23)(26), they can be rendered grammatical by the simple addition of själv: (31) a. John älskar sig själv. 'John loves himself.' b. John, berättade för Per; om sig, själv. 'John told Per about himself.' c. John, berättade för Per; om honom, själv. 'John told Per about himself.' d. Generalen, tvingade översten; att undersöka sig, själv. 'The general forced the colonel to examine himself.' Of course, ECONOMY still applies, so that honom is blocked by sig. Thus, (30) (on the indicated coindexing) remains ungrammatical even with själv added, because it loses to ( 3 I d ) o n ECONOMY.

Själv is a pronominal operator which excludes other contextually determined or implicit entities (Edmondson & Plank 1978, König 1991). In particular, emphatic Ρ 8. Suppose we assign generalen = x¡¡, översten = xt, PRO = xm, sig = x„, where (by the coindexing in (28)) xk = xm, xt = x„. At the level of interpretation, suppose we assign individual A to variable xk. Then, since xk = xm, A must be assigned to xm, and since xk = x„, A must be assigned to x„. But assigning A both to xm and to x„ violates (20b). On the other hand, the acceptable (26) can be assigned an interpretation in accord with (20). 9. Perhaps requiring preference to be given to the closer of two eligible binders (as in Hellan 1988: 178, 1990).

Disjoint reference and the typology of pronouns

191

själv means 'P, rather than someone else', i.e. 'P personally'. In the combinations sig själv and honom själv, the effect of själv is to negate Obviation. Sidestepping the interesting question how to derive this function of själv from its emphatic meaning, we can think of it as having inherently the feature [-obviative], which it adds to sig and honom (while leaving all their properties intact, including the [+anaphor] feature of sig). Therefore, addition of själv renders the sentences in (23) grammatical. But, even though combining with a pronoun to negate Obviation is perhaps själv' s most important function in the language, it can very well be used elsewhere as well, either redundantly, or for emphasis. This accounts for the appearance of själv outside of any binding context: (32) a. Det gäller honom själv. 'It concerns him [not someone else].' b. Honom själv känner jag inte. 'Him, I don't know.' c Kungen själv lo vade att komma. 'The king himself [not (just) someone else] promised to come. d. Kungen lo vade själv att komma. 'The king promised [personally, on his own] to come.' e. Kungen lo vade att komma själv. 'The king promised to come himself [in person].' In each example of (32), omitting the själv yields a grammatical sentence. In fact, wherever the simple pronoun honom or sig is permissible, it is usually possible to add själv, if clarity or emphasis require it:10 (33) a. John betraktar sig själv som kompetent. 'John regards himself as competent.' b. John anser sig själv ha blivit bedragen. 'John believes himself to have been cheated.' c. John hittade en bild av sig själv i tidningen. 'John found a picture of himself in the newspaper.' d. Generalen, tvingade overstep att befalla löjtnanten* att raka honom?, själv. 'The general forced the colonel to order the lieutenant to shave him himself {or: ... by himself).' One of the virtues of the present approach is that these other uses of själv need not be treated as unrelated "emphatic" reflexives. More than just economy is gained thereby.

10. If this is equally true in Dutch, then Roster's objections to the coargument condition do not go through (Köster 1986: 328). In so far as they are based on cases of optional zichzelf outside of the coargument domain, they could be interpreted in the same way as suggested here for Swedish.

192

Paul Kiparsky

If there really are two självs, we would expect that they could occur together. But this is not possible, see (34c):11 (34) a. Ministern angrep kungen själv. 'The minister attacked the king himself [no less a personage than the king].' b. Ministern själv angrep sig själv. 'The minister himself attacked himself [lesser officials attacked themselves too].' c. *Kungen angrep sig själv själv. 'The king attacked himself [no less a personage than himself].' 12 If there are two distinct elements själv, this is mysterious. If we take the "reflexives" sig själv, honom själv to be simply "emphatic" själv associated with sig and honom, the data in (34) are as expected. A further consequence is that coreference between the subject and the possessive counterpart of sig is allowed, because they are not coarguments. So in (35), the bare possessive reflexive sin can refer to the subject: (35) John, angrep sina, vänner. 'John attacked his friends.' Unlike the Russian possessive reflexive (see (17), (18)), the Swedish possessive reflexive is ambiguous between the bound variable and coreferential readings in the local domain, and correspondingly the non-reflexive possessive (e.g. *hans in (35)) is then excluded. The possessive reflexive can be reinforced with the adjective egen 'own' (corresponding to själv), if emphasis or other reasons require it: (36) John, angrep sina, egna vänner. 'John attacked his own friends.' For the same reason, in (37), coreference between the object Per and the genitive pronoun hans is permitted. Obviation is inapplicable because they are not coarguments, and Blocking is inapplicable because objects cannot antecede reflexives. 11. On the English emphatic reflexive he himself see (Bickerton 1987). The literature on emphatic reflexives is divided on the question whether the "emphatic" and ordinary reflexive uses can be subsumed under the same lexical element. The unitarian view that I am assuming here has been argued for Trique ma 3? à 13 (Hollenbach 1984), for Dutch zelf (Everaert 1986: 42), for Italian stesso (Napoli 1989: 335), and for Old English self (Mitchell 1985: section 475, Ogura 1989: 46, who notes that se//"has no reflexive function by itself' (p. 23), and Levinson 1991). The contrary position is taken for Persian by Moyne (1971), for Chinese by Tang (1989), and for Norwegian by Hellan (1988: 63). Hellan suggests that the two words selv differ in stress, but this claim deserves closer examination. At least in Swedish, it does not seem to be true (nor in Dutch, according to Everaert, voce). 12. Though (34c) could conceivably mean 'The king attacked himself by himself/on his own', with adverbial själv having scope over the VP, to the extent that this makes sense.

Disjoint reference and the typology of pronouns

193

(37) Vi gav Per, hans, lön. 'We gave Per his salary.' It also follows that sig without själv is fine in (38), again because the conjuncts of the co-ordinate NP are not coarguments of the subject, the coargument rather being the entire conjoined NP sig och Maria, as in (38a). Naturally, since the antecedent of sig must be a superior (e.g. c-commanding) subject it cannot be found in a coordinate phrase, as in (38b). (38) a. Han, berättade om sig, och Maria. 'He told about himself and Maria.' b. Maria, och {*sin, /hennes, } vän, berättade om henne,j. 'Maria and her friend talked about her.' Finally, Obviation directly accounts for the gap seen in (39), which (like its English analog (43)) cannot be had by Blocking alone: (39) John, och Lisa, föraktar ? honom/*honom, själv/*sig,/*sig, själv. 'John and Lisa despise him(self).' The analysis. Honom and sig are [+obviative]; sig is moreover reflexive, i.e. specified as requiring a (subject) antecedent in the finite domain; själv is an operator which exempts elements from Obviation, making them [-obviative] (i.e. proximate). The system is thus fully compositional, in that själv makes a unitary contribution to the pronouns it combines with, which is related to the intrinsic meaning which it has outside of those combinations: (40) a. b. c. d.

sig: sig själv: honom\ honom själv:

[+reflexive,+obviative] [+reflexive,-obviative] [reflexive,+obviative] [-reflexive,-obviative]

We are now ready to lay out the constraint tableaux for Swedish sig in the local domain. Tableau (41) shows the analysis of reflexive objects, first of the coargument type, such as (23), and then of the non-coargument type, such as (24).

13. Also in Norwegian, sentences like Fris0ren barberer seg og kundene 'the barber shaves himself and the customers' are fully acceptable. Cf. Hellan (1988: 105).

194

Paul Kiparsky

(41) Reflexives and pronouns in Swedish a. Coarguments (e.g. (23)) BD OBV PROX PARSE ECON la. lb.

sig

a

le.

honom

Id.

honom själv

le.

*

*!

*

**!

*!

0

2a.

χ—» α, y - >

α

*!

sig

2b.

sig själv

2c.

honom

2e.

*!

*

*!

**

*!

honom själv

2d.

*

0

3a.

χ-»

a,y->

b

3b. 3c.

*!

sig själv

ss>

BS'

*

sig

*!

sig själv

*!

*

*

*!

**

honom

*

honom själv

3d. 3e.

*!

0 b.

Non-Coarguments (e.g. (24)) BD OBV PROX PARSE ECON

4a. "S" x->

sig

a

4b.

sig själv

4c.

honom

4d.

honom själv

4e. 5a.

*! *! **!

*!

0 a,y—ï

5b.

a

sig sig själv

5c.

honom

5d.

honom själv

5e.

**

*!

sig

*! *!

6c.

honom

6e.

*!

0 sig själv

6d.

*

*!

6b.

6a.

*!

x->a,y-*b

honom själv 0

*

*

*!

**

*

*!

The constraints account straightforwardly for the coreferential versus bound variable readings of reflexives. A s candidate sets 1 and 2 in tableau (41) show, John hatar sig själv, like its English equivalent John hates himself, predicates of John only the property of "self-hating", not the property of "John-hating".

Disjoint reference and the typology of pronouns

195

Tableau (42) shows another consequence of the obviative character of the Swedish reflexive: the ambiguity of the reflexive possessive in the local domain between the bound variable and coreferential readings, and the exclusion of the corresponding nonreflexive (see (35)). Being obviative, the reflexive sin is not subject to PROX, hence is available in both readings, and preferred to the non-reflexive hans by ECONOMY. (42) Reflexive possessive in Swedish BD OBV PROX PARSE ECON

la. lb. le. 2a. 2b. 2c.

χ —» a Kä

' x-^a,y-¥a

John* angrep hans* vänner. John* angrep sina* vänner. 0 John* angrep hans^, vänner. John* angrep sinav vänner. 0

*! *! *! *!

In this way our analysis unifies two differences between Swedish and Russian which have hitherto been ignored, or at best treated as unrelated: that simple coargument reflexives are unacceptable in Swedish and acceptable in Russian, and that possessive reflexives are ambiguous between the bound anaphora and coreferential readings in Swedish, but have only the former reading in Russian.

2.4

Collective vs. distributive readings

The second coargument disjoint reference effect, stated in (4), is illustrated in (43). (43) a. We prefer me. b. I like us. Such sentences have sometimes been claimed to be ungrammatical (Higginbotham 1985: 576, Lasnik 1989: 151). But as Reinhart & Reuland (1993) point out, only the distributive reading is excluded, while the collective reading is fine, and can be brought out clearly in appropriate contexts such as (44): (44) a. By an overwhelming majority, we preferred me. b. We have a terrific team. I really like us. The group denoted by the plural expression must participate in the event as a single collective entity, and not as separate individuals. Thus, (44a) means that the choice is made as a group, perhaps by a collective act of voting, and (44b) means that I like us qua team - not that I like all (or even any) of my teammates individually. In contrast to cases like (44), the distributive interpretation is readily available in non-coargument contexts like (45). (45) I want us to have good reputations. As Reinhart & Reuland point out, this suggests a semantic version of Principle B.

196

Paul Kiparsky

Two remarks on these cases will help clarify the issue. First, mere collaboration or joint participation of several agents in an event is not enough for the collective interpretation; (46a), for example, does not felicitously describe a situation in which John and Mary are each taking care of John part of the time, and (46b) does not felicitously describe an act of murder-cum-suicide. (46) a. *John, and Mary, are taking care of him,. b. Jim Jones killed them all with poisoned Kool-Aid. Since coreference must be with respect to the whole group denoted by the plural or conjoined NP's, and disjoint reference is distributive over the members of the group, there is actually no fully acceptable way to express coreference with respect to one of the members of the set denoted by the antecedent in sentences like (43). This is a gap in the language's expressive repertoire. Secondly, the distributivity of disjoint reference is bidirectional, as (20b) entails. Disjoint reference is distributive over the members of the group regardless of whether the group is referred to by the antecedent (43a) or by the pronoun (split antecedent cases like (43b)). Consequently, collective interpretation allows for partial overlapping of reference whether the group-denoting element is the antecedent or the pronoun, as in (44). In contrast, outside of the coargument domain, the distributive reading of the pronomináis is fully acceptable, as long as there is no possibility of a reflexive (see (47)) or Pro (see (48)). For example, in (47a), the pronominal us is not excluded (for it is not a coargument of its antecedent Γ), nor pre-empted by a reflexive (ourselves is excluded since it has only partially overlapping reference with I). (47) a. b. c. d. e. f.

I believe {us/*ourselves} to have been cheated. Richard, and Pat, both regard {him, /her ; /*himself/*herself} as innocent. I prefer to call us rape statistics. (Letter, /VÎT 21.4.1991.) John, and Mary, both have a picture of {him, /her, /*himself/*herself}. In spite of {me/*myself}, we will succeed. Both John, and Ann ; must have decided to promote everyone except {him/her/*himself,/*herself,}.

(48) a. b. c. d.

I want us to be friends. John, and Mary prefer his, going to the movies. Neither John, nor Mary, had enough film for her, to take the pictures. Mary, and I were waiting, she, having arrived slightly before me.

The correct distribution of the distributive and collective reading of plurals is accounted for by the previously developed constraints. (49) shows how / like us has only a collective reading (candidate set 1, χ —» a, y —> tí), and lacks a distributed reading (candidate set 2, χ —» a, y —» a+b+c ...), and how *I like ourselves is excluded.

197

Disjoint reference and the typology of pronouns (49) Coarguments: I like _. BD OBV PROX PARSE ECON la. χ—> a,y —> b lb. lc. 2a. χ —> a,y—> a+b+c.. 2b. 2c. cs·

2.5

us ourselves *! 0 us ourselves *! 0

* *

**

*! *

*! *

** *

Gaps

Another argument that Obviation and Economy are separate constraints comes from gaps. Obviation prohibits pronominals from referring to coarguments even in places where the binding conditions do not allow anaphors to be used in their place. In fact, in the coargument domain, gaps in the distribution of reflexives are never simply filled in by pronominals. For example, in (50), them is ruled out by Obviation even though themselves is also impossible, split antecedents being impermissible in local binding domains (I symbolize split antecedents by the plus notation, e.g. themi+J). (50) John, confronted Billy with {*themi+/ /*themselves,+/}. Since the exclusion of the pronominal in (50) cannot be due to blocking by a reflexive, such sentences give independent evidence for Obviation. In (51) and (52), the pronominal is again ruled out by Obviation, even though the reflexive is also impossible, for the prominence requirements on the antecedent are not satisfied. (51) a. I showed {*him,/*himself,} to John, in the mirror, b. I showed {*himJ /*himself, } John, in the mirror. (52) a. Mary talked about {*him, /*himself,} to John,. b. Mary talked about John, to {*him, /*himself,}. c. John, and Mary, both admire {*him, /*heij /*himself, /*herself}}. Strikingly, a pronominal cannot refer to an object coargument even where object antecedents for reflexives are disallowed generally, as in German: (53) Die Maria, hat {sich,,*,·/ihn*,·,*;} dem Hans; genau beschrieben. 'Maria described self/him exactly to Hans.' For this reason, (54) and (55) exactly reverse the pattern of (47) and (48): (54) a. b. c. d.

I believe {*me/myself} to have been cheated. Richard, regards {*him,/himself} as innocent. I call {*me/myself} Captain Nemo. Mary, has a picture of {*her,/herself}.

198

Paul Kiparsky e. f.

(55) a. b. c. d.

In spite of {*me/myself}, I will succeed. Y o u must have decided to promote everyone except {*you/yourself}. I want ( * m e ) to be your friend. John, prefers (*his,) going to the movies. Maryi didn't have enough film ( * f o r her,) to take the pictures. Mary was waiting, (*she) having arrived before me.

These are local binding domains where the pronouns are not coarguments of their antecedent. The pronominale are blocked by reflexives in (54), and by P R O in (55). Summing up: the pattern of distribution of anaphors and pronominale results from the joint application of two independent principles, Obviation and Blocking. Obviation explains the contrast between (43), (50) and (47)-(48), and Blocking explains the contrast between (54)-(55) and (47)-(48). Violation of either Obviation or Blocking by itself produces a deviant sentence but the crispest ungrammaticality comes from joint violation of both.

2.6

Morphologically marked Obviation

Finally we turn to generalization (6). Morphologically marked Obviation, at least as instantiated in Algonquian languages such languages as Cree and Ojibwa (Dahlstrom 1986, Schwartz & Dunnigan 1986: 292, Grafstein 1988, 1989, Aissen 1997), is also a manifestation of coargument disjoint reference, but one that falls within "syntax" as traditionally demarcated. These languages distinguish between two sets of pronominal affixes, obviative (the morphologically marked category), and proximate. A b o v e the clausal level, the use of proximates and obviatives is determined by discourse structure, the basic generalization being that proximate forms are reserved for the current discourse topic (or topics, since in certain cases there may be more than one concurrent topic). A switch of obviation serves, typically, to foreground another topic. Within a clause, obviation is rigorously controlled by the constraint that there can be at most one proximate third person argument.14 For example, in these Ojibwa examples (from Grafstein 1989) the obviation marker is obligatory in the coargument case in (56), but determined by the abovementioned discourse condition in (57) and (58): (56) a.

b.

John o-wa:bam-a:-an John 3-see-3-OBV 'John, sees him/ * John o-wa:bam-a: John 3-see-3(PROX) 'John, sees him,'

14. Dahlstrom notes that, although only animate nouns bear overt obviative marking in Cree, the verbal syntax shows that inanimate nouns also are obviative under the appropriate conditions.

Disjoint reference and the typology of pronouns

199

(57) a. ánimos o-gike:nda:n bo:zës ga:zo-d dog 3-know cat hide-3 'The dog knows that the cat is hiding.' b. ánimos o-gike:nda:n bo:zës-an ga:zo-ini-d dog

3-know

cat-OBV

hide-OBV-3

'The dog knows that the cat is hiding.' (58) a. Mary o-gi:-oji:ma:-an John o-gosis-an M a r y 3-PAST-kiss-OBV J o h n 3-son-OBV

'Mary kissed John's son.' b. Mary o-gi:-oji:ma:-an John-an

o-gosis-ini

M a r y 3-PAST-kiss-OBV John-OBV 3-son-FURTHEROBV

'Mary kissed John's son.' Which argument is obviative and which is proximate in this case depends on the inflection of the verb, the rule being that the subject of a direct verb is proximate, and the subject of an inverse verb is obviative (Dahlstrom 1986: 110). From the point of view of discourse organization (though not syntactically!), inverse forms are comparable to passives, and so we can say that the proximate is assigned to the syntactically more prominent argument (Dahlstrom, ibid.).15 The clause-internal distribution of Algonquian endings follows if we assume that only the obviative endings are subject to the OBVIATION constraint. The obviative endings have the same property as English pronouns, of not allowing coreference of coarguments. The proximate endings, like reflexives, are referentially dependent, but unlike them, their antecedent need not be confined to a syntactically defined domain. Rather, it is a discourse topic, which can be introduced at any point in a discourse and maintained across an arbitrarily long stretch of it. The proximates thus fall in with a large class of pronouns, discussed in the next section, which have the property that they cannot introduce a discourse referent. It is the referentially dependent status of proximates that is the primary regulator of distribution of proximates versus obviatives across clause boundaries. Obviatives are referentially independent, i.e. unrestricted as to how they can pick up their reference. Hence proximates block obviatives in discourse in Algonquian, in much the same way as reflexives block pronominals in the binding domain. Consequently, although the obviatives are analogous to familiar pronouns, their surface distribution is much more restricted because they are blocked, but by the more widely licensed morphologically unmarked category of proximates. In effect, they are confined to coargument disjoint reference function within clauses, and in discourse to new topics. From this perspective, both obviatives and proximates fit perfectly into the typology of pronominal categories, and both have, in fact, familiar counterparts in English.

15. Other discourse-related factors, such as empathy, seem to play a role as well, Dahlstrom suggests.

200

Paul Kiparsky

An important additional piece of evidence that Obviation and Blocking are independent constraints in Ojibwa as well comes from a key observation of Dahlstrom (1986). She noted that under certain conditions a narrative can maintain two concurrent proximates.16 In such cases, precisely when both are clausemates, one of them, the syntactically less prominent one, obligatorily becomes obviative. In the subsequent discourse, both then resume again as proximate.

3. 3.1

The typology of pronouns Antecedent domain

In order to flesh out the typology, we must specify the domains within which pronouns can require their antecedent to be located. They are defined by a hierarchy of constraints which impose successively more restrictive requirements on the locality relation between the anaphor and its antecedent. The domains form a strict hierarchy of inclusion:

The hierarchy is constituted by four binary divisions: (60) a.

A pronoun may be REFERENTIALLY INDEPENDENT or REFERENTIALLY DEPENDENT. Referentially independent pronouns can (but need not) introduce something new into the discourse. For example, they can have deictic and demonstrative uses. Referentially dependent pronouns cannot introduce anything new into the discourse: they must have at least a "discourse antecedent".

b.

Referentially dependent pronouns may be REFLEXIVE or NON-REFLEXIVE.

(Referentially independent pronouns are necessarily non-reflexive.) Reflexive pronouns need a syntactic antecedent. Non-reflexive pronouns can (but need not) get their reference from context/discourse. c. Reflexive (hence referentially dependent) pronouns may be FINITE-BOUND or NON-FINITE-BOUND. (Non-reflexive pronouns are necessarily nonbound.) Finite-bound pronouns require an antecedent within the same finite clause. Non-finite-bound pronouns can (but need not) have an antecedent within the same finite clause. d. Finite-bound (hence reflexive and referentially dependent) pronouns may be subject to the requirement that they be LOCALLY BOUND, or not. Locally bound pronouns require an antecedent in the first accessible subject doló. For example, she cites the sentence "When they (PROX) arrived there, he (PROX) had already shot the buffalo." (p. 114).

Disjoint reference and the typology of pronouns

201

main. Non-locally bound pronouns can (but need not) have a "longdistance" antecedent. Here are some illustrative diagnostic contexts for the antecedent domain. The category of a pronoun is defined by the maximum domain in which its antecedent may be found. The contexts define a hierarchy of five domains. (61) a.

Referentially independent'. It's _! We need to talk about _, _, and _. b. Referentially dependent'. 1. Non-reflexive: John, is here. I saw _,·. (discourse antecedent) 2. Reflexive: i. Non-bound: John, thought that I would criticize _ ¡. John, was sad. Why didn't Mary love _,·?, 1 7 ii. Finite-bound'. A. Non-locally bound: John, asked me to criticize _,·. B. Locally bound: John, criticized _,. (reflexive) John, showed Billy _ in the mirror.

Thus, him can appear in (61a), so it is referentially independent. It can appear in (61b) but not in (61a), so it is referentially dependent. Himself can only appear in (61b 2ii B), so it is locally bound. The five pronoun types cross-classify with the obviation property [±Obviative], together yielding ten logically possible types of pronominal elements. The full typology, with a representative example of each type, is given in (62). (62) The pronoun typology +ref.dep. +refl +fin.-bd

[—O]: [+0]:

+local himself —

-local sebja Swed. sig

-ref.dep. -refi

-fin.-bd.

Iceland, sig Tk. kendisi Ma. aapari Gr. o idhios

— him

17. This is the logophoric case, in the strict sense of an element whose reference is determined in relation to the speech, thought, or point of view represented in the containing discourse (Sells 1987).

202

Paul Kiparsky

The typology in (62) diverges from traditional Binding Theory in two respects. First, it differentiates by a hierarchy of binding domain specifications what Binding Theory has so far classed together as pronominale, i.e. [-reflexive] elements, and secondly, it treats Obviation as an independent cross-classifying property of all types of pronouns. Two likely functional requirements on pronouns are EXPRESSIVENESS and CONTRASTIVENESS, defined as follows. A pronominal system is expressive if it provides the means to mark both coreference and non-coreference in any domain (as presumably all languages do). This is the case if the language has at least a referentially independent pronoun, and a non-obviative one. Formally, is is equivalent to the assumption that PARSE universally outranks ECONOMY. A pronoun is contrastive if it is obviative or referentially dependent, i.e. if its relation to its antecedent is constrained in some way. I make the following conjecture: (63) a. b.

EXPRESSIVENESS: All pronominal systems are expressive, CONTRASTIVENESS: All pronouns are contrastive.

(63b) implies that the gap in the top ([—O]) line of (62) is systematic: referentially independent pronouns are universally obviative. A non-obviative referentially independent pronoun, if it existed, would be a "universal pronoun" which had neither coreference nor disjoint reference restrictions, and in particular could be used both as a demonstrative and as a reflexive. Thus, (63a) guarantees the expressiveness of a language's pronominal inventory, (63b) complements it by ensuring the functionality of each element in the inventory. The most restricted category, obviative locally bound reflexives, might at first sight seem non-existent as well. If obviation is defined as before, then a pronoun of this type would be virtually useless, for the obviation requirement prohibits it from having a coargument antecedent while its local binding requirement can only be satisfied by an antecedent in the same clause. On the other hand, it is precisely for locally bound reflexives that we need to specify subject orientation, so as to distinguish between reflexives like English himself, which can be anteceded by objects, and reflexives like German sich, which require subject antecedents. It is tempting, therefore, to take locally bound subject-oriented reflexives such as German sich as being obviative. This move at one stroke fills the empty slot in the typology and accommodates subject orientation without adding an extra feature or parameter to the system. The intuition is that although reflexives which must be locally bound cannot literally satisfy the coargument disjoint reference requirement, they can satisfy a relativized form of it, which requires disjoint reference from all coarguments but the structurally most prominent one. Formalizing this idea is however not a trivial matter.

3.2

Referentially independent pronouns

A pronoun, qua lexical element, is referentially independent if and only if it can introduce new discourse referents. It doesn't have to introduce new discourse referents; in fact the present typology excludes the possibility of pronouns that cannot get their

203

Disjoint reference and the typology of pronouns 18

reference anaphorically, and as far as I know there are none. Referentially independent pronouns include, by definition, all pronouns which allow a deictic or demonstrative use, whatever other uses they may have in addition.19 The resulting typology of pronouns, though richer, is actually more constrained in what it predicts about the range of possible anaphors. What needs to be specified about a pronoun is whether it is obviative, and what antecedent domain constraint it is subject to. Blocking and, for obviative elements, the domain of obviation, are fixed. In sections 3.3-3.8 below I document examples of the different antecedent domains, with special attention to the category of referentially dependent non-reflexive pronouns, which are not countenanced in traditional theories. In section 4 I then turn to a number of empirical issues raised by the claim that the domain of obviation is universally fixed.

3.3

Referentially dependent non-reflexive pronouns

Identifying the category. Many languages lack reflexive pronouns entirely and simply use personal pronouns in their place, such as Manam (Lichtenberk 1983: 126), Jiwarli (Austin 1989), Gumbaynggir (Levinson 1991), Old English, and Frisian (Tiersma 1985). Even more commonly, personal pronouns fill in for missing reflexives in some categories, e.g. non-third person in German, animates in Chamorro (Chung 1989), and genitives in English. These can be considered referentially dependent pronouns of the non-obviative variety. The obviative counterparts of this category are modern German masculine and feminine er, sie in reference to inanimates, and neuter es in reference to animates, and modern English it as opposed to he, she, they. Three main diagnostics distinguish referentially dependent pronouns from referentially independent pronominale, all flowing from the requirement for a discourse antecedent. First, since true deixis (pointing) introduces new discourse referents, referentially dependent pronouns cannot be used in a strictly deictic way. (64) (Pointing:) a. I mean him over there. b. Ich meine ihn da. (Pointing to a boy.) c. *Ich meine ihn da. (Pointing to a table.) (OK: Ich meine den da.) d. *I need ΓΓ, not ΓΓ. (OK: I need this, not that.) 18. The "contrastive pronouns" described in Heath (1983) come close, but they seem to be regular pronouns plus a focus clitic. 19. I take deixis in a narrow sense which does not include "ambient reference", e.g. (A to B, in reference to a man running towards them waving his hands excitedly): What do you think he wants? Here the pronoun is not introducing a new element into the discourse; rather, it is already present in virtue of the situation. - I think that my view is not really inconsistent with the idea that all pronouns are referentially dependent, (e.g. Roberts 1987, Partee 1989). Some distinction corresponding to referential dependency in my sense would have to be drawn in any case.

