Advances in the Theory of the Lexicon 9783110197815, 9783110190199

Citation preview

Advances in the Theory of the Lexicon

≥

Interface Explorations 13

Editors

Artemis Alexiadou T. Alan Hall

Mouton de Gruyter Berlin · New York

Advances in the Theory of the Lexicon edited by

Dieter Wunderlich

Mouton de Gruyter Berlin · New York

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

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

Library of Congress Cataloging-in-Publication Data Advances in the theory of the lexicon / edited by Dieter Wunderlich. p. cm. ⫺ (Interface explorations ; 13) Includes bibliographical references and index. ISBN-13: 978-3-11-019019-9 (cloth : alk. paper) ISBN-10: 3-11-019019-2 (cloth : alk. paper) 1. Lexicology. I. Wunderlich, Dieter. P326.A385 2006 895.1135⫺dc22 2006027965

ISBN-13: 978-3-11-019019-9 ISBN-10: 3-11-019019-2 ISSN 1861-4167 Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available in the Internet at http://dnb.d-nb.de. ” Copyright 2006 by Walter de Gruyter GmbH & Co. KG, 10785 Berlin All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Cover design: Christopher Schneider, Berlin. Printed in Germany.

Contents

Introduction: What the theory of the lexicon is about. . . . . . . . . . . . . . . . . Dieter Wunderlich

1

Concepts of the lexicon in theoretical linguistics. . . . . . . . . . . . . . . . . . . . . Martin Neef and Heinz Vater

27

Towards a structural typology of verb classes . . . . . . . . . . . . . . . . . . . . . . . Dieter Wunderlich

57

From rags to riches. Nominal linking in contrast to verbal linking . . . . 167 Barbara Stiebels A theory of lexical event structure and its cognitive motivation . . . . . . . 235 Stefan Engelberg Ontology and diachrony of polysynthesis . . . . . . . . . . . . . . . . . . . . . . . . . . . 287 Johanna Mattissen Linguistic perspectives on morphological processing . . . . . . . . . . . . . . . . 355 Harald Clahsen The representation of inflectional morphology in the mental lexicon. An overview on psycho- and neurolinguistic methods and results . . . . . 389 Martina Penke Inheritance-based models of the lexicon . . . . . . . . . . . . . . . . . . . . . . . . . . . . 429 James Kilbury, Wiebke Petersen and Christof Rumpf Appendix 1: List of subprojects in the SFB 282 . . . . . . . . . . . . . . . . . . . . . Appendix 2: List of major publications from SFB 282. . . . . . . . . . . . . . . . Subject index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Language index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

479 481 495 501

Introduction: What the study of the lexicon is about Dieter Wunderlich

For twelve years during the period 1991 to 2002 the Sonderforschungsbereich (special research cluster) SFB 282 ‘The theory of the lexicon’ was hosted by the Heinrich-Heine-Universität Düsseldorf (in collaboration with the Bergische Universität Wuppertal and the Universität zu Köln), and was sponsored by the German Science Foundation (DFG). Within more than 20 subprojects, some over the whole period and some of a shorter duration, various kinds of research emphasizing the lexicon was done, following different perspectives, based on phonological, morphological, or semantic backgrounds, and pursuing theoretical and typological studies, computer simulations as well as psycholinguistic and neurolinguistic experimentation. In the appendix, a list is given of the individual projects and of some of the major publications that grew out of the larger project. Among many other activities, the SFB 282 organized three international conferences (1995 Lexical Structures, 1998 Lexicon in Focus, 2001 The Lexicon in Linguistic Theory), and the papers of the second were published in Stiebels & Wunderlich (2000). It is a pleasure for me to present a further volume in which some of the more recent work of the research project is surveyed. Before I characterize the individual contributions, I will first in retrospect present some of the ideas that made a more comprehensive study of the lexicon in the early nineties worthwhile, and also add some still relevant questions of a more general nature.

Under the lexicon we understand a component of a language in the sense of an I-language (Chomsky 1995: 17), that is, an individual, internal and intentional capacity of human beings that enables them to speak and to understand. The most simple assumption about any language is that it has a lexicon and a grammar. In the lexicon all actual ‘words’ are stored, pairing sounds and meanings (or gestures and meanings), whereas the grammar is a computational system that characterizes which sequences of ‘words’ can constitute meaningful ‘utterances’. The last 50 years of linguistics can be described as an ongoing debate on what an adequate theory of the grammar should be, and despite the fact that each of the proposed theories delegated certain information to the lexicon, there was relatively little debate about it. In a way, the conception of the lexicon remained constant as the inevitable prerequisite of any language, although individual researchers vacillated

2 Dieter Wunderlich between viewing the lexicon as a passive storage of atomic items (Selkirk 1982; Baker 1988) and viewing it as a more creative component of language (Pustejovsky 1995). In the development of the main stream theory, concepts such as rules and constructions, levels of representations, constraints on rules, underlying principles and parameters, as well as economy (minimalist) principles worked in shifts – nothing comparable is found for the lexicon. Besides the main stream, however, certain theories were developed that emphasized a more lexical perspective: Categorial Grammar – in particular Montague Grammar (Dowty 1979), Lexical Functional Grammar (Bresnan 1982, 2001), Head-driven Phrase Structure Grammar (Pollard & Sag 1987, 1994). It should also be mentioned that in the late eighties a strong tendency towards favouring the lexicon as a structure-giving reservoir emerged: concepts such as the Lexicalist Approach (di Sciullo & Williams 1987), Lexical Phonology (Kiparsky 1982, 1985), and Lexical Learning (Pinker 1984) come to mind. It is these concepts with which our research project has embarked. To put it in a nutshell: The lexicon is considered the nucleus of a language; each item, bearing phonological, semantic and categorial properties, conforms to certain principles of grammar, and the collection of all those items is highly structured (partly due to those principles); by its distributed information the lexicon widely determines the working of the components of the grammar. If, for example, propositional attitude verbs such as believe or speech verbs such as say are subcategorized for a clausal complement, recursivity is already built into the lexicon (you said that he believes that…). Pinker also points out that the rules of a grammar are acquired by learning lexical items that instantiate them. In planning a common enterprise towards the study of the lexicon, the basic question was, somewhat simplifying: What type of lexical information can properly trigger computations of the grammar, and how must lexical information be organized in order to minimize the components of a grammar? Such a question indeed presupposes the lexicon as a highly connected and at the same time minimized potential for structure formation rather than a mere classified inventory. The original proposal presented three domains of research and also focussed certain issues and ideas for clarifying their potential range: the structure of lexical items, the architecture of the lexicon, and the role of the lexicon in the grammar. Let me briefly address these three issues.

Introduction: What the study of the lexicon is about

3

The structure of lexical entries Lexical items combine phonological information (PHON) with semanticconceptual information (SEM), and are indexed for their role in the grammar through categorial information (CAT). That just the bundle of properties shown in (1) characterizes a lexical item is an idiosyncratic fact of English, in principle, at least the values of PHON and SEM could be different. For this reason one speaks of an arbitrary mapping between SEM and PHON. The activity of eating is described in most languages by a transitive verb; it is this categorial information that allows the speakers to make use of the lexical item within utterances. However, the relationship between meaning and category is not always as straightforward as in example (1). For example, Fongbe, a Kwa language of Benin, coins ‘eat’ as ù nú ‘eat thing’, ‘help’ as dó àl ‘put/give hand’, and ‘escape’ as dó zín ‘put/give invisible’ (Lefebvre & Brousseau 2002: 236, 248). (1)

eat

PHON: SEM: CAT:

/i:t/ x devours some food y transitive verb

All idiosyncratic linguistic knowledge is assumed to be encoded in the lexicon. Lexical items with a single, noncomposable meaning, however, can be of different complexity: on the one hand, they can be minimal units, roots or affixes, while on the other hand, they can be formed morphologically or syntactically, like idioms and fixed phrases that have a nontransparent meaning. Regarding the more complex lexical objects, the question arises by which general structural principles they are constrained and how their internal structure has to be represented. For instance, kick the bucket ‘die’ simultaneously is a verb phrase (qua its own structure) and an intransitive verb (qua its idiomatic meaning). One of the central issues of the lexicon concerns the specification of lexical items according to their category. Which categories and subcategories have to be distinguished, and which further properties such as semantic type and argument structure could follow from such information? A first major distinction divides the vocabulary into lexical or major categories (open sets of content words), including noun, verb, adjective, and adposition, and functional categories (closed sets of function words); each of them has to be represented by a certain general feature structure (Chomsky 1970; Jackendoff 1977; Bresnan 1982; Wunderlich 1996c; Baker

4 Dieter Wunderlich 2003). Elements of the major categories are predicates, but the categories differ in that their inventories relate to different conceptual values and dimensions. Nouns prototypically relate to objects, verbs to situations, adjectives to properties, and adpositions to certain constellations. Consequently, the elements of the major categories have different types of argument structure and are subjected to different conditions of referential anchoring. Nouns can be used referentially or predicatively, but verbs only referentially, and their referential variable must be checked by tense or mood. Finally, adjectives and adpositions are referentially anchored only by participating in a nominal or verbal structure; consequently, they are usually used as modifiers. By contrast, functional categories are related to the major categories in that they specifically serve to realize their grammatical potential. Their function is twofold: (i) they bind the referential variables conceptually attributed to the major categories, and (ii) they realize the syntactic potential of the major categories by higher syntactic projections. Articles and quantifiers are specific for nouns, while tense, aspect and mood are specific for verbs. Dependent on the type of language, the inventory of a particular functional category (say, past, or the definiteness operation) can comprise independent functional words, can consist of phonological operations or affixes, or can be empty. In the latter case, the specific function must be taken over by the context. Another point concerns the way in which lexical information is represented. It should best be recognized as underspecified, that is, redundancyfree: everything that can be derived on the basis of given information does not need to be represented explicitly (Bierwisch 1996). To comply with this economy requirement one needs a special technology of feature representation and feature unification. Moreover, featural composition as well as the semantic composition of lexical items should be incremental, that is, information might be added but not deleted in the course of composition. This does not exclude that certain kinds of information can be saturated, or neutralized, or even be detached in the derivation. (2)

a. Saturation of argument requirement in synthetic compounds gamekeeper, nutcracker, letter opener b. Neutralization of gender in the German plural feminine: die Katze ‘the cat’, die Angestellte ‘the employee’ masculine: der Affe ‘the monkey’, der Angestellte ‘the employee’ plural: die Katze-n, die Affe-n, die Angestellte-n

Introduction: What the study of the lexicon is about

5

c. Detachment of agent in anticausatives (Haspelmath 1993) Japanese: hirak-u ‘open’(tr.) – hirak-e-ru ‘open’(itr.) Maltese: feta ‘open’(tr.) – n-feta ‘open’(itr.) Moreover, on each compositional level certain defaults are possible, which can be overridden by the more specific information of lexical items. An example is shown in (3). (3)

Case assignment in German prepositions (Bierwisch 1996) a. Default: stative prepositions assign dative (bei ‘at’, in ‘in’), b. Specific rule: dynamic prepositions assign accusative (in ‘into’), c. Lexical: zu ‘to’, a dynamic preposition, assigns dative.

Finally, lexical items are the first meaningful elements that children identify when they acquire a language; therefore, they should not bear information that one could not detect on the basis of just a few positive instances. Hence, features related to inflectional classes or rule application are rather questionable. Instead, one should try to replace those features by other, independently required features of PHON, SEM, or CAT (such as theme vowel, animacy, or gender in nominal inflection classes). The architecture of the lexicon The lexicon can be seen as both locally and globally structured, and consequently, it can be extended creatively in manifold ways, following the structure of individual items or groups of items and thus forming local clusters per analogy, or following the more general categorial traits that make up the global structure of the lexicon. In a way, the lexical units can be considered to be nodes of a ‘network’ that spreads the features used in characterizing lexical information. Subnets have the structure of hierarchies or inheritance trees, type lattices, and so on (see Kilbury et al., this volume). In each of these subnets, certain nodes can be more salient or more frequently instantiated than others, and thus give rise to additional items clustering with them, which in turn can enforce further generalization. Thus, the lexicon simultaneously gets enriched and becomes more generalized in the course of time. This is particularly true of the lexicon of a language learner, but also holds for the lexicon of an adult speaker and as well for the lexicon that is shared by speakers in the historical perspective (notwithstanding the fact that certain entries get lost).

6 Dieter Wunderlich In a rather abstract sense, generalizations in the lexicon can be understood as feature structures superimposed on the more idiosyncratic structures of individual items through unification. For instance, German feminines ending in schwa take -n in the plural (Tanne, Tannen ‘fir’), while masculines ending in schwa moreover take -n in all singular cases (Affe, Affen ‘monkey’). English nouns denoting container objects (such as bottle) can be used as denominal verbs with the reading ‘put something in it’ (bottle the wine ‘put the wine in bottles’). Nouns denoting fruits can also be used for the tree on which these fruits grow (he ate a peach, he planted a peach). Verbs denoting transition events can be used both causatively and inchoatively (he bleached his shirts, his shirts bleached). Verbs such as refuse and nouns such as refusing and refusal are lexically related: they have the same argument structure and the same selectional restrictions (a fact that prompted Chomsky 1970 to his lexicalist hypothesis). And so on. Generalized feature structures of the lexicon have been observed by various researchers; they have been called macros, schemes, templates, lexical rules, among others, and both monotonic and non-monotonic aspects have been attributed to them. The idea of a global architecture of the lexicon can be further elaborated along the two complementary forces of lexicalization and grammaticalization (Lang 1989). Lexicalization means that lexical items spell out the nodes of an abstract cognitive space which is formed by a finite set of conceptual dimensions relevant for human behavior (SEM), while grammaticalization means that lexical items specify the featural nodes of an abstract linguistic space which is formed by the relevant dimensions inherent to the grammatical categories (CAT). If communication means constitute a relevant dimension, then verbs such as to telegraph, to fax, to email come into being. If gender is a relevant category, than all nouns are assigned a gender value. The major lexical categories can each be characterized by a specific cognitive and grammatical potential. Relevant for nouns are referential dimensions (such as individuation parameters, ontological sorts, and gestalt characteristics), relevant for verbs are cognitive schemes for the temporal classification of states and events, and relevant for adjectives are perceptual dimensions. Therefore, each of these categories has its own structural profile regarding lexicalization and grammaticalization. Also, the kind of variation possible for lexical items is category-specific: nouns mainly vary regarding sorts (peach: fruit or tree), verbs regarding event types (bleach: causative or inchoative), and adjectives regarding perceptual dimensions (wide: internal

Introduction: What the study of the lexicon is about

7

or external distance, according to the opposites wide – narrow/tight vs. wide – near/close). The possible values in the cognitive and grammatical dimensions can be distributed differently, so that both systematic and arbitrary lexical gaps can arise in a language, as well as corresponding margins for lexicon extension. Likewise, these values can be differently distributed cross-linguistically, so that languages widely differ in how fine-graded particular lexical nets of them are. Memorized phrasal units or items that bear cliticized elements can also be reanalysed lexically and thus let new lexical regularities come into existence. Being predicators, the items of the lexical categories have a certain number of arguments to be realized by morphological means or syntactic complements. The argument structure can be represented by means of thematic roles (such as agent, patient, goal) and, more abstractly, by means of proto-roles (Dowty 1991) or certain parameters of semantic decomposition (Jackendoff 1990; Bierwisch 1989; Wunderlich 1997). Principles of argument linking determine how these roles correspond to syntactic subcategorizations and case assignment. Verbs, which project a clause and therefore are in the center of grammar, have the most complex argument structure, which also interacts with event structure, and further show additional complexities such as inherent (lexical) case. Moreover, verbs together with a complement often form an idiomatic structure (such as kick the bucket). Thus, the verbal ‘net’ inherent to the brain must have a highly complex organization. Another aspect of the architecture of the lexicon involves the fact that lexical items are of different morphological complexity; they can be roots, affixes, compounds, or reduplicated, derived, inflected forms, in various ways. Usually, only some of the more complex items have predictable properties, while others have idiosyncratic properties, for instance, a noncompositional semantic reading. Therefore, part of the architecture of the lexicon is the amount of structure introduced by morphology. However, it is still a matter of debate whether morphology constitutes a separate module of grammar which is nearer to the lexicon than syntax (hence, presyntactic) (which is the traditional assumption) or even part of the lexicon (because it deals with word forms), or just a subcomponent of syntax in that it deals with word syntax and is mainly controlled by syntactic principles (Lieber 1992). A variant of the latter view is Distributed Morphology (Halle & Marantz 1993, 1994), which claims that morphology spells out

8 Dieter Wunderlich certain features of derived syntactic representations and is therefore postsyntactic. One can also find the position of split-morphology, assuming that derivational morphology belongs to the lexicon, while inflectional morphology, at least insofar as it deals with case and agreement, belongs to syntax (Anderson 1982, 1992). Regardless of how one sees the proper place of morphology, one has to deal with the phenomenon of lexical integrity (Di Sciullo & Williams 1987): complex words mostly behave as atoms at the level of phrasal syntax and phrasal semantics, that is, the structure of words usually is not ‘visible’ in the syntax, except perhaps with regard to certain inflectional features. The morphological complexity levels of the lexicon could be characterized by categories such as root, stem, word, by level features such as ±minimal, ±maximal, or by prosodic categories such as syllable, foot, phonological word. It is not yet clear whether one of these options is to be preferred, or whether they should be taken as co-existent, inducing different views on the lexicon, or different ways in which the lexicon works crosslinguistically. It seems that the Semitic languages need different optimality evaluations on the stem level than on the word level (Kiparsky to appear). Other languages have a [+maximal] affix, which closes up a possible word, and no further affix can be attached. Many languages show more than one type of reduplication, differing according to whether a syllable, a foot or a phonological word gets reduplicated. The minimal morphological units (roots, affixes) are often assumed to be underspecified regarding their category or meaning; they are then subject to certain principles that regulate morphological specification. Given underspecification on the one hand, and certain principles of specification on the other, it is not necessary that the combination of morphological items is controlled by explicit rules. Rather it could be assumed to be free in principle and only restricted by the principles of specification: only those items can be combined that yield a combination for which a categorial value can be found (Wunderlich 1996a,b). Under this view, the morphological complexity levels would follow inherently and need not be given explicitly. Morphological forms based on a common root, in particular inflected forms, are often ordered in paradigms, according to the grammatical values they realize. It is again an open question whether these paradigms are real objects, subjected to specific paradigm structure rules (stem and paradigm approach: Aronoff 1994, Stump 2001; word and paradigm approach: Robins 1959, Matthews 1991, Blevins 2004) and optimization principles (Carstairs

Introduction: What the study of the lexicon is about

9

1983, Carstairs-McCarthy 1994), or rather an epiphenomenon resulting from the information inherent to a number of related forms (Bobaljik 2002). A final distinction between actual, potential and virtual words has to be made. Given the possibilities of creating new words, no limits can be given for the set of actual words. Accidental gaps can be filled at some time (for instance through borrowing) and therefore can be viewed as nodes at which a word potentially could appear. Moreover, actual words can be morphologically complex and semantically transparent, so that they do not belong to the idiosyncratic knowledge, but if they are frequently used they can nevertheless be memorized and thus belong to the ‘mental lexicon’. If morphology (vis-à-vis syntax) is taken as a generative part of the lexicon, it is clear that it produces many virtual words, even combinations that as such never surface but get integrated into more complex items; and some of these virtual words can become actual if they acquire some idiosyncratic aspects. For an agglutinating language such as Turkish, however, it would not be reasonable to consider all possible morphological combinations as entities that could be stored, even if they make up virtual words (Hankamer 1989). It is in particular the inflecting type of morphology that has impact on the global structure of the lexicon. The role of the lexicon in the grammar Lexical items are combined into syntactic phrases on the basis of their categorial information. Possible indicators for the distinction of lexical vs. phrasal unit are the amount of idiosyncrasy (lexical elements are more idiosyncratic), integrity (syntactic rules usually cannot affect parts of a complex lexical item), the degree of productivity (phrases are more productive), the prosodic realization (lexical elements usually only have one accent), the type of memorization (lexical elements are stored), and the effects on language processing (lexical elements, at least if they are frequently used, are faster accessible). The information of lexical items is in fact a structure-building potential: arguments must be realized, functional categories must find an element on which they can function, adjectives and prepositional phrases have to share their open argument with an expression on which they can operate as a modifier, and so on. Lexical items project phrases according to their category and also license complements in virtue of their subcategorization (the formal correspondent to argument structure), so that the distinction between

10 Dieter Wunderlich intransitive, transitive and ditransitive verbs is accounted for in the syntax. There is nothing in the grammar of English that requires eat to have a complement (‘to eat something’), but if the lexical item requires it, the grammar decides that the complement is just to the right of eat. More precisely, principles of grammar determine phrasal structure (and also how much it can be disrupted by extraction), as well as prosodic structure and logical form of utterances. Studies on linguistic contacts have established that lexical borrowing also can induce new syntactic regularities. For example, Acadian French (spoken on Prince Edward Island) has borrowed particle verbs from English (ender up, finder out), generalized this construction to native verbs (parler about ‘talk about’, aller on ‘go on’, faire up ‘make up’), and finally allowed prepositions to be stranded (Où ce-qu’elle vient de? ‘where does she come from?’), which is ungrammatical in Standard French (King 2000). Phrasal information is mainly combinatoric (on the basis of what is contributed by the lexical items) and only adds requirements for linear ordering, as well as pragmatic cues (such as topic and focus) associated with alternative orders. In contrast, lexical information can be seen as integrative insofar as it abstracts over several sound-meaning occurrences in a threefold way, namely with respect to PHON, SEM, and CAT. That makes it possible to use the lexical item as a triggering element in all three modes of processing: the phonological, semantic, and syntactic processing. Thus, the lexicon is responsible for what Jackendoff (2002) calls the tripartite or parallel architecture of grammar. For the sake of discussion, let us assume that both simple and morphologically complex items belong to the domain WORD, while phrasal units belong to the domain SENTENCE. That brings up the question: To what extent are WORD items subject to the principles of grammar, and in what way do phrasal units depend on lexical information? These issues have been and still are debated controversially. However, some clarifications can be made. Typologically, linguistic information is quite differently distributed between the two domains WORD and SENTENCE. Not only do some languages nearly lack morphology, while others have very rich morphology, even if two languages have the same amount and the same type of morphology, it is often the case that words of one language are translated into phrases of the other, and vice versa. Moreover, many isomorphisms exist between word structure and phrasal structure, both in function and in structure. Elements of compounds such as woodcutter function like attributes (wooden house), derivational elements

Introduction: What the study of the lexicon is about

11

such as the suffixal verbs of Inuit (4a) function like verbs with clausal complements, and inflectional elements such as Latin -v and -isse in (4b) function like auxiliaries in English. (4b) also shows that the hierarchy of functional categories affects the ordering of inflectional affixes as well as word ordering. Arguments can be realized in the process of compounding/derivation or by syntactic complementation, as shown by synthetic compounds and their phrasal correspondents in (4c). From these observations one can infer that the principles of grammar are neither WORD- nor SENTENCE-specific, although cross-linguistically the relevant constructions are differently distributed between the two domains. (4)

a. Suffixal verbs in Inuit: paasi-sur-a-ra understand-think-TRANS-1sg/3sg ‘I think that he understands it’ b. The order of morphemes in the English phrase is reverse of that in the Latin word: fle-v-isse-s verb – aspect – mood – person cry-PERF-CONJ-2sg ‘you would have cried’ person – mood – aspect – verb c. Argument-saturation in morphology or in syntax: spaghetti-eat-er = someone who eats spaghetti

Despite many similarities, there exists one main difference between word structure and phrasal structure. Phrasal elements can be moved (displaced) or can be co-indexed by means of agreement, while elements of a word structure are positionally fixed and do not overtly agree with each other. Moreover, discourse functions such as topic and focus cannot be marked in the word, while they can be marked, often by means of displacement, within phrasal structure. Why did the morphology-syntax duality arise at all? Carstairs-McCarthy (2005) argues that they emerged under different circumstances. Rapid speech production causes phonological alternations in lexical items, and the results might be reinterpreted as encoding certain variations of a concept, such as singular or plural (Engl. foot-feet), imperfective or perfective (Engl. run-ran). “Once a pattern of relationship of this kind […] had got established, […] there would be scope to analyse in the same way relationships involving the meaningful presence versus absence of edgemost segments and segment-

12 Dieter Wunderlich clusters (proto-‘affixes’)” (2005: 179). Hence, modulatory (nonconcatenative) morphology preceded additive (affixal) morphology in language evolution. According to this view, morphology originated from allomorphy, but different circumstances were necessary to trigger the emergence of meaningful displacements (i.e. proper syntax). Through the emergence of the Minimalist Program (MP, Chomsky 1995) the role of the lexicon in the grammar has been strengthened, so that the two mainstreams of linguistic theorizing (from the view of the syntax and from the view of the lexicon) have the chance to be united. Most of the syntactic concepts proposed in earlier work have been given up in the MP. What remains are several kinds of features instantiated by the items to be merged (where ‘merge’ is a notion to describe phrasal combination at minimal expense). These items can best be thought of as lexical items; only some features such as those relating to discourse functions have to be introduced freely because lexical items usually do not have any information about these functions. By the so-called external merge (Chomsky 2005) argument requirements of the combined items are either saturated or inherited to the result, and the information of modifiers is checked for whether it is compatible with what they modify. The function of external merge is thus similar to operations of Categorial Grammar. In contrast, so-called internal merge produces a copy and thus licences dislocation. Nothing in the lexical items themselves favours dislocation; one therefore needs additional information such as ‘Realize this item at the left edge’, which would lead to such a move. That lexical items can determine the result of external merge must be due to their categorial information, so that the study of syntax and that of the lexicon have to complement each other, which, regrettably, is not a point of view very much deliberated in MP. Nevertheless, I would like to claim that categorial information provided by the lexicon is structure-building information. That lexical items usually cannot determine whether internal merge (that is, displacement) takes place only indicates that internal merge depends on some extra structure superimposed on lexical structure. Another point that has been strengthened by MP is the view that language interrelates two interface systems, namely phonological form (PF), interfacing with the articulatory-perceptual system, and logical form (LF), interfacing with the conceptual-intentional system. This corresponds to the traditional view that not only lexical items but also the more complex phrases built from them pair sound and meaning. It is stated in the MP that the mappings between sound and meaning are optimized on the phrasal

Introduction: What the study of the lexicon is about

13

level. In fact, to a large extent this optimization is already done in the lexicon, and it is mainly due to the additional features on which internal merge operates that further optimization is needed. This issue also raises the question of how the interface structures must be organized in order to allow for the grammar to optimize the relation between sound and meaning. For answering this question certain studies concerning lexical matters are necessary. (i) Optimality-theoretic (OT) phonology (Kager 1999; McCarthy 2004) can show that the optimal word form (concerning features such as syllable structure, stress-assignment, vowel harmony, allomorph selection) results from an evaluation that is based on a certain ranking of highly natural constraints. That suggests that OT concepts are suitable means for framing the complex relationship between grammatical form and PF. Similarly, the computation of inflectional forms (even if on the first view these forms seem rather uneconomic) can be minimized by using OT concepts (Wunderlich 2001). (ii) On the other side, (dynamic) event semantics (Rothstein 1998; Higginbotham et al. 2000; Naumann 2000, 2001) can shed light on the complex relationship between complement structure, aspectual composition and event structure of a verb. These studies suggest that, among other things, a more finelygraded aspectual theory would be able to bridge the gap between ordinary semantic representations (LF) and extralinguistic conceptual structure. The interdisciplinary lexicon The structure of lexical items triggers the tripartite architecture of grammar. That makes it necessary to involve at least the three core disciplines phonology, semantics, and morpho-syntax in the study of the lexicon. Moreover, since the lexicon constitutes the nucleus of a language, further issues require participation of lexicography, computer science, language typology and history, language acquisition research, psychology, and neurology. Lexicography aims at a complete inventarization of a certain lexicon fragment in a lexical database, which presupposes various technical decisions concerning levels, categories, and network relations, which in turn produce both local and global structures in the database, as well as corresponding accessing routes. It is important to test whether these decisions meet the intended theoretic standards. At the same time, the pragmatic experience of lexicography can encourage the researcher of the lexicon.

14 Dieter Wunderlich Computational linguistics tries to formalize the general structures of the lexicon and to develop and to implement algorithmic procedures; it thus sharpens the concepts of underspecification, monotony, and unification of feature structures. It also develops techniques for the automatic recognition of lexical information from texts, so that in the end also questions about lexical extension can be answered: How can a new word be identified and integrated into a given lexicon on the basis of partial contextual information? Since languages vary most widely in their lexicons, it is a common assumption that the comparison of lexicons also constitutes a tool of measuring the distance between languages, in particular, the point at which these languages separated from each other historically. Several methods have been proposed but remain controversial for absolute dating: glottochronology, based on measuring the percentage in which languages differ with respect to base vocabulary (Swadesh 1952, 1955, 1972; Gray & Atkinson 2003), mass comparison, based on the statistical comparison of whole lexicons examining them for words which appear similar phonologically and have a similar meaning (Greenberg 1966, 1987, 2002, Ruhlen 1994), and the comparison of slowly-changing features of grammatical and lexical structure, such as pronouns, plural markers, noun classes, and case markers (Nichols 1992). In any case, the lexicon seems to be the most individuating property of a language. In a theory of the lexicon one tries to abstract over many purely arbitrary (historically emerged) facts, for instance, that German die Sonne, der Mond contrast with French le soleil, la lune with reversed membership in the gender classes. More importantly, the French verb inventory emphasizes the expression of path of motion (monter, traverser, longer, tourner, entrer), whereas German emphasizes the expression of manner of motion (steigen, gehen, fahren, laufen, schreiten, reiten) (Talmy 1980). Even more important is the fact that some languages nearly lack the category of adjectives (Dixon 1977), or that other languages exhibit a category of classifiers, and so on. Finally, distinctions between isolating, agglutinating, inflectional, polysynthetic etc. languages, which take into account the amount and the type of morphology, of course have counterparts in the way the lexicon is organized. Linguistic typology has to pay attention to all these parameters: the choice of categories, the degree of categorial specification, the profile in particular domains of lexicalization, and the amount of morphology. Linguistic theories often claim to be guided by economy considerations. The proof for economy, however, must be provided by real situations. Two

Introduction: What the study of the lexicon is about

15

such situations play an important role for the linguist: How early are the postulated structures learned, and how fast are they processed? Studies on the acquisition of the lexicon can show in what sequences the postulated lexical features are acquired, that is, generalized on the basis of a given input, and they can also show in which domains overgeneralizations appear and, by contrast, which features are correctly acquired from the start on. If the child does not make mistakes in a certain domain, this means that the respective features are conditioned by previously acquired ones. For instance, a child learning German overgeneralizes accusative in dative contexts, but never overgeneralizes dative, so that dative proves to be a subset of accusative and is acquired second (Eisenbeiß 2002). Psycholinguistic studies investigate the processing time of simple as well as morphologically complex lexical items under various conditions either inherent to the items (e.g., frequency) or contextually given by means of priming; these studies involve both recognition (e.g., lexical decision) and production (e.g., naming). Neurolinguistic studies investigate the corresponding activation patterns of the brain by means of event-related potentials (ERP). They also try to find out in what respects people suffering from brain damage or brain diseases differ from non-handicapped persons under processing conditions. The ultimate goal of all of these experiments is to make certain inferences about linguistic representations. For instance, is a certain type of morphologically complex word memorized or not? Here, one of the underlying assumptions is that the accessing time for items that are memorized depends on their frequency, while that of non-memorized items depends on complexity. Given the family of linguistic disciplines dealing with important aspects of the lexicon, the question remains of what makes the theory of the lexicon unique. Maybe one can say that it aims at investigating the structures of the lexicon in general, its levels, categories, templates, and relational networks, as well as the principles of lexicalization. Important contributions to our understanding of the lexicon The researchers of the Sonderforschungsbereich developed a number of promising new approaches. Dynamic Event Semantics is based on the idea that verb meanings are programs for characterizing events (Naumann 2001). The multidimensional valency theory explicates the various parameters of valency in their interaction with proto-roles and event structure

16 Dieter Wunderlich (Engelberg, this volume). Lexical Decomposition Grammar accounts for the relationship between argument hierarchy (according to the rank of arguments in a semantic decomposition) and argument realization, including canonical patterns, alternations, and splits, in a typological perspective (Wunderlich, this volume; Stiebels, this volume). Minimalist Morphology applies minimalist ideas to inflection (Wunderlich & Fabri 1995), and also studies the set of constraints that determine optimal inflectional forms (Wunderlich 2001). Other work elaborated on existing or newly-developed theoretical paradigms within the domain of phonology, morpho-phonology, and prosody with respect to German (Wiese 1996; Neef 1996) as well as other languages (Walther 1999; Krämer 2003), explored the nature of clitics (Gerlach 2002), the syntax of word formation (Siebert 1999), and so on. It is not possible to document all of this work here. Further studies investigated the typological variation of the noun-verb distinction and established a taxonomy of polysynthesis (Mattissen, this volume). Kilbury et al. (this volume) developed tools for the representation and identification of lexical structures. Clahsen and his group (this volume), as well as Penke (this volume), contributed empirical data in favor of the dual processing hypothesis, which says that regular and irregular morphology are differently encoded and processed in the brain. Future studies on the lexicon I will conclude this overview by pointing to a few issues that were not in the center of the research project and have not been studied extensively, but certainly will play a role in future studies of the lexicon. One of the topics concerns linguistic relativity. Are the basic lexical categories, despite their typological variation, associated with universal concepts, or can the concepts also be framed by the existing categories? Usually one assumes that all human beings are equipped with the same conceptual capacity, so that they conceive of space, causality, temporal order, individuation, numbering, and quantification in the same way. Consequently, even if the various types of language coin these concepts in different ways and to different degrees, the speakers’ way of thinking should remain unaffected. However, Levinson and his collaborators (Levinson 1997, 2003; Majid et al 2004) showed that spatial orientation depends on whether speakers have relative terms such as left/right or only absolute

Introduction: What the study of the lexicon is about

17

terms such as north/south at their disposal. Similarly, Everett (2005) argues that the Pirahã people, living in the Amazonian area, not only have a structurally restricted language, lacking number words, quantifiers, and embedding, but also a restricted cognition, in particular, number cognition, which has been confirmed by formal experimental work (Gordon 2004). If that state of affairs would be more than cultural deprivation of a tribe living isolated, the case opens the necessity of studying the lexicon vis-à-vis certain cognitive capacities in more detail. Another promising issue would be lexicon simulation, aiming at studying the effects of self-organization in a large and still growing inventory of memorized items. At what size of a cluster of related items does it become useful for a language learner to find a generalization that transforms a flat storage into a more hierarchical storage, allowing a more economical management? Some researchers believe that categories are automatically established if the mental lexicon increases beyond a critical size. Thus, the question arises: How much of the lexicon gets self-organized if it is growing, and how much external syntax is necessary to induce the organization of the lexicon? Processes of this kind could be studied under various conditions, involving size, accessing time, inherent similarities, external triggers, and so on. Simulation experiments could also shed new light on the problem of polysemy. It is not only content words that show variation in meaning, so that contextual information must determine the specific reading, but also function words and even affixes: German als covers ‘as, than, when, while’; Spanish reflexive se covers reflexive, causative-reflexive, decausative and passive; Korean -ita covers causative and passive; Yidiny (Australian) -a covers causative and applicative – note that these are not at all rare examples. Ambiguity between readings that have contradictory effects in argument structure of course differs from conceptual specification or conceptional shift investigated by Bierwisch (1983), and from the different types of coercion discussed by Pustejovsky (1995). The general question is: Why is it more effective for so many languages to use one multifunctional morpheme, requiring contextual inferences for disambiguation or specification, rather than a small number of monofunctional morphemes that do not require additional inferences? The advantages of multifunctional vs. monofunctional morphemes vis-à-vis their respective costs require study. Finally, simulations could also shed new light on questions of evolution. Most researchers agree that the evolution of language started with a lexicon of undifferentiated, communicatively useful ‘words’. Lexical growth forces

18 Dieter Wunderlich partition, so that a certain (probably semantically based) classification of words can arise. On that basis one can combine words according to their class membership, which in turn could give rise to a rudimentary syntax. Of course, various questions arise with such a scenario. One is that the evolution of the lexicon is assumed to precede the evolution of syntax, which could be true but does not need to be. Why do all languages exhibit verbs and nouns and not some other basic semantic classification? Moreover, if it is true that verbs are stored in different regions of the brain than nouns (Saffran & Sholl 1999; Shapiro et al. 2005), one has to ask how this fact could have emerged from lexical growth. Contributions to this volume The papers collected in this volume cover quite a range of aspects of the lexicon, which constitute, of course, only a subset of the possible research fields, and only a subset of the work actually done in the Sonderforschungsbereich. Martin Neef and Heinz Vater (“Concepts of the lexicon in theoretical linguistics”) proceed from a remark by Bloomfield (1933), who said that “the lexicon is [ ] an appendix of the grammar, a list of basic irregularities”, and then discuss in what respects subsequent researchers either basically hold the same view or freed themselves from such a limitation. Particularly interesting is the fact that one of the authors propagates a declarative model of linguistic knowledge, which essentially returns to Bloomfield’s conception of the lexicon. Two papers use the framework of Lexical Decomposition Grammar, developed in Düsseldorf. Dieter Wunderlich (“Towards a structural typology of verb classes”) discusses what the universal inventory of verbs may look like, and how it is organized cross-linguistically. In accordance with traditional assumptions, he argues that the number of arguments is the major structural factor for verbs, and that transitive verbs have to be seen as the center of grammar. He continues, however, with the observation that there are more major argument linking types for transitive verbs than commonly assumed: besides the well-known (generalized) case type, the active type, the inverse type, the salience or voice type, and the positional type have to be distinguished. By contrast, the classes of intransitive verbs mostly reflect semantic factors. A major part of the article surveys the formal means by which the third argument of ditransitive verbs is realized; here, the author

Introduction: What the study of the lexicon is about

19

finds multiple ways beyond the possibilities enabled by the basic types. It seems that the study of ditransitive verbs offers a particularly interesting window into the typology of grammar. Barbara Stiebels (“From rags to riches. Nominal linking in contrast to verbal linking”) discusses the question of how limited the argument structure of nouns is. On the basis of various typological data she argues that nouns do in fact share a number of properties with verbs, which makes them ‘richer’ than one is inclined to think on the basis of limited observation: nouns often allow to introduce a higher argument (by ‘agent voice’) or a lower argument (by possessor extension), and can exhibit some kind of ‘nominal passive’ as well as possessor reduction and possessor-split. Furthermore, she adopts the view that the voice-type argument-encoding system of Austronesian languages is derived from nominalization markers. Stiebels explains the major linking types in nominalizations on the basis of the principle of strict successive argument saturation, which, among others, predicts that if a deverbal noun exhibits both a nominal and a verbal subdomain of arguments the latter must be limited to the lower arguments. Stefan Engelberg (“A theory of lexical event structure and its cognitive motivation”) postulates that the semantic representation of a verb has to get its justification from semantic facts such as selectional restrictions, interlexematic relations, inference behavior, and the ability to allow certain aspect, voice and mood categories rather than from syntactic data. He states that a lexical semantic representation SEM is about aspects of the world, and so must be described in such a way that one knows what SEM is supposed to convey about the world; consequently, he sketches a theory of lexical event structure to capture SEM of a verb. This theory is then supplemented with a linking model and further motivated by both cognitive and neurological observations. Johanna Mattissen (“Ontology and diachrony of polysynthesis”) discusses the various conceptual categories (such as localities, body parts, classifiers, medium, and manner) expressed by affixes (‘non-root bound morphemes’) in complex verb forms, thereby constituting the phenomenon of polysynthesis, which is found in a number of smaller languages around the world. Several subtypes are possible without clear correlations: independently of the serializing or incorporating character of the language, local affixes are the strongest category, but while some languages have an impressive set of local affixes, Greenlandic has none. There are languages that have noun incorporation and verb-root serialization but lack any non-root bound morphemes, and also languages with both compositional mechanisms and non-

20 Dieter Wunderlich root bound morphemes, which nevertheless cannot properly be called polysynthetic. The author further discusses possible evolutionary paths that may have lead to polysynthesis of one or the other type in recent times, and finds that again Greenlandic is an exception, maybe because its polysynthesis is quite old. Two papers are related to experimental psycholinguistic research. Harald Clahsen (“Linguistic perspectives on morphological processing”) points to the correspondence hypothesis, which states that the language processor constructs mental representations using the normal structures and operations of the grammar. On the basis of various experimental findings he shows how morphological factors such as stem formation type (composition vs. suppletion), affixation type (inflection vs. derivation), frequency (word forms vs. affixes), and specificity of inflectional affixes lead to characteristic processing effects (priming, acceleration), and are thus in line with theoretical reasoning. Martina Penke (“The representation of inflectional morphology in the mental lexicon. An overview on psycho- and neurolinguistic methods and results”) argues, too, that experimental data can be used in testing and evaluating linguistic theories. She discusses frequency effects, priming effects, and ungrammaticality effects to a greater detail, thereby including also data from Broca’s aphasia, language acquisition, and ERP measurements, and establishes evidence for lexical underspecification of affixes and for a fundamental difference between regular and irregular inflected forms. Finally, James Kilbury, Wiebke Petersen, and Christof Rumpf (“Inheritance-based models of the lexicon”) aim to integrate some independent developments in computational linguistics. First, they combine the HPSG strategy for avoiding redundancy in the representation of lexical knowledge through the use of inheritance hierarchies (Pollard and Sag 1987) with the DATR strategy for describing linguistic generalizations involving subregularities and exceptions by using nonmonotonic means (Gazdar 1987); they present their formalism QType, which preserves the efficiency of monotonic unification for parsetime processing while capturing the expressiveness of nonmonotonicity by restricting it to inheritance within type signatures. Second, they combine the strategy for the automatic construction of a lexicon from large bodies of data by using inductive procedures (Boguraev and Pustejovsky 1996) with the framework of Formal Concept Analysis (Ganter and Wille 1999) in order to develop techniques for the automatic induction of type signatures, which they illustrate with the hierarchical representation of the inflectional classes of German nouns.

Introduction: What the study of the lexicon is about

21

References Anderson, Stephen R. 1982 Where’s morphology? Linguistic Inquiry 13: 571–612. 1992 A-morphous Morphology. Cambridge University Press. Aronoff, Mark 1994 Morphology by Itself. Cambridge, MA: MIT Press. Baker, Mark C. 1988 Incorporation. A Theory of Grammatical Function Changing. Chicago University Press. 2003 Lexical Categories. Verbs, Nouns, and Adjectives. Cambridge University Press. Bierwisch, Manfred 1983 Semantische und konzeptuelle Repräsentation lexikalischer Einheiten. In Untersuchungen zur Semantik, R. R i ka and W. Motsch (eds.), 61–99. Berlin: Akademie Verlag. 1989 Thematische Rollen und Dekomposition. Seminar Papers. Third DGfS Summerschool, Hamburg. 1996 Lexical information from a minimalist point of view. In The Role of Economy Principles in Linguistic Theory, C. Wilder, H.-M. Gärtner and M. Bierwisch (eds.), 227–266. Berlin: Akademie Verlag. Blevins, James 2004 Inflection classes and economy. In Explorations in Nominal Inflection, G. Müller, L. Gunkel and G. Zifonun (eds.), 51–95. Berlin: Mouton de Gruyter. Bobaljik, Jonathan 2002 Syncretism without paradigms. Yearbook of Morphology 2001: 53– 85. Boguraev, Branimir and James Pustejovsky (eds.) 1996 Corpus Processing for Lexical Acquisition. Cambridge, MA: MIT Press. Bresnan, Joan 1982 Control and complementation. In The Mental Representation of Grammatical Relations, J. Bresnan (ed.), 282–390. Cambridge, MA: MIT Press. 2001 Lexical-functional Syntax. Oxford: Blackwell. Carstairs, Andrew 1983 Paradigm economy. Journal of Linguistics 19: 115–125. Carstairs-McCarthy, Andrew 1994 Inflection classes, gender, and the principle of contrast. Language 70: 737–787. 2005 The evolutionary origin of morphology. In Language Origins. Perspectives on Evolution, M. Tallerman (ed.), 166–184. Oxford University Press.

22 Dieter Wunderlich Chomsky, Noam 1970 Remarks on nominalization. In Readings in English Transformational Grammar, R. A. Jacobs and P. S. Rosenbaum (eds.), 184–221. Waltham, MA: Ginn. 1995 The Minimalist Program. Cambridge, MA: MIT Press. 2005 On phases. Ms., MIT. Di Sciullo, Anna-Maria and Edwin Williams 1987 On the Definition of Word. Cambridge, MA: MIT Press. Dixon, Robert M. W. 1977 Where have all the adjectives gone? Studies in Language 1: 19–80. Dowty, David D. 1979 Word Meaning and Montague Grammar. Dordrecht: Reidel. 1991 Thematic proto-roles and argument selection. Language 67: 547–619. Eisenbeiß, Sonja 2002 Merkmalsgesteuerter Grammatikerwerb: Eine Untersuchung zum Erwerb der Struktur und Flexion von Nominalphrasen. PhD diss. Düsseldorf. Everett, Daniel 2005 Cultural constraints on grammar and cognition in Pirahã: Another look at the design features of human language. Current Anthropology 46. Ganter, Bernhard and Rudolf Wille 1999 Formal Concept Analysis. Mathematical Foundations. Berlin: Springer. Gazdar, Gerald 1987 Linguistic applications of default inheritance mechanisms. In Linguistic Theory and Computer Applications, P. Whitelock (ed.), 37–67. London: Academic Press. Gerlach, Birgit 2002 Clitics Between Syntax and Lexicon. Amsterdam: John Benjamins. Gordon, Peter 2004 Numerical cognition without words: Evidence from Amazonia. Science 306: 496–499. Gray, Russell D. and Quentin D. Atkinson 2003 Language-tree divergence times support the Anatolian theory of Indo-European origin. Nature 426. Greenberg, Joseph H. 1966 The Languages of Africa. Bloomington: Indiana University. 1987 Language in the Americas. Stanford University Press. 2002 Indo-European and its Closest Relatives: the Eurasiatic Language Family – Volume II, Lexicon. Stanford University Press. Halle, Morris and Alec Marantz 1993 Distributed morphology and the pieces of inflection. In The View from Building 20, K. Hale and S. J. Keyser (eds), 111–176. Cambridge, MA: MIT Press.

Introduction: What the study of the lexicon is about 1994

23

Some key features of distributed morphology. MIT Working Papers in Linguistics 21: 275–288. Hankamer, Jorge 1989 Lexical representation and process. In Lexical Representation and Process, W. Marslen-Wilson (ed.), 392–408. Cambridge, MA: MIT Press. Higginbotham, James, Fabio Pianesi and Achille C. Varzi (eds.) 2000 Speaking of Events. Oxford University Press. Jackendoff, Ray 1977 X-bar Syntax: A Study of Phrase Structure. Cambridge, MA: MIT Press. 1990 Semantic structures. Cambridge, MA: MIT Press. 2002 Foundations of Language. Brain, Meaning, Grammar, Evolution. Oxford University Press. Kager, René 1999 Optimality Theory. Cambridge University Press. King, Ruth 2000 The Lexical Basis of Grammatical Borrowing: A Prince Edward Island Case Study. Amsterdam: Benjamins. Kiparsky, Paul 1982 Lexical morphology and phonology. In Linguistics in the Morning Calm, Linguistic Society of Korea (ed.), 3–91. Seoul: Hanshin. 1985 Some consequences of lexical phonology. Phonology Yearbook 2: 83–138. to appear Paradigm Effects and Opacity. Stanford: CSLI. Krämer, Martin 2003 Vowel Harmony and Correspondence Theory. Berlin / New York: Mouton de Gruyter. Lang, Ewald 1989 The semantics of dimensional designation of spatial objects. In Dimensional Adjectives: Grammatical Structure and Conceptual Interpretation, M. Bierwisch and E. Lang (eds.), 263–417. Berlin: Springer. Lefebvre, Claire and Anne-Marie Brousseau 2002 A Grammar of Fongbe. Berlin /New York: Mouton de Gruyter. Levinson, Stephen C. 1997 Language and cognition: the cognitive consequences of spatial description in Guugu Yimithirr. Journal of Linguistic Anthropology 7: 98–131. 2003 Space in Language and Cognition: Explorations in Cognitive Diversity. Cambridge University Press. Lieber, Rochelle 1992 Deconstructing Morphology: Word Formation in Syntactic Theory. Chicago University Press.

24 Dieter Wunderlich Majid, Asifa, Melissa Bowerman, Sotaro Kita, Daniel B. M. Haun and Stephen Levinson 2004 Can language restructure cognition? The case for space. Trends in Cognitive Sciences 8: 108–114. Matthews, Peter 1991 Morphology. Cambridge University Press. McCarthy, John (ed.) 2004 Optimality Theory in Phonology – a Reader. Oxford: Blackwell. Naumann, Ralf 2000 A dynamic temporal logic for aspectual phenomena in natural language. In Advances in Temporal Logic, H. Barringer et al. (eds.), 223–254. Dordrecht: Kluwer 2001 Aspects of changes – a dynamic event semantics. Journal of Semantics 18: 27–81. Neef, Martin 1996 Wortdesign. Eine deklarative Analyse der deutschen Verbflexion. Tübingen: Stauffenburg. Nichols, Johanna 1992 Linguistic Diversity in Space and Time. Chicago University Press. Pinker, Steven 1984 Language Learnability and Language Development. Cambridge, MA: Harvard University Press. Pollard, Carl and Ivan A. Sag 1987 Information-based Syntax and Semantics, Vol. 1: Fundamentals. Stanford: CSLI. 1994 Head-driven phrase structure grammar. Chicago University Press. Pustejowsky, James 1995 The generative lexicon. Cambridge, MA: MIT Press. Robins, Robert H. 1959 In defense of WP. Word 10: 210–234. Rothstein, Susan (ed.) 1998 Events and Grammar. Dordrecht: Kluwer. Ruhlen, Merritt 1994 The Origin of Language: Tracing the Evolution of the Mother Tongue. New York: Wiley. Saffran, Eleanor M. and Alexandra Sholl 1999 Clues to the functional and neural architecture of word meaning. In The Neurocognition of Language, C. M. Brown and P. Hagoort (eds.), 241–272. Oxford University Press. Selkirk, Elisabeth O. 1982 The syntax of words. Cambridge, MA: MIT Press.

Introduction: What the study of the lexicon is about

25

Shapiro, Kevin A., Felix M. Mottaghy, Niels O. Schiller, Thorsten D. Poeppel, Michael O. Fluss, Hans-Wilhelm Müller, Alfonso Caramazza and Bernd J. Krause 2005 Dissociating neural correlates for nouns and verbs. Neuroimage 24: 1058–1067. Siebert, Susann 1999 Wortbildung und Grammatik. Syntaktische Restriktionen in der Struktur komplexer Wörter. Tübingen: Niemeyer. Stiebels, Barbara and Dieter Wunderlich (eds) 2000 Lexicon in Focus. Berlin: Akademie Verlag. Stump, Gregory 2001 Inflectional Morphology. Cambridge University Press. Swadesh, Morris 1952 Lexicostatistic dating of prehistoric ethnic contacts. Proceedings American Philosophical Society 96: 452–463. 1955 Towards greater accuracy in lexicostatistic dating. International Journal of American Linguistics 21: 121–137. 1972 What is glottochronology? In The Origin and Diversification of Languages, M. Swadesh, 271–284. London: Routledge and Kegan Paul. Talmy, Leonard 1980 Lexicalization patterns: semantic structure in lexical forms. In Language Typology and Syntactic Description, Vol. 3: Grammatical Categories and the Lexicon, T. Shopen et al. (eds.), 57–149. Cambridge University Press. Walther, Markus 1999 Deklarative prosodische Morphologie: Constraint-basierte Analysen und Computermodelle zum Finnischen und Tigrinya. Tübingen: Niemeyer. Wiese, Richard 1996 The Phonology of German. Oxford University Press. Wunderlich, Dieter 1996a A minimalist model of inflectional morphology. In The Role of Economy Principles in Linguistic Theory, C. Wilder, H.-M. Gärtner and M. Bierwisch (eds.), 267–298. Berlin: Akademie Verlag. 1996 b Minimalist morphology: The role of paradigms. Yearbook of Morphology 1995: 93–114. 1996 c Lexical categories. Theoretical Linguistics 22: 1–48. 1997 Cause and the structure of verbs. Linguistic Inquiry 28: 27–68. 2001 How gaps and substitutions can become optimal: An OT account of argument linking in Yimas. Transactions of the Philological Society 99: 315–366. Wunderlich, Dieter and Ray Fabri 1995 Minimalist Morphology: An approach to inflection. Zeitschrift für Sprachwissenschaft 14: 236–294.




Concepts of the lexicon in theoretical linguistics Martin Neef and Heinz Vater

1. Introductory remarks Any research project needs a name. The research complex ‘Theory of the Lexicon’ opted to make the word lexicon the center of its name. The word lexicon, however, has a wide range of meanings. This diversity stems to a certain extent from the difference between the use of words in colloquial language as opposed to scientific language. Another source of the unstable characteristics of the word lexicon is the fact that words usually change their meanings in the course of time and especially in the development of scientific theories. In order to yield a better understanding of the name ‘Theory of the Lexicon’, we will start with a broad overview of the diversity of the possible meanings of the word lexicon by citing a number of entries in relevant dictionaries and handbooks. In doing so, we distinguish between the linguistic term lexicon as a word of the English language and the notion 1 LEXICON this term refers to (cf. Vater 1999). The major part of our paper will consist of an overview of the history of the word lexicon in the linguistics of the 20th century with special emphasis on structural and generative linguistics. In this context, we will concentrate on the relation of the lexicon to grammar and deal only marginally with other key questions of the research complex like the inner architecture of the lexicon and the structure of lexical entries. Since the research complex ‘Theory of the Lexicon’ was implemented in 1991, we will focus especially on the understanding of the word lexicon during that period of the development of linguistic theory. 2

———–———————————————————————————— 1

2

In accordance with Jackendoff (1983) and other authors, we represent concepts in capitals, thus e.g. making a difference between the word house and the concept HOUSE. We thank Dieter Wunderlich for his help and advice he gave us through many years in his function as the speaker of the research complex ‘Theory of the Lexicon’ and as a friend.

28 Martin Neef and Heinz Vater The term lexicon is ambiguous. Most English dictionaries mention two uses of the word, one referring to the inventory of words of a language, the other one referring to a handbook (also called dictionary): 3 lexicon /leksik
n/ noun 1 the vocabulary of a language, individual, or subject. 2 a dictionary, esp. of Greek, Latin, or Hebrew [late Greek lexikon, neuter of lexikos of words, from Greek lexis word, speech, from legein to say].

(Allen 2000: 803) The ‘vocabulary’-reading is sometimes split up into two readings referring a) to the totality of words in a speech community, b) to the totality of words controlled by an individual speaker: lexicon /'leksik
n…/ n 1 the lexicon technical all the words and phrases used in a language or that a particular person knows. 2 a book containing an alphabetic list of words with their meanings.

(Summers 1995: 812)

These two (or three) readings are also listed by most of the dictionaries of linguistics, where they are complemented by additional readings, especially ones specifying the use of the word lexicon in different approaches to linguistic theory. Such a detailed definition is given by Crystal (1992): Lexicon The vocabulary of a language (technically, its lexical items or lexemes), especially when these are listed in a dictionary as a set of lexical entries; also called lexis. The study of a language’s lexicon is lexicology. A network of semantically related lexical items, such as the words for colour or fruit, is a lexical field (or semantic field). In generative grammar, the lexicon is the component containing all the information about the structural properties of lexical items. Lexical syntax is an approach which incorporates syntactic rules within the lexicon; lexical phonology is an approach where some of the phonological rules are transferred to the lexicon. In psycholinguistics, the stored mental representation of what people know about their language is called the mental lexicon, and the study of the psychology of word meanings is sometimes called psycholexicology.

(Crystal 1992: 227–228) ———–————————————————————————————

3

Most of the German dictionaries we checked restrict their definitions and explications to the ‘dictionary’-reading. This is e.g. true of Klappenbach & Steinitz (1977: 3.2361) and Pfeifer et al. (1989: 1011). The distinction under discussion is made in Duden (1994: 819f.).

Concepts of the lexicon in theoretical linguistics

29

This definition is basically in agreement with the definition by Bußmann (1996): Lexicon [Late Grk lexicón (sc. biblión) ‘book of or for words’]. 1 An alphabetically or semantically ordered list of words for a language, dialect, or sociolect, or a list of terminology for a specific discipline. Such lists are generally compiled as reference works ( also lexicography, vocabulary). 2 In its most general sense the level of description which codifies the morphol. and semant. aspects (i.e. the forms and meanings) of the vocabulary of a lang. which cannot be derived from the regularities of the ling. system. 3 In transformational grammar, one of the basic components of grammar in the form of a subordinated list of all lexical formatives. The lex. entry consists of a phonetic-phonolog. descr. in the form of a matrix of distinct. features to which a selection of specific syntactic features is correlated ( complex symbol) (see Chomsky 1965). 4 In generative semantics, the lex. is composed of syntactically structured complexes of the smallest semantic building blocks ( semant. primitive) to which corresponding phonol. realizations are assigned ( also lexicalist vs. transformationalist hypothesis).

Bußmann (1996: 280) Although both definitions in the linguistic handbooks are quite specific (e.g. in specifying the use of the word lexicon in different linguistic schools), it will become obvious later on that they are not quite satisfactory in one point: It has to be assumed that not only words are stored in a lexicon, but also idioms (cf. to the point, ups and downs, break the ice etc.), possibly also morphemes (like learn- and -ed ) (cf. e.g. Schwarze & Wunderlich 1985: 9). Handke (1995) tries to subsume the different readings of lexicon under the common denominator of word-store (cf. fig. 1). According to Handke (1995: 49), we have to distinguish between storage for the purpose of information access (in a dictionary as a reference book or in the computer) and storage as part of the human language processing system (in the memory): 4 ———–———————————————————————————— 4

The diagram is not totally in accordance with the division Handke (1995: 49) proposes: “Word-stores that are primarily consulted for the reason of information retrieval are referred to as dictionaries. By contrast, word-stores that constitute a component within a natural language processing system are called lexicon.” According to this definition we would not expect a ‘Machine-Readable Lexicon’. The lexicon in the theoretical sense (which we treat in 2.) has been neglected altogether.

30 Martin Neef and Heinz Vater Word-Store

Dictionary

Book Dictionary

MachineReadable Dictionary

Lexicon

MachineReadable Lexicon

Mental Lexicon

Figure 1. Word-Store (modified version of Handke 1995: 49, fig. 18)

Although this word-store interpretation as a common denominator presents a progress, it has to be blamed for the same deficiency as in the case of the two former definitions: the restriction of lexical entries to words. Besides, in all the definitions quoted above – except Crystal (1992) – we can observe the effect of the ‘psycholinguistic fallacy’ (a term coined by Chesterman 1980), i.e. the blending of the lexicon conceived of as the product of linguistic description with the lexicon as a store of entities in the brain of an individual speaker. Chesterman (1980: 19) holds that Chomsky himself has contributed to this fallacy by postulating a mental grammar (as the description of the competence of a speaker-hearer), which then led to confounding the formal grammar with the psychological grammar. The term grammar as understood by Chomsky (1965) is systematically ambiguous, referring to the result of the linguist’s activity as well as to the child’s (unconscious) activity in creating a grammar on the basis of given data, when it acquires a language. Actually, Chomsky wanted to delimit the theoretical frame for competence rather than explain the real production and reception of sentences (which would be a matter of performance). But many linguists thought that the grammatical rule mechanism was in direct correspondence to psychological reality and took a transformational grammar as a performance grammar.5 But until now nobody has furnished convincing ———–————————————————————————————

5

Recently, Pullum (1996: 142) criticized Chomsky for taking “grammars to exist physically in the human brain”.

Concepts of the lexicon in theoretical linguistics

31

evidence for a direct connection between the results of linguistic descriptions and psychological activities. Chesterman’s statement applies especially to the description of the lexicon. Meibauer (2002: 16) holds that one should not expect a 1:1 correspondence between conceptions of the linguist’s lexicon and the mental lexicon, although these conceptions are not altogether independent of each other. However, theoretical conceptions should be compatible with psycholinguistic findings. Consequently, we will only be concerned with the notion of the theoretical lexicon in the following. 2.

Short history of the concept of the lexicon in linguistic theory

2.1. The lexicon in Structural Linguistics The founding father of Structural Linguistics, Ferdinand de Saussure, discusses in his (posthumously edited) lectures the lexicon only in a marginal way. He emphasizes that morphology (regarded mainly as the system of the words’ inflection) cannot be separated from syntax, because a group of inflectional items like the different forms of the Greek word phulax (‘guard’) becomes a paradigm only if we compare the functions attached to these forms (de Saussure 1916: 186): “Une déclinaison n’est ni une liste de formes ni une série d’abstractions logiques, mais une combinaison de ces deux choses: formes et fonctions sont solidaires, et il est difficile, pour ne pas dire impossible, de les séparer.” [‘A declination is neither a list of forms nor a series of logical abstractions but a combination of both: forms and functions are in complete solidarity; it is difficult if not impossible to separate them’].

For this reason, lexicology cannot be excluded from grammar, since many relationships can be expressed by words as well as by grammatical means; cf. considérer (‘to consider’) with the synonymous idiomatic expression prendre en considération. Saussure concludes that the traditional division into morphology, syntax and lexicology might have a practical value but that grammar can only be constructed on the basis of a different higher principle (‘la grammaire ne peut s’édifier que sur un principe différent et supérieur’).

32 Martin Neef and Heinz Vater Bloomfield (1933: 138) is more specific concerning the lexicon. He holds that the lexicon forms a part of semantics, along with grammar.6 He claims that the lexicon is an arsenal of irregularities: “The form-classes of English words are largely arbitrary: there is nothing to tell us that man, boy, lad, son, father are male nouns, that run, bother are verbs, that sad, red, green are adjectives, and so on. In particular, of course, the form-class of every morpheme is arbitrarily determined. A complete description of language will list every form whose function is not determined either by structure or by a marker; it will include, accordingly, a lexicon, or list of morphemes, which indicates the form-class of each morpheme, as well as lists of all complex forms whose function is in any way irregular.”

(Bloomfield 1933: 269) And he even concludes (in a very apodictic way): “The lexicon is really an appendix of the grammar, a list of basic irregularities” (Bloomfield 1933: 274). The first of these quotations shows, at the same time, that Bloomfield (1933) considers morphemes rather than words to be the relevant entries of the lexicon. In this respect, Structural Linguistics differs considerably from traditional approaches to linguistics (cf. Aronoff 1994: 18).7 The relevance of Bloomfield’s view of the lexicon in later Structural Linguistics is evidenced by the following quote from Hockett (1958: 148): “We realize that morphemes, rather than words, are the elementary building-blocks of language in its grammatical aspect, though this shift of emphasis in no sense implies that words are unimportant.” The structuralist lexicon, thus, is the domain of linguistic irregularities while the grammar of a language encompasses the regular aspects. Contrary to the definitions of the term lexicon given in ch. 1, the structuralist lexicon is not equivalent to the vocabulary of a language. This is because the regular aspects of the vocabulary do not belong to the lexicon but to grammar, more exactly to morphology as the system of regularities of word structure. Nevertheless, there is a close connection between the lexicon and the unit word that at least suggests treating the lexicon as equivalent to the vocabu———–————————————————————————————

6

7

“When the phonology of a language has been established, there remains the task of telling what meanings are attached to the several phonetic forms. This phase of the description is semantics. It is ordinarily divided into two parts, grammar and lexicon“. Aronoff (1994: 8f.) traces this development back to the work of Baudouin de Courtenay.

Concepts of the lexicon in theoretical linguistics

33

lary. Monomorphemic words are always treated as elements of the lexicon given their character as signs in the Saussurean sense. Since there is an arbitrary connection between form and meaning in simple words, they cannot be analyzed morphologically. Only complex words may be the subject of morphology. Consequently, the lexicon has to contain all root morphemes of a language plus all other linguistic irregularities that may concern both complex words and syntactic units like phrases and sentences (in the case of idiomatic expressions). The structuralist notion of the lexicon survived Structuralist Linguistics. We will show in the next paragraphs that this conception of the lexicon was relevant for generative linguistics up to 1970 and was revived in the Minimalist Program of Generative Grammar in the early 1990s.8 However, there have been some generative linguists that adhered to the structuralist conception in the interim. To give just one characteristic example: “If conceived of as the set of listemes, the lexicon is incredibly boring by its very nature. It contains objects of no single specifiable type […] and those objects that it does contain are there because they fail to conform to interesting laws. The lexicon is like a prison – it contains only the lawless, and the only thing that its inmates have in common is lawlessness.[…] Another related view that we reject is the idea that the lexicon has structure. As mentioned, it is simply a collection of the lawless, and there neither can nor should be a theory directly about it, for it can only be understood in terms of the laws it fails to obey.”

(Di Sciullo & Williams 1987: 3–4) Thus, a structuralist lexicon does not allow a theory of it. A structuralist lexicon is the counterpart of grammar lacking significant inner organization (which does not imply that lexical entries are unstructured). Obviously, a theory of the lexicon has to be a theory about a lexicon in a different understanding.

———–————————————————————————————

8

The recent reconstruction of the lexicon in terms of compositionality and reverse compositionality by Fodor (2002) obviously also relates to the structuralist lexicon.