204

Paul Kiparsky

Secondly, since pronouns heading restrictive relative clauses introduce new discourse referents, referentially dependent pronouns cannot head restrictive relative clauses: (65) a. He who hesitates is lost. (OK as restrictive) b. *Er, der zögert, ist verloren, (cannot be restrictive) c. Derjenige), der zögert, ist verloren. Or: Wer zögert, ist verloren. d. *It which he said is true. e. That which he said is true. Or: What he said is true. A third criterion is that referentially dependent pronouns, like true reflexives, do not appear freely in predicative positions. (66) a. b.

The mysterious benefactor turned out to be him. *The mysterious object turned out to be it. I wish I could become him! *I wish I could become it!

This follows if we assume that elements in predicative positions cannot be referentially dependent. 20 1 will assume that they are not referential at all. Thus the contrast between the English and German pronouns seen in (65)-(66) identifies the former as independent pronouns and the latter as referentially dependent pronouns. Old English. In Old English, the personal pronouns hë, hëo, hit 'he, she, it', in addition to their free pronominal use, are capable of referring to any argument of the same clause: (67) a.

& he hine & his öeode gelaedde to maersianne and he him and his people brought to celebrate 'and he brought himself and his people to celebrate' (Bede 5 19.468.7) b. fionne wolde heo ealra nyhst hy banian & Jjwean then would she of.all latest her bathe and wash '(having first washed the other servants of Christ that were there) then she would last of all bathe and wash herself (Bede 4 19.318.20) c. ac mid inneweardre heortan monic mid hine sprecende smeade and with inmost heart often with him speaking reflected 'in his innermost heart he often argued with himself (Bede 2 8.124.22) d. Jjaette nasnig biscopa hine oörum forbaere that no bishop him others.DAT advance.SUBJ.3PL 'that no bishop shall put himself above others' (Bede 5.278.27)

It turns out that they have exactly the properties of referentially dependent pronouns. First, Old English personal pronouns are not used deictically as in (64). Second, it appears that Old English hë, hëo, hit cannot head restrictive relative clauses. Although

20. This is superficially contradicted by constructions like John is not himself today. But himself here is not referential. Rather, it means something like 'the way he usually is'. If there were a referential dependency between John and himself the sentence would be contradictory.

Disjoint reference and the typology of pronouns

205

grammars do not seem to note it, Old English, hë, hëo, hit as heads of relative clauses seem to require specific contextually identifiable referents:21 (68) He ongann on öaere menniscnysse, seöe aefre waes and aefre biö God he began in this humanity who ever was and ever will.be God 'He [Christ], who ever was and ever will be God, began in humanity.' (/ECHom ii.8.4) In contrast, the demonstrative pronoun së, sëo, pœt with a relative clause can be generic: (69) Se öe eow gehyrsumaö, he gehyrsumaö me (/ECHom ii.50.7) He who obeys you, obeys me (restrictive) = Derjenige), der ...; Wer... (not *Er, der ..., cf. (65)) The relative clause pattern is thus as follows: (70) a. He ..., (se)Q>e)... — nonrestrictive b. Se ..., (se)(J)e) ... — restrictive or nonrestrictive Third, like modern German, Old English apparently did not allow the simple pronouns in predicative position. That is, sentences like (71) do not seem to be attested in Old English: (71) a.

*We ne magon hie weorfian. 'We cannot become them.' b. *Se cuma wies ic. 'The guest was I.' c. *öaet husel is gastlice he. 'In a spiritual sense, the eucharist is He.'

In principle, OE hë, hëo, hit could be used to refer to any discourse referent once it has been introduced. However, there seems to have been a tendency to use them in reference to the primary discourse topic. To mark a change of topic, the demonstrative pronoun së sëo pœt is typically used. The "subject-changing" function of OE së, noted in (Mitchell 1985: 129) for examples like (72), is then probably really a topic-changing function: (72) Hi habbaö mid him awyriedne engel, mancynnes feond, and se haefô andweald ... (AECHom ii.488.14) 'They have with them a corrupt angel, the enemy of mankind, and he has power over...'

21. Adherence to this rule improves the editor's translation in many instances, e.g. BIHom 19.32, 167.35, 169.4, 197.3, AECHom II 8.4, 24.12, 124.27, CP 57.18, 20, Bede 2 9.132.27. In each of these cases, the context makes clear that the head he refers to an established discourse antecedent, confirming the claim that the Old English personal pronouns are unrestricted.

206

Paul Kiparsky

It is important that referential dependency is not reducible to a pronoun's ability to be stressed or focused: referentially dependent pronominals can very well be stressed and even be put into focus position, provided only that they have a discourse antecedent: (73) a. First Max kissed the monster, and then it kissed him. b. Erst hat die Maria den Max geküsst, und dann hat er sie geküsst. 'First Maria kissed Max, and then he kissed her.' c. Ihn hat sie doch geküsst! 'Surely she kissed himV In Old English as well, contrastive uses were admissible as long as they were referentially dependent (as in (73)):22 (74) J)a Cwenas hergiaö hwilum on öa Norömen ofer 9.one mor, hwilum J)a Norömen on hy. (Orosius) 'Sometimes the Finns make raids on the Norwegians across the mountains, sometimes the Norwegians on them [i.e. on the Finns].' Therefore referential dependency is not the same as "weakness" in the sense of Rigau (1986) and others: it may be that all "weak" pronouns are referentially dependent, but the converse does not hold.23 Nor are referentially dependent pronouns necessarily "logophoric" (Hagège 1974, Clements 1975, Sells 1987, Zribi-Hertz 1989, Koopman & Sportiche 1989, Pollard & Sag 1992). That is, the reference is not necessarily characterized in terms of the speech, thought, orientation, or point of view represented by the containing discourse, as for example the logophoric reflexive in (75) is:24 (75) . . . the Kommandant, had taken good care to read the reports about himself, that Verkramp, had submitted. (Tom Sharpe, Riotous Assembly, Pan 1973: 12) Obviation: Turkish vs. Greek Another pronoun plausibly analysed as referentially dependent, of the [-obviative] type, is Turkish kendisi. It can be bound sentenceinternally, but when not so bound, it "must pick out an individual that is already in the domain of discourse, and cannot be used to introduce an individual into the domain of 25 discourse" (Enç 1983); it "cannot be accompanied by a demonstration". Greek o

22. Similarly in Middle English: It liketh hym at the wrastlyng for to be, And demeth whether he do bet or he. (Chaucer, PF 165) 23. The recent three-way classification of Cardinaletti & Starke (1996) is also based on different criteria than the one in the text. 24. First person pronouns, and to some extent also second person pronouns, are an important special case. They appear readily outside of "represented thought/speech" contexts (Reinhart and Reuland 1991). 25. Enç credits some of these observations to an unpublished paper by N. Tölek. Enç (1989: 76) on the other hand states that kendisi can be deictic; I assume she means "free", for kendisi is not, apparently, acceptable in "pointing" contexts.

Disjoint reference and the typology of pronouns

207

idhios (Iatridou 1986) probably also belongs here. C. Condoravdi (class lectures 1988) points out that it does not need to be bound within its sentence: (76) O Yanis, mas simvulepse na figume amesos. O idhios, tha efevge the Yanis us advised to leave immediately the same PRT left argotera. later 'Yanis advised us to leave immediately. He was to leave later.' It does however require at least a discourse antecedent, so that its deictic use, for example, is inadmissible, and it can have a bound variable interpretation (e.g. with a quantified antecedent). Since it need not be syntactically bound, its antecedent can be an object: (77) O Yanis, ipe ston Kosta; oti i Maria agapai ton idhio i; the Yanis told to Kostas that the Maria loved the same 'Yanis told Kostas that Maria loved him.' ('him' must be Yanis or Kostas) So Greek o idhios has the hallmarks of a referentially dependent pronoun. But unlike the other referentially dependent pronouns cited here, o idhios is restricted to emphatic/contrastive (focused) uses. It is [+obviative], and absolutely cannot refer to a coargument: (78) O Yanis agapa *ton idhio / ton eafton tu. The Yanis loves the same / the self his 'Yanis loves himself.' The genitive tu idhiu also prefers to be bound outside its clause: (79) O Yanis agapa tin mitera ??tu idhiu / tu. The Yanis loves the mother his same's /his 'Yanis loves his mother.' This can't be by obviation because they are not coarguments. In fact, the deviance is weaker than in the coargument case (78), and is presumably due to blocking of the morphologically complex tu idhiu by the simple reflexive clitic tu in the local domain. Child language. The provision of the category of referentially dependent pronouns with an unrestricted binding domain in the typology of (62) suggests a simple explanation for some otherwise puzzling acquisition data. Children, in a wide range of experimental tasks, have been found to violate principle Β more often than principle A (Grimshaw & Rosen 1990), and they seek to assign discourse antecedents to pronouns in preference to interpreting them deictically (ibid.: 200). Presumably children faced with the task of interpreting sentences like The Smurf is talking to him (perhaps in the absence of a plausible deictic context) resort to the assumption that him is a [-obviative] pronoun after all. Given that referentially dependent pronouns can have an unrestricted domain, this would not reflect a disregard of UG principles of Binding Theory, but simply a temporary resetting of the Obviation parameter for this pronoun to its marked value [-obviative], under the positive evidence provided by the experi-

208

Paul Kiparsky

menter. As we have just seen, this hypothesis would have been right on target for the real English data of a thousand years ago.26 I suspect that similar effects occur even with adult speakers. In soliciting grammaticality judgments I have noticed that even some informants who unhesitatingly reject long-distance binding at first, begin to waver after being subjected to repeated instances of it. Other instances of referentially dependent pronouns. Many pronouns which have been described as requiring "topic" antecedents probably belong here. For example, Saxon (1986: 210) states that Dogrib we- (the counterpart of the famous Navajo bi-) "seems to be appropriate only in situations where [its] referent has some status as a discourse topic"; and for Korean caki it has been claimed that there are no syntactic constraints on the antecedent, but it is normally the discourse topic (Hong 1989); caki itself cannot apparently introduce a new discourse topic.27 As discussed in section 2, the proximate pronouns of Algonquian seem to be referentially dependent pronouns as well. Standard versions of Binding Theory would have to say that referentially dependent pronouns are ambiguous between pronominals and anaphors, i.e. that they are entered in the lexicon with two distinct feature specifications, both [-anaphor.+pronominal] and [+anaphor,-pronominal]. This kind of homonymy is suggested by Köster (1986) for hem in certain Dutch dialects (p. 325), and for the English possessive pronouns (p. 343). Given the features that standard Binding Theory provides, the pronominal and the anaphor must be listed separately because there is no way to subsume them under a single underspecified lexical entry with neutralization of the relevant features.28 The 26. Grimshaw & Rosen (1990) also report evidence that children tend to reject pronominals more if the coargument antecedent is quantified, accepting e.g. Is mama bear washing her? more readily than Is every bear washing her? This fits well in the present account in that the latter sentence (having a quantified antecedent) must be a case of bound anaphora. The children making this distinction were, so to speak, learning Russian (see section 2 above). The expectation would then be that the distinction would emerge in older children with an established system of reflexives. 27. Of course, it is not always easy to distinguish unrestricted pronouns from logophoric pronouns, and there may be intermediate and mixed cases. The example that Saxon cites does suggest, though, that the antecedent Dogrib we- is not solely determined by logophoricity, and Hong explicitly argues the same for caki, although logophoricity, "empathy" and the like do play some role (see also Cho 1985). 28. Chung has argued that this apparently unhappy consequence is actually a fortunate one. She shows that in Chamorro, certain language-specific syntactic constraints single out those configurations in which the ambiguous pronouns are bound (coindexed with a commanding or preceding antecedent), and argues that this constitutes evidence for a covert distinction between pronominals and anaphors in that language (Chung 1989). Suppose however that the pronouns in question belong unambiguously to a category of pronoun which have the property that they may be bound (coindexed with an antecedent) in some local domain, but need not be bound. Then the grammatical constraints which Chung defines on anaphors can be reformulated as constraints on coindexation, and do not require multiple lexical entries. In other words, rather than referring to the distribution of elements with any particular set of intrinsically specified pronominal features, the conditions can refer to bound and free uses of pronouns. This has the additional virtue of avoiding the need to postulate lexical homonymy in a whole series of pronouns (in Chamorro,

Disjoint reference and the typology of pronouns

209

remaining types have been identified as different kinds of long-distance anaphors in the literature and can be dealt with briefly here.

3.4

Finite-bound reflexive pronouns

These pronouns require a c-commanding antecedent in the Root domain, i.e. possibly the subject of a superordinate finite clause. Icelandic sig is of this type (Thráinsson 1979). An obviative reflexive that must be bound in this domain is Marathi aapaii (Joshi 1989, Dalrymple 1993): (80) Jane-ne, John-laa, kajavle ki aapai\,,*, turangaat aahot Jane-ERG John-ACC informed that self prison-LOC was 'Jane, informed John, that self, was in prison.' Malayalam (aan (Mohanan 1982) is also obviative and require an antecedent in the Finite domain. The Eskimo null pronoun (Bok-Bennema 1984) seems to be another instance of this type.

3.5

Non-locally bound reflexive pronouns

This category is represented by Swedish sig, which may be bound out of an infinitive clause but not out of a finite clause: (81) a. b.

3.6

Generalen, tvingade översteny att hjälpa sina,¿ soldater. 'The general forced the colonel to help s e l f s soldiers.' Generalen, sade att översten, skulle hjälpa sina»,; soldater. 'The general said that the colonel should help s e l f s soldiers.'

Locally bound reflexive pronouns

Most complex reflexives are of this type. Their morphology reveals their origin as strategies for defeating Obviation. In one type, the reflexive is maneuvered into a noncoargument position, as in reflexives of the form POSS+N (where Ν = 'head', 'body' etc.). In the other, its Obviation is annulled by an element such as self

3.7

Variation

Let us assume that the hierarchy of binding domains is induced by a set of successively more restrictive constraints. For concreteness, let BD2 be the constraint that requires the saving us from having to say that all animate pronouns would be systematically homonymous between pronominals and anaphors).

210

Paul Kiparsky

antecedent to be in the same finite clause and let BDj be the constraint that requires the antecedent to be in the same clause (locally bound). Then "optional reflexivization" in Swedish is derived by allowing the ranking of BDi to vary. The glosses for the four candidate sets are: (1) 'John considers himself to be intelligent', (2) 'J asked M to criticize him', (2') 'J asked M to criticize herself, (3) 'J thinks M will criticize him', (3') 'J thinks M will criticize herself'. 29 (82) Swedish: Long-distance sig BD2 OBV ECON BDi la. J, anser n>[sig, vara intelligent] lb. J, anser n>[honom, vara intelligent] le. J, anser n>[sig, själv vara intelligent] Id. J, anser n>[honom, själv vara intelligent] 2a. J, bad M, n>[PRO, kritisera sig,] 2b. J, bad M, n>[PRO, kritisera honom,] 2c. J, bad M, n>[PRO, kritisera sig, själv] 2d. J, bad M, n>[PRO, kritisera honom, själv] 2a'. J, bad M, n>[PRO, kritisera sig,] 2b'. J, bad M, n>[PRO, kritisera honom,] 2c'. J, bad M, [PRO, kritisera sig, själv] 2d'. J, bad M, n>[PRO, kritisera honom, själv] 3a. J, tror cp[M, ska kritisera sig,] BS 3b. ' J, tror Cp[M, ska kritisera honom,] 3c. J, tror CP[M, ska kritisera sig, själv] 3d. J, tror cp[M/ ska kritisera honom, själv] 3a'. J, tror cp[M, ska kritisera sig,] 3b'. J,· tror CP[M, ska kritisera honom,] 3c'. ·®· J, tror cp[M, ska kritisera sig, själv] 3d'. J, tror CP[M, ska kritisera honom, själv]

*! *! **! *

*! *!

*

**!

*! *!

* *

**! *

*! * *

*!

*

**!

*! *!

* *

29. The PROX constraint, which plays no critical role in these cases, is omitted from these tableaux for reasons of space.

211

Disjoint reference and the typology of pronouns (83) Swedish: Local sig la. lb. lc. Id. 2a. 2b. 2c. 2d. 2a'. 2b'. 2c'. 2d'. 3a. 3b. 3c. 3d. 3a'. 3b'. 3c'. 3d'.

3.8

US' J, anser n>[ sig, vara intelligent] J, anser n>[ honom, vara intelligent] J, anser n>[ sig, själv vara intelligent] J, anser n>[ honom, själv vara intelligent] J, bad M, ip[ PRO, kritisera sig,] J, bad M, n>[ PRO, kritisera honom,] J, bad M,· n>[ PRO, kritisera sig, själv] J, bad M, n>[ PRO, kritisera honom, själv] J, bad M,· n>[ PRO, kritisera sig,] J, bad M, n>[ PRO, kritisera honom,] is· J, bad M, ip[ PRO, kritisera sig, själv] J, bad M, ip[ PRO, kritisera honom, själv] J, tror cp[ M,· ska kritisera sig,·] US- J, tror cp[ M,· ska kritisera honom,] J, tror cp[ M, ska kritisera sig, själv] J, tror cp[ M, ska kritisera honom, själv] J, tror cp[ M, ska kritisera sig,] J, tror cp[ M, ska kritisera honom,] ES' J, tror cp[ M, ska kritisera sig, själv] J, tror cp[ M, ska kritisera honom, själv]

BDi bd 2 OBV ECON *! *! **!

*! * *

*!

**i *! *!

* *

**! *!

* *

*!

*

*

**!

*! *!

* *

**!

Attack vs. defend

Having seen how the class of pronominal anaphors exemplified by Swedish sig supports the existence of an independent Obviation constraint, we now draw on its distributional properties for another piece of support for the specifically semantic character of Obviation. Hellan (1988: 108 ff.) noted a puzzling difference between the two verb types illustrated in (84), represented by angripa 'attack' vs. försvara 'defend' (I use again Swedish examples throughout).30 (84) a. Han angrep {*sig/sig själv}. b. Han försvarade {sig/sig själv}.

'He attacked himself.' 'He defended himself.'

Verbs of the attack type require the complex reflexive sig själv, as would be expected by what we have said so far. Verbs of the defend type unexpectedly may have the bare simple reflexive sig. Replacing help in (26) by such a verb therefore restores the ambiguity:

30. A similar contrast is reported for Dutch (Everaert 1986: 204).

212

Paul Kiparsky

(85) Generalen, tvingade översten, att försvara sig,j. 'The general forced the colonel to defend him/himself.' Additional examples of each type of verb are given in (86): (86) a.

Verbs requiring long reflexives: hata sig själv 'hate oneself ,föredra sig själv 'prefer oneself, undersöka sig själv 'examine oneself ,fÖrstä sig själv 'understand oneself b. Verbs allowing short reflexives: tvätta sig 'wash (oneself)', raka sig 'shave (oneself)', gomma sig 'hide (oneself)', rädda sig 'save oneself',förnedra sig 'demean oneself

Hellan (1988: 109ff.) proposes to explain the defend verbs as being intransitives of inherently reflexive form, analogous to such inherently reflexive intransitive verbs as angra sig 'repent', skynda sig 'hurry'. The reason they can have bare sig then would be that their sig is not an argument, hence a fortiori not a coargument of anything. Hellan's most convincing piece of evidence that defend-type reflexives are syntactically intransitive is that they pattern with intransitive verbs in being able to occur in presentational constructions: (87) a.

*Det tvättade mig en grupp soldater vid stranden. 'There washed me a group of soldiers by the shore.' b. Det tvättade sig en grupp soldater vid stranden. 'There washed themselves a group of soldiers by the shore.' c. Det satt en grupp soldater vid stranden. 'There sat a group of soldiers by the shore.'

There is however at least as good evidence that the defend-type reflexives are also syntactically transitive, with sig functioning as a real object. First, the reflexive can be conjoined with a full lexical NP object: (88) Han tvättade sig och sina barn. 'He washed himself and his children.' Secondly, unlike the sig of inherently reflexive verbs, the sig of defend-type verbs can have a long-distance antecedent (contrast (89) with (85)). (89) Generalen, tvingade overstep att skynda sig*,j. 'The general forced the colonel to hurry.' Third, while inherently reflexive verbs can't share their sig, either with each other (see (90a)) or with defend-type verbs (see (90b)), defend-type verbs can do so freely with each other (see (90c)): (90) a. *Han ângrade och skyndade sig. b. *Han ângrade och rakade sig. c. Han tvättade och rakade sig.

'He repented and hurried.' 'He repented and shaved.' 'He washed and shaved.'

The data in (85), (88), and (90) only make sense if defend-type reflexives can be transitive too.

Disjoint reference and the typology of pronouns

213

So defend-type reflexives must have a dual argument structure, transitive and intransitive (as Hellan himself suggests). That is, sig with these verbs is ambiguously either an object or an intransitivizing clitic which lexically saturates the object's θ-role. It follows that the bare sig in the transitive (90c) is a coargument of its antecedent, in apparent violation of Obviation. So the ability of defend-type verbs to take bare sig cannot be explained solely by their intransitivity, for the transitive variants of these verbs have exactly the same puzzling property. There is more evidence that points to the same conclusion. Consider (91): (91) Frisören rakar vanligtvis sig (själv) först och kundema barber.the shaves usually self (himself) first and customers.the efterât. afterwards 'The barber usually shaves himself first and the customers afterwards.' Clearly (91) is transitive, for each of the conjuncts sig and kunderna has to be an argument in its own right, yet själv is not required in the first conjunct. Moreover, sig can be clefted with defend-type verbs, and not with inherently reflexive verbs. Clefted sig obviously has to be an argument, yet can stand on its own without själv:31 (92) Det var sig (själv) som han försvarade/*ängrade. 'It was himself that he defended/repented'. The conclusion is that defend-type verbs probably have both a transitive and an intransitive variant, but the reflexive form of both is exempt from Obviation. We must look elsewhere for an explanation. Since Obviation is a constraint on semantic interpretation (section 2) it is reasonable to look for a semantic solution to the problem posed by the defend-type verbs as well. The key assumption will be that their internal arguments are proximate. This property will be used to explain both why they can be intransitive and why, when transitive, their reflexive object needs no själv. Suppose the special property of defend-type verbs is that their objects are inherently proximate, i.e. [-obviative]. Then, if the reflexive pronoun, normally obviative, is proximate in (84b), it is free to be coreferential with the subject there. As for the intransitive version of these verbs, let us assume that it is the result of lexical reflexivization. Clearly lexical reflexives are always proximate. If follows that lexical reflexivization is available just for this class of verbs. The difference between the defend and attack classes undoubtedly has something to do with whether the activities they denote are stereotypically "other-directed". The following contrast is suggestive: (93) a. John's defense (possibly = 'John's defense of himself) b. John's attack (Φ 'John's attack on himself)

31. See von Bremen (1984: 220) for more evidence that sig can have argument status in Swedish. The second sentence might be acceptable as a correction.

214

Paul Kiparsky

The defend-verbs typically denote physical activities, two important subclasses being "grooming" verbs such as 'wash' and 'shave' and verbs denoting position and orientation such as 'rise' and 'lean', independently of agentivity (Kemmer 1993). Along the lines of the pragmatic accounts suggested by Faltz (1976), Kemmer (1993), and Levinson (1991), one could say that the suspension of Obviation is inapplicable to stereotypically other-directed actions, which therefore would not require själv to prevent Obviation if coreference is in fact intended, and for the same reason could be intransitivized by θ-role fusion. While this is not implausible as an explanation for the origins of the distinction, it seem that it must be viewed as grammaticized to some extent, at least in the Scandinavian languages. If it were purely pragmatic in nature, it should be defeasible: that is, it should be possible to coerce a bare sig by appropriate manipulation of the context or situation. For example, if själv is motivated by a defeasible presumption of otherdirectedness, then in a discussion where specifically self-hatred is the topic it ought to be possible to say hata sig, without själv, which is not the case, I believe. 32

4.

Real and apparent complementarity failures

Formulations of Binding Theory diverge greatly in how much overlap they allow in the distribution of anaphors and pronominals. The predictions range from strict complementarity (e.g. Chomsky 1981) to several different kinds of restricted overlap (e.g. Huang 1983, Chomsky 1986, Dalrymple 1993) to quasi-independent distribution (e.g. Wexler & Manzini 1987, who allow pronoun-specific domain parametrizations for Principle A and Principle B, though restricted by certain global constraints; see below.) Strict complementarity is surely too strong: most languages show "optional reflexivization" in at least some contexts. But obviously not every conceivable kind of variation actually occurs in languages. The question is how to constrain the variation in a principled way. The proposed approach to anaphora has some direct consequences in this respect which are borne out by the cross-linguistic data. Consider in particular the distribution of ordinary pronominal elements in relation to that of reflexives. Because of Obviation and Blocking, they should never overlap in local domains. We must now consider some apparent classes of systematic exceptions to this claim.

32. However, it is sometimes possible to accommodate bare sig by focusing on a particular meaning of the verb. A striking example is the observation of Hellan (1988) that the verb meaning 'admire', normally an attack-type verb, allows the bare reflexive if it refers to admiration of the physical self (the nonstative sense, as in 'John was admiring himself in the mirror.')

Disjoint reference and the typology of pronouns

4.1

215

Snake sentences

There is apparent free variation between anaphors and pronominale in adverbial phrases of the type illustrated in (94): (94) a. John saw a snake, near him, /himself,. b. Mary, wrapped the blanket around her, /herself,. Some theories of anaphora have been set up in such a way as to predict this variation. The present approach implies that there is no free variation here, but a structural difference. The PP can be either an argument, (an adverbial or a prepositional object), in which case we get a reflexive, or functionally part of the predicate, in which case we get a pronominal. Evidence for this view comes from cases where only the pronominal is possible and cases where only the reflexive is possible. They have clearly distinguishable structural characteristics. The following descriptive generalizations hold: (95) a. A reflexive is possible if and only if a referential expression can be substituted for it. b. A pronominal is possible if and only if the PP can be predicated of the object. This is illustrated by the following bits of discourse: (96) a. John, aimed the gun at *him, /himself,. 1. *The gun is at him now. 2. Last time he aimed it at Fred. b. John, brought the gun with him,/*himself(. 1. The gun is with him now. 2. *Last time he brought it with Fred. In (96a), John's gun is not "at him" (no predication) but it is aimed "there" (locative argument), so the reflexive is obligatory. In (96b), on the other hand, John's gun is "with him" (predication) but it is not brought "there", so the pronominal is obligatory. We can test these generalization on the following contrasts (Kuno 1987: 66): (97) a.

John, wrote to Mary about *him,/himself,. (*Mary is about him. He also wrote about Fred.) b. John, referred Mary to *him,/himself,. c. John, compared Bill with *him,/himself,. d. John, addressed the letter to *him,/himself,. e. John, fell in love with *him, /himself,.

(98) a.

John, has many friends around him,/*himself(. (Many friends are around him. *He also has friends around Fred.) b. John, left his family behind him, /*himself,. c. John, has an air of aloofness about him,/*himself,.

216

Paul Kiparsky d. John, has passion in him, /*himself¡. e. John, has no gumption in him,/*himself,.

That the reflexive is obligatory in cases (96a) and (97) is not surprising. This is what would be expected on most theories, including the present one. It is the pronominal cases such as (96b) and (98) which require an explanation. A small clause analysis is implausible, for reasons discussed by Chomsky (1981) and others. For example, it would not account for the fact that the pronoun in this construction has the properties of a bound anaphor, as revealed by obligatory sloppy identity and non-substitutability of epithets: (99) a. John brought the gun with him, and so did Mary, (unambiguous) b. *John, brought the gun with the idiot,. It would also fail to generalize to cases like (100), where predication is presumably out of the question: (100) John looked around him. Another peculiarity of sentences like (96b), (97), and (100) that needs to be explained is the fact that the pronoun cannot be stressed. I will assume that the argument structure of these sentences specifies the PP as containing a proximate non-reflexive pronoun, which (as the constraints dictate) must be interpreted as a bound anaphor. This is a remnant in Modern English of the much wider proximate use of the pronominale in Old and Middle English seen in (74) ff. Semantically, the PP in effect joins with the verb into a combination that functionally constitutes a single predicate. That is, John left his family behind him is parallel to John left his family behind, and so on. In contrast the reflexive in John aimed the gun at himself is an independent argument: (101) a. Reading of (96a): λγ λχ [χ aimed the gun at >>] (John) b. Reading of (96b): λχ [χ brought the gun with x] (John) The peculiarity of the construction is that the bound anaphor is expressed by a clitic pronoun agreeing with the predicate's subject. As in the case of the Swedish attack vs. defend contrast, the fact that Obviation is sensitive to the lexical semantics of the predicate confirms its status as a semantic constraint.

4.2

Processing factors

A similar prima facie problem for Blocking approaches (including the ECONOMY constraint adopted here), which on closer analysis turns out to be an argument for Obviation, appears in several languages including Swedish. Even though Blocking is the norm in Swedish, it is sometimes suspended for possessives in casual usage, so that they can, contrary to the general rule, corefer with the subject of the same clause. This

Disjoint reference and the typology of pronouns

217

happens particularly under three conditions (Wellander 1959: 234ff.). 33 First, if the subject is passivized, especially when it is also non-human: (102) Ordet realism, är här taget i sitt,/dess, ursprungliga bemärkelse. word.the realism is here taken in self s/its original sense 'The word "realism" is here understood in its original sense.' Secondly, if the pronoun precedes its antecedent (i.e. if it is contained in a topicalized constituent): (103) Nere i Skâne, hos sin. / hennes, väninna friherrinnan C., hade hon, down in Scania, at self's/her friend countess.the C., had she âdragit sig en luftrörskatarr. contracted self a bronchitis 'Down in Scania, at self's/her friend the countess of C's (place), she had contracted bronchitis.' And third, if the pronoun is "distant" from its antecedent: (104) Han, hade träffat henney, dâ de bâda var helt unga, hos He had met her, when they both were quite young, at sin,/hans, gamie vän Β. self's/his old friend B. 'He had met her, when they both were quite young, at self s/his old friend B's place.' If this is a processing effect, we would expect it to occur in other languages as well where the same conditions obtain. In fact, similar variation in genitives occurs under corresponding conditions in other Scandinavian languages. Counterparts from Danish prose are cited in Diderichsen (1937: 52-53). One Icelandic consultant judged the pronominal to be a marginal option for the Icelandic versions of all three sentences; the other did so for (102) and rejected it for (103) and (104). In Russian, the possessive reflexive svoj, even with a third person antecedent, can sometimes alternate with the pronominal under similar special conditions (Yokoyama 1975). Wellander further observes that when several of these factors coincide, the pronominal "can satisfy even a more fastidious linguistic intuition": (105) Flickan, tvingas just under sin,/hennes, ômtâligaste âlder girl.the is.forced just under self s/her most.sensitive age att byta skola.34 to change school 'The girl is forced precisely at self s/her most sensitive age to change school.'

33. Wellander's study is based on sentences culled from journalistic prose. His material is reviewed and discussed from a generative semantics perspective by Hellberg (1980). 34. Here Wellander points out that the pronominal is more natural when the adverbial clause is separated by commas.

218

Paul Kiparsky

(106) ... faster Lotta,, som av familjen mot sin/hennes, vilja och ... aunt Lotta, who by family.the against self's/her will and trots sin,/hennes, ungdomskärlek giftes bort med brukspatron in.spite.of self's/her youth.love was.married away with factory .boss pâ J. on J. '... aunt Lotta, who was married off to the factory boss at J. by her family against self's/her will and in spite of self s/her youthful love.' Presumably what is going on here is that under conditions of processing complexity the syntactic blocking effect is weakened. Crucially, the same conditions have no weakening effect on Obviation. Thus, in sentences comparable to (102)-(106), non-possessive pronominals cannot be understood as coreferential with a coargument subject, as in (107): (107) Läsaren, mäste konfronteras med sig, själv/*honom, själv. 'The reader must be confronted with himself.' The fact that Economy is sensitive to syntactic and processing factors, while Obviation is unaffected by them, is another argument that these constraints are distinct.

4.3

Is the disjoint reference domain universal?

I have argued that the Obviation rule is universal, in the sense that if an elements is obviative, is is always disjoint in reference from its coarguments. Can disjoint reference also apply in some domain larger than the coargument domain? It has been explicitly claimed that it can (e.g. Harbert 1986, Wexler & Manzini 1987, Kapur et al. 1993, Sells 1991). The existence of such cases would require specifying additional language-specific syntactic or semantic disjoint reference domains, against the spirit and letter of the present proposal. Let us look at some of the instances where disjoint reference has been argued to apply in larger domains. Sells (1991) suggests that the pronominals as subjects of gerunds in sentences like (108a) are ungrammatical because of a disjoint reference requirement, which does not apply to subjects of "factive" gerunds, as in (108b): (108) a. John, prefers (*his,) going to the movies. (= (55b)) b. Only Churchill remembers his giving the speech about blood, sweat, and tears. He proposes that disjoint reference actually applies not just to coarguments but in a more inclusive semantic domain, the "situated domain". If this can replace the coargument domain as the domain of disjoint reference universally, the above results would go still go through. But it is not clear that this modification is necessary. Sells' central evidence for non-coargument disjoint reference could also be dealt with in another way. Suppose that nonfactive gerunds like (108a) are control structures, functionally equivalent to infinitival complements (Kiparsky & Kiparsky 1971: 357). They would

Disjoint reference and the typology of pronouns

219

thus be handled by Economy rather than by an extension of disjoint reference to a wider domain. As far as I am aware, the only potential difficulty for this proposal is that some speakers, according to Sells, exclude coreference even in cases where control is presumably out of the question. Such speakers, then, would reject sentences of the type: (109) *Maria ; told John, about him, losing the bet. Here a Blocking account is also possible: for speakers with these judgments, we may suppose that him is blocked by his. In a number of systems with long-distance binding, the disjoint reference domain is said to be the Finite clause. Icelandic is reported to be such a system (Thráinsson 1991, Kapur et al. 1990), a particularly interesting one because it allows reflexives to be bound in the Root domain. It seems, though, that there may be more variation in usage, at least for some speakers. My consultant exactly replicated the Swedish/Russian pattern of "optional long-distance reflexivization" we saw in (25): (110) a.

Jón, neyddi Gunnar, til aö skipa Pâli* aö raka hann, ; *k. 'John forced Gunnar to order Paul to shave him.' b. Hann, hjálpaói vinkonu, sinni, aö kveikja í sigarettuni sinni, ; 7hans, * ; / hennar*,j. 'He helped his date to light self s/his/her cigarette.'

And Hyams (according to Grimshaw & Rosen 1990: 202) has found that adult Icelandic speakers presented with such sentences actually prefer assigning the pronominal a long-distance antecedent to interpreting is as referring to some discourse element. In Gothic, reflexives are almost always used in the Finite domain (Kapur et al. 1990), but there is at least one example of a coreferential pronominal in this domain, so the case is not clear-cut. 35 For Russian, Peshkovskij (1956: 164) states that, while reflexive pronouns are optional in long-distance binding in modern Russian, they were required there in earlier stages of Slavic. Latin is sometimes said to be a language in which long-distance reflexives are obligatory, but as the grammars make clear, they are merely preferred, with variation conditioned by point of view (logophoricity). The following example shows that a pronominal can very well have an antecedent in the finite domain. 36 (111) ut eum in Syria sisterent, orabat that him in Syria place.3PL.SUBJ begged.3SG (Tacitus, H. 2,9) 'He begged (them) to place him in Syria.' The cross-linguistic similarity of long-distance binding patterns is underscored by preferences that look like processing effects which influence the accessibility of long35. Mark 1.10 Gasahv usluknans himinans jah ahman swe ahak atgaggandan ana ina 'He saw the heavens open up and the Holy Ghost come towards him [pronominal] like a dove.' (Diderichsen 1937: 59). 36. See Kühner & Stegmann (1966: 610-612), who also suggest that the striving to avoid ambiguity is involved.

220

Paul Kiparsky

distance binding relations, which seem to work very similarly in unrelated languages. In multiply embedded infinitive constructions, it is relatively hard to get the intermediate subject as a binder (översten in (112); the lowest subject is of course ineligible by Obviation). (112) Generalen, tvingade översten^ att befalla löjtnanten^ att raka sig 'The general forced the colonel to order the lieutenant to shave self.' The identical effect has been reported for Czech and Russian (Bily 1981) and for Chinese (Battistella 1989, Tang 1989: 109,115). We see, then, that the contexts of "optional reflexivization" examined here do not invalidate the Obviation constraint, but on the contrary support it. Needless to say, many other empirical challenges remain to be met before the universality of coargument Obviation can be confidently accepted.

4.4

Conclusion

The distribution of pronominale and reflexives reflects both Obviation, here argued to be a universal constraint which requires coarguments to have disjoint reference, and Blocking, here taken to be an economy constraint. These constraints were shown to interact with a hierarchy of binding domain constraints to generate a typology of pronouns. The constraint system predicts the category of obviative reflexives, previously described empirically in some languages, and explains a surprising relationship between the blocking of coargument reflexives, in the non-coargument case, the ambiguity of those reflexives between the bound anaphora and coreferential readings. The constraint system also predicts the novel category of referentially dependent nonreflexive pronouns, for which we provided cross-linguistic documentation.

References Aissen, Judith (1997). On the syntax of obviation. Language 73, 705-750. Anttila, Arto (1997). Variation in Finnish phonology and morphology. Ph.D thesis, Stanford University. Aoun, Joseph and Yen-hui Audrey Li (1988). Minimal disjointness. WCCFL 7, 29-40. Bach, Emmon and Barbara Partee (1980). Anaphora and semantic structure. In CLS Parasession on pronouns and anaphora, Jody Kreiman and Almerindo Ojeda (eds.), 1-28.

Battistella, Edwin (1989). Chinese reflexivization: a movement to INFL approach. Linguistics 27, 987-1012. Bickerton, Derek (1987). He himself: anaphor, pronoun, or. . . ?. Linguistic Inquiry 18, 345-348.

Disjoint reference and the typology of pronouns

221

Bfly, Milan (1981). Intrasentential pronominalization and functional sentence perspective. Lund: Lund Slavonic Monographs. Bok-Bennema, Reineke (1984). On marked pronominal anaphors and Eskimo pro. In Grammatical Representation, Jacqueline Guéron, Hans-Georg Obenauer, and JeanYves Pollock (eds.), 1-18. Dordrecht: Foris. Bouchard, Denis (1983). The Avoid Pronoun Principle and the Elsewhere Principle. NELS 13, 29-36. Bresnan, Joan, Per-Kristian Halvorsen, and Joan Maling (1985). Logophoricity and bound anaphors. MS, Stanford. Broadwell, George Aaron (1988). Reflexive movement in Choctaw. NELS 18, 53-64. Burzio, Luigi (1996). The role of the antecedent in anaphoric relations. In Current issues in comparative grammar, Robert Freidin (ed.), 1-45. Dordrecht: Kluwer. —(1998). Anaphora and soft constraints. In Is the best good enough? Optimality and competition in syntax, Pilar Barbosa, Danny Fox, Paul Hagstrom, Martha McGinnis, and David Pesetsky (eds.), 93-113. Cambridge, Mass.: ΜΓΓ Press. Cardinaletti, Anna, and Michal Starke (1996). Deficient pronouns: A view from Germanic. A study in the unified description of Germanic and Romance. In Studies in Comparative Germanic Syntax Π, Höskuldur Thráinsson, Samuel David Epstein, and Steve Peter (eds.), 21-65. Dordrecht: Kluwer. Cho, Young-Mee Yu (1985). An LFG analysis of the Korean reflexive caki. In Harvard Studies in Korean Linguistics, Susumu Kuno et al. (eds.), 3-13. Seoul: Hanshin. Chomsky, Noam (1981). Lectures on government and binding. Dordrecht: Foris. —(1986). Knowledge of Language. New York: Praeger. Chung, Sandra (1989). On the notion "null anaphor" in Chamorro. In The null subject parameter, Osvaldo Jaeggli and Kenneth J. Safir (eds.), 143-184. Dordrecht: Kluwer. Clements, George N. (1975). The logophoric pronoun in Ewe: its role in discourse. Journal of West African Languages 10, 141-177. Cole, Peter, Gabriella Hermon, and Li-May Sung (1990). Principles and parameters of long-distance reflexives. Linguistic Inquiry 21, 1-22. Coleman, Carolyn (1987). Reflexive morphology in Kunparlang. In Working papers in grammatical theory and discourse structure: interactions of morphology, syntax, and discourse, Masayo Iida, StephenWechsler, and Draga Zee (eds.). Stanford: C S U Publications. Dahlstrom, Amy (1988). Independent pronouns in Fox. In In honor of Mary Haas, William Shipley (ed.), 165-194. Berlin: Mouton. —1991). Plains Cree morphosyntax. Ph.D. Thesis, Berkeley. Dalrymple, Mary (1993). The syntax of anaphoric binding. Stanford: CSLI Publications. —; and Stuart Shieber (1991). Ellipsis and higher-order unification. Lingustics and Philosophy 14, 399-452. Diderichsen, Paul (1937). Om pronominerne sig og sin. Acta Philologica Scandinavica 13, 1-92.

222

Paul Kiparsky

Dowty, David (1980). Comments on the paper by Bach and Partee. In CLS Parasession on pronouns and anaphora, Jody Kreiman and Almerindo Ojeda (eds.), 29-40. Enç, Mürvet (1983). Anchored Expressions. WCCFL 2, 79-88. —(1989). Pronouns, licencing, and binding. Natural Language and Linguistic Theory 7, 51-92. Everaert, Martin (1986). The syntax of reflexivization. Dordrecht: Foris. Faltz, Leonard M. (1977). Reflexivization: a study in universal syntax. U.C. Berkeley Dissertation. Reprinted 1985, Garland, New York Farmer, Ann K. and Robert H. Harnish (1987). Communicative reference with pronouns. In The pragmatic perspective, Marcella Bertucelli-Papi and Jef Verschueren (eds.), 547-65. Amsterdam: Benjamins. Farrell, Patrick, Stephen A. Marlett, and David A. Perlmutter (1991). Notions of subjecthood and switch reference: evidence from Seri. Linguistic Inquiry 22, 431-456. Giorgi, Alessandra (1987). The notion of Complete Functional Complex: some evidence from Italian. Linguistic Inquiry 18, 511-518. —(1983). Toward a theory of long-distance anaphors. The Linguistic Review 3, 307361. Grafstein, Ann (1988). The Algonquian obviative and the Binding Theory. NELS 18, 175-192. —(1989). Disjoint Reference in a 'Free Word Order' Language. In Native American Languages in Theoretical Perspective, Donna Gerdts and Karen Michelson (eds.), 163-75. Albany: SUNY Press. Grimshaw, Jane (1990). Argument structure. Cambridge, Mass.: ΜΓΓ Press. —; and Sara Thomas (1990). Knowledge and obedience: the developmental status of the Binding Theory. Linguistic Inquiry 21, 187-222. Hagège, Claude (1974). Les pronoms logophoriques. Bulletin de la Société de Linguistique de Paris 69, 287-310. Haiman, John and Pamela Munro (1983). Switch reference. Typological studies in language 2. Amsterdam: Benjamins. Harbert, Wayne (1983). Germanic reflexives and the implementation of binding conditions. In Language Change, Irmengard Rauch and Gerald F. Carr and (eds.), 89-127. Bloomington: Indiana University Press. —(1986). Markedness and the bindability of subject of NP. In Markedness, Fred F. Eckman, Edith A. Moravcsik, and Jessica R. Wirth (eds.), 139-54. New York: Plenum Press. —(1990). Binding, SUBJECT, and Accessibility. In Proceedings of the Princeton workshop on comparative grammars, Robert Freidin (ed.), 29-55. Cambridge, Mass: ΜΓΓ Press. Heath, Jeffrey (1983). Reference tracking in Nunggubuyu. In John Haiman and Pamela Munro (eds.), 129-49. Heim, Irene, Howard Lasnik, and Robert May (1991). Reciprocity and plurality. Linguistic Inquiry 22, 63-101. Hellan, Lars (1988). Reflexives in Norwegian and the theory of grammar. Dordrecht: Foris.

Disjoint reference and the typology of pronouns

223

—(1990). A principle of global binding. In Grammar in progress: GLOW essays for Henk van Riemsdijk, 187-198. Dordrecht: Foris. Hellberg, Staffan (1980). Reflexiveringen som ledtrâd till svenskans struktur. Gothenburg Papers in Theoretical Linguistics 42. Herslund, Michael (1986). The double object construction in Danish. In Topics in Scandinavian syntax, Lars Hellan and Kristi Koch Christensen (eds.), 127-47. Dordrecht: Foris. Hestvik, Arild (1990). Against the notion 'governing category'. WCCFL 9, 263-275. Higginbotham, James (1983). Logical form, binding, and nomináis. Linguistic Inquiry 14, 355-420. —(1985). On semantics. Linguistic Inquiry 16, 547-593. Hollenbach, Barbara (1984). Reflexives and Reciprocals in Cópala Trique. UAL 50, 272-291. Hong, Ki-Sun (1989). Consciousness condition on the Korean reflexive Caki. BLS 15, 161-169. Huang, C.T. James (1983). A note on the binding theory. Linguistic Inquiry 14, 554561. Huang, Yan (2000). Anaphora: A cross-linguistic study. Oxford: Oxford University Press. Iatridou, Sabine (1986). An anaphor not bound in its governing category. Linguistic Inquiry 17, 766-772. Joshi, Smita (1989). Logical subject in Marathi grammar and predicate argument structure. WCCFL 8, 207-219. Kapur, Shyam, Barbara Lust, Wayne Harbert, and Gita Martohardono (1993). Universal Grammar and leamability theory: the case of binding domains and the 'Subset Principle'. In Knowledge and language: issues in representation and acquisition, Eric Reuland and Werner Abraham (eds.), 185-216. Dordrecht: Kluwer. Keenan, Edward L. (1974). The functional principle: generalizing the notion of 'subject o f . CLS 10, 278-309. Kemmer, Suzanne (1993). The middle voice. Amsterdam: Benjamins. Kiparsky, Paul and Carol Kiparsky (1971). Fact. In Semantics, Danny D. Steinberg and Leon A. Jakobovits (eds.), 345-69. Cambridge: Cambridge University Press. Klenin, Emily (1980). Sentential and discourse prominence. Russian Linguistics 4, 269-280. König, Ekkehard (1991). The meaning of focus particles: A comprehensive perspective. London: Routledge. Koopman, Hilda and Dominique Sportiche (1989). Pronouns, logical variables, and logophoricity in Abe. Linguistic Inquiry 20, 555-588. Koster, Jan (1986). Domains and dynasties. Dordrecht: Foris. Kroeger, Paul (1993). Phrase structure and grammatical relations in Tagalog. Stanford: CSLI Publications. Kühner, Raphael and Carl Stegmann (1966). Ausführliche Grammatik der lateinischen Sprache. Hannover: Verlag Hahnsche Buchhandlung.

224

Paul Kìparsky

Kuno, Susumu (1972). Pronominalization, reflexivization, and direct discourse. Linguistic Inquiry 3, 161-95. —(1987). Functional syntax. Chicago: University of Chicago Press. Lasnik, Howard (1976). Remarks on coreference. Linguistic Analysis 2, 1-22. —(1989). On the necessity of binding conditions. In Essays on anaphora, Howard Lasnik, 149-67. Dordrecht: Kluwer. Lebeaux, David (1985). Locality and anaphoric binding. The Linguistic Review 4, 343363. Lees, R.B. and E.S. Klima (1963). Rules for English pronominalization. Language 39, 17-28. Legendre, Geraldine, Paul Smolensky, and Colin Wilson (1996). When is Less More? Faithfulness and Minimal Links in wh-Chains. Rutgers Optimality Archive (ROA 117). Levinson, Stephen C. (1987). Pragmatics and the grammar of anaphora: a partial pragmatic. reduction of binding and control phenomena. Journal of Linguistics 23, 379434. —(1991). Pragmatic reduction of the binding conditions revisited. Journal of Linguistics 27,107-161. —(1991). Presumptive meanings. Cambridge, Mass.: Bradford Books. Lichtenberk, Frantisek (1983). Manam. Oceanic Linguistics Special Publications 18, University of Hawaii Press. Mitchell, Bruce (1985). Old English syntax. Oxford: Clarendon Press. Mohanan, K.P. (1982). Grammatical relations and anaphora in Malayalam. In MIT working papers in linguistics 4, Alec Marantz and Tim Stowell (eds.), 163-190. ΜΓΓ: Linguistics Department. Moyne, John (1971). Reflexive and emphatic. Language 47, 141-163. Napoli, Donna Jo (1989). Predication theory: a case study for indexing theory. Cambridge: Cambridge University Press. Nerbonne, John (1982). German impersonal passives: a non-structure-preserving lexical rule. WCCFL 1, 341-352. Ogura, Michiko (1989). Verbs with the reflexive pronoun and constructions with self in old and early Middle English. Cambridge, Mass: D.S. Brewer. Partee, Barbara H. (1989). Binding implicit quantifiers in quantified contexts. CLS 25, 342-65. Peshkovskij, Aleksandr Matveevich (1956). Russkii sintaksis ν nauchnom osveshchenij. Gosudarstvennoe Uchebno-pedagogichestvoe Izdatelstvo Ministerstva Moskva: RSFSR. Pica, Pierre (1987). On the nature of the reflexivization cycle. NELS 17, 483-499. Pollard, Carl and Ivan Sag (1983). Reflexives and reciprocals in English: an alternative to the Binding Theory. WCCFL 2, 189-203. —;— (1992). Anaphors in English and the scope of Binding Theory. Linguistic Inquiry 23, 261-303. Reinhart, Tanya (1983a). Coreference and bound anaphora: a restatement of the anaphora question. Linguistics and Philosophy 6, 47-88.

Disjoint reference and the typology of pronouns

225

—(1983b). Anaphora and semantic interpretation. London: CroomHelm. —; and Eric Reuland (1993). Reflexivity. Linguistic Inquiry 28, 178-87. Rigau, Gemma (1986). Some remarks on the nature of strong pronouns in null-subject languages. In Generative studies in Spanish syntax, Ivonne Bordelois, Heles Contreras, and Karen Zagona (eds.), 143-63. Dordrecht: Foris. Roberts, John R. (1988). Amele switch-reference and the theory of grammar. Linguistic Inquiry 19, 45-64. Rose, Marilyn Louise (1976). The syntax of the reflexive pronoun in Gothic and Old Norse. Ph.D. Thesis, University of Wisconsin-Madison. Sag, Ivan and Jorge Hankamer (1984). Towards a theory of anaphoric processing. Linguistics and Philosophy 7, 325-345. Saxon, Leslie (1986). The syntax of pronouns in Dogrib. Ph.D. Thesis, University of California, San Diego. Schwartz, Linda J. and Timothy Dunnigan (1986). Pronouns and pronominal categories in Southwestern Ojibwe. In Pronominal systems, Ursula Wiesemann (ed.), 285-322. Tübingen: Narr. di Sciullo, Anna-Maria and Edwin Williams (1987). On the definition of word. Cambridge, Mass.: ΜΓΓ Press. Sells, Peter (1986). Coreference and bound anaphora: a restatement of the facts. Ν ELS 16, 434-416. —(1987). Aspects of logophoricity. Linguistic Inquiry 18, 445-479. —(1991). Disjoint reference into NP. Linguistics and Philosophy 14, 151-69. —; Annie Zaenen, and Draga Zee (1987). Reflexivization variation: relations between syntax, semantics, and lexical structure. In Working Papers in Grammatical Theory and Discourse Structure, Masaiyo Iida, Stephen Wechsler, and Draga Zee (eds.), 169-238. Stanford: CSLI Publications. Slupesky, Ted (1989). Deixis and anaphora in the Homeric dialect. B.A. Thesis, Reed College. Tang, Chih-Chen Jane (1989). Chinese reflexives. Natural Language and Linguistic Theory 7, 93-121. Thompson, Laurence C. (1965). A Vietnamese grammar. Seattle: University of Washington Press. Thráinsson, Höskuldur (1979). On complementation in Icelandic. New York: Garland. —(1991). Long-distance reflexives and the typology of NPs. In Long-distance anaphora, Jan Köster and Eric Reuland (eds.), 49-75. Cambridge: Cambridge University Press. Tiersma, P.M. (1985). Frisian reference grammar. Dordrecht: Foris. Timberlake, Alan (1979). Reflexivization and the cycle in Russian. Linguistic Inquiry 10, 109-141. —(1980). Reference conditions on Russian reflexivization. Language 56, 777-796. Toivonen, Y.H. (1955). Suomen kielen etymologinen sanakirja I, Helsinki: SuomalaisUgrilainen Seura. Tremblay, Mireille (1990). An argument sharing approach to ditransitive constructions. WCCFL 9, 549-563.

226

Paul Kiparsky

Vikner, Sten (1985). Parameters of binder and of binding category in Danish. Working Papers in Scandinavian Syntax 23, 1-61. Lund. Walker, Alastair G. H. (1989). Frisian. In The dialects of Modern German, Charles Russ (ed.), 1-30. Stanford: Stanford University Press. Wellander, Erik (1959). Riktig svenska 3. Stockholm: Svenska Bokförlaget. Wexler, Ken and Rita M. Manzini (1987). Parameters, Binding Theory, and learnability. Linguistic Inquiry 18, 413-444. Wiesemann, Ursula (1986). Grammaticalized coreference. In Pronominal systems, Ursula Wiesemann (ed.), 437-63. Tübingen: Narr. Yadurajan, K.S. (1987). Binding theory and reflexives in Dravidian. MS, CIEFL, Hyderabad. Yang, Dong-Whee (1983). The Extended Binding Theory of anaphors. Language research 19, 169-192. Yokoyama, Olga (1975). Personal or reflexive? A functional analysis. Harvard studies in syntax and semantics I, 72-112. Cambridge, Mass.: Harvard University. Zribi-Hertz, A. (1989). Anaphor binding and point of view: English reflexive pronouns in. sentence and discourse. Language 65, 695-727.

Quirky "subjects" and other specifiers

Gisbert Fanselow

1.