34 Martin Neef and Heinz Vater 2.2. The lexicon in Transformational Generative Grammar There is some confusion concerning the defining characteristics of a generative grammar (cf. Botha 1989: 1–11). According to Chomsky (1965: 4; 1986: 3; 1991: 162), a grammar as a description of a particular language is a generative grammar if it is perfectly explicit. Thus, whether or not a grammar has the epithet ‘generative’ is not a question of linguistic content but of metascientific considerations. Many linguists, however – including Chomsky (1986: 4–5), although in an inconsistent way – reserve the term ‘generative’ for a specific kind of an explicit grammar, namely for one that is mentalistic in nature aiming at the improvement of insight into the human mind. In an even more restricted way, the notion of generative grammar is commonly used to refer to Chomsky’s own version of an explicit mentalistic grammar that has the supplementary feature of being transformational or derivational respectively. This means that a grammar consists of distinct levels of representation that are related to each other by transformational rules. Logically, however, mentalism and transformationalism are not interconnected. We restrict the meaning of Generative Grammar (in capital letters) to the mentalistic-transformational understanding of this term. This allows a better distinction between Generative and Structural Linguistics (the latter one being or not being explicit but decidedly being non-mentalistic). Declarative Linguistics, on the other hand, is definitely explicit but non-derivational. The inaugurating work of Generative Grammar, Chomsky’s Syntactic Structures (1957),9 does not deal with the notion lexicon explicitly. However, several aspects of the conception of this first model of Generative Grammar are important for the discussion of the lexicon. Within the generative paradigm, the “fundamental aim in the linguistic analysis of a language L is to separate the grammatical sequences which are the sentences of L from the ungrammatical sequences which are not sentences of L and to study the structure of the grammatical sequences”, as can be read in the initial paragraphs of Chomsky (1957: 13). A more recent formulation of the goals of the generative enterprise highlights the cognitive background and focuses on the task to model the linguistic knowledge in a native speaker’s head. While there are many differing theories conceivable in order to fulfil ———–————————————————————————————

9

This is the first work published on Generative Grammar. There is a precursor to this book, namely Chomsky (1955), whose first publication occurred as late as 1975.

Concepts of the lexicon in theoretical linguistics

35

this task, the very specific version of Syntactic Structures owes some of its fundamental properties to Chomsky’s attempt to distinguish his own approach explicitly from the prevailing scientific paradigm of his time. Thus, some of Chomsky’s decisions have more of a rhetoric force than of an indispensable theoretical rooting in the generative background. Therefore, it is not surprising that Chomsky later modified many of his initial assumptions without changing his basic generative principles. A central aim of our investigation is to add suggestions in this direction pertaining to the subject of the lexicon. Three aspects of the Syntactic Structures-model are relevant to the topic of the lexicon: First, the grammar in Syntactic Structures is a generative machine that transforms an abstract input into a less abstract output. Thus, the grammar is transformational and derivational. While the dominating paradigm of psychological research of the 50’s, behaviourism, restricts itself to the observation of the connection between stimulus and response, Chomsky is predominantly interested in specifying the mechanism deriving the output from an input. An explicit transformational machine that explains the transition from an input as a stimulus to an output as a response seems to be a promising approach for this aim. More precisely, for the speaker the meaning structure can be related to the stimulus and the sound structure to the response, while for the hearer it is the other way round. Second, the dominating linguistic school in the United States in the 50’s was the American version of structural linguistics known as Distributionalism. Among the leading figures of this school was Chomsky’s linguistic teacher Zelig S. Harris. While distributional linguistics was mainly concerned with morphology and only marginally dealt with problems of syntax, Chomsky decided to entirely neglect the grammatical relevance of a putative linguistic module of morphology. Instead, he declared morphology to be a byproduct of syntactic and phonological derivation. Ironically, for doing so he adopted Bloomfield’s view of the lexicon as a list of morphemes (which he later cautiously termed ‘formatives’). In doing so, he neglected the word as a central linguistic notion. If syntax can be devised as the grammatical module that generates all grammatical sequences of morphemes (Chomsky 1957: 32), then morphology can be dispensed with as an autonomous grammatical discipline. Consequently, most of the efforts of American Distributionalism turned out to be irrelevant for the progress in linguistics. Third, the initial stage of the grammatical derivation in Syntactic Structures is a set of phrase structure rules. The task of these rewriting rules is twofold: First, they generate abstract core-sentences by means of rules of the

36 Martin Neef and Heinz Vater type ‘Sentence  NP VP’, ending in symbols for lexical categories like N or V, and second, they insert lexical elements into the appropriate syntactic slots by rules of the type ‘V  hit, take, walk, read etc.’ or ‘Aux  C(M) (have + en) (be + ing)’ (cf. Chomsky 1957: 111). Thus, an inventory of lexical elements constitutes the output of the formation process of syntax, being itself a part of the syntax. Syntax, thus, is not a theory of sentence structure in this model but a theory of morpheme combinations, both in words and in sentences. A major step in the development of both Generative Grammar and the theory of the lexicon is Chomsky (1965). The main advantage of this retrospectively so-called Standard Model (cf. Chomsky 1972b: 5) is a clear distinction between a deep structure (derived from the notion of core sentences) and a surface structure. While the semantic interpretation is tied to the deep structure of syntactic derivation, the phonological interpretation follows the level of surface structure. Still, the grammar consists of only the three modules syntax, semantics, and phonology, lacking a morphology module. The following scheme visualizes the conception of grammar according to Chomsky (1965), in the representation of Bechert et al. (1970: 165): (1)

Standard Model of Generative Grammar (Chomsky 1965) Base (phrase structure rules + lexicon) SYNTAX SEMANTICS

deep structure transformations surface structure

meaning structures

PHONOLOGY

sound structures

In this model, there appears a grammatical component called lexicon. This lexicon replaces the lexical insertion rules of the former model:

Concepts of the lexicon in theoretical linguistics

37

“Instead, the base of the grammar will contain a lexicon, which is simply an unordered list of all lexical formatives. More precisely, the lexicon is a set of lexical entries, each lexical entry being a pair (D, C), where D is a phonological distinctive feature matrix ‘spelling’ a certain lexical formative and C is a collection of specified syntactic features.”

(Chomsky 1965: 84) Chomsky’s main motivation for this step is to simplify grammar: “Many of the grammatical properties of formatives can now be specified directly in the lexicon […] morphological properties of various kinds can be treated in this way […] Since many such properties are entirely irrelevant to the functioning of rules of the base and are, furthermore, highly idiosyncratic, the grammar can be significantly simplified if they are excluded from the rewriting rules and listed in lexical entries.”

(Chomsky 1965: 86–87) Chomsky’s conception of the relation of deep structure and semantic interpretation requires that the insertion of lexical material into abstract phrase structures takes place prior to or at the latest on the level of deep structure since only structures containing lexical elements can be interpreted semantically. This precondition, however, does not imply that the lexicon has to be a part of the syntax. It is equally well conceivable that the lexicon is a component outside syntax or even outside grammar (as in Structuralist Linguistics). Lexical information has to be derivationally available before the level of deep structure. Chomsky’s motivation for putting the lexicon inside syntax may derive from the conception of phrase structure rules in the former Syntactic Structures model. Based on this conception, it may have been natural to regard the lexicon as a part of the syntactic base although this is not a necessary conclusion. In any case, from then on the lexicon was regarded as an integral part of grammar, a fact that distinguishes Generative Grammar most strikingly from traditional and structuralist approaches to linguistics. Concerning the content of the lexicon, Chomsky explicitly adheres to the Bloomfieldian view: “In general, all properties of a formative that are essentially idiosyncratic will be specified in the lexicon.”10 (Chomsky 1965: ———–————————————————————————————

10

Formatives are defined by Chomsky (1965: 3) as ‘minimal syntactically functioning units’. The examples for lexical entries Chomsky (1965: 82–84) presents (boy, book, virtue, sincerity, frighten etc.) are words, but he admits the possibility that they might also be morphemes (Chomsky avoids the term ‘morpheme’).

38 Martin Neef and Heinz Vater 87). But, different from Bloomfield (1933), Chomsky assumes a systematic structure of lexical entries: “In particular, the lexical entry must specify; (a) aspects of phonetic structure that are not predictable by a general rule …; (b) properties relevant to the functioning of transformational rules…; (c) properties of the formative that are relevant for semantic interpretation …; lexical features indicating the positions in which a lexical formative can be inserted (by the lexical rule) in a preterminal string. In short, it contains information that is required by the phonological and semantic components of the grammar and by the transformational part of the syntactic component of the grammar, as well as information that determines the proper placement of lexical entries in sentences.”

(Chomsky 1965: 87–88) In this respect, there is a clear progress in the conception of the structure of lexical entries in the Standard Theory of Generative Grammar.11 2.3.

The lexicon under the lexicalist hypothesis

2.3.1. The lexicalist position The final chapters of Chomsky (1965) are devoted to questions of morphology. Here, Chomsky still promotes the view that both inflection and word-formation take place in the syntactic component. The meaning of complex words is determined at the level of deep structure. Syntactic transformations rearrange lexical formatives that are contained in deep structures to form complex words as elements of the surface structure. However, Chomsky realizes that derivational morphology is somewhat problematic for a syntactic treatment because of its sometimes semi-productive character. An alternative he sketches (1965: 186) consists in the immediate introduction of complex words in the lexicon.12 In his seminal paper ‘Remarks on Nominalization’ (1970a), Chomsky addresses these problems again and arrives at a fundamentally new view ———–———————————————————————————— 11

12

Botha (1968) gives a clear picture of the role of the lexicon in the Standard Model and he hints on a number of shortcomings with respect to morphological data. Earlier in the book, Chomsky (1965: 120–123) considers to even locate subcategorization rules inside the lexicon instead of among the set of rewrite rules.

Concepts of the lexicon in theoretical linguistics

39

concerning the place of morphology in grammar (without changing the overall-architecture of grammar as sketched in (2)). In order to capture the semi-productive properties of nominalizations in English, Chomsky considers two alternative theoretical positions: One option is to enrich the transformational component in order to deal with such semi-productive derivational morphology (this is called the ‘transformationalist position’ [1970a: 17]), the other is to assume that the respective complex words are already present in the base (this is called the ‘lexicalist position’ [1970a: 17]). Chomsky shows that English nominalizations cannot be systematically derived by means of transformational rules, indicating that his argumentation can be extended to concern the total of derivational morphology.13 In German, there are nominalized forms of all infinitives (cf. fahren ‘to drive’ – (das) Fahren ‘driving’), – similar to nominalizations with -ing in English – while other types of nominalization obey strong restrictions. Thus, there is no -ung-derivate *Fahrung from fahren (although Befahrung ‘driving on’ can be derived from befahren ‘to drive on’), and there is no -erderivate from stehlen ‘to steal’ (since *Stehler is blocked by the existence of Dieb ‘thief’). Syntactic explanations for these facts cannot be given. Chomsky calls his now favored view the ‘lexicalist position’ indicating that he does not want to treat complex words ‘somewhere’ in the base, but exactly in the lexicon. An alternative he does not discuss would be to assume a distinct morphological component of grammar. However, this would be a complication of the model in comparison to the option he chose, because the lexicon was already present as a component of grammar in the precursor model which did not contain morphology. Moreover, the reintroduction of morphology into grammar may have looked like a relapse into Structuralism. The new conception of the lexicon, however, makes it an original component of grammar containing rules to deal with linguistic regularities rather than a list of some kind of information. Basically, the generative lexicon is the pre-syntactic component of grammar. The lexicon, thus, developed from a mere list of individual items to a generative component in its own right, being a component that generates complex linguistic units out of smaller ones.

———–———————————————————————————— 13

Authors like Toman (1983: 13–22) have shown that Chomsky’s initial arguments for the lexicalist position are not conclusive. Hoekstra, van der Hulst & Moortgat (1980: 1) rank Chomsky’s Remarks “among the revolution-making rather than revolutionary texts.”

40 Martin Neef and Heinz Vater Chomsky (1970a) inaugurated an era of intensive study of morphology and the lexicon.14 Part of the diverse proposals concerned the question to what extent morphology was located inside the lexicon. Chomsky himself temporarily subscribed to the Extended Lexicalist Hypothesis (cf. Chomsky 1972a: 150) which states that all of derivational morphology is a lexical phenomenon while inflection is syntactic in nature. This position was later renamed the Weak Lexicalist Hypothesis to indicate the opposition to the Strong Lexicalist Hypothesis according to which all of morphology, including compounding and inflection, is handled inside the lexicon. Supporters of the Weak Lexicalist Hypothesis like Jackendoff (1975), Aronoff (1976), and Anderson (1982, 1992) as well as those of the Strong Lexicalist Hypothesis like Lieber (1981), Selkirk (1982), and Toman (1983) devised a number of different models of a lexical morphology.15 (2)

Generative model of grammar of around 1980 phrase structure rules SYNTAX SEMANTICS

deep structure

LEXICON

transformations surface structure meaning structures

PHONOLOGY

sound structures

———–———————————————————————————— 14

15

According to Hoekstra, van der Hulst & Moortgat (1980: 8), the full impact of Chomsky (1970) that replaced the overall transformational mechanism with a combination of syntactic transformations and lexical rules became fruitful as late as in the mid-seventies. Jensen & Stong-Jensen (1984) discuss the consequences of the Strong Lexicalist Hypothesis concerning the introduction of a unique morphological component into Generative Grammar with morphology being a part of the lexicon.

Concepts of the lexicon in theoretical linguistics

41

Interestingly enough, by the early eighties the lexicon as a component of Generative Grammar seemed to have left the syntactic base in order to become a distinct and autonomous component of grammar. As mentioned earlier, this move is consistent even with the derivational account of the Standard Model. Thus, it is hard to see what different empirical predictions follow from the location of the lexicon either as part of the syntactic base or as a distinct component outside syntax as long as the output of the lexicon may be inserted into syntactic deep structure representations. The scheme in (2) catches the relevant move in the conception of grammar. An interesting quote from Selkirk (1982) shows that even at that time this move went largely unnoticed: “In an earlier unpublished version of this book, I took the position that word structure rules were part of the system of base rules of the syntactic component. However, this position does not follow in any logical sense from the fact that word structure rules and phrase structure rules have the same general structural properties. It is merely consistent with that fact. Equally consistent would be a model of grammar according to which word structure rules were ‘in the lexicon’, that is, in an entirely distinct component of grammar.”

Selkirk (1982: 10) Recalling that in the Standard Model of Generative Grammar in (1) the lexicon is a part of the syntactic base, this quote shows some confusion.16 The quote makes perfect sense, however, based on a conception of grammar like in (2). In Selkirk’s word syntactic approach, the lexicon has an internal structure, comprising both a lexicon in the structuralist sense (called ‘Extended Dictionary’) and a component that generates word structures:

———–———————————————————————————— 16

The same confusion is evident in the writings of Chomsky during that period of time. For example, in Chomsky (1981: 5) the lexicon is regarded as a grammatical subsystem distinct from syntax, while some pages later (1981: 18) the syntax is said to consist “of a base which in turn consists of a categorial component and a lexicon.” Hoekstra, van der Hulst & Moortgat (1980: 11) noticed that Chomsky did not return “to the demarcation between lexicon and syntax after his introduction of the lexicalist hypothesis” during the 1970s.

42 Martin Neef and Heinz Vater (3)

Lexicon in word syntactic approaches (Selkirk 1982, in the representation of Olsen 1986: 40) LEXICON

Base Word Structure Rule: Xn  Yp Xm, 0 > n > p, m

Extended dictionary: words bound morphemes

Xn Morpholexical insertion

3

Yp

Xm

#

#







2.3.2. The lexicon in Principles-and-Parameters-Theory The model of grammar in (2), however, does not really represent the model that Chomsky assumes in his Principles and Parameters Model (also known as the Government and Binding Theory), because he realized meanwhile that the semantic interpretation just as the phonetic interpretation of syntactic structures had to be tied to the syntactic surface structure (a step already initiated in the Extended Standard Model [Chomsky 1970b: 117]), yielding the following Y-model of grammar: (4)

Model of grammar in Principles-and-Parameters-Theory (Chomsky 1981) X-bar-Theory SYNTAX

D-structure

LEXICON

move  S-structure

Logical Form

Phonetic Form

Concepts of the lexicon in theoretical linguistics

43

X-bar-Theory is the phrase structure component of the model. It replaces former rule systems of the syntactic base and generates (or defines) Deep Structures. The transformation into the Surface Structure is brought off by an extremely simplified transformational component called move  (containing only one type of transformation). The surface-structure representations are modified in such a way as to allow a semantic as well as a phonological interpretation. The Logical Form is derived from S-structure by means of a further application of move
, representing thus a further level of syntax. The relation between syntax and the lexicon as depicted in (4) consists in the fact that lexical insertion takes place at the level of DStructure. This is a viable option, but since the semantic interpretation now follows the level of S-structure, lexical insertion may also take place at this level, as some morphologists assume (e.g. Anderson 1982: 594). Other researchers have suggested that lexical insertion can be found on both levels of syntactic representation or even after S-structure (cf. the Distributed Morphology approach of Halle & Marantz 1993). Chomsky (1981: 29) attributes a prominent role to the lexicon, as witnessed by his ‘projection principle’: “Representations at each syntactic level (i.e., LF, and D- and S-structure) are projected from the lexicon, in that they observe the subcategorization properties of lexical items.” It is the lexicon where relevant syntactic properties of a word (e.g. its valency) are determined: “If persuade, for example, takes an NP object and a clausal complement as a lexical property, then an LF-representation including this verb will be well-formed only if it is assigned an NP object and a clausal complement at this level of representation” (Chomsky 1981: 29). Chomsky emphasizes that this conception of the lexicon indicates progress in comparison to the former conception of the lexicon expressed in Chomsky (1965): “Consideration of the projection principle illuminates a conceptual deficiency of theories of UG of the type considered in Chomsky (1965). In such theories, the lexicon is a set of lexical entries, each with its specific contextual features (among other properties). The categorial component of the base, quite independently, specifies a certain class of subcategorization frames in which lexical items may appear. Thus information concerning the class of subcategorization frames is in effect given twice in the grammar: once – implicitly – in the lexicon, as a property of the class of lexical items in its totality; and once – this time directly – by the rules of the categorial component.”

(Chomsky 1981: 29)

44 Martin Neef and Heinz Vater Both deficiencies – the doubling of information and the indirectness of subcategorization frames in the lexicon – were removed as a consequence of the introduction of the projection principle given in Chomsky (1981). On account of the projection principle, “the role of the categorial component of the base is reduced to a minimum” (Chomsky 1981: 31): It will only express “language-particular idiosyncrasies that are not determined by the lexicon” (Chomsky 1981: 31). It is the projection principle that takes care of a word’s combinations with other components of a syntactic structure. Although Chomsky does not explain the internal structure of the lexicon (cf. his apology in Chomsky 1981: 145, fn.81), he gives some details concerning basic properties of lexical items: “We take the lexicon to be a set of lexical entries, each specified as to category and complement structure, with further idiosyncrasies. The categorial and complement structure determine the possibilities for lexical substitution, which I will assume to be at the base level, largely as a matter of execution.”

(Chomsky 1981: 93) Summarizing, we can state that the conception of the lexicon as described in Chomsky (1981) marks progress in comparison to former views of the lexicon in Generative Grammar, since basic syntactic properties of words (especially their valency) are specified in the lexicon, which plays a “central role in the syntax by virtue of the projection principle” (Chomsky 1981: 92). On the other hand, Chomsky (1981) still considers the lexicon to be the place where primarily idiosyncrasies of language structure are stated. He does not furnish a detailed description of the internal structure of the lexicon. 2.3.3. Inside the lexicon: Lexical Phonology The relevant internal structure of the lexicon is supplied by researchers that deal with questions of morphology and phonology. Surprisingly, the most important contribution to the internal conception of the lexicon in the 1980s stems from work on phonology. In all models of the lexicon discussed so far, morphology completely precedes phonology, irrespective of the precise localisation of morphology in the grammatical derivation. However, there are data from different languages that show an interaction of morphological and phonological information. A good example is the past participle in German:

Concepts of the lexicon in theoretical linguistics

(5)

Past Participle in German past participle infinitive a. geredet reden gesucht suchen gepredigt predigen gearbeitet arbeiten getriangelt triangeln

past participle b. diskutiert versucht krakeelt palavert trompetet

45

infinitive diskutieren versuchen krakeelen palavern trompeten

These data exemplify the regular pattern of past participle formation. The question is under what circumstances the prefix ge- shows up in the participle. Underlining in the corresponding infinitives indicates the stressed syllable. Obviously, the past participle takes the prefix ge- if the base is stress-initial while it is missing if the first syllable of the base is unstressed. Considering stress to be the result of the application of a phonological rule, a morphological rule needs information that is supplied by phonological input conditions. Therefore, at least a part of phonology has to precede morphology. At the same time, morphology has to precede at least a part of phonology because morphological operations typically create the input for some phonological operations (like resyllabification). As a consequence, the model of grammar seems to require an interaction of phonological and morphological rules. This demand was included in the theory of Lexical Phonology (sometimes more adequately referred to as Lexical Phonology and Morphology) sketched in Kiparsky (1982).17 In this model, both phonology and morphology apply in the lexicon, i.e. in the presyntactic component of grammar. Both morphological and phonological rules are spread over several distinct levels in the lexicon. The exact number of levels that are necessary varies in accordance with the language to be analyzed and the author carrying out the investigation. The following diagram depicts a model that can be regarded as the standard model for West Germanic languages in the mid-eighties:

———–———————————————————————————— 17

Interaction of this type is also assumed in the approach of Anderson (e.g. 1982) but with the striking difference that both inflection and phonology apply in a postsyntactic component.

46 Martin Neef and Heinz Vater (6)

The lexicon in Lexical Phonology (from Wiese 1996: 116 and 128) LEXICON

underived lexical items MORPHOLOGY

PHONOLOGY

level 1

irregular inflection class I affixes

word stress

level 2

compounding class II affixes

compound stress

regular inflection

schwa epenthesis

level 3

POSTLEXICAL PHONOLOGY

According to this model, the lexicon has a filigree architecture comprising a list of morphemes as well as all morphological rules and many phonological rules. It is important to notice that some rules of phonology have to be applied in a component different from the lexicon, namely in the so-called postlexical part of grammar. This term falsely suggests that postlexical phonology immediately follows the lexicon in the grammatical derivation. In fact, however, the syntactic component intervenes between lexicon and postlexical phonology (which would therefore have been more aptly been termed ‘postsyntactic phonology’). The need for a postlexical phonology is supported by phonological regularities crossing the boundaries of words. Since words are the output of the lexicon and strings of words are the output of the syntax, a phonological process that applies across a word boundary is forced to be postlexical in nature. As a consequence, some rules of phonology like assimilation rules may belong to both the lexical and the postlexical part. This may be regarded as a shortcoming of this theoretical approach. The model of the lexicon given in (6) (sometimes coupled with wordsyntactic assumptions as in [3]) can be regarded as the standard version of the lexicon in generative theories around 1990. For example, Chomsky explicitly refers to the notion of the lexicon in the Extended Standard Theory (1991: 130) as the relevant conception in the current state of generative theory. A radical break is to be noted with the emergence of the Minimalist Program which is the current version of Generative Grammar. Without dis-

Concepts of the lexicon in theoretical linguistics

47

cussion, Chomsky returns to a basically structuralist view of the lexicon: “I understand the lexicon in a rather traditional sense: as a list of ‘exceptions’, whatever does not follow from general principles” (Chomsky 1995a: 235). 2.4. The lexicon in non-derivational theories The overview of the development of models of grammar using the generative framework reveals that the dominant view is a syntactic one: Syntax is regarded as the centre of grammar with other components being distributed relatively to this centre in a derivational sequence. The general idea of a derivational (or transformational) generative grammar is based on the fact that the grammaticality of a linguistic expression is determined by the sequential application of an ordered set of rules from different subcomponents to some abstract underlying representation. Thus, there is a directional relationship between an underlying representation and a surface representation. The notion of the lexicon in the Chomskyan approach to linguistics is closely connected to this overall architecture: The lexicon is a generative component of grammar that encompasses all aspects that precede the syntactic component. In short, the generative lexicon is the pre-syntactic component of a derivational generative grammar comprising both specific grammatical rules and a structuralist lexicon as a list of linguistic irregularities. In recent years, however, the derivational character of this conception has been criticized from different angles. Among the notions that were regarded as problematic the notion of intermediate levels of representation gained strongest criticism. Haider (1993: 12) remarks that a grammar is cognitively controllable only if there is a continuous relation between the grammar and the available data. This is given if a grammatical distinction has an observable equivalent in the data. If rules of grammar are distributed among several distinct levels of representation, the effects of the application of one rule may be erased by the effects of a rule that is applied at a later step of the derivation. Hence, the former rule may have effects that lack the condition of epistemological priority (a condition that Chomsky himself (1981: 10) regards as crucial). Beginning with work in computational morphology (Two-Level Morphology; Koskenniemi 1983) and with syntactic theories like Generalized Phrase Structure Grammar (Gazdar et al. 1985), Head-Driven Phrase Structure Grammar (Pollard & Sag 1994) and Projective Grammar (Haider 1993), the rule component of grammar was therefore replaced by sets of well-formedness conditions (or constraints). These con-

48 Martin Neef and Heinz Vater ditions are related to a single level of linguistic representation only. In this sense, the respective models are non-derivational, output-oriented, surfacebased, and monostratal. Another objection concerns the need of a derivational grammar for two different modes of determination of grammaticality, namely derivational rules on the one hand and representational constraints on the other hand. Prince & Smolensky (1993) argue that a theory that employs only one such mode is preferable. In their Optimality Theory, the only devices available are constraints for surface representations. In its constraint-based character, this theory resembles the above-mentioned syntactic theories. However, in Optimality Theory constraints are considered to be violable. This means that a linguistic form may be grammatical even if it violates some constraint as long as all other candidates for the realization of the intended form violate more important constraints. In declarative theories, on the other hand, constraints are conceived of as unviolable although allowing domainspecific scope. Examples are Declarative Phonology (Scobbie, Coleman & Bird 1996), Sign-Based Morphology (Orgun 1998) and Word Design (Neef 1996), the latter being an approach that emphasizes the notion of morphological constructions. Interestingly enough, a declarative approach allows to standardize the modes of description of both regular and irregular knowledge, while in Generative Grammar there is a distinction between irregular (lexical) knowledge captured in declarative statements and regular knowledge captured in procedural rules. Naturally, the lexicon has to have a different status in non-derivational theories as opposed to derivational theories discussed above. Notions like prelexical or postlexical should play no role in these conceptions because there is no derivational succession.18 However, work in the declarative paradigm shows at least two different conceptions of the lexicon. The first one stems from work on computational linguistics taking the lexicon as the list of lexemes, hence an abstraction from the actual vocabulary of a language.19 Originally conceived of as a closed list (e.g. Koskenniemi 1983), later approaches to morphology included productive aspects of vocabulary expan———–———————————————————————————— 18

19

This assessment only holds for theoretical conceptions which are strictly nonderivational. Some authors like Ito & Mester (2002) assume a serial conception of grammar within the framework of Optimality Theory by retaining the distinction between lexical and postlexical phonology. Thus, there may be a derivational residue in a basically non-derivational framework. This conception of a lexicon figures also in theoretical linguistics, e.g. in the morphological approach by Allen (1978) under the term ‘conditional lexicon’.

Concepts of the lexicon in theoretical linguistics

49

sion into the conception of the lexicon. Theories like Network Morphology (Corbett & Fraser 1993) and Type-based Morphology (Riehemann 1998) achieve this property by taking morphology basically as a consequence of lexical organization. In effect, this is more or less a declarative re-interpretation of the generative lexicon. However, Aronoff (1994) formulates sound criticism against such a conception of a “potential” lexicon that covers both fixed irregularities and productive means of lexeme-formation: “In fact, since the extensional notion of a potential lexicon plays no significant role that I know of in any theory of morphology, while the Bloomfieldian theory is crucial to a proper understanding of blocking and productivity, it seems best to simply dispense with both the notion of the potential lexicon and the name. The term lexicon should therefore be understood […] as referring only to the permanent lexicon: the list of all idiosyncratic signs, regardless of their category or complexity. The endless list of lexemes, by contrast, will remain nameless.”

(Aronoff 1994: 22)

Based on these considerations, Neef (2005a: 215, 2005b: 5) suggests a declarative conception of grammar that distinguishes between a lexicon as a collection of all types of linguistic irregularities and a grammar which is concerned with the regular portions of linguistic knowledge. Thus, the lexicon is a distinct and autonomous component of linguistic knowledge besides grammar and it resembles the structuralist conception of this notion. This declarative conception of linguistic knowledge may be sketched as follows: (7)

A declarative model of linguistic knowledge (cf. Neef 2005b: 5) GRAMMAR

phonology

morphology

syntax

semantics

LEXICON

surface forms

According to this conception, a linguistic (surface) form is grammatical if it obeys all relevant conditions of well-formedness. These conditions are spread over several linguistic modules that are defined with respect to the relevant domains. Phonology deals with the distribution of elements that have the potential to distinguish meanings, morphology deals with words as

50 Martin Neef and Heinz Vater the smallest freely occurring signs as combinations of form and meaning, syntax is concerned with the combination of words to phrases and sentences, while semantics is the locus for questions of meaning of linguistic expressions. In particular, there is no need for a generating device whatsoever. A declarative syntax, for example, is a theory that restricts the possible combinations of words in a language, with words being the basic units of syntax. If a language would lack any syntactic constraints, all combinations of words would be allowed in principle (although individual words could still entail demands on specific neighboring words). X-bar-theory can be interpreted as a set of declarative constraints prescribing possible structural relations between words. 3. Conclusion The meaning of a term depends on the theoretical framework in which it operates. If similar notions occur in different theories, it may nevertheless make sense to attribute the same term to them to keep the terminology limited. It should be kept in mind, however, that terms unavoidably have slightly different meanings in different theories.20 The term lexicon is especially interesting in this respect because the meanings related to it have an exceptionally wide range. In general, a lexicon is a store of specific aspects of linguistic knowledge with words as its central units. Restricted uses of the term lexicon can be observed in colloquial language where it is regarded as a dictionary, most likely in printed form, but also as a computerreadable data-base. In linguistics as the science of language, more abstract notions are connected with the term lexicon. In psycholinguistics, e.g., the term mental lexicon refers to linguistic knowledge in the mental equipment of individual language users (cf. Vater 1994: 226). In the preceding paragraphs, we have been primarily concerned with the observable concepts underlying the word lexicon in theoretical linguistics. We have identified three different concepts that play a major role in various linguistic theories. The first concept is that of the vocabulary of a language. The vocabulary is the stock of words of a language (in fact the existing stock of words) or the list of lexemes. Actually, it seems that the vocabulary concept is insignificant for theoretical approaches to linguistics. This is because the notion of the vocabulary blurs the border between regular and ———–———————————————————————————— 20

Cf. the elucidating criticism of the sometimes anarchistic use of linguistic terminology in Aronoff (1994: 5–8).

Concepts of the lexicon in theoretical linguistics

51

irregular linguistic knowledge. Including the mainstream in derivational Generative Grammar as well as most other approaches to linguistics, however, there is agreement that the set of irregular linguistic information plays a significant role in modelling linguistic knowledge. Especially in the Bloomfieldian tradition of Structural Linguistics, this set was captured by the term lexicon, indicating the second concept expressed by this word. The exact amount of such irregularities depends on the concomitant analysis of grammar as well as on the question of what types of items belong into the structuralist lexicon. The candidates for basic lexical items are morphemes, words, and lexemes as well as idiomatic phrases. Since morphologically simple words are signs in the Saussurean sense (i.e. arbitrary pairings of form and meaning), it is obvious that they have the closest connection to the structuralist lexicon. In quite a parallel way, Aronoff (1994: 16–22) gives a knowledgeable survey of two meanings of the adjective ‘lexical’, namely ‘idiosyncratic’ as opposed to ‘having to do with lexemes’. The main part of our considerations has been devoted to the attempt to make the line of development leading to a third concept of lexicon comprehensible. In a specific phase of derivational Generative Grammar, the lexicon was considered to be a full generative component of grammar. It is reasonable to adhere to such a reading of lexicon only in theories that are either derivational in nature or refuse morphology the status of an autonomous component of grammar, for whatever reasons. Thus, the meaning of lexicon in derivational Generative Grammar is that of the pre-syntactic component of grammar. This component contains at least the lexicon in the structuralist sense and at most such a lexicon plus all of morphology and most of phonology and maybe even some portions of syntax (disregarding questions of semantics). In retrospect, it would have been more adequate to call this concept of lexicon ‘pre-syntax’ (and to rename ‘postlexical phonology’ more adequately ‘postsyntactic phonology’), thereby reserving the term lexicon to the notion of irregular linguistic knowledge. Thus, a theory of the (generative) lexicon is a theory of derivational pre-syntax.

References Allen, Margret 1978 Morphological investigations. PhD diss., University of Connecticut. Allen, Robert 2000 New Penguin English Dictionary. Harmondsworth: Penguin.

52 Martin Neef and Heinz Vater Anderson, Stephen R. 1982 Where’s morphology? Linguistic Inquiry 13: 571–612. 1992 A-Morphous Morphology. Cambridge: CUP. Aronoff, Mark 1976 Word Formation in Generative Grammar. Cambridge, MA: MIT Press. 1994 Morphology by Itself. Stems and Inflectional Classes. Cambridge, MA: MIT Press. Bechert, Johannes, Danièle Clément, Wolf Thümmel and Karl Heinz Wagner 1970 Einführung in die generative Transformationsgrammatik. München: Hueber. Bloomfield, Leonard 1933 Language. New York: Holt. Botha, Rudolf P. 1968 The Function of the Lexicon in Transformational Generative Grammar. The Hague/Paris: Mouton. 1989 Challenging Chomsky. The Generative Garden Game. Oxford: Basil Blackwell. Bußmann, Hadumod 1996 Routledge Dictionary of Language and Linguistics. London: Routledge. Chesterman, Andrew 1980 Contrastive generative grammar and the psycholinguistic fallacy. Papers and Studies in Contrastive Linguistics 11: 17–24. Chomsky, Noam 1955 The Logical Structure of Linguistic Theory. Ms., Harvard University and MIT (Published in part 1975, New York: Plenum). 1957 Syntactic Structures. The Hague/Paris: Mouton. 1965 Aspects of the Theory of Syntax. Cambridge, MA: MIT Press. 1970 a Remarks on nominalizations. In Readings in English Transformational Grammar, R.A. Jacobs and P.S. Rosenbaum (eds.), 184–221. Waltham: Ginn & Co (Reprinted in Chomsky 1972b: 11–61). 1970 b Deep structure, surface structure, and semantic interpretation. In Studies in General and Oriental Linguistics Presented to Shiro Hattori an the Occasion of his Sixtieth Birthday, R. Jacobson and S. Kawamoto (eds.). Tokio: TEC (Reprinted in Chomsky 1972b: 62–119). 1972 a Some empirical issues in the theory of transformational grammar. In The Goals of Linguistic Theory, S. Peters (ed.). Englewood Cliffs: Prentice-Hall (Reprinted in Chomsky 1972b: 120–202). 1972 b Studies on Semantics in Generative Grammar. The Hague: Mouton. 1981 Lectures on Government and Binding. Dordrecht: Foris. 1986 Knowledge of Language: Its Nature, Origin, and Use. New York: Praeger Publishers.

Concepts of the lexicon in theoretical linguistics 1991

53

Some notes on economy of derivation and representation. In Chomsky 1995 b: 129–166. 1995 a Categories and transformations. In Chomsky 1995b: 219–394. 1995 b The Minimalist Program. Cambridge: MIT Press. Corbett, Greville G. and Norman A. Fraser 1993 Network morphology: a DATR account of Russian nominal inflection. Journal of Linguistics 29: 113–142. Crystal, David 1992 The Cambridge Encyclopedia of Language. Cambridge: CUP. Di Sciullo, Anna Maria and Edwin Williams 1987 On the Definition of Word. Cambridge, MA: MIT Press. Duden 1994 Das große Fremdwörterbuch. Mannheim etc.: Dudenverlag. Fodor, Jerry 2002 The lexicon and the laundromat. In The Lexical Basis of Sentence Processing: Formal, Computational and Experimental Issues, P. Merlo and S. Stevenson (eds.), 75–84. Amsterdam /Philadelphia: Benjamins. Gazdar, Gerald et al. 1985 Generalized Phrase Structure Grammar. London: Blackwell. Haider, Hubert 1993 Deutsche Syntax – Generativ. Tübingen: Narr. Halle, Morris and Alec Marantz 1993 Distributed Morphology and the pieces of inflection. In The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger, K. Hale and S. J. Keyser (eds.), 111–176. Cambridge, MA: MIT Press. Handke, Jürgen 1995 The Structure of the Lexicon. Berlin: de Gruyter. Hockett, Charles 1958 A Course in Modern Linguistics. New York: Macmillan. Hoekstra, Teun, Harry van der Hulst and Michael Moortgat 1980 Introduction. In Lexical Grammar, T. Hoekstra, H. van der Hulst and M. Moortgat (eds.), 1–48. Dordrecht: Foris. Ito, Junko and Armin Mester 2002 Lexical and postlexical phonology in Optimality Theory: Evidence from Japanese. In Resolving Conflicts in Grammars, G. Fanselow and C. Féry (eds.), 183–207. (Linguistische Berichte, Sonderheft 11) Hamburg: Buske. Jackendoff, Ray 1975 Morphological and semantic regularities in the lexicon. Language 51: 639–671. 1983 Semantics and Cognition. Cambridge, MA: MIT Press.

54 Martin Neef and Heinz Vater Jensen, John T. and Margaret Stong-Jensen 1984 Morphology is in the lexicon! Linguistic Inquiry 15: 474–498. Kiparsky, Paul 1982 Lexical Morphology and Phonology. In Linguistics in the Morning Calm, I. S. Yang (ed.), 3–91. Seoul: Hanshin. Klappenbach, Ruth and Wolfgang Steinitz 1977 Wörterbuch der deutschen Gegenwartssprache. Berlin: AkademieVerlag. Koskenniemi, Kimmo 1983 Two-level morphology: a general computational model for wordform recognition and production. PhD diss. University of Helsinki. Lieber, Rochelle 1981 On the Organization of the Lexicon. Bloomington: Indiana University Linguistics Club. Meibauer, Jörg 2002 Lexikon und Morphologie. In Einführung in die germanistische Linguistik, J. Meibauer et al., 15–69. Stuttgart/Weimar: Metzler. Neef, Martin 1996 Wortdesign. Eine deklarative Analyse der deutschen Verbflexion. Tübingen: Stauffenburg. 2005 a Die phonologischen Einheiten des Deutschen aus Sicht einer Autonomen Deklarativen Phonologie. Linguistische Berichte 202: 207– 249. 2005 b Die Graphematik des Deutschen. Tübingen: Niemeyer. Olsen, Susan 1986 Wortbildung im Deutschen. Stuttgart: Kröner. Orgun, Orhan Cemil 1998 Cyclic and noncyclic phonological effects in a declarative grammar. Yearbook of Morphology 1997: 179–217. Pfeifer, Wolfgang et al. 1989 Etymologisches Wörterbuch des Deutschen. Berlin: Akademie-Verlag. Pollard, Carl and Ivan Sag 1994 Head-Driven Phrase Structure Grammar. Chicago: Chicago University Press. Prince, Alan and Paul Smolensky 1993 Optimality Theory. Constraint interaction in Generative Grammar. (Rutgers University Center for Cognitive Science Technical Report 2) Ms., Rutgers University. Pullum, Geoffrey 1996 Nostalgic views from MIT. Review article on The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger, K. Hale and S. J. Keyser (ed.). Journal of Linguistics 32: 137–147.

Concepts of the lexicon in theoretical linguistics

55

Riehemann, Susanne 1998 Type-based derivational morphology. Journal of Comparative Germanic Linguistics 2: 49–77. Saussure, Ferdinand de 1916 Cours de linguistique générale. Paris: Payot. Schwarze, Christoph and Dieter Wunderlich 1985 Einleitung. In Handbuch der Lexikologie, C. Schwarze and D. Wunderlich (eds.), 7–23. Königstein: Athenäum. Scobbie, James M., John S. Coleman and Steven Bird 1996 Key aspects of Declarative Phonology. In Current Trends in Phonology: Models and Methods. Volume II, J. Durand and B. Laks (eds.), 685–709. Salford: European Studies Research Institute. Selkirk, Elisabeth 1982 The Syntax of Words. Cambridge, MA: MIT Press. Summers, Della Longman Dictionary of Contemporary English. München: Langen19953 scheidt-Longman. Toman, Jindrich 1983 Wortsyntax. Tübingen: Niemeyer. Vater, Heinz 1994 Linguistik: Natur- oder Geisteswissenschaft? In Germanistentreffen Bundesrepublik Deutschland – Polen: 16.9.–30.9.1993, W. Roggausch (ed.), 219–232. Bonn: DAAD. 1999 Wort und Begriff. Eine terminologische Klärung. In Deutsch von innen und außen. Ulrich Engel zum siebzigsten Geburtstag, A. Ktny (ed.), 147–153. Pozna: Wydawnictwo Naukowe UAM. Wiese, Richard 1996 The Phonology of German. Oxford: Clarendon.

Towards a structural typology of verb classes Dieter Wunderlich

After the introduction, section 2 considers prototypical transitive verbs. Section 3 turns to intransitive verbs and their relationship to transitive ones, and also considers the active-inactive type of argument linking (3.4). Section 4 deals with further argument linking types for transitive verbs: the inverse type (4.1), the salience or voice type (4.2), the positional type (4.3), and the generalized case type (4.4), the latter comprising accusative, ergative and split systems, and the possibility of dative. Section 5 considers ways of marking special semantic classes of verbs lexically. Section 6 discusses how a third argument is integrated, and thus extends the typology of section 4. It deals with the semantic decomposition of ditransitive verbs (6.1) and general principles of constraining it (6.2), considers serial verb constructions and noun incorporation as argument-reducing operations (6.3), turns to constructions where the recipient is treated like the object of a transitive verb (6.4) or differently from the object of a transitive verb (6.5), and closes with a new look on the English ‘dative’ alternation, which is compared with the double object/serial verb alternation in some West African languages.

1. Introduction: The grammar is a set of structural generalizations What one needs to know about verbs is whether they have one, two, three or even more arguments: intransitive laugh has one argument, transitive see has two arguments, and ditransitive give has three arguments. If one is concerned with a particular language, one also needs to know how these arguments are realized. Turning from English to Turkish, to Georgian or to one of the indigenous languages of the Americas, one finds quite different argument linking types, i.e. ways in which the arguments of a verb are realized. This article tries to summarize some recent work on both argument structure and argument realization. The first question is how certain we are that every language has verbs. The majority of linguists is convinced that the existence of two well-distinguished lexical categories, namely verbs and nouns, is one of the most cer-

58

Dieter Wunderlich

tain universals of human language. Some linguists, however, dispute such a universal claim. They present two kinds of counter-evidence: (i) Some languages do not exhibit clear differences between morphological means that apply to verbs and those that apply to nouns (Sasse 1993a on Cayuga, an Iroquoian language). (ii) Other languages exhibit large sets of lexical roots that function as verbs when combined with an aspect or tense marker, but function as nouns when combined with a definite article (Broschart 1997 on Tongan, an Oceanic language). For a survey of such reservations see Sasse (1993b). A closer inspection of these ‘counter-examples’, however, shows that these languages do not represent exceptions to the abovementioned universal. Their grammar shows clear asymmetries between verbs and nouns, in the morphological behavior, the incorporation of nouns into verbs, and the forming of new nominal stems derived from verbs (Wunderlich 1996a; Baker 2003). In a recent debate on Munda (Austroasiatic languages of India), for which the existence of lexical categories is also questioned, Evans & Osada (2005a: 384) conclude that “Munda clearly distinguishes nouns from verbs”. There is of course some semantic background for this distinction. Prototypically, verbs (such as sleep, stay, hit, give) denote temporally changing entities (events or states) in which one or more objects are participating, while nouns (such as man, house, bottle, salt) denote temporally constant entities (objects determined merely spatially) which can participate in events or states. Usually, a complex scene where a man is sleeping is decomposed into the temporal predicate SLEEP (with an argument slot) and some instance of the nontemporal predicate MAN (an argument that fills the slot); SLEEP is associated grammatically with a verb and MAN with a noun, rather than the other way around. Nearly every language encodes ‘a manN sleepsV’ rather than ‘a sleepN occurs-to-manV’. However, the classification into verbs and nouns is not purely semantics-driven. Given the rich conceptual variation into punctual vs. extended events, on-going activities vs. achievements, permanent vs. temporally restricted states, masses vs. individual objects, concrete vs. abstract activities/ objects one must wonder why there are just two major lexical classes rather than, say, eight or ten. Moreover, a closer inspection of the vocabulary of a language shows that some items seem to be wrongly classified: nouns such as journey, war, and game denote events rather than objects, while verbs such as resemble, exist, be above, and be tall do not denote events. One point to be made here is that some languages (such as English) in addition to verbs and nouns also have prepositions (aboveP) and adjectives (tallA), so

Towards a structural typology of verb classes

59

that a more detailed classification arises. Other languages, such as Chinese, do not make such a clear addition to the categorial inventory. More important is another point: lexical categories evolve in a set of lexical items from generalization towards certain structural patterns, a fact that can be summarized by a distinctive grammatical feature by which the lexicon is partitioned into two classes, a designated one (such as verbs) and the remaining (unmarked or default) class (called nouns). Such a classification is more robust and more economical than a purely semantic classification, however, it leaves certain ‘less motivated’ assignments, which in the history of a language may survive as relics. A child that starts out to learn the words of a language seems to need only a short time to detect the verb class vs. the noun class, and as soon as this happens, the child produces overgeneralizations, i.e., classifies not always in accordance with the language to be learned. If one has acquired the basic classification of the vocabulary, there is no further need to motivate it semantically. Only for a new item does the question arise to which class it belongs, which is mostly decided on the basis of semantic (or, sometimes, phonological) similarity to already existing items – if a semantically similar word is classified as a verb, the new item will also be assigned to verbs. Furthermore, semantic factors continue in playing a contextual role when the particular meanings of items are in focus. Within the class of verbs, various kinds of semantic subclassification come into mind: verbs with animate or inanimate arguments, verbs of movement, position or placement, verbs of manipulation, experience, perception, communication, and so on. Nearly none of these possible semantic factors is decisive for the further grammatical subclassification of verbs, except animacy in some languages. For instance, the Algonquian languages from North America make a formal distinction between stems with animate and those with inanimate objects (e.g. waapam ‘see sth. animate’ vs. waapaht ‘see sth. inanimate’ in Plains Cree). The most robust subclassification of verbs concerns the number of arguments: intransitive verbs have one, transitive verbs have two, and ditransitive verbs have three nominal arguments. (Verbs with zero valency are extremely rare – one possible semantic class of this kind are weather verbs, such as Latin pluit ‘it rains’, however, note that English uses here an expletive pronoun, which masks the verb to be intransitive.) Besides that, verbs are subclassified of whether they take a clausal complement (verbs of mental attitudes), which under some conditions can also be reduced to an infinitive or another non-finite verb form (he hopes to win vs. he hopes that he

60

Dieter Wunderlich

will win). Furthermore, at least some languages have a subclass of verbs that take a locational argument, e.g., a prepositional phrase (he sits on the bank, he puts the cans on the bank). Sometimes one also finds a class of verbs that take prepositional objects in which the preposition is lexically fixed (without contributing a particular meaning): an jdn denken ‘think at someone’, auf etw. hoffen ‘hope for sth.’, an etw. glauben ‘believe in sth.’ If two nominal arguments occur with a verb, the meaning of the verb sometimes requires one argument to be animate and the other to be inanimate (read, sew, enter), however, more than often this is not the case. The second argument of see, for instance, can be inanimate or animate; in the latter case, the two arguments can in principle be exchanged, thereby shifting the intended reading (the man saw the lion and the lion saw the man mean different things). It is even possible that a verb has three arguments being similar in their nature; for example, send and introduce allow any ordering of three arguments referring to a person (the neighbor sent/introduced the woman to a specialist; a specialist sent/introduced the neighbor to the woman; etc.). If the number of arguments counts, there must be some way to make the arguments recognizable in their specific role: which nominal functions as which argument. In English, this decision is made by position: usually the ‘subject’ precedes the verb, and the ‘object’ follows it. Other languages primarily use morphological case on nominals, or they use pronominal affixes or clitics attached to the verb. The particular device that maps argument roles onto morphosyntactic patterns is called ‘argument linking’. Various types of argument linking are known, some of which will be discussed in section 4.1 Each argument linking type represents an economic way of avoiding ambiguities with verbs having more than one nominal argument. As we will see, each of these devices simultaneously enforces a certain subclassification of verbs according to structural parameters. In any case, the classification of verbs seems to depend more on the possible realization patterns for their arguments than on inherent semantic features. Sometimes, however, it is possible to use a formal device to mark an exceptional semantic class of verbs. For instance, German has a small class of intransitive experiencer verbs with accusative marking, which is unusual for intransitives: mich friert ‘I am cold’, mich fröstelt ‘I am shivering with cold’. ———–————————————————————————————

1

My classification of argument linking types goes back to Wunderlich (2002b).

Towards a structural typology of verb classes

61

Among the various argument linking types one type is found that seems to be rather uneconomical: the portmanteau-type. A portmanteau affix simultaneously specifies more than one type of information. For instance, Latin passive is realized by a set of suffixes that inform us about both passive and person-number (ama-mini ‘you (pl) are loved’, with -mini ‘2pl.PASS’). Another type of portmanteau affix simultaneously specifies more than one argument. Hungarian uses the single suffix -lak/-lek to express the combination I you2, similarly, Ayacucho Quechua uses the suffix -yki to express this combination, see (1 a,b). (1)

I you portmanteau suffixes in Hungarian and in Quechua a. szeret-lek ‘I love you (sg/pl).’ b. riku-yki ‘I see you.’

To express the combination I you is very special because here the speaker ‘acts upon’ the addressee in the propositional content similar to the speech act itself, so it does not wonder that a portmanteau morpheme adapted to this special task is found in several languages. However, if all possible combinations of subject values and object values are expressed by portmanteaux, the number of these affixes increases, and it is not possible for the learner to make any separate generalizations about subjects and objects. That portmanteau affixes in general are uneconomical can be seen from a simple calculation. With three persons (1st, 2nd, and 3rd person) and two numbers (singular and plural) one would need at most 6 affixes for the subject and 6 affixes for the object (together 12 affixes); however, there are 28 possible combinations (if all reflexive settings such as I/we I/we and you you are excluded). The portmanteau linking type, therefore, is strongly disfavored. However, Kiowa, a Tanoan-Kiowa language spoken in west central Oklahoma, obviously exhibits this type, as illustrated by the examples in (2). Here, the prefixes simultaneously specify person and number of the two arguments of a transitive verb (2a), and in addition, they specify the number (dual or plural) of a third argument (2b).

———–————————————————————————————

2

The arrow is used to express subjectobject information.

62

Dieter Wunderlich

(2)

Portmanteau prefixes in Kiowa (Watkins 1984) a. Transitive verbs: a single prefix encodes information about both arguments (góp ‘hit’). èm-góp ‘I hit you/him’ èm: 1sg 2  3sg

dO´-góp ‘you/he/they hit us’ dO´ : 2  3 1pl

é-góp ‘you/he hit me’ é: 2  3sg 1sg

gO` -góp ‘we/he/they hit you’ gO` : 1pl  3 2sg

b. Ditransitive verbs: a single prefix encodes information about all three arguments (O ‘give’, kut ‘book’) kút né-O´ ‘you/he gave me two books’ né: 2  3sg / 1sg/ dl kút bédê-O´ ‘you (pl) gave me/us/him two books’ bédê: 2pl/ 1  3sg/ dl

kút nén-O´ ‘I gave you/him two books’ nén: 1sg/ 2  3sg/ dl kút gyâ-O´ ‘they gave me/us/him some books’ gyâ: 3pl/ 1  3sg/ pl

This situation is not so bad from the perspective of the speaker (in terms of economy) because nearly all prefixes are ambiguous (e.g., cover both 2nd person and 3sg), but that is certainly not welcomed from the perspective of the hearer. Therefore, Kiowa is not a language that invites people to participate. Indeed, the population is small and isolated (according to the 1990 census, only about 1,000 middle-aged and older speakers out of a population of 6,000). Probably, the observed prefixes are the result of massive fusion (forced by the pressure towards ‘one prefix only’), and the portmanteau type observed in this language is not a real candidate of typological choice. In the following, I will disregard such extreme but unfelicitious cases and rather concentrate on the more frequent structural types. I will start with prototypical transitive verbs in section 2 because these verbs constitute the core of a grammar. I then turn to intransitive verbs, including their relationship to transitive ones, in section 3; properties of these verbs determine what is known as active-inactive type. Further argument linking types, which all are based on transitive verbs, are considered in section 4. Section 5 demonstrates how special semantic classes can be marked by morphosyntactic means already available for a language. Section 6 is concerned with the structural possibilities of dealing with a third argument in ditransitive verbs.

Towards a structural typology of verb classes

63

2. Prototypical transitive verbs are in the center of a grammar Transitive verbs, which denote a relation between two participants of an event, belong to the most ‘ingenious’ inventions of human language. They are in the center of any grammar; if they were absent from the lexicon, the grammar would be much simpler than it actually is. Typical (or canonical) transitive verbs are chase, hit, kill, eat, kiss, and many more. They do not form a characteristic semantic class (representing similarities in some basic field of activities such as hunting-gathering, nourishing, social and sexual behavior), but rather reflect a very deep structural generalization (which must have been an important step in the evolution of human language). Every language allows the enrichment of the class of transitive verbs by less canonical items such as see, hear, meet, ask, obtain, surround, as well as the derivation of new transitive verbs such as enlarge and to open – although the languages differ widely in the details. So what is the common property of transitive verbs? Canonical transitive verbs are two-place predicates with two clearly distinguished argument roles, which is illustrated with the verb eat in (3).3 (3)

y x EAT(x,y) x (= the eater) is the higher argument, and y (= the eaten) is the lower argument. Each  abstracts over an argument and thus represents the argument role or slot. All -roles (y, x) must be saturated in the course of composition, i.e. the two arguments must be realized by some linguistic expressions.

———–———————————————————————————— 3

We say that apple denotes a certain individual fruit, which is represented as x APPLE(x), where x counts as the referential argument of the noun. It is manipulated (specified and bound) by the functional categories on the noun, such as determiners and quantifiers (the apple, every apple). Similarly, one can say that eat denotes a certain individual event, which is represented as y x e EAT(x,y)(e), where e counts as the referential argument of the verb. It is manipulated (specified and bound) by the functional categories on the verb, such as aspect, tense, and mood. Since this article is concerned with the argument structure of verbs, which I believe to be independent of the verb’s capacity of denoting events, the referential argument of verbs is disregarded.

64

Dieter Wunderlich

For instance, one can first apply ‘eat’ on ‘the apple’, and then apply the result on ‘John’, which yields ‘John eats the apple’. (4)

a. (eat the apple):

y x EAT(x,y) (the.apple) = x EAT(x, the.apple)

b. (John (eats the apple)): x EAT(x, the.apple) (John) = EAT(John, the.apple) An important observation is that transitive verbs are always asymmetric. It is easy to make a semantic distinction between the eater and the eaten, but is not so easy to make a semantic distinction between the two persons who marry each other. MARRY is a symmetric predicate (if John married Anne, than Anne also married John), nevertheless, the verb marry is in the same way grammatically asymmetric as eat: in ‘John married Anne’, ‘John’ realizes the higher argument, and ‘Mary’ the lower one. This is the cost for putting marry into the class of transitive verbs. The source for the generalization to the concept of transitive verb is of course semantic. In a prototypical transitive verb (such as eat rather than marry), the higher argument functions as agent or actor (‘the participant which performs, effects, instigates, or controls the situation denoted by the predicate’, Foley & Van Valin 1984: 29), while the lower argument functions as patient, or undergoer, or affected. Several theories try to describe the realization of arguments as a mapping from a semantic hierarchy of thematic or eventive roles to a hierarchy of grammatical functions (subject, object) or cases (nominative, accusative). Thematic hierarchies include various roles such as agent, patient, theme, experiencer, beneficiary, recipient, goal, source, but it is discussed controversially how these roles are demarcated and what their exact ordering is (Grimshaw 1990, Jackendoff 1990; see also Bornkessel et al. to appear). Only two proposals succeed in reducing the number of thematic roles to just two (corresponding to the two arguments of a transitive verb): the two protoroles in Dowty’s (1991) account (protoagent and protopatient) are determined by the relative weight of several semantic factors, while the two macroroles in the account of Foley & Van Valin (1984) (actor and undergoer) reflect semantical and grammatical factors. Eventive roles account for the participation of arguments in possible subevents such as an ongoing activity and a change of state (Pustejovsky 1991, among others). A summary is given in (5).

Towards a structural typology of verb classes

(5)

65

The asymmetry of transitive verbs hierarchy:

the lower argument (nearer to the verb)

the higher argument

protoroles:

protopatient

protoagent

macroroles:

undergoer

actor

conceptual inferences:

affected

controller

eventive roles:

specifies the result

specifies the ongoing activity

grammatical roles:

object

subject

case (accusative system):

accusative

nominative

case (ergative system):

nominative

ergative

features:

+hr

–hr

–lr

+lr

As a facon de parler, it is often convenient to use the semantic notions of agent vs. patient, or the more structural notions of subject vs. object. However, one must be careful to avoid misinterpretations: in ‘a wall surrounds the garden’, the higher argument, namely ‘the wall’, is certainly not an agent semantically, and in ‘the bear was killed’, ‘the bear’ is considered to be the (grammatical!) subject, although it certainly functions as patient. The most neutral way is to encode the asymmetry (i.e. the hierarchy of roles) by two simple relational features: +hr ‘there is another role which is higher than this one’ (for short: ‘there is a higher role’) characterizes the lower argument, while +lr ‘there is a lower role’ characterizes the higher argument (Wunderlich 1997a).4 These features, representing ‘abstract case’, have the advantage that their positive value corresponds to the information of mor———–———————————————————————————— 4

+hr could as well be spelled out as ‘I am a lower role’. The particular choice of notation is due to a historical accident. In any case, the comparative features are different from Kiparsky’s (1992, 1998) superlative features such as +HR ‘I am the highest role’ (which is identical to our –hr). The reason to deviate from Kiparsky’s proposal is a more principled conception of markedness. As we will see in the following, the plus values in our system correlate with marked instances, and the minus values with unmarked instances. This is in line with phonological conventions according to which, e.g., a voiced consonant (such as d, b or g in contrast to t, p, k) is marked by a plus value (+voiced) rather than by a minus value such as –voiceless.

66

Dieter Wunderlich

phological case: accusative bears the feature +hr (and thus marks the lower argument), while ergative bears the feature +lr (and marks the higher argument). The subject of an intransitive verb is designated as –hr,
lr; therefore, neither accusative nor ergative are compatible with this argument. However, the nominative (bearing no feature specification) is compatible with an intransitive subject; it is indeed compatible with any argument. But in the morphology usually the most specific option is chosen: if accusative is possible, then it should be chosen. This specification of accusative and ergative conforms to the traditional way in which these notions are used cross-linguistically. If the agent (A) is realized like the subject of an intransitive verb (S), and the patient (P) is realized differently (by means of a morphological case, a postposition, or a pronominal affix), the marker that encodes P is called ‘accusative’ (ACC). However, if P is realized like S, and A is realized differently, the marker that encodes A is called ‘ergative’ (ERG), see (6). The respective other type of realization (regardless of whether it is specifically marked or simply unmarked, for instance, if the stem is used) is called ‘nominative’ (NOM). (6)

The definitions of ‘accusative’ (ACC) ACC

intransitive transitive

P

vs.

‘ergative’ (ERG)

NOM

NOM

S

S

A

P

ERG

A

Accusative is illustrated with an example from German (7), and ergative with one from Basque (8). Note that the auxiliary in Basque agrees with all the structural cases occurring with a verb; ‘E’ indicates agreement with ergative, and ‘N’ indicates agreement with nominative. (7)

Accusative in German a. intransitive: Hans schläft. ‘John sleeps.’ b. transitive: Hans sieht ein-en Mann. John sees a-ACC man ‘John sees a man.’

(8)

Ergative in Basque a. intransitive: Jon joan da. John go.PERF AUX.3N ‘John went.’

Towards a structural typology of verb classes

b. transitive:

67

Jon-ek zopa jan du. John-ERG soup eat.PERF AUX.3N.3sgE ‘John ate the soup.’

The definition in (6) suggests that accusative and ergative are complementary to each other, so one must wonder why accusative and ergative are unevenly distributed among the languages of the world. Languages with accusative are more frequent than those with ergative. It seems that A is the more prominent argument, and that it is preferred to realize the more prominent argument by nominative because it can then much easier function as syntactic pivot, which controls the formation of more complex structures (such as coordinations, relative clauses, or control verb constructions of the type I want to see him). This inherent asymmetry is expressed by the scale +hr > +lr ‘It is better to mark a lower argument (an object) than a higher argument (the subject)’, see section 4.4 below. It is important to notice that accusative and ergative are not at all universal notions. Although the majority of languages seem to have developed some device that can properly be described by these terms, there are nevertheless many languages with well-defined argument linking types of different properties. It is therefore interesting to study these different types and to see how they frame possible generalizations of verb classes. One further remark is at place here. Proponents of a semantic account of argument structure often deny that the arguments of a verb are ordered. I think that the assumption of a strict ordering is conceptually much simpler than the alternative of assuming variation in ordering, or considering the arguments to be an unordered set. One reason is that a strict lexical ordering allows a simple mapping onto linear syntactic order. As one can indeed observe, the default ordering in the syntax mostly conforms to argument hierarchy, regardless of whether the arguments are marked by case or not. For instance, in (7b) and (8b) the object follows the subject, and so it does in the respective English translations. Moreover, the Barss & Lasnik (1986) tests – if properly adapted for the language under question – are able to identify the ranking of arguments independently in most cases. These tests include anaphoric binding, bound pronouns, weak crossover, negative polarity items, and multiple questions, and are briefly illustrated in (9). (9)

Tests for argument hierarchy. (The co-indexed expressions are assumed to refer to the same entity.) a. The ape sees himself. *Himself sees the ape.

68

Dieter Wunderlich

b. All mothersi like theiri baby. *Theiri mothers like all childreni. c. Which mani beats hisi son? *Whose soni did hisi father beat? d. No one ate anything. *Anyone ate nothing. e. Who prefers what? *What does who prefer?

What disturbs who? *Who does what disturb?

These tests can also be applied to show the relative ordering of two objects in a verb with three arguments. It is still open to discussion to what extent these tests are sensitive to the semantic hierarchy of arguments, or merely to morphological asymmetry or syntactic ordering. I will come back to this issue in section 4.1 when I discuss inverse morphology. 3. Intransitive Verbs reflect semantic classes Intransitive verbs are simpler than transitive ones because they have only one argument. As I pointed out above, a grammar does not start with the intransitives because they do not show the features of ordering and asymmetry. However, the intransitives are interesting for their own. They most clearly show the possibilities and effects of semantic classification, they constitute the basis of intransitivizing vs. transitivizing morphology, and they give rise to one of the major types of argument linking. 3.1. Two or four classes of intransitive verbs? Regarding the grammatical behavior of intransitive verbs, many languages exhibit just two classes, whereby two types of semantic criteria play a role. In Lakhota, a Siouan language, one finds a distinction between agentive and nonagentive intransitives (10): in the 1st or 2nd person, agentive verbs encode their subject differently from nonagentive verbs (which is discussed in more detail in section 3.4). In contrast, the Mayan language Yukatek shows a distinction according to inherent aspect (11): inherently imperfective verbs take the perfect marker -(n)ah but are unmarked in the imperfect, whereas inherently perfective verbs take the imperfect marker -Vl (with the vocalic segment V underlying vowel harmony) but are unmarked in the perfect.

Towards a structural typology of verb classes

69

(10) Intransitive verbs of Lakhota, ordered according to agentivity (Van Valin 1977) a. agentive verbs hi iyotaka psica slohã thasosa

b. nonagentive verbs arrive sit down jump crawl spit

hãska h"‚xpaya ile khata puza

be tall fall down burn be hot be dry

(11) Intransitive verbs of Yukatek, ordered according to telicity (Krämer & Wunderlich 1999) a. inherently imperfective verbs b. inherently perfective verbs /aalkab/ balak/ hé/esíin kirits/ naay nóok/ ook/ot tóoS tSe/eh Sik/nal

run roll sneeze squeak dream snore dance rain laugh fly

/eem-el /ok-ol kim-il k/uch-ul lub/-ul luk/-ul naats/-al sih-il tS/iih-il waak/-al

descend enter die arrive fall leave, escape/flee approach be born grow explode, burst

Obviously, Lakhota and Yukatek make a different cut in the set of intransitives. The two classes of Lakhota contain both inherently imperfective and inherently perfective verbs, and the two classes of Yukatek contain both agentive and nonagentive verbs. These examples thus suggest a semantic cross-classification of intransitive verbs into four classes, a classification which seems to be exhaustive, although most languages do not reflect all these possibilities in their grammar. All formal classes of a language derive from certain generalizations forced by prototypical instances, and so do the intransitive verb classes. In the course of lexicalization and subsequent semantic shift, however, certain items may turn out to be ‘wrongly’ classified semantically; they remain as idiomatic. Note also that one of the Lakhota classes includes statives, whereas statives of Yukatek form a separate class. The cross-classification in the set of intransitive verbs suggested by the examples above (and proposed by Dowty 1991) is summarized in (12) and

70

Dieter Wunderlich

there illustrated with some English examples. The distinction agentivenonagentive is represented by the feature [control], which is associated with the subject of the verb; [+control] verbs denote an event that is brought about by an agent. Transversely, the distinction perfective-imperfective is represented by the feature [telic], which is associated with the event denoted by the verb; [+telic] verbs denote an event that brings about a certain result. (One can decompose these verbs by means of the predicate BECOME, so that DIE is represented as BECOME DEAD.) Since control is often associated with an ongoing activity, and a result with the absence of an agent, there is a semantic correlation between [+control] and [–telic], and, vice versa, between [–control] and [+telic]. Correspondingly, one finds more items in the [+control,–telic] and the [–control,+telic] classes than in the two other ones. (12) Cross-classification of intransitive verbs

[+telic];

agentive [+control]; x+contr VERB(x)

nonagentive [–control]; x -contr VERB(x)

come, stand up

‘unaccusatives’: arrive, die, fall, grow

‘unergatives’: dance, laugh, run, speak

lie, stink

BECOME

[–telic]

Syntacticians often distinguish between ‘unaccusative’ and ‘unergative’ verbs (in considering them as would-be transitives). The ‘unaccusatives’ are said to have an underlying object that cannot receive accusative; correspondingly, the ‘unergatives’ are said to have an underlying subject that cannot be realized by ergative (just because accusative and ergative are special for transitive verbs).5 This distinction seems to correlate best with the two classes characterized by [+control, –telic] and [–control,+telic]. The notion of an ‘underlying object’ of an intransitive verb, however, remains doubtful, since what is meant is a semantic distinction somewhat reflected in the grammatical behavior of verbs. Nearly every language shows some grammatical reflexes of unaccusativity vs. unergativity. In German, one finds a distribution of the properties ———–———————————————————————————— 5

The unaccusative verbs were first called ‘ergative’ verbs (Burzio 1981, 1986, den Besten 1984, Keyser & Roeper 1984). The notions ‘unaccusative’/‘unergative’ go back to Perlmutter & Postal (1984).