Introductory remarks

*

Theoretical concepts which were useful during a certain period in the progression of science often have to be abandoned later to give way to a necessary shift of perspective. The notion "quirky subject" (subjects bearing oblique Case) is, arguably, a case in point. It has guided important empirical research, with a major emphasis on Icelandic. The recent proliferation of "quirky subjects" to further languages has, however, eroded the fundament on which the concept stands. The distinction between languages with and without "quirky subjects" turned out not to be based on a large set of diagnostics that always go hand in hand. Rather, in many languages, only some of the criteria are fulfilled, while the others are not, often because the criteria are inapplicable. Thus, the once clear picture of the grammatical landscape around "quirky subjects" is blurred. Drawing a clear-cut line between constructions that exemplify "quirky subjects" and those that do not therefore involves some arbitrariness. In particular, the discussion of "quirky subjects" suffers from the fact that there is no easily identifiable and grammatically meaningful notion of "subject", which could decide which of the criteria are essential, a point that was stressed by Sigurösson (2000a: 27). Questioning the usefulness of the notion of a subject (as part of grammar in addition to Case and the hierarchy of arguments) thus appears more fruitful than further attempts of answering the question "Is X a quirky subject?" on partially arbitrary grounds. In this spirit, Wunderlich (2001) has developed a contrastive analysis of Icelandic and German that * This paper has profited substantially from discussions with Joanna Blaszczak, Ina Bornkessel, Karin Donhauser, Caroline Féry, Susann Fischer, Werner Frey, Gunnar Hrafn Hrafnbjargarson, Diana Pili, Florian Schäfer, Matthias Schlesewsky, Halldór Sigurösson, Arthur Stepanov, and Ralf Vogel, and from the comments of the editors and an anonymous reviewer. Several of the analyses in this paper have been presented in my lectures at the Aristotle University at Thessaloniki (9/2001), at the Head Movement Workshop at UCLA (10/2001), and at the Workshop on Adverbs in Wuppertal (11/2001). I am grateful to the audiences for helpful comments. The research reported here was partially supported by DFG grants to the project A3 of the Research Group "Conflicting Rules" at the University of Potsdam, and to the project "Nonstructural Case" at the University of Potsdam and the Humboldt-University at Berlin.

228

Gisbert Fanselow

avoids reference to the notion of (quirky) subjects. The present paper is in line with his view, it elaborates his basic insights in a different framework. It also owes a lot to the insights Bhatt (1999) formulates for Kashmiri. On the one hand, Reis (1982) has shown that the laws governing German syntax can be formulated properly in terms of Case and argument roles/hierarchies, while the notion "subject" is superfluous. On the other hand, the explanation of unmarked constituent order in German verb second main clauses justifies the postulation of a specific structural position that hosts the highest argument of a predicate (among other elements). In the model proposed here, Icelandic differs minimally from German in that this position has to be filled in embedded clauses as well. The further differences between Icelandic and German concerning this position, which were noted, e.g., by Zaenen et al. (1985), can be reduced to factors involving Case, or they are due to the absence of particular construction types in German. The postulation of "quirky subjects" is thus not warranted. In this respect, we concur with Wunderlich (2001), but details of our analyses differ. The paper is organized as follows. In the next section, we introduce the notion of "quirky subjects", and discuss some of the data that seem to motivate the postulation of a major difference between Icelandic and German in the domain of subjects. Section 3 develops and justifies an analysis of German normal word order (see Lenerz 1977), which integrates fundamental insights of Travis (1984), Rizzi (1997), and Pili (2001). Section 4 shows how Icelandic fits into the model developed for German. In section 5, we discuss the criteria for labelling a phrase as a "quirky subject" one by one. It will be shown that the empirical facts do not support the view that Icelandic, but not German, possesses such a grammatical function. Either both languages have quirky subjects, or none of them does. In the concluding section, we argue that the latter interpretation is the more promising one.

2.

Setting the stage

In Icelandic and German, noun phrases bearing a Case different from nominative can appear in preverbal/sentence initial position. Given that the two languages allow the topicalization of objects, the presence of oblique (non-nominative) noun phrases in clause initial position constitutes no surprise by itself. In both Icelandic and German, there are, however, constructions in which an oblique noun phrase needs no special pragmatic licensing for occupying the preverbal/clause-initial position. This is the case for sentences such as (1). (1)

a. Honum him.DAT

b. Ihm him.DAT

var

hjalpáó.

was

helped

wurde geholfen. was

'He was helped.'

helped

(Icelandic) (German)

Quirky "subjects" and other specifiers

229

At exactly this point, the traditions of grammatical descriptions of the two languages (within the generative framework) bifurcate. For Icelandic, a battery of tests such as the one listed in (2) (borrowed from Sigurösson 2000a) has been developed (see also Andrews 1982, Zaenen et al. 1985, among others) that seems to establish that the preverbal noun phrase in (la) is a subject, in spite of its being marked with dative case. With respect to the domains mentioned in (2), oblique preverbal noun phrases of the type in (la) behave like standard nominative subjects. For example, they can (and must) be realized as PRO in a control infinitive (criterion (2e)), as (3) illustrates. The idea that non-topicalized preverbal oblique nomináis are "quirky subjects" has become the standard analysis for Icelandic. (2)

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

Reflexivization Subject-verb inversion (in VI and V2 environments) Subject position in ECM infinitives Raising Control (i.e. the ability of being a controllee) Conjunction reduction

(3)

Ég vonaöist til aö PRO veröa hjàlpaô. I hoped for to PRO.DAT be helped Ί hope that one helps me.'

In contrast, the fact that certain oblique noun phrases may occupy the clause-initial position without bearing a special pragmatic function has often been noted for German (see, e.g., Lenerz 1977), but there are hardly any attempts of accounting for it. Den Besten (1985) is a notable exception - insofar as his proposal allows that the dative pronoun in (lb) can occupy the subject position [Spec,IP]. However, the construction in (lb) fails to meet certain of the diagnostics in (2). For example, the dative nominal cannot be realized as PRO in an infinitive in German. (4)

* Ich hoffe geholfen zu werden. I hope helped to be Ί hope that one helps me.'

Such contrasts between Icelandic and German suggest that the dative pronoun ihm in (lb) is not a subject (see Zaenen et al. 1985, Sigurösson 2000a). This view is the standard approach for German, but it leaves an important question unanswered: why is it the case that some objects may always occupy the preverbal position, while others can do so only when they bear special pragmatic functions?

230

3.

Gisbert Fanselow

The prefield: filling the preverbal position in German

The standard analysis of German main clauses is based on insights of Köster (1975) and Thiersch (1978), and is illustrated in (5). It takes the right-peripheral position which the finite verb occupies in embedded clauses (5a) as basic, and derives the second position effect of main clauses by moving the finite verb to an empty Comp slot. In addition, exactly one constituent of the clause is moved to the specifier position of CP [Spec,CP] preceding the Comp slot, which is filled by the finite verb (5b). (5)

a.

dass der Peter den Hans eingeladen that the.NOM Peter the.ACC Hans invited b. [cp[Den Hans], [[chatk] [ip der Peter t¡ the.ACC Hans has the.NOM Peter 'Peter has invited Hans.'

hat has eingeladen tk]]] invited

That this standard approach has shortcomings was first noted by Travis (1984). One fact that calls for an explanation is the privileged access of the nominative argument of standard transitive verbs to the preverbal position (the prefield). Objects of such verbs may appear in the prefield only if they bear a special pragmatic function, like being topics or being in focus. Thus, (5b) could not be used as an answer to "what has happened?". It can only be used as an answer to "who has Peter invited?", or "who has invited Hans?". The nominative argument, however, can always show up in the prefield. (6) is a possible answer to "what has happened?". It represents the default order. Travis (1984) takes similar observations as evidence for her claim that preverbal nominative noun phrases occupy the subject position [Spec,IP] rather than [Spec,CP] in German main clauses, cf. (6).1 In other words, there are two types of main clauses in German: IPs, which represent unmarked order and begin with subjects, and CPs, which may begin with any category provided that it bears a special pragmatic function like topic or focus. See also Zwart (1993) for a related approach, among others. (6)

[n>[Der Hans] [ [ i h a t j the.NOM Hans has 'Hans has invited Peter.'

[vpden the.ACC

Peter eingeladen tk]]] Peter invited

It has been noted by Lenerz (1977) that the basic or unmarked word order nominative > dative > accusative does not hold in all clauses. There are two constellations in which dative > nominative is "unmarked" or "normal", viz., passive clauses and clauses involving unaccusative and psychological predicates. In other words, we find deviations from the nominative-first pattern only when the nominative argument is an (under-

1. Travis (1984) cites further data involving the placement of unstressed pronouns to the prefield position as supporting her analysis, but data from more varieties of German question the validity of her pertinent argument, see Gärtner & Steinbach (2001) for details.

231

Quirky "subjects" and other specifiers

lying) direct object, that is, when the dative noun phrase (the indirect object) arguably is higher in the hierarchy of arguments than the nominative noun phrase.2 For simple embedded clauses such as (7), one can explain the unmarked status of dative > nominative order in a straightforward way. One only needs to assume that the lexical hierarchy of arguments must be reflected by the structural hierarchy of arguments in the VP when they are merged, and that nominative noun phrases are not forced to move to [Spec,IP] in German (see den Besten 1985 for the latter assumption).3 (7)

[CP dass [n>[vp einem Kind das

Fahrrad gestohlen wurde]]]]

that a.DAT child the.NOM bike 'that the bike was stolen from a child'

stolen

was

Identical unmarked order facts characterize main clauses, as illustrated by (8) and (9). Both (8) and (9) are potential answers to an out-of-the-blue question like "what has happened?" (unlike their nominative-initial counterparts!). (8)

(9)

Einem Kind wurde das Fahrrad a.DAT child was the.NOM bike 'The bike was stolen from a child.' Einem Schauspieler ist der Text a.DAT actor is the.NOM text 'An actor has forgotten the text.'

gestohlen. stolen entfallen. forgotten

That this constitutes a problem for the standard analysis of the German prefield construction was (to the best of my knowledge) first noted by Hubert Haider in a talk at the 1988 GGS-conference in Passau. Why should the dative object have a privileged access to α in (10) if what is involved is movement to [Spec,CP], that is, to an operator position? How can the dative object move at all to an operator position if it is neither a topic nor a focus phrase? (10) [cp α

wurde was

[vp einem Kind [das a.DAT child

the.NOM

Fahrrad gestohlen]]] bike

stolen

The natural analysis in the spirit of Travis (1984) seems to be that the clauses in (8) and (9) are IPs, with the subject position [Spec,IP] being filled by the dative noun phrase, cf. (10') and see den Besten (1985) for a proposal allowing such an account. (10') [IP Einem Kind wurde a.DAT child was

[VP[das Fahrrad gestohlen]]] the.NOM bike stolen

2. That this holds for psychological predicates with a dative argument, too, has been argued for in Fanselow (1992). 3. This may be due to a principle such as SEMHIER(ARCHY) proposed in Wunderlich (2001): "The linear order of arguments should correspond to their semantic ranking (with the highest argument first)." For base structures, the relevant generalization also follows from the strict compositionality of interpretation in a straightforward way, see Fanselow (1991), Stechow (1992).

232

Gisbert Fanselow

Such a solution seems to work well for the data considered so far, but other observations call for a modification. What (7) and (8), (9) show is that unmarked order facts of embedded and main clauses are identical:4 if α can be the leftmost element of an unmarked complement clause, it can occupy the preverbal position in an unmarked main clause, too, and vice versa. This, however, does not only hold for phrases that may be considered subjects in some extended sense. It is also true for adverbials. Thus, Frey (2000) shows that temporal adverbs may freely precede nominative subjects in pragmatically unmarked embedded clauses. The same is true of main clauses: temporal adverbials may precede the finite verb in a main clause without bearing any special pragmatic function. (11) Am Sonntag hat ein Eisbär einen Mann gefressen, on Sunday has a polar.bear a man eaten Ά polar bear ate a man on Sunday.' Likewise, sentence level adverbs can always precede subjects, and they are tolerated in the prefield in unmarked situations, as has already been noted by Köster (1978) for Dutch. (11) and (12) are perfect answers to "what has happened?" (12) Vielleicht hat der Schauspieler seinen Text vergessen, perhaps has the actor his text forgotten 'Perhaps, the actor forgot his text.' Köster (1978) observes furthermore that sentence level adverbs originating in an embedded clause cannot be placed into the prefield position of the matrix clause, as (13) and (14) illustrate for Dutch and German, respectively. Since long-distance operator movement is possible in Dutch and German, the ungrammaticality of the b-examples in (13) and (14) suggests that sentence level adverbs cannot undergo operator movement. If so, the adverbs cannot have reached the clause-initial position by operator-movement in the a-examples either. (13) a.

Waarschijnlijk is hij ziek. probably is he sick 'Probably, he is sick.' b. * Waarschijnlijk zegt Jan dat hij ziek is probably says Jan that he sick is 'John says that he is probably sick.'

(14) a.

Glücklicherweise hat er angerufen. fortunately has he called 'Fortunately, he has called.'

4. This is not entirely correct, since clitic non-subject pronouns and certain particles may be adjacent to the complementizer in an embedded clause, although they cannot precede the finite verb in a main clause. We will ignore this problem here because of limitations of space. See Fanselow (2002) for a sketch of an account.

233

Quirky "subjects" and other specifiers b. * Glücklicherweise denke ich dass er t fortunately think I that he Ί think that fortunately he has called.'

angerufen hat. called has

The position preceding the finite verb in a pragmatically unmarked clause is, thus, not an operator position, but it is also not a "subject" position in an interesting sense of the term: it is obvious that nominative subjects and sentence level adverbs have too little in common from a grammatical perspective. The claim that the sentence level adverbs occupy the "subject position" in (13), (14a) may be unobjectionable in a certain technical sense, but it implies that the content of the notion "subject" is reduced to "leftmost element in a pragmatically unmarked clause" In other words, there is a non-operator position, an A-position, at the left periphery of German main clauses which is not confined to subjects, but can be occupied by quite diverse types of phrases. There are several ways of making this idea precise. For example, let us assume that German main clauses are projections of a Finiteness-head Fin (see, e.g., Rizzi 1997). Fin is the target of verb movement in main clauses. The specifier position [Spec,Fin] must be filled obligatorily, and movement to it is governed by some version of the Minimal Link Condition (MLC): only the element which is closest to [Spec,Fin] and which meets the requirements for the position targeted by movement can undergo movement (see Chomsky 1995).5 Since Fin itself imposes no restrictions at all on its specifier position, nothing but β can move to α in (15a). The MLC thus implies that normal order facts of main and complement clauses are identical. What may appear in β in main and embedded clauses is determined by laws governing the hierarchy of arguments and the placement of adverbials, which need not concern us here. 6 (15) a. a F i n f o . β [...]] b. Compfepß...] Leaving the domain of unmarked word order, we may first observe that non-focal, thematic elements may appear in the first position following the complementizer in embedded clauses (16a). The same is true for matrix clauses (16b). This follows straightforwardly in our model: because of the MLC, exactly the highest element β is attracted (moved) to [Spec,FinP]. It does not matter whether the phrase occupying β shows up there because it was merged (base-generated) in this slot, or because it was moved there by scrambling. Thus, under the pragmatic conditions that license scrambling, both (16a) and (16b) are correctly predicted to be fine: the scrambled element is mapped further to [Spec,FinP].7'8

5. This may be formulated, e.g., as follows (in the spirit of Chomsky 1995): α cannot move to Σ if there is a β that could also move to Σ, such that β c-commands a. 6. Since Wunderlich's constraint SEMHIER mentioned above applies to surface representations, it has, essentially, the same effects as a Minimal Link Condition restricting the movement of XPs out of their base positions to [Spec.FinP]. 7. The term "scrambling" is used here in a loose, non-technical sense. See Fanselow (2001) for arguments against deriving free constituent order by movement.

234

Gisbert Fanselow

(16) a.

dass den Wagen niemand reparierte that the.ACC car nobody fixed 'that nobody fixed the car' b. Den Wagen hat niemand repariert, the car has nobody fixed 'Nobody has fixed the car.'

Rosengren (1993) and Krifka (1998) argue convincingly that the conditions for scrambling cannot always be formulated in terms of conditions that are met by the scrambled phrase itself. Rather, the scrambling of α at least sometimes serves the purpose of allowing another phrase β to be in focus. If correct, their analyses imply that den Wagen does not reach its position in (16a) and (16b) by operator movement (because it has no operator features). Consequently, den Wagen occupies an A-position in (16b) (= Spec,FinP) - again, it would make little sense to equate that slot with a subject position. 9 Finally, the prefield can also host w/i-phrases and focused material, as (17) shows. The simplest analysis would seem to be that Fin is also able to bear operator features such as wh or foe. Recall that the Minimal Link Condition requires that a phrase Ρ can move to a position Q only if Ρ is the phrase closest to Q that meets the requirements imposed for Q. Thus, if Fin has a [+wh]-feature in (18), its specifier can only be filled by a phrase bearing a [+wh]-feature, if one makes the standard assumption that the features of specifiers and heads must agree (so that the [+wh]-feature of Fin can be checked, as required). Consequently, β can move to α in (18), provided that it possesses a [+wh]-feature while γ lacks it. In other words, when Fin has a [+wh]-feature, the closest wh-phrase will move to its specifier position, possibly crossing other [-wh]phrases on its way up. This accounts for (17b). (17) a.

Was hat sie ihm gegeben? what has she him given 'What has she given to him?' b. Nichts hat sie ihm gegeben, nothing has she him given 'She has given him NOTHING.'

(18) [a

[FIN (¿wh), (±foc)] [ ΣΡ γ .. .β ...]]

8. Movement to [Spec.FinP] is an instance of Α-movement. Our analysis therefore presupposes that short scrambling involves Α-movement (see Deprez 1989, Mahajan 1990) or no movement at all (Fanselow 2001), because Α-movement to [Spec.FinP] following scrambling would be improper otherwise. 9. But see Borer (1995: 570ff.) for an argumentation that XPs in preverbal and pre-subject position in fact occupy [Spec,IP], at least in Hebrew.

Quirky "subjects" and other specifiers

235

In a similar way, a [+focus]-specification of Fin allows the fronting of focus phrases, as in (17b).10 When Fin is not specified positively for any operator feature, [Spec,Fin] is an A-position, and the constellation discussed in the context of (15a) arises again: the phrase closest to [Spec,Fin] moves to that position. Bhatt (1999: 90) makes a proposal very similar to ours for Kashmiri. In a slightly more complex analysis, Fin is not able to bear operator features. One then needs (at least) one additional functional layer above FinP, as shown in (19), where Comp bears the operator features. (19) [ C P a

[Comp (±wh), (±foc)] [rmpFin ... [ Σ Ρ γ ...β ...]]]

The choice between (18) and (19) depends, partially, on the analysis of further languages. In Russian (see Baylin 2001), there seems to be a verb second effect comparable to the one in German, but only in clauses representing "unmarked" order, and in clauses involving the preposing of thematic material (as in (16b)). The verb does not, however, move to second position when wh-phrases or phrases involving (contrastive) focus are preposed. Relative to (19), this difference between German and Russian can be explained easily: the finite verb moves only to Fin in Russian, whereas it climbs up to Comp (when present) in German. Kashmiri main clauses (see Bhatt 1999) resemble their German counterparts very closely, but in questions that also contain topic material, the order topic-wA-phrase(s)-verb (20) is preferred over simple verb second structures. This seems to indicate that there is yet another functional head above Comp that fails to attract the verb in Kashmiri. (20) Rameshan kyaa dyutnay tse Ramesh what gave you 'What did Ramesh give to you?' The resulting model comes close to the one proposed in Rizzi (1997). Since the proper analysis of wh- and focus-movement is not crucial for the point to be made, we will simply assume that (19) is correct, and leave the issue in the interest of space. In this section, we have argued for the existence of a non-operator A-position [Spec,FinP] at the left periphery of German main clauses, an idea that elaborates basic insights of Travis (1984) and Zwart (1993). [Spec,FinP] is the landing site for the highest argument and certain adverbials in sentences involving unmarked order, and it may host a scrambled phrase under the pragmatic conditions that license this type of reordering. If the nominative arguments of active transitive predicates are prototypical subjects, then [Spec,FinP] is a "subject position" in the sense that prototypical subjects typically move to it in unmarked sentences. If a "subject position" is a structural slot that is reserved for the category of subjects, and relative to which the subject properties of a phrase are established, then [Spec,FinP] is not a subject position - unless we are willing to make the notion of subject void of empirical content.

10. It need not concern us here whether topichood corresponds to a further operator feature, as is implied by the analyses of Frey (2000) and Pili (2001). See in particular Frey (2000), for the need of assuming (clause-internal) topic-movement in addition to scrambling.

236

4.

Gisbert Fanselow

Linking German and Icelandic

The analysis (19) for German main clauses developed in the preceding section resembles the proposal Holmberg (2000) puts forward for Stylistic Fronting in Icelandic in certain respects. Holmberg follows the standard minimalist assumption (see Chomsky 1995) that the subject position [Spec,IP] must be filled because Infi has an "EPPfeature", that must be checked by a phrase moving to its specifier position.11 Originally, the EPP-feature was equated with a determiner-feature, because only DPs (that is, categores that have determiner features) appear to be able to fill the subject position, i.e., [Spec,IP], in English. One may question the appropriateness of such a decision on the basis of examples like (21), in which a PP occupies the subject position. Dissociating the "EPP-feature" from the determiner-feature becomes inevitable, however, when one tries to account for movement to the specifier position of CP in terms of an EPPfeature, too, as Chomsky (2000) and Fanselow & Mahajan (1996, 2000) have proposed, because the specifier position of CP in German and other languages can be filled by all types of phrases, not just by DPs. The EPP-feature is, thus, a feature that only marks that certain positions be filled, it has no additional content.12 (21) Into the room came a man from India. Consider now Icelandic Stylistic Fronting, as exemplified by (22). When there is no argumentai noun phrase that could fill the position preceding the finite verb in an Icelandic clause (= [Spec,IP] in the standard analysis), two possibilities exist. The expletive pad may be inserted, or a non-finite verb or verbal particle may be fronted, as in both the main and the embedded clause in (22). According to Holmberg (2000), Stylistic Fronting is by no means an exceptional construction. Rather, Stylistic Fronting results from standard movement to [Spec,IP]. (22) Fram hefur komiô aö fiskaö hefur veriô ί leyfisleysi. forth has come that fished has been illegally 'It has come forth that it has been fished illegally.' It has been established by Jacobson (1987) and Larson (1988) that the verb and the verbal complex can be the highest category in the projection that is embedded by Infi (in particular, in verb-object languages), because a structure such as (23) is established.13 When the "external" argument is absent, or when no other argument ccommands the verbal complex, the verbal complex is the highest element below Infi.

11. Recall that EPP stands for Extended Projection Principle, that is, the requirement that all clauses have a subject. 12. One may wonder whether that insight refutes rather than supports the claim of the Minimalist Program that movement is only triggered when there is a need to check an uninterpretable feature of an attracting head. The only raison d'être for the EPP-feature is the fact that movement is necessary. We will not pursue this issue here. 13. Several proposals have been made in the literature concerning the nature of F, the choice among which is, again, not crucial for the point we wish to make.

Quirky "subjects" and other specifiers

237

(23) α Infi [[F verbal complex]¡ [vp —1¡...]] Given the Minimal Link Condition, α can only be filled by an element from the verbal complex in (23), because these are closest.1 Since the finite verb must occupy Infi itself, it is only the non-finite parts of the verbal complex that can go to [Spec,IP], such as the participle fiskad or the verbal particle fram. Holmberg (2000) shows in detail why this account of Stylistic Fronting is superior to previous analyses. For our purposes, it is of particular importance that he has established that [Spec,IP] can be filled by a particle or a participle. Icelandic [Spec,IP] thus is a "subject position" only in the uninteresting sense introduced above: it is the position which prototypical subjects can move to - not because of any intrinsically important "subject property", but because movement to the position is governed by the Minimal Link Condition. Other categories are tolerated in the position as well. It is thus more appropriate to speak of a FinP rather than an IP (as we did for German), in order to avoid unwanted connotations of [Spec,IP]. Icelandic differs from German, however, in that the verb moves to second position in embedded clauses introduced by a complementizer as well. Pili (2001) argues that German and Icelandic embedded clauses possess the abstract structure [cp ... Comp [πηΡ ..· Fin ...]]. In Icelandic, complementizers are inserted into Comp, and the verb always moves to Fin. In German, complementizers are inserted in Fin, and move to Comp. Consequently, the verb cannot move to an embedded Fin in German (because the position is occupied by the trace of the complementizer).15 We will adopt this solution for the purposes of the present paper. Thus, German and Icelandic share the structure (24). FinP is always projected, while Comp is present only in embedded clauses, and in structures involving wh- or focus features. [Spec,FinP] is an A-position to which the highest element in the category below it moves. It is not a subject position in any interesting sense. (24) [CPY [Comp (±wh), (±foc)] [ F m P a Fin ... [ ΣΡ β ...]]] We can now return to the issue of quirky subjects. If [Spec,FinP] is able to host elements other than nominative subjects in Icelandic and German, because it is a position not restricted to subjects (as evidenced by (22) and related constructions), we do not have to assume that dative henni is a quirky subject in (25) if we want to explain why henni rather than their occupies the preverbal position in an unmarked clause. It suffices to assume that the dative corresponds to the higher of the two arguments in (25), so that it will move to [Spec,FinP] under the restrictions imposed by the Minimal Link Condition.

14. We have simplified the proposal of Holmberg (2000) in a number of respects that are irrelevant for our discussion. 15. In Pili's model, the question arises whether [Spec,FinP] can and must be filled when a complementizer is inserted into Fin. We will simply assume that complementizers are not compatible with specifiers. See Pili (2001) for a different view concerning this issue.

238

Gisbert

Fanselow

(25) Henni leiddust/leiddist their. she.DAT be.bored.3PL/3SG they.NOM 'She was bored with them.' The literature on Icelandic (Zaenen et al. 1985, Sigurösson 1989), Faroese (Barnes 1986) and Spanish (Masullo 1993, Fernández-Soriano 1999) claims that linearization facts constitute evidence for the subject status of preverbal oblique noun phrases. The claim is made that phrases appearing in the first position of a clause without bearing a special pragmatic function (topic or focus) are subjects. It is the same phrases that immediately follow a verb moved to Comp in a yes-no-question, and a non-subject whquestion. In the light of our discussion, this argument is invalid. Adverb placement data (in German) and participle placement facts (in Icelandic) show that the highest Aposition of a clause, [Spec,FinP], is not restricted to subjects. Consequently, we cannot conclude anything concerning grammatical functions from the fact that a phrase is able to show up in [Spec,FinP]. Special reference to subject status would be necessary only if the oblique noun phrase appearing in [Spec,FinP] would not be the highest argument in the construction. In a mono-argumental structure, this cannot be the case for obvious reasons. In structures with two arguments, oblique noun phrases can appear in [Spec,FinP] only in passive and unaccusative and psych-verb constructions (see Sigurösson 1989), and they always correspond to the higher of the two arguments (the indirect rather than the direct object). Therefore, no reference to subject status is necessary in an account of their placement into the clause-initial position. 16

16. This does not imply that subjects of active transitive constructions never show up with a Case different from nominative in other languages. If the Case of (such) subjects is determined by Infi, then variation in the feature composition of Infi is able to trigger Case differences. Indeed, finite Infi combines with nominative subjects in Latin, while infinitives are constructed with an accusative subject (Jensen 1983, Pillinger 1980). There are aspect-related Case differences in Hindi (Mohanan 1994, Alexiadou 1999, Stiebels, 2000) and Georgian (Anderson 1984), and modality has an influence on subject Case in Finnish (see Laitinen & Vilkuna 1993), and Russian (Moore & Perlmutter 2000). The existence of ergative languages also proves that nominative is not the only Case that can go to subjects (see also Stiebels 2000 for a detailed discussion). Non-nominative "quirky" subjects pose a problem in those construction types only in which the Case to be expected for a subject is nominative because, e.g., Infi assigns nominative Case only (as is true for Icelandic, Faroese, and German). See Sigurösson (2000a) for a discussion of the importance of this distinction between the two types of non-nominative subjects. The Case of the oblique noun phrases in [Spec.FinP] may be a structurally assigned dative (as in a passive of a double object construction, and, perhaps, unaccusative and psychological predicates), or be the result of an idiosyncratic Case government property of individual verbs. See Blume (1998, 2000) and Fanselow (2000) for different attempts of reducing the amount of idiosyncrasies in lexical Case government.

Quirky "subjects " and other specifiers

5.

239

Differences between Icelandic and German

It is, of course, not only the word order facts (which German shares with Icelandic) that have motivated the postulation of quirky subjects. Rather, preverbal oblique noun phrases behave differently in Icelandic and German in a number of respects. In this section, we discuss these differences, and show that they do not involve a difference in the grammatical function (subject vs. object) of the phrase in [Spec,FinP].

5.1

C o o r d i n a t i o n reduction

Icelandic shows a peculiar behavior in (26) with respect to coordination reduction. The deletion of an oblique noun phrase in [Spec,FinP] can be licensed by a nominative subject antecedent (and vice versa): (26) Ég haföi mikiö aö géra og (mér) var samt ekki hjàlpaô I.NOM had a.lot to do and (me.DAT) was nevertheless not helped Ί had a lot to do, and no one helped me.' If conjunction reduction is sensitive to the identity of grammatical functions, (26) would be an argument for the subject status of a dative noun phrases in [Spec,FinP] in Icelandic. In contrast, (27) fails to be grammatical, which - under the assumption that coordination reduction presupposes an identity of grammatical functions - would show that German [Spec,FinP] does not host quirky subjects. (27) * Der Mann mag die Bibel und dem Mann gefällt der Koran. the.NOM man likes the Bible and the.DAT man pleases the Koran 'The man likes the Bible, and the Koran pleases him.' Should we indeed formulate the theory of conjunction reduction in such a way that the identity of grammatical functions is the major factor? The answer must be negative, since Faroese has problems with (26), too, as (28b) taken from Barnes (1986) (=his (75) and (80)), illustrates. Faroese shares the other evidence for quirky subjects with Icelandic. If (26) is a criterion for the subject status of an oblique noun phrase in [Spec, FinP], only Icelandic possesses quirky subjects, and "subject" would be a term relevant to coordination reduction only. (28) Faroese a. henni dámdi vael at lesa og nyttist ikki sjónvarp her.DAT liked well to read and needed not television 'She liked to read a lot, and needed no television' b. ?? honum leiddist viö liviö og tveítti seg á bláman him.DAT bored with life and NOM threw self in ocean 'He was bored with life and threw himself into the ocean.'

240

Gisbert Fanselow

An explanation for the contrast between (26) and (27) can be found quite easily in the domain of identity requirements for Case, however. German object-initial structures may be reduced only if the Cases of the two objects match, as the contrast between (29a,b) and (29c) illustrates. This suggests that Case identity is a necessary condition for coordination reduction in German. (29) a. Den Arzt unterstützt Hans und den Arzt behindert Maria. the.ACC doctor supports Hans and the.ACC doctor impedes Maria 'Hans supports the doctor, and Mary impedes him' b. Dem Arzt hilft Hans und dem Arzt assistiert Maria. the.DAT doctor helps Hans and the.DAT doctor assists Maria 'Hans helps the doctor, and Mary assists him' c. *Dem Arzt hilft Hans und de» Arzt unterstützt Maria. the.DAT doctor helps Hans and the.ACC doctor supports Maria 'Hans helps the doctor, and Mary supports him.' For obvious reasons, pairs of dative and nominative noun phrases in [Spec,FinP] as in (27) fail to meet the Case identity requirement, so the contrast between (26) and (27) is accounted for without any reference to subjecthood if we assume that Icelandic does not impose a Case identity requirement. There is independent evidence for this claim, as (30) shows. 18 The two objects bear different Cases in (30a). They can be topicalized simultaneously (30b). The deletion of the second object yields marginal results (30c), but the structure improves considerably when the identical subject is left out as well, as in (30d). (30d) suffices to establish that Case identity is in general not a requirement for the wellformedness of coordination reduction in Icelandic. What we observe in (30) for Icelandic contrasts with the facts in (29), and the pertinent difference is also responsible for the contrast between (26) and (27) on obvious grounds. (30) a.

Viö aetlum aö kjósa Jjennan we intend to elect jjessum forseta. this.DAT president

forseta

og

viö viljum treysta

this.ACC president and we

want

trust

b. Pennan forseta aetlum viö aö kjósa og jjessum forseta viljum viö treysta. c. ??I>ennan forseta aetlum viö aö kjósa og η .τ viljum viö treysta. d. (?)Pennan forseta aetlum viö aö kjósa og DAT viljum NOM treysta. 'We intend to elect this president, and we want to trust this president.' The coordination reduction criterion therefore cannot tell us anything about the subject status of a phrase.

17. Wunderlich (2001) suggests that a high-ranked visibility requirement for dative Case is responsible for the ungrammaticality of (27). However, *dem Mann gefällt der Koran und mag die Bibel 'the man.DAT pleases the Koran and likes the Bible' is as ungrammatical as (27), although the dative Case now finds a visible representation. 18. Thanks to Halldór Sigurösson for providing me with the relevant examples.

241

Quirky "subjects" and other specifiers 5.2

Infinitives I: raising

Another crucial property of Icelandic and Faroese [Spec,FinP]-phrases is their behavior in Raising-to-Object contexts. In a normal Raising-to-Object/Exceptional Case Marking (ECM) construction such as I expect her to come early, the subject of the infinitive receives a Case assigned by the matrix verb. In the Faroese counterpart of such constructions, the quirky Case of a noun phrase in [Spec,FinP] is overwritten by the accusative Case assigned by the fee/ieve-matrix-verb, as (31) illustrates. (31) Faroese (Barnes 1986: his (38) and (39)) a. Jógvani t0rvaôî ein nyggjan Jogvann.DAT needed a new 'Jogvann needed a new car.' b. Eg helt Jógvan t0rva I believed Jogvann.ACC need Ί think Jogvann needs a new car.'

bil. car ein nyggjan bil. a new car

Faroese allows the replacement of an oblique Case by a structural one in a number of further contexts: the dative assigned by some verbs falls victim to nominative Case assignment by Infi in the passive, and many (though not all) oblique noun phrases in [Spec,FinP] may also be realized with nominative Case. (32) Faroese (Barnes 1986: 33) a. Maer dámar vael hasa bókina. me.DAT likes well that book.ACC b. Eg dámi vael hasa bókina. I.NOM like well that book.ACC Ί like that book very much.' Icelandic shows constructions comparable to (3 lb) as well, but we do not observe the Case shift typical of Faroese in these (and the other) contexts, see (33). (34) shows that the highest argument of the embedded clause may appear in front of a matrix clause adverb (i" barnaskap minum) in the èe/i'eve-construction. This fact strongly suggests that (33) and (34) involve a raising construction in the strict sense: the highest argument of the infinitive is extracted from it, and placed into a position in the matrix clause. If so, there is no need to make the (possibly problematic) assumption that the raising infinitive is a FinP itself. (35) taken from Zaenen et al. (1985: 448) is most telling in this respect: the complement clause element appearing in the matrix clause may be a PP as well. The position targeted by movement is thus category-insensitive. (33) É g I

tel believe

henni hafa her.DAT have

alltaf always

¡DÓtt Ólafur thought Olaf.NOM

leiöinlegur. boring.NOM

Ί believe her always to have found Olaf boring.' (34) Ég taldi Guórúnu ί barnaskap minum sakna. Haraldar I believed Gudrun.ACC in foolishness my miss Harald.GEN 'In my foolishness, I believed Gudrun to miss Harald.'

242

Gisbert Fanselow

(35) Ég taldi undir rúminu íbarnaskap mínum vera goöan I believed under bed.the in foolishness my be good felustaö. hiding.place 'In my foolishness, I believed that a good hiding place would be under the bed.' Which categories move to the relevant position in the matrix clause seems entirely determined by the Minimal Link Condition. Reference to the subject status of what is attracted is superfluous. The same holds for Raising-to-Subject constructions. It must be guaranteed that only the highest argument of the embedded infinitive can move to the [Spec,Fin] position of the matrix clause in constructions like Icelandic (36), but that is already an automatic consequence of the Minimal Link Condition. Nothing more needs to be said in this context, in particular, no reference to subject status is necessary. (36) Ólafi virtist hafa leiöst. Olaf.DAT seemed have be.bored German differs substantially form Icelandic and Faroese in the domain under consideration. There is no Raising-to-Object construction at all. Believe-type verbs enter control constructions, and Case is assigned directly into perception verb and causative complements (see den Besten 1985). Verbs like scheinen 'seem' trigger clause union obligatorily, so one cannot check for the presence or absence of raising. The differences between German and Icelandic are, therefore, again due to quite independent factors. The subject status of a phrase in [Spec,FinP] is not involved.

5.3

Infinitives Π: control

Both in Icelandic (37) and in Faroese (38), oblique noun phrases may be realized as PRO in a control infinitive. This is clearly impossible in German (39). (37) Icelandic Ég vonaöist til aö PRO veröa hjálpaó. I hoped for to PRO.DAT be helped Ί hope that one helps me.' (38) Faroese Eg kann ikki torga PRO at I.NOM can not bear PRO.DAT to Ί cannot bear lacking money.'

vantar pengar. lack money

(39) * Ich versuche, PRO nicht Geld zu fehlen. I try PRO.DAT not money to lack Ί try to not lack money.'

Quirky "subjects" and other specifiers

243

Infinitive tests often figure prominently among the criteria for subjecthood, so we may be confronted with the decisive - and only - subject-related difference between German and Icelandic. Note that such a result would be a meagre one, because subjectstatus could then only predict differences concerning infinitives. Note that the special status which "subjects" seem to play in control infinitives is not really captured by the theory of syntax, if one wants to go beyond a restatement of the facts. Chomsky (1981) assumes that non-finite Infi cannot assign Case. For noun phrases that have to rely on Infi in order to be able to satisfy a requirement such as the Case Filter (all visible noun phrases must have Case), Chomsky's conjecture implies that they cannot show up in infinitives. The subject of an infinitive therefore must lack a phonetic matrix, it has to be PRO. Clausal subjects that might show up in infinitives because CPs do not need Case pose an obvious problem. Even if this difficulty can be circumvented, the problem created by oblique noun phrases remains. The only argument in a passive construction of helfen 'help' does not depend on Infi for its Case, it receives dative Case from the verb in German (and also in Icelandic). For the presence of such a dative noun phrase, the Case marking potential of Infi should not really matter. Thus, one should expect to find something like (40) as an infinitive for verbs combining with oblique noun phrases in both languages - but we do not do so in either! (40) * Ich versuche mir geholfen zu werden. I try me.DAT helped to be Ί try to be helped.' In more recent work, Chomsky (1995) assumes that a special null Case can and must be assigned in infinitives. If this 'null' Case cannot be realized on noun phrases with a phonetic matrix, and if it must be assigned to some noun phrase in a control infinitive, it follows that one noun phrase position must be left unrealized phonetically in an infinitive. Alternatively, one can simply say that a control infinitive corresponds to a oneplace predicate, semantically, as Wunderlich (2001) does with his NOPROP-Constraint: "An infinitive clause must have an open argument role". Both approaches correctly imply that one argument position must be left unrealized phonetically in an infinitive. The ungrammaticality of (40) is thus explained in a straightforward way: all argument places in the infinitive are realized, in contrast to what the null Case idea or a constraint like NOPROP imply. (41), on the other hand, is incompatible with fundamental facts of Case realization. Chomsky (1981) observes that the idiosyncratic Cases assigned by individual lexical items must not be replaced by structural Cases in a passive construction or a nominalization. 19 If the null Case of Chomsky (1995) is a structural one that must be assigned to PRO in (41), and if the dative Case governed by helfen must not be left unrealized, the ungrammaticality of (41) is explained if null and dative Case may not co-occur on the same noun phrase: either null or dative Case would have to be left unassigned, but both options violate a principle of grammar.

19. This corresponds to Wunderlich's (2001) visibility condition for lexical Case.

244

Gisbert Fanselow

(41) * Ich versuche PRO geholfen zu werden. I try helped to be Ί try to be helped.' The wellformedness of (37) and (38) presupposes, then, that lexically governed Cases can be replaced by structural ones (or be left unrealized) in Faroese and Icelandic. (31) and (32) have already established this for Faroese. The wellformedness of (38) is thus accounted for. Icelandic keeps idiosyncratic Cases when they compete with accusative Case assigned by an ECM verb, or nominative Case assigned by Infi, but there are contexts in which idiosyncratic Case gives way to structural Case. Thus, whenever semantic conditions as identified by Maling (2001) allow it, lexically governed dative objects can be mapped onto genitive noun phrases in nominalizations (quite in contrast to German), as (42) taken from Maling (2001: 449) shows. (42) a.

Jpeir björguöu sjómanninum. they rescued sailor.DEF.DAT 'They rescued the sailor.' b. björgun sjómannsins rescue.DEF sailor.DEF.GEN 'the rescue of the sailor'

In Icelandic free relative clauses (see Vogel 2002), the relative pronoun must bear the Case assigned by the matrix verb. There is no matching requirement. Consequently, the Case assigned by the complement verb to the relative pronoun is suppressed, even if it is a lexically governed dative, as in (43). (43) a.

Ég elska *J>eim/ I like *those.DAT/ Ί like those that I help.' b. ?Eg elska *hvequm/ I like *who-DAT/ Ί like whoever I help.'

{)ann sem ég hjálpa. those.ACC that I help hvern (sem) ég hjálpa. who.ACC (that) I help

Under conditions different from those of Faroese, lexical Cases need not be realized in Icelandic either, so that the crucial condition for the grammaticality of (37) can be met. It remains to be explained why it is always the higher argument (irrespective of its Case) that must be realized as PRO in Icelandic. One can assume, for example, that the assignment of null Case is also subject to a Minimal Link Condition effect: it is assigned to the closest NP it can go to. Given that lexical Case need not always be realized in Icelandic, null Case will therefore always end up on the highest argument in Icelandic (this is equivalent to the constraint NOHIGH "The highest argument is not allowed to stay in an infinitive clause" of Wunderlich 2001). In German, lexical dative and genitive Case must not be left unexpressed, so that lower noun phrases have a chance of receiving null case when the higher one bears a lexical specification.

Quirky "subjects" and other specifiers

245

This section has established the same point as section 5.1: the differences between German and Icelandic involve the Case system, and not grammatical functions.

5.4

Agreement

In Icelandic and German (as in many other languages), agreement is a relation which the verb enters with nominative noun phrase only, and never with oblique noun phrases. Agreement facts could, however, establish the subject status of certain oblique noun phrases indirectly - viz., if certain nominative noun phrase cannot (fully) agree with the verb, because they are not subjects. In German, the verb always shows full agreement with the nominative noun phrase, irrespective of whether the latter is a direct object (as in passive and unaccusative clauses, see Grewendorf 1989) or a subject (i.e., the highest argument of an active transitive predicate). The situation is different in Icelandic. On the one hand, there are two verbs (see Sigurösson 1996) which may fail to show agreement with their nominative objects, but since the phenomenon is restricted lexically, it does not imply anything concerning their grammatical function. On the other hand, agreement is in general impoverished with (and only with) postverbal nominative noun phrases (which are, arguably, objects): the verb cannot agree with such a noun phrase when the latter is a 1st or 2nd person pronoun. In such a situation, the relevant content may simply be inexpressible (Sigurösson 2000b) or the verb may show up in its default 3rd person singular form (Hrafnbjargarson 2001), see (44). (44) J)ér fotti viö fyndin. you.DAT found.3SG we.NOM amusing 'You found us amusing.' Impoverished agreement depending on position is not restricted to constructions that potentially involve a non-nominative subject. Samek-Lodovici (2002) gives an overview of agreement impoverishment in various Italian and Arabic dialects, in which the verb shows full agreement in SV-order, but only partial agreement or no agreement at all in VS-constructions. Since no other phrase needs to occur preverbally in the VScase in Italian and Arabic, one can at most conclude that postverbal nominative noun phrases are not full subjects (whatever merits that proposal may have). It would not be justified to derive this from the assumption that another noun phrase has taken over the subject function. Several solutions for impoverished agreement have been developed (Alexiadou 2001, Hrafnbjargarson 2001, Samek-Lodovici 2001) which explain the phenomenon in terms of the hierarchical position the nominative NP occupies - "subject" status plays no role.20

20. However, there is no impoverished agreement when an adverb occupies [Spec.FinP]. Apparently, it is the position of the nominative noun phrase relative to the base position of the verb, and not relative to Fin that determines the richness of agreement.

246

5.5

Gisbert Fanselow

Islandhood

The contrast in (45)-(46) (see Zaenen et al. 1985: 451-52) is a further piece of evidence that might establish the subject status of oblique noun phrases in the specifier position of a FinP. Long distance wA-movement is grammatical in Icelandic, as (45b) illustrates. Topicalization within a complement clause as in (45c) makes that clause an island for extraction. The grammaticality of (46a) implies thus that henni has not been topicalized, while the nominative noun phrase Olafur in (46b) cannot simply occupy the subject position. (45) a.

Jon telur aö Maria hafi kysst Harald John believes that Mary has kissed Harald 'John believes that Mary kissed Harald yesterday.' b. Hvenaer telur Jon aô Maria hafi kysst when believes John that Mary has kissed 'When does John believe that Mary kissed Harold?' c. *Hvenaer telur Jon aö Harald hafi Maria kysst

(46) a.

b.

ígaer. yesterday Harald? Harald

Hvenaer telur Jon aö henni hafi {jótt Ólafur leiöinlegur? when believes John that her.DAT has found Olaf.NOM boring 'When does John believe that she found Olaf boring?' *Hvenaer telur Jon aö Olafur hafi henni J)ótt leiöinlegur?

The contrast certainly shows that there is a difference between the two complement clauses in (45c) and (46a), in spite of the fact that both begin with an oblique noun phrase. In order to account for the island status of the complement clause in (45c), it suffices to say that no other operator position may intervene between the trace and the target position of an operator movement. Harald has moved to such an operator position in (45c), and exerts a blocking effect for further extraction. (46a) is grammatical if henni does not occupy an operator position. Nothing requires that the slot it is sitting in is a subject position.

5.6

Summary

The present section has reiterated a point frequently made (e.g., by Sigurösson 2000a): German and Icelandic noun phrases in [Spec,FinP] differ in a number of respects (coordination, control and raising infinitives). It also stresses a point that is rarely made: they cannot but differ from each other, because independently motivated differences in the domain of conjunction reduction, the realization of lexical Case or the availability of constructions like Raising-to-Object imply that phrases in [Spec,FinP] must show a different behavior in the two languages.

247

Quirky "subjects" and other specifiers

6.

Consequences for grammar

The first detailed arguments for the analysis of Icelandic oblique clause initial noun phrases as quirky subjects were formulated in contributions to Lexical Functional Grammar (LFG). That this should be so comes as no surprise. In this model, grammatical functions are the basic units of grammatical analysis, so the peculiar behavior of (la), repeated here as (47a), more or less had to be analysed in terms of subject status. (47) a. Honum him.DAT

b. Ihm him.DAT

var

hjalpâô.

was

helped

wurde geholfen. was

(Icelandic) (German)

helped

'He was helped.' Just like Wunderlich (2001), the present paper shows that one would have to analyse the clause initial phrase of German (47b) as a quirky subject, too: whatever differences there are between the two languages must be accounted for in terms of the grammatical behavior of Case. In addition, the parallelisms in the behavior of clause initial oblique phrases such as the ones in (47) can be shown to follow from the hierarchy of arguments. The notion "subject" is not necessary at any point in the analysis of the construction. This insight is in line with independent developments in the theory of grammar. In standard GB-theory (Chomsky 1981), grammatical functions played no official role, but there was a subject position, viz. [Spec,IP], that was quite distinct in grammatical behavior from other structural slots for arguments. Thus, its presence was forced by a specific principle (the Extended Projection Principle), is was considered argumentai even if it did not receive a thematic role, it received Case in a different way (by agreement with Infi rather than by government), etc. Gradually, specific assumptions concerning [Spec,IP] were eliminated. For example, the Case of all noun phrases can nowadays be assigned/checked in a specifier-head relation. Furthermore, more than one functional projection seems justified between the verbal projection and the Complayer. In a minimalist account of syntax, there is thus no position left that one might call a subject position. All we are left with is distinctions of Case and hierarchy. This is much reminiscent of what is also assumed in Lexical Decomposition Grammar (Wunderlich 1997), although details differ. In the light of such overall developments, it would be quite surprising if the concept "quirky subject" would be more than a purely descriptive label. The present paper has shown that this is in line with the empirical facts.

248

Gisbert Fanselow

References Alexiadou, Artemis (2001). Functional Structure in Nomináis. Amsterdam: Benjamins. —; and Elena Anagnostopoulou (1998). Parametrizing AGR: Word order, V-movement, and EPP-checking. Natural Language and Linguistic Theory 16, 491-39. Anderson, Stephen (1984). On Representation in Morphology: Case Marking, Agreement, and Inversion in Georgian. Natural Language and Linguistic Theory 2, 157— 18.

Andrews, Avery (1982). The Representation of Case in Modern Icelandic. In The Mental Representation of Grammatical Relations, Joan Bresnan (ed.), 427-503. Cambridge, Mass.: ΜΓΓ Press. Bailyn, John (2001). Generalized Inversion. To appear in Natural Language and Linguistic Theory. Barnes, Michael (1986). Subject, Nominative, and Oblique Case in Faroese. Scripta Islandica 37, 13—46. den Besten, Hans (1985). The Ergative Hypothesis and Free Word Order in Dutch and German. In Studies in German Grammar, Jindrich Toman (ed.), 23-64. Dordrecht: Foris. Bhatt, Rakesh Mohan (1999). Verb Movement and the Syntax of Kashmiri. Dordrecht: Kluwer. Blume, Kerstin (1998). A contrastive analysis of interaction verbs with dative complements. Linguistics 36, 253-80. —(2000). Markierte Valenzen im Sprachvergleich. Tübingen: Niemeyer. Borer, Hagit (1995). The Ups and Downs of Hebrew Verb Movement. Natural Language and Linguistic Theory 13, 527-606. Chomsky, Noam (1981). lectures on Government and Binding. Dordrecht: Foris. —(1995). Categories and Transformations. In The minimalist program, 219-394. Cambridge, Mass.: ΜΓΓ Press. —(2000). Minimalist inquiries: The framework. In Step by Step: Essays on minimalist syntax in honor of Howard Lasnik, Roger Martin, David Michaels and Juan Uriagereka (eds.), 89-55. Cambridge, Mass.: ΜΓΓ Press. Déprez, Viviane (1989). On the typology of syntactic positions and the nature of chains. Doctoral dissertation, MIT, Cambridge, Mass. Fanselow, Gisbert (1991). Minimale Syntax. Unpublished habilitation thesis, University of Passau. Special edition of Groninger Arbeiten zur germanistischen Linguistik 32. —(1992). Die Struktur des deutschen Mittelfeldes und die Ergativitätsproblematik. In Deutsche Syntax. Ansichten und Aussichten, Lutz Hoffmann (ed.), 276-03. Berlin: de Gruyter. —(2000). Optimal Exceptions. In The Lexicon in Focus, Barbara Stiebeis and Dieter Wunderlich (eds.), 173-209. Berlin: Akademie Verlag. —(2001). Features, 0-roles, and free constituent order. Linguistic Inquiry 32, 405-437.

Quirky "subjects" and other specifiers

249

—(2002). Miinchhausen-Style Head Movement. To appear in Proceedings of the Head Movement Workshop. UCLA Working Papers in Linguistics. —; and Anoop Mahajan. (1996). Partial Movement and Successive Cyclicity. In Papers on Wh-Scope Marking. Arbeitspapier 76 des Sonderforschungsbereich 340. Uli Lutz and Gereon Müller (eds.), 131-61. Stuttgart/Tübingen. —;— (2000). Partial movement and successive cyclicity. In Wh-scope marking, Ulrich Lutz, Gereon Müller, and Arnim von Stechow (eds.), 195-230. Amsterdam: Benjamins. Fernández-Soriano, Olga (1999). Two Types of Impersonal Sentences in Spanish: Locative and Dative Subjects. Syntax 2, 101-140. Frey, Werner (2000). Über die syntaktische Position der Satztopiks im Deutschen. Ms., Zentrum für Allgemeine Sprachwissenschaft, Berlin. Gärtner, Hans Martin, and Markus Steinbach (2001). Conditions on the Scrambling and Fronting of Reduced Pronominale in Dutch and German. Ms., Potsdam/ Mainz. Grewendorf, Günther (1989). Ergativity in German. Dordrecht: Foris. Holmberg, Anders (2000). Scandinavian Stylistic Fronting: How any category can become an expletive. Linguistic Inquiry 31, 445^4-83. Hrafnbjargarson, Gunnar Hrafn (2001). An Optimality Theory Analysis of Agreement in Icelandic. Ms., Stuttgart. Jacobson, Pauline (1987). Phrase Structure, Grammatical Relations, and Discontinous Constituents. In Syntax and Semantics 20: Discontinuous Constituency, Geoffrey Huck and A. Ojeda (eds.), 27-69. San Diego: Academic Press. Jensen, John T. (1983). Case and Thematic Role in Latin: Evidence from Passive Constructions. Bloomington, Ind.: IULC. Koster, Jan (1975). Dutch as an SOV language. Linguistic Analysis 1, 111-136. —(1978). Locality Principles in Syntax. Dordrecht: Foris. Krifka, Manfred (1998). Scope inversion under the rise-fall contour in German. Linguistic Inquiry 29, 75-112. Laitinen, Lea, and Maria Vilkuna (1993). Case marking in necessive constructions and split intransitivity. In Case and other Junctional categories in Finnish syntax, Anders Holmberg and Urpo Nikanne (eds.), 23-28. Berlin: de Gruyter. Larson, Richard K. (1988). On the Double Object Construction. Linguistic Inquiry 19, 335-391. Lenerz, Jürgen (1977). Zur Abfolge nominaler Satzglieder im Deutschen. Tübingen: Narr. Mahajan, Anoop. (1990). The A/A-bar-distinction and movement theory. Doctoral dissertation, MIT, Cambridge, Mass. Maling, Joan (2001). Dative: The heterogeneity of the mapping among morphological case, grammatical functions, and thematic roles. Lingua 111, 414—464. Masullo, Pascual Jose (1993). Two Types of Quirky subjects: Spanish vs. Icelandic NELS 23, 303-316. Mohanan, Tara (1994). Argument Structure in Hindi. Stanford: CSLI Publications. Moore, John, and David Perlmutter (2000). What does it take to be a dative subject?. Natural Language and Linguistic Theory 18, 373—416.