Towards a structural typology of verb classes

71

given in (13) roughly into two classes, which have been associated with the notions ‘unergative’ and ‘unaccusative’ (Haider 1985). The members of one large class of intransitives can be passivized, can undergo er-nominalization, have haben ‘have’ as the perfect auxiliary, and their participles cannot function as attributes. The members of another large class of intransitives behave in every respect conversely: they cannot be passivized, cannot undergo er-nominalization, have sein ‘be’ as the perfect auxiliary, and their participles can function as attributes. This distribution is illustrated with the examples in (14). (13) The grammatical behavior of intransitive verbs in German (tendencies)

have a passive undergo er-nominalization allow an attributive participle select the perfect auxiliary

(14)

unergatives (+control,
telic)

unaccusatives (
control,+telic)

yes yes no haben

no no yes sein

unergatives Im Sommer wurde oft getanzt. Passive of ‘In the summer they often danced.’ die besten Tänzer ‘the best dancers’ *die in der Luft getanzten Schmetterlinge ‘the in the air danced butterflies’ Sie haben oft getanzt. Perfect of ‘They danced often.’

unaccusatives *Im Sommer wurde schnell gewachsen. Passive of ‘In the summer they grew fast.’ *die langsamsten Wachser ‘the slowest growers’ die am Haus gewachsenen Pflanzen ‘the at the house grown plants’ Sie sind schnell gewachsen. Perfect of ‘They grew fast.’

As (15) illustrates, there are also verbs that undergo passive but select for the auxiliary sein ‘be’, and (16) shows verbs that do not undergo passive but select for the auxiliary haben ‘have’. Aufstehen ‘stand up’ is both agentive and telic, whereas liegen ‘lie’ is both nonagentive and static. (15) [+control,+telic] intransitives in German a. Von niemandem wurde aufgestanden. passive of ‘Nobody stood up.’

72

Dieter Wunderlich

b. Er ist schnell aufgestanden. perfect of ‘He stood up fast.’ (16) [–control,–telic] intransitives in German a. *Von niemandem wurde im Bett gelegen. passive of ‘Nobody lied in the bed.’ b. Er hat im Bett gelegen. perfect of ‘He lied in the bed.’ These examples suggest that the possibility of passive depends on the feature [+contr], whereas the auxiliary sein ‘be’ is selected if the feature [+telic] is present. (Again, there may be idiomatic verbs for which no prediction can be made on semantic grounds.) If that is true, the set of the four criteria would be characteristic for the four classes in (12) rather than the two classes in (13), which have been suggested by syntacticians. Since ernominalization patterns with the possibility of passive, and the possibility of an attributive participle patterns with the choice of sein, no more than four classes need to be assumed. These four classes turn out to be determined by semantic rather than syntactic criteria, as Kaufmann (1995a) has pointed out. 3.2. The relationship between intransitives and transitives Many intransitive verbs can be transitivized, and many transitive verbs can be intransitivized. Some of these shifts have no overt marking on the verb. But many languages have a transitivization marker, and often one finds two (or even more) of these markers: the causative, which adds an agent functioning as causer, and the applicative, which adds an object of some sort (a resultative object, a beneficiary, a possessor, a location or even an instrument). Transitive verbs formed from intransitive ones are mostly prototypical in the sense of section 2, and those that undergo intransitivization (overtly marked or not) again are mostly prototypical transitives. Thus, the property of control seems to play a crucial role: [+control] marks the presence of an agent, and [–control] marks the presence of an (affected) object. The relationship between transitives and intransitives can then be systematized as in (17); this relationship is overwhelmingly symmetric.

Towards a structural typology of verb classes

73

(17) Intransitive-transitive shifts +control intransitive  transitive


x VERB(x) 
y
x VERB(x,y)

–control
y 
y
x

VERB(y) VERB(x,y)

The four types of shift represented in (17) are the following:6 (a) (b) (c) (d)

[+control] intransitives add a lower argument, a so-called ‘cognate’ or ‘internal’ object (dream a nightmare, dance Tango); [–control] intransitives add a higher argument, an agent or causer (dry the shirts, gallop a horse); transitive verbs with subject-related meaning allow the existential binding of the object (object deletion or antipassive), which yields a [+control] intransitive (he was eating, I can see again); transitive verbs with object-related meaning allow the absence of the subject (subject deletion or anticausative/middle), which yields a [–control] intransitive (break, ring).

(18) illustrates these possibilities with examples from German. (18) a. b. c. d.

intransitive Sie fuhr schnell. ‘She drove fast.’ Die Wäsche trocknete. ‘The shirts dried.’ Sie aß langsam. ‘She ate slowly.’ Die Glocken läuteten. ‘The bells were ringing.’

 



transitive Sie fuhr den Fiat schnell. ‘She drove the Fiat fast.’ Sie trocknete die Wäsche. ‘She dried the shirts.’ Sie aß die Suppe langsam. ‘She ate the soup slowly.’ Sie läutete die Glocken. ‘She rung the bells.’

For [–control] verbs it is not always clear whether the basic verb is intransitive or transitive, i.e., whether it belongs to class (b) or (d). Intransitive trocknen ‘dry’ is a +telic verb derived from the adjective trocken ‘dry’; it ———–————————————————————————————

6

The relationship between items that undergo the [-control] shifts has been studied cross-linguistically by Haspelmath (1993) and Comrie (2006).

74

Dieter Wunderlich

denotes a process possible without human instigation, and can therefore classified as a basic intransitive. Verbs of this class can easily be transitivized, which is the background for considering them to have an ‘underlying object’ – the subject of the intransitive verb becomes object of the transitive one; in other words, the transitive verb has an object-related meaning. In contrast, läuten ‘ring’, although it also derives from an adjective (laut ‘loud’), is [–telic]; it became idiomatic (for ringing bells) in the history of German and was already transitive in Old High German; since it denotes a process that usually needs a human instigator it is classified as basic transitive. (This is not necessarily true for English ring.) A further possibility not considered so far is that the two variants coexist, forming a hybrid (ambitransitive) verb. This can especially be expected for deadjectival verbs. From adjectives both intransitive and transitive verbs are derivable, and only conceptual factors determine whether the verb is preferred as intransitive (welken ‘fade’, reifen ‘ripe’) or transitive (leeren ‘empty’, öffnen ‘open’); nothing excludes that both options are equally possible (enlarge, widen, cool off). Such a hybrid pattern for deadjectival verbs could have been generalized to cover also underived verbs such as brechen ‘break’, which also has both options. For the transitive variants of basically intransitive verbs it is evident that they are lexically decomposed. Class-a transitive verbs, derived from an agentive intransitive (unergative) verb, not only add a cognate object but also some predicate that licenses this object. The predicate COGNATE in (19a) allows participants that have an internal relationship to the event denoted by the base predicate; e.g., possible cognates of DRIVE are vehicles (drive a car), transported persons/objects (drive the minister/the luggage), or distances (drive a mile). Similarly, class-b transitive verbs, alternating with a nonagentive intransitive (unaccusative) verb, have the predicate ACT (denoting some unspecific action) to license the causer/agent. (19) The decomposition of transitive verbs derived from intransitives a. fahren ‘drive’: y x+contr {DRIVE(x) & COGNATE(y)} b. trocknen ‘dry’: y–contr x {ACT(x) & BECOME DRY(y)} Most transitive verbs can undergo passivization, which binds the subject existentially (be eaten, be broken), as well as certain variants of the middle, which suppresses the subject (e.g. by inherent reflexivization, where the subject is identified with the object (wash, dress)). Derived transitives nearly always allow passivization, as shown in (20). With respect to the

Towards a structural typology of verb classes

75

middle (21), German and English differ: German usually marks this operation by an overt reflexive, while English does not. (English does not have lexical reflexives such as sich schämen ‘be ashamed’, sich fürchten ‘fear’.) (20) Passive of derived transitives of German and English a. Das Auto wurde schnell gefahren. ‘The car was driven fast.’ y
x {DRIVE(x) & COGNATE(y)}

b. Die Wäsche wurde im Garten getrocknet. ‘The shirts were dried in the garden.’ y
x {ACT(x) & BECOME DRY(y)}

(21) Middle a. Das Auto fuhr sich [REFL] gut. ‘The car drove well.’ b. Er kleidete sich [REFL] schick. ‘He dressed trendy.’ Cross-linguistically, the relationship between intransitives and transitives is subject to much variation. Some languages mark all transitive verbs by means of a transitivization affix, other languages classify into basic intransitives and basic transitives but always mark a shift. There are also languages that mark a shift only in a few instances (like German in the inherent reflexives). At the extreme end, there would be languages that never mark a shift. Near to this extreme is Basque, which never marks intransitivization, as illustrated in (22b). The intransitive use of ‘clean’ has two readings: the passive reading (i) and the inherent reflexive reading (ii), which are represented here by a semantic formula. (22) Intransitive readings in Basque (Joppen & Wunderlich 1995: 144) a. Jon-ek kotxe-a garbi-tu du. John-ERG car-DET clean-PERF AUX.3N.3sgE ‘John cleaned the car.’ y x {ACT(x) & BEC CLEAN(y)} b. Garbi-tu da. clean-PERF AUX.3N i. ‘He/it was cleaned.’ y
x {ACT(x) & BEC CLEAN(y)} ii. ‘He cleaned himself.’ x {ACT(x) & BEC CLEAN(x)}

76

Dieter Wunderlich

Basque has also a class of hybrid verbs that shift between intransitive and transitive readings without any marking (including igo ‘move.up/take.up’, jais ‘move.down/take.down’, hil ‘die/kill.PERF’, among others). (23) Intransitive-transitive hybrids in Basque (Joppen & Wunderlich 1995: 143) a. Jon igo-tzen da. John move.up-IMPF AUX.3N ‘John moves up.’ b. Jon-ek maleta-k igo-tzen ditu. John-ERG suitcase-pl take.up-IMPF AUX.3plN.3sE ‘John takes the suitcases up.’ However, Basque also has developed the causative morpheme -eraz (from the verb eragin ‘make/let’), which is productively used with all types of verbs. For the hybrid class of verbs one can assume that the transitive member is causativized without marking, probably because these verbs are historically old. If these verbs are used intransitively, they receive the two intransitive readings available for all transitive verbs (as demonstrated in (22b)), and furthermore, a third ‘anticausative’ reading because of their nature as hybrids (iii). (24) Three intransitive readings with Basque hybrids (Joppen & Wunderlich 1995: 144) Itziar hil da. Itziar die/kill.PERF 3N.AUX i. ‘Someone has killed Itziar.’ y
x {ACT(x) & BEC DEAD(y)} ii. ‘Itziar has killed herself.’ x {ACT(x) & BEC DEAD(x)} iii. ‘Itziar has died.’ y BEC DEAD(y)7 Thus, the lack of an overt device necessarily produces ambiguity, which has to be resolved by the respective context.

———–———————————————————————————— 7

It is in dispute how this middle reading is derived (see Kaufmann 2004).

Towards a structural typology of verb classes

77

3.3. On the decomposition of transitive verbs As we have seen above, transitive verbs derived from intransitive ones (or from intransitive adjectives) can be decomposed into at least two predicates, one that characterizes the higher argument (the subject), and another that characterizes the lower argument (the object). The question is whether a similar decomposition is also possible for basic transitives. There are indeed two classes of verbs where this seems feasible, the production and the annihilation verbs. Production verbs denote an event in which something comes into existence, while annihilation verbs denote an event in which something vanishes. Thus, ‘bake a cake’ could be analyzed that a cake comes into existence by the activity of baking, and ‘eat a cake’ – that a cake disappears by the activity of eating. (The naive ontology involved in language does not necessarily conform to the law of mass preservation!) This proposal is represented in (25). (25) Production and annihilation verbs a. bake: y x {BAKE(x) & BECOME EXIST(y)} b. eat: y x {EAT(x) & BECOME NON-EXIST(y)} However, nothing in the grammatical behavior of these transitive verbs hinges upon such a decomposition; one could as well consider bake and eat to be non-decomposed. Moreover, there are many transitive verbs (such as love, hit) for which a decomposition doesn’t make much sense. Therefore, for the time being, I assume that basic transitive predicates (such as LOVE, HIT) do exist. I find Kratzer’s (1996) proposal that transitive verbs are basically intransitive and add an agent only by means of active voice not particularly illuminating (for more specific points, see Wunderlich 2001).

3.4. The active-inactive type of argument linking The semantic classification of intransitive verbs can be transferred to transitive verbs, considering that the subject usually is the controller and the object the affectee. Various indigenous languages of the Americas have two sets of pronouns or pronominal affixes to be attached to the verb: items of one set are used to mark a 1st or 2nd person transitive subject, while items of the other set are used to mark a 1st or 2nd person object. (The 3rd person usually is unmarked, except in the plural.) If it comes marking the only ar-

78

Dieter Wunderlich

gument of an intransitive verb, items of both sets can be used, depending on whether it is more similar to the subject or to the object of a transitive verb. These markers can neither be said to be accusative nor to be ergative because there is a clear split in the intransitives and the definition given in (6) of section 2 fails. Therefore, this type of argument linking is assumed to reflect the active or inactive status of an argument rather than its position in the argument hierarchy.8 For illustration, (26) represents the full set of personal affixes of Dakota, a group of Siouan dialects spoken in North and South Dakota as well as in other regions in the US and Canada. The prefixes for the 1st and 2nd person show the relevant distinction, whereas no distinction is made in the dual ‘I and you (sg)’. There are three further affixes: wicha-, which is only used for animate 3pl objects (and therefore can truly be classified as accusative), -pi, which is separated from person and encodes plural, and the portmanteau prefix chi- to be used for the special communicative situation 1 2. (26) Personal affixes of Dakota (Schwartz 1979) active [+act]

inactive [–act]

accusative [+hr]

1

wa-

ma-

—

2

ya-

ni-

—

u) (k)-

1dl.incl 3pl.anim

—

— —

wicha-

-pi : Plural [+pl] chi- : Portmanteau 1 2

The use of these affixes is demonstrated in (27) to (29). It should become clear from these examples that the prefix ma- ‘1’ can mark the subject of an intransitive verb, the patient (object) of a transitive verb, and the recipient ———–————————————————————————————

8

Actually, one needs more evidence for the conclusion that one is dealing with the active type. Choctaw, a Muskogean language of Oklahoma and Mississippi, has a similar split in the intransitives. Nevertheless, the markers on the Choctaw verb are classified as accusative (applied to inactive intransitives) and nominative (applied to active intransitives) because there exists a third set of markers, namely dative (Davies 1986). Dakota and the other languages subsumed under the active type do not have such a third set of markers. For the time being, I assume that the accusative-marked intransitives of Choctaw are lexically marked (see section 5 and footnote 13 below).

Towards a structural typology of verb classes

79

of a ditransitive verb. One can conclude that ma- encodes the inactive status of the 1st person rather than its semantic role (as patient or recipient) or its grammatical function (as subject or object). (27) Intransitive verbs in Dakota a. wa-niwa ‘I swim’ wa-niwa-pi ‘we swim’ b. ma-t’a ma-t’a-pi

‘I die’ ‘we die’

(Active) (Inactive)

(28) Transitive verbs in Dakota a. wa-kte ‘I killed him’ h wic a-wa-kte ‘I killed them’ b. ma-kte ma-kte-pi

‘he killed me’ ‘they killed me’/‘he killed us’

c. ma-ya-kte chi-kte

‘you killed me’ ‘I killed you’

(Active) (Inactive) (Mixed)

(29) Ditransitive verbs in Dakota a. wa-k’u ‘I gave it to him’ wicha-wa-k’u ‘I gave it to them’ b. ma-k’u ma-k’u-pi

‘he gave it to me’ ‘they gave it to me’/’he gave it to us’

c. ma-ya-k’u chi-k’u

‘you gave it to me’ ‘I gave it to you’

(Active) (Inactive) (Mixed)

As one can see from these examples, the argument role of a 1st or 2nd person is always made explicit; thus, the main task of an argument linking device is fulfilled. (The role of a 3rd person singular, however, must be specified by additional syntactic nominals.) One clear disadvantage of this type of argument linking is the necessity of classifying intransitive verbs according to their semantic properties. What are the exact conditions under which a verb is classified into the active class? For a number of active type languages, Mithun (1991) investigated the semantic factors for the active/non-active classification of intransitive verbs. Some of her results are summarized in (30): active ([+act]) is the marked option for dynamic verbs in Guaraní (a Tupi-Guarani language

80

Dieter Wunderlich

spoken in Argentina, Bolivia and Paraguay), for verbs with a participant which instigates the situation denoted by the verb (Lakhota) or controls it (Caddo in Western Oklahoma), or for an inherently imperfective verb with an agent argument in a broad sense (Mohawk, an Iroquoian language of Canada). Only in Central Pomo (which belongs to the Pomoan family), [+affected] of human beings (corresponding to [–act]) is the marked option. The complementary class always contains all other verbs (including the borderline cases), and is thus the default class. Inherently stative verbs are [–act] in Guaraní, Lakhota, Caddo and Mohawk, but [–aff] (corresponding to [+act]) in Central Pomo. (30) a. b. c. d.

Guaraní: Lakhota: Caddo: Mohawk:

dynamic [+act] vs. stative [–act] instigator [+act] vs. non-instigator [–act] controller [+act] vs. non-controller [–act] agent & non-perfect [+act] vs. non-agent or perfect [–act] e. Central Pomo: patient [+aff] vs. non-patient [–aff]

This variation demonstrates the difficulty of supporting a split required for grammatical reasons by a semantic partition. As a reaction to this problem the intransitive split is often lexicalized in the course of historical development, so that the decision becomes independent of semantic factors. This is observed for Cherokee (an Iroquoin language) by Scancarelli (1987). (Similar observations have been made for the choice of sein ‘be’ as the perfect auxiliary in German, which is often lexicalized.) On the other hand, a semantic split allows us to distinguish between the possible readings of an intransitive verb, as shown in (31) for Lakhota. In quite a similar way the choice of the German perfect auxiliary sometimes distinguishes between possible readings of a verb, as shown for the same semantic contrast in (32). (31) Different semantic readings for an intransitive verb in Lakhota (Foley & van Valin 1984: 96) a. wí cexélka b. há: cexélka 1sg.ACT slip/slide 1sg.INACT slip/slide ‘I’m slipping.’ ‘I’m sliding.’

Towards a structural typology of verb classes

81

(32) Different semantic readings for an intransitive verb in German a. Er ist gerutscht. b. Er hat gerutscht. he is slipped/slid he has slipped/slid ‘He was slipping.’ ‘He was sliding.’ Furthermore, if the feature [+act] (or conversely, [+aff]) is semantically interpreted, it becomes possible to mark exceptions from canonical transitive verbs. For instance, transitive experiencer verbs of Lakhota can be marked by means of double-inactive, as shown in (33). (33) Double-inactive in Lakhota (Mithun 1991) i-ni-ma-ta ?-2.INACT-1.INACT-PROUD ‘I am proud of you’ For Central Pomo not only double-inactive verbs (34a) but also doubleactive verbs are reported; with regard to the latter one has to assume that no participant is affected (34b). (34) Double-inactive and double-active in Central Pomo (Mithun 1991) / yáqan? a. to=wa mto 1.INACT =Q 2.INACT remember ‘Do you remember me?’ / b. mul /a yáq-an-ka-w chó-w. 3.ACT 1.ACT remember-IMPF-CAUS-PERF not-PERF ‘I couldn’t think of him.’ These observations suggest that the semantic reading of an active vs. inactive marker is the relevant factor. However, Central Pomo also allows a person-related split in the class of emotional verbs where the 1st person (higher in its person status) is always marked as affected (i.e. inactive), regardless of whether this person is the source or the target of emotion (35a,b). (35) Person-related split in Central Pomo (Mithun 1991) / údaw. a. Mul qadála to 3.ACT hate= IMMEDIATE 1.INACT really ‘I [+aff] really hate him [–aff]’

82

Dieter Wunderlich

b. To qadál-m-ad=a mul. 1.INACT hate-NON.EMPATHETIC-IMPF=IMMEDIATE 3.ACT ‘He [–aff] hates me [+aff]’ Clearly, a 3rd person could as much get affected as a 1st person. It seems therefore more appropriate to consider the hierarchy active > inactive as a purely grammatical scale which can be associated with any arbitrary semantic scale, so that something like the person-related split of Central Pomo might have evolved. This conclusion is supported by the observation that the ‘different subject’ (DS) marking of Lakhota not only reflects differences in the reference to persons but also in their active/inactive status. (36a) illustrates coordination by means of the ‘same subject’ marker -y; here, one and the same person is active in both events. In (36b) the DS marker -qan encodes that two different persons are involved, whereas in (36c) it encodes that a single person is involved in two different activity states. (36) Coordination in Lakhota (Foley & Van Valin 1984: 119) a. Há: káluhu-y si:má:mérqaki:hi. 1sg.ACT go-home-SS went.to.bed ‘I went home and then went to bed.’ b. Há: káluhu-qan mí:p mérqaki:hi. 1sg.ACT go-home-DS 3sg.ACT went.to.bed ‘I went home and he went to bed.’ c. Há: xá: qákki-qan wi qa:lál tá:la. 1sg.ACT water bathe-DS 1sg.INACT sick become ‘I took a bath and got sick.’ 4.

Further argument linking types in transitive verbs

In this chapter, I introduce more basic types of argument linking. The inverse type (section 4.1) encodes whether the subject or the object is higher on a person scale than the respective other argument; that is, two independent scales are aligned, which makes it possible that in principle only one set of person affixes is needed. In contrast, in the salience type (section 4.2) several sets of person affixes are used, one for each semantic or grammatical role if it is the most salient; this type exhibits certain similarities with the

Towards a structural typology of verb classes

83

voice type, and therefore I put them together. In the positional type (section 4.3) syntactic positions are grammaticalized for encoding argument roles, and no person affixes are needed. Finally, by far the most widespread is the generalized case type (section 4.4), which comprises several varieties of accusative and ergative marking systems. Implicit in the distinction of these types is the claim that they can neither be reduced to each other nor regarded as subtypes of a more comprehensive or abstract type. Only the positional type could in some sense be related to the accusative type. The salience and voice types, which are still poorly defined, could turn out as subclasses of a single type or as a collection of several distinct types. Presently no one can make a survey which is close to being complete. 4.1. The inverse type The inverse type is best-known from the Algonquian languages, which stretch from Labrador south into present North Carolina and west across the Plains into Alberta and Montana; Plains Cree, the language with which we will illustrate this type, belongs to a group of Canadian dialects stretching westwards. Most remarkable for Algonquian is the way in which the roles of subject and object are encoded in the class of transitive animate (TA) verbs in which both arguments are animate. Consider (37a) with the reading ‘we see the dog’, and (37b) with the inverted reading ‘the dog sees us’: this difference is triggered by the theme (or voice) markers that directly follow the stem, whereas the person-number morphemes are not altered. /-a/ is called a ‘direct’ morpheme (triggering that the higher person is subject), and /-ikw/ is called an ‘inverse’ morpheme (where the higher person is object). In contrast, transitive inanimate stems (TI) do not undergo such an alternation because an inanimate entity cannot become subject of ‘see’, see (37c). (37) Transitive verbs in Plains Cree (Wolfart 1973; Wolfart & Carroll 1981; Dahlstrom 1991; Fabri 1996) a. Ni-wap-am-a-nan atim. 1- see-TA-DIR-1pl dog ‘We see the dog.’ b. Ni-wap-am-iko-nan atim. 1- see-TA-INV-1pl dog ‘The dog sees us.’

84

Dieter Wunderlich

c. Ni-wap-aht-e-naan. 1-see-TI-LOC.PERS-1pl ‘We see it.’ A first observation concerns the central role of animacy. Plains Cree classifies transitive verbs according to whether the object is animate or inanimate by means of derivational suffixes. Similarly, intransitive verbs are classified according to whether their subject is animate (IA) or inanimate (II). (38) shows some derived stems of Ojibwe (a related Algonquian language), which are based on the adjectival root biin ‘clean’. (38) Some derived stems in Ojibwe (Valentine 2001) animate

inanimate

Intransitives: the subject is …

biin-zi (IA)

biin-ad (II)

‘be clean’

Transitives: the object is …

biin-ih (TA)

biin-toon (TI)

‘make clean’

These four classes of verbs are differently inflected insofar as only animates can be 1st or 2nd person. There is only one set of person-number affixes available, and these affixes always fill the same positions. The TA verbs therefore need a further device for determining whether an affix relates to subject or object. The inflected TA verbs show a strong person hierarchy at work. In the prefixes, 2nd person has preference over 1st person, as attested by the examples in (39) in which 1st and 2nd person are combined (1st inclusive person refers to ‘I and you’); furthermore, both local persons (1st and 2nd) have preference over 3rd person. This suggests the scale 2 > 1 > 3. (39) Preference of 2nd person in Plains Cree a. Ki-tasam-i-n. b. Ki-tasam-iti-n. c. Ki-tasam-i-nan. 2-feed-INV-1 2-feed-DIR-1pl 2-feed-DIR-1 ‘You feed me.’ ‘I feed you.’ ‘You(sg/pl) feed us.’ d. Ki-tasam-iti-nan. e. Ki-tasam-a-naw. 2-feed-INV-1pl 2-feed-DIR-1.incl ‘We feed you(sg/pl).’ ‘Weincl feed him.’ In the suffixes, however, plural of the 1st person is preferred over plural of the 2nd person (pl/1 > pl/2), as shown by (39c,d).

Towards a structural typology of verb classes

85

In addition, all 3rd person arguments must be ranked to each other. Only one proximate 3rd person (the foregrounded one) is possible in a certain stretch of discourse, all other 3rd persons must be obviative (sometimes called 4th person) and cannot be marked for plural. Obviative must be marked on nouns or noun phrases, so that one gets informed about the obviation status of each nominal referent. It is sufficient, however, to use only verb forms, and as the examples in (40) show, the theme markers indicate whether the proximate person is subject or object. The use of /-e/ and /-ikw/ in (40) is consistent with the ranking 3 > obv. (40) Obviative in Plains Cree a. wapam-e(w)-ak see-DIR-3pl ‘They see himobv.’

b. wapam-ikw-ak see-INV-3pl ‘Heobv sees them.’

As has become evident, the grammatical function of the person-number morphemes in TA verbs wholly depends on the theme (voice) markers attached directly to the stem. The set of these markers in the independent mode (characterizing unembedded sentences) is represented in (41); all instances below the shaded diagonal encode direct constellations (where the subject is higher on the person scale than the object), whereas all instances above it encode inverse constellations (where the subject is lower than the object). The person hierarchy is extended here to 2 > 1 > 3 > obv > inanimate, thus including further variants of the 3rd person. It is never possible that subject and object have an equal person status; reflexives in which subject and object are identical are realized by intransitive verbs. (41) Theme markers in Plains Cree (independent forms) O\S

2

1

2 1

3

obv

inanimate

-iti -ikw

-i

3

-a

obv

-im-a

inanim

-e

-e

more forms

To account for the function of the theme markers formally, one can introduce the feature [+hp] ‘there is a higher person’, which expresses relative salience analogously to the feature [+hr] ‘there is higher role’. The theme

86

Dieter Wunderlich

marker then associates this feature with one of the argument roles. More precisely, a direct marker associates it with the lower role, and an inverse marker associates it with the higher role. This is shown in (42). (42) a. direct

-a:

b. inverse y x VERB (x,y) +hr –hr +hp 3 1

y x VERB (x,y) +hr –hr -ikw: +hp 1 3

The last line indicates a possible distribution of person values. The direct voice is compatible with the 1st person assigned to the higher role (the subject), whereas the inverse voice requires the 1st person to be assigned to the lower role (the object). In principle it would be possible to have only one direct and one inverse marker. Actually, however, there are several markers. They can be represented as partly generalized portmanteau morphemes bearing information about both arguments, as shown in (43); only -ikw is a fully generalized inverse morpheme. (43) Theme markers direct

- -i -

1/2 3 2 1 3 obv

+hp/+hr +hp,+1/+hr +hp,+obv/+hr

inverse

-ikw -iti

INV

1 2

+hp/–hr +hp,+1/–hr

Indeed, theme markers (which classify verbs according to the semantic values of their arguments) may have been derived from former portmanteau morphemes, thereby regularizing the inflectional patterns. Dahlstrom (1988) reports that the conjunct (dependent) forms of Plains Cree underwent a rapid change during the second half of the 19th century. (The conjunct forms lack a person prefix, but instead have sometimes an initial conjunct marker, - or k-.) Presently they exhibit the same theme markers as the independent forms (except some combinations with 1sg and 2sg, which lack a theme marker), but that was not always the case: they involved much more irregular portmanteaus in the 1855 translation of the Gospel according to St. John.

Towards a structural typology of verb classes

87

The conjunct forms are interesting also for another point. Since they do not have a position for a person prefix, ambiguities would arise if only were the suffixes known from the independent forms available. The conjunct forms indeed exhibit more person-number suffixes, and some of them (in forms without a theme marker) are truly portmanteau morphemes, as illustrated in (44). (44) Singular conjunct forms in Plains Cree a. -wpam-at b. -wpam-isk CONJ-see-2sg 3sg CONJ-see-3sg 2sg ‘that you saw him’ ‘that he saw you’ In principle it is possible that portmanteau forms collapse, for instance, 3 2sg and 1 2sg could be expressed by the same suffix; such a suffix would have to be classified as 2sg.accusative. The sporadic emergence of accusative morphemes indeed has been observed, although not in Cree. But in the conjunct mood of Ojibwe (dialect of Manitoulin Island, Lake Huron) the two accusative morphemes -i (1.ACC) and -inin (2.ACC) have developed (Valentine 2001). These morphemes, interesting as they are, must nevertheless be judged as sporadic because similar morphemes have already been attributed to Proto-Algonquian by Goddard (1979). The system as a whole was and is robust enough to reject further generalization to accusative. As a whole, the morphology of the inverse type can be described as symmetric. In every use of a TA verb the two arguments must have a different person status so that the lexical asymmetry of transitive verbs is outbalanced by the person hierarchy. Disregarding the theme markers themselves, no asymmetry remains: each person-number affix can relate to either subject or object. If two nouns are expressed in a clause (such as ‘man’ and ‘duck’ in (45)), one must be marked as obviative. Together with the possibility of direct vs. inverse marking, this leads to four versions, illustrated in (45). (45) Four-way transitive alternation in Plains Cree (Wolfart 1981: 30) a. wpam-w npw siisiip-a. man duck-OBV see-DIR ‘The man sees the duck (obv)’ b. wpam-ik npw siisiip-a. see-INV man duck-OBV ‘The duck (obv) sees the man’ [‘The man is seen by the duck’]

88

Dieter Wunderlich

c. wpam-w npw-a siisiip. man-OBV duck see-DIR ‘The duck sees the man (obv)’ d. wpam-ik npw-a siisiip. see-INV man-OBV duck ‘The man (obv) sees the duck’

[‘The duck is seen by the man’]

Although passive might be the best translation equivalent of the inverse construction, the latter would wrongly be classified as ‘some kind of passive’. Besides the transitive inverse form, Cree has an impersonal passive, in which the TA stem is inflected intransitively, shown in (46). (46) wpam-aw npw / siisiip. see-3 man / duck ‘The man /the duck was seen.’ Moreover, the inverse construction does not alter the syntactic status of the two arguments. This conclusion is forced by the observation that Plains Cree does not show any syntactic subject-object asymmetry of the kind known from English or other case-based languages (Dahlstrom 1991; Bailin & Grafstein 1991; Wunderlich 2005). English exhibits so-called weak cross-over effects: for instance, a subject can bind the possessor of the object (47a), while the object cannot bind the possessor of the subject (47b); the intended meaning of (47b) can only be expressed by the passive in (47c). (47) Weak cross-over effects in English /Whichi woman loves heri son? a. Annai loves heri son. /*Whichi woman does heri son love? b. *Heri son loves Annai. c. Annai is loved by heri son. /Whichi woman is loved by heri son? No such restriction holds for Plains Cree. Not only can the subject bind the possessor of the object, but also the object can bind the possessor of the subject, as shown in (48b).9

———–———————————————————————————— 9

The possessor of a noun is considered to be more salient than the possessed; therefore, the possessed noun usually gets an obviative marking.

Towards a structural typology of verb classes

89

(48) No weak cross-over effects in Plains Cree (Dahlstrom 1991: 99) a. kahkiyaw iskww-ak skih--w-ak o-tnis-iww-a. all woman-pl love-DIR-3-pl 3P-daughter-3plP-OBV ‘All womeni love theiri daughters.’ b. kahkiyaw iskww-ak skih-ik-w-ak o-tnis-iww-a. all woman-pl love-INV-3-pl 3P-daughter-3plP-OBV ‘Theiri daughters love alli women.’ [All women are loved by their daughters.] This binding property is independent of position. The sentences in (49) are positional variants of (48b) but get the same reading. (49) Different positions of quantifier and nouns (Dahlstrom 1991: 99, 87). a. o-tnis-iww-a skih-ik-w-ak kahkiyaw iskww-ak. woman-pl 3P-daughter-3plP-OBV love-INV-3-pl all ‘Theiri daughters love alli women.’ [All women are loved by their daughters.] b. kahkiyaw skih-ik-w-ak o-tnis-iww-a iskww-ak. all love-INV-3-pl 3P-daughter-3plP-OBV woman-pl ‘Theiri daughters love alli women.’ [All women are loved by their daughters.] (49b) shows quantifier floating, with the quantifier in focus position before the verb. As Dahlstrom (1991) points out, quantifier floating in Plains Cree is only possible for object quantifiers – thus, ‘all woman’ is a true object. There are of course no logical restrictions for weak cross-over. Both formulas in (50), representing the meanings of (48a,b), are sound. (50) a. x y [WOMAN (x) & DAUGHTER.OF.x (y) LOVE (x,y)] b. x y [WOMAN (x) & DAUGHTER.OF.x (y) LOVE (y,x)] Nevertheless, (50b) cannot be expressed by an English active sentence. English, like many other languages, has acquired a syntactic notion of ‘subject’ and ‘object’, which is lacking in Plains Cree. Unlike English, ‘subject’ in Plains Cree always refers to the highest argument; neither case nor position can help us to identify the subject; only the theme marker determines which argument functions as subject. Therefore, disregarding all superficial differences, it is correct to say that inverse morphology represents a type of argument linking of its own, it neutralizes subject-object asymmetry and is

90

Dieter Wunderlich

therefore incompatible with case. A change into a case system is improbable, even if sporadic generalization to accusative may occur. We are unable to see Plains Cree as a mere ‘notational’ variant of English. Yet, inverse morphology is clearly a rare (typologically marked) option. It can emerge only under very specific circumstances, including head marking, a rigid classificatory system of verbs related to argument properties, a strong person hierarchy, and the existence of person-number morphemes unspecified for grammatical functions. The major classification of verbs, being forced by the inverse morphology, is into transitive verbs that have an animate object and those that have not. Correspondingly, ditransitive verbs formed by various derivational means (causative, benefactive, comitative) are always classified as transitive animate (TA) in virtue of the fact that they add an animate argument. As we have seen, inverse morphology (which grammaticalizes person hierarchy) also forces the introduction of a fourth person (obviative) because no two arguments can have the same person status. This all together indicates a strong interaction between the emergence of certain verb classes and the structural means of identifying the arguments of a verb unambiguously. 4.2. The salience or voice type In the active type (section 3.4) we found two sets of pronominal affixes (active and inactive), while the inverse type (section 4.1) exhibits only one set of pronominal affixes together with direct and inverse morphemes, and additional portmanteau morphemes that seem to undergo several changes. In contrast, the salience type exhibits several sets of pronominal affixes, together with the restriction that only one affix is possible at a verb – it encodes the most salient argument. This type is exemplified by Arizona Tewa, a Tanoan language spoken in a number of pueblos along the Rio Grande in New Mexico and Arizona. (51) Six sets of person prefixes in Arizona Tewa (Kroskrity 1985) a. STAT: encodes the argument of an intransitive verb b. REFL: identifies agent and patient of a transitive verb c. POSS: encodes a possessor related to the subject of an d. BEN: encodes a beneficiary argument of a transitive verb e. AGT: encodes the agent of a transitive verb f. PAT: encodes the patient of a transitive verb

Towards a structural typology of verb classes

91

STAT and REFL

are unproblematic because intransitives and reflexive transitives (such as ‘he washes (himself)’) only have one argument to be realized. POSS adds an argument to an intransitive verb (52a), while BEN adds an argument to a transitive verb (without any further marking of the verb) – these arguments count as the most salient ones because they mostly refer to a human person and are often in the focus of predication. AGT and PAT encode the person which is more salient, see (52b–f). 1st and 2nd person are more salient than the 3rd person, and ‘these girls’ in (52f) is more salient than ‘those boys’ because of the proximate-distal contrast. With a PAT marking, the respective less salient agent receives an oblique marking (52c,e,f), which resembles the obviative marking of Cree (section 4.1). (52) The one-prefix restriction of Arizona Tewa (Kroskrity 1985) a. semele dín-han pot 1sg.POSS-break ‘My pot broke’ b. he’i-n sen-en dó-khwE!di this-pl man-pl 1sg.AGT-hit ‘I hit these men’ c. he’i-n sen-en-di dí-khw E!di this-pl man-pl-OBL 1sg.PAT-hit ‘These men hit me’ d. Ne’i kwiyó ná:-tay this woman 2sg.AGT-know ‘You know this woman’ e. Ne’i kwiyó-di wó:-tay this woman-OBL 2sg.PAT-know ’This woman knows you’ f. nE!’i-n ‘ayú-n ’o:’i-n ’enú-n-di ’ó:bé- khw E!di this-pl girl-pl that-pl boy-pl-OBL 3pl.PAT-hit ‘Those boys hit these girls’ The prefixes of the AGT set cannot be identified with active in the sense of the active type because they do not occur with intransitive verbs; likewise, the prefixes of the PAT set cannot be identified with inactive. Neither can the elements of PAT be identified with accusative (or the elements of AGT with ergative) because there is no contrast to nominative, which would be required on the basis of the definitions given in (6), section 2. In principle,

92

Dieter Wunderlich

however, it is imaginable that one of these markers develops to accusative or ergative. A more detailed comparative study of the Kiowa-Tanoan languages seems promising, given the interesting fact that Kiowa represents an extreme portmanteau type (see section 1), but Arizona Tewa the salience type, and that both Tiwa and Towa, the two other Tanoan branches, have an inverse marker (according to Klaiman 1993). Similar in its function, but much more systematic, is the so-called voice system of the Philippine languages, which we illustrate below with Tagalog. In this language, only one argument can be encoded on the verb: it is usually understood as definite and most prominent syntactically (for instance, only this argument can be relativized). The marked argument can be the agent (actor voice AV), the patient (object voice OV), a location or recipient (LV), a beneficiary (BV) or an instrument (IV); the three latter voices can add an argument without any further marking on the verb – thus they have the combined effect of applicative+passive (except that no argument is demoted). Some of these markers are also found with intransitive verbs, which shows that they are sensitive to semantic roles. The noun or noun phrase that corresponds to the designated (the most prominent) argument is marked by the proclitic ang, which could be identified with nominative (pronouns and proper names have a different marker). All other structural arguments are marked by ng (pronounced as /nang/), which is called genitive (but could as well be called accusative), while locatives and recipients are marked by sa (possibly to be analyzed as dative). Event nominalizations, as well as the recent past, lack any voice marking, therefore, no nominative ang can appear in them. The choice of the voice morphemes is often lexically determined. All attempts to relate the Philippine voice system to either accusative or ergative have failed, and attempts to interpret the most prominent argument as topic or focus have failed, too. (There exists a different topic construction, and the tendency of being definite conflicts with the interpretation as focus.) What one finds are syntactic effects such as the possibility to get relativized. Similar systems are found in other Austronesian languages, too, partly reduced or reorganized. The name voice reminds us of active and passive, which also promote one of the arguments to the most prominent one. However, passive binds the agent existentially (it could only be expressed by an oblique adjunct, such as English (He was seen) by the policeman), while the corresponding object voice (OV) leaves the agent in a structural case. Moreover, a separate passive construction can coexist with the OV construction; Balinese, for instance, exhibits both (Arka 2003).

Towards a structural typology of verb classes

93

The examples in (53) illustrate the five voices of Tagalog with the verb bili ‘buy’; it should become clear that the voices neatly correspond to the semantic roles of the participants of a buying event. Thus, the voice affixes classify verbs according to their potential of bearing certain semantic roles; they cannot be considered to be pronominal affixes because they do not bear any information about person and number.10 (For phonological reasons, some of the prefixes surface as infixes; < > indicates such a prefix. In (53a), for example, /um -/ is infixed to the stem bili.) (53) Voices in Tagalog (Foley & Van Valin 1984: 135) a. Bili ang=lalake ng=isda ng=pera sa=tindahan. buy NOM=man GEN=fish GEN=money LOC=store ‘The man bought fish in the store with money.’ b. Bili ng=lalake ang=isda ng=pera sa=tindahan. buy GEN=man NOM=fish GEN=money LOC=store ‘The man bought the fish in the store with money.’ c. Bilh-an ng=lalake ng=isda ng=pera ang=tindahan. buy-LV GEN=man GEN=fish GEN=money NOM=store ‘The man bought fish in the store with money.’ d. I-bili ng=lalake ng=isda ang=bata. BV-buy GEN=man GEN=fish NOM=child ‘The man bought fish for the child.’ e. Ipan-bili ng=lalake ng=isda ang=pera sa=tindahan. IV-buy GEN=man GEN=fish NOM=money LOC=store ‘The man bought fish in the store with the money.’ The person from whom one buys something is also marked by the location clitic sa; correspondingly, it is advanced to the most prominent argument by the location voice (LV), as shown in (54b). (54) Source in Tagalog (Foley & Van Valin 1984: 63) a. Bili ng=isda sa=bata ang=lalake. -buy GEN=fish LOC=child NOM=man ‘The man bought some fish from the child.’ ———–———————————————————————————— 10

If one assumes that the voice markers originated from affixes that make one of the arguments the referential argument of a derived noun (Ross 2002; Stiebels, this vol.), one indeed wouldn’t expect them to bear person-number information.

94

Dieter Wunderlich

b. Bilh-an ng=lalake ng=isda ang=bata. buy-LV GEN man GEN fish NOM child ‘The man bought some fish from the child.’ With the verb bigay ‘give’, the clitic sa marks the recipient as a goal (55a,b). The examples in (55) also illustrate how relativization functions in Tagalog: the recipient can be relativized in the location voice (LV; 55a), and the patient can be relativized in the object voice (OV; 55b), but the patient cannot be relativized if the verb shows actor voice (AV; 55c). (55) Relative clauses in Tagalog (Foley & Van Valin 1984:141) a. bata-ng bigy-an ng=lalake ng=isda child-COMP give-LV GEN=man GEN= fish ‘the child which was given fish by the man’ b. isda-ng i-bigay ng=lalake sa=bata fish-COMP OV-give GEN=man LOC=child ‘the fish which was given to the child by the man’ c. *isda-ng nag-bigay ang=lalake sa=bata fish-COMP PERF.AV-give NOM=man LOC=child The common properties of voice systems seem to be the following: only one argument can be encoded on the verb, where the marker reflects the semantic role, and, possibly, the semantic value (person or definiteness) of this argument. However, unlike passive, no other argument gets demoted (which is particularly true for Tagalog). Arizona Tewa fuses voice with person-number, whereas Tagalog does not. In Tagalog, the argument that is encoded on the verb achieves a special grammatical status, i.e., functions as syntactic ‘pivot’. Regrettably, the information about Arizona Tewa is too poor for evaluating whether the designated argument gets a similar syntactic function. It is therefore not yet possible to describe the general profile of the voice type more precisely. It might turn out in the end that Arizona Tewa and Tagalog belong to different types. 4.3. The positional type The positional type is found in various languages, particularly in English, Chinese, West African and Oceanic languages, but also in other parts of the world, and it is often combined with some residual agreement. Recall that

Towards a structural typology of verb classes

95

the default ordering of arguments usually follows the argument hierarchy, with the highest argument in the first position and the lowest argument in the last position (independently of where the verb is positioned). It is this default ordering that is fixed in the positional type. Argument position can also depend on the information status of arguments. Topics, which express something which has already been given in the context, are mostly realized initially, sometimes followed by the focus element, which expresses the new thing – but there are many means to express focus differently, whereas topics usually are preferred in the first position. Under the condition that arguments should be recognizable even if of one of them is topicalized, the order SVO turns out to be the most robust type. If the object is topicalized, one gets the order OTSV, which clearly differs from SVO because two nominal arguments precede the verb. Thus, the roles of S and O can always be distinguished, as illustrated in the Chinese example (56). (56) SVO and object topicalization in Chinese a. wo kan-le shu le. I read-ASP book ASP ‘I read books.’ b. shu, wo kan-le. book I read-ASP ‘The book(s), I read.’ The SOV type (verb-final) would yield the ordering OTSV as well, which, however, produces ambiguity because the first nominal could be S or O. Therefore, SOV type languages usually use morphological case rather than position to distinguish S and O. The verb-initial languages often have VOS as the basic order, but in this case no clear distinction is possible between OTVS and STVO; one can only recognize that one of the arguments is topicalized but doesn’t know which one; ambiguity is thus unavoidable. It is therefore reasonable that the positional type is restricted to SVO languages. The subject is realized pre-verbally, and the object in the position that immediately follows the verb; this position can be considered a variant of accusative marking. The SVO type has a further advantage: one can form series of predications with a shared subject such as S(V1O1)(V2O2) … – such a series is called a serial verb construction SVC (see section 6.3). One of the verbs of an SVC can be grammaticalized for a specific function. As Peyraube (1996)

96

Dieter Wunderlich

points out, already the earliest Chinese inscriptions (more than 3,000 years old) have SVO together with SVOV+complement and fixed elements (‘prepositions’) that appear in either the first or the second V-position. Modern Chinese shows constructions such as those in (57): S ba O1VO2, and SVO1 gei O2. Here, ba (in one of its several functions) marks a possessor of O2, and gei (formerly ‘give’) marks a goal. (‘CL’ stands for classifier.) (57) ba and gei in Chinese (Sybesma 1999: 137, 105) a. Li Si ba Lao Li duan-le tui. Li Si BA Lao Li break-ASP leg ‘Li Si broke Lao Li’s legs.’ b. Zhang San song yi-ben shu gei Li Si. Zhang San give one-CL book to Li Si ‘Zhang San gave a book to Li Si.’ Obviously, the main verb in both (57a) and (57b) has three arguments. We will see in section 6.3 that the serial verb construction is one of the possible ways to express a third argument in the positional type. Another way is the double object construction to be discussed in section 6.4; it is also found in Chinese (58a). The observation that gei ‘to’ can optionally appear in this construction (58b) and also occurs in the ba-variant (58c) leads Sybesma (1999) to the conclusion that all realizations of Chinese song ‘give’ go back to an underlying change of location (rather than a change of possession) structure (see also sections 6.1 and 6.6). (58) Further variants of expressing ‘give’ in Chinese (Sybesma 1999: 100ff., 136) a. Zhang San song Li Si yi-ben shu. Zhang San give Li Si one-CL book ‘Zhang San gave Li Si a book.’ b. Zhang San song gei Li Si yi-ben shu. c. Zhang San ba zhei-ben shu song gei Li Si. Zhang San BA this-CL book give to Li Si ‘Zhang San gave this book to Li Si.’ These examples indicate that the SVO type can pave the way to more complex constructions, including the possibility of introducing prepositions for the marking of arguments. It is even conceivable that such a preposition

Towards a structural typology of verb classes

97

develops to a case marker. Chinese ba seems to function as an accusative marker in some instances (such as (59a)), while in other instances (59b,c) it is not so clear whether this is true. (59) ba as a possible accusative marker in Chinese (Sybesma 1999: 137ff.) a. Ta ba zhu mai-le. he BA pig sell-ASP ‘He sold the pigs.’ b. Ta ba huaping cha-le yi-ba hua. bunch he BA vase stick-ASP one-CL flower ‘He stuck a handful of flowers into the vase.’ c. Ba ge zei pao-le. BA CL thief escape-ASP ‘A thief escaped.’/ ‘They had a thief escape.’ 4.4. The generalized case type Let us finally turn to the most frequent type of argument linking, which is so wide-spread that many linguists take it to be the universal type. In any case, it is central for all theories about argument structure. In systems of this type the fundamental lexical asymmetry of transitive verbs is preserved in the morphology and syntax. In section 2 I introduced the notions ‘accusative’ and ‘ergative’: a marker is called ‘accusative’ if it encodes objects (patients) differently from agents and intransitive subjects; it is called ‘ergative’ if it encodes transitive subjects (agents) differently from objects and intransitive subjects. Both ‘accusative’ and ‘ergative’ are contrasted with ‘nominative’. I also introduced the two features [+hr] and [+lr] that encode the argument hierarchy and, simultaneously, characterize the two alternatives: accusative marks [+hr] argument roles, and ergative marks [+lr] argument roles, whereas nominative (bearing no feature specification) is compatible with any argument role. Furthermore, I argued that marking the lower role is preferred over marking the higher role, which results in the preference scale +hr > +lr (for convenience: ACC > ERG). An argument linking system that conforms to these conditions is called a ‘generalized case’ system. It can be accusative-based, ergative-based, or both (i.e., a mixed system). Moreover, there can be different ways to spell out the accusative or ergative property: The verb can be combined with

98

Dieter Wunderlich

pronominal affixes or agreement morphemes that belong to an ACC or ERG set (head-marking), or the nominal arguments (pronouns or full NPs) themselves are combined with particles (adpositions) or morphological case (dependent-marking). The notion of generalized case thus includes both the flagging of the verb and the flagging of syntactic complements (where ‘flagging’ can be done by certain kinds of affixes, clitics or particles). It is even possible to subsume the positional SVO type under generalized case: the position to the right of the verb is reserved for objects; it can thus be characterized as accusative (bearing the feature [+hr]). 4.4.1. Head- vs. dependent-marking In the types discussed in the preceding sections (the active type, the inverse type, and the salience type, but not in the Philippine voice type), the arguments are marked on the verb by means of pronominal affixes; such a mechanism is called head-marking. There can be two sets of pronominal affixes (active and inactive), only one set of pronominal affixes (in the inverse type), or several sets (in the salience type). In section 4.1 we found that portmanteau morphemes can be generalized to accusative; in principle, also inactive affixes or elements of the PAT set of Arizona Tewa could develop to accusative – all these instances lead to an accusative headmarking. Head-marking is one of the major characteristics of languages with ‘rich’ morphology (Nichols 1986, 1992). Pronominal affixes can further develop to pure agreement markers; the verb then needs to be complemented by free syntactic elements that instantiate person and number values the verb agrees with. In contrast, dependent-marking usually does not emerge within sets of pronominal elements, but rather from elements that are attached to a noun phrase in order to specify its argument role. In section 4.3, I argued that Chinese ba could be regarded as being in the process of becoming an accusative marker. Similarly, a topic marker could get reanalyzed as an accusative morpheme. There are various ways in which systems with morphological case may have developed. In principle, head-marking and dependent-marking emerge independently of each other. One finds languages with both kinds of marking, languages that have only one of them, and languages with no marking (except by position), see (60). Basque exhibits ergative, dative and nominative, both on the auxiliary verb and on free nominals. Yimas (a language of New Guinea) exhibits ergative, accusative, dative and nominative sets of pronominal

Towards a structural typology of verb classes

99

affixes, but has no case marker on nouns. Japanese shows accusative, dative and nominative postpositions with nouns, but no agreement morphemes on the verb. Chinese, an ‘isolating’ language, represents the positional type SVO without agreement and case (except some tendency towards an accusative preposition). If both case and agreement are lacking, the language may nevertheless show ‘rich’ morphology; an example is the Siberian isolate Nivkh: the only way to mark an argument in Nivkh is incorporation of the lowest argument (see section 6.3 below). (60) Head marking and dependent marking are independent of each other with ‘agreement’

without ‘agreement’

with case

Basque (E, D, N)

Japanese (A, D, N)

without case

Yimas (E, A, D, N)

Chinese: isolating (SVO) Nivkh: polysynthetic (SOV)

4.4.2. Mixed accusative-ergative systems Pure accusative systems are found in Indo-European and in many other families around the world. A pure ergative system is found in Basque as well as in many more languages, but more than often ergative occurs together with accusative. Such a system is called a mixed system. Yimas (New Guinea) exhibits three sets of pronominal affixes in the 1st and 2nd person. As the examples in (61) show, a- can properly be called an accusative morpheme and ka- an ergative morpheme because both contrast with ama-, used with intransitive verbs. However, the 3rd person lacks accusative. Since an inflected verb usually begins with a nominative form, accusative and ergative never cooccur with a transitive verb. A detailed analysis of the pronominal affix paradigms of Yimas is presented in Wunderlich (2001). (61) Accusative and ergative in Yimas (Foley 1991) a. ama-tmuk-t. 1sgN-fall-perf ‘I fell down.’

b.

pu-a-tpul. 3plN-1sgA-hit ‘They hit me.’

c. pu-ka-tpul. 3plN-1sgE-hit ‘I hit them.’

d.

na-mpu-tpul. 3sgN-3plE-hit ‘They hit him.’

100 Dieter Wunderlich Accusative and ergative can, however, cooccur with transitive verbs of Dyirbal, a Pama-Nyungan language of Australia (Northeast Queensland). This language has morphological case, and again shows a split between local person (1st or 2nd) and 3rd person. Pronouns of local person are marked by accusative (-na), while demonstratives as well as full nouns are marked by ergative (-gu). (62) The linker inventory of Dyirbal (incomplete; Dixon 1994: 10/14) NOM

ACC

ERG

I

noun DEM: F.SG DEM: M.SG

 balan bayi

— — —

-gu bagun bagul

III

1PL ‘we all’ 2PL ‘you all’

ana
urra

ana-na
urra-na

— —

The reader can easily verify that this kind of split allows four combinations, summarized in (63). (‘loc’ indicates 1st or 2nd person.) (63) Four possible case patterns in Dyirbal Direct setting (loc/3) ‘We see the man.’ NOM

NOM

Symmetric setting (loc/loc) ‘We see you.’ NOM

ACC

Inverse setting (3/loc) ‘The man sees us.’ ERG

ACC

Symmetric setting (3/3) ‘The man sees him.’ ERG

NOM

In Udi, a Northeast Caucasian language, one finds these four case patterns, too, yet for slightly different reasons. As in Dyirbal, ergative pronouns are lacking in the 1st and 2nd person, and accusative is restricted to definite, animate or pronominal objects. The examples in (64) illustrate these possibilities. (The consonant t’ which precedes the case ending is a stem augment. The subject marker on the verb is infixed to the verb stem uk.) (64) The four-way case split of Udi (Schulze 2001) a. e-t’-in s/um-ax uk-sa. DIST-t’-ERG bread-ACC eat-PRES ‘(S)he eats the bread.’

Towards a structural typology of verb classes

101

b. e-t’-in s/um uk-sa. DIST-t’-ERG bread.NOM eat-PRES ‘(S)he eats bread.’ c. zu s/um-ax uk-sa. I.NOM bread-ACC eat-PRES ‘I eat the bread.’ d. zu s/um uk-sa. I.NOM bread.NOM eat-PRES ‘I eat bread.’ Also Hindi shows these four case patterns, but here the ergative is restricted to perfect, whereas accusative (similar to Udi) is found with human, specific animate or definite inanimate objects. (65) The four-way case split of Hindi (Mohanan 1994) a. niinaa-ne baalikaa-ko uthaa-y-aa.
lift-PERF-M Nina.F-ERG girl-ACC ‘Nina lifted up a/the girl.’ b. niinaa baalikaa-ko uthaa-eg-ii. Nina.F.NOM girl-ACC lift-FUT-F ‘Nina will lift up a/the girl.’ c. niinaa-ne roTii khaa-y-ii. Nina.F-ERG bread.F.NOM eat-PERF-F ‘Nina ate bread.’ d. niinaa kelaa khaa-eg-ii. Nina.F.NOM banana.M.NOM eat-FUT-F ‘Nina will eat a banana.’ The different case patterns thus signal not only what the subject and what the object is but moreover a certain distribution of values regarding animacy, definiteness, and aspect. Linguists observed surprisingly similar splits in unrelated languages from different parts of the world, a fact that calls for explanation. Aissen (1999) and, on a different theoretical background, Stiebels (2000, 2002) proposed to consider these similarities as resulting from the harmonic alignment of scales. I follow here Stiebels’ proposal. A first scale concerns the morphological features of means selected for marking a certain property; cross-linguistically, accusative is more preferred than ergative (66a). A second scale concerns the interacting contextual fea-

102 Dieter Wunderlich tures such as person, animacy, definiteness, or aspect; in each of these domains, the respective plus-valued feature is considered to be more salient than the minus-valued one, see (66b). These two scales are harmonically aligned if higher features are combined with higher features, and lower features with lower features, which yields, among others, acc/loc > acc/3 (‘accusative is preferred for a local person’) and erg/3 > erg/loc (‘ergative is preferred for the 3rd person’). Finally, in accordance with the assumption that a grammar has to be restrictive, the reverse ranking is expressed as a markedness hierarchy, e.g. *ACC/3 » *ACC/loc (‘accusative for the 3rd person is more strongly blocked than accusative for a local person’). This procedure yields a whole collection of markedness hierarchies, see (66c). (66) Harmonic alignment of morphological and semantic scales (Stiebels 2000, 2002)11 a. Morphological scale: +hr > +lr (for mnemotechnical ease: ACC > ERG) It is better to mark a lower argument (an object) than a higher argument (the subject). b. Contextual semantic scales: person: loc > 3 animacy: +anim > –anim definiteness: +def > –def aspect: +perf > –perf c. Contextualized markedness hierarchies: » *ERG/3 *ACC/3 » *ACC/loc *ERG/loc *ACC/–anim » *ACC/+anim *ERG/+anim » *ERG/
anim *ACC/–def » *ACC/+def *ERG/+def » *ERG/–def *ACC/–perf » *ACC/+perf *ERG/+perf » *ERG/–perf A split in the realization of ergative (accusative) results if the requirement (or MARK(ACC)) intervenes between the two markedness constraints, which could be done differently in different scales. Each of the hierarchies in (66c) allows three options, as shown in (67). MARK(ERG)

(67) a. MARK(ACC) » *ACC/3 » *ACC/loc: All three persons have accusative. ———–———————————————————————————— 11

For convenience, I use the case names ERG and ACC rather than the respective features, which would be more adequate in the general framework.

Towards a structural typology of verb classes

103

b. *ACC/3 » MARK(ACC) » *ACC/loc: Only 1st and 2nd person have accusative. c. *ACC/3 » *ACC/loc » MARK(ACC): Accusative is absent. The reader will realize that the collection in (66c) includes all the conditions that account for the splits reported above. For instance, *ERG/loc » MARK(ERG) » *ERG/3 is relevant for both Dyirbal and Udi (where ergative is restricted to the 3rd person), and *ACC/–def » MARK(ACC) » *ACC/+def is relevant for both Udi and Hindi (where accusative is ruled out for indefinite objects). A split can also occur between two coexisting linking devices, as it happens in Warlpiri, a Pama-Nyungan language of the northern territory of Australia. Here, nouns show ergative case, but agreement is accusativebased: -rna ‘1sg’ agrees with the subject of both intransitive and transitive verbs (68a–c), while -ju ‘1sg’ agrees with the object (68d); consequently, -ju has to be classified as accusative (A), and -rna as nominative (N). (68) Ergative case and accusative agreement in Warlpiri (Andrews 1985: 106f.) a. Ngaju ka-rna parnka-mi. I.NOM PRES-1sg.N run-NONPAST ‘I am running.’ b. Ngajlu-rlu ka-rna yankirri wajilipi-nyi. PRES-1sg.N emu.NOM chase-NONPAST I-ERG ‘I am chasing an Emu.’ c. Ngajulu-rlu ka-rna-ngku nyuntu nya-nyi. PRES-1sg.N-2sg.A you.NOM see-NONPAST 1-ERG ‘I see you.’ d. Ngaju ka-npa-ju nyuntulu-rlu nya-nyi. see-NONPAST 1.NOM PRES-2sg.N-1sg.A you-ERG ‘You see me.’ Such a mixed system in which head-marking exhibits accusative and dependent-marking ergative is found in several languages, whereas the reverse constellation does not exist. This fact again follows from harmonic alignment. Since argument affixes on the verb are pronominal, one has to regard the scales given in (69a,b). (69) Harmonic alignment regarding pronouns and nouns a. ACC > ERG (as above)

104 Dieter Wunderlich b. pro > N Pronouns are referentially more salient than nouns. c. *( ACC/N) » *( ACC/pro) Accusative marking on nouns is more strongly blocked than that on pronouns. d. *( ERG/pro) » *( ERG/N) Ergative marking on pronouns is more strongly blocked than that on nouns. Summarizing, the invention of a generalized case system has the advantage that it accounts for argument hierarchy purely structurally and at the same time allows the consideration of different semantic weights of arguments. 4.4.3. The emergence of dative Many linguists consider dative to be a semantic or lexically-inherent case, which does not emerge under structural conditions. There is, however, overwhelming evidence that dative is a structural case: it often marks the medial argument of a ditransitive verb (the recipient of ‘give’, ‘send’ etc., or the causee of a causativized transitive verb), and can alternate with nominative; a dative argument can be coindexed with an agreement morpheme, and can be suppressed in certain constructions. On the basis of these diagnostics, Wunderlich (2002a) argued that dative is a structural case in Hungarian; similar arguments can be put forward for other languages, too. There are, of course, many languages that do not have a dative, and, as Eisenbeiß (2002) has shown, the acquisition of dative always follows the acquisition of accusative in an accusative-based language like German. Dative can also appear in a pure ergative language such as Basque; and here the acquisition of dative follows that of ergative. I claim that dative is a general option of a generalized case system, but whether it arises or not depends on historical circumstances. One advantage of having the two features [+hr] and [+lr] is the possibility of combining them and thereby characterizing the medial argument of a ditransitive verb; correspondingly, dative is defined by the feature combination [+hr,+lr].12 ———–———————————————————————————— 12

In the preceding paragraphs, I used ACC and ERG instead of +hr and +lr for mnemotechnical reasons. Dative would then be the combination [ACC,ERG]; to avoid unnecessary confusion, I go back to the original features.

Towards a structural typology of verb classes

105

Such a decompostion of dative also conforms to Dowty’s (1991) observation that a dative argument accumulates properties of both proto-agent (being a controller) and proto-patient (being affected), and it makes at once clear that the existence of dative presupposes the existence of either accusative or ergative. (70) illustrates that dative can arise in an accusative- or an ergative-based system independently, but is certainly facilitated in a mixed system (in which the two features for marking case are already used). Dative cannot appear in a language of the inverse type. (70) Encoding argument roles and generalized case intransitives transitives ditransitives y x VERB(x,y) z y x VERB(x,y,z) x VERB(x) –hr +hr –hr +hr +hr –hr –lr –lr +lr –lr +lr +lr NOM

ACC NOM NOM ERG

ACC DAT NOM NOM DAT ERG

We call the case patterns associated with the argument roles in (70) ‘canonical’ because they offer an optimal distinction by case. Such a canonical system is very effective and also robust enough to allow various aberrations (‘noncanonical’ patterns) in the marking of minor lexical classes. 5. The lexical marking of verbs Many languages with generalized case, either head-marking or dependentmarking, use the respective features in establishing minor lexical classes, for instance, classes of noncanonical ‘transitive’ verbs. In section 3.4 we observed that an active type language allows double-inactive as well as doubleactive verbs, both deviating from a canonical transitive. Similarly, languages with generalized case can establish classes such as nominative-dative verbs, double-accusative verbs, ergative-marked intransitives, and so on. I consider these special classes to be marked lexically. The presence of a particular lexical feature is not totally idiosyncratic, one rather expects it to be (weakly) correlated with some semantic factor. For instance, ergative-marked intransitives of Hindi-Urdu are expected to be agentive rather than nonagentive. On the other hand, the semantic properties alone usually do not suffice to predict the membership in such a class. Often, the presence of lexical marking can only be motivated on historical grounds or by analogy. Paul (1919)

106 Dieter Wunderlich cites many examples where a particular verb shifts between canonical case and lexically marked case several times in the history of German. A positional system lacks the possibility of lexical case features. Verbs such as ‘help’, ‘like’ etc, which are lexically marked in Icelandic and German, are canonically transitive in English. A positional system thus clearly reduces the set of lexical classes. Similarly, the inverse type, where a semantic scale is aligned with argument hierarchy, does not allow lexical marking apart from inversetantum (‘make someone thirsty/drunk/sick’, ‘give someone a headache’ in Ojibwe) and less expected classifications of nouns (cars are animate). This does not exclude that semantic classes of verbs can be characterized by derivational means. 5.1. Two classes of lexically marked dative verbs The existence of dative decomposed into the feature combination [+hr,+lr] allows the introduction of two subclasses in the set of 2-place verbs by means of lexical marking. The two classes are symmetric to each other: either the lower argument is exceptionally marked by [+lr], or the higher argument is exceptionally marked by [+hr]. Through adding the remaining inherent features one yields the combination [+hr,+lr] in both instances. These two types of noncanonical datives should be distinct semantically. The first class is exemplified by the NOM-DAT verbs of German (such as helfen ‘help’, folgen ‘follow’, danken ‘thank’). These verbs seem to invite the inference that the object is a controller on its own, i.e., behaves as an active being in the event denoted by the verb. Indeed, the verbs often have an animate object that is independently in action. (71b) illustrates this class of ‘active object verbs’. Similar verbs are found in several languages, including the ergative language Basque, in which they compete with ERGNOM verbs (Joppen & Wunderlich 1995). (71) NOM-DAT verbs in German a. Sie folgte ihm. She.NOM followed he.DAT ‘She followed him.’ b. y x FOLLOW(x,y) designated: +lr inherent: +lr +hr –hr DAT NOM case :

Towards a structural typology of verb classes

107

The second class is exemplified by the DAT-NOM verbs of German (such as gefallen ‘like’, schmecken ‘taste, enjoy’, schmerzen ‘pain’, gehören ‘belong’). These verbs seem to invite the inference that the subject is affected, namely an experiencer rather than an agent. Indeed, many verbs of this class are experiencer verbs, and those verbs that alter between dative subject and accusative object (such as schmecken ‘taste, enjoy’) have a slightly more active reading with accusative (see (72)), which is expected from a semantic point of view. Yet, this class also includes verbs that do not have an experiencer subject (such as gehören ‘belong’), and, on the other hand, many experiencer verbs (such as fürchten ‘fear’, mögen ‘like’) instead belong to the canonical transitives. This suggests that this class cannot be defined on semantic grounds. Such an ambivalence with regard to semantic factors is characteristic for all types of lexical marking. (72) Alternating DAT-NOM and NOM-ACC in German a. Mir schmeckte der Braten/ ?der Thymian. I.DAT enjoyed the.NOM roast meat/ ?the thyme b. Ich schmeckte den Thymian/ ?den Braten. I.NOM tasted the.ACC thyme/ ?the roast meat How lexical marking functions in this class is illustrated in (73). The lower argument resists accusative because of the working of DEFAULT (‘every clause displays the default linker nominative’, Wunderlich 2003). On the surface, the shift from (72a) to (72b) seems to involve two features (appearing on two different arguments), but it can actually be described by the loss of just one lexical feature. Such a loss must have happened to English like historically, in view of the fact that cognate Icelandic likar ‘like’ still behaves as a DAT-NOM verb. (73) DAT-NOM verbs in German a. Ihm gefiel der Roman. he.DAT liked the.NOM novel ‘He liked the novel.’ b. y x LIKE(x,y) designated: +hr inherent: +hr –lr +lr NOM DAT case:

108 Dieter Wunderlich Many languages have productive classes of experiencer DAT-NOM verbs (or ACC-NOM verbs if dative is lacking), in particular ergative languages – perhaps because the shift from ergative to dative affects only one argument. Some languages also exhibit intransitive verbs with accusative marking; the lexical feature is again +hr (‘being affected’), which yields accusative because an intransitive verb does not have the inherent feature +lr. German examples are mich friert ‘I am cold’ and mich fröstelt ‘I am shivering’, both clearly experiencer verbs. In Choctaw, a Muskogean language of Oklahoma and Mississippi, all nonagentive intransitives (such a ‘be hungry’, ‘sweat’, ‘be tall’, ‘be old’) are marked accusative (Davies 1986), in contrast to agentive verbs, which canonically realize nominative.13 Choctaw also exhibits several transitive verbs that alternate between two case patterns regarding the pronominal (agreement) affixes on the verb. Some examples are given in (74). (74) Case pattern alternations in Choctaw a. NOM-ACC and NOM-DAT (Davies 1986: 110, 112) i. Chi-alikchi-li-tok. ii. Chim-alikchi-li-tok. 2A-doctor-1N-past 2D-doctor-1N-past ‘I doctored you.’ ‘I doctored you.’ anoksita ‘admire’, shilli ‘comb’, akammi ‘corral’, … b. NOM-DAT and ACC-DAT (Davies 1986: 128, 121) i. Chi-nokkilli-li-h. ii. Chi-sa-nokkilli-h. 2D-hate-1N-pred 2D-1A-hate-pred ‘I hate you.’ ‘I hate you.’ komota ‘fear’, nokpalli ‘desire’, noktalha ‘mistrust’, …

classa

classb

———–———————————————————————————— 13

As mentioned in footnote 8, one could as well assume that Choctaw belongs to the active type. Nominative would have to be identified with [+active], and accusative with [+affected], in full correspondence with what the case features [+lr] and [+hr] suggest semantically. In that case, dative would have to be characterized as [+active, +affected]. The case pattern alternations shown in (74) could then be described by lexical assignments of [+act] and [+aff], so that one would get the lattice in (75b), which is isomorph to that in (75a). Thus, nothing forces us to classify Choctaw as belonging to the nominative-accusative or to the active-inactive type. Obviously, this language is at the border between the two types, and it illustrates how the more abstract generalized case type could have emerged.

Towards a structural typology of verb classes

109

classc c. NOM-DAT and DAT-ACC (Davies 1986: 112, 86) i. Chim-ihaksi-li-tok. ii. Chi-am-ihaksi-tok. 2D-forget-1N-past 2A-1D-forget-past ‘I forgot you.’ ‘I forgot you.’ lhakoffi ‘miss’, ahchiba ‘tired of’, kania ‘lose’, achokma ‘like’ (75) Lexically marked transitive verbs of Choctaw (for b see footnote 13) a. Marking by case features y x b +lr +hr

b. Marking by active features y x +act +aff

DAT ACC

y +lr

x a,b,c

y

DAT NOM

y

x c +hr

y x +act

y

x +aff

ACC DAT

x

a

y

x

ACC NOM

The lattice in (75a) shows the relationship between these lexical markings. The members in classa and classb alternate between the presence or absence of just one lexical feature, while the members in classc alternate between the presence of a lexical feature on either the lower or the higher argument. Of course, some further assumptions must be made about how the grammar of Choctaw deals with these lexical features. For instance, DEFAULT (which requires nominative in a domain) must be ranked lower than in German (where +hr on the higher argument leads to DAT-NOM rather than to DATACC as in Choctaw). In a comparative study on lexical marking in Icelandic and German, Wunderlich (2003) accounts for all occurring case patterns on the basis of a few general constraints, which for Choctaw would have to be slightly reranked. Interestingly, lexical marking in Choctaw only affects the morphology of verbs. Nouns (noun phrases) only show nominative or accusative (the latter called ‘oblique’ by Davies), so that all the verbs shown in (74) realize their arguments syntactically in the canonical NOM-ACC pattern. This once again demonstrates that verb-external syntax is much more regularizing than verb-internal morphology.

110 Dieter Wunderlich 5.2. Two classes of experiencer verbs with an expletive argument Another, totally different type of lexical marking is mimicry. We say it rains, but cannot ask *what rains? because RAIN is a zero-place predicate, which does not allow any nominal argument. Therefore, the it in it rains is called expletive; it fulfills a structural rather than a semantic condition on verbs in English. Here, a zero-place predicate mimics a structurally intransitive verb. Similarly, one-place predicates can mimic structurally transitive verbs. This is possible in two ways: either a higher or a lower argument role is added for purely structural reasons. German allows both options: a higher expletive is realized by es ‘it’, whereas a lower expletive is realized by a reflexive (+refl). The first option yields impersonal verbs: the true argument is shifted to a lower role and thus receives accusative, as if there were a higher argument; this is demonstrated in (76b). This shift invites the inference that the argument is affected by some process, which is typical for experiencers. (76) Impersonal transitives in German a. (weil) es ihn ekelte (because) it he.ACC disgusted ‘(because) he was disgusted.’ b. y  x BE_DISGUSTED (y) inherent: +hr –hr es The second option yields the so-called inherent reflexives (such as sich erinnern ‘remember’): they have a fake object, which is necessarily bound to the subject. This again invites for an experiencer reading: one and the same argument is both agentive and affected. (77) Inherent reflexives in German a. Er schämte sich. he.NOM was.ashamed himself ‘He was ashamed.’ b.  y x BE_ASHAMED(x) inherent: +hr –hr +refl

Towards a structural typology of verb classes

111

Some verbs allow both patterns without any shift in meaning, which shows that these patterns themselves do not contribute to compositional meaning. (78) Alternation between impersonals and inherent reflexives es ekelte ihn er ekelte sich ‘it made him sick’ es freute ihn er freute sich ‘he was pleased’ es wunderte ihn er wunderte sich ‘he was surprised’ German exhibits more patterns involving impersonals and inherent reflexives, including dative reflexives (sich etw. aneignen ‘take possession of sth.’, sich etw. anmaßen ‘usurp sth.’ sich etw. verbitten ‘beg to decline sth.’). Consider the examples in (79). (79) Inherent dative reflexives in German a. Ich eignete mir das an. I.NOM got.possession 1.DAT.REFL that.ACC at ‘I took possession of that./ I acquired that.’ b. Ich trank mir einen an. I.NOM drunk I.DAT.REFL INDEF.ACC at ‘I drunk too much.’ c. Ich arbeitete mir einen ab. I.NOM worked I.DAT.REFL INDEF.ACC off ‘I worked too much.’ (Here, not only the reflexive but also einen is semantically empty.) These examples exhibit the canonical NOM-DAT-ACC pattern. In general, the introduction of expletive arguments preserves the canonicality of case patterns, whereas lexical case marking introduces noncanonical case patterns. They are thus opposite options for marking special classes in the lexicon. 6. Ditransitive verbs: The emergence of a third argument Every language needs a device to realize triadic predicates because even the most elementary transaction and communication events involve three participants: the giver, the recipient, and the given object, or the speaker, the addressee, and the uttered message. If such a device exists, it can easily be generalized to cover also the expression of more complex or derived three-

112 Dieter Wunderlich participant events. In any case, the ditransitive option extends the grammatical potential of transitive verbs. It is the recipient or addressee who constitutes the third argument not found in a prototypical transitive verb. The following survey will attest that it is not a simple step to go beyond the argument linking system developed for transitive verbs. Several variants of accounting for a third argument are imaginable, more conservative ones and more innovative ones. Either the issue of realizing a third argument is reduced to what can be done with transitive verbs, or the existing option for transitive verbs is slightly extended: a third position is introduced in a positional system, and a third case is introduced in a case system. In any case, how a third argument is integrated crucially depends on the particular linking type for transitive verbs. In subsequent sections I will present three major typological variants (see also Sedlak 1975; Faltz 1978; Dryer 1986; and Haspelmath 2005). In the first, the number of syntactic arguments is nontrivially reduced by a serial verb construction or by noun incorporation; both options are innovative for human language in general, although they are nearly absent in the IndoEuropean family (section 6.3). In the second (perhaps the most conservative), the recipient is regarded as primary object and therefore treated like the object of a transitive verb (section 6.4). In the third, the recipient is treated differently from the object of a transitive verb, namely as the intended location or the intended possessor (section 6.5). The two latter alternatives also seem to underlie the English ‘dative’ alternation, which is discussed in section 6.6. Before that, I will discuss the semantic decomposition of ditransitive verbs in section 6.1, and a few of the general constraints that help us to grasp ditransitive verb formation more theoretically (section 6.2). 6.1. How complex are ditransitive verbs semantically? Why do we care about ditransitive verbs? On the one hand, speakers of an informed community must be able to express three-participant events of the kind one finds in transactions and communications. On the other hand, all argument linking types found in the languages around the world seem to have specialized on the basis of transitive verbs, and thus the question arises: how can a third argument be integrated? From the semantic point of view the answer is relatively simple. An elementary three-participant event involves a transitive event directed to a goal. One gives (or says) something to an intended recipient; similarly, one

Towards a structural typology of verb classes

113

puts something to an intended location. All ditransitive events include a transitive action with an intended result, which itself is stative. Essentially, two types of two-place stative predicates play a role: POSS(y,z) and LOC (z,AT(y)); something is in the possession of a person, or it is located at a certain place or object. The combination of a transitive action with a two-place stative result is usually linked by a shared argument. If I sent a letter, and you received the letter, ‘the letter’ is the shared argument. Similarly, if I say “hallo”, and you hear “hallo”, this ‘hallo’ is shared in our minds. In order to integrate two predicates of this sort in just one (ditransitive) verb, the respective object must be the shared argument. What one has first to find out is how the third argument is integrated semantically, and, then, how it is expressed. The former problem concerns us in this section. 6.1.1. Two types of elementary ditransitive verbs As already indicated, all verbs with three arguments are semantically complex, they can be decomposed into two predicates: one describing a certain activity and the other describing a certain result. Nearly every language exhibits some of these verbs in its primitive lexical repertoire (with no sign of morphological derivation). A closer inspection of these underived ditransitive verbs shows that they belong to two well-defined semantic classes: change of possession and change of location verbs. The former denote a transition into a state of possession (as defined by BECOME POSS), while the latter denote a transition into a state of location (as defined by BECOME LOC). Typical change of possession verbs are give, lend, and buy (in general, ‘give’ verbs of any arbitrary language, and often also ‘show’ and ‘ask’). The third argument of these verbs is typically a recipient, a human or animate being who comes into the possession of an object. The three participating semantic roles can be called agent (=x), recipient (=y) and theme (=z). (80b) shows possible semantic representations of these verbs. (80) Change of possession verbs a. Anna gave Max a book. Anna bought Max a book. b give: z y x e {ACT(x) & BECOME POSS(y,z)}(e) buy: z y x e {BUY(x,z) & BECOME POSS(y,z)}(e)

114 Dieter Wunderlich Typical change of location verbs are put, throw, push, and glue (and many similar verbs in other languages). The third argument of these verbs is typically a goal, an object or a place where the theme becomes located. Many languages have developed locative prepositions or locative cases to encode location in combination with verbs. Thus, what we actually observe in the English examples in (81a) is a third syntactic argument in form of a directional prepositional phrase (PP), and the goal is an argument of the preposition rather than of the verb itself. We can speak of ‘indirect linking’ in such a case. (81) Change of location verbs a. Anna put the glasses behind the tree. Max threw the book behind the tree. b. throw: P y x e {THROW(x,y) & P(y)}(e) behind the tree: u (BEC) LOC(u, BEHIND* the.tree) throw behind the tree: y x e {THROW(x,y) & BEC LOC(y, BEHIND* the.tree)}(e) (81b) shows how such an indirect linking takes place. The first line states that ditransitive throw requires a predicative argument P, which can, for instance, be instantiated by the PP behind the tree. Semantically, this PP provides the predicate LOC specified for a BEHIND* region related to ‘the tree’. The last line shows the full VP as a result of combining these two pieces of information. The prepositional content often gets incorporated into the verb, as happens in the English word enter, a verb that can be decomposed into {GO(x) & BECOME LOC(x, IN z))}. Verbs that undergo the so-called locative alternation quite systematically incorporate the predicate LOC into the verb. (82a) shows paste/kleben as a change of location verb in the format of (81), and (82b) shows the alternative in which LOC is incorporated, so that the goal becomes a direct argument of the verb. (For reasons that are discussed in section 6.2, the theme becomes oblique in this case.) Notice that German marks the alternation with the prefix be- on the verb, while English leaves it unmarked. (82) Locative alternation in English and German a. Anna pasted all her photos on the wall. Anna klebte alle ihre Photos an die Wand. glue: P y x e {PASTE(x,y) & P(y)}(e)

Towards a structural typology of verb classes

115

b. Anna pasted the whole wall with her photos. Anna beklebte die ganze Wand mit ihren Photos. glue: z x e y {PASTE(x,y) & BEC LOC(y,z)}(e) Some verbs (such as German schicken ‘send’) denote an event in which change of possession and change of location cooccur; these verbs have both options. (83) Change of possession & location verbs a. Anna schickte dem Verleger die Photos. Anna sent the.DAT publisher the.ACC photos b. Anna schickte die Photos an den Verleger / in die Bibliothek / Anna sent the.ACC photos at the publisher / into the library / auf den Speicher. onto the store. ‘Anna sent the photos to the publisher/to the library/to the store.’ (It is a peculiarity of English that only to is possible here, see section 6.6 below.) If the added argument is human (or animate) the recipient construction is preferred, but if it is inanimate the goal construction is preferred. (§Anna schickte dem Speicher die Photos sounds odd.) The two classes of primitive ditransitives verbs, then, can schematically be summarized by the semantic representations in (84), with the proviso that LOC in (84b) could be provided by a prepositional argument. (84) Two classes of ditransitive verbs a. ACT(x) & BEC POSS(y,z) y is a recipient – a medial argument b. ACT(x) & BEC LOC(z,AT(y)) y is a goal – the lowest argument As before, ACT(x) is assumed to be an unspecific agentive predicate, which encodes that x does something intending a certain result. (Regarding the most unspecific verbs give and put one only knows that some transitive action is involved, but which kind of action depends on the context.) One could instead assume an unspecific manipulation predicate MANIP(x,z), which encodes that x does something concerning z, but that wouldn’t change anything in the analysis because z must be a shared argument. However, verbs such as buy and throw involve a more specific manipulation predicate. In the result always a third argument is added, a recipient, a goal,

116 Dieter Wunderlich or sometimes a source, as happens in ‘steal’. Jim stole his neighbour a knife denotes an event where the neighbour possessed the knife in the beginning rather than in the end. Double object constructions of Chinese are even preferred with the source reading (Zhang 1998). That semantic decompositions such as those proposed in (84) are relevant becomes evident if one considers two of the major classes of denominal verbs: locatum vs. location verbs). The former class includes to bridle/ zäumen, to saddle/satteln, to salt/salzen, to sugar/zuckern (where the noun refers to some supplement or substance), whereas the latter class includes to cellar/kellern, to store/speichern, to shoulder/schultern, to bottle, to box (where the noun refers to a container or a supporting object). (85) Two major classes of denominal verbs a. Locatum verbs have the generalized reading {ACT(x) & BECOME POSS(y,z)} b. Location verbs have the generalized reading {ACT(x) & BECOME LOC(z,AT(y))} How can the meaning of a denominal verb be determined? Following Kiparsky (1997) and Stiebels (1998) I assume that an abstract semantic template is chosen where the respective noun is integrated as the lowest argument. Thus, to bridle the horse gets the reading ‘supply a horse with a bridle’ (i.e., make the horse to have a bridle), whereas to cellar the wine gets the reading ‘put the wine into a cellar’ (i.e., make the wine to be located in a cellar). Which template is chosen partly depends on the prototypical reading of the noun, and partly on the argument structure used with the verb. This can be shown with a neologism I owe to Urbas (1990). If one is confronted with the statement John matchboxed the crumbs, one might come up with the reading that John collected the crumbs in a matchbox (because a container noun triggers the template of location verbs), whereas the intransitive clause John matchboxed the whole time suggests a different reading, for instance, that John produced matchboxes, and John matchboxed the crumbs into a box (where matchbox itself is a change of location verb) suggests an instrumental reading. The possibility of deriving a complex verb from a simple noun constitutes one of the major arguments for semantic decomposition. Without having access to semantic templates it would remain mysterious why speakers can so productively (and also regularly) turn nouns into verbs. The templates used for the generation of denominal verbs probably derive from

Towards a structural typology of verb classes

117

generalizations concerning semantic classes of simple verbs. Given that both types of denominal verbs cited in (85) are productive, it is clear that both change of possession and change of location are available options. 6.1.2. Derived ditransitives There are various ways of supplying an ordinary transitive verb with a third argument. Such an argument extension operation can be marked morphologically (e.g., by an affix added to the verb), or it can remain unmarked. Typical derivations are the following: – the causative adds a causer – the applicative adds a beneficiary, a location, or an instrument – the so-called possessor raising adds a possessor of the manipulated thing German does not have morphological means to perform these derivations, except that a particle or prefix can do such a job in one of its semantic functions (e.g., be- in the locative alternation) (Stiebels 1996). Nevertheless, all three types of derivations can be found in German: a causative verb (86a), a benefactive dative (86b), and a possessor dative (86c). Schematic representations are given in (87). (86) a. Sie ließ ihn einen Brief schreiben. ‘She let him write a letter.’ b. Sie kochte ihm eine Suppe. ‘She cooked him a soup.’ c. Sie verband ihm die Hand. She bandaged he.DAT the hand ‘She bandaged his hand.’

(alienable possession)

(inalienable possession)

(87) a. ACT(x) & WRITE(y,z) b. COOK(x,z) & BECOME POSS(y,z) c. BANDAGE(x,z) & POSS(y,z) Since an extra argument must be licensed by some predicate, operations that add an argument are always incremental from the semantic perspective. The causative adds the higher predicate ACT(x), in which x is the causer (an agent), while the agent of the embedded verb (y) becomes a causee. (In the German verb complex schreiben lassen ‘let write’, the causee is realized by

118 Dieter Wunderlich accusative.) The two other operations, illustrated in (86b–c), add POSS (or BECOME POSS); in such a case the extra argument (y) is the medial one. The main difference is that the beneficiary (86b) tends to be an alienable possessor (one who comes into possession), while the possessor (86c) is of course an inalienable possessor (the one the hand belongs to). Most grammatical systems treat these argument extensions in the same way as they treat the basic ‘give’ verb, probably on the basis of generalization. (For instance, the German sentences in (86b–c) display the same canonical ditransitive pattern NOM – DAT – ACC as geben ‘give’.) However, whether the causative is involved in this generalization depends on how far the ‘give’ pattern is generalized. For instance, the causee of the Hungarian causative (88b) is differently realized from a recipient (88a). One can assume that the Hungarian causative morpheme imports instrumental case. (88) ‘Give’ and causative in Hungarian a. Anna Péter-nek adott egy könyv-et. Anna Peter-DAT gave a book-ACC ‘Anna gave Peter a book.’ b. Anna Péter-rel olvas-tat egy könyv-et Anna Peter-INST read-CAUS a book-ACC ‘Anna lets Peter read a book.’

.

The languages of the world differ widely in the way they add a higher argument (a ‘new’ subject) or a lower argument (a further object) to a transitive verb. Surveys on causative constructions are found in Comrie (1985) and Dixon (2000), among others. I will briefly illustrate two ways of adding a lower argument, in considering Indonesian and Tzotzil, a Mayan language of Mexico. The formation of ditransitive verbs in Indonesian is very productive. Transitive verbs are used with a prepositional phrase (89), but can also be suffixed with -kan (an applicative morpheme); this results in a double object (DO) construction (90) (see also section 6.4 below). According to most of the object tests applied by Chung (1983), the first (i.e., the added) object behaves like the object of a transitive verb. (89) Transitive verbs with PP in Indonesian (Chung 1983) a. mereka mem-bawa daging itu [kepada dia]PP. TRANS-bring meat the to him they ‘They brought the meat to him.’

Towards a structural typology of verb classes

119

b. Ali mem-beli telefisi [untuk ibu-nja]PP. mother-his Ali TRANS-buy television for ‘Ali bought a televison for his mother.’ (90) Ditransitive extension in Indonesian (Chung 1983) a. mereka mem-bawa-kan dia daging itu. TRANS-bring-BEN him meat the they ‘They brought him the meat.’ b. Ali mem-beli-kan ibu-nja telefisi. Ali TRANS-buy-BEN mother-his television ‘Ali bought his mother a televison.’ Unlike derived ditransitives, a few basic ditransitive verbs can be used in the DO construction without -kan (beri ‘give’, kasih ‘give’, bajar ‘pay’), but curiously, -kan is used here to mark the prepositional construction optionally (that is, has an opposite function to that in (90)). (91) Alternation of ‘give’ in Indonesian (Chung 1983: 234) a. Ali beri wanita itu surat. Ali give woman the letter ‘Ali gave the woman a letter.’ b. Ali beri (-kan) surat kepada wanita itu. woman the Ali give-BEN letter to ‘Ali gave a letter to the woman.’ Ditransitive verbs of Tzotzil (an ergative language) must be formed through be-suffixation applied to transitive verbs. The added argument, the recipient in (92a,b) and a possessor in (92c), is then marked by nominative agreement like the object of a transitive verb. Example (92c) shows that the (nominative) possessor on the verb agrees with the (ergative) possessor on the object (corresponding to German er hielt mir meinen Kopf). (92) Ditransitive extension in Tzotzil (Aissen 1987: 106, 107, 126) a. Ch-a-k-ak’-be. INCOMPL-2.NOM-1.ERG-give-DITRANS ‘I’ll give it to you.’ b. ’i-j-chon-be chitom li Xune COMPL-(3.NOM)-1.ERG-sell-DITRANS pig the Xun ‘I sold pigs to Xun.’

120 Dieter Wunderlich c. Ch-i-s-toyilan-be j-jol. INCOMPL-1.NOM-3.ERG-keep.lifting-DITRANS 1.ERG-head ‘He kept lifting my head.’ In this language, ‘give’ is not a basic ditransitive but rather a transitive verb that has to undergo the applicative -be, which is generalized to cover also possessor raising. This observation nicely supports the assumption that recipients are a kind of possessors. Sections 6.3 to 6.5 below address the question how a third argument is realized if a certain argument linking potential for transitive verbs is given. In languages that amply use local prepositions, the problem of realizing a third argument mainly reduces to the question how a recipient gets realized. 6.2. Some general constraints Semantic composition by which two or more predicates are combined increases the generative power of a language. One therefore aims at restricting the possible combinations, so that the generative power remains delimited. (If everything were possible in a language, utterances would produce semantic noise rather than articulated speech.) More specifically, we are looking for a way to restrict the combinations of predicates in a verb or verb-like complex (such as verb-verb compound or serial verb construction). In Lexical Decomposition Grammar (Wunderlich 1997a,b), the semantic form (SF) of a verb is intended to be a minimal semantic decomposition, i.e., the meaning of the verb is decomposed only as far as necessary for predicting the argument hierarchy and other structural factors. Perhaps it is never possible to specify the conceptual meaning of a verb completely, at least, this would not be the aim of linguists who are interested in the structural impact of decomposition. In order to determine argument hierarchy, SF decompositions of verbs must be asymmetric. Logically (conceptually), A & B and B & A are equivalent, but as SF decompositions they are not. By convention, the first predicate in our representations is considered to be the higher one, so that A & B is actually [A [& B]]. According to various authors (Kaufmann 1995b; Kaufmann & Wunderlich 1998; Wunderlich 2000; Gamerschlag 2005), the SF of a simple or derived verb obeys at least the following constraints: the combination of predicates denotes a coherent situation (93a), a causal predicate precedes the result predicate (93b), and there is a referential chain between the components (93c).

Towards a structural typology of verb classes

121

(93) a. COHERENCE. Subevents encoded by the predicates of a decomposed SF structure must be connected contemporaneously or causally/consequentially. b. ICONICITY. In a decomposed SF structure, cause precedes result, and consequences follow their instigation. c. CONNECTION. In a decomposed SF structure, each predicate must share at least one argument with another predicate, either explicitly or implicitly. COHERENCE requires that the individual predicates in A & B either relate to the same time span or are connected by a causal/consequential relation. In the latter case, ICONICITY informs us about the function of the individual predicates: in A & B, B is the result of A, or B belongs to a set of consequences enabled by A. Because of such a strong requirement, ACT(x) & BECOME POSS(y,z) can only receive the interpretation that ACT(x) causes y to get into the possession of z because an ongoing activity and a transition (expressed by BECOME) can never be contemporaneous (at most they could overlap temporally). Therefore, a connecting predicate such as CAUSE need not be encoded explicitly. Although serial verbs (see also 6.3) seem to form a syntactic construction, they nevertheless obey the condition of coherence. It has been unanimously pointed out by all researchers concerned with serial verbs that they denote just one coherent event, in contrast to a syntactic coordination, which denotes two independent events. For instance, a serial verb construction never allows an anaphoric pronoun (94a), while a coordination often requires it (94b). This suggests that serial verbs are a lexical complex in which the argument that separates the two verbs is infixed, similarly to a particle verb, in which the particle often is separated from the verb. (The examples are from Edo, a Benue-Congo language of Nigeria.) (94) Serial verb vs. coordination in Edo (Stewart 2001: 60) a. Òzó lé èvbàré ré. Ozo cook food eat ‘Ozo cooked and ate food.’ b. Òzó lé èvbàré rrí ókè. Ozo cook food eat it ‘Ozo cooked food and ate it.’ CONNEXION requires that at least some of the arguments of different predicates are identified with each other. Thus, COOK(x,z) & BECOME POSS(y,z)

122 Dieter Wunderlich in (87b) above characterizes a possible verb (e.g., ‘John cooks porridge for Anne’), whereas COOK(x,u) & BECOME POSS(y,z) does not. (‘John cooks porridge and Anne gets a cake’ could not be expressed by a single verb.) An internal argument of a predicate that is followed by another predicate in SF cannot be realized structurally. For instance, in the resultative construction (95b) one cannot express the stuff that was drunk. (95) Strong resultatives in English (Kaufmann & Wunderlich 1998) a. The guests drank all of the red wine. b. The guests drank the wine cellar empty. c. DRINK(x,u) & BECOME EMPTY(z) The object u of DRINK becomes nonstructural if a resultative is added that predicates of a different argument than u. Wunderlich (1997a,b) proposed the constraint (96), which restricts the mapping of SF representations into morphosyntax. (96) STRUCTURAL ARGUMENT. An argument is structural only if it is either the lowest argument or (each of its occurrences) L(exically)-commands the lowest argument. L-command is defined for the nodes in SF, which represent logical types, as follows:  L-commands  if the node , which either directly dominates  or dominates  via a chain of nodes type-identical with , also dominates . With these preparations, we can establish the argument hierarchies for the two classes of ditransitive verbs outlined in (84), as well as for causativized transitives, in (97). (97) Argument hierarchy of ditransitive verbs a. Change of possession: ACT(x) & BECOME POSS(y,z) argument hierarchy: x > y > z (agent > recipient > theme) b. Change of location: ACT(x) & BECOME LOC(z,AT(y)) argument hierarchy: x > z > y (agent > theme > goal) c. Causativized transitives: ACT(x) & VERB(y,z) argument hierarchy: x > y > z (causer > causee > patient) In these minimal representations it is sufficient to satisfy CONNEXION by implicit arguments: indeed, x would have to act on z in both (97a) and (97b), and on y in (97c).

Towards a structural typology of verb classes

123

In the remainder of this chapter I briefly survey the various constructions for realizing three arguments. Several of these constructions can coexist in a language. Some ditransitive verbs may follow one of the patterns, while other verbs follow another pattern, or a single verb may allow variation according to contextual circumstances. It is also possible that the pattern for dependent-marking (morphological case) differs from that for head-marking (e.g., agreement). 6.3.

The number of syntactic arguments is reduced

In this section two constructional types are considered which avoid the introduction of any further means for dealing with three structural arguments: serial verb construction and noun incorporation. As Mattissen (this volume) points out, these two constructions largely determine what is known as the polysynthetic type of language. 6.3.1. Serial verb constructions The serial verb construction (SVC) is found in many South-East Asian and West African languages, and in other regions of the world as well. It is not restricted to particular language families and also occurs in Creole languages, as well as in sign languages. The examples in (98) from Yoruba, a Benue-Congo language of Nigeria, illustrate an SVC encoding change of possession. The first verb is transitive and realizes subject and object positionally. The second verb with the meaning of ‘give’ shares these two arguments and adds a third one, a recipient. Literally, the construction means ‘Baba took the gown and then he gave it to the chief’, but it actually lacks anaphoric pronouns such as he and it. Such a construction is sometimes the only way to realize the ditransitive verb ‘give’. Several tests show that an SVC denotes a single coherent event rather than the coordination of two independent events; it is thus more similar to a single compound verb than to a syntactic coordination. (98) Change of possession: serial verbs in Yoruba, left-headed (Baker 1991: 80f) a. Bàbá fi èwù fún oba. Baba take gown give chief ‘Baba gave the gown to the chief.’

124 Dieter Wunderlich b. Ó ra isu fún mi. 3sg buy yam give 1sg ‘He bought me a yam.’ Yoruba is an SVO positional type language, where the verb is to the left of a verb phrase. The SVC is then left-headed, too, i.e., the first verb is its head, and the shared object is realized right-adjacent to it. Although the two verbs are separated by the shared object, they function semantically as a compound. Most languages with SVCs belong to the SVO type, but Ijo, an Atlantic-Congo language, shows the rare case of an SVC based on the SOV pattern (99). Here, the head is to the right: the object precedes the verb, and the second verb constitutes the head of the SVC, as indicated by tense marking. The two verbs are again separated by a noun, but in this case by the recipient, which exclusively belongs to the second verb. (99) Change of possession: serial verbs in Ijo, right-headed (Williamson 1965: 54) Erí opúru-mo àki tobóu pìri-mi 3sg crab-DET.PL take boy give-PAST ‘He gave the boy the crabs.’ The serial verb construction is able to express many more semantic relationships between two verbs (such as instrumental ‘take a knife and cut the tree’, consequential ‘go to the market, buy a fish and cook (it)’, or manner ‘go around and search’). (100) shows an example of change of location in the SVO-type. Here, the first verb introduces agent and theme, which function as the shared object, while the second verb adds a goal. The goat in the example is the being that is pushed and thereby forced to fall into the hole. (100) Change of location: serial verbs in Edo, left-headed (Baker & Stewart 1999: 20) Úyì sùá èwé lá ùvún. Uyi push goat enter hole ‘Uyi pushed the goat into the hole.’ The serial verb construction has the advantage that it offers additional argument positions for a triadic predicate. This is obvious in examples where the same predicate concept or even the same verb is repeated. In Oaxaca Chontal, a Hokan language of Mexico, the two ‘give’ verbs are sligthly dif-

Towards a structural typology of verb classes

125

ferent (101), whereas Cantonese repeats the same verb in just one construction (102). In both examples the recipient is added by the second verb. (101) Double-give in Oaxaca Chontal (Sedlak 1975) liw’á. kúpa elmelyu páypa 3sg.gave the.money 3sg.gave.to his.son ‘He gave the money to his son.’ (102) Double-give in Cantonese (Sedlak 1975: 146) O pei ts’in pei koey. 1sg give money give 3sg ‘I give money to him.’ If serial verbs are a means of realizing additional arguments, one expects that one of the verbs can become a marker for a specific type of argument. Interesting in this connection is that Mandarin Chinese has generalized markers in both verb positions. The alternation in (56) shows that ‘give’ in the second position marks a recipient, but if it is in the first position it marks a beneficiary. (103) Alternation between recipient und beneficiary marking in Mandarin (Luo 1999: 4) a. Lisi xie xin gei wo. (Recipient or goal) write letter give 1sg ‘Lisi wrote me a letter.’ b. Lisi gei wo xie xin. (Beneficiary) give 1sg write letter ‘Lisi wrote a letter for me.’ 6.3.2. Noun incorporation Another, totally different way of reducing the number of syntactic arguments is noun incorporation. A verb can integrate its lowest argument by lexical incorporation so that a more complex verb results whose valency is reduced by one: a transitive verb shifts to intransitive, and a ditransitive verb shifts to transitive. This makes it possible to account for all ditransitive verbs by the same grammatical means that are used in transitives. One such language is Arizona Tewa (see also section 4.3 above). The example in

126 Dieter Wunderlich (104) shows that the theme argument is incorporated, so that the PAT prefix can successfully refer to the recipient. (104) Ditransitive verbs in Arizona Tewa (Kroskrity 1985) na:’in dí-k’ú:wá-mE!gí we 1.PAT-sheep-give ‘We were given sheep (by you or some third person).’ Yukatek Maya is another language with incorporation. The examples in (105) illustrate that in both change of possession and change of location first the respective theme is integrated by incorporation, and then the resulting verb undergoes applicative, which allows a further argument. This further argument is a recipient realized by a suffix in (105a), while it is a goal realized syntactically by a noun phrase in (105b). (105) Noun incorporation, followed by applicative, in Yukatek (Krämer & Wunderlich 1999: 466f) a. taan=u kon-lol -t-ik-et

INCOMPL=3.ERG sell-flower -APPL-IMPF-2.NOM ‘He is selling flowers to you.’ b. h b/in-et a lam -k/ak/-t le kòl-o/ COMPL go-2.NOM 2.ERG stick.into-fire-APPL DEF cornfield-DEM ‘You went to set fire to the cornfield.’ These examples show that, similarly to the serial verb construction, change of possession and change of location can be treated alike. Both require a third argument: the argument which is added by the second verb (of the SVC) or by applicative (after noun incorporation) is either a recipient or a goal. This is predicted by the templates given in (84a) and (84b). At first glance, the Siberian language Nivkh seems to constitute a counter-example to this prediction – surprisingly, the third argument is always incorporated into the verb, regardless of whether there is a change of possession, as in (106), or a change of location, as in (107). Under the assumption that only the lowest argument of a verb can be incorporated, one expects that a goal can be incorporated, like in (107), but not a recipient – because a recipient is not the lowest argument.

Towards a structural typology of verb classes

127

(106) Change of possession in Nivkh (Mattissen 2001: 158ff.) a. ´m´k karandas p’-o“la-k’im-d. REFL-child-give-INDIC mother pencil ‘The mother gave a pencil to her child.’ b. ñi to h´-ñivx-ar-d. 1sg fish that-person-feed-INDIC ‘I am feeding fish to him/her.’ (107) Change of location in Nivkh a. ñi seta ir-t’i-d. 1sg sugar dish-put-INDIC ‘I put the sugar into the cup.’ b. ma t’om-hupu-d. dried.fish fat-dip-INDIC ‘I dipped dried fish into fat.’ The only possible way to account for data such as those in (106) is to assume that recipients are treated as goals. More precisely, one has to assume that the dominant relation in Nivkh is LOC, even if other languages would prefer POSS. This assumption is compatible with the fact that Nivkh does not have any possession verb: in order to express simple possession (or lack of possession) existential verbs are used with locative case (108a). On can further observe that locations can generally be incorporated in Nivkh: (108b) shows an example of locative incorporation into a stative intransitive verb. (108) Possession verb and locative incorporation in Nivkh (Mattissen 2001) a. ñ-uin h´-bitV´ q’au-d. 1sg-LOC that-book not.exist-INDIC ‘I do not have that book.’ b. p’-r´u-o“la parta-ñv-d. REFL-teach-child desk-sit(down)-INDIC ‘The pupil sits at a desk.’ The only relevant mechanism of indicating the role of an argument is noun incorporation in Nivkh, and in this respect the language is unique. There is a clear advantage of such a mechanism. Given that a verb denoting a change of possession has two animate arguments, it is wise to make a distinction

128 Dieter Wunderlich between these two arguments. The solution of Nivkh is: if an animate argument is incorporated in a verb it must be a recipient or beneficiary; it cannot be an agent. One can regard noun incorporation as a separate argument linking type (Wunderlich 2002c). Let us finally see how causative and noun incorporation can interact. The causative adds an argument, and noun incorporation binds an argument. This can be done in either order, and even iteratively, as illustrated by examples from Alutor, a Chukotko-Kamchatkan language. (109a) shows that a normal transitive verb results if first the object of the underlying verb is incorporated and then causative applies (literally ‘I made my wife meat-eat’). If causative applies first (109b), a ditransitive verb results, and since only two structural arguments are possible, one of the objects must be realized obliquely (here: the theme by instrumental case); subsequent incorporation then leaves the verb intransitive, and the agent is realized by nominative (literally ‘I wife-fed with meat’). (Note that the information about the subject is repeated in the final slot of an intransitive verb. Causatives are formed with the prefix n- and a simultaneous suffix.) (109) Different orderings of causative and noun incorporation in Alutor (Koptjevskaja-Tamm & Muravjova 1993) av´ a. g´m-nan t´-n´-t´rg´-// aw´´ j-at-´n 1sg.S-CAUS-meat-eat-SUFF-3SG.O wife.NOM I-ERG ‘I fed my wife with meat.’ b. g´mm´ t´-av´-n-aw´´ j-at-´k t´rg-a I.NOM 1sg.S-wife-CAUS-eat-SUFF-1sg.S meat-INSTR ‘I fed my wife with meat.’ If causative is applied iteratively, noun incorporation can be repeated, too. (110) shows a derivation which starts with an intransitive verb (‘dry’) and finally ends in an intransitive verb (which of course is much more complex semantically). Interestingly, in each stage of the derivation, the number of structural arguments is either one or two, corresponding to the number of arguments that can be encoded on the verb. (110) Iterated causative and noun incorporation in Alutor (Koptjevskaja-Tamm & Muravjova 1993) g´mm´ t- akka-n-nalg´- n-kuww-at-av´-tk-´n I.NOM 1sg.S-son-CAUS-skin-CAUS-dry-SUFF-SUFF-PRES-1sg.S ‘I am making my son dry skins.’

Towards a structural typology of verb classes

6.4.

129

The recipient is treated like the object of a transitive verb

In this section I consider various ways of encoding the recipient similarly to the object of a transitive verb. In a system based on animacy, the animate recipient counts as more prominent than an inanimate theme. It therefore can be marked by accusative like the transitive object, or conversely, the theme, corresponding to the transitive object, can be demoted by oblique marking. In the positional linking type, the recipient takes the verb-adjacent position like the transitive object. Several tests such as passive and object agreement show that the recipient then indeed occupies the grammatical role of transitive objects. Therefore this solution of the third-argument problem can best be characterized by the term ‘primary object’, introduced by Dryer (1986). His clarification is welcomed in order to avoid confusion caused by undifferentiated uses of the notions ‘direct’ and ‘indirect’ object. Argument hierarchy is again the best background on which a useful notion of ‘object’ can be based. First one has to delimit the set of objects, which all are different from the highest argument (the subject) and therefore designated as +hr. Then, a primary object is defined as the highest object, whereas a direct object is defined as the lowest object. In this section we are concerned with the primary-object solution. 6.4.1. The recipient is the more salient object semantically If animacy plays an essential role in a language one expects that the recipient is treated alongside with transitive objects, but that the theme argument (prototypically inanimate) remains unmarked. This indeed can be observed in languages of the active-inactive type, as well as in languages with inverse marking. In section 3.4 regarding the active-inactive type we already saw that recipients are marked by inactive affixes on the Dakota verb (see (29)). Another example that illustrates this is (111). (111) Inactive recipient marking in Lakhota (Van Valin 1977: 19) Bill wówapi kin (miye) ma-k’u 1sg.INACT-give Bill book DEF 1sg ‘Bill gave me the book.’ In an inverse system (section 4.1), the grammatical role of animate arguments is determined by direct vs. inverse marking. Direct marking encodes

130 Dieter Wunderlich the less salient person as object, whereas inverse marking shifts its role to subject. The two animate arguments involved are either agent/ patient of a transitive verb or agent/recipient of a ditransitive verb. (112) shows a direct form of Fox (Algonquian), marking a 3pl recipient. Similarly, examples from Ojibwe (Algonquian) attest that the 1st person (the more salient one) is agent in the direct form (113a), but recipient in the inverse form (113b). (112) Recipient in Fox (Dahlstrom 1995: 4) ne-minw-aki ne- ish-aki m omakini. 1-give-3pl 1-uncle-pl that(obv).which.I.shot ‘I gave my uncles the game which I killed.’ (113) Recipient in Ojibwe (Dryer 1986: 812) a. n-gi:-mi:n-a: mzinhigan za:bdi:s. John 1-PAST-give-3.ANIM book ‘I gave John a book.’ b. n-gi:-mi:n-ik-a: mzinhigan za:bdi:s. 1-PAST-give-INV-3.ANIM book John ‘John gave me a book.’ Derived ditransitive verbs can add an animate causer, or companion, or beneficiary (114a–c), which all get preference. For instance, (115) illustrates that a 1st person beneficiary requires inverse marking. (114) Some derived ditransitives in Ojibwe (Valentine 2001: 435, 463 ff.) a. Causative b. Comitative baak-nam-oo-h wiid-oopo-m WITH-eat-WITH open-TRANS.ANIM-epenth-CAUS ‘get someone to open something’ ‘eat with someone’ c. Benefactive dkw-aabiit-maw short-stringlike.object-BEN ‘shorten something (stringlike) for someone’ (115) Benefactive in Ojibwe (Valentine 2001: 700) Aw kwe n-dazht-amaa-g n-babgiwyaan. 1P-shirt that woman 1-make-BEN-INV ‘That woman is making me a shirt.’

Towards a structural typology of verb classes

131

6.4.2. The recipient is marked by accusative A case that marks the object of a transitive verb is called ‘accusative’. Khasi, a Mon-Khmer language of Assam, uses ya for this purpose (116a). With a ditransitive verb, the accusative marker ya however switches to the third argument (the beneficiary, i.e., the person who profits from the action of teaching), while the original object remains unmarked (116b). (116) Accusative marking in Khasi (Dryer 1986: 816) a. /uu hiikay [ya ka ktien phare]. he teach ACC the language English ‘He teaches English.’ b. /uu hiikay [ya a] ka ktien phare. he teach ACC 1sg the language English ‘He teaches me English. In the West Tibetan language Kham, a 3rd person subject is ergativemarked, and an animate object is accusative-marked, while an inanimate object remains unmarked (117a,b); furthermore, the verb agrees with a 1st person object (117c). Now, turning to a ditransitive verb, exactly these options are combined: the 1st person recipient is marked like the corresponding patient, and the (inanimate) theme argument remains unmarked (117d). Ditransitive verbs thus show the case pattern ERG-ACC-NOM, which is typical for a mixed ergative-accusative language lacking dative. (117) Accusative-marking and object agreement in Kham (Dryer 1986: 817) a. nga: zihm nga-jxy-ke (inanimate object) 1sg house 1sg-build-PAST ‘I built a house.’ b. no-e ka:h-lay poh-ke-o. (animate object) 3sg-ERG dog-ACC beat-PAST-3sg ‘He beat the dog.’ c. no-e nga-lay cyu:-na-ke-o. 3sg-ERG 1sg-ACC watch-1sg-PAST-3sg ‘He watched me.’ d. no-e nga-lay bxhtanji ya-n-ke-o. 3sg-ERG 1sg-ACC potato give-1sg-PAST-3sg ‘He gave me a potato.’

132 Dieter Wunderlich Another possibility for ditransitive verbs is double accusative, which one finds in Yaqui, a Uto-Aztecan language of Mexico (118a). The passive data given in (118b,c) reveal that the recipient behaves as the primary object under passivization. (118) Double accusative and passive in Yaqui (Van Valin 2002: 20) a. Joan Peo-ta /uka vaci-ta miika-k. Juan Pedro-ACC DET.ACC corn-ACC give-PERF ‘Juan gave Pedro the corn.’ b. Peo /uka vaci-ta miik-wa-k. Pedro DET.ACC corn-ACC give-PASS-PERF ‘Pedro was given the corn.’ c. *U/u vaci Peo-ta miik-wa-k. DET.NOM corn Pedro-ACC give-PASS-PERF ‘The corn was given to Pedro.’ Moreover, a ditransitive verb can be causativized, which then yields three accusatives in series (119a). Again, only the highest object (which in this case is the causee) can become nominative under passive (119b). (119) Triple accusative and passive in Yaqui (Van Valin 2002: 20) a. U/u maejto usi-ta mansana-ta yoem-ta miik-tua-k. DET.NOM teacher child-ACC apple-ACC man-ACC give-CAUS-PERF ‘The teacher made the child give the man the apple.’ b. U/u usi mansana-ta yoem-ta miik-tua-wa-k. DET.NOM child apple-ACC man-ACC give-CAUS-PASS-PERF ‘The child was made to give the man the apple.’ Yindjibarndi, a Pama-Nyungan language of Australia, behaves similarly: The two objects of a ditransitive verb are realized by double accusative (120a), but only the recipient can become nominative under passive (120b,c). (120) Double accusative and passive in Yindjibarndi (Dryer 1986: 829) a. Ngaarta yungku-nha ngayu murla-yi. man give-past 1sg.ACC meat-ACC A man gave me the meat. b. Ngayi yungku-nguli-nha murla-yi ngaarta-lu. 1sg.NOM give-PASS-PAST meat-ACC man-INST I was given the meat by a man.

Towards a structural typology of verb classes

133

c. *Murla yungku-nguli-nha ngayu ngaarta-lu. meat give-PASS-PAST 1sg.ACC man-INST The meat was given to me by a man. 6.4.3. The Theme is marked obliquely Another way to treat recipients as primary objects is to mark the theme argument. In Tahitian, a Polynesian language with the basic word order VSO, the theme of a ditransitive verb is marked by the preposition / i, while the recipient remains the unmarked object (121). (121) Tahitian (Sedlak 1975: 153) /ua horo/a a:tu vau [/i te ho:/e tao/a] /ia:na. PAST give away 1sg OBL the one present 3sg ‘I gave him the present.’ 6.4.4. The recipient is the primary object in a double object construction Languages with positional SVO marking use a double object construction SVO1O2 for ditransitive verbs, in which O1 counts as the primary object, and O2 as the secondary object. As a rule, the primary object O1 is the more prominent one: it is affected by passive (i.e., becomes S), can be relativized or topicalized, or can be co-indexed with an object affix. Languages with this property, among others, are Chinese, Vietnamese, the Oceanic and the Bantu languages, as well as most of the creoles. For English we know two alternative constructions with ditransitive verbs, one is the double object (DO) construction, in which the recipient is the primary object, and the other is the prepositional object (PO) construction, in which the theme is the primary object, while the recipient (goal) is marked by the preposition to. As predicted by the above-mentioned rule, only the object adjacent to the verb (the primary object) is affected by passive, and it is preferred for relativization and topicalization, as shown in (80a,b). (122) Alternating ditransitive constructions of English a. He gave the woman a book. (double object: DO) Passive: The woman was given a book. Relative clause: The woman he gave a book is over there. Topicalization: The woman, he gave a book.

134 Dieter Wunderlich b. He gave the book to the woman. (prepositional object: PO) Passive: The book was given to the woman. Relative clause: The book he gave to the woman is over there. Topicalization: The book, he gave to the woman. The same is true for many other languages. As, however, Bresnan & Moshi (1990) have pointed out, apart from Bantu languages with asymmetric objects (Kikuyu, Chichea, Swahili) there exist also Bantu languages with symmetric objects (Sesotho, Kichaga, Kinyarwanda, Marathi). In the asymmetric object languages only the primary object is affected by passive or can be marked by object agreement, while in the symmetric object languages both objects can be affected. (By definition, there can be no SVO1O2 marking language in which the secondary object becomes subject under passive while the primary object does not.) Moreover, the possibility of a secondary object to become affected is mostly controlled by further semantic factors for avoiding ambiguity. Indeed, in a language with strict symmetric objects (both being animate) a certain amount of ambiguity is unavoidable. (123) shows for Sesotho of South Africa that the patient and the beneficiary (the latter introduced by the applicative operation) can appear in either order, which invites for two interpretations. In the passive, illustrated in (124), each of the two objects can become subject, which again produces ambiguity, i.e., both (124a) and (124b) have two interpretations. Thus, Sesotho is one of the rare languages that have no clear grammatical mechanism for distinguishing between a second and a third argument. One expects, however, that the speakers will avoid these ambiguities by contextual cues, or by encoding different information structures (topic, focus). (123) Symmetric objects in Sesotho, inviting for two interpretations (Lee 2000) Sello o-shap-el-a Lineo bashanyana. / bashanyana Lineo. Sello I.SBJ-beat-APPL-FV Lineo boys i. ‘Sello beats the boys for Lineo.’ ii. ‘Sello beats Lineo for the boys.’ (124) Symmetric passive in Sesotho, inviting for two interpretations (Lee 2000) a. Morena ó-bítsel-íts-o-é baná. chief I.SBJ-call-APPL-PASS-PAST children

Towards a structural typology of verb classes

135

b. Baná ba-bítsel-íts-o-é morena. children II.SBJ-call-APPL-PASS-PAST chief i. ‘The children benefitted from having the chief called.’ ii. ‘The chief benefitted from having the children called.’ In contrast, asymmetric object languages make a clear distinction between primary and secondary object. The beneficiary object of Kikuyu (introduced by the applicative operation) must be verb-adjacent, thus, the reading of (125a) crucially differs from that of (125b). (125) Asymmetric objects in Kikuyu (Lee 2000) a. Karioki O:n-E-irE gari dereba. Karioki SBJ.find-APPL-PAST car driver ‘Karioki found a driver for the car.’ b. Karioki O:n-E-irE dereba gari. Karioki SBJ.find-APPL-PAST driver car ‘Karioki found a car for the driver.’ However, if it comes to passive, alternation is possible, provided that the semantic role of the participants can be inferred from the meaning of the verb. In (126) it is clear that the flower is given to the teacher because it would be bizarre to give a teacher to a flower. (126) Alternating passive in Kikuyu (Lee 2000) a. Moarimo ne a-hE-ir-uO ihoa ne mo:do. teacher FOC SBJ-give-PAST-PASS flower by man ‘The teacher was given the flower by the man.’ b. Ihoa ne re-hE-ir-uO moarimo ne mo:do. flower FOC SBJ-give-PAST-PASS teacher by man ‘The flower was given to the teacher by the man.’ As one can observe, the positional linking of Bantu is supported by subject agreement, and, under special circumstances, also by object agreement. This could be the reason why these languages admit so much structural variation under the influence of topic and focus. In the symmetric languages it is even possible that SVO switches with OVS, see the possible readings of (124). More precisely, we are probably dealing here with STVOF vs. OTVSF, where the topic is preverbal and the focus postverbal, regardless of what the subject and what the object is.

136 Dieter Wunderlich Can we account for primary objects in a more formal way? I think that a simple logical consideration can guide us. Recall from (70), section 4.4 above, that –hr characterizes the highest argument and –lr the lowest argument from two different perspectives (what is higher, or, what is lower?). This yields the representation in (127b) below, in which the medial argument is the most marked one (+hr,+lr); it is therefore called ‘indirect’ object according to a long tradition, based on case-marking languages such as Latin or Greek. This is a fine solution in the presence of generalized case, which is independent of linear ordering. Positional argument linking, however, uses linear ordering, which is always from left to right and makes a second perspective from right to left dubious. Thus, the feature +lr seems unsuitable. The concept of linear ordering rather suggests something like the feature +hr to be applied iteratively. In a first step, all objects are assigned +hr, while the highest argument remains –hr. In a second step, all lower objects are assigned +ho (‘there is a higher object’), and the highest object remains –ho. (127a) shows how this procedure yields the intended result. The lowest object comes out to be the most marked one (+hr,+ho), so that the medial argument can truly be characterized as the primary object. Notice that most grammatical constructions affecting objects pick out the least-marked object, be it the primary or the direct one. (127) Encoding primary vs. indirect object a. z y x {ACT(x) & BECOME POSS(y,z)} +hr +hr –hr +ho –ho secondary primary object object b. z y x {ACT(x) & BECOME POSS(y,z)} +hr +hr –hr –lr +lr +lr direct indirect object object There can be only these two ways of encoding a hierarchy, so that a fourth or fifth argument would not alter the typology based on the choice between +ho and +lr as the second feature. Usually speakers do not like too much complexity, even if it is grammatically possible. However, constructions with four structural arguments have been tested by researchers, see example (119) above, and also chapter 6 of Joppen-Hellwig (2001).

Towards a structural typology of verb classes

137

Given this more precise concept of primary object, how can we integrate the various observations made in this section? In a Bantu language with symmetric objects, the feature +ho is ignored, so that both objects are undistinguishably +hr. In a language with double or even triple accusative such as Yaqui (118) (119), the feature value –ho is decisive to single out the highest object. A language in which the theme is obliquely marked as in Tahitian (121), obviously is compatible with this account because the lowest argument is most marked. Differently is Khasi (116), which picks out the highest object to be marked by accusative: either because it is more animate or because it is judged as the indirect object – that can only be decided by further inspection. If, however, the language has ergative (+lr), the feature +lr must be present. Therefore, the ERG-ACC-NOM pattern shown above for Kham (117) is not an instance of primary object marking in the sense of (127a), it rather follows from the role of animacy in this language. In terms of animacy, the recipient mostly outranks the theme. Therefore, the primary object status of recipients in active or inverse type languages follows from independent reasons. One could even speculate that animacy is the source for generalizing the structural notion of primary object. This conclusion makes the introduction of an indirect object, to be discussed in the next section, a puzzle. As we will see, there are at least two forces working against the primary object solution: the existence of ergative (and the kind of featuremarking it implies), and the fact that recipients are added to transitive verbs, and thus, intuitively, are the more marked objects. 6.5.

The recipient is treated differently from the object of a transitive verb

This section deals with various solutions of the third-argument problem in which a special status of the third argument is accepted. It is either marked as the goal, or indirectly as the possessor of the theme, or by an extra case, the dative. The two latter solutions indeed justify the notion of indirect object. Every mechanism of realizing transitive verbs remains as before, but is augmented by an additional mechanism. 6.5.1. The recipient is marked as the goal by means of a preposition A very common way is marking the recipient as the goal, i.e. by means of a local preposition. Such a device usually exists independently of ditransitive

138 Dieter Wunderlich verbs, but can be generalized to cover also recipients. This is illustrated by an example from Tamazight, a Berber language, in (128). The same option is found in the English PO construction. (128) Goal-marking in Tamazight (Faltz 1978: 77) i- a urgaz le Ta i Tmattutt 3msg-give man book to woman ‘The man gave the book to the woman.’ 6.5.2. The recipient is marked as an alienable possessor of the theme In contrast, the option in which the recipient is realized as the possessor of the theme is very rare cross-linguistically. It is found in Tongan, a Polynesian language, as a second option, which in this language, however, is not rare. The construction in (129b) literally means ‘I give your food’; ha’o is an alienable possessor which syntactically belongs to the following noun. (129) Tongan: Alternation between goal and possessor marking (Broschart, Lambert p.c.) a. na’e ‘oange ‘e Pita ha tohi [kia Mele]PP. PAST give.away ERG Peter ART.UNSPEC book DIR Mary ‘Peter gave Mary a book.’ b. ‘oua mu’a ke u ‘oatu [ha’o me’atokoni]NP. wait first ERG 1sg give 2sg.POSS.AL food.POLITE ‘Wait first before I give you some food.’ A similar ‘recipient lowering’ construction is discussed by Van Valin (2002) for Dyirbal, an Australian language. Dyirbal allows the alternation between dative (130a) (see also below), oblique (instrumental) marking of the theme (130b), and marking the recipient as the possessor of the theme (130c). Here, ‘the woman gave the man’s beans’ exactly means what English expresses by the sentence the woman gave the beans to the man. The endings on the demonstratives refer to the nominal class to which the following noun belongs (I, II, or III).

Towards a structural typology of verb classes

139

(130) Dyirbal (Van Valin 2002: 14, 22; Dixon 1972) a. Ba-la-m mira
ba-gu-n yugumbi-ru wuga-n DEM-NOM-III beans DEM-ERG-II woman-ERG give-TNS ba-gu-l yara-gu. DEM-DAT-I man-DAT ‘The woman gave the beans to the man.’ b. Ba-yi yara wuga-n ba-gu-n dyugumbi-ru DEM-NOM.I man give-TNS DEM-ERG-II woman-ERG ba-gu-m mira
-dyu. DEM-INST-I beans-INST ‘The woman gave the man beans.’ c. Ba-la-m mira
ba-gu-n dyugumbi-ru wuga-n DEM-NOM-III beans DEM-ERG-II woman-ERG give-TNS ba-gu-l yara-u. DEM-INST-I man-GEN ‘The woman gave the mani’s beans (to himi).’ Conversely, it is also possible that the theme is marked as possessed by the recipient, provided that the language has a proprietive case, which however is rare and only found in some Australian languages, for example, in Kayardild. In (131), nguki-wuru has to be translated as ‘having water’, so that the construction literally means ‘(I) give iti to the men having wateri’. (131) Kayardild (Australian, Van Valin 2002: 24) …nguki-wuru wuu-ja dangka-y. water-PROP give-ACT person-MLOC ‘… [and I] will provide mankind with water’ At first glance, both constructions, the possessor marking of the recipient and the proprietive marking of the theme, support our analysis that recipients are intended possessors. It is not totally unexpected that the realization of a component of the verb is delegated to one of the arguments. The proprietive solution of Kayardild can best be compared with the delegation of LOC to an external PP, which we found in (81) above. The respective compositions are schematized in (132). The proprietive in this function is so rare because the proprietive itself is rare, compared to local prepositions.

140 Dieter Wunderlich (132) Composition with a delegated predicate a. PUT(x,z) & P(z) yields PUT(x,z) & BECOME LOC(z, AT y) b. GIVE(x,y) & P(y) yields GIVE(x,y) & BECOME POSS(y,z) (or WITH(y,z)) However, such a composition is not possible for the possessor marking of the theme because the possessor is the higher argument of POSS, and the noun that bears the possessor is a sister argument of the recipient. We have to conclude that some inference must be made: ‘A possessor of the theme can be identified with the intended possessor.’ A possessor of an argument is often delegated to the verb (by the so-called possessor raising), which is the opposite of recipient lowering. Consider the following examples from Choctaw, a Muskogean language. Each of the verbs in (133) has four arguments; the dative that is nearer to the stem marks the recipient, and the other marks the possessor of the theme. (The word order of the syntactic arguments follows the argument hierarchy; only the subject is marked by nominative case, all other structural arguments can be left unmarked.) (133) Possessor raising in Choctaw (Davies 1986: 54) a. Hattak-at ohoyo iskali am-im-a:-tok. man-NOM woman money 1.DAT-3.DAT-give-PAST ‘The man gave my money to the woman.’ b. Alla towa chim-i-pila-li-tok. child ball 2.DAT-3.DAT-throw-1.NOM-PAST ‘I threw your ball to the child.’ We have good reason to believe that possessor raising is a generalization enabled by the existence of a recipient dative (or accusative). On the other hand, possessors found on nouns are not readily generalized to cover also recipients as a sister argument of the noun. Summarizing, this option is not especially advantagous, in contrast to what we turn to now. 6.5.3. The recipient is realized by dative The introduction of dative as a special case for the ‘indirect object’ is very common, not only in the Indo-European languages. The following two examples show that dative is compatible with both a pure accusative and a

Towards a structural typology of verb classes

141

pure ergative system. Japanese is an accusative language of the SOV-type; all cases are marked by a postposition, and ni is a postposition that marks certain locative arguments but also the recipient of ditransitive verbs (134a), as well as the causee of a causativized transitive verb (134b). In contrast, Basque is an ergative language of the SOV-type; ergative and dative are marked by a suffix on the noun, and the final auxiliary agrees with all three arguments. Again, the recipient of a ‘give’ verb (135a) and the causee of a causativized transitive verb (135b) are encoded by dative. (134) Dative marking in Japanese a. otoka-ga onna-ni hon-o age-ta. man-NOM woman-DAT book-ACC give-PAST ‘The man gave the book to the woman.’ b. John-ga Mary-ni sakana-o tabe-sase-ta. John-NOM Mary-DAT fish-ACC eat-CAUS-PAST ‘John let Mary eat the fish.’ (135) Dative marking in Basque a. Ni-k pobre-ei diru-a ema-ten diet. 1sg-ERG poor-pl.DAT money-DET give-IMPERF AUX.3N.3pD.1E ‘I give money to the poor.’ b. Ama-k haurr-a-ri zopa jan-araz-ten dio. mother-ERG child-DET-DAT soup.DET eat-CAUS-IMPERF ‘The mother lets the child eat the soup.’ Dative is also possible in a pure head-marking language such as Yimas, a polysynthetic language of New Guinea that lacks morphological case. In Yimas, pronominal affixes are attached to the verb; they belong to different sets realizing the function of case. Dative affixes are restricted to 3rd person (136a), otherwise the recipient is marked by an accusative affix (136b). (The nominative affixes agree with the respective class of the noun, such as IV or VI.) (136) Alternation between dative and accusative marking in Yimas (Foley 1991) a. irpm mu- ka- tkam-t -nakn. coconut palm.IVsg IVsg-1sgE-show-PERF-3sgD ‘I showed him a coconut palm.’

142 Dieter Wunderlich b. ura k- mpu- a- tkam-t coconut.VIsg VIsg-3plE-1sgA-show-PERF ‘They showed me the coconut.’ Many more languages display a canonical case pattern for ditransitive verbs, in the sense outlined above in (70) (section 4.4): NOM – DAT – ACC in accusative systems, and ERG – DAT – NOM in ergative systems. On the basis of the two features +hr and +lr, dative is indeed maximal; it is not possible to encode a fourth argument by a further structural case. In other words, the feature lattice for structural cases in (137), based on argument hierarchy, is complete. (137) Four possible structural cases dative accusative nominative

[+hr,+lr] [+hr]

[+lr] [

ergative

]

The reader may ask about genitive: Isn’t it a structural case, too? I consider genitive to be encoded like accusative but restricted to nouns; usually it marks the possessor of the noun (see Stiebels this volume). Only if the genitive enters the verbal system in addition to accusative, it offers a special semantic reading, for instance, a partitive or a negative polarity reading. In languages such as Russian and Finnish, the genitive/partitive acquired many functions of the accusative, which could happen because the two cases bear the same feature +hr. In the remainder of this section I first consider a case in which dative is just one option (Inuit), and then turn to two instances of dative-accusative syncretism, where dative and accusative have collapsed into one marker (Hindi and Georgian). 6.5.4. Alternative options for ditransitive verbs At the beginning of section 6 I said that many languages have more than one option to express a third argument. One example is Dyirbal, demonstrated in (130). The two case patterns shown in (130a,b) are also found in Inuit (West Greenlandic), in which this alternation is subject to an interesting re-

Towards a structural typology of verb classes

143

striction. The basic ditransitive verbs of Inuit encode the recipient like the object of a transitive verb, and the theme argument is marked by instrumental (as in Dyirbal (130b)), see (138a). This seems to be a lexical property of these verbs because all derived ditransitives verbs mark the medial argument by dative, thereby showing the canonical case pattern of ditransitives. Surprisingly, all basic ditransitives can be shifted into this canonical pattern, too, by suffixation with -ut, as shown in (138b). It is questionable whether -ut has any semantic contribution. The portmanteau agreement suffixes on the verb relate to the NPs showing ERG/NOM. In the examples below, the verb ending therefore shifts from 3pl in (138a) to 3sg in (138b). (138) Basic ditransitives in Inuit (Bittner 1994: 20) a. Juuna-p miiqqa-t atuakka-mik nassip-p-a-i. J.-ERG child-PL book-INST send.IND-TR-3sg/3pl ‘Juuna sent the children a book.’ b. Juuna-p atuagaq miiqqa-nut nassi-up-p-a-a. J.-ERG book child-DAT send-UT-IND-TR-3sg/3sg ‘Juuna sent a book to the children.’ Both variants of such a ditransitive verb can be causativized. The causee, then, is expressed by dative, and the theme argument is realized by either instrumental or nominative, depending on whether -ut is applied or not. The preferred position of arguments is always ERG – NOM (the two arguments the verb agrees with), followed by the other objects according to their hierarchy. Therefore, (139a) exhibits the ordering ERG – NOM – DAT – INST, and (139b) the ordering ERG – NOM – DAT – DAT. Because of this strict ordering, the double dative in (139b) does not lead to any confusion. (139) Causative in Inuit (Bittner 1994: 86) a. Kaali-p qimmi-t irnir-mi-nut puisi-mik nirlir-sip-p-a-i. K.-ERG dog-PL son-3sgP-DAT seal-INST feed-CAUS-IND-TR-3sg/3pl ‘Kaali let his son feed the dogs seal meat.’ b. Kaali-p puisi irnir-mi-nut qimmi-nut nirli-ut-sip-p-a-a. K.-ERG seal son-3sgP-DAT dog-DAT feed-UT-CAUS-IND-TR-3sg/3sg ‘Kaali let his son feed the dogs seal meat.’ A suffixal verb of Inuit adds some other higher predicate (such as ‘believe’, ‘want’, ‘think’), so that the subject of the embedded predicate becomes medial: it is always realized by dative. Again, an instrumental appears if a basic

144 Dieter Wunderlich ditransitive verb is affected. (140a) illustrates the embedding of a transitive verb, which leads to the surface pattern ERG – NOM – DAT, while (140b) illustrates the embedding of simple ‘give’. (140) Suffixal verbs in Inuit (Bittner 1994: 59f) a. Aani-p miiqqa-t Juuna-mut paasi-sur-a-i. understand-think.IND-TR-3sg/3pl A.-ERG child-PL J.-DAT ‘Aani thinks that Juuna understands children.’ b. Kaali-p miiqqa-t Maala-mut atuakka-mik tuni-qqu-va-a-i. K.-ERG child-PL M.-DAT book-INST give-want-IND-TR-3sg/3pl ‘Kaali wants Maalat to give the children a book.’ A summary is given in (141). (141) Ditransitive verbs in Inuit (West-Greenlandic): the verb agrees with ERG and NOM z +hr –lr

y +hr +lr

x –hr +lr

derived

NOM

DAT

ERG

basic ‘give’

INSTR

NOM

ERG

VERB(x,y,z)

As we have seen, dative indicates the medial position of an argument also in Inuit. If typologically and genetically different languages behave so similar, there is reason to believe that the correlation dative medial argument reflects some basic structural property of human language. My explication is that the encoding of argument hierarchy yields two possible features (+hr and +lr), which, if they are combined, automatically lead to this correlation.