250

Gisbert Fanselow

Pili, Diana (2001). On A- and A'-Dislocation in the Left Periphery: A Comparative Approach to the Cartography of the CP-System. Doctoral dissertation, University of Potsdam. Pillinger, Otto (1980). The accusative and infinitive in Latin: a refractory complement clause. Journal of Linguistics 16, 55-83. Reis, Marga (1982). Zum Subjektbegriff im Deutschen. In Satzglieder im Deutschen, Werner Abraham (ed.), 171-211. Tübingen: Narr. Rizzi, Luigi (1997). The Fine Structure of the Left Periphery. In Elements of Grammar, Liliane Haegeman (ed.), 281-337. Dordrecht: Kluwer. Rosengren, Inger (1993). Wahlfreiheit mit Konsequenzen - Scrambling, Topikalisierung und FHG im Dienste der Informationsstrukturierung. In Wortstellung und Informationsstruktur, Marga Reis (ed.), 251-312. Tübingen: Niemeyer. Samek-Lodovici, Vieri (2001), Agreement Impoverishment under Subject Inversion. A Crosslinguistic Analysis. To appear in Linguistische Berichte. Sigurösson, Halldór (1989). Verbal Syntax and Case in Modern Icelandic. Doctoral dissertation, Lund. —(1996). Icelandic Finite Verb Agreement. Working Papers in Scandinavian Syntax 57, 1-46. —(2000a). To be an oblique subject: Russian vs. Icelandic. Working Papers in Scandinavian Syntax 66, 1-32. —(2000b). The Locus of Case and Agreement. Working Papers in Scandinavian Syntax 65, 65-108. Stechow, Arnim von (1992). Kompositionsprinzipien und grammatische Struktur. In Biologische und soziale Grundlagen der Sprache, Peter Suchsland (ed.), 175-248. Tübingen: Niemeyer. Stiebeis, Barbara (2000). Typologie des Argumentlinking. Habilitation thesis, Düsseldorf. To appear in Studia Grammatica. Berlin: Akademie Verlag. Thiersch, Craig (1978). Topics in German Syntax. Doctoral dissertation. MIT, Cambridge, Mass. Travis, Lisa (1984). Parameters and Effects of Word Order Variation. Doctoral dissertation, MIT, Cambridge, Mass. Vogel, Ralf (2002). Free Relative Constructions in OT Syntax. To appear in Linguistische Berichte. Wunderlich, Dieter (1997). Cause and the Structure of Verbs. Linguistic Inquiry 28, 27-68. —(2001). The force of lexical case: German and Icelandic compared. To appear in The nature of the word: essays in honour of Paul Kiparsky, Kristin Hanson and Sharon Inkelas (eds.). Cambridge, Mass.: ΜΓΓ Press. Zaenen, Ann, Joan Maling, and Höskuldur Thráinsson (1985). Case and Grammatical Functions: The Icelandic Passive. Natural Language and Linguistic Theory 3, 441483. Zwart, C. Jan-Wouter. (1993). Dutch syntax. Doctoral dissertation, University of Groningen.

Why case marking?

Wolfgang

Klein

En vieillissant on devient plus fou, et plus sage. (La Rochefoucault)

1.

Introduction*

In the memorable year of '68, on a hot summer day, Dieter Wunderlich and I were lying at the beach of Laboe, north of Kiel, and wondered about why languages can be so different. In those days, I tried to learn Chinese and Basque at the same time, two languages which are structurally as different as one can imagine. In both cases, the success was somewhat limited. But there was one thing I gained from this experience: this is the deep conviction that inflectional morphology is largely superfluous, because other than the Basque, the Chinese easily do without it. Dieter did not quite share in this view, but the sun was hot, and the sea was cool, and we did not settle the issue on the spot. Some 33 years and an endless number of morphology papers later, he finally took the bull by the horns: 'Why is there morphology?' was the title of his talk at the 23rd Annual Meeting of the German Linguistic Society - that society whose cofounder and first president he was. The following remarks are a variation on this theme, and again, as in Laboe, they are from the perspective of the language learner. Their gist is this: If we want to understand the nature of the human language faculty and the nature of linguistic systems, we should see what this capacity does when it brings about such a system, and not just look at the final result. In the first part of this paper, I will elaborate on this view, and in the second part, I will illustrate what it might tell us about the function of case marking.

I wish to thank Ingrid Kaufmann and Barbara Stiebels for most helpful comments on an earlier version of this paper.

252

2.

Wolfgang Klein

Learner varieties are the normal case Perhaps the saddest aspect of this linear ranking lies in the acceptance of inferiority by bottom dwellers, and their persistent attempt to ape inappropriate methods that may work higher up on the ladder. When the order itself should be vigorously challenged, and plurality with equality asserted in pride, too many [bottom end] scientists act like the prison trusty who, ever mindful of his tenuous advantages, outdoes the warden himself in zeal for preserving the status quo of power and subordination. (Stephen Jay Gould 1989: 279)

Within the various disciplines that investigate the manifold manifestations of the human language faculty, research on how people learn a second language does not rank very high. Second language acquisition (SLA) researchers are bottom dwellers. Does this fact only reflect irrational but rock-solid caste prejudices on the part of those who want to protect their privileges? Or has it anything to do with SLA research itself - its object, its methods, its theoretical or empirical standards, its potential benefit for mankind? To begin with the latter, it appears that among the various linguistic disciplines, SLA research is probably the only one that is, or at least can be, of any substantial practical use. This should be a solid base of self-confidence, and a good reason to be held in some esteem by others. In fact, when linguists find themselves in a situation where they are urged to justify their existence in the eye of the common beholder, this is one of their arguments (together with aphasia, machine translation and automatic speech recognition). As to the second explanation, I do not think that the field of SLA in general scores so badly as regards the standards of cogent argumentation, of conceptual clarity, of clean data collection and of empirical validation. This is not to mean that it could not improve considerably in many ways, and any effort in this direction should be made. But there is little reason to assume that the empirical basis of typological comparison is on the average more solid than what is normally done in SLA research. If, for example, a study of word order typology is based on 400 languages, then this means that the author cannot have spent much time on understanding what the word order regularities of each language are. Nor is there a good reason to believe that notions commonly used in theoretical linguistics are of necessity clearer and better defined than those used in the study of second language acquisition; we shall discuss this for ubiquitous notions such as 'subject' or 'object' below. The problem seems to be rather that no one sees how the analysis of the odd productions of the second language learner, this distorted, flawed, ridiculous, chaotic mimicking of 'real language', could tell us something substantial, something principled, something fundamental about the nature of the human mind. It is this perception that must be changed in the first place. Learner varieties are a genuine manifestation of the human language faculty, and the careful and systematic investigation of how they are

Why case marking?

253

internally structured and how they develop over time is a genuine contribution to the understanding of this faculty. In fact, I believe that learner varieties are its core manifestations, and 'real languages' - or a speaker's perfect knowledge of a 'real language' - are just borderline cases. They are particularly interesting for social and cultural reasons, they are also interesting because they often exploit the structural potential of the human language faculty to a particularly high extent. But to the linguist, they should be no more privileged than is the noble lion over the humble drosophila melanogaster to the biologist. We are used to take perfect mastery of a language to be the normal case, and the linguistic knowledge of a perfect speaker - a speaker who masters a 'real language' up to perfection - to be the primary object of the linguist's efforts. But what does it mean that a speaker masters a language perfectly well, what must his or her knowledge be like in order to qualify as native? Our common façon de parler in these matters somehow implies that there are such entities as 'real, fully-fledged languages', such as German, Greek or Yukatec, and speakers 'know' them to a higher or lesser degree. But this is simply a myth. There are five thousand languages on earth. There are about 200 countries on earth. This means that there are - on the average - 25 languages per country, with a range between 1 and several hundred. As a consequence, the normal case is simply that a person has varying knowledge of different languages. That would be the good way to state the facts for the layman who believes that there are well-defined entities called 'languages'. But there aren't. What really happens is this: HUMAN BEINGS, EQUIPPED WITH THIS SPECIES-SPECIFIC MENTAL CAPACITY CALLED HUMAN LANGUAGE FACULTY, MANAGE TO COPY, WITH VARYING DEGREES OF SUCCESS, THE WAYS IN WHICH OTHER PEOPLE SPEAK. They develop LEARNER VARIETIES. Under specific conditions, they

push this process to a degree where their own competence to speak and to understand does not perceivably differ from that of their social environment (or, perhaps, a special group within their social environment, like school teachers). Then, we speak of 'perfect mastery'. But this perfect mastery is just a special case of a learner variety - that case in which neither the learner nor his neighbours notice any difference, or at least no difference they would consider to be of particular social importance. Normally, the speaker's language faculty also allows him or her to develop more than just one such learner variety; the degree to which these come close to 'perfection' varies considerably. But all of them are manifestations of the human language faculty. Many learner varieties do not exploit the full potential of this faculty, for example in terms of syntactic or morphological structure or of lexical repertoire. But even my Russian learner variety, which is very elementary indeed, uses more of the human language faculty's morphological potential than the 'fully-fledged language' with most native speakers on earth, Chinese. If we really want to understand the nature of the human language faculty, we must investigate how its manifold manifestations are organised and how they develop over time. This includes the study of 'fully-fledged languages' - or more precisely, the speaker's knowledge of a fully-fledged language - as a special case. This case is perhaps particular interesting for cultural and sometimes even for structural reasons. After

254

Wolfgang Klein

all, the ways in which Friedrich Schiller, Thomas Bernhard and the average inhabitant of Niederkassel put their words together are more complex, more refined, more multifarious and therefore perhaps more instructive to an understanding of the human language capacity than the learner variety of Keiko Watanabe, Ergün Üzlemir or Giuseppe Scorcese after five years in Oberbilk. But we should keep in mind that the learner varieties of the latter are the normal case, nowadays as well as in the history of mankind; and here as everywhere in science, the investigation of the normal case should not be something peripheral, left to those at the bottom end, who are graciously, and with occasional friendly applause, allowed to borrow from those working higher up on the ladder. The systematical and careful study of how people process linguistic input in communicative situations, of how they use their innate capacities in order to turn this input into learner varieties and of how they abandon these for other, more complex or just differently organised learner varieties until this process eventually comes to a halt, in short: the study of second language acquisition is not a minor, a derived branch within the various disciplines that set out to investigate the human language faculty. It is central to an understanding of that remarkable capacity with which a friendly nature has endowed us. Whilst I believe that this argument is perfectly logical, I realise that it would be more convincing to demonstrate how work on learner systems can lead to new insights about the structure and the functioning of language. In the remainder of this paper, I will try to do this for an area which has always been in the focus of research from the days of the Greek grammarians to the days of SFB 282 - inflectional morphology.

3.

Two questions Fortwährend schiebt sich die Tradition zwischen die Tatsache und den Betrachter. (Jellinek 1913: 21)

There is considerable research on how second language learners acquire the inflectional morphology of the target language. Numerous studies document the learners' struggles with the oddities of German noun declension, Spanish irregular verbs or agreement marking in French. By far most of this work deals with SLA in the classroom, where - as especially those among us who had to learn Latin or Greek in school will remember (forsan et haec olim meminisse iuvabit, Aeneis I) - the memorization of morphology plays an eminent role.1 Implicitly or explicitly, this research is strongly norm-oriented: there are clearly defined rules of how words should change their form, these rules are made explicit to the learner, and acquisition research measures the learner's successes and failures to apply them This work does not require any deep understanding of why certain morphological regularities are as they are. What counts is 1. Throughout this paper, the term 'morphology' tout court relates to inflectional morphology only.

Why case marking?

255

the mere fact that it is not die Flücher but die Flüche, whereas it is not die Tüche but die Tücher, or that it is j'ai fait la communication and not j'ai faite la communication, but la communication que j'ai faite rather than la communication que j'ai fait. Therefore, this research is of only moderate interest to the linguist. There is much less research on how learners approach inflectional morphology outside the classroom. It is this research that might help to answer the question of how learners construct inflectional morphology out of the material with which they are confronted. This material takes the form of a more or less continuous sound stream uttered in a communicative context, and by processing and interpreting this input, the learner must somehow derive how certain lexical entries change their form under certain conditions - in other words, how words are inflected and what this inflection is good for. Other than in the classroom, this process is not determined by the particular way in which the rules to be learned are presented to the learner. It is entirely governed by the inherent properties of the learner's innate language faculty, on the one hand, and by whatever he knows about other languages, in particular about his first language, on the other. Hence, the investigation of this process should inform us about the natural principles of second language acquisition. But it can also help us to answer a second question which goes beyond the immediate concerns of the acquisition researcher. This is the question of what these findings can tell us about the nature of inflectional morphology and hence, as our tradition has it, about a core part of the human language faculty. In the Western tradition of linguistics, the notion of 'grammar' was for two millenia almost equivalent to 'inflectional morphology'. Most of Donatus or Priscianus deals with the rules according to which words change their form, rather than with the rules according to which they are put together. Even phenomena which we now tend to view as syntactic or semantic, such as argument structure, were mainly seen from the perspective of case marking: uti requires the genitive, persuadere requires the dative, videre requires the accusative. The study of how time is expressed in language was, and still is, mostly concerned with what some morphological changes on the verb contribute to this task. The first grammars of 'modern' languages, such as English, German or French, readily adopted this 'morphology bias', and the fact that the inflectional systems of these languages were less elaborate than in Latin or Greek was generally seen as a sign of erosion and decay (see Jellinek 1913 for a most instructive documentation of this tradition). This view was hardly ever challenged before the end of the 19th century. How is it at the beginning of the 21st century? The answer to this question is more difficult as might appear. First, the fact that some languages, such as Chinese or Vietnamese have virtually no inflectional morphology renders the 'classical view' obsolete. In the design of human language, inflectional morphology is a common but by no means indispensable part. Second, there is good reason to assume that the way in which we investigate firmly established grammatical categories such as Tense, Mood or Case is still strongly determined by this traditional perspective and, as a consequence, is often led the garden path. Third, odd as the notion of a grammatical 'decay' may seem to us - it is simply a fact that to the extent to which we have clear historical records, languages tend to reduce or to give up inflectional morphology rather than to elaborate it. There are some

256

Wolfgang Klein

exceptions, such as the formation of future marking in Romance languages {aimerai from amare habeo and similar ones). They are often referred to in the linguistic literature, but all in all, they are rare and do not affect the overall picture. 2 English, Dutch, and even German show very reduced inflectional systems when compared to their common Westgermanic origin, let alone when compared to older stages of Indoeuropean. Chinese, the paradigmatic example of a language without inflection is assumed by some scholars to once have had it, but alas, it is gone and has only left some traces in form of lexical tones. So, some exceptions aside, the entire historical development seems to go away from morphology, or to vary on a theme by Dr. Samuel Johnson: 'Inflectional morphology has, like governments, a natural tendency to degenerate'. These observations face the general linguist with two questions -

Why do we have this asymmetry? and Why do languages have inflectional morphology in the first place?

This now is the point where the study of other manifestations of the human language capacity than 'fully-fledged languages' with all their oddities inherited from the past can help us. We should have a look at how this capacity constructs learner varieties when exposed to some input. I do not want to argue that this broader perspective provides the final answer; but it can give us some evidence on how and why inflectional morphology is bom. In earlier work on untutored second language acquisition (see Perdue 1993, Klein and Perdue 1997), it could be shown that after some time, learners regularly develop a special type of linguistic system - the 'basic variety' (BV). The BV is a relatively stable and well-structured form of language not found in first language acquisition and tutored second language acquisition; its structure seems to be independent of source and target language, i.e., it seems to reflect universal properties of the human language faculty. About one third of the learners we have investigated fossilise at the level of the BV, i.e., they stick to its structural characteristics and only enlarge their vocabulary; others go more or less beyond that stage, but hardly anyone comes close to the language of the learning environment. As any other language, the BV has a lexical repertoire, i.e., a set of elementary expressions, and a grammar, i.e., rules which turn these elementary expressions into more complex ones. But there is no functional inflection of words. Still, the BV is a wellorganised and a highly efficient linguistic system. As a rule, the absence of inflectional morphology does not seriously harm the speakers' communicative potential - in contrast to the absence of appropriate lexical items. Nevertheless, many of them go beyond the BV level - not only in lexical but also in structural respects. They develop morphology. Why? First observations show that this is a very complicated and tedious process. Do they start this long march just because their social environment exhibits this kind of linguis-

2. I do not think, therefore, that there is strong evidence in favour of a 'morphological cycle', as discussed by Wunderlich (2001b). It is correct, however, that we can only oversee - at the very most - the last ten percent in the evolution of human language, and in the dark ages, there could have been some such cycles.

Why case marking?

257

tic behaviour? Or are there inherent reasons - communicative and structural deficits in the form of language they use? To the extent to which the latter is the case, that is, to which the acquisition of morphology is more than mimicking the input, we might have some evidence on the question why and how inflectional morphology develops at all. At this point, this is only a possibility and a reason to rethink the role of inflectional morphology.

4.

Learner varieties and 'real' languages

The study of second language acquisition has its origin in practical concerns - in problems of second language teaching. This background has naturally led to a particular view on SLA, for which two assumptions are constitutive: -

There is a well-defined target of the acquisition process, and only this is a 'real language', and learners miss this target at varying degrees and in varying respects - they make 'errors' in production as well as in comprehension.

We may call this view the 'target deviation perspective', and it is this view which dominates the teacher's as well as the linguist's attitude towards SLA. The alternative to the target deviation perspective is to understand the learner's performance at a given time as an immediate manifestation of his or her capacity to speak and to understand: form and function of these utterances are governed by principles, and these principles are those characteristic of the human language faculty. This 'learner variety perspective' can be characterised by three key assumptions (Klein and Perdue 1997: 307): -

-

-

During the acquisitional process, the learner passes through a series of LEARNER VARIETIES. Both the internal organisation of each variety at a given time as well as the transition from one variety to the next are essentially systematic in nature. There is a limited set of organisational principles of different kinds which are present in ALL learner varieties. The actual structure of an utterance in a learner variety is determined by a particular interaction of these principles. Learner varieties are not imperfect imitations of a 'real language' but systems in their own right and characterised by a particular lexical repertoire and by a particular interaction of organisational principles. Fully developed languages, such as Japanese, Chinese or Kpelle are special cases of learner varieties. They represent a relatively stable state of language acquisition - that state where the learner stops learning because there is no difference between his variety and the input - the variety of his social environment.

In other words, the process of language acquisition is not to be characterised in terms of errors and deviations, but in terms of the two-fold systematicity which it exhibits: the inherent systematicity of a learner variety at a given time, and the way in which

258

Wolfgang

Klein

such a learner variety evolves into another one. If we want to understand the acquisitional process, we must try to uncover this two-fold systematicity. And if we want to understand 'fully fledged languages', we should try to understand how our innate language faculty constructs them, when exposed to a certain input. This applies to all parts of a language - including inflectional morphology.

5.

Lexical repertoire and rules of composition

In whichever way views vary on the nature of human language, two points seem uncontroversial: There must be a set of elementary expressions ('lexemes'), and there must be 'rules of composition' which describe how complex expressions are formed from simpler ones. This holds for all manifestations of the human language faculty, ranging from very elementary learner varieties to 'fully fledged languages'. A lexeme is a cluster of minimally three types of features -

semantic, i.e., those which describe the lexical meaning (or 'lexical content') of an expression phonological, i.e., those which describe its phonological shape categorial, i.e., those which characterise its behaviour with respect to rules of composition.

Other properties may be linked to a word, such as graphematical features; but what seems crucial are the three types mentioned above. This does not preclude, however, that in some specific cases, semantic features are absent or phonological features are absent. What seems indispensable are categorial features; but this is perhaps a matter of dispute. Grammatical rules are traditionally divided into morphological and syntactic, depending on whether they operate within the shape of a word or go beyond the individual word; there are some borderline cases, just as there are borderline cases between 'lexicon' and 'grammar' ; these will not be discussed here. So far, I have more or less re-stated the obvious. The next point is much less uncontroversial. I would want to make a rigid distinction between two types of 'rules of composition' - those which operate on lexical information, on the one hand, and those which serve to integrate the complex expression into the context, on the other. I will call the former 'LC-rules' (for lexical content) and the latter 'CI-rules' (for context integration), respectively. LC-rules serve to form complex lexical contents from simple ones; in doing so, they also affect categorial and phonological features of the times on which they operate. Typical examples are -

the constituent which expresses agent comes first (based on semantic features) the plural of German nouns of class 17 is formed by attaching -n (based on categorial information) a lexeme of type 'article' and a lexeme of type 'noun' form an expression of type 'noun phrase' (again based on categorial information)

Why case marking?

259

and so on. Note that rules of this type are not purely 'syntactic' or 'morphological' in the traditional sense; they may also exclusively operate on semantic information provided by lexemes. There are also rules ('Sandhi') which only have to do with phonological information. All that matters is that they are stated in terms of the information provided by the lexemes involved. Typical CI-rules are, for example -

focus constituents come last lexemes which preserve information from the preceding sentence come first lexemes which preserve information from the preceding sentence are deaccented, i.e., deprived of their suprasegmental information lexemes which preserve information from the preceding sentence can be deprived of their segmental phonological features ('ellipsis')

and so on. They also include rules which concern the 'illocutionary status' of a sentence, when made in some communicative context, such as -

a question is marked by a final rise an assertion is marked by having finite component of the verb in second position an imperative is marked by bare stem in initial position

and the like. Clearly, these rules are not based on merely lexical information. After all, nothing in the meaning of the lexeme schweig- 'be silent' says that it should be used as a question, an assertion or an imperative, just as nothing in the lexical information of this lexeme tells us whether, in a given utterance, this information is new or maintained from a preceding utterance. The distinction between 'LC-rules' and 'CI-rules' is a principled one. It does not preclude that in a given language, bits and pieces of both types are clustered together to one complex rule. In fact, I believe that the apparent opacity of 'fully fledged languages' is very often due to such a clustering, whereas the separation is relatively neat in more elementary manifestations of the human language capacity. In the 'Basic Variety', we seem to have very simple rules such as 'Controller [agent] first' or 'Focus last'; the problem there is that under specific communicative circumstances, the two types of rules are in conflict and hence, when applied simultaneously, do not allow the formation of a complex expression. Such cases call for additional devices, and this, we believe, is one potential source of morphological marking, a point to which I will return in section 7.1 below. If we take 'grammar' to be the overarching term (in contrast to opposing it to 'lexicon'), then a grammar is organised as follows:

260

Wolfgang Klein

(1)

grammar lexicon clusters of features - semantic - categorial - phonological - perhaps other

rules of composition LC-rules

relate only to lexical information

CI-rules

relate to contextual information

Where does inflectional morphology fit into this picture? It belongs to the rules of composition, but to which sort? This will be discussed in the next section.

6.

What is inflection good for?

Inflection is not just a change in the shape of a word. As a rule, it operates on all three types of lexical information - phonological, semantic and categorial. Take, for example, the rule which turns the German verb stem schweig- 'be silent' into the 'past participle' geschwiegen. It changes the phonological properties in various ways: it adds the prefix ge-, the suffix -en, and it turns the diphthong [ai] into the long vowel [i:]. Next, it somehow modifies the meaning, an effect which is much harder to describe - maybe it indicates a 'posttime' ; we shall turn to this point in section 7.2 below. Finally, it also affects the categorial properties; thus, the resulting form cannot be made finite, as is the case with the bare stem; but it can, for example, be combined with an auxiliary to form a present perfect. The role of temporal marking and of finiteness in acquisition beyond the Basic Variety has repeatedly been addressed in the literature on SLA outside the classroom (see, e.g., Starren and van Hout 1996, Giacalone Ramat 1997). Therefore, I will not elaborate on these two inflectional categories here but turn to another no less fundamental such category - case marking. The more specific question to be addressed in the remainder is thus: What is case marking good for? Hard to tell. In the BV, it is strikingly absent. What do its speakers loose apart from the fact that their language does not sound like the language of their social environment? Could it be that the wisdom of the adult language learner, when not under the teacher's whip, simply chooses to ignore something because there is no reason to learn it? The idea sounds blasphemous; but then, we should come up with a clear idea what case marking is good for. Traditionally, case marking on a noun phrase may be 'absolute' or 'governed'. Examples of the first type include the 'ablativus absolutus' (his rebus gestis, Caesar pontem fecit) and 'adverbial noun-phrases', for example Roma 'in Rome' vs. Romam 'towards Rome'. The latter type also exists in German, but it is restricted to some isolated adverbials such as den ganzen Tag 'all day'. In any event, it seems a very different

Why case marking?

261

phenomenon than 'governed case marking', where a noun-phrase must be inflected in a particular way because some other element with which it is combined requires such a case marking. In German, as in most other languages, these 'case-requiring' elements are primarily prepositions and verb stems; others, such as adjectives, are marginal. Consider, first, prepositions. With one remarkable exception, case marking in prepositional phrases is just decorum; neither to the Β V speaker nor to the linguist is it transparent why ohne 'without' governs the accusative, mit 'with' governs the dative and wegen 'because of governs the genitive. In fact, native speakers were well-advised if they adopted the learner's way in this regard - no case marking at all. The only exception is variable case marking with prepositions such as in 'in', a u f on', vor 'in front of : with the dative, they denote a location, and with the accusative, they mark in addition that this location is the target of some change of location. Thus, auf dem Tisch 'on the table' indicates a place which is higher than and in contact with the table; auf den Tisch indicates in addition that this place is the endpoint of some movement. Whereas there are only a few prepositions, there are thousands of verb stems which require a particular case on their arguments; what is this marking good for? In what follows, I will discuss this question first for the 'Basic Variety' and then for 'Standard German', a language which is notorious for its complex noun declension in general and its case marking in particular ( Ί rather decline two beers than a single German noun', Mark Twain).

7. 7.1

Case marking and the 'Argument-Time Structure' of verbs Where the BV fails

Elementary utterances in the BV usually consist of an uninflected verb and one or two nominal arguments. There are three types of organisational principles; in Klein and Perdue (1997), these were called phrasal, semantic and pragmatical; the most important ones are:3

3. The following exposition is confined to what is essential in the present context; copula constructions, for example, or constructions with three arguments - rare anyway - are not discussed here. For a more comprehensive presentation, see Klein & Perdue (1997) and the literature quoted there.

262 (2)

Wolfgang Klein Structural constraints PHI. NPi PH2. NPi PH3. V -

in the Β V V V NP 2

KFZ

SEM.

The NP-referent with highest control comes first 4

PR.

Focus expression comes last

Phrasal rules exclusively operate with categorial features; hence, they are LC-rules in the terminology used above. Semantic rules operate with semantic features. Note, however, that these features do not come from the NP itself but from the verb; it is not inherent to the referent of an NP to be an agent ('high control') or a patient ('low control')· The pragmatic constraint, finally, is a clear example of an CI-rule. The fact that some constituent is focussed is not a property of its lexical information. Hence, we have a clean separation of these rule types. This is a very elegant and versatile system. But problems arise when these neat principles get into conflict. The clearest case we noted is a scene in the re-telling of Chaplin's 'Modern Times', in which one of the protagonists - the girl - is accused to have stolen a (loaf of) bread, a situation which can be easily described by (3): (3)

Mädchen stehle Brot. 'Girl steal bread.'

There are two nominal arguments, the first one is the 'controller', the second one is focussed (probably together with the verb, a point which does not matter here). These three rules taken together result in an utterance such as (3). But as the story goes on, the speaker has to express that (allegedly) it was not the girl who stole the bread but Charlie. Now, the speaker must either violate PR, as in (4a), since Charlie is focussed and hence should be in final position, or SEM, as in (4b), since Charlie is the controller and hence should be in first position: (4)

a. Charlie stehle Brot. 'Charlie steal Bread.' b. Brot stehle Charlie. 'Bread steal Charlie.'

Here, the BV system is structurally too simple: it cannot handle these conflicting requirements. There are two ways to deal with this problem. The first one consists in a ranking of the two principles, for example as in (5):

4. The 'control asymmetry' is based on the idea that the arguments of a verb can be ranked by the greater or lesser degree of control that their referents exert, or intend to exert, over the referents of the other argument(s). Strength of control is a continuum, including the possibility that two arguments rank equally high (in which case, of course, the 'control principle' cannot be decisive in what comes first).

Why case marking? (5)

263

Semantic constraints outweigh pragmatic constraints.