6.5.5. Dative-accusative syncretism I briefly show for two languages (Hindi and Georgian) that even if the distinction between dative and accusative disappears, a language can retain its indirect object type (i.e., the recipient is treated differently from the object of a transitive verb). Hindi has the overt object clitic ko, which traditionally is called sometimes accusative and sometimes dative, depending on the further context (Mohanan 1994; Butt 1995). Similarly, Georgian has the overt

Towards a structural typology of verb classes

145

object suffix -s, which traditionally is called dative. For the sake of clarity, I will gloss these morphological elements as accusative markers, specified as +hr. Both Hindi and Georgian show an accusative-ergative split (see (65) for Hindi), which might have influenced the collapse of dative and accusative. In both languages it is the passive that can indicate the indirect object. In the passive of ditransitive verbs in languages such as German or Japanese the indirect object (the medial argument) remains dative, whereas the direct object (the lowest and therefore less marked argument) is shifted to nominative, as shown in (142). (This shift is forced by the requirement that at least one argument should be realized in the nominative.) Similarly, in both Hindi and Georgian the accusative of the medial argument remains in the passive. (142) Passive of ditransitive verbs in German and Japanese14 a. Ihm wurde ein Orden überreicht. he.DAT AUX.PASS a decoration presented ‘A decoration was presented to him.’ b. Hon-ga onna-ni age-rare-ta. book-NOM woman-DAT give-PASS-PAST ‘A book was given to the woman.’ Let us first consider the case-marking of ditransitive verbs in Hindi. (143a) shows a basic ditransitive in the perfect, (143b) a derived ditransitive verb in the perfect, and (143c) the imperfect. The reader will at once observe that the various salience conditions (such as animacy and definiteness) determining the alternation between accusative and nominative, discussed above with regard to (65), do not apply here; on the contrary, the lowest argument is always marked nominative, and the medial argument is always marked accusative. The explanation is that double accusative (which could result from the salience conditions) is forbidden in Hindi, so that the more marked argument role (the medial one) is uniformly preferred to be associated with the more marked case (the accusative).

———–————————————————————————————

14

Japanese passive has several functions. It is also possible that the indirect object becomes nominative, similar to the German kriegen-passive: Er (NOM) kriegte einen Orden (ACC) verehrt ‘he got presented a decoration’.

146 Dieter Wunderlich (143) Ditransitive verbs in Hindi (Mohanan 1994) a. Ravii-ne baalak-ko/*baalak baccaa/*bacce-ko diy-aa. Ravi-ERG boy-ACC/*boy.NOM child.NOM/*child-ACC ‘Ravi gave a/the child to a/the boy.’ b. Ravii-ne gaay-ko/*gaay kelaa/*kele-ko Ravi-ERG cow-ACC/*cow.NOM banana.NOM/*banana-ACC khilaay-aa eat.CAUS.PERF-M ‘Ravi fed a/the cow a/the banana.’ c. Ravi mãã-ko baccaa detaa hai. Ravi.NOM mother-ACC child.NOM give.IMPERF be.PRES ‘Ravi gives a/the child to the mother’ In the passive of these verbs, nothing changes in the realization of case, except that the highest argument is unrealized. (144) Passive of ditransitive verbs in Hindi mãã-ko baccaa diyaa gayaa. mother-ACC child.NOM give.PERF go.PERF ‘The mother was given a/the child’ Thus, the lack of a distinctive dative and the ban against double accusative lead to a simplification of the canonical ditransitive patterns, summarized in (145). Note that the perfect still exhibits three distinct cases, a fact that could have played a role for the dative-accusative collapse to emerge. (145) Ditransitive verbs in Hindi: the verb agrees with the higher NOM z +hr –lr

y +hr +lr

x –hr +lr

perfect

NOM

ACC

ERG

non-perfect

NOM

ACC

NOM

passive

NOM

ACC

*

VERB(x,y,z)

Let us now turn to Georgian, which behaves in many respects similar to Hindi. Like Hindi, Georgian has developed an ergative that is restricted to some sort of past or completion. The Georgian ergative is exclusively found in the so-called aorist series (of tense-aspect-mood combinations),

Towards a structural typology of verb classes

147

exemplified in (146a). In all other combinations the ergative is blocked. Differently from Hindi, the present series of Georgian shows double accusative (146b). However, in the passive it is the recipient that remains accusative (146c). (146) Aorist, present and passive in Georgian (Joppen-Hellwig 2001: 50) a. Ketino-m Eka-s xalit a a-uk-a. Ketino-ERG Eka-ACC carpet V-present-AOR.3N ‘Ketino presented Eka with a carpet.’ b. Ketino Eka-s xalit a-s s-ukni-s. Ketino Eka-ACC carpet-ACC 3D-present-PRES.3N ‘Ketino presents Eka with a carpet.’ c. xalit a e-uk-eb-a Eka-s. carpet PASS-present-TH-PRES.3N Eka-ACC ‘The carpet is presented to Eka.’ Georgian exhibits a third series of tense-aspect-mood, which is best described as evidential (encoding a hear-say constellation) and illustrated in (147). Here, the highest argument is marked by +hr (a situation that we found above – in section 5.1 – as characteristic for experiencer verbs). Since Georgian does not allow the combination of two accusatives of animate nouns, the medial argument gets instead expressed by a semantic case. (147) Evidential in Georgian Ketino-s Eka-tvis xalit a u-uk-eb-i-a. Ketino-ACC Eka-for carpet V-present-TH-EVID-3N ‘As one says, Ketino presented Eka with a carpet.’ The agreement facts of Georgian would deserve their own chapter (see Wunderlich 1996b; Joppen-Hellwig 2001). It is enough to say that the agreement system is always accusative-based: the verb agrees with the subject and a 1st or 2nd person (highest) object. In addition, consonant-initial verb stems show dative agreement for the 3rd person, see (146b). The lexical feature of the evidential morpheme forces both case reversal and agreement reversal in this mood. The table in (148) summarizes the case patterns of ditransitive verbs found in Georgian. As in Hindi, the aorist pattern exhibits three distinct structural cases.

148 Dieter Wunderlich (148) Ditransitive verbs in Georgian: the verb agrees with NOM and the higher ACC z +hr –lr

y +hr +lr

x –hr +lr

aorist

NOM

ACC

ERG

present

ACC

ACC

NOM

passive

NOM

ACC

*

evidential

NOM

semantic case

ACC

VERB(x,y,z)

Georgian also allows derived ditransitive verbs: the initial vowel of a verb can function as an applicative morpheme. In the literature on Georgian, a distinction is made between the objective version, in which a possessor or beneficiary is added (149a), and the superessive version, in which a location (an object on which something is placed) is added (149b). As expected, we find double accusative in the present. (149) Objective and superessive version in Georgian (Joppen-Hellwig 2001: 117) a. mzia u-cmend-s da-s pexsacml-eb-s. Mzia APPL-clean-PRES.3N sister-ACC shoe-PL-ACC ‘Mzia cleans the shoes for her sister.’ b. Mama a-cer-s misamart-s konvert-s. father APPL-write-PRES.3N address.ACC envelope-ACC ‘Father writes an address on the envelope.’ The passive of these applicatives again preserves the case of the medial argument, as illustrated in (150) and (151). Thus, the medial argument indeed is more marked than the lowest argument. Another, so far unmentioned fact of Georgian is the behavior of plural agreement: in this regard, the verb always agrees with the highest argument rather than with a nominative one, that is, in the passive the verb agrees with the higher object. Therefore, the verb is plural in (150a), but singular in (150b), complying with the number of the beneficiary, and in opposition to what the respective nominative phrases reveal about number.

Towards a structural typology of verb classes

149

(150) Passive of the objective version in Georgian (Joppen-Hellwig 2001: 132) a. deb-s pexsacmel-i e-cmind-eb-a-t. sister.PL-ACC shoe-NOM APPL.PASS-clean-TH-PRES.3N-PL ‘The shoe is cleaned for the sisters.’ b. da-s pexsacml-eb-i e-cmind-eb-a. sister-ACC shoe-PL-NOM APPL.PASS-clean-TH-PRES.3N ‘The shoes are cleaned for the sister.’ The same choice of number agreement can be observed in the passive of the superessive version, shown in (151). (151) Passive of the superessive version in Georgian (Joppen-Hellwig 2001: 133) a. konvert-eb-s misamart-i e-cer-eb-a-t. envelope-PL-ACC address-NOM APPL.PASS-write-TH-PRES.3N-PL ‘The address is written on the envelopes.’ b. konvert-s misamart-eb-i e-cer-eb-a. envelope-ACC address-PL-NOM APPL.PASS-write-TH-PRES.3N ‘The addresses are written on the envelope.’ This clearly indicates that ‘envelope’ is a higher argument than ‘address’, and thus behaves as a ‘possessor’ of the address rather than its location, contrary to what the notion ‘superessive’ suggests. Again we see that structural generalization can go beyond semantic boundaries. Above we observed (with respect to Nivkh) that the grammatical effects of the predicate LOC are generalized to instances of change of possession (section 6.3, (106)), and here we can observe that the predicate POSS is generalized to instances of change of location. Both are unexpected from a purely semantic perspective. Observations like these make the study of languages a fascinating topic: not only differ the surface forms from language to language but also the ranges of semantic templates shaping the contents. Within certain limits, languages differ in their semantic perspectives. In discussing two alternatives of expressing three-place concepts (and the corresponding grammatical means for realizing the three arguments), I claimed that the predicates POSS and LOC offer the basic semantic machinery to perform this task. Consequently, each language has to find a proper generalization for constructions

150 Dieter Wunderlich based on either POSS or LOC, so that the borderline between the POSS field and the LOC field is not determined by independent semantic considerations but rather by language-specific grammaticalization. However, grammaticalization processes are not arbitrary. If a language develops alternative constructions, there is usually some inherent division of labor at work. English has developed two constructions for ditransitive verbs (based on POSS, and LOC), and the choice between them follows a clear semantic gradience, as we will see in the next section. 6.6. The English ‘dative’ alternation Above I postulated two types of ditransitive verbs from a semantic point of view, change of possession and change of location verbs. Usually these two types are distinguished by formal means. English uses these means not only for a distinction between two classes of ditransitives, but also as alternative realizations for most individual verbs. The two alternatives are the double object (DO) construction and the prepositional object (PO) construction, and the alternating verbs are said to undergo the ‘dative alternation’ (a somewhat misleading term because English doesn’t have a dative.) A question much discussed in the literature is whether this alternation follows from the semantic difference between change of possession and change of location. Pinker (1989), Krifka (2004) and others have argued that this indeed is the case. However, other authors are sceptical. How can it be possible that the verb give shifts it semantic reading from (152a) to (152b)? (152) DO-PO alternation in English a. Anna gave Max a photo. b. Anna gave a photo to Max.

(DO) (PO)

The DO-PO alternation is found rather frequently. Bresnan & Nikitina (2003) observed that there are only few verbs that totally resist the DO-PO alternation. The DO construction often is possible with a pronominal recipient (153c), while it is blocked with a nominal recipient (153a). Conversely, the PO construction can be possible with pronouns (154c), while it is blocked with nouns (154a).

Towards a structural typology of verb classes

151

(153) DO is possible with pronouns in verbs of imparting a force (push, pull, carry, lift, lower), and verbs of communication (whisper, yell, mumble, mutter) (Bresnan & Nikitina 2003) a. *Susan pushed John the box. *Susan whispered Rachel the news. b. Susan pushed the box to John. Susan whispered the news to Rachel. c. Susan pushed him the chips. Susan whispered me the answer. (154) PO is possible with pronouns (Bresnan & Nikitina 2003) a. The car cost Beth $5000. b. *The car cost $5000 to Beth. c. It would cost nothing to the government. Similar observations can be made with respect to definiteness, topichood, length of expression etc. The more definite, topical or shorter the expression of the recipient is the better it fits with the DO construction. Bresnan & Nikitina (2003) therefore argue that the choice of construction has to do with other factors than a difference in meaning. On the contrary, I think that it is exactly a difference in the semantic representations that helps us to understand the observations made by Bresnan & Nikitina. Let me first say that these authors overestimate the difference in meaning. The predicates POSS(y,z) and LOC(z, AT y) are weakly equivalent because usually if one of them is true, the other is true, too. If z is located at y, then y is able to exert some ownership on z. Conversely, if y has possession of z, than z must be located in some neighborship to y for y being able to exert his possessorship. For this reason, many languages can express possession by means of a locative preposition. (155) shows this for Russian. (155) Possession in Russian: a. U menja kniga. at me.GEN book ‘I have the book.’

b. U nego bylo mnogo druzej. at him.GEN was many friends.GEN ‘He had many friends.’

Moreover, the POSS  LOC alternation is generally possible in languages such as German and English, as exemplified in (156).

152 Dieter Wunderlich (156) POSS-LOC alternation in German and English a. Das Haus hat drei Bäder. The house has three bathrooms. b. In dem Haus sind drei Bäder. There are three bathrooms in the house. Therefore, no logical reasons prevent us from interpreting a change of possession as a change of location, and vice versa, given that the verbs allow that alternation. The difference in meaning mainly reduces to a difference in argument hierarchy. The possessor is the higher argument of POSS, while it participates in the lower argument of LOC. (Because the possessor only participates in the lower argument of LOC, the predicates POSS and LOC are only weakly equivalent.) In German, dative verbs compete with verbs that have a full PP argument (with LOC being external to the verb), which has been exemplified with schicken ‘send’ in (83) above (section 6.1). The dative verb has three nominal arguments, with the recipient as the medial one (157a), while the PP verb has a prepositional phrase (related to the goal) as the third argument (157b). Generally, a PP argument occupies the lowest position. (157) Dative verbs vs. PP verbs in German a. Dat: z y x {ACT(x) & BEC POSS(y, z)} b. PP:

ACC

DAT

NOM

P PP

z

x

ACC

NOM

{ACT(x,) & P(z)}

In the English PO construction, the preposition is fixed, which can be considered a formal reflex of LOC being incorporated in the semantic representation of the verb. Hence, both variants of English have three nominal arguments, with the only difference that the recipient is medial in DO (158a) but the lowest argument (shaped as a goal) in PO (158b). (158) Double object and prepositional object in English : a. DO: z y x {ACT(x) & BEC POSS(y, z)} secO primO b. PO: y z x {ACT(x) & BEC LOC(z, AT y)} to primO

Towards a structural typology of verb classes

153

These decomposed representations allow us to make some important predictions. In the DO construction, the recipient should behave consistently as the higher object: it can bind a reflexive theme, can license a negative polarity item (such as any), can be moved in multiple questions, etc., which is true according to the several tests applied by Larson (1988). In contrast, in the PO construction the recipient (shaped as a goal) should behave consistently as the lower argument: it can be bound by a theme, cannot be moved in multiple questions, etc. This is demonstrated here for binding and scope. A quantifier in the higher argument can bind the possessor of a lower argument, but not vice versa, as shown in (159) and (160). The recipient binds the possessor of the theme in the DO construction (159a), and the theme binds the possessor of the recipient=goal in the PO construction (160a), while the respective reversed bindings are excluded (159b, 160b). (159) DO construction a. He gave every womani heri baby. Which womani did you give heri baby? b. *He gave itsi mother every babyi. (160) PO construction a. He gave every babyi to itsi mother. Which babyi did you give to itsi mother? b. *He gave heri baby to every womani. Similarly, the higher argument has scope over the lower argument; scope reversal is a highly marked option and scarcely acceptable, as shown in (161b) and (162b). The theme can be distributed to multiple recipients in the DO construction, where each girl gets her own telescope in (161a). Conversely, the recipient (=goal) can be distributed to multiple themes in the PO construction, where each telescope ends up at some girl in (162a). (161) DO construction (Bresnan & Nikitina 2003) a. Ozzy gave each girl a telescope. y
z {ACT(ozzy) & BEC POSS(y,z)} b. *Ozzy gave a (different) girl each telescope. z
y {ACT(ozzy) & BEC POSS(y,z)}

154 Dieter Wunderlich (162) PO construction (Bresnan & Nikitina 2003) a. Ozzy gave each telescope to a girl. z
y {ACT(ozzy) & BEC LOC(z,AT y)} b. ??Ozzy gave a (different) telescope to each girl. y
z {ACT(ozzy) & BEC LOC(z,AT y)} The examples (159) to (162) clearly indicate that both constructions have their own semantic potential to express more complex states of affairs, and therefore it is justified to represent them differently. We are now in the position to come back to the observations made by Bresnan & Nikitina. The representations in (158) let us predict an asymmetric distribution of the two constructions according to the contextual features listed in (163). (163) Differential object marking in ditransitive verbs more salient less salient animate inanimate 1st or 2nd person 3rd person pronoun full noun definite indefinite short expression long expression topic focus Each of these contextual features forms a salience scale, which has to be associated with the argument hierarchy found in the DO or PO construction. The concept of harmonic alignment of scales then suggests the following: The more salient an argument is the higher its position should be in argument hierarchy, given that there is a choice. This amounts to the competition scenario in (164). (164) DO-PO competition: a. If the recipient is more salient than the theme, DO is chosen (alternatively, PO is blocked). b. If the recipient is less salient than the theme, PO is chosen (alternatively, DO is blocked). This prediction turns out to be true. Collins (1995) calculated several factors determining the choice of construction on a broad text basis. He found the following distribution in the DO construction. (In the PO construction, the (reversed) preferences are less striking – around 1.5).

Towards a structural typology of verb classes

155

(165) Preferences in the DO construction (Collins 1995: 47) Recipient Theme focus 1 90 topic 14 1 pronoun 11 1 definite 4.5 1 longer 1 3 In the DO construction the recipient is more likely to be the topic (14 times), a pronoun (11 times), or definite (4.5 times) than the theme, but it is less likely to be in the focus or the longer expression. Bresnan & Nikitina (2003) present much more data pointing in the same direction, but are still arguing for one underlying semantic representation. However, the fact that these asymmetries can be predicted by harmonic alignment should convince us that different argument hierarchies play a role. The essential function of the lexical decompositions given in (158) is to provide us with a principled way of stating argument hierarchies: the recipient is the higher object in the DO construction but the lower object in the PO construction. Collins’ data also indicate that DO is the marked alternative because it leads to a sharper profile under varying contextual circumstances. For this reason, a child who is confronted with a mixed input of DO and PO data could detect the advantage of the DO construction (in the sense outlined in (164)) more easily than that of the PO construction. This is in accordance with Adone’s (2002) observation that Creole-speaking children of the Seychelles overgeneralize the DO construction more readily than the PO construction. Another more general conclusion can be drawn. The existence of two alternative constructions has the advantage that the speakers can give different contextual weights to an argument. In the choice between DO and PO the speakers weigh the relative salience of the two objects, which is similar to the choice between active and passive, where they weigh the relative salience of subject versus object. It would be interesting to study other types of alternatives under this perspective, for instance, the competition between DO and the serial verb construction, about which only little is known so far. Sedlak (1975) reports a few observations about West African languages. Twi, which is a dialect of Akan, a Kwa-language of Ghana, prefers the DO construction with a nominal or indefinite theme (166a,b), while a pronominal or definite theme requires a serial verb (SV) construction (167a,b). A

156 Dieter Wunderlich similar observation is made by Campbell (1996) for Kwawu, another dialect of Akan (166c, 167c). (166) Double object in Akan (Sedlak 1975; Campbell 1996: 101) a. O-fEmm me ne pO^nkó nó. 3sg-lent 1sg 3sgP horse that ‘He lent me his horse.’ b. O-maa me siká bí. 3sg-gave 1sg money some ‘He gave me some money.’ c. Me-maa Kofi nwoma (*no). 1sg-gave Kofi book (that) ‘I gave Kofi a/*the book.’ (167) Serial verbs in Akan (Sedlak 1975: 142f.; Campbell 1996: 101) a. O^-de nó fEmm me. 3sg-take that lent 1sg ‘He lent it to me.’ b. O-de siká nó maa me. 3sg-take money the gave 1sg ‘He lent me the money.’ c. Me-tOnn nwoma no maa Kofi. 1sg-sold book that gave Kofi ‘I sold the book to Kofi.’ From these observations one can conclude that the SV construction offers the ranking theme > recipient (goal) and is therefore optimal for expressing a theme with a more salient contextual value, in contrast to the DO construction with the ranking recipient > theme. This conclusion is further supported by data from Fongbe, a closely related Kwa language of Benin. The sentences in (168) with DO illustrate that a quantifier in the recipient can bind a pronoun in the theme (regardless of how the two arguments are linearly ordered) but not the other way, while (169) with SV shows the reverse, namely that only a quantifier in the theme (the shared argument) can bind a pronoun in the recipient (with only one possible ordering).

Towards a structural typology of verb classes

157

(168) DO construction in Fongbe with the ranking recipient > theme (Lefebrve & Brousseau 2001: 455) a. Ùn xlE! mE dòkpódòkpó fO` tóò tO` n. (or: …fO` tóò tO` n mE~ dòkpódòkpó) 1sg show person every picture GEN ‘I showed every person his picture.’ b. *Ùn xlE! fO` tóò dòkpódòkpó fO` tóó’tO´ tO`n 1sg show picture every picture.owner GEN *‘I showed its owner every picture.’ (169) SV construction in Fongbe with the ranking theme > recipient (=goal) (Lefebrve & Brousseau 2001: 463) a. Ùn sO´ fO` tóò dòkpódòkpó xlE fO` tóó’tO´ tO` n. 1sg take picture every show picture.owner GEN ‘I showed every picture to its owner.’ b. *Ùn xlE! fO` tóò tO` n xlE! mE~ dòkpódòkpó 1sg take picture GEN give person every * ‘I showed his picture to every person.’ It follows from these observations that xlE! ‘show’ has to be represented by BEC POSS in the DO construction, but by BEC LOC in the SV construction, similarly to English PO in (158b). (That the SV construction is similar to PO is not surprising, given the possibility that prepositions can emerge from a serial verb, see section 4.3.) However, it is fair to say that more elaborated studies are needed in order to see how the choice between DO and SV construction is determined in general. 7. Summary This survey could have given the impression that the structural profiles of languages are more different and more complicated than one ever had imagined. Despite all necessary differentiations along the way, such a view would be wrong. Languages could be much more diverse and chaotic than they actually are. I tried to show that structural generalizations determine the design of a language and hence the actual classifications of verbs one can find cross-linguistically. Structural generalizations are based on certain semantic similarities and extend them. Once these generalizations have

158 Dieter Wunderlich been made, the conditions for further generalizations, and thus for an optimal design, can differ. It is like a journey: if you have started in a certain direction, the conditions for making further choices have changed. The most obvious classification of verbs is that they can have one, two, three (or even more) arguments. If a verb has only one argument, a further classification is based on semantic factors. The common property of all linguistic types, however, is the fact that they base their design on the asymmetry of transitive verbs, which is both semantical and structural in nature. It is semantical insofar as the difference between agent and patient (or control and affected) is concerned, and it is structural in that this difference is generalized to argument hierarchy, with the agent being higher than the patient. The semantic asymmetry of transitive verbs can also be used to import a classification in the set of intransitive verbs. The concept of argument hierarchy plays the major role. Several constructional devices translate this hierarchy into linear order or react otherwise to it. Serial verb constructions and the SVO positional type translate the hierarchy, whereas noun incorporation, generalized case, as well as the choice of syntactic pivot, specifically react to the hierarchy. ‘Syntactic pivot’ is the name of an argument that determines syntactic constructions in that it, for instance, can be relativized or left out in subordinations or coordinations. Cross-linguistically, either the most salient (unmarked) argument or the highest argument are syntactic pivots, and sometimes both under different conditions. It is, moreover, an advantage if in principle every argument can be promoted to achieve this function. Another important factor for the realization of arguments are the sortal and the referential properties of the arguments themselves (person, animacy etc., as well as specificity and definiteness), whereas their information status only plays a minor role. There are various possibilities to react on these two demands, i.e., to make visible the argument hierarchy and to account for the individual properties of arguments. In the way of a taxonomy I argued that the following argument linking types have been established cross-linguistically. 1. The active type grammaticalizes the semantic factors that lead to subjectobject asymmetry. 2. The inverse type grammaticalizes person hierarchy in the context of argument hierarchy, so that in the end subject-object asymmetry gets neutralized. 3. The voice type grammaticalizes the semantic roles of arguments.

Towards a structural typology of verb classes

159

4. The generalized case type grammaticalizes argument hierarchy by a closed feature set, but makes the realization of these features sensitive to semantic factors . 5. The positional type grammaticalizes argument hierarchy by linearization, independent of semantic factors but possibly dependent on information status. Each of these types favors different subclassifications of verbs. The active type, the voice type, and the generalized case type also allow the expression of noncanonical verb classes by a certain ‘deviant’ use of their features, i.e., by means of lexical marking. When it comes to the integration of a third argument, one can clearly see the different potential of the individual types. By and large, the taxonomy gets much more differentiated, but only a few principled ways are added. Particularly interesting are serial verbs (which are on the borderline between morphology and syntax) and noun incorporation (which can even constitute an argument linking type of its own, see section 6.3 above). It is also interesting that languages often have alternative means to realize a third argument, so that the study of the choice between these alternatives (such as the English ‘dative’ alternation) becomes a worthwhile topic. Above, I listed five argument linking types. There could be more types, or different ones. Only future research will show us how many different types are possible and in what respect each type constitutes an optimal solution under certain conditions. One might believe that some types are less optimal than others, so that in the long run only a few types will survive, say the generalized case type and the positional type. I am sceptical about it. Under the view of design, the Philippine voice type is quite an elegant solution, and that is even more true for the inverse type. If a linguist without knowledge of Algonquian would have been asked to characterize which argument linking types are possible, I am quite certain that the inverse type would not have been included in his list. This shows the difficulty: even the best linguists are unable to imagine the range of constructions that can perform a comparatively simple task. But the actual problem lies even deeper because the inverse type is more than just a construction, it offers an ‘ingenious’ semantics-syntax mapping in which the syntactic subject-object asymmetry, one of the cornerstones of generative grammar, is sacrificed. In acknowledging this fact many introductory books to generative grammar would have to be rewritten. Thus, one aim of my survey is to pledge for more liberality towards typological richness, and at the same time to encourage future research.

160 Dieter Wunderlich References Aissen, Judith 1987 Tzotzil Clause Structure. Dordrecht: Reidel. 1999 Markedness and subject choice in Optimality Theory. Natural Language and Linguistic Theory 17: 673–711. Andrews, Avery 1985 The major functions of the noun phrase. In Language Typology and Syntactic Description, Vol. I: Clause Structure, T. Shopen (ed.), 62– 154. Cambridge University Press. Arka, I Wayan 2003 Balinese Morphosyntax: A Lexical-functional Approach. Canberra: Pacific Linguistics. Bailin, Alan & Ann Grafstein 1991 The assignment of thematic roles in Ojibwa. Linguistics 91: 397– 422. Baker, Mark 1991 On the relation of serialization to verb extension. In Serial Verbs: Grammatical, Comparative and Cognitive Approaches, C. Lefebvre (ed.), 79–102. Philadelphia: John Benjamins. 2003 Lexical Categories. Verbs, Nouns, and Adjectives. Cambridge University Press. Baker, Mark and O.T. Stewart 2001 A serial verb construction without constructions. Ms., Rutgers University. Barss, Andrew and Howard Lasnik 1986 A note on anaphora and double objects. Linguistic Inquiry 17: 347– 354. Bittner, Maria 1994 Case, Scope, and Binding. Dordrecht: Kluwer. Bornkessel, Ina et al. (eds.) to appear Semantic Role Universals: Perspectives from Linguistic Theory, Language Typology and Psycho-/Neurolinguistics. Berlin /New York: Mouton de Gruyter. Bresnan, Joan and Lioba Moshi 1990 Object asymmetries in comparative Bantu syntax. Linguistic Inquiry 21: 147–185. Bresnan, Joan and Tatiana Nikitina 2003 On the gradience of the dative alternation, Ms., University of Stanford. Broschart, Jürgen 1997 Why Tongan does it differently: Categorial distinctions in a language without nouns and verbs. Linguistic Typology 1: 123–165.

Towards a structural typology of verb classes

161

Burzio, Luigi 1981 Intransitive Verbs and Italian Auxiliaries. PhD dissertation MIT. 1986 Italian Syntax: A Government Binding Approach. Dordrecht: Reidel. Butt, Miriam 1995 The Structure of Complex Predicates in Urdu. Stanford: CSLI publications. Chung, Sandra 1983 An object-creating rule in Bahasia Indonesia. In Studies in Relational Grammar 1, D. M. Perlmutter (ed.), 219–271. University of Chicago Press. Collins, Peter 1995 The indirect object constructions in English: an informational approach. Linguistics 33: 35–49. Comrie, Bernard 1985 Causative verb formation and other verb-deriving morphology. In Language Typology and Syntactic Description, Vol. III: Grammatical Categories and the Lexicon, T. Shopen (ed.), 309–348. Cambridge University Press. 2006 Transitivity pairs, markedness, and diachronic stability. Linguistics 44: 303–318. den Besten, Hans 1984 The ergative hypothesis. In Studies in German Grammar, J. Toman (ed.) 23–64. Dordrecht: Foris. Dahlstrom, Amy 1988 Morphological change in Plains Cree verb inflection. Folia Linguistica Historica 9: 59–71. 1991a Plains Cree Morphosyntax. New York: Garland. 1991b Topic, focus and other word order problems in Algonquian. Winnipeg: Voices of Rupert’s Land. Davies, William D. 1986 Choctaw Verb Agreement and Universal Grammar. Dordrecht: Reidel. Dixon, Robert M. W. 1972 The Dyirbal language of North Queensland. Cambridge University Press. 1994 Ergativity. Cambridge University Press. 2000 A typology of causatives: form, syntax, and meaning. In Changing Valency. Case Studies in Transitivity, R. M. W. Dixon and A. Y. Aikhenvald (eds.), 30–83. Cambridge University Press. Dowty, David R. 1979 Word Meaning and Montague Grammar: The Semantics of Verbs and Times in Generative Semantics and in Montague’s PTQ. Dordrecht: Reidel. 1991 Thematic proto-roles and argument selection. Language 67: 547–619.

162 Dieter Wunderlich Dryer, Matthew S. 1986 Primary objects, secondary objects, and antidative. Language 62: 808–845. Evans, Nicholas and Toshiki Osada 2005a Mundari: The myth of a language without word classes. Linguistic Typology 9: 351–390. 2005b Author’s response: Mundari and argumentation in word-class analysis. Linguistic Typology 9: 442–457. Fabri, Ray 1996 The inverse morphology of Plains Cree (Algonquian). Yearbook of Morphology 1995: 17–41. Faltz, Leonard M. 1978 On indirect objects in universal syntax. CLS 14: 76–87. Foley, William A. 1991 The Yimas Language of New Guinea. Stanford University Press. Foley, William A. and Robert D. Van Valin Jr. 1984 Functional Syntax and Universal Grammar. Cambridge University Press. Gamerschlag, Thomas 2005 Komposition und Argumentstruktur komplexer Verben. Eine lexikalische Analyse von Verb-Verb-Komposita und Serialverbkonstruktionen. Berlin: Akademie Verlag. Goddard, Ives 1979 Delaware Verbal Morphology. New York: Garland. Grimshaw, Jane 1990 Argument Structure. Cambridge, MA: MIT Press. Haider, Hubert 1985 Über sein und nicht sein: Zur Grammatik des Pronomens sich. In Erklärende Syntax des Deutschen, W. Abraham (ed.), 221–252. Tübingen: Narr. Haspelmath, Martin 1993 More on the typology of inchoative/causative verb alternations. In Causatives and Transitivity, B. Comrie and M. Polinsky (eds.), 87– 120. Amsterdam: Benjamins. 2005 Ditransitive constructions: The verb ‘give’. In The World Atlas of Language Structures, M. Haspelmath et al., 426–429. Oxford University Press. Jackendoff, Ray 1990 Semantic Structures. MIT Press. Joppen, Sandra and Dieter Wunderlich 1995 Argument linking in Basque. Lingua 97: 123–169. Joppen-Hellwig, Sandra 2001 Verbklassen und Argumentlinking. Tübingen: Niemeyer.

Towards a structural typology of verb classes

163

Kaufmann, Ingrid 1995a O- and D-predicates. A semantic approach to the unaccusativeunergative distinction. Journal of Semantics 12: 377–427. 1995b What is an (im)possible verb? Restrictions on semantic form and their consequences for argument structure. Folia Linguistica 29: 67–103. 2004 Medium und Reflexiv. Eine Studie zur Verbsemantik. Tübingen: Niemeyer. Kaufmann, Ingrid and Dieter Wunderlich 1998 Cross-linguistic patterns of resultatives. Working Papers “Theory of the Lexicon”, 109. University of Düsseldorf. Keyser, Samuel Jay and Thomas Roeper 1984 On the middle and ergative constructions in English. Linguistic Inquiry 15: 381–416. Kiparsky, Paul 1992 Structural case. Ms., Institute for Advanced Study, Berlin. 1997 Remarks on denominal verbs. In Complex Predicates, A. Alsina et al. (eds.), 473–499. Stanford: CSLI Publications. Klaiman, Mimi H. 1993 The relationship of inverse voice and head-marking in Arizona Tewa and other Tanoan languages. Studies in Language 17: 343–370. Koptjevskaja-Tamm, Maria and Irina A. Muravjova. 1993 Alutor causatives, noun incorporation and the Mirror Principle. In Causatives and Transitivity, B. Comrie and M. Polinsky (eds.), 287– 313. Amsterdam: Benjamins. Kratzer, Angelika 1996 Severing the external argument from its verb. In Phrase Structure and the Lexicon, J. Rooryck and L. Zaring (eds.), 109–138. Dordrecht: Kluwer. Krifka, Manfred 2004 Lexical representations and the nature of the dative alternation. Korean Journal of English Language and Linguistics 4: 1–32. Kroskrity, Paul 1985 A holistic understanding of Arizona Tewa passives. Language 61: 306–328. Lee, Hanjung 2000 Bidirectional optimality and ambiguity in argument expression. Ms., Stanford University. Luo, Lyih-Peir 1999 Studien zu seriellen Verbkonstruktionen: ihre Syntax und Semantik im Chinesischen. Tübingen: Niemeyer. Mattissen, Johanna 2003 Dependent-head Synthesis in Nivkh and its Contribution to a Typology of polysynthesis. Amsterdam: Benjamins.

164 Dieter Wunderlich Mithun, Marianne 1991 Active/agentive case marking and its motivations. Language 67: 510–546. Mohanan, Tara 1994 Argument Structure in Hindi. Stanford: CSLI Publications. Nichols, Johanna 1986 Head-marking and dependent-marking grammar. Language 62: 56– 119. 1992 Linguistic Diversity in Space and Time. Chicago University Press. Paul, Hermann 1919 Deutsche Grammatik, Bd. III: Syntax. Perlmutter, David M. and Paul M. Postal 1984 The 1-advancement exclusiveness law. In Studies in Relational Grammar 2, D. M. Perlmutter and C. G. Rosen (eds.), 81–125. University of Chicago Press. Peyraube, Alain 1996 Recent issues in Chinese historical syntax. In New Horizons in Chinese Linguistics, C.-T. J. Huang and Y.-H. A. Li (eds.), 161–213. Dordrecht: Kluwer. Pinker, Steven 1989 Learnability and Cognition. The Acquisition of Argument Structure. Cambridge, MA: MIT Press. Pustejovsky, James 1991b The syntax of event structure. Cognition 41: 47–81. Ross, Malcolm 2002 The history and transitivity of western Austronesian voice and voicemarking. In The History and Typology of Western Austronesian Voice Systems, F. Wouk and M. Ross (eds.), 17–62. Canberra: Pacific Linguistics. Sasse, Hans-Jürgen 1993a Das Nomen – eine universale Kategorie? Sprachtypologie und Universalienforschung 46: 187–221. 1993b Syntactic categories and subcategories. In Syntax. Ein interdisziplinäres Handbuch zeitgenössischer Forschung, J. Jacobs et al. (eds.), 646–686. Berlin /New York: Mouton de Gruyter. Schwartz, Linda J. 1979 The order of Dakota person affixes: an evaluation of alternatives. Glossa 13: 3–12. Schulze, Wolfgang 2001 The Udi Language: A Grammatical Description. Online, Universität München. [http://www.lrz-muenchen.de/~wschulze/] Sedlak, Philip A. S. 1975 Direct/indirect object word order: a cross-linguistic analysis. Working Papers on Language Universals 18: 117–164. Stanford University.

Towards a structural typology of verb classes

165

Stewart, Osamuyimen T. 2001 The Serial Verb Construction Parameter. New York: Garland. Stiebels, Barbara 1996 Lexikalische Argumente und Adjunkte: Zum semantischen Beitrag von verbalen Präfixen und Partikeln. Berlin: Akademie Verlag. 1998 Complex denominal verbs in German and the morphology-semantics interface. Yearbook of Morphology 1997: 265–302. 2000 Linker inventories, linking splits and lexical economy. In Lexicon in Focus, B. Stiebels and D. Wunderlich (eds.), 211–245. Berlin: Akademie Verlag. 2002 Typologie des Argumentlinkings: Ökonomie und Expressivität. Berlin: Akademie Verlag. this vol. From rags to riches. Nominal linking in contrast to verbal linking. Urbas, Martina 1990 Denominale Verben: Semantische und konzeptuelle Aspekte. Master’s Thesis. University of Düsseldorf. Van Valin, Robert D. 1977 Aspects of Lakhota Syntax. PhD dissertation, University of California, Berkeley. 2002 The role and reference grammar analysis of three-place predicates. Ms. Watkins, Laurel 1984 A Grammar of Kiowa. Lincoln: Univ. of Nebraska Press. Williamson, Kay 1965 A Grammar of the Kolokuma Dialect of Ijo. Cambridge University Press. Wolfart, H. Christoph 1973 Plains Cree: A Grammatical Study. Transactions of the American Philosophical Society, New Series, Vol. 63, Part 5. Wolfart, H. Christoph and Janet F. Carroll 1981 Meet Cree. A Guide to the Language. Lincoln: University of Nebraska Press. Wunderlich, Dieter 1996a Lexical categories. Theoretical Linguistics 22: 1–48. 1996b A minimalist model of inflectional morphology. In The Role of Economy Principles in Linguistic Theory, C. Wilder, M. Bierwisch and H.-M. Gärtner (eds.), 267–298. Berlin: Akademie Verlag. 1997a Cause and the structure of verbs. Linguistic Inquiry 28: 27–68. 1997b Argument extension by lexical adjunction. Journal of Semantics 14: 95–142. 2000a Predicate composition and argument extension as general options. In Lexicon in Focus, B. and D. Wunderlich (eds.), 247–270. Berlin: Akademie Verlag. 2000b Optimal case in Hindi. Ms., University of Düsseldorf.

166 Dieter Wunderlich 2001a

How gaps and substitutions can become optimal: the pronominal affix paradigms of Yimas. Transactions of the Philological Society 99: 315–366. 2001b Prelexical syntax and the voice hypothesis. In Audiatur Vox Sapientiae, C. Féry and W. Sternefeld (eds.), 487–513. Berlin: Akademie. 2002a On the nature of dative in Hungarian. In Papers from the Budapest Conference (= Approaches to Hungarian, Vol. 8), I. Kenesei and P. Siptár (eds.) 161–184. Budapest: Akadémiai Kiadó. 2002b Argument linking types – approached from the perspective of LDG. In Report of the Special Research-project for the Typological Investigation of Languages and Cultures of the East and West 2001, Part II, H. Suzuki (ed.), 777–799. University of Tsukuba, Japan.

002c Argumentlinking als strukturelle Generalisierung (Inkorporationstyp Nivkh). In Sprache als Form. Festschrift für Utz Maas zum 60. Geburtstag, M. Bommes, C. Noack and D. Tophinke (eds.), 11–21. Wiesbaden: Westdeutscher Verlag 2002. 2003 Optimal case patterns: German and Icelandic compared. In New Perspectives on Case Theory, E. Brandner and H. Zinsmeister (eds.) 331–367. CSLI publications. 2004 Why assume UG? In What Counts as Evidence in Linguistics? – The Case of Innateness, M. Penke and A. Rosenbach (eds.), special issue of Studies in Language 28: 615–647. 2005 The challenge by inverse morphology. Lingue e Linguaggio 4: 195– 214.

o appear Argument hierarchy and other factors of argument realization. In Bornkessel et al. (eds.) Valentine, J. Randolph 2001 Nishnaabemwin Reference Grammar. University of Toronto Press. Zhang, Ning 1998 The interactions between construction meaning and lexical meaning. Linguistics 36: 957–980. Zúñiga, Fernando 2002 Inverse system in indigenous languages of the Americas. PhD diss., University of Zurich.

From rags to riches. Nominal linking in contrast to verbal linking Barbara Stiebels

1. Introduction The comparison of nouns and verbs usually focuses on some important differences between these two major lexical categories: In ontological terms, nouns and verbs differ in their typical referential potential; verbs refer to eventualities, nouns to concrete objects. The differences in the referential potential also have consequences for the set of functional categories that is associated with each category (see Wunderlich 1993b, 1996). Verbs may bear specifications and markings for aspect, tense, mood, polarity etc., whereas nouns may bear specifications regarding definiteness/specificity, gender etc. In Chomsky’s (1970) feature system, nouns ([+N,–V]) and verbs ([–N,+V]) do not show any categorial overlap, they are maximally distinct. This classification raises doubts as to whether nouns and verbs could have anything in common at all. According to Wunderlich’s (1996) classification, nouns and verbs share the feature specification [–dep], i.e. they are both referentially independent because they provide a referential argument. The two categories differ with respect to the feature [art] ‘articulated argument structure’: verbs have an articulated argument structure ([+art]), nouns do not. Under this perspective, the major difference between nouns and verbs is one in terms of argument structure, disregarding the differences in the nature of their referential argument for the moment. Therefore, argument linking of nouns and verbs is likely to be affected by the differences in argument structure. That nouns are usually attributed a non-articulated or ‘poor’ argument structure has to do with the fact that most underived nouns are non-relational and denote simple sortal predicates (house, dog). The class of relational nouns (kinship terms or body part expressions) is generally far smaller. I would like to make use of the metaphor of ‘being in rags’ for the impoverished argument structure of nouns. However, there are contexts in which nouns exhibit a more complex argument structure: Event nominalizations

168 Barbara Stiebels based on polyadic verbs usually pose a challenge for the nominal linking resources since they inherit the arguments of the underlying verb. This facet of nominal argument structures is one of ‘riches’. It is the aim of this paper to show which properties of nominal linking fall under the aspects of ‘rags’ and which under the aspects of ‘riches’. The discussion then touches on the question of which aspects of nominal argument linking are really specific for nouns and nominals (= the products of nominalizations) and which linking aspects are rather parallel to those of verbs. As has been discussed thoroughly in the literature, the noun-verb distinction is not as clear-cut as one may expect (there are nouns such as Hochzeit ‘wedding’, which refer to eventualities, and verbs such as sich erstrecken ‘extend (itr.)’, which refer to permanent states); some languages show a rather weak morphosyntactic differentiation of nouns and verbs (see the discussion of Iroquoian, Salishan or Oceanic languages: Kinkade 1983; van Eijk & Hess 1986; Sasse 1993; Jelinek & Demers 1994; Broschart 1997; Mithun 1999a/b). One might speculate that those languages that exhibit a weak noun-verb distinction show the strongest similarities of argument linking in nouns and verbs. This, however, is not necessarily true. The Salishan languages, for instance, use a separate linker (possessor agreement) for arguments of noun-like elements. Likewise, the Oceanic language Tongan has a noun-specific linker (the alienable genitive). In contrast to these languages, quite a number of languages with a stronger noun-verb distinction on the morphosyntactic level display more categorial linker overlaps, i.e. linkers that are used both with nouns and verbs (e.g. the identity of accusative and genitive in Yaqui and of possessor and object agreement in Kilivila; see Stiebels 2002). In Stiebels (1999) I have shown that Classical Nahuatl, though displaying a clear noun-verb distinction and distinct linker inventories, exhibits an unexpected symmetry of argument linking in nouns and verbs: Nominal arguments underlie the same restrictions of argument saturation as verbal arguments and they are not optional to the extent postulated in the literature (e.g. Bierwisch 1989; Grimshaw 1990). Given that underived nouns and verbs differ in the potential complexity of their argument structures, the question arises which consequences this has for the potential linking strategies. The evaluation and comparison of nominal and verbal linking should not only include the linker inventories and their use, but also consider the problem of verbal vs. nominal voice (is there a category of nominal voice?) and the study of the potential range of nominal linking strategies. The richness of verbal linking strategies has been exhaustively documented in the literature (see Comrie 1989; Wunderlich this

From rags to riches. Nominal linking in contrast to verbal linking

169

volume for an overview). This leads to the question: To which extent do verbal linking types also occur in nouns and nominals? As I will show, the parallelism of nouns/nominals and verbs is quite remarkable. In the remainder of this paper I will compare the argument structure of nouns and verbs (section 2), the voices that may be associated with each category (section 3), and their linking patterns and types (section 4). I will show that nouns/nominals, although being semantically/structurally poorer than verbs, still show unexpected similarities with verbs in their linking strategies. The linking spectrum found in verbs is almost completely mirrored in nouns/nominals. In section 5 I will discuss the strategies languages use in dealing with the complex argument structure of event nominals; since the nominal linking resources are based on underived relational nouns, the linking of event nominals derived from polyadic verbs poses a serious challenge and requires an extension of the nominal linking mechanisms in one way or another. The existence of these compensatory strategies and the differences in information structure (section 6) are the major differences between verbal and nominal argument linking. 2. In ‘rags’: How limited is the argument structure of nouns? As already indicated in the introduction, verbs and nouns differ in the potential complexity of their argument structure. Verbs can be structurally complex, allowing up to three structural arguments besides the referential argument, i.e. the situational variable, in their base entries, as the following simplified representations of impersonal, intransitive, transitive and ditransitive verbs in (1) show. Following the convention of Lexical Decomposition Grammar (LDG; Joppen & Wunderlich 1995; Wunderlich 1997a, this volume; Stiebels 1999), I will represent nouns and verbs as predicate-argument structures (= Semantic Form SF). The argument structure (i.e. the grid) is derived by -abstraction of the variables according to their depth of embedding, starting with the most embedded variable, which is assumed to be the lowest-ranked argument and written as left-most -abstractor on the -grid. ACT(x) in (1d) denotes that there is an activity of x; the causal relation between the activity of x and the change of possession is inferred (see Kaufmann 1995).

170 Barbara Stiebels (1)

Argument structure of verbs a. rain s RAIN(s) b. cry x s CRY(x)(s) c. read y x s READ(x,y)(s) d. give z y x s [ACT(x) & BECOME(POSS(y,z))](s)

es,t e,es,t e,e, es,t e,e,e,es,t

In (1) I have also indicated the semantic types of the verbs, which are based on the simple Montagovian types e for entities and t for truth values; the situational variable s is also of type e, but highlighted as eS. e,t is a complex type that takes an entity expression and yields a truth value. The referential argument of verbs, i.e. the event/situational variable s, may be further split up into a world and a time parameter. However, I will stick to the simple notation s. This event variable does not play a role in verbal argument linking; it is targeted by tense, mood, and aspect. Depending on the availability and admissibility of argument-extending operations (see section 3), some languages may allow verbs with up to five or even more (structural) arguments. (2) shows an example of a 5-place verb in the Muskogean language Creek: the transitive verb ‘cut’ is turned into a 5-place verb by means of the causative (CAUS) and two applicatives (APPL):1 (2)

5-place verb in Creek (Martin 2000: 400) (i)s-in-tac-ipéyhc-ís APPL.INST-APPL.BEN-cut-CAUS-IND ‘they/people [1] are making him/her [2] cut it [3] for him/her [4] with it [5]’

In contrast, most nouns display a very simple argument structure. Non-relational (i.e. sortal) nouns, which represent the majority of nouns, denote 1place predicates, shown in (3a); inherently relational nouns, which encom———–———————————————————————————— 1

The list of abbreviations is given in the appendix. Note that I gloss the actual morphological markers, no abstract syntactic case or agreement. In addition, I treat the agreement markers as mirror image of case. Thus, N designates the ‘nominative-like’ default marker, A the ‘accusative-like’, D the ‘dative-like’ and E the ‘ergative-like’ marker, which means, for instance, that I call an agreement marker an E-linker if it has the same distribution as ergative case (see also section 4.1, especially (36)).

From rags to riches. Nominal linking in contrast to verbal linking

171

pass kinship terms and body part expressions, denote 2-place predicates and are represented in (3b/c). In general, most sortal nouns may be turned into a relational noun, i.e. undergo possessor extension (see section 3.3). (3)

Argument structure of nouns a. cat u CAT(u) b. mother v u MOTHER(u,v) c. nose v u [NOSE(u) & PART_OF(u,v)]

e,t e,e,t e,e,t

Nouns (or the noun phrases they project) can be used either referentially (the doctor) or predicatively (he is a doctor). In the noun’s referential use its referential argument is bound (e.g. by a determiner that turns the nominal predicate into an entity expression) – consequently the NP can saturate an argument role of another predicate. This argument-saturating function crucially differs from those available for verbs. However, if nouns are used predicatively, they may structurally resemble intransitive or transitive verbs. The following example from Classical Nahuatl illustrates this resemblance: nouns and verbs take subject agreement prefixes (glossed as N-linkers). The internal argument of a transitive verb is indexed by object agreement (glossed as A-linker), the internal argument of a noun by possessor agreement (P-linker). (4)

Predication in Classical Nahuatl a. ni-koi 1SG.N-sleep ‘I sleep’ b. ni-tek i 1SG.N-lord ‘I am a lord’ c. ni-mits-itta 1SG.N-2SG.A-see ‘I see you’ d. ni-mo-siww 1SG.N-2SG.P-woman ‘I am your wife’

Note that the linker of the possessor argument does not change if the noun is not used predicatively: The possessor agreement linker (P) in (5a) cannot be substituted by the default agreement linker (N) as shown in (5b). (5)

a. mo-siww 2SG.P-wife ‘your wife’

b. *ti-siww 2SG.N-wife intended: ‘your wife’

172 Barbara Stiebels There are no (underived) nouns that exhibit more than two structural arguments. Since the referential argument does not play a role for argument linking in non-predicative uses of nouns, structural linking is generally rather limited, often restricted to one case/agreement linker, which, in addition, is barred from being doubled in many languages. These restrictions on structural linking also have an impact on the realization of inherited arguments in nominalizations. Often, only one argument may receive a structural nominal linker (see section 5). The following example from the Austronesian language Tukang Besi illustrates the challenge to argument linking posed by nominalization: Here, a ditransitive verb is nominalized. Since the situational variable is now an argument that could be projected syntactically (in the predicative use of the nominal), argument linking has in principle to deal with a 4-place predicate. Tukang Besi is special in that it allows the doubling of the genitive: (6)

Tukang Besi: nominalization of ditransitive verb (Donohue 1999: 319) [te pa-manga-’a-u u Aswi nu ika atu], no-marasai CORE CAUS-eat-NOML-2SG.P GEN Aswin GEN fish that 3.N-difficult na ’ita-ra-no NOM see-NOML-3.P ‘the way you fed the fish to Aswin, it didn’t look easy’

Here, the agent argument is indexed by possessor agreement and the two lower arguments are realized by genitive. The differences between nouns and verbs in their argument structures also play a role in category-shifting derivations: nominalizations often only involve a category shift without any semantic enrichment (see Bierwisch 1989; Stiebels 1999), as shown for the event nominalization (6) in (7a/b), whereas the derivation of denominal verbs (e.g. to salt (the soup), to butter (the bread), to saddle (the horse)) usually requires semantic enrichment, namely the addition of the situational argument and often the extension by verbal predicates such as ACT or BECOME (see Wunderlich 1987; Kiparsky 1997; Stiebels 1998), which is indicated in (7c/d).

From rags to riches. Nominal linking in contrast to verbal linking

(7)

a. b. c. d.

pa-mangaV pa-manga-aN saddleN saddleV

173

y x u s
s’ [ACT(u) & FEED(x,y)(s’)](s) y x u s
s’ [ACT(u) & FEED(x,y)(s’)](s) u SADDLE(u) y x s [ACT(x) & BECOME(POSS(y, z 2 SADDLE(z)))](s)

3. Do nouns have voices like verbs? The phenomenon of ‘voice’ or ‘diathesis’ is well-known for verbs (e.g. Comrie 1985; Baker 1988; Wunderlich 1993a, 1997a; Dixon 2000). In many languages verbs display several voices/diathesis operations, which differ regarding the resulting number of arguments or the realization of the respective arguments. Given the impoverished argument structure of nouns, one may ask whether the notion of voice is extendable to some phenomena in nouns/nominals. As I want to show in the following, there are operations in nouns/nominals that may be regarded equivalents of verbal voice in terms of argument extension/reduction and the selection of the argument affected by the operation. Nevertheless, nouns and verbs differ in their distributional patterns of admissible voice operations. (8) provides an overview of the possible operations, which I will discuss more thoroughly in the following subsections: (8)

Voice/diathesis operations in verbs and nouns

Arg. extension Arg. reduction

Affected arg.

Verb

Noun

highest

causative

agent nouns

lowest

applicative

possessor extension

highest

passive

not productively available

lowest

antipassive

‘antipossessive’

Voice operations in nouns fall into two classes depending on the category of the base: N-N derivations and V-N derivations (i.e. nominalizations) differ clearly in their preferential and admissible patterns, as I will show below. Note that only N-N derivations could be regarded as nominal voice proper because only they are category-preserving. ———–———————————————————————————— 2

For the sake of simplification I assume that the referential argument of the base noun is bound by the -operator, a type-shifting operator (e,t,e) that selects an entity from a set without specifying a concrete one.

174 Barbara Stiebels 3.1. Verbal voice: causative and applicative As mentioned above, verbal voice/diathesis operations are well-studied. I will briefly recapitulate their properties in terms of argument structure. In verbs the most typical instance of argument extension with respect to the highest argument is the causative, which introduces a causer argument, thus demoting the former highest argument to an internal argument role (causee).3 (9) shows the causativization of an intransitive verb in K’iche’; the ensuing transitivization licenses the use of the ergative agreement marker in (9b). (9)

K’iche’: causative (Campbell 2000: 277) a. -atin-ik ASP-bathe-ITR ‘he bathed’ b. -r-atin-is-x ASP-3SG.E-bathe-CAUS-TR ‘she bathed him’ c. x s BATHE(x)(s)  x u s’
s [ACT(u) & BATHE(x)(s)](s’)

(9c) represents the operation of causativization: The causer argument is introduced by the predicate ACT(u); the situational variable of the base verb is existentially bound. Apart from the integration of a highest argument, many languages also allow the integration of a new lowest or second-to-lowest argument, which is realized as direct object (see also Wunderlich 2000). Whereas the integration of a lowest argument in (strong) resultatives (e.g. run the lawn flat)4 is generally not marked in the verb, the integration of an instrument, beneficiary, goal, locative, manner or reason argument is generally marked. Following the tradition in Bantu linguistics I will call this type of diathesis operation applicative. ———–———————————————————————————— 3

4

Operations that likewise introduce a highest argument include the assistive in Quechua (van de Kerke 1996; Stiebels 2003) or the affective in Basque (Joppen & Wunderlich 1995). Kaufmann & Wunderlich (1998) distinguish between strong and weak resultatives: In strong resultatives, the result predicate predicates of a newly introduced argument (*run the lawn vs. run the lawn flat), whereas in weak resultatives the result predicate predicates of an argument of the base verb (cut the bread vs. cut the bread in pieces).

From rags to riches. Nominal linking in contrast to verbal linking

175

The following example from K’iche’ illustrates the instrumental applicative. In the base verb, an instrumental DP may be realized obliquely as in (10a); in (10b) the instrument argument has been turned into an argument of the verb, which is realized as direct object; the instrumental applicative is used in K’iche’ if the instrument is focused; therefore, the instrumental DP appears in the preverbal focus position. (10c) shows the simplified derivation of the applicative verb. (10) K’iche’: applicative (Campbell 2000: 278) a. -at-in-’ay i  ASP-2SG.N-1SG.E-hit with wood ‘I hit you with a stick’ b.  -in-’aya-b’e-x w-h wood ASP-1SG.E-hit-APPL-TR 2SG.E-GEN ‘I used a stick to hit you’ c. y x s HIT(x,y)(s)  z y x s [HIT(x,y)(s) & INSTR(z)(s)] (10b) shows that the former object loses its structural character; it can only be realized obliquely, i.e. by means of a relational noun. In general, argument extensions that apply on polyadic verbs pose a challenge for structural linking. 3.2. Nominal voice: agent nouns Do nouns have similar options to obtain an additional argument? There are N-N derivations that correspond to the causative, i.e. they introduce a new highest argument, which can be superficially analyzed as some kind of ‘agent’ argument. These ‘agent nouns’ may encompass various readings, often encoded by the same morphological marker: the derived noun denotes in a more general reading a person who is associated with the referent of the base noun or refers in a more specific reading to the inhabitants of a place derived from toponyms. (11) shows this pattern in German (derived by the suffix -er) and (12) in the Mayan language Mam (derived by the prefix aj-). (11) N-N derivations in German a. Garten  Gärtn-er b. Berlin  Berlin-er

‘gardener’ ‘person from Berlin’

176 Barbara Stiebels (12) N-N derivations in Mam (England 1983: 117–18) a. cheej ‘mule’  aj-cheej ‘mule driver’ miis ‘mass’  aj-miis ‘Catholic’ b’iitz ‘song’  aj-b’iitz ‘singer’ b. Xhniil ‘Colotenango’  aj-xhniil ‘person from C.’ Itzal ‘Ixtahuacán’  aj-itzal ‘person from I.’ One can assume that the respective derivational morpheme integrates an abstract relation REL whose specific interpretation depends on the possible prototypical relations between the higher argument, which denotes a person, and the lower argument, the referent of the base noun. The specific instantiation of REL is determined on Conceptual Structure (CS).5 (13) Representation of -er/ajN u
v [REL(u,v) & N(v)]

e,t,e,t

A more specific instantiation of REL can be found in Classical Nahuatl: The suffixes -e and -wa introduce a new referential argument that is understood as the owner of the base noun referent. (14) Nahuatl: new ‘owner‘ argument (Andrews 2003: 330–31) a. n-e ‘one who has a home’ home-NOML b. -wa water-NOML ‘one who has water’ c. ts-wa cave-NOML ‘one who has a cave’ = ‘fox’ The derivation of (14a) is shown in (15): the suffix takes a nominal predicate and yields a semantically enriched predicate. (15) Derivation of (14a) N u
v [POSS(u,v) & N(v)] e,t,e,t a. -e e,t b. n-(i) x HOME(x) u
v [POSS(u,v) & HOME(v)] e,t c. n-e ———–———————————————————————————— 5

(13) is a simplification. It assumes that the base noun regularly denotes a predicate. However, with toponyms, an entity expression of type e is integrated. Therefore, one may assume the additional entry: v u REL(u,v); e,e,t.

From rags to riches. Nominal linking in contrast to verbal linking

177

This pattern of ‘owner nouns’ differs from the cross-linguistically well-established pattern of possessor extension, in which the possessor argument is not the referential argument; I will discuss this pattern in the next section. Far more patterns of argument extension regarding the highest argument can be found with deverbal nouns. Besides the nominalization patterns in which an argument of the underlying verb becomes the referential argument of the nominal (in event, agent or patient nominalizations), there are nominalizations in which the new referential argument is not an argument of the underlying verb (instrument, locative, result or manner nominalizations); here, the referential argument must have been introduced in the course of derivation. Whereas German and English confound agent and instrument nominalizations (in terms of the derivational suffix -er: read-er vs. dry-er), Mam uses different markers for the derivation of agent, instrument and locative nominalizations, thus distinguishing agent and instrument nominalization: (16) V-N derivations in Mam (England 1983: 119–20) a. agent nominalization yoola ‘speak’  yoola-l ‘speaker’  waa-l ‘eater (of tortillas)’ waa ‘eat’ b. instrument nominalization luk ‘pull up’  luk-b’il ‘instr. for pulling up’ tzutz' ‘sit’  tzut’-b’il ‘bench, chair’ c. locative nominalization juus ‘burn’  jus-b’een ‘burned place’ aq'n ‘work’  aq’an-b’een ‘worked place’ In Stiebels (1999) I have shown for Classical Nahuatl that a language may distinguish agent and instrument nominalizations structurally even though they are not distinguished in terms of verbal marking: Agent nominalizations in Nahuatl are derived from active verbs (by conversion of the habitual form as in (17a)), thus rendering the agent the referential argument of the derived nominal, whereas instrument nominalizations are derived from passive verbs (again by conversion of the habitual form as in (17b)); this has the consequence that the agent is no longer available because it is existentially bound by the passive morpheme. Therefore, the instrumental argument must have been introduced by the derivation and cannot be conceived of as conceptually shifted ‘agent’. In both cases in (17), the lowest-ranked argu-

178 Barbara Stiebels ment is saturated by an ‘unspecific object’ prefix; these arguments are also not available as referential arguments. (17) Classical Nahuatl: agent vs. instrument nominalization (Andrews 2003: 340, 346) a. t-y-wa-ni USP.H-war-make-HAB ‘person who makes war against someone, warrior’ b. y x
s MAKE_WAR_AGAINST(x,y)(s) c. a-tek--ni USP.NH-cut-PASS-HAB ‘thing with which one cuts s.th., knife’ d. y u
x
s [INSTR(u)(s) & CUT(x,y)(s)] Locative nominalizations in Nahuatl are likewise based on passive verbs, also yielding evidence for the introduction of a new referential argument. Regarding instrumental, locative or manner arguments one can observe a clear difference between nouns and verbs: Whereas these arguments are introduced as ‘applied’, i.e. internal arguments with verbs, they are introduced as the highest, i.e. referential arguments with nouns.6 3.3. Nominal voice: possessor extension Whereas the applicative is often restricted to certain semantic classes of verbs (mainly activity or accomplishment verbs) and does not show the same range of distribution in all languages, the corresponding argument extension in nouns, the possessor extension, seems to be universal. Hereby, a sortal noun is extended by a possessive relation, which licenses the possessor argument as lower-ranked argument (see also Ortmann 1994; Barker 1995; Partee & Borschev 2003): (18) Possessor extension (POREX) u N(u); e,t  v u [N(u) & POSS(v,u)]; e,e,t ———–———————————————————————————— 6

Only rarely is an instrumental or similar argument introduced as the highest argument of a verb (by the circumstantial voice in Malagasy, see Keenan & Polinsky 1998, or the instrumental voice in Jacaltec, see Craig 1978). Here, the highest (agent) argument is blocked from being realized structurally, which indicates that it is ousted by the new argument.

From rags to riches. Nominal linking in contrast to verbal linking

179

Generally, there are only a few nouns that cannot undergo possessor extension. Mam, for instance, has a class of ‘impossessible’ nouns (following Lehmann’s 1998 terminology), i.e. nouns that cannot be used with a possessor; this class only includes nouns that denote natural phenomena (e.g. kya j ‘sky’ and che w ‘star’). Many languages do not mark the operation of possessor extension, but simply add a possessor argument. This is true for German, English and many other languages. Some languages must first apply a morphological operation of argument/possessor extension; then, the respective noun can take a possessor argument. Mayan languages exhibit this pattern; they make use of certain nominal markers, as shown in (19) for Mam and Huastec; the relevant affixes are glossed as POREX. Note that the possessor itself is realized by an ergative agreement marker. (19) Possessor extension in some Mayan languages a. Mam (England 1983: 68) xjaal  n-xjaal-al-a person 1SG.E-person-POREX-12 ‘my people’ xiinaq  n-xinaq-iil-a ‘man’ 1SG.E-man-POREX-12 ‘my man’ b. Huastec (Edmonson 1988: 350) u=bakn-il bakan  tortilla 1SG.E-tortilla-POREX  in=ha-ll ha water 3SG.E=water-POREX i  in=i-l chile 3SG.E=chile-POREX

‘my tortilla’ ‘its water’ ‘its chile’

3.4. Verbal voice: passive and antipassive Besides argument extension languages make ample use of argument reductions such as passive or antipassive; argument reduction may also be triggered by noun incorporation, reflexivization, reciprocalization or mediopassive. (Agentless) passive involves argument reduction regarding the highest argument (the external argument is existentially bound); argument reduction regarding the lowest (or some internal argument) is called antipassive (in ergative languages) or unspecified object deletion, the latter remaining

180 Barbara Stiebels unmarked on the verb in most languages. The following example in (20b) shows the antipassive pattern (ANTIPASS) in K’iche’: the verb is marked by the suffix -on and the internal argument can no longer be realized structurally; it may be optionally realized by means of an oblique marker (h). The resulting agreement pattern is that of an intransitive verb; only the remaining argument is indexed by a nominative/absolutive agreement marker. (20) K’iche’: antipassive (Campbell 2000: 266–67) a. k-a-yoq’ ri a-nn ASP-2SG.E-mock the 2SG.E-mother ‘you mock your mother’ b. k-at-yoq’-on (-h ri a-nn) ASP-2SG.N-mock-ANTIPASS to.3SG.E-GEN the 2SG.E-mother ‘you mock (your mother)’ c. y x s MOCK(x,y)(s)  x s
y MOCK(x,y)(s) Besides ergative languages, there are also accusative languages that do not allow unspecified object deletion; in Classical Nahuatl, for instance, the respective argument must be bound by one of the two ‘unspecific’ object prefixes t- (human object) or a- (non-human): The internal argument is either specified as in (21a), triggering object agreement, or unspecified; in the latter case, the ‘unspecific’ object prefix is obligatory (compare (21b/c)). (21) Nahuatl: unspecified object saturation a. ni-k-tmoa (i-) 1SG.N-3SG.A-seek flower-ABS ‘I seek a/the flower’/‘I seek him/her/it’ b. ni-a-tmoa 1SG.N-USP.NH-seek ‘I seek something’ c. *ni-tmoa 1SG.N-seek ‘I’m seeking’ 3.5. Nominal voice: possessor reduction (‘antipossessive’) Does one find argument reduction in nouns as well? Can a relational noun be turned into a non-relational noun that no longer takes a possessor argument?

From rags to riches. Nominal linking in contrast to verbal linking

181

It seems that the number of languages that exhibit an overt exponence for such an operation of argument reduction is quite small. The Mayan languages belong to those few languages that mark possessor reduction. (22a) is an example from Mam: the suffix -b’aj triggers argument reduction. I will call this operation ‘antipossessive’ (ANTIPOSS) in the following. Huastec even distinguishes between two types of derived non-relational usages: the general antipossessive form (-lek) and a form used in generic readings (-lab), see (22b). With some nouns, the two markers yield unpredictable meaning differences (see (22c/d)).7 (22) Possessor reduction in Mam and Huastec a. Mam (England 1983: 69) n-qan-a  qam-b’aj 1SG.E-foot-12 foot-ANTIPOSS b. Huastec (Edmonson 1988: 354) u=ukul  uk-lek stomach-ANTIPOSS 1SG.E-stomach  uk-lab stomach-ANTIPOSS c. Huastec (Edmonson 1988: 355) u= akan  akan-lek 1SG.E=foot foot-ANTIPOSS d.  akan-lb foot-ANTIPOSS

‘stomach’

(absolutive)

‘stomachs’

(generic)

‘a plant’ ‘a forked pole’

The antipossessive in (22a) may be represented as follows; the possessor argument is existentially bound: (23) Representation of qam-b’aj u
v [FOOT(u) & PART_OF(u,v)] ———–————————————————————————————

7

In Nahuatl, for many nouns both the non-possessed and the possessed form are morphologically marked. The picture is not as neat as in Mayan languages because the morphological marking does not correlate with an inherent or established relational character of the possessed noun. Non-possessed forms are marked by the absolutus marker -(i), possessed forms by the suffix -w(i) (with possessum singular) or -wn (possessum plural). However, even non-relational nouns may exhibit an absolutus (kal-li house-ABS, - water-ABS), whereas the relational variant of a sortal noun may lack the marker for possessor extension (no-kal 1SG.P-house vs. n--w 1SG.P-water-POSS). This indicates that additional factors play a role.

182 Barbara Stiebels 3.6. Nominal voice: ‘nominal passive’ There is no true nominal equivalent to passive, which is not very surprising: The single argument of sortal nouns cannot be existentially bound because this (referential) argument must be available for referential binding by a determiner or other operations that are necessary for the semantic integration of the noun; likewise, impersonal verbs such as rain cannot be passivized in any language because existential binding of the situational variable would yield a non-predicate expression. There is one type of conceptual shift (mainly in body part expressions) that could be analyzed as some very abstract kind of ‘nominal passive’: The noun changes its reference from a body-part expression to the ‘owner’ of the body part, which could be accounted for by existential binding of the highest argument; these nouns are often analyzed as instances of exocentric structures: (24) German: conceptual shift (‘nominal passive’) a. Der Schmoll-mund hat sich mal wieder rassistisch AUX.3SG REFL once again racially the sulk-mouth geäußert. uttered ‘the pout once again has uttered racial remarks’ b. v
u [MOUTH(u) & PART_OF(u,v)] The conceptual shift is represented in (24b) for the noun Mund ‘mouth’; due to existential binding of the highest argument, the lower argument becomes the referential argument.8 This option does not seem to be available productively; kinship terms cannot be manipulated in this way. In addition, there is no language attested yet in which this option is productive and morphologically marked.9 ———–———————————————————————————— 8

9

Many languages allow the formation of ‘patient nouns’, i.e. nominalizations that refer to the internal argument of the base verb (e.g. English -ee: employee). The higher arguments of the base verb may be existentially bound (especially the situational argument), but it is also possible that a possessor argument in these nominals may be interpreted as underlying agent argument (my employee: ‘the person I employ’). In German there are some derived nouns based on body part expressions that resemble the ‘nominal passive’ pattern, e.g. Tausend-füß-ler thousand-foot-NOML ‘millipede’ or Links-händ-er ‘left-hand-er’. One can either invoke the nominal passive analysis sketched above or the agent nominal analysis; in the latter case, the new argument would be identified with the possessor argument of the base noun.

From rags to riches. Nominal linking in contrast to verbal linking

183

3.7. Voice interaction Verbal voices show a complex interaction, in which, for instance, argument reduction may precede or follow argument extension (see Krämer & Wunderlich 1999) or in which various operations of argument extension may be combined (see e.g. Baker 1988; Stiebels 2003). (25a) illustrates the antipassivization of a causativized verb in Huastec, which turns the causee into an oblique argument; (25b) shows the combination of causative and benefactive applicative, yielding a ditransitive verb. (25) Huastec: voice interaction (Edmonson 1988: 189–90)  -om-ts k’al an uk’el-om a. hah kal- he go.out-CAUS-ANTIPASS-COMPL with the drown-NOML ‘he was the hauler out of drowning persons (a lifeguard)’ b. tikin pa -b-an--i tsakam 2/1 descend-CAUS-EP-APPL-EP ‘get them down for me, child’

child

In contrast to the verbal pattern, nominal voice interaction seems rather marginal. One apparent case is the combination of possessor extension and possessor reduction in Huastec (26). (26) Huastec (Edmonson 1988: 355, 374) a. u= akan 1SG.E-foot

b. u=t’u l 1SG.E=flesh ‘my (own) flesh’   akan-lb t’u-lek flesh-ANTIPOSS foot-ANTIPOSS ‘a forked pole’ ‘flesh’   u=akan-lb-il u=t’u-lk-il 1SG.E=flesh-ANTIPOSS-POREX 1SG.E=foot-ANTIPOSS-POREX ‘my forked pole’ ‘my meat (that I sell)’

In (26a) a lexicalized antipossessive form is extended, in (26b) the antipossessive binds the inherent possessor argument of the body-part expression, thus allowing the addition of a possessor argument that shows no inherent relation with the body part. This pattern of combined argument reduction

184 Barbara Stiebels and argument extension mirrors the pattern found in the combination of antipassive or noun incorporation and applicative, which is attested for the Mayan language Yucatec. (27) Yucatec: voice interaction (Bricker 1978: 16) k=in ’ak-e -t-ik in kòol PRES=1 chop-wood-APPL-IMPF 1 cornfield ‘I wood-chop my cornfield’ First, the verb is intransivitized by incorporation of the noun (’ak), then the locative (kòol) is integrated as direct object by means of the ‘applicative’ morpheme –t. Another case of nominal voice interaction is the recursive application of the ‘nominal causative’, i.e. the integration of an owner argument in Classical Nahuatl (see section 3.2, (14)): (28) Classical Nahuatl: recursion of ‘owner noun’ formation (Andrews 2003: 331) a. [mi-wa ]-k-wa fish-NOML-CONN-NOML ‘one who has owned one who has owned fish’ = ‘master of fishermen’ b. [k -k aw-e ]-k-wa head-tree-NOML-CONN-NOML ‘one who owns a horn owner’ = ‘cattle owner’ In both cases the derived noun is input to another application of the possessor noun formation. It seems that an intervening connector (-k) is necessary in order to avoid strict adjacency of the two morphological exponents of this operation. 3.8. The reinterpretation of nominalization as verbal voice There is one language type in which verbal voice morphology is derived from nominal morphology, i.e. markers of nominalization: the voice system of West-Austronesian languages, most thoroughly studied in the Philippine language Tagalog (Schachter & Otanes 1972; Kroeger 1993). In these lan-

From rags to riches. Nominal linking in contrast to verbal linking

185

guages, a special verbal marker indicates the argument role of the most prominent syntactic argument; the prominent argument receives a special case marker, often called nominative although it does not function as default linker. Actor voice (AV) is used if the agent argument is most prominent, while object voice (OV) is used if the lowest argument is most prominent; instrumental voice and locative voice often have the double function of introducing a new verbal argument and designating it as the most prominent argument. Interestingly, in many of these languages the voice markers are also used for the derivation of deverbal nouns – at least partially. Let me illustrate this with data from Mayrinax Atayal, a Formosan language. In this language, nominalization markers overlap completely with verbal voice markers; there are no other means of nominalization, as can be found in other West-Austronesian languages (e.g. Tagalog). Whereas in (29a) the actor voice infix -um- is used as a verbal voice marker, it is used in (29b) as marker of agent nominalization. Note that the category of the form q-um-uriq cannot be determined in isolation; in (29b) the preceding case marker ku indicates the nominal use of the form. (29) Mayrinax Atayal: actor voice and agent nominal (Huang 2002: 213) a. quriq ru m-aiyay la steal and AV-escape PART ‘he stole (things) and escaped’ b. m-aiyay ku quriq la AV-escape NOM steal PART ‘the thief has escaped’ (30a) illustrates the use of -un as object voice in the verb and (30b) its use as marker of patient nominalization, which requires irrealis reduplication in addition. Object voice is used with a prominent internal theme argument, which receives nominative; the agent argument must then be realized in the genitive. (30) Mayrinax Atayal: object voice and patient nominalization (Huang 2002: 202/215) a. niq-un ni sayun ku ßua NOM sweet.potato eat-OV GEN S ‘Sayun ate sweet potatoes’

186 Barbara Stiebels b. maniq cu aqih na na~niq-un ru si-putak=nia AVeat ACC bad GEN IRR~eat-NOML and IV-vomit-3SG.GEN ‘he ate spoiled/bad food [=the eaten stuff] and threw up’ The following list of usages of verbal voice markers in Mayrinax Atayal demonstrates the strong parallelism of verbal voice marking and corresponding nominalization: in general, the argument that is most prominent in the verb becomes the referential argument of the nominalization. There is only one deviation: since there is no specific verbal voice marker if there is no salient argument, event nominalizations are also derived by means of the actor voice marker. (31) Mayrinax Atayal: voice marking and nominalization (Huang 2002) a.

actor voice (realis) /m-, ma-, -/ m-aquwas ‘sing’ m-aniq ‘eat’

m-aquwas m-aniq

b.

actor voice (realis) /pa-/ pa-paquwas ‘sing’

agent nominalization pa-paquwas ‘singer’

c.

actor voice (realis) /m-, ma-, -/ m-aquwas ‘sing’ quriq ‘steal’

agent nominalization

d.

event nominalization

m-aquwas quriq

‘singing’ ‘eating’

‘singer‘ ‘thief’

object voice (irrealis) /-un/;  na~niq-un ‘eat’ pa~putu ‘cut’

patient nominalization na~niq-un pa~putu

e.

locative voice (irrealis) /–an/ na~niq-an ‘eat’ pa~pqwas-an ‘sing’

locative nominalization na~niq-an ‘restaurant’ pa~pqwas-an ‘sing hall’

f.

instrumental voice (irrealis) /Ca~-/ pa~patiq ‘write’ ka~kalo ‘comb’ ca~capux ‘sweep’

instrument nominalization pa~patiq ka~kalo ca~capux

‘food’ ‘things to cut’

‘pen’ ‘comb’ ‘broom’

Linguists such as Ross (2002) have argued that the voice markers originated from nominalization structures; the prominent argument was integrated

From rags to riches. Nominal linking in contrast to verbal linking

187

into a predicative structure in which it was identified with the referential argument of the nominalization. Ross illustrates his assumption with the following example from the Formosan language Paiwan, for which he reconstructs the meaning indicated below. According to Ross, the suffix -n derived a patient nominal and only later became a marker of object voice. (32) Paiwan (Ross 2002: 39) tkl-n nua qaa a vaua drink-NOML/OV GEN stranger DET wine ‘the stranger will drink the wine’ reconstructed: ‘the wine is something of a stranger’s to be drunk’ Accordingly, all nominalization markers were reinterpreted as verbal voice markers and ousted the original neutral verb forms. To some extent, these markers continued to be used in their nominalizing function; in some languages, additional means for nominalization emerged. This scenario would explain the parallelism of verbal voice marker and nominalization marker regarding the selected argument (= the prominent argument with verbs, but the referential argument with nouns/nominals).

4. Linking All languages have developed means to realize all the arguments of a functor, be it by case, agreement, position or a mixture of these. Argument linking makes use of various resources: it can be purely structural in exploiting the argument hierarchy, it can refer to the semantic properties of arguments according to the predicates they are part of (‘semantic roles’ in a broad sense), or it can refer to the sortal/referential properties of arguments. The last option does not yield an efficient linking pattern because arguments that have the same sortal properties cannot be discriminated. Argument linking in terms of semantic roles refers to labels such as agent, patient, theme, instrument etc. Although various linking theories make use of semantic roles (in their usual sense in Lexical Functional Grammar, Bresnan 2001, in the abstract sense of actor and undergoer in Relational Grammar, van Valin 1993, or in terms of Proto-Agent and Proto-Patient, Primus 1999), there seem to be no clear cases in which argument linking solely relies on semantic roles, i.e. the linkers directly match the semantic roles to be distinguished independently in the respective language.

188 Barbara Stiebels 4.1. The generalized case type Linking systems of the generalized case type (see Wunderlich, this volume) have been studied most thoroughly in the linguistic literature. Therefore, I do not want to focus on this type. Linking of the generalized case type refers to some notion of abstract case. In LDG abstract case is determined by the argument hierarchy, which itself is derived from the semantic representation of the respective item according to the depth of embedding, as shown for the simplified representation of give: (33) Argument hierarchy give z y x s [ACT(x) & BECOME(POSS(y,z))](s) z < y < x (< s) z, being the lower argument of the POSS relation, is the most embedded argument and, hence, the lowest-ranked argument. LDG accounts for the argument hierarchy by means of the two structural features [+hr] ‘there is a higher argument’ and [+lr] ‘there is a lower argument’. (34a) shows the evaluation of the argument structure of a ditransitive and an intransitive verb in terms of the two features. Note that the situational variable s does not count for the evaluation of these features in verbs; there is no possibility to project this argument into the morphosyntactic structure of the clause headed by the corresponding verbal predicate. (34) Evaluation of verbal argument structures a. ditransitive verb

b. intransitive verb

+hr  +hr  –hr  z



y

x

–lr  +lr  +lr

–hr s

x



 s

–lr

Not all languages use both [hr] and [lr] in verbal linking. In contrast to verbs, the referential argument of nouns, which is syntactically active in predicative uses, does count for the evaluation in terms of these hierarchy features (see (35)). As shown in section 2, languages do no distinguish between predicative and non-predicative uses of nouns: The internal arguments are linked invariantly independent of the binding/saturation of the referential argument. Therefore, I assume that the referential argument is evaluated in terms of [hr] in any case.

From rags to riches. Nominal linking in contrast to verbal linking

189

(35) Evaluation of the argument structure of relational nouns v +hr

u 


hr



For nouns, only [hr] plays a role; the feature [lr] only comes in via lexical designation (see below). LDG assumes that the structural linkers are specified in terms of these features: (36) Specifications of case/agreement linkers in LDG [ ] Nominative (NOM/N) [+hr] Accusative (ACC/A) [+lr] Ergative (ERG/E) [+hr,+lr] Dative (DAT/D) [+hr] Genitive (GEN/P) Nominative case (NOM) and agreement (N) are default linkers; they are maximally underspecified. Accusative and genitive only differ in their categorial designation (both being [+hr]). Which of these two linkers bears a categorial signature (as verbal or nominal linker) depends on how the two linkers compete in the respective language. In Russian, for instance, accusative bears a categorial signature as verbal linker, whereas the genitive does not; therefore, genitive is also used with verbs if the respective animate nominal argument does not have a distinct accusative form. Note that a potential categorial underspecification of the linkers accounts for linker overlaps between the various functor categories quite easily (see Stiebels 2002): Nouns, verbs and adpositions may share some or all of their linkers. The most natural overlap is that of accusative/genitive (NOM/ACC) or object/possessor agreement (A/P) because both are [+hr]. If a language lacks ACC/A and GEN/P, NOM/N may be used for the linking of possessor arguments (or other internal arguments of nouns). These two types of linker overlaps are structurally unmarked because they do not need any additional mechanism besides the lexical specification of the linkers. Another pattern of overlap is the use of DAT/D or ERG/D as structural nominal linker. The Eskimo-Aleutan language Yupik, for instance, uses ERG as possessor linker:

190 Barbara Stiebels (37) Yupik: ERG as possessor linker (Mithun 1999b: 250) tan’gaurlur-m taluyar-i boy-ERG fishtrap-3SG/3PL ‘the boy’s fishtraps’ If DAT or ERG are used as structural linkers of nouns/nominals, the internal argument of the noun must be lexically specified as [+lr], which is motivated by the fact that the possessor argument v is a quasi-medial argument on SF; (38) shows the representation of possessed nouns in these languages: When nouns undergo possessor extension, the possessor argument is designated as [+lr]. (38) Possessed nouns with DAT/D or ERG/E possessor linking v u [NOUN(u) & POSS(v,u)] +hr –hr +lr DAT ERG (if

the language lacks DAT)

Depending on the presence of DAT/D or ERG/E in the linker inventory of the language, one of these linkers may be used for nominal linking. These overlaps are structurally marked due to the required lexical designation of the possessor argument (see Stiebels 2002 for details). The systems I want to discuss in the following are not based on a purely structural notion of argument linking in the sense of generalized case. The following systems more or less implement semantic notions (situation control, sortal features, discourse features etc.). 4.2. The active type In the active type (see also Comrie 1989; Merlan 1985; Mithun 1991; Dixon 1994), verbs are distinguished according to the subject’s disposition to control the situation denoted by the verb.10 Many languages restrict the active pattern to intransitive verbs, distinguishing active and inactive verbs ———–———————————————————————————— 10

Note that sortal features play a role as well, i.e. active languages may distinguish between usages of a predicate with a human or animate subject or those with an inanimate subject; the former will trigger the active pattern, the latter the inactive pattern.

From rags to riches. Nominal linking in contrast to verbal linking

191

by means of the linkers, as shown in (39b) for the Muskogean language Koasati. Active verbs in Koasati select agreement markers from the active linker inventory (ACT), whereas inactive verbs select agreement markers from the inactive linker inventory (INACT). In Koasati, many active verbs have an inactive counterpart, as shown for ‘sleep’ and ‘laugh’ in (39); the shift from active to inactive reading may also involve a shift in aktionsart. (39) Active type Koasati (Kimball 1991: 252) active (ACT-linker)

inactive (INACT-linker)

a. nó:ci-li-t sleep-1SG.ACT-PAST ‘I slept’

ca-nó:ci-t 1SG.INACT-sleep-PAST ‘I fell asleep’

b. afá:ka-li-t laugh-1SG.ACT-PAST ‘I laughed’

ca-afá:ka-t 1SG.INACT-laugh-PAST ‘I burst out laughing’

A few inactive verbs such as ‘be hungry’ (see (40a)) take the agreement marker from another inventory of linkers (D-linkers), which can be regarded a lexicalized pattern. (40) Active type Koasati (Kimball 1991: 132, 254) inactive (INACT-linker)

inactive (D-linker)

a.

kom-akásn 1PL.D-be.hungry.PL ‘we are hungry’

b. ca-hó:p 1SG.INACT-be.sick ‘I am sick’

am-hó:p 1SG.D-be.sick ‘I am hurt’

In addition, some inactive verbs exhibit an INACT-linker/D-linker variation (see (40b)). Kimball (1991) points out that in these cases, the INACT-linker is used for more permanent situations, whereas the D-linker is used for temporally restricted situations. Some languages of the active type extend the active/inactive distinction to polyadic verbs, as shown in (41) for Koasati. In contrast to the active transitive verb in (41a), the highest argument of the transitive inactive verb in (41b) receives an INACT-linker. The mediopassive marker -ka productively derives inactive verbs from transitive active verbs such as (41c); the

192 Barbara Stiebels resulting inactive verb links the highest argument by means of the D-linker as in (41d). Therefore, intransitive and derived transitive inactive verbs differ in their subject agreement linker. (41) Koasati: transitive active vs. inactive verbs (Kimball 1991: 128/256/77) a. mán haci-hí:ca-li-lahõ again 2PL.INACT-see-1SG.ACT-IRR ‘I will see you all again’ b. ci-ca-bàn 2SG.INACT-1SG.INACT-need ‘I need you’ c. a:nipó-n onaksóhli-li-t meat-ACC char.TR-1SG.ACT-PAST ‘I charred the meat (on purpose).’ d. a:nipó-k am-onaksóh-ka-t meat-NOM 1SG.D-char-MPASS-PAST ‘I charred the meat by accident.’ The nominal concept that comes closest to the verbal active type is the alienable-inalienable distinction in possessive relations. A possessor of an alienable possessive relation is conceived of as an entity controlling the possessive relation (being able to establish or dissolve a possessive relation), whereas a possessor of an inalienable possessive relation is unable to dissolve the possessive relation. Koasati displays this distinction: Alienable possession triggers D-agreement markers for the possessor, whereas inalienable possession triggers INACT-agreement markers. This distribution mirrors the less permanent – more permanent distinction found in some inactive verbs (see (39d) above). (42) Koasati: possessor split (Kimball 1991: 433–34) a. ca-ttii 1SG.INACT-eye ‘my eye’ [inalienable ~ more permanent] b. am-ifá 1SG.D-dog ‘my dog’ [alienable ~ temporally restricted] Interestingly, one can find reflexes of the alienable-inalienable distinction also in the verbal domain: The Bantu language Kinyarwanda does not mark

From rags to riches. Nominal linking in contrast to verbal linking

193

possessor raising in case of inalienable possession (see (43a/b)); the possessor úmwáana ‘child’ of the body part expression ukúboko ‘arm’ is turned into an object of the verb. Whereas the verb lacks a special marker in (43b), it receives the applicative marker -er as in (43d) in case of alienable possession: (43) Kinyarwanda: possessor raising (Kimenyi 1980: 97–98) a. umugóre y-a-vun-nye [ukúboko k’úúmwáana] of.child woman 3SG.N-PAST-break-ASP arm ‘the woman broke the arm of the child’ b. umugóre y-a-vun-nye [úmwáana] [ukúboko] woman 3SG.N-PAST-break-ASP child arm ‘the woman broke the child’s arm’ c. umuhuûngu a-ra-som-a [igitabo cy’ûmukoôbwa] boy 3SG.N-PRES-read-ASP book of.girl ‘the boy is reading the book of the girl’ d. umuhuûngu a-ra-som-er-a [umukoôbwa] [igitabo] boy 3SG.N-PRES-read-APPL-ASP girl book ‘the boy is reading the girl’s book’ Noun incorporation in Nahuatl also reflects the specific character of inherently relational, i.e. inalienable nouns. With the incorporation of these nouns, the possessor argument becomes an additional argument of the verb: In (44a) the possessor argument (1SG) is indexed as the subject of the verb, which is a result of noun incorporation. No such effects are documented in Nahuatl for the incorporation of non-relational (alienable) nouns. (44) Classical Nahuatl: Incorporation of inalienable nouns (Andrews 2003: 270–71) a. ni-m-sepw-k 1SG.N-hand-become.numb-PAST ‘I became numb at my hand’ = ‘my hand became numb’ b. x s BECOME(NUMB(x))(s)  y s BECOME(NUMB(x (HAND(x) & PART_OF(x,y))))(s) (44b) illustrates the effect of argument extension via noun incorporation in formal terms: Since the incorporated noun is relational, it is integrated via Functional Composition with the verb; the unsaturated possessor argument y is inherited to the verb.

194 Barbara Stiebels 4.3. The salience and the inverse type There are actually two linking types that implement sortal properties of the argument referents into their linking system: the salience type and the inverse type (see also Wunderlich this volume). Both linking types are based on a salience hierarchy of sortal features such as (45a). The degree of individuation may also be reflected in a hierarchy of referential salience as in (45b). However, the latter does not play a role in the systems I want to discuss in this subsection. (45) Salience hierarchies a. 1,2 > human > animate > inanimate b. definite, proper names > specific > unspecific Nouns and verbs differ in their expected salience patterns. With verbs, the highest argument, being the instigator or controller of the situation in the prototypical case, is expected to be the more salient argument, i.e. animate or human. This is due to the apparently universal tendency of languages to lexicalize situations from the perspective of the potentially more salient entity (e.g. as cause-effect relations). In nouns, in which kinship terms and body-part expressions represent the typical relational nouns, the possessor, being human or at least animate, is expected to be the more salient argument. I will call the expected pattern the direct setting, which is unmarked, and the unexpected or less expected pattern the inverse setting, which is marked. The direct setting, illustrated for a 1st person subject/possessor and a 3rd person object/possessum, is given in (46a)/(47a), the inverse setting, illustrated for a 3rd person subject/possessor and a 1st person object/possessum, is given in (46b)/(47b). (46) Salience relations in verbs a. expected pattern (direct) VERB(x,y) y x 3 1 +ls +hr

b. unexpected pattern (inverse) y x VERB(x,y) 1 3 +ls +hr

From rags to riches. Nominal linking in contrast to verbal linking

195

(47) Salience relations in nouns (v: Possessor) a. expected pattern (direct) b. unexpected pattern (inverse) v u NOUN(u,v) v u NOUN(u,v) 1 3 3 1 +ls +ls +hr +hr Thus, in the direct setting of verbs the more salient argument (identified by the feature [+ls] ‘there is an argument with lower salience’) is [–hr], whereas in nouns the more salient argument is [+hr] in the direct setting. In the salience type, most impressively represented by Arizona Tewa (Kroskrity 1985; Wunderlich this volume), the indexing of arguments by agreement markers is restricted such that the most salient argument of a polyadic verb is to be indexed. This constraint is often due to the morphological restriction that only one person/number marker may occur, which enforces a selection procedure for the argument to be indexed. The argument role of the respective argument is identified by a set of distinct linkers that are specified with respect to the argument role of the indexed argument. However, the salience type is not yet attested for nouns, which might be due to fact that there are less contexts for a salience competition than in verbs: a) A salience competition may occur in predicatively used relational nouns that allow its referential argument to be more salient than a possessor argument (usually kinship terms). b) A salience competition may occur in event nominalizations that are based on transitive or ditransitive verbs; here the non-referential arguments may compete with each other. In the inverse type, normally both arguments are indexed. In contrast to the salience type there are contexts/structures in which it is possible to index a less salient argument. A marker on the verb signals whether the external argument referent is more salient than the internal one or vice versa. In the Algonquian languages both settings are marked, in other languages only the inverse setting is marked. The Algonquian language Potawatomi marks the direct setting by –a as in (48a), the inverse setting by -uk as in (48b). Potawatomi exhibits the salience hierarchy 2 > 1 > 3prox > 3obv, i.e. third person is further differentiated between proximate and obviative discourse referents. (48) Potawatomi: Inverse system (Hockett 1948) a. k=wapm-a-wa-k ‘you (pl) see them’ 2=see-DIRV-PLS-PLO

196 Barbara Stiebels b. k=wapm-uk-wa-k 2=see-INV-PLO-PLS

‘they see you (pl)’

Note that the person proclitics and the plural suffixes are not designated for a specific grammatical function. The proclitics index the more salient argument; the plural suffix –wa likewise indexes the more salient argument, i.e. the subject in (48a), the object in (48b). Inverse systems can also be observed in nouns. Amith & Smith-Stark (1994a/b) have described several nominal inverse systems in Uto-Aztecan languages. The system in Guerrero-Nahuatl is quite illustrative in this respect. The direct/inverse system plays a role in relational nouns that are used predicatively and may differ in the relative salience of the two arguments: kinship terms allow both arguments to show different degrees of salience. The salience hierarchy in Guerrero-Nahuatl is 1/2 > 3 (with some preference for 1 > 2). In the direct setting, the possessor is marked by the possessor affix as in (49a), whereas the highest argument is indexed by subject agreement (N-marker), which is non-overt for third person. In the inverse case, the possessor is marked by N-agreement, whereas the highest argument is marked by A-agreement as in (49b/c). The direct pattern is represented in (49d), the inverse pattern in (49e). (49) Guerrero-Nahuatl: Inverse system (Amith and Smith-Stark 1994a: 351) a. no-nn ‘she is my mother’ 1SG.P-mother b. n-nn 1SG.A-mother ‘I am his/her mother’ c. ti-n-kon-w 2SG.N-1SG.A-child-POREX ‘I am your child’ d. v u MOTHER (u,v) [direct] +ls P

e. v N

N

u +ls

MOTHER (u,v)

[inverse]

A

Since Guerrero-Nahuatl does not make use of a separate direct/inverse marker, the two settings can only distinguished by a partial substitution of the linkers; in the inverse setting, a verbal linker (A) comes into play. In

From rags to riches. Nominal linking in contrast to verbal linking

197

Stiebels (2002) I have provided an analysis of the inverse patterns in Guerrero-Nahuatl, Cahuilla and Chinook. Though all three languages show an equivalent sensitivity for the relative salience of the two arguments of the noun, these languages each require subtly different mechanisms in the analysis of their nominal inverse system. A somewhat different nominal inverse system is found in the Algonquian languages, which is triggered by the obviation system of these languages. In Plains Cree, for instance, the possessor is regularly more prominent than the possessum; therefore, a third person possessor is proximate in the default case, whereas the possessum then must be obviative because only one proximate argument is allowed. With a first person possessor as in (50a), the possessum does not need to be obviative; it can be obviative as in (50b). However, with a third person (proximate) possessor, the possessum must be obviative as in (50c). The unexpected situation of an obviative possessor is marked by the nominal inverse marker (-iyi) as in (50d). (50) Plains Cree: nominal inverse system (Wolfart 1973: 31) a. ni-tm ‘my horse’ 1SG-horse b. ni-tm-a 1SG-horse-OBV ‘my horse (obv)’ c. o-tm-*(a) 3SG-horse-OBV ‘his horse (obv)’ d. o-tm-iyi-wa 3SG-horse-INV-OBV ‘his (obv) horse’ The data of this section show that the less frequent type of inverse linking can be found both in verbs and nouns. The crucial difference between their inverse linking is, however, the alignment of the salience hierarchy with the argument hierarchy.

4.4. The portmanteau type In the portmanteau type (Wunderlich this volume), most consequently represented by Kiowa, the linking features of subject and object are mapped onto a fused morpheme, which encodes both person and number features of subject and object. (51) provides some examples for Kiowa.

198 Barbara Stiebels (51) Kiowa: portmanteau type (Watkins 1984) a. èm-góp I hit you/him èm b. é-góp you/he hit me é

c. d-góp you/he/they hit us d  d. g-góp we/he/they hit you(sg) g In the extreme case, languages of the portmanteau type would mark each possible subject-object setting with a distinct marker. However, there is no such case attested; the markers in Kiowa show some degree of syncretism. Apart from the fact that Kiowa argument linking lacks the extent of generalization found in systems of generalized case, it is purely structural. However, there could be portmanteau systems that also implement semantic features; I will discuss such a case below. The number of possible possessor-possessum settings is smaller than the number of subject-object settings because most nouns only allow a third person possessum. This raises the question of whether there are nominal portmanteau systems and how they look like. There are traces of the portmanteau type in nouns as well. This is true for languages that display an overlap of verbal and nominal linking patterns, as shown in the following for Yupik and Huastec: Whereas the ergative language Yupik chooses a structural assignment of the features (consequently aligning -features (person/number) to the [+lr] or [–lr] argument), Huastec makes use of a salience-based portmanteau type (Zavala 1994; Edmonson 1988). The following example from Yupik shows that -nka assigns the feature [+1] to the [+lr] argument, whereas the other argument receives the number feature [+pl]; compare the nominal pattern in (52d) with the verbal pattern in (52e). Since a lexical possessor argument is marked by ERG, as shown in (37), it must be lexically designated as [+lr], as shown in (52d). (52) Yupik: portmanteau type (Mithun 1999b: 73, 250) a. ila-nka ‘my relatives’ relative-1SG/3PL b. paqete-a-nka visit-IND-1SG/3PL ‘I visited them’ c. -nka [+1,+lr;+pl] d. v u RELATIVE(u,v) +lr +1 +pl

From rags to riches. Nominal linking in contrast to verbal linking

e. y –lr +pl

x +lr +1

199

s VISIT(x,y)(s)

In Potosina-Huastec the two fused pronominal markers tin and tu differ in terms of salience regarding their distribution in purely local subject-object settings. In the direct setting, in which the subject outranks the object and the possessor the possessum on the basis of the scale 1 > 2, tu is used (see (53a/b)); in the inverse setting, tin is used.11 (53) Potosino-Huastec: nominal portmanteau pattern (Edmonson 1988: 125–26) a. nan( ) tu tak-a( )-al I 1/2 touch-TR-INCOMPL ‘I touch you’ b. pl tu ebi-il be 1/2 teacher-POREX ‘you are my teacher’ c. nna tat tin k a-a hale t-it tsakl momma you 2/1 hit-TR because COMP-2.N angry ‘Momma, did you hit me because you’re angry?’ d. (tat ) pl tin ebal-il (tat ) (you) be 2/1 student-POREX (you) ‘I am your student’ The analysis of the linker distribution in Potosino-Huastec must align two salience relations: the inherent category-specific disposition for salience (let me call this feature [+ils] ‘there is an inherently less salient argument’) and the overall salience in terms of the sortal hierarchy. (54) shows the distribution of the patterns in Huastec.

———–————————————————————————————

11

That Potosino-Huastec works on basis of a salience hierarchy is not obvious from the isolated examples given here. In view of the overall paradigm (see Edmonson 1988; Zavala 1994), the salience analysis is, however, justified.

200 Barbara Stiebels (54) Huastec: distribution of tu/tin in nouns and verbs a. tu V : y x TOUCH(x,y) 2 1 +ils direct +ls b. tu N: v u TEACHER(u,v) 1 2 +ils direct +ls c. tin V: y x HIT(x,y) 1 2 +ils inverse +ls d. tin N: v u STUDENT(u,v) 2 1 +ils inverse +ls

(53a)

(53b)

(53c)

(53d)

There seems to be no nominal linking system of the portmanteau type that has evolved independently from a corresponding verbal one. However, given the rare occurrence of pure portmanteau systems, it is not astonishing that this type is also only rarely represented among the nominal linking types. 4.5. Minor linking types There are two linking types whose distribution is rather limited: the incorporational type (Wunderlich 2002) and the Philippine-style voice system (see 3.8). In the incorporational type, represented by some Paleo-Sibirian languages (e.g. Nivkh, Matthissen 2003, this volume) and to some extent by other polysynthetic languages, internal arguments of the verb are incorporated unless they are information-structurally salient. In general, the arguments are incorporated according to their position on the argument hierarchy, starting from the lowest-ranked argument.12 Again, one may ask ———–———————————————————————————— 12

In Stiebels (1998) I have made use of the Principle of Lexical Incorporation, which states that the lowest-ranked argument is occupied by the base noun of a denominal verb; the restriction also applies in the derivation of admissible patterns of argument-saturating incorporation.

From rags to riches. Nominal linking in contrast to verbal linking

201

whether the verbal linking strategy of incorporation is extended to argument linking in nouns as well. (55) shows some representative patterns of noun incorporation (NI) in Nivkh. (55a) shows NI with a transitive verb, (55b/c) with a ditransitive verb. In (55b) a theme object is incorporated, i.e. the lowest argument of an abstract possessive relation, in (55c) a locative argument, the lowest argument of an abstract local relation (see also Wunderlich 2002). The secondto-lowest argument is realized as external NP/DP. (55) Nivkh: noun incorporation (Mattissen 2003: 146, 143) a. t f-ke- house-buy-IND/NOML ‘someone buys a house’ b. ñi Xevgun caqo-asqam- 1SG X knife-take.away-IND/NOML ‘I took the knife away from Xevgun’ c. co-

-ñivx ke toqo-xro- fish-hunt-person net fence-hang-IND/NOML ‘the fisherman hung the net on the fence’ d. z y x s [TAKE(x,y) & BECOME(¬POSS(y,z))](s)  y x s [TAKE(x,y) & BECOME(¬POSS(y,
z KNIFE(z)))](s) e. z y x s [HANG(x,y) & BECOME(LOC(y,R[z]))](s) 13  y x s [HANG(x,y) & BECOME(LOC(y,
z [FENCE(z)]))](s) (55d/e) show the derivation of the incorporation patterns in (55b) and (55c), respectively. In both cases, the lowest-ranked argument is saturated by the incorporated noun. Besides the well-established verbal patterns, Nivkh also exhibits incorporation into a nominal functor, i.e. possessors may be incorporated as in (56b/c). Even WH-elements questioning a possessor may be incorporated as in (56d). Relational nouns, which correspond to adpositions, may also incorporate their internal argument; the resulting complex may be incorporated into the verb as in (56e). Note that incorporation affects the realization of the stem-initial consonant of the head noun tf in (56b); see Mattissen (2003) for details. ———–————————————————————————————

13

R in (55e) denotes an abstract neighboring region with respect to the relatum.

202 Barbara Stiebels (56) Nivkh: other patterns of incorporation (Mattissen 2003: 225/148) a. tf ‘house’ father-house ‘father’s house’ b. tk-rf Seus-house ‘Seus’ house’ c. Seus-tf who-boat-Q ‘whose boat …?’ d. a
-mu-
a e. [ño-vj]-si- barn-under-put-IND/NOML ‘put sth. under the barn’ Besides these polysynthetic structures of NI, synthetic compounds (e.g. man-eater) could be analyzed as nominal instantiations of NI: The internal argument of the underlying verb is incorporated as the non-head of the compound, suggesting a semantic bracketing such as [[man-eat]-er]. Alternatively, one might say that the argument requirements of the underlying verb are inherited to the nominalized verb, hence man is incorporated into eater. These synthetic compounds are also available in languages that lack NI in verbs (e.g. German and English). In languages that exhibit verbal and nominal NI, synthetic compounds are ambiguous between a structure [[NV]-NOML] (based on verbal NI) and [N-[V-NOML]] (nominal compounding). The Philippine-style voice system discussed briefly in 3.8 is another minor linking system. As shown for Atayal, the verbal voice markers are also used as nominalization markers. Each marker affects a specific argument of the underlying verb, be it in its voice marker function or in its nominalizing function. However, the parallelism does not go beyond the affected argument. As a voice marker, it indicates syntactic prominence of the respective argument. No such function applies for its use as nominalization marker, i.e. the voice markers are not used to signal syntactic prominence of the respective nominal argument. I will return to the differences between nouns and verbs in terms of information structure in section 6. 4.6. Linking splits Both verbal and nominal linking may exhibit linking splits, which are structures of inhomogeneous linking: Depending on the context, different linkers are used for an argument of a given functor. For instance, objects may be marked by nominative instead of accusative, subjects may be marked by nominative instead of ergative. Triggering factors for a split are referential features (definiteness, specificity), sortal features (e.g. animacy) or aspectual/temporal features. Such splits may also reflect inhomogeneities between categories: Pronouns or pronominal affixes are likely to instantiate an

From rags to riches. Nominal linking in contrast to verbal linking

203

accusative system, whereas lexical nouns may instantiate an ergative system. These splits have been thoroughly documented in the literature (see e.g. Silverstein 1976; Hopper & Thompson 1980; Givon 1984; Comrie 1989; Dixon 1994). A purely verbal character is exemplified by those linking splits that are triggered by verbal categories such as tense, aspect or polarity; the aspectually conditioned ergative split in Hindi would be a case in question: Ergative is licensed with perfective verbs as in (57a), but excluded with imperfective verbs as in (57b). (57) Hindi: aspectually conditioned ergative split (Mohanan 1994: 70) a. raam-ne ravii-ko piiTaa Ram-ERG Ravi-ACC beat.PERF ‘Ram beat Ravi’ b. raam ravii-ko piiTtaa hai Ram.NOM Ravi-ACC beat.IMPF be.pres ‘Ram beats Ravi’ In contrast, linking splits that are determined by the referential or sortal properties of the respective argument (‘differential object marking’, e.g. Aissen 2003; Stiebels 2000, 2002) can be observed for both verbs and nouns. An illustrative case is the use of a [+hr] linker (accusative/genitive) for specific arguments in Turkish languages, shown here for Tyvan. Accusative can only be used for specific objects, see the examples in (58a/b). The genitive underlies a similar constraint; it is only licensed with specific possessors (see (58c/d)) or if the noun is modified, thus being restricted in its set of potential referents as in (58e). (58) Tyvan (Anderson & Harrison 1999: 17/25) a. men nom nomudum ‘I read a book’ I book read.PAST.1SG b. men nom-nu nomudum I book-ACC read.PAST.1SG ‘I read the book’ c. dilgi ke-i fox fur-3SG.P ‘fox fur’ d. dilgi-ni ke-i fox-GEN fur-3SG.P ‘the fur of the fox’ e. dilgi-ni ımak ke-i fox-GEN soft fur-3SG.P ‘soft fox fur’

204 Barbara Stiebels Linking splits reflect morphosyntactic economy: the unexpected occurrence of a category/feature is marked. Thus, an animate or definite object is less expected than an inanimate or unspecific object. This type of markedness distribution can be easily modeled by means of Harmonic Alignment (see Aissen 1999, 2003; Stiebels 2000, 2002). However, once again a specific difference between verbs and nouns emerges: the internal argument of verbs is expectedly non-salient, whereas the internal (possessor) argument of nouns is salient. Therefore, an animate possessor is unmarked, an inanimate possessor is marked. The Mayan language Itzaj Maya neatly reflects this salience preference: Possessor extension is not morphologically marked if the possessor (of a body part) is alive; however, if the possessor is dead (inanimate), possessor extension is indicated by a suffix. (59) Itzaj Maya: possessor split (Hofling 2000: 267) a. u-tz’omen aj-Jwan ‘the brain of Juan (Juan is alive)’ 3SG.E-brain MASC-Juan b. u-tz’omen-il aj-Jwan 3SG.E-brain-porex MASC-Juan ‘the brain of Juan (Juan is dead)’ The distinction of alienable-inalienable possession, which I have discussed in 4.2, is often treated as a nominal split, not as a distinct linking type per se. However, it deviates from many splits in that it does not reflect morphosyntactic economy; the active or alienable pattern is not more or less marked than the inactive or inalienable pattern. Both patterns simply indicate underlying semantic differences of the SF’s predicates of the relevant items. As I have tried to show in 4.2, the alienable-inalienable pattern is equivalent to the active type in that in both cases the argument referent’s disposition to control the situation/relation denoted by the verb or noun plays a significant role. Even languages that do not encode the alienable-inalienable distinction in morphosyntactic terms (such as Kinyarwanda or Nahuatl) show some sensitivity to it. The subsections in 3 and 4 have shown that nouns and verbs show more similarities in terms of voice and linking strategies/resources than one might expect on the first view. The following section deals with the problems nominals encounter with the challenging ‘verbal heritage’ in case of event nominals and thus highlights the major difference between verbs and nouns.

From rags to riches. Nominal linking in contrast to verbal linking

205

5. The ‘riches’: event nominals As already indicated in section 2, event nominals pose a serious challenge to the structural linking potential of nouns because the number of arguments of nominalizations may exceed the number of structural arguments in underived nouns. Languages show different strategies to compensate for the structural deficits of nouns, for instance, the use of mixed categories or the transfer of arguments to other functors. The restriction of argument realization in nominalizations can be illustrated by two cases in German: It is not possible to realize a canonical dative as in (60a) or a lexical dative as in (60c) in the corresponding nominalization (see (60b/d)); the only alternative available (for some of the nominals) is the use of an oblique marker for the respective argument. It is not possible to use the structural nominal linker (genitive), as indicated in (60b). Likewise restricted is the use of reflexives with the nominalization of inherently reflexive verbs as sich schämen ‘be ashamed’ (see (60e)). The reflexive must be omitted; even the replacement by the emphatic anaphor selbst ‘self’ is excluded. (60) German a. den Kinder-n ein Buch schenken the.DAT.PL child.PL-DAT a book donate ‘donate a book to the children’ b. * die Schenk-ung eines Buch-es den Kinder-n/ the give-NOML a.GEN book-GEN the.DAT.PL child.PL-DAT/ der Kinder the.PL.GEN child.PL ‘the giving of a book to the children’ c. den Kinder-n helfen the.DAT.PL child.PL-DAT help ‘help the children’ d. * die Hilfe den Kinder-n/ die Hilfe für die Kinder det help the.DAT.PL child-DAT.PL/ the help for the.PL child.PL ‘the help for the children’ e. sich schämen REFL be.ashamed die Scham the shame * die Sich-Scham the REFL-shame * die Selbst-Scham the self-shame

206 Barbara Stiebels In the following subsections I want to show how languages might compensate for their structural deficits in nominals. The discussion is based on Koptjevskaja-Tamm’s (1993) exhaustive study of nominalization patterns. However, I will not focus on the role of retained verbal functional categories in the nominals (e.g. aspect or mood) and the degree of ‘nouniness’ (see Malchukov 2004 for some discussion). 5.1. Major linking types in nominalizations Koptjevskaja-Tamm (KT in the following) distinguishes four major types of event nominalizations: a) the nominal type, which exhibits a complete nominal linking pattern, b) the sentential type, which exhibits a complete verbal linking pattern, c) the ACC-POSS type, which shows a mixed verbalnominal linking pattern, and d) the POSS-ERG type, which also exhibits a mixed structural/non-structural linking pattern.14 The last two types differ regarding the argument that is marked by the structural nominal linker (‘POSS’ in the following): in the ACC-POSS type, the highest argument is realized nominally, in the POSS-ERG type, the lowest argument is realized nominally. I consider the notion of a POSS-ERG type a real misnomer; there is no (structural) ergative involved. The respective argument receives an oblique linker; therefore, I will call this type the POSS-OBL type. In addition, I will call the sentential type the ‘verbal type’. The structure of the various patterns will become clear in view of the relevant examples. Neither KT’s presentation nor my discussion presuppose that a language should make use of just one linking pattern; languages may have various options, often related to a specific type of nominalization. Only few languages allow the doubling of the structural nominal linker; the Austronesian language Tukang Besi is a representative of the option that KT classifies as nominal type; here, all inherited arguments of the base verb may be realized in the nominalization by means of the genitive or possessor agreement. In (6), repeated here as (61), the causee and the theme argument of the verb ‘make eat’ are marked by genitive, whereas the causer is marked by possessive agreement. ———–———————————————————————————— 14

KT actually uses the labels of ‘POSS-ACC’ type and ‘ERG-POSS’ type; in her terminology, the left linker refers to the higher argument, the right linker to the lower argument. In order to reconcile the labels with my representations of argument structure, I have changed the order: the left linker refers to the lower-ranked argument, the right to the higher-ranked argument.

From rags to riches. Nominal linking in contrast to verbal linking

207

(61) Tukang Besi: nominal type (Donohue 1999: 319) [te pa-manga-’a-u u Aswi nu ika atu], core CAUS-eat-NOML-2SG.P GEN Aswin GEN fish that no-marasai na ’ita-ra-no 3.N-difficult NOM see-NOML-3.P ‘the way you fed the fish to Aswin, it didn’t look easy’ German is much more restricted in this respect: it allows the realization of at most one preverbal genitive, which is restricted to proper names, and one postverbal genitive: (62) FriedrichsGEN Armee [der langen Soldaten] GEN ‘Friedrich’s army of long soldiers’ Languages may circumvent the restrictions of nominal linking (specifically the ban on linker doubling) by the use of so-called mixed categories. All arguments but one are realized verbally (i.e. by means of verbal linkers), the remaining argument is realized nominally. The following example illustrates the mixed use of linkers in Turkish nominalizations. As far as intransitive verbs are concerned, the nominal linker suffices as in (63a). The nominalization of a transitive verb, however, requires a mixed linking pattern: The internal argument of temizlemek ‘clean’ in (63b) is marked by accusative, which is restricted to verbs, whereas the agent is realized by genitive. (63) Turkish: event nominalization; ACC-POSS type (Borsley & Kornfilt 2000: 106) a. ua-ıı n hastalan-ma-sın-dan dolayı servant-GEN fall.ill-NOML-3SG.P-ABL because ‘because the servant got sick …’ b. Hasan [ua-ıı n oda-yıı dikkat-li-ce Hasan servant-GEN room-ACC care-with-ADV temizle-me-sin]-i söyle-di clean-NOML-3SG.P-ACC say-PAST ‘Hasan said that the servant should clean the room carefully’ Basque is a representative of the verbal type, i. e. all arguments of the nominalization may receive the verbal linker, which is shown for the ditransitive verb ‘give’: the nominal takes the canonical verbal NOM-DAT-ERG pattern; the nominal linker genitive is not used.

208 Barbara Stiebels (64) Basque: verbal type of nominalization (Saltorelli 1988: 155) Mikel-ek ama-ri diru-a ema-te-a … Mikel-ERG mother-DAT money-NOM give-NOML-DET.NOM … ‘Mikel’s giving the money to mother [appears very good to me]’ The Salishan language Shuswap exhibits a mixture of nominal and verbal type: with nominalizations of intransitive verbs, the nominal linker (possessor agreement) is used (see (65b)), whereas with nominalizations of transitive verbs, the regular verbal pattern (A-agreement and E-agreement) is used (see (65c/d)).15 (65) Shuswap: partly nominal, partly verbal type of nominalization (Kuipers 1974: 41/43) a. k’ex-n-s dry-TR-3.E ‘he dries it’ b. -n-s- °í °lx DET-1SG.P-NOML-jump ‘my jumping’ c. -s-k’ex-n-s DET-NOML-dry-TR-3.E ‘his drying it’ d. -s-q°l-n-cécm-s DET-NOML-summon-TR-1SG.A-3SG.E ‘his summoning me’ Languages of the verbal type may either behave like Shuswap, i.e. triggering a nominal linker if only one argument needs to be linked, or instantiate the verbal linking pattern in all contexts. The POSS-OBL type can be illustrated with data from the Caucasian language Abkhaz. Here, the possessor is indexed by the dative agreement marker (D) as in (66a). This marker is also used to index the single argument of a nominalized intransitive verb as in (66b), the lowest argument of a nominalized transitive verb as in (66c), and the two lowest arguments of a nominalized ditransitive verb as in (66d). With nominalized transitive and ditransitive verbs the agent argument can only be realized obliquely.16 ———–———————————————————————————— 15

16

Other Salishan languages are more complicated; the use of nominal or verbal linking also depends on morphological conditions and the presence of auxiliaries (see Kroeber 1999). According to Koptjevskaja-Tamm, the doubling of D-linkers in (66d) is not fully acceptable, which is expected in view of the fact that Abkhaz is not an instance

From rags to riches. Nominal linking in contrast to verbal linking

209

(66) Abkhaz: POSS-OBL type (KT 1993: 135–36) a. à-’k°’n ye-y°n the-boy 3SG.D-house ‘the boy’s house’ b. larà l-ca-rà she 3SG.D-go-NOML ‘her going’ c. [sarà s-la] b  -ba-rà I 1SG-by 2SG.D-see-NOML ‘my seing you’ d. ?[darà r-la] s-w-ta-ra they 3PL-by 1SG.D-2SG.D-give-NOML ‘their giving me to you’ In view of the four major types of linking patterns in nominals the question arises as to which factors determine the existence and properties of these patterns. Regarding the possible linking patterns in nominals, Croft (1991) has formulated the following Deverbalization Hierarchy, which is based on observations by Comrie (1976): (67) Deverbalization Hierarchy (Croft 1991: 83) If a verbal form inflects tense-aspect-modality (TAM) like a predicated verb, then it will take subject and object dependents like a predicated verb. If a verbal form takes a subject dependent like a predicated verb, then it will take an object dependent like a predicated verb. On the one hand, Croft relates the verbal type to the presence of verbal functional elements in the nominal; if a nominal includes a verbal category of tense, aspect or mood, it should display a verbal linking pattern. (67) does not state that a verbal linking pattern implies the presence of verbal functional categories in the nominal. On the other hand, Croft predicts that nominalizations of polyadic verbs cannot show a pattern in which the lower argument receives a nominal linker and the higher argument a verbal linker, hence there should be no POSS-ACC (or POSS-NOM) pattern. In my opinion, ———–———————————————————————————— of the nominal type, which allows the doubling of the structural nominal linker. I would speculate that the D-linker for the recipient argument is an oblique linker, whereas the D-linker for the lowest argument is a structural linker.

210 Barbara Stiebels (67) is one step towards an explanation of the linking patterns, but does not suffice to derive all four patterns. In the following, I will try to argue that the linking patterns of nominals are confined by and large to the major types discussed here because nouns differ from verbs in their preference for strict successive argument saturation. I will also outline which additional factors play a role, thus accounting for the verbal and the POSS-OBL type. 5.2. The role of strict successive argument saturation in nominals Which factors might influence the presence the four major types of linking in nominalizations? In this section I would like to show that the four pattern emerge from a specific requirement of nouns and nominalizations: argument saturation has to apply successively, starting from the lowest-ranked argument; argument swapping is strongly dispreferred. Consequently, the lowest argument gets strong preference in nominal structural linking. Another factor for the instantiated linking type is the presence of verbal subdomains on the -grid of nominals. Let us compare the four types. I will use the hypothetical event nominalization of a ditransitive verb for the sake of illustration because it illustrates the challenge for structural linking most clearly. The nominalization of intransitive verbs does not pose any challenge for nominal linking as (68) illustrates. The situational variable s now counts for argument linking; as the highest argument it is [–hr]; I will assume that it is referentially bound by a D-element. There is only one internal argument to be saturated. For the sake of illustration, I will use the label POSS for the structural nominal linker, ignoring for the moment the fact that languages differ in their structural nominal linker (e.g. genitive, dative, ergative). (68) Nominalization of intransitive verbs x s V(x)(s) +hr –hr POSS D The picture changes with nominalizations of transitive and ditransitive verbs; there are two or three [+hr]-arguments, respectively. In structures of the nominal type, all arguments are part of the nominal linking domain.

From rags to riches. Nominal linking in contrast to verbal linking

(69) Nominal type z y x +hr +hr +hr [ POSS

POSS

POSS

211

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

s –hr ]N D

A crucial prerequisite for the nominal type is the possible doubling of nominal linkers because, apart from those languages that show a possessor split, no language has more than one structural linker for nouns/nominals.17 The POSS-OBL type also instantiates a purely nominal linking domain. Since the structural linker cannot be doubled, the nominal linker is restricted to the lowest argument of the linking domain. The other arguments must be realized obliquely – by oblique linkers that are either restricted to nouns/nominals or that occur with both verbs and nouns. (70) POSS-OBL type z y x +hr +hr +hr [ POSS ? OBL

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

s –hr ]N D

Note that the POSS-OBL type (ERG-POSS in KT) is often analyzed as an ergative structure (e.g. KT; Laczkó 2000; Alexiadou 2001) because the single internal argument of nominalized intransitive verb is treated like the lowest internal argument of a nominalized transitive/polyadic verb, whereas the agent argument must receive an oblique linker. The ‘ergative makeup’, however, is an epiphenomenon, resulting from the fact that always the lowestranked argument gets preference in structural nominal linking. The two other linking types do not have a homogeneous nominal linking domain. Their linking domain is split up into two subdomains. The referential argument is always part of the nominal domain [ ]N. The difference between the verbal type and the ACC-POSS type lies in the extension of the subdomains. In the ACC-POSS type the highest and next-to-highest argument ———–———————————————————————————— 17

There are various patterns regarding the use of the alienable and inalienable linker in nominals. KT (pp. 210ff.) discusses some patterns. Tongan, for instance, shows a GEN-ERG pattern as well as a NOM-OBL pattern in the nominals based on transitive verbs. GEN is used as inalienable linker, NOM as alienable linker. Due to the lack of systematic studies, I don’t want to discuss this topic any further.

212 Barbara Stiebels are in the nominal subdomain; all other arguments are in the verbal subdomain. In the subdomains category-specific linkers are used. The nominal subdomain is maximal, integrating as many arguments as can be realized structurally by nominal linkers.18 (71) ACC-POSS type z y x +hr +hr +hr [[–lr +lr ] V [ ACC DAT

s [ACT(x) & BECOME(POSS(y,z))](s) –hr ] N] POSS D

In the verbal type, the nominal subdomain is not maximal; actually, no nominal linker is used. Only the highest argument is bound by some determiner-like element. (72) Verbal type z y +hr +hr [[–lr +lr NOM DAT

x s [ACT(x) & BECOME(POSS(y,z))](s) +hr –hr ] N] +lr ] V [ ERG D

That a nominal linker is completely avoided in the verbal type is astonishing in view of the fact that a structural nominal linker is available; at least one argument could be linked in a nominal subdomain. A possible explanation for this pattern might be that a language of this kind avoids inhomogeneous linking, i.e. the use of a mixed verbal-nominal linking pattern. Before turning to the problem of the subdomains in section 5.3, I would like to focus on another aspect of the order of argument saturation in nominals. The comparison of the four types shows that, unless GEN is doubled as in the nominal type, only the lowest argument of the nominal domain is marked by GEN. In order to mark another argument by means of this linker, the argument that would be targeted by GEN must receive an alternative (oblique) linker. In this case, the structural linking resources of the nominal are not exhausted so that GEN can be assigned to some other argument. Such shifts of linking patterns occur in the interaction with voice, i.e. with passive and antipassive. ———–———————————————————————————— 18

The feature [lr] is only instantiated if it is lexically assigned or if the argument is part of a verbal subdomain.

From rags to riches. Nominal linking in contrast to verbal linking

213

In languages of the POSS-OBL type, demotion of the lowest argument renders another argument accessible for the structural linker; antipassive is the operation that does have this effect. Let me illustrate this with examples from West Greenlandic. In West Greenlandic a lexical possessor is structurally marked by ergative (like the Yupik pattern in (37)). (73a) shows the nominalization of an intransitive verb. The inherited argument is additionally indexed by 3SG agreement. With the nominalization of a transitive verb, the lowest argument receives the ergative marker as in (73b). If the lowest argument is demoted by means of antipassive, it receives an oblique linker (instrumental) as in (73c). Consequently, the structural linker is available for the agent argument, which, in addition, is indexed by the agreement marker. (73) West Greenlandic: event nominals (Fortescue 1984: 212/46/213) a. Urtu-p ajurnaar-nir-a Ortu-ERG have.accident-NOML-3SG ‘Ortu’s accident’ b. anguti-p tuqun-nir-a man-ERG kill-NOML-3SG ‘the killing of the man’ [= patient] c. piniartu-p tiriannia-mik aallaa-nnin-nir-a hunter-ERG fox-INSTR shoot-ANTIPASS-NOML-3SG ‘the hunter’s shooting of a/the fox’ (74) shows the change in the potential linking patterns. With the nominalization of the base form, represented in (74a), the structural linker, which can only be used once, is reserved for the lowest-ranked argument; the feature [+lr] is assigned according to the nominal possessor linking mechanism. In (74b) the lowest argument can no longer be targeted by the structural linker, therefore, the next higher argument receives the structural linker. (74) West Greenlandic: POSS-OBL type a. y x s KILL(y,z) (s) +hr +hr –hr [+lr ]N ERG – D

214 Barbara Stiebels b. Antipassive
y x +hr +hr +obl [ +lr INSTR ERG

s –hr

kill(y,z) (s)

]N D

Note that (74b) does not represent a new type of nominalization (namely OBL-POSS). The oblique linking is a result of antipassivization. Whereas the pattern in West-Greenlandic can be characterized as an upward shift of nominal structural linking, the mirror image, the downward shift of nominal structural linking, occurs in languages of the ACC-POSS type. The higher argument must receive an alternative (oblique) linker so that a lower argument may be linked with the nominal structural linker; this linking pattern is triggered by passive; Turkish nominals provide a case in question. (75a) shows the base pattern: the lowest argument receives a verbal linker, the higher argument a nominal linker. In (75b) the agent argument is demoted to oblique realization (PP) by means of passive; as a consequence, the next lower argument may receive genitive. (75) Turkish: nominalization of passive verb (Comrie & Thompson 1985: 364–65) a. Hasan-ın mektub-u yaz-ma-sı Hasan-GEN letter-ACC write-NOML-3SG.P ‘Hasan’s writing of the letter’ b. mektub-un [Hasan-ın tarafından] yaz-ıl-ma-sı letter-GEN Hasan-GEN by write-PASS-NOML-3SG.P ‘the letter’s being written by Hasan’ The shift of the nominal domain is shown in (76). (76) Turkish: ACC-POSS type a. y x s WRITE(y,z) (s) +hr +hr –hr ] N] [[ ]V [ ACC GEN D

From rags to riches. Nominal linking in contrast to verbal linking

b. Passive y s
x +hr –hr [ ]N GEN D OBL

WRITE(y,z)

215

(s)

Strict successive argument saturation explains the major types of linking patterns in nominalizations. Are there any languages or structures that violate this constraint? There is evidence that this can occur in certain circumstances. One case in question is the formation of present participles in German. These participles take the external argument of the verb as referential argument (der Lesen-de ‘the reader’); the remaining arguments are realized verbally. Some participle formations now yield a conflict, namely those in which the highest argument is marked dative (e.g. with gehören ‘have, belong’); this is a consequence of its lexical marking as [+hr]: (77) German: verbs with dative-marked highest argument gehören y x s POSS(x,y)(s) +hr –lr +lr NOM DAT

If the highest argument were chosen as referential argument, the lexical marking would become invisible. If the non-highest argument were chosen as referential argument, argument swapping would have to take place in order to retain the visibility of the lexical case and to provide a referential argument. As the examples in (78) show, the option of argument swapping is chosen; the present participle refers to the lowest verbal argument. (78) a. der den

Kinder-n Gefallen-de DET DET.DAT.PL child.PL-DAT like-PTC.PRES ‘that which the children like’ b. das den Kinder-n Gehören-de DET DET.DAT.PL child.PL-DAT belong-PTC.PRES ‘that which belongs to the children’

Ehrich & Rapp (2000) observe that the lowest-ranked argument gets preference in the structural linking of German event nominals, i.e. is realized by genitive. If, however, the verb does not contain a change of state predicate, a genitive DP may be interpreted as some non-lowest argument (die Befra-

216 Barbara Stiebels gung des Kanzlers ‘the questioning of the chancellor (= agent/patient)’), although this is not the preferred reading. These data show that a violation of strict successive argument saturation can also be triggered by the aspectual properties of the underlying verb and the referential properties of the resulting nominal. 5.3. The role of subdomains and the derivation of the four major types The assumption of subdomains immediately raises the question of how the generation of subdomains is restricted. Could there be a non-left-peripheral verbal subdomain within a nominal subdomain as in (79a), or could verbal and nominal subdomain be interchanged as in (79b)? (79) a.

y +hr [

b.

x s +hr –hr ]N [+lr] V

y x +hr +hr [ [ ]N[

s –hr ]V]

Regarding the distribution of subdomains, syntactic accounts (e.g. Borsley & Kornfilt 2000) seem to provide a natural account: they claim that a verb may be turned into a noun at various head positions of syntactic projections (e.g. V-, VP- or IP-nominalization): Therefore, everything below that position of category change must be realized verbally, everything above that position must be realized nominally. As a consequence, verbal and nominal domain cannot be interchanged, nor can there be an intertwining of verbal and nominal domain as in (79a). Malouf (2000) characterizes these syntactic approaches as accounts of ‘phrasal coherence’. However, there are structures that seem to contradict Phrasal coherence: Nominals in Arabic exhibit the structure in (80); here, the higher argument, i.e. the agent, which receives the structural nominal linker, is realized closer to the event nominal than the lower-ranked arguments, which receive a verbal linker. (80) Arabic: ACC- POSS type (Fassi-Fehri 1993: 247) salb-u zayd-in r-ajul-i maal-a-hu depriving-NOML Zayd-GEN DET-man-ACC money-ACC-3SG.P ‘Zayd’s depriving the man of his money’

From rags to riches. Nominal linking in contrast to verbal linking

217

In order to maintain a syntactic account, one must assume additional movements (e.g. movement of the nominal from some lower position). This may work in syntactic terms, but undermines the appeal for a syntactic analysis.19 In a lexical account, the exclusion of the structures in (79a/b) must refer to the mechanisms of semantic composition. The nominalizing morpheme combines with the embedded verb via Functional Composition; in this case, all arguments that are not yet saturated are inherited to the derived noun. If all arguments of the embedded verb are inherited, one obtains a simple nominal domain. If some arguments of the underlying verb are already saturated, they form a verbal subdomain. Under the regulations of strict successive argument saturation, only the lowest-ranked arguments can enter the verbal subdomain. The arguments that are not saturated at the point of the composition with the nominalizer enter the nominal subdomain. There is no further step in the composition (since there is no further functor) in which unsaturated arguments could enter another domain. Hence, the point of the composition at which the nominalizer is integrated determines the (maximally bipartite) segregration of linking domains, which is illustrated in (81) with the nominalization of a ditransitive verb in a language with an ACC-POSS system: It is assumed that in (81f) and (81e) the verb combines with its lowest arguments according to the verbal linking patterns. In (81d) it combines with the nominalizer; the arguments are inherited to the resulting complex. The two remaining arguments are linked nominally or bound by a determiner, respectively. If the nominalizer combined with the verb at stage (81f), all arguments would enter the nominal domain (= nominal type). If the nominalizer combined with the verb at stage (81b), all arguments besides the referential argument would be linked verbally (= verbal type). If the nominalizer combined with the verb at stage (81e), only the lowest argument would be linked verbally. However, given that the structural nominal linker can only be used once, the nominal linker resources might not suffice to link all remaining arguments. Under the assumption (i) that two functors (the verb and the nominalizer) are combined, (ii) that there are various stages at which this composition ———–———————————————————————————— 19

Syntactic accounts predict that verbal functional categories are retained in the nominal if the verb is nominalized in a high syntactic projection, e.g. above some functional projection. Moreover, these accounts have to exclude the stranding of modifiers in the lower projections by some additional mechanism given that these modifiers are often not allowed.