I suspect that native speakers of English indeed have such a ranking. They would always consider the first argument to be the 'controller'. Therefore, sentence (4b) intuitively sounds very bizarre to a native speaker of English - but much less so to a native speaker of German, where the 'controller' might easily be in final position; hence, in cases of ambiguity, they tend to follow the opposite ranking. Whichever ranking is chosen - the fact remains that one of the constraints is violated. If we adopt the English strategy, for example, it is not clear which argument is in focus; if we adopt the 'German strategy', it is not clear which element is the 'controller' (except that, in this particular example, it is unlikely that the bread is the 'controller'). 5 The other way, and in fact the only principled way, to solve the problem is the 'invention' of another device which allows the speaker to mark either what is focus or what is controller. In the case of a BV speaker, this 'invention' is not free - it is directed by what is the case in the target language, maybe also by what is the case in the source language. By contrast, the first homines sapientes, whilst in principle in the same situation, had no model to lean on: they had to create something freely. It appears that natural languages have used two options for the 'additional device' - they either use suprasegmental means, or they create a specific segmental expression, a 'morpheme'. This morpheme may be free, or it may be attached to one of the relevant words. In the present context, I will not go into suprasegmental devices. Let me just note, first, that they are widely used to this end, and second, that to the best of my knowledge, they only mark an expression as 'focussed' or 'non-focussed' but never as 'agent', 'patient' or the like. They serve CI-functions, not LC-functions. The other choice, the formation of a specific morpheme, has both options. It is possible to invent/adopt a morphological 'focus marker' (or 'non-focus marker'), and it is possible to invent/adopt a morphological 'controller marker', a 'patient marker' and so on. In SLA, the first possibility is exemplified by some learners of French who use a 'particle' [se] to mark an element in initial position as focussed - a precursor of the cleft construction 'c'est... que' (see Klein & Perdue 1997: 330). The other possibility is tantamount to case marking, either by inflection or by some free morpheme. Various options to achieve this are possible, for example -

'controller' is marked by a special suffix, and 'non-controller' is marked by another suffix only 'controller' is marked only 'non-controller' is marked 'non-focus' is marked by a special suffix (thus indicating something like 'topichood')

and so on. It may also be that the relevant marking only occurs when (at least) two arguments are present (otherwise, confusion can arise), but it is also possible that the 5. Languages differ in their ranking, and as we argued in Klein and Perdue (1989), there is some evidence to assume that learners 'transfer' the priorities of their source language into their learner varieties.

264

Wolfgang Klein

'case role' is marked in all occurrences, no matter whether there is a second argument with which it can be confused. In the case of language acquisition, the learner is not free to choose between these various options and to build his or her own system. Eventually, the learner has to copy what the social environment does, irrespective of whether he really understands it or not. Adult learners may be somewhat reluctant to do this, if it is difficult for them and if they do not see what it is good for. This may be one of the reasons why they often get stuck at one point. Children normally don't; this may be either due to the fact that they are simply better in imitating things which they do not understand, or that they are more willing to do it. No one understands why ohne has an accusative and mit has a dative; but some people simply learn it without asking, and others don't. Our ancestors, who first invented inflectional morphology including case marking, were not under the influence of an already existing system. But we have no direct evidence of what they did: all we have is the result of a long process of transformation, elaboration, reduction. A fully-fledged language, and its inflectional morphology in particular, resembles an old city on which many generations have left their traces, to the better or to the worse. This explains many oddities, such as the quaint case assignment of some verbs or prepositions; but it does not preclude a very systematic basis - a set of 'default principles'. In the next section, I will discuss what this set of default principles could look like in the case of Standard German.

7.2

Case marking and the 'argument-time structure' of German verbs

Conventional wisdom has it that verbs, in contrast to nouns, refer to 'events' or, more generally speaking, to 'situations'. This is a very misleading notion. To which situation does schlaf- 'sleep' refer? It is the entire sentence which refers to a situation, and the verb makes a - substantial - contribution to the description of this situation. Consider, for example, the situation referred to by following sentence, when uttered on some occasion: (6)

Tessi öffnete das Pförtchen. 'Tessi opened the little door.'

The verb stem öffn- 'open' indicates some properties which the two NP-referents have at some time intervals. The little door must first be not open and then open, Tessi must do something, for example turn a knob and push the door into a certain direction, or push a button, or say 'Sesame, open! - whatever; more generally speaking, she must somehow be 'in control of the situation, in contrast to the referents of the other arguments.6 If all of this is essentially correct, then THE FUNCTION OF THE VERB IS TO INDICATE PROPERTIES OF ARGUMENT-TIME PAIRS. These pairs themselves are not expressed by the verb but by noun phrases, by adverbials, by morphological variation of the stem and perhaps by other means. Sometimes, they are to be derived by context.

6. Exactly this is the origin of the 'control asymmetry' in the Basic Variety.

Why case marking?

265

What the verb itself provides are open slots to be filled appropriately, i.e. pairs of argument-time variables. In what follows, I shall use A, B, C, ... as variables for arguments and ti, t2, t3, ... as variables for time spans; an argument-time pair (briefly ATpair) is denoted by . It will be helpful to consider some examples. In (7), there is only one argument variable and one time variable, and the descriptive property is 'open'. The argument variable is specified by das Pförtchen, the time variable is only vaguely restricted by the morphological tense marking on the copula: (7)

Das Pförtchen war offen. 'The little door was open.'

In (8), there is only one argument variable, as well; but properties are assigned to it at two times, which I will call FT ('first time') and ST, respectively: (8)

Das Pförtchen öffnete sich. 'The little door opened.'

The FT-property of the single argument is 'not open', the ST-property is 'open'. In this case, the argument-variable is filled twice, by an NPn and by sich? The two timevariables are not specified (but they are restricted by the past tense marking, i.e., they must be before the utterance time). Let us now return to (6). Here, the verb assigns varying properties to two entities at different times. The entity which specifies the first argument variable is said to do something, whatever this may be, and the entity which specifies the second argument variable is first said to be not open and then, to be open. Hence, we have three ATpairs which are assigned descriptive properties by the verb stem. The lexical meaning of the verb can then be described as a Boolean cluster of elementary predications over AT-pairs (leaving aside for the moment whether the descriptive properties are adequately described by terms such as 'active', etc.): (9)

a. offenseib. öffhc. öffn-

open not-open & open not-open & open & 'active'

This does not exhaust the lexical content of verb stems. If there is more than one ATspecification, as in (9b) and (9c), then the relationship between these must be indicated, too. In the first place, this includes the temporal relation between the time spans. Thus, tj must be AFTER t¡; this is what we covered above by the labels first time FT and second time ST. If there is a third time span, as in (9c), the relation between tk (the time of B's being active) and t¡ as well as tj must be indicated, as well. For sentence (4) to be true, Tessi may still push the button of her automatic door opener, although the door is already open. But the sentence is not true if the door opened but he started his activity

7. Here and in what follows, N P n is a noun phrase marked as nominative, NP D is a noun phrase marked as dative, and NP A is a noun phrase marked as accusative.

266

Wolfgang Klein

only when it was already open. Thus, tk must overlap with t¡; it may but need not overlap with tj.8 As any lexical entry, a verb is a cluster of phonological, categorial and semantic features, the latter being called here lexical content (cf. section 6 above). The lexical content of a verb has a STRUCTURAL COMPONENT and a DESCRIPTIVE COMPONENT. The structural component is the AT-structure, which consists of the various AT-pairs together with a specification of the temporal and non-temporal relations between these. The descriptive component consists of the various qualitative or spatial properties assigned to an AT-pair. These two components can be coupled in different ways. They may be conflated into a single morpheme, for example, as is the case with öffii- (in both variants). In offensei-, the descriptive property is contributed by offen, whereas the verb sei- does not specify a qualitative or spatial property; it only has an AT-structure, and it can be made finite (in contrast to the other component offen). Other cases are possible, but I will not go into these details here. Before now turning to what all of this has to do with case marking, it will be useful to sum up in which way the present view differs from the traditional perspective. Traditionally, lexical verbs are assumed to have an 'argument structure'. It is also assumed that verbs (and more complex expressions such as full VPs) can be classified according to their inherent temporal properties into 'event types', 'Aktionsarten' etc. They have an ARGUMENT STRUCTURE as well as an EVENT STRUCTURE. The present view takes these two notions together: verbs have an ARGUMENT-TIME STRUCTURE. This has a number of consequences. One of these relates to the way in which semantic and formal 'government' is analysed, i.e., the semantic and formal restrictions which the verb imposes on the 'filling' of its argument variables.9 By semantic government, I mean 'case roles' or 'thematic roles' such as agent, theme, experiences, benefactive, patient etc. I believe that these notions, whose fuzziness has often been lamented, are nothing but a gross classification of the descriptive properties which Vs may assign to an AT-pair. I see little use in such a classification beyond an initial orientation of the 'descriptive component' - except it can be shown that such an assignment has clear structural consequences. But this already relates to the other side of government, i.e., to

8. There are other than merely temporal relations between different AT-specifications. In (6), for example, it does not suffice that Tessi did something and the door made a transition from not open to open. For this sentence to be true, this temporal coincidence must not be accidental: we assume that the latter were not the case if the former were not the case. Hence, a sort of counterfactual relation may obtain between various AT-specifications; this, I believe, is the background of the predicate CAUSE found in many decompositional analyses of verbs. As is usually the case for lexical entries, such a relation, where it exists, may be individually marked for each entry or covered by a lexical default rule; this is an empirical issue which we will not follow up here. 9. Another one concerns the notion of 'event time'. If, for example, the verb in itself contains several temporal variables, a notion such as 'event time' turns out to be a gross oversimplification: what, for example, is the 'event time' in (6): is it t¡, tj or tk ? Or is it some interval which contains all or some of these? Similarly, notions such as 'anteriority' or 'posteriority' or even 'simultaneous' tum out to be highly problematic. What, for example, is the 'posttime' in (6)? Is it the time where the little door is open, or is it the time after Tessi's activity?

Why case marking?

267

constraints on morphological properties such as accusative, dative or to syntactic properties such as 'subject' or 'direct object'. These constraints are traditionally seen as a part of the verb's categorial features. It would be much more elegant, however, if they could be derived from the AT-structure or from the Descriptive Component. Thus, one might look for principles such as 'If an argument is described at two times, then it is realised as NPA' or 'An argument which is assigned the property 'active' is always encoded as an NP n '. It is unlikely that this is possible for the entire verb lexicon of a language; but it should be possible for the default case. Languages are the product of a complex historical development, and just as there are idiosyncrasies in other parts of the lexical information, we should expect them here, as well. With this caveat in mind, it appears that German has a number of very simple default principles for case assignment by the verb, and a rule which determines their relative weight in cases of conflict. These are: (10) DP DP DP DP

A: B: C: D:

DP E:

One argument variable is filled by an NPNTwo-times argument variables are filled with NP A . One-time argument variables are filled with NP D . If the verb assigns the property 'active' to an argument, then this argument is realised as NPNIf the verb is lexically empty, then the argument which expresses the descriptive property is realised as NP A . n

In cases of conflict, DP A is strongest. This means that an NP n can encode one-time arguments as well as two-times arguments; it also encodes a 'controller', if there is such an argument. Other NPs are much more restricted in what they can encode. The rules in (10) are extremely simple, and if we are to believe the Scholars that simplicitas est sigillum veritatis, then there is good reason to assume that they are true. But there are also various problems, one of which I will discuss now (for some other problems, see Klein 2000). Under the analysis suggested here, an NP A is always a 'two-times argument'. This is plausible in cases such as Dieter opened the door, where the door is first not open and then open. But it is not plausible for other cases, such as Dieter hated his uncle, where the second argument is NPA. This fact is nicely reflected in Dowty's idea that the 'prototypical' patient is a 'change-of-state' argument (Dowty 1991). It appears to me, however, that the notion of 'change-of-state' in general confounds two interrelated but in principle independent features of verb meaning, which are clearly kept apart in the notion of AT-structure. These are its DESCRIPTIVE PROPERTIES, such as being open or being in Heringsdorf, and its TEMPORAL STRUCTURE i.e., the intervals and subintervals at which some argument is assigned these descriptive properties. Compare, for exam10. Under this view, there is not so very much a distinction between 'structural case' and 'lexical case' but between 'assignment by default' and 'assignment by exception'. 11. This principle primarily concerns the descriptively empty 'two-place copula' haben 'have', as in Karl hat Angst/rote Haare 'Karl is scared (lit.: has fear)/Karl has red hair.'.

268

Wolfgang Klein

pie, the two sentences Dieter slept and Dieter woke up. In the first case, a descriptive property ('asleep') is assigned to Dieter at some time t¡; morphological marking indicates that this time is (normally) in the past. In the second case, two distinct properties are assigned to Dieter; first, say at t¡, he must be asleep, and then, say at tk, he is not asleep. Hence, the lexical content of the verb to wake up comprises two temporal variables, tj and tk, which are sequentially ordered and which are associated with different descriptive properties. In this case, the properties are mutually exclusive, a constellation which is perfectly well covered by the notion of a 'change-operator' (such as the familiar BECOME). But it is also imaginable that the lexical content of a verb lexeme has two time variables with less divergent properties. Take, for example, The temperature fell and The temperature rose. For the first sentence to be true, it is necessary that at some interval tk, the temperature is 'lower' on some scale than at some earlier interval tj; for the second sentence to be true, the temperature must be 'lower' at the first interval tj than at the second interval tk- Verbs of this sort are not 'telic' or 'resultative'. The Vendler tests identify them as activities, rather than as accomplishments or achievements. I am not sure whether they should be described by a change-operator. Is it possible that the lexical content of a verb provides two time variables with IDENTICAL descriptive properties? At first, this idea sounds bizarre: why should this be the case? But compare the two sentences Dieter was in Heringsdorf and Dieter remained in Heringsdorf. They both assign a 'static' spatial property to Dieter. But somehow, the second sentence gives the impression that Dieter was there at some time tj and then, at some time tk, could have gone but hasn't. The difference is brought out more clearly if we add a modal verb, such as in Dieter was allowed to be in Heringsdorf and Dieter was allowed to remain in Heringsdorf. In the first case, the permission concerns his entire stay, whereas in the second case, it only concerns the second subinterval. In order to describe the semantic effect of this morphosyntactic operation appropriately, we must assume that it has SELECTIVE access to the verb content - to a subinterval which is descriptively not different from the first interval. The addition of a modal verb is not the only morphosyntactic process which demonstrates this. Negation is another case. In Dieter was not in Heringsdorf, his entire stay there is denied; in Dieter did not stay in Heringsdorf, it is only denied that he was not there at a second subinterval during which he could have been there. The possibility to have two temporal intervals with the same descriptive properties is not confined to verbs with only one argument. It is also found with 'transitive verbs' such as to leave in sentences like Could you please leave the door open. Here, the idea is that the door is open at a first time and should also be open at a second time, in contrast to the possibility that, due to the addressee's action, it is closed at this second time. The conclusion is therefore, that we must carefully distinguish whether a verb content has one, two or even more temporal variables and whether the descriptive properties assigned to these variables are identical, slightly different or 'radically' different. An operator such as BECOME conflates these notions. There is no change in a verbs such as to remain or to stay, and similarly in German bleiben; still, there are two subin-

269

Why case marking? 12

tervals which are selectively accessible to morphosyntactic operations. The linguist's decision of whether a verb stem involves one or more temporal variables can therefore not be based on mere semantic intuition; it must explore how the content of this verb stem can be modified by all sorts of morphosyntactic operations. Such an operation is the formation of the 'past participle', briefly mentioned in section 6.1 above. 13 1 will call this operation GE-; V is the lexeme to which it applies. The way in which the attachment of GE- affects the phonological information of V is complex but well studied; it brings about the forms of the past participle. The change in categorial information is somewhat less clear; some operations applicable to the verb are now blocked, for example, the past participle can no longer be made finite (unless some other verb is attached). How does GE- affect the semantic features provided by V? Under the present approach, it changes its AT-structure and possibly adds further descriptive properties. I assume that the latter is not the case but that GE- only operates on the AT-structure as follows: (11) The past participle denotes ST-properties of V. Remember that under the present approach, ST-properties are relative to an argument, typically the NP A . If the verb does not provide any ST-properties, because there is no appropriate AT-pair specified for two times, then the past participle can still be formed; but when attached to an argument, the resulting construction is not interpretable. This is the case for verb stems such as schweig- 'be silent' or gehör- 'belong'; therefore, das geschwiegene Kind 'the been-silent child' or der (mir) gehörte Apfelsaft 'the (to me) belonged apple juice' should not make sense, and so it is. A stem such as einschlaf 'fall to sleep', by contrast, does have two time slots for its single argument, and therefore, das eingeschlafene Kind should be possible, and so it is. The stem öjfn- has two time slots only for one of its arguments, and therefore, das geöffnete Tor assigns the SL-properties (being open after having been not open) to this argument, i.e., to the door. In conclusion, the assumption that an accusative indicates, in the default case, that an argument has two time variables makes perfect sense, so long as we do not confuse the existence of two time variables with a 'change of state'. 14 On the other hand, the AT-analysis suggested here explains a number of additional facts of German. Only transitive objects 'passivise' - that is, only these objects provide the two time slots that are necessary for a past participle to apply. As just stated, it explains why das geschwiegene Kind is odd, whereas das eingeschlafene Kind or das gestillte Kind 'the breast-fed child' are fine. It naturally accounts for the ambiguity between 'unergative' 12. There are also exceptions in the opposite direction. The situation described by Dieter worked in a shoe factory includes many quite heterogenous subintervals, hence many 'changes'. None of these subintervals, however, is selectively accessible to any morphosyntactic operation, such as negation, adverbial modification or addition of another verb stem. 13. For a detailed discussion of how its various usages can be captured under the present treatment, see Klein (2000). 14. I do think, however, that there are accusatives which encode a 'single-state argument', for example measure phrases such as 'The colossus of Rhodes weighed one hundred tons'. Characteristically, these cannot form a passive.

270

Wolfgang Klein

and 'unaccusative' verbs with one argument. The argument of an 'unergative verb' is specified at one temporal interval, and thus, it behaves like the 'subject' of a transitive verb. The argument of an 'unaccusative verb' is specified at two time intervals, that is, it behaves like the 'direct object' of a transitive verb. Finally, the analysis also makes plausible why auf dem Tisch, i.e., with dative, is 'static', whereas auf den Tisch, with its accusative, is 'dynamic' - the latter involves two temporal variables. I do not claim that principles such as 10 account for all instances of case marking. They are DEFAULT PRINCIPLES. After all, languages are the product of a complex historical development, and just as there are idiosyncrasies in other parts of lexical information, we should expect them here.

8.

What is a subject, what is an object Right after I had joined the project, I began to study linguistics, and very soon, I was deeply bewildered by the fact that there is not the faintest unanimity in this apparently so precise, this allegedly so mathematiced and physicalised branch of science. In this discipline, the authorities don't even concur on most elementary, quasi introductory issues such as what's a morpheme or a phoneme. (Stanislaw Lem, The master's voice)

It has often been noted (see, e.g., Reis 1982) that what is called 'subject' is actually a cluster of heterogenous features - morphological features such as case marking, syntactic features such as position, semantic features such as agentivity, and pragmatical features such as topic status; these may but need not be present. In other words - this notion is a cloud. This is in no way different for 'direct object'. How is it defined? In school grammar, no definition is given at all; normally, these notions are illustrated by examples, and the relevant generalisations are left to the reader. Modem approaches, lest they simply continue this tradition, define them either in terms of case roles, such as 'theme, benefactive, patient', etc., or in terms of tree geometry. The first way is unsatisfactory because these notions themselves are most unclear a fact that has often been deplored (see the extensive discussion in Helbig 1973). Moreover, they do not make sense in many cases. It may be justified to call the NPp a 'benefactive' in Er half ihr 'He helped her', but surely not in Errötend folgt er ihren Spuren 'Blushing he followed her traces'. The tree geometry approach, generally used in Generative Grammar and in some other branches of structural linguistics, looks much clearer - in fact, so much clearer that it is most often taken as self-evident. One might say, for example: 'The direct object is the first NP immediately dominated by V on D-structure' (this is the definition in Chomsky's 'Aspects', where this idea was first worked out, the argument is analogous for other variants of generative grammar). But this clarity is only apparent. Such an account only shifts the problem to the question WHY A PARTICULAR TREE

Why case marking?

271

It is neither the Lord nor the Pope who places einen Apfel 'an apple' immediately under V in Adam aß einen Apfel 'Adam ate an apple'. It is the linguist who draws pictures. Trees, as this term is used in linguistics, are abstract structures based on two types of structural relations between its elements dominance and precedence. These, and only these, relations are available to represent the structural relationship between simple or complex expressions. It is the linguist's task to decide why certain elements are connected to each other by a vertical stroke, i.e., by a dominance relation, and why a certain element is placed in the tree such that it precedes some other element. Very often, the two relations available turn out to be insufficient, and the linguist's way out is usually to stipulate several trees, together with some mechanism to relate these to each other ('transformations', 'reanalysis', and other ones). In any case, it is not the tree which says why something is connected to something else in a particular way - why, for example, an NP is in the 'direct object relation' to a V. It is the linguist who constructs the tree in a given case, and this decision must be based on clear and reasonable criteria. Mere reference to the position in the tree is no solution. Recently, a number of linguists, in particular Manfred Bierwisch, Paul Kiparsky and Dieter Wunderlich, developed an analysis of this problem which potentially overcomes these difficulties, although it operates with trees, as well. Details vary; the most elaborate version is found in Wunderlich (1996), see also Wunderlich (1997, 2001a). In this approach, the asymmetry of argument variables is defined by the features 'higher role available (in the same clause)' - 'lower role available'. Consider, for example, the sentence Tessi gab ihm den Schlüssel 'Tessi gave him the key'. The variable filled by den Schlüssel 'the key (ACC)'is assigned the feature complex 'higher role available, no lower role available', the variable which is filled by Tessi gets assigned the feature complex 'no higher role available, lower role available', and finally, the third variable, which corresponds to ihm 'him (DAT)' gets assigned 'higher role available, lower role available.' Morphological case marking is then easily defined on the basis of such a feature complex, such as, for example, 'no higher role available, lower role available' is marked by nominative in German. This is a very elegant approach, indeed. But it raises two basic problems. First, which independent criteria are crucial to decide why some element is 'higher' than some other element? They cannot be based on morphological marking, because this would render the analysis circular. Are they based on case roles - say 'benefactive' is lower than 'agent' but higher than 'theme'? Then, we are back to the familiar problems with these notions. Or do they exploit the depth of embedding in lexical decomposition? But how would this work for verbs such as beobacht- 'watch' which are not assumed to be lexically decomposable? - Second, I do not see how this analysis works for VI with NP n and NPD alone, such as helf- 'help', gebühr- 'be due to' or ähnel'resemble'? In Wunderlich (1997), this case is analysed as a lexically marked deviation from the case assignment dictated by the 'role hierarchy'. This is surely not false; but it is not satisfactory, either. After all, there should be a reason why it is possible to say die von uns unterstützten Flüchtlinge 'the refugees supported by us', but not die von uns geholfenen Flüchtlinge 'the refugees helped by us'. In other words, this deviation STRUCTURE IS ASSUMED IN A SPECIFIC CASE.

272

Wolfgang Klein

is NOT just a lexical idiosyncrasy of case marking. The lexical content must contain some feature which predicts that THIS FACT IS ON A PAR WITH THE PECULIAR CASE MARKING; precisely this is done by the 'argument-time'-analysis suggested above.

9.

Conclusion Was soll der Scheiß? (Unknown linguist, repeatedly)

I surely do not believe that the analysis sketched in the preceding sections is the final answer to the problem of why there is case marking. There is hardly an area in linguistics in which this is so manifest as inflectional morphology with which linguists have now been struggling for more than two thousand years. Here as elsewhere, being the inheritor of a long tradition is both beneficial and burdensome. But I firmly believe that looking with an open eye to the way in which second language learners try to make sense of the sounds that hit their ears may help us to get a fresh understanding of the principles that rule all manifestations of the human language faculty.

References Dowty, David (1991). Thematic proto-roles and argument selection. Language 67, 547-619. Giacalone Ramat, Anna (1997). Progressive periphrases, markedness, and second language data. In Language and its Ecology. Essays in Memory of Einar Haugen, Stig Eliasson and Ernst H. Jahr (eds.), 261-285 Berlin: Mouton de Gruyter. Gould, Stephen Jay (1989). Wonderful Life. The Burgess shale and the nature of history. New York: Norton. Helbig, Gerhard (1973). Die Funktionen der substantivischen Kasus in der deutschen Gegenwartssprache. Halle: Niemeyer. Jellinek, Max Hermann (1913). Geschichte der neuhochdeutschen Grammatik. Erster Halbband. Heidelberg: Carl Winters Universitätsbuchhandlung. Klein, Wolfgang (1998). Assertion and Finiteness. In Issues in the Theory of Language Acquisition, Norbert Dittmar and Zvi Penner (eds.), 225-245. Frankfurt/M.: Lang. —(2000). The argument-time structure of recipient constructions in German. To appear in Typological Studies on West Germanic, Werner Abraham and Jan-Wouter Zwart (eds.). Amsterdam: Benjamins. —(2000). An analysis of the German Perfekt. Language 76, 358-382. —; Perdue, Clive (1989). The learner's problem of arranging words. In The crosslinguistic study of sentence processing, Brian MacWhinney and Elizabeth Bates (eds.), 292-327. Cambridge: Cambridge University Press.

Why case marking?

273

—;— (1997). The basic variety, or Couldn't natural languages be much simpler. Second Language Research 13, 301-347. Perdue, Clive, (ed.) (1993). Adult language acquisition: Cross-linguistic perspectives. Cambridge: Cambridge University Press. Reis, Marga (1982). Zum Subjektbegriff im Deutschen. In Satzglieder im Deutschen, Werner Abraham (ed.), 171-211. Tübingen: Niemeyer. Starren, Marianne and Roeland van Hout (1996). The expression of temporality in a second language. Zeitschrift für Literaturwissenschaft und Linguistik 104, 35-50. Wunderlich, Dieter (1996). Dem Freund die Hand auf die Schulter legen. In Wenn die Semantik arbeitet. Klaus Baumgärtner zum 65. Geburtstag, Gisela Harras and Manfred Bierwisch (eds.), 331- 360. Tübingen: Niemeyer. —(1997). CAUSE and the structure of verbs. Linguistic Inquiry 28, 27-68. —(2001a). The force of lexical case: German and Icelandic compared. To appear in The Nature of the Word: essays in honor of Paul Kiparsky, Kristin Hanson and Sharon Inkelas (eds.). Cambridge, Mass.: ΜΓΓ Press. —(2001b). Why is there morphology? Paper presented at the 23rd Annual Meeting of the Deutsche Gesellschaft für Sprachwissenschaft, Leipzig, March 2001. [References relate to the handout].

Structural cases in Russian

Ilse Zimmermann

1.

Objectives*

On the basis of data from contemporary noncolloquial Russian, this paper tries to bring together 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 the structural arguments of verbs and those of event nominalizations. The particular questions to be raised are the following: -

Which types of cases are to be distinguished? Which complements of lexical categories count as structural arguments? Which systematic case alternations of complements show up in connection with morphological operations on the lexical governor? Which rules guarantee the correct case realizations of argument expressions? Which types of configurations and case features are involved?

I will argue for a strict differentiation of universal semantico-syntactic case features and language-specific morphosyntactic ones. Furthermore, I will argue that there are regular correspondences between them. 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 of the argument structure in the lexical entries of functor expressions. Section 3 demonstrates the far-reaching parallelism of verbal constructions and their nominalizations, with systematic case variance of structural arguments. Section 4 is concerned with case licensing. It offers a system of rules correlating abstract seman-

* This investigation is dedicated to Dieter Wunderlich with great respect and many thanks for stimulating discussions, human warmth and intellectual openness for alternative solutions. I am indebted to the editors of this volume for kind support and to Natalja Gagarina for help with the translation of the examples into English. Valuable questions of the editors and of one reviewer helped me to clarify some of my assumptions on cases.

276

Ilse Zimmermann

tico-syntactic case features in the argument structure of lexical governors and the morphosyntactic case features of the corresponding argument expressions.

2.