218 Barbara Stiebels can take place, (iii) and that argument swapping is strongly dispreferred, the four major types of nominal linking patterns emerge quite naturally. (81) Nominalization of a ditransitive verb in ACC-POSS-type a. e wo b. D es,t wo es,t,e ex,es,t c. XGEN wp NOML d. ex,es,t wo es,t,es,t ey,ex,es,t e. YDAT wo V f. ZACC ez,ey,ex,es,t There is still one issue that needs to be resolved: Under which circumstances do languages select one of the four options? I assume that this results from the interaction of the following constraints: (82) FA (strict Function Application): The arguments are saturated according to their position on the -grid (starting with the lowest argument). (83) *SUBDOMAIN: The -grid should not contain distinct linking domains. (84) UNIQUENESS: Each linker should only be used once in a given linking domain. (85) H-L (homogeneous linking): All arguments of a functor should be marked by linkers of a given lexical category (e.g. only nominal or only verbal linkers). (86) MAX(+hr): A [+hr] argument in the input should have a correspondent in the output, i.e. it must be marked by a linker that contains [+hr] in its lexical specification. FA is the constraint that regulates strict successive argument saturation. *SUBDOMAIN is the ban on the instantiation of (verbal) subdomains. UNIQUENESS is the ban on the doubling of (structural) linkers, e.g. the doubling of genitives. It is clear that languages of the nominal type (e.g. Tukang Besi) have a low-ranked UNIQUENESS constraint, whereas it is ranked rather

From rags to riches. Nominal linking in contrast to verbal linking

219

highly in all the other types. The constraint H-L, the requirement for homogeneous linking, motivates the verbal type: Although it is not necessary to link all internal arguments verbally – one argument could receive a nominal linker – this option is chosen in languages such as Basque. MAX(+hr) is one of the constraints that regulate structural linking. I have integrated it here as a constraint that favors structural linking over oblique linking. Ignoring the details of the concrete linking patterns, the four types result from the following constraint rankings; FA is assumed to be ranked rather high in general: (87) Constraint ranking of nominal linking types a. b. c. d.

Nominal type: *SUBDOMAIN » MAX(+HR) » UNIQUENESS, H-L verbal type: UNIQUENESS, H-L » *SUBDOMAIN POSS-OBL type: UNIQUENESS, *SUBDOMAIN » H-L ACC-POSS type: UNIQUENESS » *SUBDOMAIN, H-L

The nominal type emerges from a ban on subdomains and favoring nominal structural linking over avoidance of linker doubling. The verbal type emerges from a ban on linker doubling and the requirement for homogeneous linking. In the POSS-OBL type linker doubling is barred as well as the formation of subdomains. Finally, in the ACC-POSS type it is not allowed to double linkers; however, the formation of subdomains is possible. 5.4. Argument transfer Another strategy of circumventing the restrictions of nominal linking is the transfer of arguments to some other functor whose linking resources are not exhausted. Hungarian is an interesting case in question. Here, all arguments besides the one that receives structural dative or nominative are transferred to the present participle való of the verb van ‘be’ (Laczkó 1995, 2000). The participle functions as a nominal modifier and integrates the arguments of the derived nominal into the AP it heads: in (88a) it integrates a DP in the illative case, in (88b) a DP in the inessive case. The argument that is linked structurally by means of dative or nominative (Anna-nak/Janos) is realized outside the AP.

220 Barbara Stiebels (88) Hungarian: argument transfer (Kenesei et al. 1998: 207; Laczkó 1995: 112) a. Dénes-nek [Mártá-ba való]AP vele-szeret-és-e M-ILL be.PTC into-love-NOML-3SG.P D-DAT ‘Dénes’s falling in love with Martha’ b. Janos [London-ba való]AP meg-érkez-és-e be.PTC PV-arrive-NOML-3SG.P J.NOM L-INE ‘John’s arrival in London’ This type of argument transfer is rather unusual because arguments of a higher functor are transferred to a modifying head and not to a regular complement-taking head. In the general case (e.g. with verb incorporation, raising verbs or light verbs), argument transfer is directed upwards, from the embedded functor to the selecting functor, which I want to illustrate for light verb constructions in Japanese (Grimshaw & Mester 1988). The light verb suru (past: shita) inherits the arguments of the embedded event nominal and licenses verbal linking of these arguments, which may include structural linkers, oblique linkers (like kara as in (89b)) and complementizers as in (89c). The event nominal itself is marked by accusative; due to the ban on doubling accusatives (‘double-o constraint’), an argument of the event nominal cannot be realized by accusative.20 (89) Japanese: light verb construction (Grimshaw & Mester 1988: 207) a. John-wa Mary-ni hanashi-o shita J-TOP M-DAT talk-ACC LV.PAST ‘John talked to Mary’ b. John-wa Tookyoo-kara shuppatsu-o shita J-TOP T-from departure-ACC LV.PAST ‘John departed from Tokyo’ c. John-wa murabito-ni [ookami-ga kuru-to] keikoku-o shita J-TOP villager-DAT wolf-NOM come-COMP warn-ACC LV.PAST ‘John warned the villagers that the wolf was coming’ ———–———————————————————————————— 20

It is possible to realize such an argument as the topic of the clause (Grimshaw & Mester 1988: 216): i. [kaiketsu-no hookoo]-wa sono hookokusho-ga Mary-ni shisa-o shiteiru Mary-DAT suggestion-ACC LV solution-GEN direction-TOP that report-NOM ‘that reports suggests to Mary the direction of the solution’

From rags to riches. Nominal linking in contrast to verbal linking

221

If the arguments of the nominal are not transferred to the light verb suru (past tense: shita), they are marked by genitive (in addition to their inherent case), which is the structural nominal linker: (90) Japanese: argument realization in nominals (Grimshaw & Mester 1988: 207) a. John-no Mary-e-no hanashi J-GEN M-to-GEN talk ‘John’s talk to Mary’ b. John-no Tookyoo-kara-no shuppatsu departure J-GEN T-from-GEN ‘John’s departure from Tokyo’ c. John-no murabito-e-no [ookami-ga kuru-to]-no keikoku J-GEN villager-to-GEN wolf-NOM come-COMP-GEN warning ‘John’s warning to the villagers that the wolf is coming’ As (90) shows, argument transfer is not compulsory in Japanese. The arguments may stay within the NP/DP projected by the event nominal. However, argument transfer allows the arguments to enter verbal projections of information structure (e.g. being realized as topic). According to Grimshaw & Mester at least two arguments of the event nominal must be transferred to the light verb, namely the highest argument of the nominal (usually the ‘subject’) and some other internal argument of the nominal. A complete transfer of all arguments to the light verb is also possible. If not all arguments are transferred, the transfer must obey the argument hierarchy: It is not possible to transfer a lower-ranked argument if some higher-ranked argument is not transferred. The argument transfer can be analyzed as Functional Composition of suru and the embedded event nominal. The light verb takes a predicative argument (the event nominal) and designates it as an argument that receives a structural [+hr] linker (accusative). In the course of the composition of light verb and event nominal, the event nominal is marked by accusative. The arguments that are not saturated are inherited to the complex as in (91c):
P
u
s [DO(u) & P(u)](s); +hr eu,es,t,eu,es,t b. hanashi
y
x
s [TALK(x) ey,ex,es,t & DIRECT_TOWARDS(x,y)](s); c. hanashi-o suru
y
u
s [DO(u) & TALK(u) ey,ex,es,t & DIRECT_TOWARDS(u,y)](s);

(91) a. suru

222 Barbara Stiebels The Hungarian transfer pattern, i.e. the composition of való with the event nominal, is different because való is a nominal modifier. I do not want to work out a complete analysis for való in all its intricacies, but briefly sketch a tentative analysis. In the Montagovian tradition, modifiers are analyzed as functors of type X/X (with X ranging over possible simple/complex types/ categories); hence, they do not change the syntactic category or semantic type of its argument (the noun/nominal in this case).21 való is different in that it selects an unsaturated nominal functor and saturates some of the functor’s arguments in its own domain. Therefore, I propose to analyze való as a more complex functor. Its semantic type is X,X,Y,Y, i.e. it selects an element (X) that is selected by some other functor (X,Y) and discharges it (yielding something of type Y). The general structure is shown in (92a). X and Y range over possible simple and complex types. (92b) shows the analysis for a nominalized intransitive verb that has a predicative argument (e.g. (88b)). (92) a.

b.

… p NPY qp NX,Y APX,Y,Y qp valóX,X,Y,Y ZPX NPe,t qp Ne,e,t NPe qp Ne,t,e,e,t APe,t,e,e,t,e,e,t qp valóe,t,e,t,e,e,t,e,e,t PPe,t

The predicative argument (a PP) is realized within the való-phrase. The agent argument is realized by means of the nominal structural linker; the referential argument is bound outside of the NP. ———–———————————————————————————— 21

There is no consensus that all modifiers should be analyzed as elements of type X/X. Especially, intersective modifiers have been analyzed differently (see Wunderlich 1997b).

From rags to riches. Nominal linking in contrast to verbal linking

223

6. Information structure The preceding sections have shown that nouns show a richness of linking strategies that resembles that of verbs. If one takes into account that the argument structure of nouns does not exhibit the complexity found in verbs, the richness is remarkable. However, there is an important difference between nouns and verbs in terms of information structure. Nouns do not instantiate a projection of information structure independent of the verbal head that selects the noun/NP/DP. If an argument of a noun/NP/DP is topicalized or focused and if the language has a designated position for these discourse functions, then this position will be one projected from a verbal head (e.g. a sentence-initial topic position and an immediately preverbal/ postverbal focus position). This follows from three conditions: (i) Nouns/ NPs/DPs are natural arguments of verbs, not vice versa. (ii) Verbs, and not nouns, project clauses (with the exception of certain nominal predicates in the default tense). (iii) Information structure is a property of clauses and not attributed to its parts. Therefore, a focused or topicalized possessor argument has to leave the NP/DP and target the respective verbal projection of information structure. An alternative to syntactic movement would be a specific morphological marking. This is illustrated with examples from Hungarian and Mam. If a possessor NP/DP is focused in Hungarian (indicated by the subscript F), it must be moved to the preverbal focus position, either piedpiping the whole DP/NP as in (93a), or stranding the rest of the DP/NP as in (93b). The same is true for a possessed nominal head. Note that only dative-marked possessors may be extracted in Hungarian; possessors in the nominative have to stay within the DP. (93) Hungarian: possessor focus (Kenesei et al. 1998: 170) könyv-é-t] olvas-ta el a. Anna [Péter-nekF a DET book-3SG.P-ACC read-PAST.3SG PV A P-DAT ‘it’s Peter’s book that Anna has read’ b. Anna Péter-nekF olvas-ta el [a könyv-é-t] read-PAST.3SG PV DET book-3SG.P-ACC A P-DAT ‘it’s Peter’s book that Anna has read’ Mam differs from Hungarian in that possessor focus does not involve movement, whereas the topicalization of contrastive possessors does (topicalization is indicated by the subscript T). Focusing of verbal arguments in

224 Barbara Stiebels Mam requires the movement of the focused NP/DP to the immediate preverbal position, only separated from the verb by directional or aspectual/ temporal markers (see subject focus in (94b) and object focus in (94c)). Focusing of transitive subjects either requires either an oblique marking in addition or the use of a special verbal ‘agent focus’ marker (see also Stiebels 2006). (94) Mam: argument focus (England 1983: 225–26) a. ma kub’ ky-tzyu- n xiinaq cheej REC.PAST DIR 3PL.E-grab-DIR man horse ‘the men grabbed the horse’ [VSO] b. [ky-u n xiinaq]F ma kub’ ky-tzyu- n cheej REC.PAST DIR 3PL.E-grab-DIR horse 3PL.E-REL man ‘the men grabbed the horse’ [SFVO] [‘it were the men who grabbed the horse’] c. [aa cheej]F ma kub’ ky-tzyu- n xiinaq DEM horse REC.PAST DIR 3PL.E-grab-DIR man ‘the men grabbed the horse’ [OFVS] [‘it was the horse that was grabbed by the men’] Focused or topicalized possessors differ from non-focused or non-topicalized possessors in Mam: they are marked twice on the head noun (if they are pronominal). In (95a), the possessor marker is prefixed doubly, which happens with focused possessors; in (95b) the possessor is marked by a prefix and a suffix, which happens with topicalized contrastive possessors.22 Whereas the simply focused possessor is not moved into the preverbal focus position, the topicalized contrastive possessor is moved including its head to the sentence-initial (topic) position. (95) Mam: possessor focus/topicalization (England 1983: 144) a. ma kub’ tiil-j w-nF-jaa REC.PAST DIR knock.down-PASS 1SG.E-1SG.E-house ‘MY house was knocked down’ [… not your’s …]

———–————————————————————————————

22

According to England, the double marking of the possessor by prefix and suffix is used for possessors that are topicalized and focused. I assume that this characterization corresponds to that of contrastive topics.

From rags to riches. Nominal linking in contrast to verbal linking

225

ma kub’ tiil-j b. nF-jaa-waT 1SG.E-house-1SG.E REC.PAST DIR knock.down-PASS ‘it was my house that was knocked down’ ‘regarding MY house, it was knocked down’ [B.S.] It seems as though the morphological marking of the focused possessor in Mam suffices to indicate its focus function, whereas topicalization requires movement in any case. Argument transfer as found in possessor raising (or as discussed with the Japanese light verb construction) can also be seen as means to enable underlying nominal arguments to enter verbal projections of information structure. 7. Conclusions I have started from the contrast of the impoverished argument structure of underived nouns (the ‘rags’) and the potential rich argument structure of event nominals (the ‘riches’), which pose a serious challenge to the linking resources of nouns/nominals. As I have tried to show, one cannot maintain the assumption that because the argument structure of simple nouns is impoverished, the linking resources will likewise be impoverished/insufficient. It is true that the nominal linker inventory is smaller than the verbal linking inventory in most languages, especially in languages with a linking system of the generalized case type. However, there are far more parallels between nouns and verbs as one might expect from a superficial perspective. Apart from passive one can find nominal equivalents for voices and diathesis operations such as applicative, antipassive, and causative. There is actually almost no verbal linking type that does not have an equivalent with nouns: the active type, the inverse type, the portmanteau type, and the incorporational type. Even the West-Austronesian voice system is used with nouns, bearing some overlap with the verbal voice function in terms of the affected argument. Following Ross (2002) one might assume that this voice system emerged from structures of nominalization. When the nominalizing affixes were reinterpreted as verbal affixes, they developed a new function as markers of syntactic prominence of the relevant argument. The discussion of the linking patterns of event nominals has shown that languages use various strategies to compensate for their lack of elaborated linking inventories in nouns. Only languages of the nominal type extend the

226 Barbara Stiebels nominal linking resources to structures that are more complex than those of underived nouns (being at most 2-place predicates) by doubling the structural linker. The integration of an oblique linker in the POSS-OBL type makes use of the non-structural linking resources of nouns. The other two types (verbal type and ACC-POSS type) increase the categorial complexity of the derived noun by the use of verbal subdomains. A further option of dealing with the complexity of event nominals is the transfer of arguments to another functor. I have highlighted one specific property of nouns, namely their strict successive argument saturation. Often one cannot find similar restrictions in verbs. The verbal option of argument swapping (used, for instance, in scrambling) may be related to another specific properties of verbs: their potential to project additional structures of information structure. Argument swapping often serves to align arguments (and adjuncts) according to their information-structural status. Nouns cannot head specific nominal projections of information structure. Therefore, focused or topicalized arguments of nouns have to target the respective verbal projection. In terms of the two metaphors of ‘rags’ and ‘riches’ one might say that nouns are not bound to their inherent stage of ‘rags’; they have the potential to deal with the ‘riches’ of the complex heritage in event nominals, but they also act in ‘riches’ regarding the variety of linking systems a language may instantiate for nouns/nominals.

Acknowledgements I would like to thank Manfred Bierwisch and Dieter Wunderlich for many helpful remarks. The research was funded by the German Science Foundation (project grant within the SFB 282 ‘Theory of the lexicon’ and Heisenberg grant STI 151/2-1).

From rags to riches. Nominal linking in contrast to verbal linking

Abbreviations  = 1, 2, 3

 A

ABL ABS ACC ACT ADV ALL ANTIPASS ANTIPOSS APPL APPL.BEN APPL.INST ASP AUX AV CAUS COMP COMPL CONN CORE D DAT DET DIR DIRV E EP ERG FUT GEN HAB ILL IMPF

exclusive OR clitic boundary first, second, third person infix reduplication ACC-like agreement ablative absolutus accusative active agreement adverb allative antipassive antipossessive applicative benefactive instrumental aspect auxiliary actor voice causative complementizer completive aspect connector core case DAT-like agreement dative case determiner directional marker direct verb form ERG-like agreement epenthetic material ergative future tense genitive habitual illative case imperfective

INACT INCOMPL IND INE INSTR INV IV IRR ITR LV MASC MPASS N NOM NOML OBV OV P PART PASS PAST PERF PL POREX PRES PROX PTC PV Q REC.PAST REFL REL SG TOP TR USP.H USP.NH

inactive agreement incompletive aspect indicative inessive case instrumental case inverse verb form instrumental voice irrealis intransitive light verb masculine mediopassive NOM-like agreement nominative nominalization obviative object voice possessor agreement particle passive past tense perfective plural possessor extension present tense proximate present participle preverb question marker recent past reflexive relational noun singular topic marker transitive unspecific human unspecific non-human

227

228 Barbara Stiebels References Aissen, Judith 1999 Markedness and subject choice in optimality theory. Natural Language and Linguistic Theory 17: 673–711 2003 Differential object marking: Iconicity vs. economy. Natural Language and Linguistic Theory 21: 435–483. Alexiadou, Artemis 2001 Functional Structure in Nominals: Nominalization and Ergativity. Amsterdam: Benjamins. Amith, Jonathan D. and Thomas C. Smith-Stark 1994a Transitive nouns and split possessive paradigms in Central Guerrero Nahuatl. International Journal of American Linguistics 60: 342–368. 1994b Predicate nominal and transitive verbal expressions of interpersonal relations. Linguistics 32: 511–547. Anderson, Gregory David and K. David Harrison 1999 Tyvan. München: Lincom Europa. Andrews, Richard J. 2003 Introduction to Classical Nahuatl. 2nd edition. Norman: University of Oklahoma Press. Baker, Mark 1988 Incorporation: a theory of grammatical function changing. University of Chicago Press. Barker, Chris 1995 Possessive descriptions. Stanford: CSLI publications. Bierwisch, Manfred 1989 Event nominalizations: Proposals and problems. Linguistische Studien 194: 1–73. Berlin: Akademie der Wissenschaften der DDR. Borsley, Robert D. and Jaklin Kornfilt 2000 Mixed extended projections. In The Nature and Function of Syntactic Categories, R. D. Borsley (ed.), 101–132. New York: Academic Press. Bresnan, Joan 2001 Lexical-functional Syntax. Oxford: Blackwell. Bricker, Victoria Reifler 1978 Antipassive constructions in Yucatec Maya. In Papers in Mayan Linguistics, N. C. England (ed.), 3–24. Columbia: University of Missouri. Broschart, Jürgen 1997 Why Tongan does it differently: Categorial distinctions in a language without nouns and verbs. Linguistic Typology 1: 123–65. Campbell, Lyle 2000 Valency-changing derivations in K'iche'. In Dixon and Aikhenvald (eds.), 236–281.

From rags to riches. Nominal linking in contrast to verbal linking

229

Chomsky, Noam 1970 Remarks on nominalization. In Readings in English Transformational Grammar, R. A. Jacobs and P. S. Rosenbaum (eds.), 184–221. Waltham, MA: Blaisdell. Comrie, Bernard 1976 The syntax of action nominals: a cross-language study. Lingua 40: 177–201 1985 Causative verb formation and other verb-deriving morphology. In Language Typology and Syntactic Description, Vol. III: Grammatical Categories and the Lexicon, T. Shopen (ed.), 309–348. Cambridge University Press. 1989 Language Universals and Linguistic Typology. 2nd edition. Oxford: Blackwell. Comrie, Bernard and Sandra A. Thompson 1985 Lexical nominalization. In Language Typology and Syntactic Description, Vol. III: Grammatical Categories and the Lexicon, T. Shopen (ed.), 349–398. Cambridge University Press. Craig, Colette G. 1978 On the promotion of instrumentals in Mayan languages. In Papers in Mayan Linguistics, N. C. England (ed.), 55–68. Columbia: University of Missouri. Croft, William 1991 Syntactic Categories and Grammatical Relations: The Cognitive Organization of Information. The University of Chicago Press. Dixon, R. M. W. 1994 Ergativity. Cambridge: Cambridge University Press. 2000 A typology of causatives: form, syntax and meaning. In Changing Valency. Case Studies in Transitivity, R. M. W. Dixon and A. Aikhenvald (eds.), 30–83. Cambridge University Press. Dixon, R. M. W. and Alexandra Aikhenvald (eds.) 2000 Changing Valency. Case Studies in Transitivity. Cambridge University Press. Donohue, Mark 1999 A Grammar of Tukang Besi. Berlin /New York: Mouton de Gruyter. Edmonson, Barbara W. 1988 A descriptive grammar of Huastec (Potosino dialect). PhD diss., Tulane University. Ehrich, Veronika and Irene Rapp 2000 Sortale Bedeutung und Argumentstruktur: ung-Nominalisierungen im Deutschen. Zeitschrift für Sprachwissenschaft 19: 245–303. Eijk, Jan P. van and Thom Hess 1986 Noun and verb in Salish. Lingua 69: 319–331.

230 Barbara Stiebels England, Nora C. 1983 A Grammar of Mam, a Mayan Language. Austin: University of Texas Press. Fassi Fehri, Abdelkader 1993 Issues in the Structure of Arabic Clauses and Words. Dordrecht: Kluwer. Fortescue, Michael 1984 West Greenlandic. London: Croom Helm. Givón, Talmy 1984 Syntax: A Functional-typological Introduction, Vol. I. Amsterdam: Benjamins. Grimshaw, Jane 1990 Argument Structure. Cambridge, MA: The MIT Press. Grimshaw, Jane and Armin Mester 1988 Light verbs and -marking. Linguistic Inquiry 19: 205–232. Hockett, Charles 1948 Potawatomi. International Journal of American Linguistics 14: 1–10, 63–73, 139–49, 213–25. Hofling, Charles Andrew 2000 Itzaj Maya Grammar. Salt Lake City: University of Utah Press. Hopper, Paul J. and Sandra A. Thompson 1980 Transitivity in grammar and discourse. Language 56: 251–99. Huang, Lillian M. 2002 Nominalization in Mayrinax Atayal. Language and Linguistics 3 (2): 197–224. Academia Sinica, Taiwan. Jelinek, Eloise and Richard A. Demers 1994 Predicates and pronominal arguments in Straits Salish. Language 70: 697–736. Joppen, Sandra and Dieter Wunderlich 1995 Argument linking in Basque. Lingua 97: 123–169. Kaufmann, Ingrid 1995 Konzeptuelle Grundlagen semantischer Dekompositionsstrukturen: Die Kombinatorik lokaler Verben und prädikativer Komplemente. Tübingen: Niemeyer. Kaufmann, Ingrid and Dieter Wunderlich 1998 Cross-linguistic patterns of resultatives. Working papers of the SFB 282, ‘Theory of the lexicon’, 109. University of Düsseldorf. Keenan, Edward L. and Maria Polinsky 1998 Malagasy (Austronesian). In The Handbook of Morphology, A. Spencer and A. M. Zwicky (eds.), 563–623. Oxford: Blackwell. Kenesei, István, Robert M. Vago and Anna Fenyvesi 1998 Hungarian. London: Routledge.

From rags to riches. Nominal linking in contrast to verbal linking

231

Kerke, Simon van de 1996 Affix order and interpretation in Bolivian Quechua. PhD diss., University of Amsterdam. Kimball, Geoffrey D. 1991 Koasati Grammar. Lincoln: University of Nebraska Press. Kimenyi, Alexandre 1980 A Relational Grammar of Kinyarwanda. Berkeley: University of California Press. Kinkade, M. Dale 1983 Salish evidence agains the universality of ‘noun’ and ‘verb’. Lingua 60: 25–40. Kiparsky, Paul 1997 Remarks on denominal verbs. In Complex Predicates, A. Alsina, J. Bresnan and P. Sells (eds.), 473–499. Stanford: CSLI Publications. Koptjevskaja-Tamm, Maria 1993 Nominalizations. London: Routledge. Krämer, Martin and Dieter Wunderlich 1999 Transitivity alternations in Yucatec, and the corrrelation between aspect and argument roles. Linguistics 37: 431–479. Kroeber, Paul D. 1999 The Salish Language Family: Reconstructing Syntax. Lincoln: University of Nebraska Press. Kroeger, Paul 1993 Phrase Structure and Grammatical Relations in Tagalog. Stanford: CSLI publications. Kroskrity, Paul 1985 A holistic understanding of Arizona Tewa passives. Language 61: 306–28. Kuipers, Aert H. 1974 The Shuswap Language. The Hague: Mouton. Laczkó, Tibor 1995 The Syntax of Hungarian Noun Phrases: A Lexical-functional Approach. Frankfurt am Main: Peter Lang. 2000 Derived nominals, possessors, and lexical mapping theory. In Argument Realization, M. Butt and T. H. King (eds.), 189–227. Stanford: CSLI Publications. Lehmann, Christian 1998 Possession in Yucatec Maya. Unterschleissheim: Lincom Europa. Malchukov, Andrej L. 2004 Nominalization /Verbalization: Constraining a Typology of Transcategorial Operations. München: Lincom Europa.

232 Barbara Stiebels Malouf, Robert P. 2000 Mixed Categories in the Hierarchical Lexicon. Stanford: CSLI Publications. Martin, Jack. B. 2000 Creek voice: beyond valency. In Changing Valency. Case Studies in Transitivity, R. M. W. Dixon and A. Aikhenvald (eds.), 375–403. Cambridge University Press. Mattissen, Johanna 2003 Dependent-head-synthesis in Nivkh: A Contribution to a Typology of Polysynthesis. Amsterdam: Benjamins. this vol. Ontology and diachrony of polysynthesis. Merlan, Francesca 1985 Split intransitivity: functional oppositions in intransitive inflection. In Grammar Inside and Outside the Clause: Some Approaches to Theory from the Field, J. Nichols & A. C. Woodbury (eds.), 324– 362. Cambridge University Press. Mithun, Marianne 1991 Active/agentive case marking and its motivations. Language 67: 510–546. 1999a Noun and verb in Iroquoian languages: Multicategorisation from multiple criteria. In An Anthology of Word Classes, P. M. Vogel and B. Comrie (eds.), 379–420. Berlin /New York: Mouton de Gruyter. 1999b The Languages of Native North America. Cambridge University Press. Mohanan, Tara 1994 Argument Structure in Hindi. Stanford: CSLI publications. Ortmann, Albert 1994 Possessorkongruenz: Eine Fallstudie zum Verhältnis von Semantik, Morphologie und Syntax. Master’s thesis, University of Düsseldorf. Partee, Barbara H. and Vladimir Borschev 2003 Genitives, relational nouns, and argument-modifier ambiguity. In Modifying Adjuncts, E. Lang, C. Maienborn and C. Fabricius-Hansen (eds.), 67–112. Berlin /New York: Mouton de Gruyter. Primus, Beatrice 1999 Cases and Thematic Roles – Ergative, Accusative and Active. Tübingen: Niemeyer. Ross, Malcom 2002 The history and transitivity of western Austronesian voice and voicemarking. In The History and Typology of Western Austronesian Voice Systems, F. Wouk and M. Ross (eds.), 17–62. The Australian National University: Pacific Linguistics. Saltorelli, Mario 1988 Basque. London: Croom Helm.

From rags to riches. Nominal linking in contrast to verbal linking

233

Sasse, Hans-Jürgen 1993 Das Nomen – eine universelle Kategorie? Sprachtypologie und Universalienforschung 46: 187–221. Schachter, Paul and FeT. Otanes 1972 Tagalog Reference Grammar. Berkeley: University of California Press. Silverstein, Michael 1976 Hierarchy of features and ergativity. In Grammatical Categories in Australian Languages, R. M. W. Dixon (ed.), 112–171. Australian Institute of Aboriginal Studies No. 22. New Jersey: Humanities Press Int. Stiebels, Barbara 1998 Complex denominal verbs in German and the morphology-semantics interface. In Yearbook of Morphology 1997, Geert Booij and Jaap van Marle (eds.), 265–302. Dordrecht: Kluwer. 1999 Noun-verb symmetries in Nahuatl nominalizations. Natural Language and Linguistic Theory 17: 783–836. 2000 Linker inventories, linking splits and lexical economy. In Lexicon in Focus, B. Stiebels and D. Wunderlich (eds.), 211–245. Berlin: Akademie Verlag. 2002 Typologie des Argumentlinkings: Ökonomie und Expressivität. Berlin: Akademie Verlag. 2003 Transparent, restricted and opaque affix orders. In Syntactic Structures and Morphological Information, U. Junghanns and L. Szucsich (eds.), 283–315. Berlin /New York: Mouton de Gruyter. 2006 Agent focus in Mayan languages. Natural Language and Linguistic Theory 24: 501–570. Van Valin, Robert D. Jr. 1993 A synopsis of role and reference grammar. In Advances in Role and Reference Grammar, R. D. Van Valin (ed.), 1–164. Amsterdam: Benjamins. Watkins, Laurel 1984 A Grammar of Kiowa. Lincoln: University of Nebraska Press. Wolfart, H. Christoph 1973 Plains Cree: A grammatical study. Transactions of the American Philosophical Society 63. Wunderlich, Dieter 1987 An investigation of lexical composition: the case of German be-verbs. Linguistics 25: 283–331. 1993a Diathesen. In Syntax. Ein Internationales Handbuch zeitgenössischer Forschung I, J. Jacobs, A. von Stechow, W. Sternefeld and T. Vennemann (eds.), 730–747. Berlin / New York: Mouton de Gruyter.

234 Barbara Stiebels 1993b

Funktionale Kategorien im Lexikon. In. Theorie und Praxis des Lexikons, F. Beckmann and G. Heyer (eds.), 54–73. Berlin /New York: Mouton de Gruyter. 1996 Lexical categories. Theoretical Linguistics 22: 1–48. 1997a Cause and the structure of verbs. Linguistic Inquiry 28: 27–68. 1997b Argument extension by lexical adjunction. Journal of Semantics 14: 95–142. 2000 Predicate composition and argument extension as general options – a study in the interface of semantic and conceptual structure. In Lexicon in Focus, B. Stiebels and D. Wunderlich (eds.), 247–270. Berlin: Akademie Verlag. 2002 Argumentlinking als strukturelle Generalisierung (Inkorporationstyp Nivkh). In Sprache als Form. Festschrift für Utz Maas zum 60. Geburtstag, M. Bommes, C. Noack and D. Tophinke (eds.), 11–21. Wiesbaden: Westdeutscher Verlag. this vol. Towards a structural typology of verb classes. Zavala Maldonado, Roberto 1994 Inverse alignment in Huastec. Funcion 15–16: 27–81.

A theory of Lexical Event Structures and its cognitive motivation Stefan Engelberg

1. Introduction The rise of event semantics in many domains of formal semantics like tense, aspect, quantification, adverbial modification, etc. has had an obvious impact on the semantic translation of verbs: verbs are represented as having a referential event argument in addition to their thematic arguments. This, in turn, forces lexical semantic theories to recur to this event argument in their semantic representations of verbs. Among the various lexical semantic approaches, the ones that assume that event structures constitute the nucleus of a verb’s meaning probably deviate the most from earlier approaches to verb meaning based on theta roles or decompositions. The development of such a theory of Lexical Event Structures as meaning representations for verbs is the topic of this paper. This venture is motivated by the demand for an adequate lexical counterpiece to sentential event semantics. It aims at developing a theory with a greater breadth of empirical coverage than is achieved by theories mainly concerned with linking. Furthermore, this undertaking is guided by the conviction that lexical semantics needs a more solid cognitive foundation to yield empirically contentful semantic representations. Thus, the questions to be answered in this paper are: how do we conceptualize events and how is this reflected in the grammatical and semantic behavior of the verbs denoting these events? The paper will give a concise account of the theory of Lexical Event Structures. The paper has three objectives which correspond to the following three sections. In section 2 I will sketch the theory and discuss the empirical goals the theory pursues (section 2.1) and the semantic components Lexical Event Structures consist of (section 2.2). Section 3 is devoted to linguistic phenomena whose explanation depends on Lexical Event Structures. In section 3.1 I will briefly illustrate in how far Lexical Event Structures are related to phenomena from five central empirical domains of lexical semantics and in section 3.2 it will be shown how Lexical Event Structures function

236 Stefan Engelberg in a linking theory. Section 4 aims to show how the central semantic concepts in Lexical Event Structures can be anchored to concepts which are well-founded in cognitive science. Section 4.1 discusses the event concept employed and illustrates the relation between the perception of movements and the use of verbs of movement. Section 4.2 deals with the concept of volition with respect to the licensing conditions for intransitive verb passives. In section 4.3 the distinction between durativity and punctuality, which has proven relevant for a number of verb semantic phenomena, is tied to the way we perceive events and structure our own actions. Section 5 provides a conclusion.1 2.

Lexical Event Structures

2.1. Empirical issues If one proposes a theory on how a verb’s meaning shall be represented, it seems worthwhile to spend a few thoughts on how such a theory and the particular meaning representations are empirically justified. Most lexical semantic theories that have been developed especially to capture verb meanings deal primarily with the question how a verb’s meaning determines the syntactic surrounding it occurs in. Starting from this particular research interest, it seems that it is often assumed that the main justification for the semantic representation proposed comes from syntactic data. This conception, I think, is misleading (cf. Engelberg 2000a, 2001). Let us assume that a linguistic expression LE is assigned the lexical semantic representation SEM as in (1a) and is claimed to occur with case patterns (or in syntactic configurations) as represented in the morphosyntactic representation SYN in (1b). Let’s further assume that a linking rule like (1c) is postulated.

———–————————————————————————————

1

This paper reports on one of the central topics of the research project “Valency in the Lexicon” (part of the project cluster SFB 282 “Theory of the Lexicon”), headed by Joachim Jacobs, and with Kerstin Blume, Ingrid Kaufmann, Ulrich Klein, and Barbara Lenz as collaborators. The project was also concerned with the processing of valency information, the lexical representation of non-obligatory arguments, marked syntactic valencies, and the valency of nominalized infinitives.

A theory of Lexical Event Structures and its cognitive motivation

(1)

237

a. SEM: PREDA(x, PREDB(y, z)) b. SYN: case1, case2, case3 (or [NP1 [NP2 NP3]] ) c. Linking: If the semantic representation of a lexeme is SEM, then y is assigned case2 (or, respectively, y enters the syntactic structure as NP2).

Now, if we want to check whether the statement in (1c) is true, we have to check whether (1a) is the case, whether (1b) is the case and whether the conditional in (1c) is borne out by the data. In order to check (1a) we indisputably have to know how SEM is to be interpreted. Only if this is sufficiently clear does a statement like (1c) have any empirical content. This is even more important if a theory is assumed that claims that statements like (1c) are based on a general homomorphism from lexical semantic onto syntactic structures. Theories of this kind are often particularly susceptible to circular argumentations in just imposing syntactic structures onto semantic ones.2 To assign a semantic representation SEM to a certain linguistic expression LE is to claim that LE corresponds to some configuration in the world we talk about. In this sense SEM is a description of the world on a certain level of abstraction. In order to check the correctness of an assignment of SEM to LE we at least have to check whether LE is indeed about the aspects of the world described in SEM and in order to do that we must of course know what SEM is supposed to convey about the world. The interpretation of SEM is based on the assignment of individual variables to entities and on truth conditions for the predicates, relations and operators used in SEM. On this basis, the meaning of complex expressions is computed from that of its parts and the way they are combined. While on the one hand those concepts used in SEM which are traditionally employed by sentential semantics – propositional relators, quantifiers, temporal operators, etc. – have been provided with fairly precise truth conditions, predicates like SEAFOOD, FLABBERGASTED or CELEBRATE are rather taken at their face value. Fortunately, we ———–————————————————————————————

2

It should be noted that while I assume that the structure of the entities verbs refer to greatly matters for lexical semantics I do not exploit any alleged homomorphic correspondence between the structural aspects of lexical meaning and syntactic structures. I wouldn’t categorically exclude that such structural correspondences exist although I doubt that it is as often the case as has been suggested in verb semantics. This paper is simply not about syntax-semantic interfaces of this sort. It neither requires such structural similarities nor is it incompatible with them.

238 Stefan Engelberg can often be quite generous with respect to what these predicates mean exactly, as long as it has no impact either on semantic composition or on the lexical-semantic licensing of syntactic or semantic phenomena. But this is of course different with respect to those “soft” lexical concepts like ‘agent’, ‘event’, ‘volition’, ‘causation’ which occur in lexical decompositions or meaning postulates and which on the one hand seem to elude the assignment of precise truth conditions, but on the other hand are used to make strong claims about the syntactic and semantic behavior of lexemes. Anyhow, it cannot be overlooked that lexical semantics has often shied away from giving a clear account of such basic semantic concepts. Thus, they remain quite shapeless, such that theories that employ these concepts are easily immunized against falsification. In other words, the empirical content of a lexical semantic theory is proportional to the degree of exactness with which the truth conditions for their basic predicates are given. One might of course object to this demand by claiming that it is so obvious who the agent and who the patient is in Mary hit John that any answer to the question what AGENT and PATIENT mean wouldn’t challenge our semantic representations anyways. But, firstly, to pursue its goals a theory is obligated not only to use intuitively clear concepts but also to explicate them. And secondly, there are numerous cases in which our ignorance of the meaning of semantic termini obviously obstructs progress in the field. Psych-verbs and the question in how far agentivity and causation are involved in their meaning (with numerous consequences for predictions about their syntactic behavior) constitute one out of many examples. As long as it is not sufficiently clear what CAUSE stands for, one might argue that in both fear and frighten causation is at stake (with the stimulus being the causer; e. g. Dowty 1991: 579) or that only frighten involves causation (e. g. Grimshaw 1990: 22) – whatever suits the theory and allows for the desired prediction about the mapping of arguments into syntax. What needs to be done to make progress in explicating our basic semantic concepts? I adhere to the not uncontroversial but quite unspectacular epistemological view that our experience and knowledge of the world is mediated by our cognitive predispositions. Under this moderate realistic view, the entities linguistic expressions are about are entities in this cognitively mediated world. If semantic representations are representations of the world we talk about, then we should find ample evidence for the nature of the entities and predicates employed in our semantic representations in the domain of those cognitive sciences which are devoted to research on the way we experience the world. In section 4 I will firstly show that this expectation is

A theory of Lexical Event Structures and its cognitive motivation

239

not unfounded and secondly that there is linguistic evidence that the world as cognitively mediated is indeed the world linguistic expressions are about. While a theory that does not provide truth conditions for the semantic predicates it employs has no empirical content, a theory that cannot account for the syntactic or semantic effects a verb has is linguistically insufficient. This conviction explains how the efforts in developing a theory on Lexical Event Structures have been distributed. Besides the search for a cognitive foundation for the basic semantic concepts, the aim of our research into verbs and events was to show that Lexical Event Structures provide the right conceptual grounding for the explanation of a wide range of syntactic and semantic phenomena. The empirical coverage of Lexical Event Structures encompasses phenomena in the following five domains, which I consider to be the central domains of lexical semantics (historical and acquisitional phenomena aside): (i) (ii)

(iii)

(iv) (v)

Semantics-syntax mapping: Semantic properties of verbs determine to a large degree the syntactic realization of arguments and the ability to take part in valency alternations, resultative constructions, etc. Selectional restrictions: The combination of a verb with other lexemes is subject to certain semantic restrictions. This concerns the co-occurrence of particular adverbials or derivational morphemes with certain classes of verbs on the one hand and verb-dependent restrictions on the NPs filling argument positions on the other. Grammatical-categorical restrictions: Verbs are semantically classified with respect to their ability to allow the expression of progressive aspect, imperative, particular voices or other contrasts of grammatical categories. Interlexematic relations: Verbs stand in semantic relations to each other, such as antonymy, hyponymy or synonymy. Inference behavior: Semantic properties of verbs influence the inference behaviour of lexical items in complex expressions. In particular, there are lexically-based inference peculiarities that show up in regular alternations of sentence patterns such as in diatheses, or when the tense or grammatical aspect of the sentence is changed. E. g., inferences about the internal temporal structure of an event expressed by the simple form of a verb might carry over to the progressive form of the verb or not, depending on particular lexical verb semantic properties.

240 Stefan Engelberg 2.2. Sketch of a theory of Lexical Event Structures The mereological structure of an event, its temporal properties, and the kind of involvement of participants in the event lie at the core of the theory that I will refer to as ‘Lexical Event Structure Theory’.3 Its basic idea is that the meaning of a verb is to be represented as a Lexical Event Structure (LES) which has the following characteristics: (i) (ii) (iii)

(iv)

(v)

Complexity of events: Verbs refer to events that are internally structured in the sense that they can consist of different subevents (e1, e2, …) and possibly a state (s). Sorts of subevents: The subevents are durative (eDUR) or punctual (ePCT). Relations between subevents: Subevents stand in temporal relations to each other, e. g., a subevent e1 can precede a subevent e2 completely (e1 < e2), or e1 can overlap with e2 (e1 o e2) (cf. for details Engelberg 2000a, 2004). Participation in subevents: The event participants which correspond to the arguments of the verb are not necessarily involved in all subevents, but rather only in some of them; semantic functions like CONTROL, MOVE, VOLITION etc. relate participants and subevents. Implication vs. presupposition: The occurrence of a subevent is either entailed ( I) or presupposed ( P) by the open proposition in the semantic translation of the verb, i.e. by an expression like “VERB(x,y,e)”.

This basic idea is illustrated by the following examples of the German verbs abtrocknen ‘dry off’, fahren ‘drive’, erschießen ‘kill by shooting’, and fangen ‘catch’. The two-place verb abtrocknen ‘to dry off’ refers to a complex event where the first subevent e1 has a certain duration (“DUR”), i. e., is not punctual, and involves two participants, an agent and a patient. In Ron hat den Bierkrug abgetrocknet ‘Ron dried off the beer mug’, e1 is Ron’s acting upon the beer mug (probably with a towel). Almost simultaneously (“o”), a second durative event e2 occurs which only involves the patient, namely the beer mug becoming dry. This results in a following (“ accusative/ergative > dative > genitive > other cases The degree of markedness of the syntactic valency increases with the number of these conditions that are violated. Thus, /nom/acc and /erg/abs are unmarked, /nom/dat and /abs/dat are slightly marked, /erg/dat is more marked (since both case specifications violate condition (ii) and /gen/gen is extremely marked (and probably not attested in natural languages) (Blume 2000: 221). The more marked a valency is, the more rarely it is found in natural languages, e. g. /abs/dat valencies are much more frequent than /erg/dat valencies. Starting from these assumptions, Blume (2000), investigating 11 languages from the Indo-European, Finno-Ugric and Austronesian language families, answers two questions: (i) How are marked syntactic valencies licensed? (ii) How are the arguments of marked and unmarked valencies linked to their respective cases? As far as the first question goes, the data suggest that crosslinguistically unmarked valency patterns can be found with verbs of any semantic class, whereas marked valencies can be found only with verbs that fulfill certain semantic conditions. At first sight, though, the group of verbs showing up with marked valencies seems rather heterogeneous. Looking only at German verbs with the marked /nom/dat valency, at least five groups emerge and are dealt with in Blume (2000): verbs of interaction, psych- and perception verbs, unaccusative verbs, verbs of possession, and certain kinds of stative verbs. Despite this heterogeneity, what Blume (2000: 180ff) identifies as the licensing condition for marked syntactic valencies of two-place

256 Stefan Engelberg verbs is limited semantic transitivity. That is to say, neither verbs that are transitive to a high degree, nor verbs with a particularly low degree of transitivity allow marked syntactic valencies. The notion of transitivity as well as the linking principle are based on a prototype approach to thematic roles in the vein of Dowty (1991). As we have seen in section 2.2, semantic relations connect subevents and event participants. In Blume’s (2000: 169f) approach, one set of these entailed relations characterizes a prototypical agent and another set a prototypical patient. The proto-agent properties are (i) CONTROL of the subevent, (ii) SENTIENCE with respect to the subevent, (iii) pursuing a particular interest or INTENTION with respect to a subevent, and (iv) exhibiting ACTIVITY or a particular function. Among these entailments ACTIVITY and the combination of CONTROL and INTENTION are considered to be particularly “potent” proto-agent properties. The proto-patient properties are (i) being CONTROLLED by another participant in a subevent, and (ii) being AFFECTED by the physical or mental impact of another participant. Participants which do not entail any proto-agent or proto-patient entailments with respect to a subevent are marked as THEMATIC. Blume’s concept of semantic transitivity is characterized by four properties (Blume 2000: 186). A verb is semantically transitive if (i) it has at least two distinct argument positions, (ii) at least one argument is a participant in the event, (iii) at least one argument has potent proto-role properties, and (iv) there is a clear asymmetry between the two arguments with respect to the distribution of proto-role properties. The more of these criteria apply, the more transitive the verb is. A verb is said to be of limited transitivity if the first three of these conditions are fulfilled but not the last one, i. e. there is no clear asymmetry in the distribution of proto-roles. This kind of limited semantic transitivity is what licenses marked syntactic valencies. A few examples will show how this condition cuts the right piece out of the verbal lexicon. Verbs like essen ‘eat’ fulfill all the transitivity conditions in having two arguments standing for event participants with potent protorole properties and a clear asymmetry between proto-agent properties for the subject and proto-patient properties for the object. A verb like sich verspäten ‘be late’ as a syntactically two-place verb with an expletive reflexive violates condition (i), a verb like kosten ‘cost’ where both arguments do not realize event participants violates condition (ii), and a verb like besitzen ‘own’ violates condition (iii) in not exhibiting any proto-role properties for one of its arguments. All of these verbs typically have unmarked valencies; in German they are /nom/acc-verbs. In the case of verbs like essen, this is due to high transitivity, in the other cases due to low transitivity.

A theory of Lexical Event Structures and its cognitive motivation

257

Three /nom/dat verbs will serve here to illustrate the case of limited transitivity, the perception verb schmecken ‘taste (good)’, the interaction verb folgen ‘follow’ and the two-place unaccusative zerbrechen ‘break’ (cf. also Blume 2000: 191ff): (15) a. Die Suppe schmeckte ihm. the soup.NOM tasted (good) he.DAT ‘The soup tasted good to him.’ b. Sie folgte dem Mann. she.NOM followed the man.DAT ‘She followed the man.’ c. Die Vase zerbrach ihm. the vase.NOM broke he.DAT ‘The vase (went and) broke on him. / He inadvertently caused the vase to break.’ The verb schmecken, which is event-structurally simple and has two event participants, implies SENTIENCE for the person doing the tasting, but exhibits no other proto-role properties. Thus, there is only a weak asymmetry of proto-role properties indicating limited transitivity (Lex 15).12 schmecken

SYN:

/nom/dat

SEM:


y
x
e[SCHMECK(x,y,e)]

LES:

( I e1[+DUR]: xSENTIENCE, yTHEMATIC)

Lex 15. Lexical entry for German schmecken ‘taste (good)’.

Verbs of interaction like folgen ‘follow’, antworten ‘answer’, nachgeben ‘give in’, trotzen ‘defy’, etc. are characterized by a presupposed subevent, in which one of the two participants acts (moves away, asks something, exerts pressure, etc.) and a second overlapping implied subevent in which the other participant acts. The distribution of proto-role properties is obviously not asymmetric while all the other transitivity conditions are fulfilled. Thus, the marked valency is licensed. ———–———————————————————————————— 12

The Lexical Event Structures of the verbs illustrated here slightly deviate from the structures given in Blume (2000). This does not affect her argumentation, though.

258 Stefan Engelberg folgen

SYN:

/dat/nom

SEM:


y
x
e[FOLG(x,y,e)]

LES:

( P e1[+DUR]: yACTIVITY/…) o ( I e2[+DUR]: xACTIVITY/…)

Lex 16. Lexical entry for German folgen ‘follow’.

Finally, the unaccusative two-place zerbrechen ‘break’ (and similarly for verbs like verrotten ‘rot’ or überkochen ‘boil over’) expresses that some object x breaks as a consequence of some presupposed event the other participant y was involved in and that the breaking of x is to the disadvantage of y. Since y neither intends nor controls the breaking of x and x exhibits none of the two proto-patient properties, there is only a slight asymmetry in proto-role properties, such that the marked valency is licensed. zerbrechen

SYN:

/nom/dat

SEM:


y
x
e[ZERBRECH(x,y,e)]

LES:

( P e1: xACTIVITY, yTHEMATIC) < ( I e2[+PCT]: yTHEMATIC) < ( I s: yTHEMATIC)

Lex 17. Lexical entry for the two-place variant of German zerbrechen ‘break’.

The licensing condition for marked valencies explains which verbs can occur with /nom/dat and other unmarked valency patterns. What it doesn’t tell us, though, is which argument in the semantic valency description has to be linked to which case in the syntactic valency description. This is accomplished by a two-part universal linking principle. (16) Universal Linking Principle (A) For unmarked valencies: (i) The argument which exhibits potent proto-agent properties in the first implied subevent will become nominative or ergative. (ii) The argument which has the most proto-patient properties and the least proto-agent properties will become accusative or absolutive. (B) For marked valencies: The argument which exhibits potent protoagent properties in the first implied subevent will receive the highest ranked case in its syntactic valency.

A theory of Lexical Event Structures and its cognitive motivation

259

The linking principle for unmarked valencies not only predicts the distribution of cases for two-place verbs but also for three-place verbs, e. g. for those denoting ‘giving’ events (Lex 18, ex. 17a). The giver in Basque eman ‘give’ appears in ergative case since it has potent proto-agent properties in the initial subevent, the given object occurs as an absolutive NP since it has the least proto-agent properties, while the recipient, which has agentive properties in the second ‘receiving’ subevent, gets the dative as the remaining case of the syntactic valency of the verb. eman

SYN:

/abs/dat/erg

SEM:


z
y
x
e[EMAN(x,y,z,e)]

LES:

( P e1: xACTIVITY/CONTROL, yAFFECTED, zAFFECTED) < ( I e2: yACTIVITY/CONTROL, zAFFECTED) < ( I s: yTHEMATIC, zTHEMATIC)

Lex 18. Lexical entry for Basque eman ‘give’.

(17) a. Zuek lagunei opari poliak ematen You.PL:ERG friend.PL:DAT present nice.PL:ABS give.IMPF dizkiezue. 3:ABS.have.DAT.PL:ABS.3:PL:DAT.2:PL:ERG ‘You always give nice presents to your friends.’ (Joppen & Wunderlich 1995: 129) b. Oku ou muimui ‘i he ka.13 PROG I.ABS follow DAT(OBL) DET car ‘I am following the car.’ (Chung 1978: 54) For verbs with marked valencies, Blume’s linking principle correctly predicts that interaction verbs like German folgen ‘follow’ (Lex 16), Tongan muimui ‘follow’ (Lex 19, ex. 17b) or Hungarian felel ‘answer’ (Lex 20, where it is presupposed that y asks x something), whose distribution of proto-role properties is often completely symmetric, link the proto-agent in the implied subevent to the highest ranked case in the syntactic valency, which is absolutive for muimui and nominative for felel (cf. also Blume 1998). ———–———————————————————————————— 13

I follow Blume (2000: 204) here in naming the subject case with intransitive verbs in Basque ‘absolutive’ (‘nominative’ in other sources) and the non-ergative, non-absolutive case in Tongan ‘dative’ (‘oblique’ in other sources).

260 Stefan Engelberg muimui

SYN: SEM:

/dat/abs
y
x
e[MUIMUI(x,y,e)]

LES:

( P e1[+DUR]: yACTIVITY/CONTROL…) 3s=but ‘However, they will say that he is a great entertainer, but …’

The example shows that such an organization is different from the structural type exhibited by the best known languages outside of linguistics, the European ones. It alone serves to rectify the ideas often held by non-linguists (and even by some linguists) that all languages must be similar in their structure, with Latin serving as a quasi blueprint for grammar, and that anything outside of Europe is somehow primitive. At a first glance, lexical category distinction and polysynthesis seem unconnected topics; on a closer look, however, among the languages with a weak lexical category distinction are polysynthetic languages such as Salishan and Wakashan languages (North America; cf. Sasse 1993a/b; Kinkade 1983). These languages are characterized by complex word forms in which the semantics that would be expressed by nouns and verbs in European languages are encoded in the form of bound morphemes (not roots) within a single word form.

288 Johanna Mattissen These observations fostered the detailed study of the components of polysynthetic word forms pursued in the project to see to what degree the information encoded in nouns and verbs in more analytically structured languages (like the European ones) is expressed by morphemes in complex words forms in polysynthetic languages and renders the noun-verb distinction “superfluous”. For this purpose, a database was set up to file and compare the morphemes of about 75 languages in terms of the ontological domains their meanings can be sorted into. For reasons of space, a choice of these languages is treated here, with the affiliation and sources listed once and for all in table 1: Table 1. Sample of polysynthetic languages Language

Affiliation

Present Location

Source

Abkhaz

NorthwestCaucasian

Caucasus

Hewitt 1989, Lucassen 1985

Ainu

(Paleosiberian) Hokkaido, Japan

Shibatani 1990, Tamura 2000

Atsugewi

Palaihnihan

Pacific coast, USA

Mithun 1999

Awtuw

Papuan

Papua New Guinea

Feldman 1986

Bininj Gun-wok

Gunwinjguan

Arnhemland, Australia

Evans 1996, 2002, 2003

Blackfoot

Algonquian

Western Canada

Taylor 1969

Caddo

Caddoan

Southern USA

Mithun 1999, 1984

Cahuilla

Uto-Aztecan

Southwestern USA

Seiler 1977

Capanawa

Panoan

Peru

Loos Ms

Cayuga

Iroquoian

Great Lakes, USA/ Canada

Sasse 1993a, 2002

Cherokee

Iroquoian

Great Lakes, USA/ Canada

King 1975, Blankenship 1997

Chinook

Chinookan

Northern Pacific coast, Boas 1911d USA

Choctaw

Muskogean

Southeastern USA

Haas 1941a

Chukchi

ChukchoKamchadal

Chukotka, Siberia

Kämpfe/Volodin 1995, Spencer 1995

The Ontology and diachrony of polysynthesis

289

Classical Nahuatl Uto-Aztecan

Mexico

Newman 1967, Andrews 1975

Greenlandic

Eskimo

Greenland

Fortescue 1983, 1984, 1990

Haida

(Athapaskan)

Queen Charlotte Islands, Canada

Swanton 1911, Haeberlin 1921–23

Hupa

Athapaskan

Pacific coast, USA

Goddard 1911, Golla 1996

Ket

Yeniseian

Eastern Siberia

Drossard 2002, Werner 1997

Kiowa

Kiowa-Tanoan West central USA

Watkins 1984

Klamath

—

Pacific coast, USA

Barker 1964, DeLancey 1991, 1999

Kwakwala

Wakashan

Pacific coast, Canada

Boas 1911c, 1947

Lakhota

Siouan

Middle West, USA

Boas /Deloria 1979, De Reuse 1994

Maidu

Maiduan

Pacific coast, USA

Dixon 1911

Miwok

Miwokan

Pacific coast, USA

Broadbent 1964

Mohawk

Iroquoian

Great Lakes, USA /Canada

Baker 1996, Mithun /Corbett 1999

Muskogee

Muskogean

Southeastern USA

Haas 1941a

Navaho

Athapaskan

Southwestern USA

Young /Morgan 1980

Nez Perce

Sahaptian

Northwestern USA

Aoki 1970

Ngandi

Gunwinjguan

Arnhemland, Australia Heath 1978

Nivkh

(Paleosiberian) Sakhalin Island & Amur basin, Siberia

Mattissen 2003

Pano

Panoan

Peru

Loos 1999

Pirahã

Mura

Brazil

Everett 1986

Quileute

Chimakuan

Northern Pacific coast, Andrade 1933 USA

Sarcee

Athapaskan

Western Canada

Cook 1984

Slave

Athapaskan

Central Canada

Rice 1989

290 Johanna Mattissen Sora

Munda

Eastern India

Pinnow 1966, Steever 1986

Southern Tiwa

Tanoan

Southwestern USA

Allan /Gardiner /Frantz 1984

Spokane

Salishan

Northwestern USA

Carlson 1972, 1990

Takelma

—

Pacific coast USA

Sapir 1922

Tarasco

—

Mexico

Friedrich 1971, Foster 1969

Tiwi

Non-PamaNyungan

Bathurst & Melville Islands, Australia

Osborne 1974

Tonkawa

—

Southern USA

Hoijer 1933

Tsimshian

Tsimshian

Pacific coast, Canada

Boas 1911b

Tunica

Gulf

Southern USA

Haas 1941b

Washo

—

Pacific coast USA

Kroeber 1907

Wichita

Caddoan

Central USA

Rood 1976, 1996

Yagua

Peba-Yaguan

Peru/Colombia

Payne/Payne 1990

Yana

Hokan

Pacific coast, USA

Sapir 1911, Mithun 1999

Yimas

Papuan

Papua New Guinea

Foley 1991

Yucatec Maya

Mayan

Yucatan, Mexico

Sullivan 1984, Blair 1964

The finer-grained analysis of the morphemes involved in polysynthesis serves to delimit polysynthesis in the first place. Mattissen (2003, 2004a, 2004b, 2004c) furnishes a first approach to the delimitation and classification of polysynthetic languages, taking the work of Boas (1911a), Mithun (1984), Fortescue (1994) and Drossard (1997) as a starting point. Before, polysynthesis had been a heuristic category in the common use of the term, lacking a generally acknowledged definition. I started out by examining those languages which have been considered polysynthetic in the literature to find their common denominator. Such languages are encountered all over the world with the exception of Africa and arguably Europe, particularly in North, Meso-, and Lowland South America, in Australia, Papua New Guinea, Siberia, India and the Caucasus. As it turned out, however, they are quite heterogeneous in their structure. Language internally, polysynthesis is not a homogeneous principle of language

The Ontology and diachrony of polysynthesis

291

structure, but comprises a range of heterogeneous phenomena, such as polypersonalism (marking of more than one participant on the verb), noun incorporation (integration of a noun stem into a finite verb), verb root serialization (integration of another verb stem into a finite verb), derivation, and affixation. Based on these observations, in Mattissen (2003 and 2004c) a language is acknowledged to be polysynthetic because of (2) (i) the existence of complex, polymorphemic verb forms which allow, within one word unit, for components in the form of non-root bound morphemes with quite ‘lexical’ meaning or optionally for the concatenation of lexical roots; (ii) these components express several of the following categories: Event or participant classification and quantification, setting (e.g. ‘in the night’), location or direction, motion, instrument (e.g. ‘by hand’), manner (‘by pulling’, ‘quickly’), modality (including evidentiality), degree, scale (‘only’, ‘also’) and focus, chronology (e.g. ‘first’, ‘again’), as well as the usual categories such as valence, voice, central participants, tense, aspect (phase), mood, and polarity. “Non-root bound morphemes with rather lexical meaning” are understood to be affixes with meanings which would be expressed by independent forms, especially lexical roots, in more analytic languages, but which never make up free word forms on their own in the languages concerned. They are not roots, nor are they clearly derivational or inflectional morphology. To count as a root, a morpheme with lexical semantics must constitute a free word form on its own or in combination with inflectional affixes. Examples of non-roots are the morphemes -sirsur- ‘provide’ (a verbalizer) and -ssuar- ‘big’ in the Greenlandic verb form above, which are not obligatory, but productive, and do not create lexicalized items. They are known as “lexical affixes” or “field-affixes” from grammars of polysynthetic languages. Even though the existence of non-root bound morphemes is a necessary criterion for polysynthesis in our definition, it is not a criterion with sharp contours, as Mattissen (2004c, 2004b) shows. With the help of the database and against the background of the subclassification of polysynthetic languages (Mattissen 2003 and 2004c) which allows one to deal with the above-mentioned structural heterogeneity I shall attempt to spell this out. Note that we restrict ourselves to the verbal domain here. The nominal side of polysynthesis, i.e. complex noun forms, has been studied in detail in Mattissen (2003 and 2004a).

292 Johanna Mattissen 2. Subtypes of polysynthesis One parameter for such a subclassification (established in Mattissen 2003 and 2004c) is the assignment of the categories listed in the polysynthesis catalogue in (2) to non-root bound morphemes or lexical morphemes. Thus, the fundamental difference between polysynthetic languages in this respect is the number of lexical roots allowed within one verb complex: Just one lexical verb root or more than one root. More than one root is found in the case of noun incorporation and verb root serialization. Furthermore, the languages vary in the number of different lexical roots they can integrate, i.e. in the existence of noun incorporation, adverb incorporation, verb root serialization, etc., possibly also adjective incorporation (depending on the number of lexical categories actually distinguished in the language). This axis describes the word-formational type exhibited by a polysynthetic language. The two extremes on this axis are (i) Languages which use non-root bound morphemes for all of the categories in (2) mentioned above and allow only one root per verb complex, which we will henceforth call the affixal strategy. A good example is Greenlandic. As example (1) shows, there is only one root, aliikkut-, the rest are affixes. (ii) Languages which ad hoc combine more than one lexical root in a verb form to attain a polysynthetic verb form, according to a compositional strategy, besides having non-root bound morphemes. Chukchi exemplifies this type. In (3), all morphemes except the inflectional circumfix are roots. (3)

Chukchi (Skorik 1961: 103) t-tor-ta-plwnt-pojg-pela-rkn 1sA-new-good-metal-spear-leave-PRSI:1s>3s ‘I am leaving a good, new, metal spear’

“Affixal” and “compositional” were chosen in Mattissen (2003, 2004c) as handy terms for the phenomena, which correspond to the terminology for mechanisms in word formation. However, the formation of polysynthetic word forms does not necessarily lead to lexicalized products. The affixal strategy is not generally inflectional as most affixes are not obligatory. Their

The Ontology and diachrony of polysynthesis

293

status is actually somewhere in the grey area between derivation and inflection, as they are fully productive and marked on the verb ad hoc. In between subtypes (i) and (ii), there are languages with noun incorporation, but without verb root serialization, and languages with verb root serialization without noun incorporation. The second parameter (established in Mattissen 2003, 2004c) is the internal organization of the polysynthetic verb form. The language can either be characterized by a verb form which offers a fixed number of slots for different elements which are fixed in their position and their order relative to each other. The maximally complex verb form is determined by the number of slots, although not every slot need to be filled in a single verb form. The verb form may thus be described by establishing a template, e.g. in Navaho (see Young & Morgan 1980: 107) and Ket (Werner 1997: 154–155). Or no such template can be set up, as the components of a verb form are not fixed in their position, but in their scope, which allows them to be ordered according to the intended meaning. The affixes are thus chained rather freely, conditioned by semantics and compatibility restrictions. At the same time, the verb form is not restricted in its complexity and length. Greenlandic is a good example of this organizational type: Only the initial (root) and the final (person & mood inflection, enclitics) positions are fixed, as in (1). Between these morphemes, derivational suffixes and suffixes with fairly lexical or with more or less grammatical meanings are combined in different orders. Chukchi is of the same type, as between circumfixes roots can be chained according to the intended semantics, as in (3). A third subtype, the mixed organization, is exhibited by word forms which have recurrent positions in a template or have templatic and scopeordered sections in the verb form, e.g. a prefixal and a suffixal domain. (4)

Miwok (Broadbent 1964: 40) a. /etal-nuku-lumhu- return-CST-ready-PRS.IMPF:3s ‘he is ready to take him home’ (lit. ‘he is ready to make him go home’) b. /etla-lamhy-nuku- return-ready-CST-PRS.IMPF:3s ‘he is making him ready to go home’

Cross-linguistically, polysynthesis proved to be a different bundling of these features, cf. table 2 (Mattissen 2004c: 210). The first three lines correspond

294 Johanna Mattissen to subtypes of compositionally polysynthetic languages, the last three lines to those of affixally polysynthetic ones. Table 2. Polysynthesis types non-root bound morpheme

scopeordered

templatic

noun incorporation

verb root serialization

+ + +

+ + –

+ – +

+ + +

+ + +

Nivkh, Lakhota Pano Sora, Ket, Wichita

+

+

+

+

–

+

+

–

+

–

+

–

+

+

–

Blackfoot (specimen wanted) Takelma

+ + +

+ + –

+ – +

– – –

+ + +

Maidu, Yimas Capanawa, Yagua Tonkawa

+

+

+

–

–

+

+

–

–

–

+

–

+

–

–

Quileute, Klamath Greenlandic, Kwakwala Navaho

language

These synchronic structures of the polysynthetic verb form have a diachronic dimension. In Navaho (Young & Morgan 1980: 107) and Athapaskan languages, in Wichita (Rood 1976: 17–22) and Ket (Werner 1997: 154– 155), for instance, we encounter different layers of morphemes within the verb form which seem to have been taken up on both sides of the root over time, and are fixed today. Categories belonging to the domains of person, number, aspect etc. are found in different places in the verb form, and the verb form is not further expandable. I would like to term this the “onion type”, compare the Wichita template (cf. Rood 1976: 17–22) with 24 positions in front of and 7 positions after the verbal root: (5)

NSG.S-QUO.T/A-AGT-PRV-UDG-AOR-UDG-REC-PRV-PST.NEG-PL.ADEF-FOC/MANNER/CHRON-NSG.REC-SIZE.INCORP-PAT.INCORPCOLL.U-NSG.U-CLASS-TRV-INS.INCORP-LOCL-DU-VERB.ROOT-DISTRITER-CST-ASP-SERIAL.VERB.ROOT-ASP.SUB-IRR-LOC.PTCL

The Ontology and diachrony of polysynthesis

295

In contrast to this, in Greenlandic and Chukchi the verb forms are expandable by morphemes which are chained inside the verb form, between lexical and/or inflectional morphemes which constitute a frame. In Greenlandic (Fortescue 1984), the first morpheme in a verb form is always a lexical root, the last morpheme is the inflectional suffix indicating person and mood. The non-root bound morphemes are chained and squeezed between the two fixed positions. Greenlandic polysynthesis seems to be quite old, as there is no trace of any lexical origin or other source for the non-root bound morphemes. In Chukchi (Spencer 1995), the first and last morphemes are inflectional, between them, a number of lexical roots and non-root bound morphemes may be chained. (6)

Greenlandic (Fortescue 1984: 315) an-niru-lir-sin-niqar-sinnaa-suri-nngik-kaluar-pakka big-more-begin-CST-PASS-POT-think-NEG-but-IND:1s>3p ‘I do not think they can be made any bigger, but …’

(7)

Chukchi (Spencer 1995: 459) nko mt-mec-qora-grke-pltko-mk then 1pS-almost-deer-hunt-finish-AOR:1p ‘then we almost finished hunting reindeer’

This type I would like to call the “sandwich type” here. In a third type, exhibited by Klamath (Barker 1964), Yimas (Foley 1991), Takelma (Sapir 1922) (and Nivkh, Mattissen 2003), evidence points to a coalescence of at least two word forms adjacent in origin, resulting in a synchronic single word unit. For Klamath, DeLancey compares the morpheme -nannwi- ‘as soon as, right away’, which is transcribed as a separate word by Gatschet in 1890, in the two forms (8)

a. Klamath (Gatschet 1890: 113) qtanc’an nannwi sq’ol-lG-o:t-a-k fall.asleep as.soon.as SG.lie-down-while-IND-EMPH ‘he fell asleep as soon as he lay down’ b. Klamath (Barker 1964: 157, 158, 154) qtan-c’n-nannwi-ank sleep-just.do.and.no.more-right.away/at.once-PST.PTCL ‘having gone right to sleep’

296 Johanna Mattissen and concludes that in Gatschet’s time it “already had a distinct grammatical status, governing an uninflected verb form which could not occur independently, but – accepting Gatschet’s transcription at face value – that it was not yet completely morphologized”

(DeLancey 1991: 434–435). In Yimas, we encounter inside a complex verb form composed of several verbal roots sequential markers which are reminiscent of converb suffixes (which are characteristic in that region). The complex verb form may thus have arisen from the fusion of a converb and a main verb. (9)

Yimas (Foley 1991: 344) na-mp -mampi-kwalca-mpi-tay-cut 3sU-3dA-again-rise-SEQ-see-REMPST ‘they both got up again and saw him/got up and saw him again’

This third type I would like to call the “burdock type”1. These evolutionary types can give us a clue as to how polysynthetic structures may have developed in the first place. And this is where we come full circle: In all of the three types of evolution of complex verb forms, there is grammaticalization of erstwhile lexical roots, which are integrated into such forms, to non-root bound morphemes via an ambiguous stage, as traced in Mattissen (2004b). Grammaticalization of lexical morphemes not only allows us to establish a connection between the seemingly disparate subtypes of polysynthetic languages as sketched in table 2, but raises the question of lexical category distinction with regard to the possible source constructions for synchronic non-root bound morphemes, e.g. noun incorporation and verb root serialization. The precondition is, of course, a nounverb distinction at least in a diachronic stage of the language in question. The questions arising from these reflections are straightforward: – What is the variety of non-root bound morphemes in a single language and cross-linguistically? – Are they distinct in terms of semantics from derivational or inflectional verbal morphemes in non-polysynthetic languages, e.g. preverbs? – Are there recurring ontological domains which may be ordered in an implicational hierarchy? ———–———————————————————————————— 1

The flower-head of the burdock hooks to any host which comes close enough.