The framework

Within a minimalist framework of sound-meaning correlation (Chomsky 1995), the analysis follows a lexicalist conception of morphology (Stiebeis & Wunderlich 1994, Wunderlich & Fabri 1995, Wunderlich 1997b) and the differentiation of Semantic Form and Conceptual Structure (Bierwisch 1983, 1987, 1997, Bierwisch & Schreuder 1992). I assume Phonetic Form (PF), Logical Form (LF) and Semantic Form (SF) to be the relevant grammatically determined levels of representation. My conception of syntax is restrictive. I presuppose the structural layers in (1) and (2) for CPs and DPs, respectively. (1) (2)

CP MoodP TP NegP vP* VP DP FP nP* NP

In the base structure, argument expressions appearing with structural cases of verbs are placed in [Spec VP] or [Spec vP]. Correspondingly, argument expressions appearing with structural cases of de verbal nouns are placed in [Spec NP] or [Spec nP]. DPs with nonstructural cases and PPs appear in the complement position. The verb raises to Mood or to C (Zimmermann 1999) and - in parallel to sentence structures - the de verbal noun overtly moves to a high functional projection F (Alexiadou 1999), so that all argument expressions of Ν are to its right (Haider 1992). I will not discuss the nature of the category F. Possibly, it is a further n. The syntactic configurations on the level of LF are the input for semantic interpretation. This implies that syntactic movements of constituents can have an effect in SF (Zimmermann 1999). 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 expression DP¡, occupies some derived position whereas the tail of the chain t¡ is in the complement or specifier position of V, ν, Ν or n. (3)

(DP;,... , ti)

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 G¡. They are associated with λ-operators λχ; which represent the argument positions of the respective functor expression.

277

Structural cases in Russian (4)

λχ η ... λχι Gn Gì argument structure

[... χι ... xn ... ] predicate-argument structure

The argument positions Xx¡ are ordered from right to left according to the relative depth of embeddedness of the arguments x¡ 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 mnemonical reasons, I will represent it as s (referring to situations).1 The other arguments constitute participant, propositional or predicate arguments. (5)

λχ η ... λχι Às [... s ... Χι... x„ ... ] Gn Gì with s G e, x¡ e {e, t, }

λχι in (5) represents the argument position of the external argument, and λχ η is the argument position of the lowest internal argument. For DP arguments, the grammatical features G¡ are case requirements (Zimmermann 1967) which must be fulfilled by the corresponding DPs as heads in LF chains. So we have the following hierarchy of argument positions:2 (6)

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 in (6) is also valid for the classification of argument expressions with respect to their predictable or unpredictable case forms. Typically, the lowest internal argument expression can have idiosyncratic (lexically determined) case marking.3 As regards the nature of case requirements on argument expressions, I assume the following: Morphologically expressed cases of DPs are understood as analysable morphosyntactic characteristics of inflected nouns, pronouns, adjectives, determiners and quantifiers. These characteristics are represented by a small number of morphosyntactic case features, which allow morphological cross-classification and underspecification (see section 4). I assume that idiosyncratic case requirements on argument expressions are directly referred to by these case features as instantiations of G¡ in the argument structure of lexical governors (see the examples in (13)-(17)). In contrast, 1. As in Bierwisch (1987, 1989) and in Bischof (1991), the referential argument s of verbs and event nominalizations is integrated into the predicate-argument structure by the constant INST ('instantiates') of type . 2. Concerning the coincidence of the external and referential argument positions of nouns, see Williams (1981) and Stiebels (1999, 2000a, 2001). 3. I disregard here idiosyncratic case marking of the external argument (see Wunderlich 1999).

278

Ilse Zimmermann

predictable case forms of argument expressions are understood as structural cases. They are conditioned by the semantic argument hierarchy. Wunderlich (1997a, 1999, 2000), Stiebels (1996, 1997, 2000a, 2000b, 2001) and Wunderlich & Lakämper (2001) posit configurational case features for complements of verbs and nouns in general, without discriminating between semantico-syntactic and morphosyntactic feature systems. Their features are [±hr] (there is a/no higher role), [±lr] (there is a/no lower role), as proposed analogously by Kiparsky (1992). These features are used to characterize case requirements on argument expressions and the morphological case realizations. 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 morphosyntactic case features of the pertinent argument expressions. With respect to German, I understand morphological case classifications as they are found, for instance, in Bierwisch (1967) or Gallmann (1998), to be autonomous morphosyntactic qualifications which are not reducible to a system of semanticosyntactic 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 the semantically conditioned abstract case features [±hr] and [±lr] as argument addresses in the argument structure of functor expressions. Lexical cases are unpredictable case forms of argument DPs idiosyncratically required by the respective governor. They are represented by language-specific morphologically conditioned features. Semantic cases are morphosyntactic case feature bundles associated with DPs which get a semantic interpretation (Zimmermann to appear).

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. The main concern is DPs with structural (i.e. predictable) cases, in short structural arguments. It will be shown which case requirements on argument expressions remain unchanged in nominalizations and which arguments receive alternative case realisations.

3.1

The argument structure of verbs and their nominalizations

Words as syntactic atoms are fully inflected items. They are represented syntactically with all affixes of word formation and inflection. With Bierwisch (1989) and Bischof

Structural cases in Russian

279

(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. 4 Verbs and event nominalizations have the same semantic characterization. The nominalising suffix simply converts the verb into a noun without changing the SF. 5 The following noun phrases with deverbal heads (in (7a)-(10a)) illustrate the case realizations of the pertinent argument expressions, contrasted with infinitival phrases (in (7b)-(10b)). The examples are given with normal word order. 6 (7a) shows the nominalization of an intransitive base verb, (8a) of a transitive base verb and (9a) of a ditransitive base verb. (7)

a.

vyzdorovlenie pacient-a recovery patient-GEN 'the recovery of the patient' b. vyzdorovet' 'recover'

(8)

a.

znanie rebënk-om jazyk-a knowledge child-lNSTR language-GEN 'the knowledge of the language by the child' b. znat' jazyk know language.NOM 'know the language'

(9)

a.

nemedlennoe soobSöenie institut-ami firm-e immediate informing institute-lNSTR.PL firm-DAT svoich zakaz-ov their order-GEN.PL 'the institutes' immediate informing of the firm about their orders' b. nemedlenno soobäöit' firm-e svoi zakaz-y immediately inform firm-DAT their order-ACC.PL 'inform the firm immediately about their orders'

In (10) and (11), the internal argument idiosyncratically shows up in the dative and in the instrumental, respectively. (12) illustrates a nominalization which corresponds to a causative and to a reflexive anticausative verb with an idiosyncratic dative complement.

4. In contrast to this position, see Schoorlemmer (1995) and Alexiadou (1999). 5. I do not consider here the omissibility of argument expressions in constructions with a deverbal noun as head. 6. I do not consider derived word order variations 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 genitive complement need not be adjacent to the noun. In the nominalizations, the deverbal noun precedes the highest structural argument expression. This results from head movement of Ν to F (see section 2, (2)).

280

Ilse Zimmermann

(10) a.

izmena muz-a zen-e betrayal husband-GEN wife-DAT 'the betrayal of the wife by the husband' b. izmenit' zen-e betray wife-DAT 'betray the wife'

(11) a.

obmen tovariSô-ej opyt-om exchange comrade-GEN.PL experience-lNSTR 'the exchange of experience by the comrades' b. obmenjat'sja opyt-om exchange experience-lNSTR 'exchange experience'

(12) a.

obuôenie (mater'-ju) rebënk-a ötenij-u teaching/learning (mother-INSTR) child-GEN reading-DAT 'the teaching of reading to the child by the mother' 'the learning of reading by the child' b. obuöit' rebënk-a ötenij-u teach child-ACC reading-DAT 'teach the child reading' c. obuöit'-sja òtenij-u learn-REFL reading-DAT 'learn reading'

In (13)-(17), I give the lexical representations of some verbs and event nominalizations. I include semantic and morphosyntactic information which is relevant for the case realization of argument expressions. The predicate-argument structure is unanalysed. Positions for structural arguments are associated with the abstract case features [±hr] and [±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 morphosyntactic 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 (13)-(15), the case information is systematic, redundant and therefore omissible. In contrast, the internal argument of the lexical entries in (16)-(17) is lexically marked, a property that must be learnt. (13) vyzdorovet' 'recover' - vyzdorovlenie vozniknut' 'emerge' - vozniknovenie λχ λβ [... s ... χ ... ] -hr -Ir V: NOM Ν: GEN

'recovery' 'emergence'

Structural cases in Russian

281

(14) znat' 'know' - znanie 'knowledge' usvoit' 'acquire' - usvoenie 'acquisition' Xy λχ Xs [... s ... χ ...y ... ] +hr - h r -Ir +lr V: ACC NOM Ν:

GEN

INSTR

(15) soobscit' vrucit' λζ +hr -Ir V: ACC

'inform' - soobscenie 'information' 'hand in' - vrucenie 'handing in' Xy λχ λβ [... s ... χ ... y ... ζ ... ] +hr - h r +lr +lr

(16) izmenit' pomoc' Ày +R +P V: DAT

'betray' - izjnena 'betrayal' 'help' - pomose' 'help' λχ λβ [... s ... χ ... y ... ] -hr -Ir

DAT NOM N : GEN DAT INSTR

NOM Ν : DAT GEN

(17) obmenjat'sja 'exchange' - obmen 'exchange' zanimat'sja 'be engaged' - zanjatie 'engagement' λy λχ λβ [... s ... χ ... y ... ] +Ρ -hr -Ir V : INSTO NOM Ν : INSTO GEN

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 on argument expressions of verbs are inherited by event nominalizations. The same is true for the structural dative, as shown by (15). Bayer et al. (2001) observe that dative complements behave like complements with idiosyncratic case marking insofar as 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 persistence 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 (15) and (9)). Bayer et al. (2001) classify only the nominative and

282

Ilse

Zimmermann

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. In my view, what is to be regarded as structural cases are the dative of complements of ditransitive verbs and that of the complements of their nominalizations along with the genitivas subjectivus and the genitivus objectivus in nominalizations. Furthermore, I propose that the respective phrases be represented as DPs. Alternating case realizations of structural arguments are connected with the external argument position λ χ in (13)-(17) and with the lowest position for a structural argument, Xy in (14) and λζ in (15). The accusative argument of transitive and ditransitive verbs corresponds to the genitive argument of the deverbal noun, whereas the nominative argument of these verbs corresponds to the instrumental. The nominative argument of intransitive verbs as in (13), (16) and (17) appears in the genitive, in nominalizations. These regularities concerning alternating case forms of structural arguments must be captured by the rules which define the relationship between the abstract semanticosyntactic case features [±hr] and [±lr] of structural argument positions and the morphosyntactic case features of the respective DPs as argument expressions (see section 4).

3.2

Reflexive verbs and their nominalizations

In Russian, nominalizations of reflexive verbs, as in (17) and (18a), do not combine with the reflexive morpheme -sja as (18c) shows. This is in contrast to Polish (cf. formowac (siq)lformowanie (siç) 'form/formation') and to Nahuatl (Stiebels 1999). (18) a.

Vertolët prizemlil-sja. helicopter.NOM landed-REFL 'the helicopter landed' b. Vertolët prizemljal-sja/byl prizemlën (odnim passazir-om). helicopter.NOM landed-REFL/AUX landed one passenger-INSTR 'the helicopter was landed (by one of the passengers)' c. prizemlenie vertolët-a (odnim passazir-om) landing helicopter-GEN one passenger-INSTR 'the landing of the helicopter (by one of the passengers)'

The deverbal noun prizemlenie 'landing' in (18c) is ambiguous. It corresponds to the reflexive verb prizemljat'sja/prizemit'sja in (18a) and to the transitive verb prizemljat' /prizemit', which in (18b) 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 morphosyntactic 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

Structural cases in Russian

283

elements are always correlated with the lowest structural argument position, i.e. they appear where the absent true argument in the structural accusative would. Passive, middle and anticausative constructions are characterized by the absence of the external argument. 7 In addition to (18a) with an anticausative reflexive verb, (19) is a case of reflexivum tantum, which characteristically does not combine with an accusative argument. (20) illustrates the reflexive passive of an imperfective verb, and (21) the middle. (19) Mal'öik boy.NOM

smeët-sja. laughs-REFL

'The boy is laughing.' (20) Plan plan.NOM

razrabatyvaet-sja. worked.out-REFL

'The plan is worked out.' (21) Knig-a choroso prodaët-sja. book-NOM well sells-REFL 'The book sells well.' In all these constructions, the reflexive exponent is added on the basis of the morphosyntactic feature [+refl] of the respective verb. In nominalizations, this marking does not apply. In (22)-(23), I give complex lexical representations for two causative/anticausative verb pairs and their event nominalizations. These lexical entries respect the farreaching 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. 8 As in German, the deverbal noun of such pairs is systematically ambiguous.

7. With Bierwisch (1997), I assume that anticausative verbs are related to the corresponding causative verbs in the lexicon. The respective pairs constitute a complex lexical entry with the systematic absence of the causer argument and the semantic constant DO-CAUSE in the SF of the anticausative verb. In contrast, the verb's SF is not reduced in passive and middle constructions. Its external argument is simply blocked. 8. In (22) and (23), α and β are variables for the values + and - of morphosyntactic features and for the presence or absence of formatives and semantic components put in parentheses. Causative verbs like prizemljat'/prizemit' 'land' are nonreflexives. Anticausative verbs like prizemljat'sja/ prizemit'sja 'land' lack the semantic component of causation and the causer argument x.

284

Ilse Zimmermann prizemit'f sja ).a 'land' Ày ahr a. b.

+lr

ACC

NOM

Ν

GEN

INSTR

V

NOM

Ν

GEN

obucit'(sja) .a

a. b.

3.3

V Ν V Ν

prizemlenie

'landing'

-hr

-Ir V

-

( λ χ ) α Xs [... s ... (... χ ... ) α ... y ... ]

'teach, learn'

Xy

+R

ahr

+Ρ

-lr

+lr

DAT

ACC

NOM INSTR

DAT

GEN

DAT

NOM

DAT

GEN

-

obucenie

'teaching

( λ χ ) α Xs [ s ... (...x ... ) a . . . y ... ζ

λζ

-hr

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 a structural case. It should not be confused with the agentive instrumental phrase in passive constructions. (8) illustrates a nominalization of a stative verb which does not have any passive. Nevertheless, the external argument appears in the instrumental. Moreover, (24) demonstrates that the agent argument in the instrumental cannot bind the reflexive pronoun svoj- in passive constructions - in contrast to the instrumental in nominalizations (see (9)): (24) * Nemedlenno soobäöalis' immediately were.informed

firm-e firm-DAT

institut-ami

svoi zakaz-y

institute-INSTR.PL their order-NOM.PL

'the firm w a s informed immediately b y the institutes about their orders'

Thus, I differentiate between the structural instrumental of the external argument in event nominalizations as in (8), (9) and (12), the lexical instrumental of the internal argument as in (11) and the semantic instrumental of modifiers including the so-called argument adjunct of passive constructions.9

9. I leave it undecided whether the predicative instrumental in copula sentences or in constructions with secondary predicates is a structural or semantic case (see Geist 1999, Demjjanow & Strigin 2000a, 2000b).

Structural cases in Russian 3.4

285

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 structural adnominal genitive of event nominalizations.10 As is evident from the examples and the lexical entries in 3.1 and 3.2, 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: (25) a. deverbal noun based on intransitive verb (...) λχ Xs [... s ... χ ... ] -hr -lr GEN

b. deverbal noun based on transitive verb (...) Xy λχ Às [... s ... χ ... y ... ] +hr - h r -lr +lr GEN

c. deverbal noun based on ditransive verb λζ λy λχ λβ [... s ... χ ... y ... ζ ... ] +hr +hr - h r -Ir +lr +lr GEN

d. relational noun λy λχ t... χ ... y ... ] +hr - h r GEN

(25a) represents deverbal nouns of intransitive verbs like vyzdorovlenie 'recovery', pomose' 'help' and deadjectival nouns like zavisimost' 'dependence'. (25b) and (25c) characterize deverbal nouns of transitive and ditransitive verbs like usvoenie 'acquisition', soobscenie 'information', and (25d) 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 1999, 2000a, 2001) in following Bierwisch (1989) and Bischof (1991) with regard to the treatment of event nominalizations. I do not accept Stiebels' proposal to handle

10. For the argument-modifier distinction of genitive phrases, see Zimmermann (1991) and Partee & Borschev (2000).

286

Ilse Zimmermann

relational, semantically enriched sortal nouns and event nominalizations alike. In Stiebeis' 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: (26)

λζ λγ λχ Xs [ - s ... χ ... y ... ζ ... ] +hr +hr +hr - h r

Since the adnominal structural genitive is classified by Stiebels as a [+hr] linker, all non-highest argument expressions of nouns could be genitive phrases. Several additional principles of case realization are necessary to avoid this. I do not need such additions. I assume that the argument structure of verbs is preserved in nominalizations, including the information on structural cases (see (25a-c)).

4.

Case realizations and case licensing

DPs can emerge with morphosyntactic case information of the language-specific case system.11 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 (Stiebels 2000a, Wunderlich & Lakämper 2001). It must be emphasized that my conception of syntax does not assume any movements of a DP in order for it 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

11. In addition to inflectional case markings as in Russian, "cases" of DPs can be realized by adpositions or by syntactic configurations.

Structural

287

cases in Russian

features and by adding the feature obl(ique), whose specifications I borrow from Franks ( 1 9 9 5 ) : ^ (27) Jakobson's (1936) enriched case system R NOM ACC DAT INSTR GENL LOCI GEN2 LOC2

Ρ

+ +

u

G

obi

+ +

+ + + + + +

+ + + + + +

+ + 13

with the correlations 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) 14 A s the features of Jakobson's system serve morphosyntactic and semantic generalizations, it is no accident that L D G ' s features [±hr] and [±lr] roughly coincide with Jakobson's features [±R] and [±P], respectively: (28) The case system of L D G

NOM ACC DAT GEN

hr

lr

+ + +

+

GEN

dir

instr

...

+

12. For more or less modified systems of case features for Russian, see Jakobson (1958), Neidle (1988) and Franks (1995). I do not follow the sophisticated case system for Russian proposed by Franks (1995). It takes the accusative as the most unmarked case and does not allow the fulfilment of the markedness constraint (29). But it is interesting in itself. It, too, widely respects the case distinctions of Jakobson's case classification and is able to characterize case syncretisms of Russian as feature coincidence. 13. I adopt the German names of the case correlations from Jakobson (1936). The English translations are as in Franks (1995). The semantic characterizations are taken from Jakobson (1958). 14. Russian has two special, very restricted cases: the genitive2 with the case suffix -u in declension class I (as in caska caju 'a cup of tea') and the locative2 with the accented linker -u in declension class I (v lesu 'in the wood') and with the accented linker -i in declension class III (v stepi 'in the steppe'). For details, see Isaôenko (1962).

288

Ilse Zimmermann

It is essential to keep in mind that Wunderlich and his colleagues do not assume a special system of features for the morphosyntactic case realizations of DPs. In their system of case linkers, the semantico-syntactic features [±hr] and [±lr] play a role along with further differentiations of various origin. Characteristically, the additional features do not serve the purpose of cross-classification. This is in sharp contrast to the system proposed in (27). Whereas the features [±hr] and [±lr] correspond to the semantic hierarchy of complements and allow the characterization of the nominative, accusative, dative and the ergative, the genitive needs further specifications in order to be distinguished from the accusative. I included GEN in (28) (cf. Wunderlich 2000) as a provisional feature for an adequate 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 accusative of verb complements, it seems questionable to have the specification [+hr] (or [+R] of the system in (27)) for the genitive. The additional features [dir], [instr] and others in (28) would characterize so-called semantic cases in richer case systems, as, for instance, Hungarian. In sum, I cannot imagine how such an enlarged system of case features could deliver homogeneous criteria for cross-classification and generalizations. It seems more enlightening to keep morphosyntax 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 on the semantic decomposition of functor expressions and on 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 & 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 (27), all cases - except for the nominative as the most unmarked case - are identifiable by positively specified features. Thus, the markedness constraint (29) for lexical representations of morphemes, which has been proposed by Stiebels (2000a,b), can be observed. (29) *[-]: Avoid the negative specification of a morphosyntactic feature in the lexical representation of morphemes.

Structural cases in Russian

289

Furthermore, at least some syncretisms can be characterized more or less economically. In (30), some examples are given. The syncretisms in (30a)-(c) are systematic in my system because they involve natural classes; additional features may be regarded as neutralised. The syncretisms in (30d) and (e) are unsystematic because they can only be accounted for by disjunctive feature specifications. (30) Some case syncretisms a. NOM=ACC: [-obi]; neutralisation of [R] Examples: step' 'steppe' (ΙΠ, -pi) okno 'window' (I, +neutr) roman 'novel' (I, -anim) b. GEN=DAT=INSTR=LOC: [+obl]; neutralisation of [R], [P] and [U] Examples: sta 'hundred' novoj 'new' [+fem] c.

GEN=LOC: [+U]; neutralisation of [P] 15

Examples: d.

GEN=DAT=LOC: [ + R , + P ] ν [ + U ]

Example: e.

tech 'those' [+pl] novych 'new' [+pl] stepi 'steppe'(m, -pi)

GEN=ACC: [ + R , - Ρ ] ν [ + U , - P ]

Example:

mal'äka 'boy' (I, +anim)

For further case syncretisms in Russian, see Franks (1995). The examples given in (30a)-(c) illustrate the possibility of characterizing syncretisms as feature neutralizations. The conditions are more or less complex and demand principled limitations. I must leave it to further investigation which case syncretisms are systematic and 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. Stiebeis 2000a, Franks 1995). It should be clear from the foregoing assumptions about the system of morphosyntactic case features for Russian that nouns, adjectives, pronouns and numerals can enter the syntax underspecified for case distinctions. Their combination with other caseinflected entities can add case information. The same is true for the combination of DPs with governors requiring specific cases of their complements. Agreement of nouns, adjectives, numerals and determiners is thus a syntactic means of additional case specification, as is the fulfilment of a governor's case requirements by its combination with its complements.

15. More precisely, adjectives, some numerals and pronouns show case syncretism in the genitive and locative plural in the context [-animate] and the syncretism GEN=ACC=LOC in the context [+animate]. This fact raises the general question whether the syncretisms GEN=ACC and NOM=ACC are genuine phenomena of feature neutralisation (see Franks 1995: 59, fn. 44).

290

4.2

Ilse

Zimmermann

Analysis of argument realisations in terms of Correspondence Theory

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 morphosyntactic case features. For structural cases, it is necessary to interrelate the abstract case features [±hr] and [±lr] in the argument structure of functor expressions with the morphosyntactic 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, genitivel and instrumental: (31) Correspondence constraints for structural arguments a. cORR(ahr) = a R b. coRR(alr) = a P The constraints take into account the semantic background of the morphosyntactic case features [±R] and [±P] and the far-reaching correspondence of the feature systems (27) and (28). Note that the correspondence constraints presuppose the existence of a correspondent; in this respect, they function like the iDENT-constraints in Standard Correspondence Theory. Other principles of argument realization are borrowed from LDG's system of constraints and adapted to my assumptions on morphosyntactic case features. (32) 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 in the input. f. DEFAULT: One structural argument must be realized in a default case, the nominative for verbs, the genitivel for nouns as respective governors. 16 g. UNIQUENESS: The argument expressions must be realized in different morphosyntactic cases. h. *[+F]: A positively specified morphosyntactic case feature must be avoided. i. *[+R,-obl]/[aV,+N]: The accusative for complements of nouns and adjectives must be avoided (abbreviated as *ACC/N). 16. Both the nominative and the genitive as default cases are underspecified for the features [R] and [P], according to the system (27). 17. The constraint *[+F] is a simplification. It must be differentiated for the particular morphosyntactic case features with specific ranking.

Structural cases in Russian

291

j. *[-R,-obl]/[-V,+N]: The nominative for complements of nouns must be avoided (abbreviated as *NOM/N). These principles and the correspondence constraints in (31) constitute a system of faithfulness, expressivity and economy constraints, which are ranked and determine the degree of acceptability of case co-occurrences for argument expressions of lexical governors. Concerning the faithfulness constraints, it is essential to understand the possible input-output relations of features in my system of assumptions on structural cases (cf. Stiebeis 2000a: 62). In (33), F is a feature variable for hr and Ir. C is a variable for morphosyntactic case features, u represents underspecification, F/C pairs are assumed to underlie faithfulness constraints such as DEP and MAX and the correspondence constraints in (31). The introduction of features to an underspecified input constitutes a DEP violation. If F and its correspondent have conflicting feature values, this yields a violation of the respective CORR constraint. (33) Input-output relations of case features input output DEP(C) MAX(+F) CORR(aF) * u +C * u -C u u +F +C lie +F -C * +F u * +C -F -F -C u -F I assume the following constraint ranking (cf. Wunderlich 2000): (34) Constraint ranking MAX(lex), MAX(+hr,+lr) »

DEFAULT, UNIQUE, *ACC/N, * N 0 M / N

»

CORR(ahr), CORR(alr), MAX(+hr)

»

MAX(+lr), DEP(F), * [ + F ]

The tableaux in (35)-(38) show the optimal candidates for znat' 'know'/znanie 'knowledge' (14), soobäät' 'inform'Isoobitenie 'information' (15), izmenit' 'betray'/izmena 'betrayal' (16) and for prizemit'(sja) 'land'/prizemlenie 'landing' (22). The input of the candidates in the tableau representations consists of the syntactic category of the governor and its non-referential argument positions with abstract case features and - if necessary - with idiosyncratic case requirements on the pertinent argument expressions. Output of the representations are arrays of exclusively positively specified morphosyntactic case characterizations abbreviated by traditional case names (see table (27)). The syntactic configurations of the corresponding DP complements are

292

Ilse Zimmermann

assumed to be given in the respective non-derived 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. (35) shows the evaluation of the linking patterns of a transitive verb and its derived nominal. With verbs, DEFAULT is violated if the linking pattern does not include NOM, with nomináis, it is violated if the linking pattern does not include GENI. Since NOM is excluded for nouns, INSTR becomes the optimal linker for the agent argument because it makes [+lr] visible by means of its [+P] specification. GENI, which is less faithful for the internal argument than ACC, is preferred in nomináis due to high ranking of DEFAULT. Recall that ACC is specified as [+R], DAT as [+R,+P,+obl], GENI as [+U, +obl], INSTR as [+P,+obl] and NOM as [ ].

DEP(F)

*

es- ACC NOM *!

NOM NOM GENi NOM NOM GENi *!

E* jf *

*

*

*

*!

*

**

*!

*

**

**

*

*

***

*

****

*!

DAT NOM ACC DAT

MAX(+lr)

MAX(+hr)

CORR(alr)

O *κ

CORR(ahr)

*ACC/N

χ

UNIQUE

y

DEFAULT

(35) Evaluation of case patterns of 2-place verbs (verbs vs. nomináis) a. Input: Àyf+hr-lr] Àx[-hr,+lr]; [+V,-N] (znaf 'know')

* *

**

*

***

ACC INSTR *!

*

***

NOM INSTR

*

**

***

****

NOM DAT

GENi INSTR *!

*!

*! *

293

Structural cases in Russian

*!

*

1

GENi NOM

*!

NOM

*!

*

*

*

ACC NOM

INSTR *1

*

**

**

*

**

**

*

***

*

INSTR * !

**

**

***

*

***

****!*

*

***

****

*!

***

*****

*

***

****

*!

ACC

GENi GENj

*

INSTR GENi

* *

GENi DAT INSTR

*

* *

DAT

β * GENi

E" *

* *

GENi

MAX(+lr)

*

MAX(+hr)

*

o *ζ

DEP(F)

ACC NOM

1 υ