The Ontology and diachrony of polysynthesis

297

– Is there a correlation between the subtypes of polysynthesis and the ontological domains represented in the non-root bound morphemes? – Can the semantics expressed by the non-root bound morphemes give us a clue to an earlier structure of the language, e.g. via a correlation of ontological domains to specific source constructions? – How were they integrated, and into which kind of construction? – Are the affixally polysynthetic languages and compositionally polysynthetic languages two different morphological types which should not be subsumed under one structure? – Can a language develop into a polysynthetic language? – Can a language develop away from polysynthesis? – Is polysynthesis a terminal point in morphological evolution or rather a starting point? In the remainder of the paper, I will try to answer these questions and lay the basis for further reflection, with the emphasis on the ontology of nonroot bound morphemes.

3. Ontological domains of non-root bound morphemes The study of non-root bound morphemes in 75 languages showed an impressive range of semantics. Most morphemes may be sorted into a dozen or so larger fields of meaning, others seem to be idiosyncratic. Among the larger fields, the most common ontological domains are the following (see also Mattissen 2003: chs. 7, 10): (10) Ontological domains localities (direction and position) body parts classifiers animals, plants, human beings artefacts/man-made environment medium/natural force motion manner of action

degree chronology (‘first’, ‘again’ etc.) scale/focus (‘only’, ‘just’ etc.) reversative (‘undoing’, see (46a)) quantification phase of action (‘begin’,‘complete’, ‘keep’) modality.

We will turn to examples presently. The category of “instrument” mentioned in (2) above is not an ontological, but a functional class. Morphemes

298 Johanna Mattissen referring to body parts, classifiers, the medium, humans, animals, plants and artefacts may specify a patient or an instrument, e.g. in Spokane (Egesdal 1981), Maidu (Dixon 1911: 693) and Washo (Kroeber 1907). (11) Washo (Kroeber 1907: 288) di-liwi-lup-gic-ue-hi 1s-foot-with-roll-thither-FUT ‘I will roll it this way with the foot’ This, incidentally, is also true for noun incorporation. In Takelma (Sapir 1922: 69) and Cayuga (Sasse 1993a, 2002), the same root morphemes may be incorporated for indicating patients or instruments, with the concrete interpretation in Takelma being dependent on the slot the morpheme appears in. (12) Takelma (Sapir 1922: 64, 65) a. -p! i-nukwa  n hand-fire-warm:1s ‘I warm my hands over the fire’ b. dan-ba- -sget!e’sgidi n rock-up-hand-lifted:1s ‘I lifted up the rocks with my hands’

(first slot: patient)

(third slot: instrument)

In non-polysynthetic languages, the verbal categories tend to be restricted to mood, tense, aspect, person, number and/or valence, and voice. Thus, the list of polysynthetic categories in (10) above is sufficiently distinct as a whole. As a matter of fact, however, – TAM, person and voice are marked on the verb in polysynthetic languages as well and – any one polysynthetic language does not exhibit all of the above-mentioned categories, but different numbers of them in different combinations. This makes the delimitation of polysynthetic languages difficult – if not arbitrary. The Siberian language Ket, for example, only has one category of nonroot bound morphemes: Those denoting localities (see Drossard 2002). Is there a qualitative difference between these morphemes and local preverbs, as in German, Slavic and South Caucasian languages?

The Ontology and diachrony of polysynthesis

(13) Ket d-la-l-aq 1sS-outside-PRF-go ‘I went outside’

German raus-geh-en out-go-INF ‘to go out’

Russian vy-xodit’ out-go:INF ‘to go out’

299

Laz (S. Caucasian) gama-b-ul-ur out-1.A-go-PRS.1s ‘I go out’

In German, the local preverbs exist as prepositions or adverbs as well. In Russian, some preverbs exist as prepositions, others do not have transparent sources. In Laz, the preverbs are all bound morphemes without any plausible source (Mattissen 1995b, in prep.). In Ket, some of the non-root bound local morphemes go back to nominal sources (body part or part of space, cf. Drossard 2002: 239–240), others cannot be traced back to lexical items. Obviously, there is no line of demarcation between the local morphemes of these languages. From the point of view of semantics, all four languages have deictic elements. In addition, German, Russian and Ket use individuocentric local systems, based on the three-dimensional model of axes a human being projects from itself: Up–down, front–back, left–right. Ket and Laz both have a(n additional) system oriented at landmarks, viz. the river in Ket and the mountain/slope in Laz (cf. Mattissen in prep.): (14) Ket orientational system (cf. Drossard 2002) downstream woods  river
bank upstream = inland (15) Laz orientational system (Mattissen in prep.) (up, vertical) cile cin cile  (upwards, slope)   mele  Ego  mele (cross-axis)  (downwards, slope)
ale tude
ale (down, vertical) Thus, Ket does not have a system which does not find its match outside of polysynthetic languages (see also below). The qualitative difference between Ket and the other languages mentioned does not lie in its bound morphemes, but in other characteristics of the language: Ket has productive noun incorporation into the finite verb

300 Johanna Mattissen and verb root serialization. Although there are slight traces of noun incorporation in German (not into a finite verb, however), German, Slavic and South Caucasian languages do not share these features. Turning the argument around then and making noun incorporation and verb root serialization sufficient criteria for polysynthesis is doomed to failure as well. On the one hand, a lot of polysynthetic languages, such as Greenlandic, Navaho, Quileute and Spokane, do not manifest these features, on the other, languages such as Swedish or Japanese have them, but lack non-root bound morphemes (see Mattissen 2003, 2004c for details). Besides Ket, Sora and Nahuatl are polysynthetic languages with a single category of non-root bound morphemes, in fact they only have one and two local morphemes, respectively. Like Ket, both have noun incorporation and verb root serialization (see Mattissen 2003: 173, 182, 280–284). All the other polysynthetic languages studied have more than one of the categories in (10) and are thus easily delimited as polysynthetic. This is especially true for the languages without noun incorporation and verb root serialization, in which polysynthesis relies on the existence of non-root bound morphemes with lexical-like meaning alone: All the relevant candidates tested exhibit at least five different categories. Thus, they may be clearly distinguished from Laz. Polysynthesis may thus not be defined on the basis of a single feature, but is characterized by a bundle of features. Considering the extant non-root bound morphemes in different languages, we come across a wide variety in semantics, presented below. For reasons of space, we will spare the fairly well-known domains of phase (‘begin’, ‘end’, ‘continue’ etc.) and modality. Suffice it here to mention that bound modal morphemes express epistemic, epistemological and deontic modalities, e.g. potential, desiderative, evidentials, intentional, and others. Phase and modality morphemes are exemplified in the Greenlandic verb form below. (16) Greenlandic (Fortescue 1984: 315) an-niru-lir-sin-niqar-sinnaa-suri-nngik-kaluar-pakka big-more-begin-CST-PASS-POT-think-NEG-but-IND:1s>3p ‘I do not think they can be made any bigger, but …’

The Ontology and diachrony of polysynthesis

3.1.

301

The local domain

Most prominent in polysynthetic languages, both in number and in variety, are the morphemes denoting local concepts (mostly “path” in the sense of Talmy 1985). The presence of a non-root bound local morpheme is almost a sufficient criterion for polysynthesis, as only a handful of languages which look polysynthetic otherwise lack local bound morphemes (cf. Mattissen 2003: 280–281). The local domain is subdivided here into three semantic sections – sometimes a subtle distinction. Subsection one comprises the parts-of-space relations found in individuocentric local systems. The individuocentric system is characterized by three axes projecting from a human being, an animal or a thing and opening up a three-dimensional space: Top-bottom, front-back, left-right. In contrast, a geocentric local system is oriented towards landmarks or the natural environment (e.g. mountains, rivers, sea) or cardinal points (north, south, etc.), as in Ket or Laz (see (14) and (15) above)2. Geocentric systems are taken up in the third subsection, whereas subsection two comprises a type which combines individuocentric relations and classifiers. 3.1.1. Individuocentric Systems For some European individuocentric systems, Johnston & Slobin (1979) evaluated in which order the different relations are learned in first language acquisition. Drossard (1992) tested their results against the morphological encoding of these relations in Uralic and Caucasian languages, which are rich in local relations. He found that the relations learned earlier tend to be encoded by case, and those learned later, by postpositions and relational nouns. With this background in mind, it is interesting to see whether in polysynthetic languages, such a hierarchy of local relations is borne out, as well, viz. whether the local relations learned earlier tend to be integrated into the verb forms more frequently than the other ones, and in which form they are integrated (non-root bound morphemes or lexical roots), in order to learn more about the diachronic aspects of the evolution of lexical roots into affixes and of the development of verbal complexity. A survey of our sample shows that individuocentric systems are best represented in Kwakwala, Tsimshian and Spokane non-root bound mor———–———————————————————————————— 2

For local systems see Senft (1992), Levinson (1996) and Pederson et al. (1998).

302 Johanna Mattissen phemes, but absent from Greenlandic ones, although the language has a great number of non-root bound morphemes. Examples for integrated local morphemes are: (17) Takelma (Sapir 1922: 65) dan-ba- -sget!e’sgidi n rock-up-hand-lifted:1s ‘I lifted up the rocks’ (18) Wichita (Rood 1976: 75) kiya:kí-riwa:c-/áras-a-ri-kita-/a-hí:rik-s EVID.AOR:3-big-meat-COLL.U-TRV-top-come-ITER-IMPF ‘by making many trips, he carried the large (quantity of) meat up into it [the tree]’ The following table displays the representation of local relations in different polysynthetic languages, organized according to the Johnston & Slobin (1979) and Drossard (1992) hierarchies. The abbreviations A (affixal), C (compositional), NI (noun incorporating compositional) and VR (verb root serializing compositional) stand for the word-formational subtypes of polysynthesis. The local relations are as follows: Interiority (‘in’), superiority (‘on’), inferiority (‘under’), proximity (‘near’), posteriority (‘behind, after’), mediality 2 (‘between’), mediality 3+ (‘among’), anteriority (‘in front of, before’), exteriority (‘out of’), laterality (‘next to’). Table 3. Individuocentric morphemes I category  inter super infer prox poster med2 med3 anter exter later  language Kwakwala A

+

+

+

+

+

+

+

Tsimshian A

+

+

+

+

+

+

Slave NI

+

+

+

+

+

Wichita C

+

+

+

Ket C

+

+

+

+

Spokane A

+

+

+

+

Klamath A

+

+

Maidu VR

+

+

+

+ +

+

+

+

+

+

+

+

+

+

+ +

+ +

+ +

+

The Ontology and diachrony of polysynthesis

303

category  inter super infer prox poster med2 med3 anter exter later  language Haida

+

Tarasco A

+

+

+

+

Yimas VR

+

+

+

Hupa A

+

+

+

Takelma NI

+

+

Quileute A

+

+

Washo A

+

Nez Perce C

+

Awtuw VR Bininj Gunwok C Tonkawa VR

+ +

+ + + + + +

Greenlandic A

Obviously, the Johnston & Slobin and Drossard’s hierarchies (which were set up for fusioning and agglutinative languages) do not hold for the integration of non-root bound morphemes in polysynthetic languages, except perhaps for the first three relations; note, however, that the interior relation (‘in’) is not basic. There is no other pattern visible, either, except a predilection for ‘up’ and ‘down’. The second most important axis is ‘front’– ‘back’. The concepts ‘left’–‘right’ are not found in integrated form in any language, nor as case/postposition for that matter. The presence or absence of a category does not correlate with the polysynthetic subtype. We half-expected the compositionally polysynthetic languages to have less affixes, as they may express things either by way of integration of lexical roots or as independent words. This is, however, not generally the case. Wichita and Ket have a considerable number of local morphemes. In Ket, some local morphemes betray their nominal origin, i.e. they result from the grammaticalization of erstwhile integrated nouns, which is still a productive mechanism in the language (cf. Drossard 2002). The same is probably true for Slave, where we can trace non-root bound morphemes back to nominal roots (cf. Mattissen 2004b; Rice 1989; and section 3.3).

304 Johanna Mattissen Nevertheless, local non-root morphemes need not necessarily be denominal. Payne (1990: 223) points out for South American languages that “the location, direction, orientation, and especially movement formatives commonly (if not always) derive from verbs. The stages in development from full verb roots which occur in serial-like constructions, to compounding, to bound affixes, could be illustrated from a single language.”

DeLancey (1999: 57) considers the locational suffixes of Klamath to stem from verb roots, as well. Compare also local adpositions, which may be of nominal or verbal origin cross-linguistically (e.g. in Nivkh, Mattissen 2003: 10). Thus, no specific source may be postulated for local non-root morphemes. Table 4. Individuocentric morphemes II category  language

deictics

Kwakwala A

travers

circum

edge

citer

other

+

+

+

+

+

Maidu VR

+

+

+

Klamath A

+

+

+

Tsimshian A

+

+

Hupa A

+

+

+

Nez Perce C

+

+

Slave NI

+

Takelma NI

+

Ket C

+

Wichita C

+

Tarasco A

+

+ +

+ +

+ +

+

+

+

+ +

+ +

+ +

Spokane A

+

+

Washo A

+

+

Awtuw VR

+

+

Quileute A

+

+

Haida

+

+

Bininj Gun-wok C

+

+

Chukchi C

+

Yimas VR

+

Greenlandic A

The Ontology and diachrony of polysynthesis

305

Table 4 displays further relations together with deictics (as in (19)) in individuocentric systems. The relations are: Transversity (‘across’), circumferentiality (‘around’), citeriority (‘this side’), ulteriority (‘beyond’), citerioranteriority (‘opposite’) and ‘edge’ (as in (20)). (19) Mohawk (Mithun 1999: 139) a. ta-s-at-áwei-a’t b. ia’-s-at-áwei-a’t CISL-2.A-MED-be.in-CST TRSL-2.A-MED-be.in-CST ‘come in!’ ‘go in!’ (20) Kwakwala (Boas 1947: 238, glosses missing) a. d -’nx- nd b. ts! m- ’nx- nd ‘take hold of edge’ ‘melt at edge’ Deictics and relations such as ‘back & forth’ (subsumed under “other”) are common in polysynthetic languages as integrated morphemes. 3.1.2. Individuocentric + Classifier Subsection two constitute semantically more complex morphemes which combine individuocentric parts of space with a classificatory (or landmark) component, e.g. ‘across a liquid’ or ‘through a narrow channel’. (21) Tarasco (Friedrich 1971: 232) kor ó-kor ó-nda-pa-ni drag.making.noise-RED-along.the.ground-en.route.going-citat.form ‘drag sth. along on the ground, noisily’ Such a “cumulative exponence” of local features seems to be rare in nonpolysynthetic languages, even in classifier languages, but is not completely alien to them: Laz (South Caucasian), for instance, has preverbs such as dolo- ‘down in a cylindrical object’, moo- ‘thither on a hook’ etc. Below, translation equivalents are listed for the relevant morphemes in the languages evaluated. Kwakwala: inside hollow object, on top of long standing object, on flat object, on round object, top of surface, to sit nearby on floor, to hide nearby outside, fore, stern, inside of material; side of

306 Johanna Mattissen round object, side of long object, side of house; across liquid; end of long horizontal object Tsimshian: through a hole, through a long thing, along a valley, along surface of a long thing, lengthwise along middle line, lengthwise on either side of middle line, along edge, along through middle, along through middle of long thing, through narrow channel Klamath: repeatedly in air, into container, taking home first, to limit/end, up out of water, down a hill, over a mountain, on top of a full load, ending in a dead-end, stuck in tight place, out of tubular object, in narrow place Haida: across a body of water, across a strip of land, toward open place (sea/fire), toward delimited place (woods/inland/island), into canoe Hupa: into enclosure, toward human, shelf above, apart (mass), permanently away, out of receptacle, against/along vertical surface Tarasco: on small flat surface, around the side of/ground, near ground Quileute: side of a canoe, side of object, top of a bag Maidu: down into hole, down into house, down into box Washo: with end of long object Again, the three languages with the greatest number of local morphemes in subsection one excel here as well, while the category is not represented in Greenlandic. Languages with a small number of local affixes do not have affixes combining individuocentric and classifying semantics. 3.1.3. Geocentric Systems The third subsection of locals contains morphemes oriented towards a geocentric system, i.e. a system of localization operating on landmarks or the natural environment (e.g. rivers, mountains) or cardinal points (north– south, etc.). In such systems, the location of a referent is described in relation to salient points in the landscape which may or may not constitute axes in a parallel fashion to up–down, front–back etc. Such orientation systems are not specific to polysynthetic languages. They have been shown to exist in African, Australian and Polynesian languages, in the Himalaya and the Caucasus, without these languages being suspected of polysynthesis (e.g. Islandic (Haugen 1969); Laz (Mattissen in prep.); Hai
om (Widlok 1996);

The Ontology and diachrony of polysynthesis

307

Belhare (Bickel 1994); Tamil (Pederson 1993); Malagasy (Ozanne-Rivierre 1987, 1997); Manam (Lichtenberk 1983); Kilivila (Senft 2001); Longgu (Solomon Island; Hill 1993); Marquesan, Cemuhi, Nemi, Iaai, Lifou (Ozanne-Rivierre 1997); Dyirbal; Guugu Yimidhirr (Levinson 1992); Arrernte (Wilkins 1989); Yankunytjatjara (Lewis 1976); Warlpiri (Laughren 1978); Tzeltal (Brown and Levinson 1991). Even the fact that a geocentric non-root morpheme is integrated into a verb form is no sufficient criterion for polysynthesis, as Laz (cf. (15)) manifests such a construction, as well.




e-bulur ‘go up (vertically)’ ele-bulur ‘go upward (on a slope)’
me-bulur ‘go there (on horizontal plane)’
  cele-bulur ‘go downward’ ce-bulur ‘go down’



(22) Laz

Here are schematical presentations of the geocentric systems of some polysynthetic languages: (23) Hupa (cf. Goddard 1911) north/downstream west/downhill 


 east/uphill south/upstream (cf. Boas 1911b)



upriver water  shore/land

downriver

up (vertical)
upward (slope) houses  woods  downward (slope) down (vertical) 

(24) Tsimshian

rear  middle of house  front/door of house fire
side of house



(25) Takelma (cf. Sapir 1922) into air into house   out of house

downriver  upriver from mountain to plain/into river

308 Johanna Mattissen Compare  



(26) Nivkh (cf. Krejnovi 1982) upriver up (vertical) 

river


shore/bank inland down  

downriver The corresponding local morphemes are integrated into the verb form, as in the following examples: (27) Tsimshian (Boas 1911b: 325) q’ai-hwag9ait-tsag99 am-yu’k-det still-CPL-toward.shore-move-3p ‘they moved still far away toward the shore’ (28) Haida (Swanton 1911: 255) Gien l’ L-g-gî’l-gAn then 3s shape.of.man-swim-landward-PST ‘then he swam landward’ The following list furnishes an overview of translation equivalents of geocentric morphemes in different polysynthetic languages. In some languages, they constitute oppositions, in others, they rather seem unsystematic. Tsimshian: up (vertical), up (slope), down (vertical), down (slope), water shore, landwater, woods behind houses, woodshouses, rearmiddle of house, middlefront/door of house, side of housefire, fireside of house, upriver, downriver, out of water, at right place Kwakwala: out to sea, up from beach, down to beach, opposite side downriver, across a hill, downriver, upriver, into woods, along bank of river, shoreward, mouth of river, top of hill/bank of river, bank of river, on surface of water, in house/floor, into house, outside front of house Quileute: beach, sky, gravel at bottom of sea, prairie, territory, river, river bank, point/peak, coast, hill, cape Takelma: upriver, downriver, into air, from mountain to plain/in river, into house, out of house

The Ontology and diachrony of polysynthesis

Awtuw: Slave: Wichita: Hupa: Klamath: Nez Perce:

309

until dawn, upstream, downstream, uphill, downhill into air, to shore, to a point, out of the water, into water ditch, open area/plain, brush under surface of water, into/out of water surface up against/to shore into brush

The most elaborate systems are again manifest in Tsimshian and Kwakwala, two polysynthetic languages which stand out for the number of their local morphemes. Although the greatest number of geocentric morphemes is found in affixally polysynthetic languages, most polysynthetic languages exhibiting a geocentric system are of the compositional type. To conclude the local section, we did not find systematic correlations which could serve as a necessary or sufficient criterion for polysynthesis even if the local domain is the largest one. The local systems are moulded in an idiosyncratic fashion. The source of the morphemes can be both a nominal or verbal lexical root. It is interesting to note that the languages manifesting the greatest number of non-root bound morphemes on the whole, Kwakwala and Tsimshian, dispose of the largest inventories of local morphemes, i.e. the majority of their non-root bound morphemes has local semantics. At the same time, these languages do not have a clear distinction of nouns and verbs as lexical categories. Greenlandic, however, despite its large set of non-root bound morphemes, does not have local morphemes, but a clear noun-verb distinction.

3.2. Motion Close to the local morphemes in semantics, there is a small class of non-root bound morphemes signaling an accompanying (directed) motion. Again, translational equivalents are given below. Nez Perce: as object passes by, trail behind, move in order to/around, go away to, return from Klamath: arriving, go to V, come to V, while moving, going to meet Awtuw: and go away, and get up, and throw away Haida: push/outward hands, pull Hupa: motion away Washo: motion

310 Johanna Mattissen Among the languages without local non-root bound morphemes, Cahuilla and Kiowa have motional morphemes, which could count as a compensatory class for the missing local morphemes (Mattissen 2003: 280). (29) Kiowa (Watkins 1984: 180) O` hègO´ yá -kO´y-tò-mO` ·gO` -O` mdè-hO now 3sA>1sU.PL-Kiowa-speak-adept-become-go ‘I am getting better at speaking Kiowa’ This would allow to make the presence of locational/motional bound morphemes a necessary criterion for polysynthesis. However, motional morphemes are not mutually exclusive with local morphemes; Klamath (Barker 1964), Haida (Swanton 1911) and Hupa (Goddard 1911) have both. In Awtuw (Feldman 1986) and Chinook (Boas 1911d), motion verbs get integrated into a complex verb form via verb root serialization. (30) Awtuw (Feldman 1986: 52) ka-d-ma-taw-owra-t-akla-kow-kay-e NEG-FAC-go-yet-again-DU-dig-BEN-PRF-PST ‘(two) had not gone and dug again for (s.o.) yet’ Such a construction could be the source for a grammaticalization of motional morphemes, as seems to be the case for Klamath: DeLancey (1991, 1999) sketches an evolution from verb root serialization to affixal polysynthesis (see Mattissen 2004b). He assumes that the non-root bound morphemes in positions 10 through 14 of the Klamath verbal template, which contribute locational, directional or manner meanings, originated in verb roots in a “verb-compounding construction”, with non-last verbs in the string uninflected (1991: 442–443; 1999: 61, 78–79). In Greenlandic, motional verbalizers never suffix to verb stems, but only to nominal(ized) stems. The language does not have verb root serialization, in fact no two roots are allowed in one complex word form. 3.3. Body parts Besides the local domain, body parts play an important role as non-root bound morphemes, and indeed in the integration of material into a complex verb form: Body parts are also the most important category integrated via

The Ontology and diachrony of polysynthesis

311

noun incorporation. Body part roots or morphemes integrated into the verb refer to the patient, instrument or location, and can be possessed by the actor. (31) Spokane (Carlson 1972: 99) a. n-l-áw’s-q n-t-m in-chop-half/middle-head-TR-MED ‘his head got chopped in the middle’ b. pxw- n-ést -m-n-t-xw blow-back-hand-TRV-CTRL-TR-2sA ‘you scatter it out from the palm of your hand’ The following list gives some examples of the inventory of non-root body part morphemes in form of translational equivalents: intestines/sinew, nose, leg/foot, fish tail, anus, anus/buttocks , skin/hide, tooth, forehead, elbow, mind/heart, breast/ trunk/lungs, breasts, eye, hand/twig/branch, tail of a quadruped, head, vulva, eyebrow, thigh, throat, body, heel, hair, neck, muscles, hip, fur, meat/flesh, human hair, feather/wing/gill, penis, shoulder, testicles, bucal cavity, back of head/nape, rib, shoulder/top part, foot/leg, shoulder, foot/leg, nose, ear, mouth, face, tongue, hand, palm Tarasco: guts/field, neck/throat/inner eave, buttocks/genitals/bottom, hand/finger, mouth/border, taste, crotch, mouth/vulva/bud/ door, loins/groin/genitals/home village, chest/heart/interior of canoe/house, eye/cheekbone/face/lap/inner wall/cob/memory, heart/field, back/roof/outer wall/corn husk/outside/pot belly, nose/forehead/flower/seed, arm, side/thigh/floor/side of pot/ upper leg, ear/shoulder/corner/ground, jaw/teeth/chin, lower leg/exterior, buttocks, outside of head/brains/top of corn/hill/ tree Kwakwala: body, in front of body, surface of body, eye/door, head, face, temples, cheek, forehead, ear, nose, inside of nose, mouth/ vessel, tooth, neck, throat, shoulder+upper arm, hand, chest, back, lap, crotch, knee, lower leg, foot, penis, shin, in mind Spokane: skin/body/dwelling, eye/fire, lips/tongue/food, hand, head, bottom, back, foot, throat, nose, urine, tooth, stomach, ear, arm, breast, belly, meat, neck, meat body, front, side, shoulder, body covering, tail, thigh, side Quileute:

312 Johanna Mattissen Tiwi:

Klamath:

Maidu: Washo: Haida: Lakhota: Hupa: Bininj G.:

head, face, forehead, eye, ear, nose, mouth, tongue, chin, beard, neck, shoulder, upper arm, lower arm, hand, chest, breast, belly, navel, penis, vagina, pubic hair, kidney, side, back, buttocks, upper leg, knee, lower leg, ankle, foot back, head first, fist/foot, finger(nail), buttock, mouth, elbow, from hand to hand, opening mouth wide, with hand outstretched, along the back nose/snout, sitting on, foot, fingers, foot with hand, with foot, with head, with teeth/rubbing back, shoulder, teeth, hands/grasping foot, (hands), (mouth) into mouth lying on back

The category is strongest in affixally polysynthetic languages, compositional ones rather having noun incorporation instead, as for instance Takelma. (32) Takelma (Sapir 1922: 64–65) a. dan-ba- -sget!e’sgidi n rock-up-hand-lifted:1s ‘I lifted up the rocks’ b. -p! i-nukwa  n hand-fire-warm:1s ‘I warm my hands over the fire’ Body part morphemes furnish evidence for an evolution from root integration to affixation: Polysynthetic languages such as Haida, Slave and Lakhota have both incorporation of body part nouns and integration of non-root morphemes referring to body parts into the same position in the verb (cf. Mattissen 2004b). Haida has a range of instrumental body part prefixes, with lexical roots such as in (33) being used in the same way: (33) Haida (Swanton 1911: 221–226) used as lexical (instrumental) prefix st!a- ‘by kicking’ sL!‘with the fingers’ k!ut- ‘with the lips’

lexeme ‘foot’ ‘hand’ ‘lips’

The Ontology and diachrony of polysynthesis

xAñ
iLu-

‘with the face’ ‘with the arms’ ‘by canoe’

313

‘face’ ‘arm, wing’ ‘canoe’

For Slave, Rice (1989) identifies “position 4” in a complex verb form in which both noun roots are incorporated and non-root bound morphemes occur. Some morphemes, such as ‘head’ in (34a) below, are identical in shape and semantics when forming free nominal forms with the help of an inflectional affix and when integrated into position 4. These are cases of noun incorporation. Note that for ‘head’, there are already several allomorphs for incorporation which the root does not seem to exhibit. This may constitute a first phase of development separating the integrated form(s) and the root. The concept ‘bone’ in (34b) is expressed by two distinct morphemes in position 4, one of them identical in shape to the lexical root, the other of a different but apparently related form. The latter observation is also true for the morpheme(s) ‘hair’. This state of affairs is open for interpretation as to the status of the integrated morphemes. Are they allomorphs of the roots, or are they affixes which are most probably etymologically related to the roots? The morphemes in (34c) and (34d), although found in the noun incorporating position of the verb, are undoubtedly affixes in Slave, without any relation to a lexical root language-internally. In the case of ‘work’, however, there is an obvious cognate in the affiliated language Dogrib with the same meaning, which is a lexical root. This leads us to suspect the former existence of a lexical root -la- ‘work’ in Slave as well. (34) Slave (Rice 1989: 647–669) root morpheme integrated a. ‘stick’ -téh -téh‘head’ -fí -fí- ~ -kwí- ~ -tthí- ~ -píb. ‘bone’ -w’éné -w’éné-, -w’i‘hair’ -ghá -xac. ‘work’ — -lad. ‘food’ — -shé‘rope’ — -ah-

cognate (Dogrib)

la ‘work, job’ — —

For Tiwi, Osborne (1974: 48) notes that “incorporated forms [non-root bound morphemes, JM] and free forms [lexical roots] are generally not cognate, however, as the free forms which were cognate with the existing

314 Johanna Mattissen incorporated forms have long since disappeared, owing to the rapid rate of lexical replacement in Australian languages”. Examples are (35) Tiwi (Osborne 1974: 48–50) non-root morpheme reconstructed root pula *jimpula akli *jikala mumu *mumuta

meaning ‘knee’ ‘upper leg’ ‘back’

For Washo, an affixally polysynthetic language, Jacobsen (1980) is able to trace four (out of ca. 125) prefixes to noun roots language internally, which suggests earlier noun incorporation. (36) Washo (Jacobsen 1980: 86) dulˇ- ‘with the hand’ : á:du ‘hand’ Language-external evidence can be adduced within the Muskogean languages, which display affixal polysynthesis, i.e. synchronically allow only one root per complex verb form. Only Muskogee manifests petrified remains of noun incorporation which allow to suspect a development to have taken place in the language family. Haas (1941a: 312–315) identifies the element nok- in Choctaw verbs, whose semantic contribution is ‘throat’ or ‘neck’, which does not, however, exist as a lexical root in Choctaw. The word for ‘neck’ is the non-related iko:la (-ko:nla:). (37) Choctaw (Haas 1941a: 312–315) affix-root nok-taka:li throat-stick ‘have sth. stuck in the throat’ Haas demonstrates that in related Muskogee there is a cognate root -nok- as in ca-nók-wa (1s-throat-suffix) ‘my neck’, which is also found in fossilized incorporation, while in Koasati the root is -no:bi ‘throat’ (< *nok-pi), with nok- occurring integrated into verbs. Thus, Haas draws the conclusion that there was noun incorporation in “Proto-Muskogean” (1941a: 315). At that stage, then, the language was of the compositional type. The examples of Slave, Haida, Tiwi, Washo and Choctaw sketch a possible path of development from a noun incorporating language to a language with non-root bound morphemes.

The Ontology and diachrony of polysynthesis

315

Although the language is affixally polysynthetic and has a large set of nonroot bound morphemes, body part morphemes are absent from Greenlandic. In contrast to Quileute, Kwakwala and Spokane, Greenlandic has a sound noun-verb distinction. The language has no noun incorporation, but deverbal nouns instead. These facts strengthen the idea that body part morphemes may develop from noun incorporation and are blocked if a language does not have noun incorporation at all during its history. 3.4. Classification Next in number are non-root bound morphemes which classify a participant with respect to shape, animacy, consistency, organization/configuration etc. and, from the point of view of semantics, are reminiscent of noun classification in Asian languages. (38) Atsugewi (Talmy 1985: 74) cw-a-ss aa q-íct-a 3S-from.wind.blowing.on.it-runny.icky.material.moves-into.liquid-FAC ‘the guts blew into the creek’ (39) Spokane (Carlson 1990: 74) ’ -/o xw-álqw under-go-long.cylindrical.object ‘he crawled along under a log’ The category is well represented in a lot of polysynthetic languages, both of the affixal and the compositional types, especially in Haida, Maidu, Kwakwala, but again not represented at all in Greenlandic. It exists alongside noun incorporation in Bininj Gun-wok (Evans 1996, 2002) and Wichita (cf. Rood 1976: 23). Haida:

Maidu:

sack-full, cubic, lying on ground, coiled/wrapped, spoon/ feather, slender, ring, cylindrical, round, rope, long, heap, textile/two dimensional, flat+thick, short, small, flat/two dimensional, roundish, flexible objects tangled together, animate, spike-full, flat+thin, branching, large inside, large cylinder, small, slim+blue, red+short+protruding, small, open space, dumb-shell curled, small+slender round/massive, edge of long thing by steady motion, end of long thing by steady motion, edge/side of thing by sudden mo-

316 Johanna Mattissen tion, end of long thing by sudden motion, end of thing by sudden motion, endways/in direction parallel to length, scattered object, animate, inanimate, round/massive object Kwakwala: human, tribe, flat, long, round, finger-width, fathom/span, times, at a time, days, bundles, five pairs, pair, dish, strings of fish, hole, layers, one way, price Quileute: day (with numerals), place, bunch/handful, piece, (back-) pack, box, load, occasion/turn/time, year Klamath: round, long, live, flat, sharp/pointed, extend in line, clothlike Spokane: day, scattered object, long cylindrical object, kind, band, round object Bininj G.: flat, long, spread/parallel, on side Wichita: animate, liquid, activity Slave: wooden, soft+oily Hupa: adult Hupa, other Tarasco: flat surface Washo: with long object Tiwi: sth. empty In a diachronic perspective, the morphemes in question most probably originate in lexical roots. A case in point is Klamath. DeLancey traces the classifying prefixes of Klamath to noun roots in a noun incorporating construction (1999: 57, 79). An evolution of a classifier morpheme from a lexical root can be sketched along the lines of different syntactic stages in modern incorporating languages. The source construction is noun incorporation, as in (40) Yucatec Maya (Mithun 1984: 857) a. k-in-’ak-ik e’ (iil in-kool) (free N) ICPL-1sA-chop-IMPF.TR tree in 1sPOSS-cornfield ‘I chop a tree’ b. k-in-’ak-e’ (N incorporation) ICPL-1sA-chop-tree (IMPF.ITR) ‘I wood-chop’

The Ontology and diachrony of polysynthesis

317

(41) Chukchi (Kozinsky, Nedjalkov & Polinskaja 1988: 667) a. tlg-e tkec/-n (utkuc/-k) pela-nen father-ERG bait-ABS.SG (trap-LOC) leave-AOR:3s>3s ‘father left bait in the trap’ b. tlg-n t kec// -pela-g/e father-ABS bait-leave-AOR:3s ‘father left bait’ Note that there are two basic types: In type one (exemplified above), noun incorporation is detransitivizing, i.e. the referent3 of the incorporated noun is not a participant marked on the verb as a pronominal affix or encoded outside the verb as an overt constituent. In type two, noun incorporation is not detransitivizing, and the referent of the incorporee is marked on the verb as a pronominal affix or encoded as a free pronoun. In the example from Southern Tiwa below, the portmanteau morpheme encodes first singular actor acting on animate singular undergoer, and the latter refers to the referent of ‘man’. The personal prefix does not change after noun incorporation, i.e. now both the pronominal prefix and the incorporee have the same referent. I call this type of encoding “internal doubling”. (42) Southern Tiwa (Allen, Gardiner & Frantz 1984: 294–295) a. seuanide ti-m-ban man 1s>As-see-PST ‘I saw a/the man’ b. ti-seuan-m -ban compare plural: c. bi-seuan-m -ban 1s>As-man-see-PST 1s>Ap-man-see-PST ‘I saw the/a man’ ‘I saw the men’ Now, in some languages in which there is detransitivizing incorporation, incorporation may also be promoting peripheral participants and thus function as an applicative. In the examples below, the incorporating verb is transitive, and its undergoer in the absolutive is promoted from a local relation; compare (40a) and (41a).

———–————————————————————————————

3

I use reference in the sense of pointing to a referent without implying anything about discourse referentiality here.

318 Johanna Mattissen (43) Yucatec Maya (Mithun 1984: 858) k-in-’ak-e’-t-ik in-kool ICPL-1sA-chop-tree-TR-IMPF 1sPOSS-cornfield ‘I clear my cornfield’

(applicative)

(44) Chukchi (Kozinsky, Nedjalkov & Polinskaja 1988: 667) tlg-e utkuc/-n t kec// -pela-nen father-ERG trap-ABS.SG bait-leave-AOR:3s>3s ‘father left bait in the trap’ A second type of transitive incorporating verb is possessor raising, with the possessor of the incorporee being the undergoer in the absolutive. (45) Chukchi (Kozinsky, Nedjalkov & Polinskaja 1988: 682–683) a. nan remkl/-in pojg-n mce-tku-nin 3s.ERG guest-POSS spear-ABS break-ITER-AOR:3s>3s ‘he broke the guest’s spear’ (transitive active) b. nan pojg -mca-tko-nen remkl/-n 3s.ERG spear-break-ITER-AOR:3s>3s guest-ABS lit.: ‘he spear-broke the guest’ (possessor raising) Both constructions have the necessary constellation for a further development: An incorporating verb, an incorporee and a discourse-referential undergoer, and in this they parallel the non-detransitivizing construction. Chukchi, however, which does not have classifying affixes, stops at this point. Languages with internal doubling have promotional and possessor-raising incorporation, as well. Compare the following examples: (46) Mohawk (Mithun 1984: 868, 871) a. wa-hshako-hkwvny-ahrá:-ko-’ ne akohsá:tvs PST-3sm>3sm-harness-set-REVERS-PUNC PART horse ‘he harness-removed the horses’ (applicative) b. wa-hi-’sereht-anv´hsko PST-3sm>1s-car-steal ‘he car-stole me’ (possessor raising) c. kv´tsyu v-kuwa-nya’t-ó:’ase fish FUT-3p>3sf-throat-slit ‘they will throat-slit a fish’

The Ontology and diachrony of polysynthesis

319

(47) Caddo (Mithun 1984: 864) dahku’-’á-nníw’áhah 2s>1s-eye-remove ‘take my eyes out’ As in Southern Tiwa a third person undergoer is overtly marked, we observe all three persons involved marked on the verb (1st, 2nd, 3rd of paradigm A), with internal doubling still extant: (48) Southern Tiwa (Allen, Gardiner & Frantz 1984: 307) ka-kuchi-thã-ban (possessor raising) 1s>2s|A.s-pig-find-PST ‘I found your pig’ The evolution towards classifying constructions starts out from enlarging the pool of possible undergoers to verb-external nominal participants. An overt form identical in reference to the incorporee is used with the incorporating verb. I call this stage “external doubling”. (49) Mohawk (Baker 1996: 310, 318) a. Uwári -ye-nakt-a-núhwe’-ne’ ne Sak rao-nákt-a’ FUT-3fs.A-bed-epV-like-PUNC PART S. 3ms.U-bed-STAT M. ‘Mary will like Sak’s bed.’ b. ra-wir-a-núhwe’-s thík owirá’a 3msS(>nU)-baby-epV-like-HAB that baby ‘he likes those babies’ (50) Gunwinjgu (Mithun 1984: 867) bene- ed-na  ed-gereeni 3d-camp-saw camp-new ‘they saw a freshly made camp’ (51) Yana (Sapir 1911: 269) gi -áuna ‘ái- y-au-si-ndja carry-inorganic-fire-PRS-1sA NON.SUBJECT-fire ‘I have carried fire’ The further development exploits the principle of enlarging the pool of possible undergoers, this time in the semantic perspective. The next step is taxonomic incorporation, as in

320 Johanna Mattissen (52) Mohawk (Baker 1996: 310) … rabahbót wa-h-tsy-a-hnínu-’ bullhead FAC-3ms.A-fish-epV-buy-PUNC ‘he bought bullheads’ (53) Gunwinjgu (Mithun 1984: 867) bene-dulg-na
mangaralaljmayn 3d-tree-saw cashew.nut ‘they saw a cashew tree’ The incorporated root is a hyperonym to the referent of the free NP. In this stage, the 1:n relation between incorporees and free nouns arises, laying the foundation for classification. One and the same incorporee cooccurs with free nouns of different semantic domains. In Caddo, for instance, /ah-uh is the free form of ‘eye’ (with suffix -uh). The examples below show the root as an incorporee with different overt NPs. (54) Caddo (Mithun 1984: 865) a. kassi/ háh-/i-/ á-sswí/-sa/ bead PROG-prefix-eye-string-PROG ‘she is stringing beads’ b. ka/ás háh-/i-/ ah-/í/-sa/ plum PROG-prefix-eye-grow-PROG ‘plums are growing’ The evolution towards a classifying affix then presumably proceeds as follows: Roots with more general semantics specialize in incorporation and are lost as an independent lexeme. At the same time, they are usually used in co-occurrence with an external noun. In Mohawk, there are two morphemes (‘body’, ‘animal’) which are only found integrated into a verb, whereas free forms with the same or a similar meaning are non-related roots. (55) Mohawk (Mithun 1984: 864) a. shakoti-ya’t-í:sak-s ne ronú:kwe 3p>3p-body-seek-IMPF PART mp-person ‘they were looking for the men’ b. sk hrao-naskw-y ne takos one msU>nA-animal-own PART cat ‘he has one cat’

The Ontology and diachrony of polysynthesis

321

As these two morphemes have lost their connection to a lexical root, they must be considered non-root bound morphemes, and are classifiers then. There is further evidence for such an evolution to non-root bound morphemes within the Iroquoian language family. Sasse (2002: 215–216) shows different stages within Cayuga, taking the lexeme ‘be in liquid’ as an example: (56) Cayuga (Sasse 2002: 215) a. productive root incorporation: h-k-at-ahsí’t-o-h ‘I’ll put my feet into the water’ TRSL-3snA-SRFL-foot-be.in.liquid-CST b. compositional: t-k-at-rnáw’t-o-h ‘I’ll put sugar in liquid’ PRV-3snA-SRFL-sugar-be.in.liquid-CST c. monolithic but transparent: -at-rihst-o-h ‘trap’ (lit. ‘put metal in water for self’) -SRFL-metal-be.in.liquid-CST d. monolithic opaque: -ahs-o-h ‘paint’ (-ahs- ‘??’) The element -ahs- is not a root in Cayuga, but is found in the incorporee position, suggesting a lexical source. In Cherokee (Southern Iroquoian), root incorporation is restricted to a few body part roots and roots denoting items of clothing, otherwise we encounter classificatory verbs, in which the classificatory elements are nonproductive and sometimes not even segmentable (cf. Blankenship 1997: 98–99). Language-internal (and, of course, language-external) evidence point to a stage with more productive incorporation, however. Example (57a) presents a morpheme for which we may establish an etymological relation to an extant root. (57b) illustrates the Cherokee element -neeh-, which, although lexicalized on verbs and non-productive, can be identified in classificatory verbs as indicating a liquid patient. It is cognate with the incorporated root ‘liquid’ *-hnek- in the Northern Iroquoian languages, which still make use of root incorporation synchronically (cf. Blankenship 1997: 97), e.g. Mohawk -hnek-ihr- ‘liquid-consume’.

322 Johanna Mattissen (57) Cherokee (King 1975: 98) a. a-sgwal-niha : 3s>3s-head-hit:EMPH ‘he hit him on the head’ b. ga-ne-hiya’a : 3s>3s-liquid-leave:PRS ‘he leaves sth. liquid behind’

askoli ‘head’

a-hiya’a 3s>3s-leave:PRS ‘he leaves sth. behind’

With incorporation fossilized in Cherokee and the connections to roots lost, the morphemes have become non-root bound morphemes. Thus, Cherokee is further advanced than its sister languages in an evolution from root incorporation to non-root bound morphemes and from compositional to affixal polysynthesis. When such morphemes eventually become opaque, the construction reaches the stage of classificatory verbs.

3.5. Medium and natural force This domain comprises the non-local elementary surroundings something is in or an action happens in, i.e. fire, ground, water, air, heat, cold, and natural forces such as ‘sun’ and ‘wind’. (58) Wichita (Rood 1976: 63, 271) ke/-iyákic-/i/ássi:s-hí:ko:-te/es-íki-h QUO-PTCL.INDEFS.REFLPOSS-father-deep.inside.water-die-CST-SUB ‘the ones who drowned their father’ It is a small, but fairly common class, of which ‘fire’ is the most common member, and is found in both affixal and compositional polysynthesis. Here is a list of translational equivalents: Kwakwala: Quileute: Klamath: Wichita: Bininj G.: Tarasco: Haida:

on fire, water/air, flames, adrift, floating, on rock, on ground rock, fire, sun, (body of) water, ground/soil/dirt fire, into fire, on a fire, out of fire/water, in the sun water, fire, stuck in ground fire, water, sun liquid/wound/mind, on the ground fire in body, under water

The Ontology and diachrony of polysynthesis

Tiwi: Spokane: Lakhota: Hupa: Slave:

323

fire, stone water, sth. in water heat into fire into fire

The languages with the largest inventories of non-root bound morphemes are best equipped here as well, and once again, the class is not represented in Greenlandic. In Slave and Haida, concepts such as ‘fire’, ‘land’, ‘current of water’ are integrated into the verb via noun incorporation, non-root morphemes are the exception, but again this points to a possible origin of the class. (59) Haida (Swanton 1911: 221, 223) integrated morpheme dal- ‘by means of a current of water’ Al- ‘by fire from without’

root ‘rain’ ‘sunshine’

In Haida, however, there is also the morpheme ‘wind/breath’ identical to the verb t- ‘blow’ in verb root serialization (Swanton 1911: 224). Up to this point, the most conspicuous fact is that Greenlandic with its large set of non-root bound morphemes does not share the categories with the largest numbers of morphemes and those which are probably of nominal origin. 3.6. Manner Another fairly common class among both compositionally and affixally polysynthetic languages encoded with non-root bound morphemes is manner of action. Out of 62 clearly polysynthetic languages in the sample, local morphemes are represented in 59, whereas manner morphemes are manifest in 25. (60) Kwakwala (Boas 1947: 245) a. d ’ln-ak˚ b. d ’nxala-k˚ina damp-easy sing-nice ‘get damp easily’ ‘sing nicely’

324 Johanna Mattissen (61) Nez Perce (Aoki 1970: 87–88) a. wilé·-ke/yk-se b. siwí-ke/yk-se c. / ipsqí-ke/yk-se swim-move-SG on.foot-move-SG running-move-SG ‘I am running’ ‘I am swimming’ ‘I am walking’ (62) Tonkawa (Hoijer 1933: 28) ge-nec-xa-ile-nec’-o-’ 1sU-CST-with.force-hit-DU-DECL-3.PRS ‘the two of them made me hit him with force’ Here are the translation equivalents of manner morphemes found in the sample: Tiwi:

sneak up, run, crawl, swim, sleep, full, hungry, talk, work, fight, cry, eat, dance, burn, hot, cold, be frightened, finish, look, listen, hunt, make spears, get up, standing up, walking along, go along, get water Greenlandic: suddenly, suddenly a bit, half-heartedly, badly, with difficulty, hard, carefully, well, at ease, powerfully, violently, strongly, quickly, hurriedly, in vain, with emotion Awtuw: originally, quickly, separately, together, properly, wrong, badly, in vain, secretly, detain s.o., cover sth., block sth. Klamath: run, sit, stand, drag, burn, tie, look, peek, smell, slit open, in game/fun, holding, replacing, spread out, avoiding, rolling in snow Kwakwala: accidentally, easily, exclusively, real, similative, as expected, as usual Lakhota: quickly/suddenly, cutting, shooting/punch, push with hands, pulling, pressure Nez Perce: deprive of sth., in vain, competitive, half-heartedly, as someone Bininj G.: properly, wrong, afraid, secretly, violent intent Hupa: sit/carry/in air, into pieces, crossing/stretched Ainu: truly, alone, suddenly Chukchi: truly, quickly, simply Maidu: spontaneously/by accident, by force/by pulling Wichita: quickly, accidentally Quileute: sick, kind/sort

The Ontology and diachrony of polysynthesis

Tonkawa: Spokane: Haida: Washo:

325

with force/to a distance successfully tread, chop/club, lead/pull/tow, tie by turning

The manner morphemes manifest a great variety, and thus do not seem to form a veritable class. Nevertheless, ‘quickly’, ‘accidentally’ and ‘suddenly’ recur. This time, Greenlandic exhibits the category as well, while not having adverbs as a lexical category. In Haida, there are lexical verbs homophonous to manner components integrated into the verb, i.e. the language makes use of verb root serialization for the expression of manner besides the non-root morphemes (if they are non-roots). (63) Haida (Swanton 1911: 220–226) morpheme meaning tc!ît‘by shooting’ nan‘by grinding’ qea‘by looking’ q!eît‘by cutting, with a knife’ L‘by contact’

verbal root identical in form ‘shoot’ ‘grind’ ‘look’ ‘cut’ ‘touch’

This is a probable source construction for the category as a whole. In Ngandi (Heath 1978), productive verb root serialization is possible with ‘throw’, ‘get’, ‘make’, verbs of sense perception and other bodily activities. All other components integrated into a complex verb form are nonroot bound morphemes. DeLancey (1999: 57) traces the manner prefixes of Klamath to verb roots, as well. Although verb root serialization is not a productive mechanism in Klamath, DeLancey mentions three residual examples which constitute the missing link: The two roots sdig- ‘smell’ and nqen- ‘shout’ may form a free verb form with an inflectional suffix but may also precede another verb root, i.e. still allow verb root serialization (1999: 77). Furthermore, he establishes the following connection: (64) Klamath (DeLancey 1991: 433–435; 1999: 76) non-root bound morpheme assumed source root ci:- ‘live, stay, dwell’ (verb root) cv- ‘sitting’ (prefix)

326 Johanna Mattissen The grammaticalization of verb root serialization advances as it becomes obligatory to integrate certain verb roots which thus lose their root status (see Mattissen 2004b for details). The root dies out and may later be renewed, as Jacobsen (1980: 98) alludes to for Washo by mentioning “the absence of synonymous independent stems to match most of the type I prefixes, which suggests that they have lost their status as stems too recently for new stems to have arisen to complement them. Thus there is no other way to express, for example, ‘running’ or ‘throwing’ or ‘raining’ than by the use of the appropriate lexical prefix.”

When the connections to roots are lost, the morphemes have become nonroot bound ones. Verb root serialization then becomes unproductive as a mechanism and fossilizes. 3.7.

Smaller categories

Further ontological domains are less common, but nonetheless characteristic for both compositional and affixal polysynthesis when integrated as nonroot bound morphemes. Again, translation equivalents are listed below. Greenlandic not only exhibits these categories, but is well-equipped with relevant morphemes. 3.7.1. Degree A morpheme of degree is Chukchi ‘almost’: (65) Chukchi (Spencer 1995: 459) nko mt-mec-qora-grke-pltko-mk then 1pS-almost-deer-hunt-finish-AOR:1p ‘then we almost finished hunting reindeer’ Greenlandic: rather, more or less, somewhat, greatly, a little, completely, not especially, almost, most, more, much more, a little more, partly, just a little, barely, a lot, enormously, very much, so, a bit, not so much, too, really Kwakwala: a little more, very, too much, completely Chukchi: intensive, a little, gradually, superlative, almost

The Ontology and diachrony of polysynthesis

Ainu: Quileute: Hupa: Klamath: Nez Perce: Awtuw: Bininj G.: Tonkawa:

327

completely, low, totally, well, thoroughly to excess, thoroughly nearly, hardly almost, at once in excess incompletely partly nearly/almost

In Wichita, the roots ‘big’ and ‘little’ have to be incorporated into the copula verb or a noun form, they do not form word forms on their own (Rood 1976: 11–13). Incorporation is in line with the generally compositional type of the language and may constitute a first step of grammaticalization of a lexical root to a non-root bound morpheme of degree. (66) Wichita (Rood 1976: 12) né:rhir/a has/a:kí-riwa:c-/i buffalo QUO.AOR.3s-big-be ‘it was a big buffalo’ In Ainu, there is verb root serialization for the expression of ‘over-’ besides non-root bound morphemes of degree: (67) Ainu (Tamura 2000: 195) mokor-kasu sleep-exceed ‘oversleep’ Haas presents three verbal suffixes in Tunica, among them -(h)ila- ‘on the point of, almost’, which she considers to be related to verbal roots (1941b: 69–70), hi’la ‘move’ in the case of ‘almost’. In comparison to the root, the suffix is atonic and undergoes morphophonemic changes when attaching to a verb root; on the other hand, it follows the verb root and precedes other suffixes as a serial verb root does. In Tunica verb root serialization, inflection follows the rightmost root, e.g. (68) Tunica (Haas 1941b: 72) na’ra-ya’ka-po’-w-h rise-come-look-3ms-when ‘when he got up, came and had a look, …’

328 Johanna Mattissen Haas herself proposes verb root serialization (although not using this term) as the source construction from which the morphemes in question were grammaticalized. Thus, we have evidence as to a verbal origin of morphemes of degree. 3.7.2. Chronology Morphemes of chronology comprise affixes meaning ‘first’, ‘again’ etc., as in the example from Yimas: (69) Yimas (Foley 1991: 344) na-mp -mampi-kwalca-mpi-tay- cut 3sU-3dA-again-rise-SEQ-see-REMPST ‘they both got up again and saw him/got up and saw him again’ Kwakwala: Bininj G.: Greenlandic: Awtuw: Wichita: Yimas: Hupa:

in advance, one after the other, in row/piles, immediately again, at proper point in time, then/next, previously/already first, again, one after another, early, formerly first, for a long time, yet, again for a while, first first, again pursue/again, first

3.7.3. Focus (70) Nez Perce (Aoki 1970: 96) hi-i·q-cimi-se 3S-talk-only-SG ‘all he does is talk’ Greenlandic: Kwakwala: Bininj G.: Nez Perce: Chukchi: Awtuw:

just/only, exactly still/only, also, just just only only just

The Ontology and diachrony of polysynthesis

329

3.7.4. Quantifier Quantification marked on the verb is less frequent than expected: Bininj G.: Awtuw: Washo: Chukchi: Greenlandic: Tiwi:

collective, all, distributive all, much, half distributive/collective, distributive distributive distributive collective

Chronology is a domain which is seldom found as a verbal grammatical category outside of polysynthesis. Whereas degree turns up as comparatives, superlatives or elatives in more analytic languages, chronology and quantifiers may only be found in integrated form in lexicalized preverbs, e.g. in German, Slavic, and South Caucasian: German vor-kochen ‘prepare food in advance’, wieder-sehen ‘see again’, Russian pro-spat’ ‘sleep for some time’, Laz oo -pxer (in.middle.of.heap-sit) ‘I sit in the middle of a heap’ and in the Romance prefix re- ~ ri- (e.g. re-voir ‘see again’). Nevertheless, morphemes of the above-mentioned categories seem to be highly grammaticalized in the languages in question. There are, however, hints as to a possible evolution. Chukchi, for example, has morphemes expressing degree and focus, but also incorporates lexical adverbs into the verb, such as janot ‘first’ in the following example. (71) Chukchi (Spencer 1995: 456) … mggtkal/n qejwe n-janot-pker-g/an many.legged.one indeed 3sS-first-come-CND:3s ‘indeed, the many legged would come first’ In Nivkh, focus morphemes can be traced back to a verbal origin, with a homophonous verbalizer still extant. (72) Nivkh (Mattissen 2003: 82) scalar operator -para ~ -vara ~ -bara ‘yet’ -park ~ -vark ~ -bark

‘only’

verbalizer -varad ~ -barad ‘sth. is similar to sth.’ -vark ~ -park ~ -bark ‘sth. is contained exclusively’

330 Johanna Mattissen The example below shows the focus morpheme in a root serial-like construction, in the position a second verb root takes in verb root serialization, indicating the probable source construction. Verb root serialization is a productive mechanism in Nivkh (Mattissen 2003: ch. 6). (73) Nivkh (Panfilov 1965: 22)  Q: atak-a ! i oli-nonq-!pr-lo ? grandfather-VOC 2s stag-young-bring.in-whether ‘Grandpa, have you brought in a young stag?’  A: q aukra ñi i-nd-bark-d-ra no 1s 3sU-see-only-IND/NML-HILI ‘No, I only saw it.’ In more than one respect, Nivkh seems to be in a pre-stage of polysynthesis (Mattissen 2003: chs. 7, 10) and operates with lexical roots where polysynthetic languages have non-root bound morphemes. (74) Nivkh (Mattissen 2004c: 199) Integration into a complex verb form of material expressing in the form of phase non-root bound morphemes degree non-root bound morphemes focus non-root bound morphemes (grammaticalized from verb root serialization) modality non-root bound morphemes (grammaticalized from verb root serialization) manner verb root serialization location noun incorporation & verb root serialization quantification noun incorporation body part noun incorporation So Nivkh may tell us where polysynthesis came from, and polysynthesis shows where Nivkh might go (if it survived). 3.8. Further specialized classes The ontological domains which remain to be looked at are small but all the more interesting. These non-root bound morphemes have nominal trans-

The Ontology and diachrony of polysynthesis

331

lation equivalents in more analytic languages with fairly specific semantics – something we do not generally expect to be found in bound form because of pragmatic salience. Such morphemes even turn up in two compositionally polysynthetic languages alongside noun incorporation. The most surprising among these is the class of morphemes denoting humans. 3.8.1. Artefact and man-made environment (75) Tiwi (Osborne 1974: 50) --ma-t ili anta-jamanipa 1s-LINKER-with-basket-get.up ‘I got up and took a basket’ Quileute:

stone arrow-head, basket, bow/gun/weapon, bow, bow-string, sealing canoe, (fish) trap, knife, river canoe, arrow, canoe/vehicle, tool/instrument/utensil/artifice, place/location/dish/container, spear, fishing line, thing/ground, gill net, small basket, arrow point, food, decorated blanket, wall, hat, dress, fishing equipment, food to be eaten in a journey/lunch, dwelling/indoors, village, door, bed, man’s coat, strand of a rope, custom, roof, remainder/waste, food, platform, blanket/bed covers, stump, pack strap, village, bait, road, magic Tiwi: bark basket, bark canoe, dug-out canoe, bundle of spears, stick, bark shelter, grave post, firewood, bundle of firewood Spokane: cover/outside, bow, thing, clothes, vehicle, knob, sickness Wichita: settlement, garden, road Bininj G.: house, road, song Kwakwala: canoe/aboard, bow of canoe, roof Tarasco: midriff/crossroads, road/street

3.8.2. Human (76) Spokane (Carlson 1990: 76) kwén-l-t-n take-child-TR-1sS ‘I carry his child’

332 Johanna Mattissen Quileute: Spokane: Bininj G.: Tiwi:

spouse, shaman, wife, friend, canoe mate, guardian spirit, companion, female, child, youngster child, spouse, person spouse, child sick person, dead body

3.8.3. Animal (77) Tiwi (Osborne 1974: 50) a-wunt i-ma-ti puant-‘mu d ugong-sit 3s-DUR-with-d ‘he’s sitting with a dugong’ Tiwi:

Quileute: Spokane: Bininj G.:

turtle’s egg, uncooked small animal, raw shellfish, cooked shellfish, raw fish, raw crab, cooked crab, raw bird/flying fox, cooked bird/flying fox, raw meat, cooked meat/fish, cut up cooked meat, live wallaby, dead wallaby, raw possum, dugong/turtle, crocodile, buffalo, dog, wild honey flounder, dead whale, whale, egg of salmon, trout/smelt/ sucker, egg of bird cattle, roaster, animal, egg track, egg

3.8.4. Plant Tiwi:

fruit, raw xamia nut, roast xamia nut, prepared xamia nuts, raw yam, paper bark, stringy bark Quileute: wood, tree/log, plant/bush/tree, kelp, leaf, wood (combust.), grass/hay Spokane: fruit, wood, plant Wichita: grass, tall grass Kwakwala: branch/leaf/body hair 3.8.5. Other (generics) (78) Tiwi (Osborne 1974: 36) m arikuap-puipa ji-ni-mni-k-m 3sA.PST-distance-1sU-ee vening-dance-hit.eye ‘he came and hit me in the eye while I was dancing in the evening’

The Ontology and diachrony of polysynthesis

333

Quileute:

language, color, weather, foot-prints, odor, manner/way, size/ space, noise/sound, stone Tiwi: fresh water, morning, evening, ochre Bininj G.: by night, in the morning, contents Spokane: winter, sickness Haida: cause/thing Greenlandic: early These domains cover the things and creatures humans come across in their everyday lives, housing, vehicles, eggs being the most common categories. In the human domain, ‘spouse’ and ‘child’ keep recurring; Quileute has lots more. Interestingly, all these classes are best represented in Quileute and Spokane, in which a noun-verb distinction is less pronounced. Thus, translation equivalents of nouns of more analytic languages are found among the nonroot bound morphemes. However, there is a compositional language with a clear noun-verb distinction which has morphemes of this kind as well: Bininj Gun-wok (Evans 1996, 2002, 2003). Nevertheless, these “nominal-like” classes are small. Kwakwala (without a clear noun-verb distinction) is poorly equipped in these categories, although its overall number of morphemes is so large. Obviously, such concrete semantics are rather encoded otherwise, which makes Quileute all the more conspicuous. 4. The search for correlations As a first conclusion, we can state that – the number of morphemes in a language does not correlate with the categories represented in that language, compare Greenlandic and Kwakwala – evidence for the evolution of non-root bound morphemes is extant in only a few cases. The question is now whether there are any correlations between the nonroot morphemes and the subtypes of polysynthesis, compositional or affixal, with templatic, scope-ordered or mixed internal structure (as established in Mattissen 2003: ch.10) and of the onion, sandwich, or burdock evolutionary type.

334 Johanna Mattissen 4.1. Affixally polysynthetic languages On the whole, affixally polysynthetic languages exhibit the largest set of non-root bound morphemes both in number and variety. Affixal polysynthesis is best represented in North America and practically absent from Asia (except for Eskimo languages) and Oceania/Australia.

Kwakwala

Quileute

Klamath

Tarasco

Spokane

Tiwi

Tsimshian

Navaho

Washo

category

local

+

+

+

+

+

+

+

+

+

classifier

+

+

+

+

+

+

+

+

manner

+

+

+

+

+

+

+

phase

+

+

+

+

+

+

body part

+

+

+

+

+

+

artefact

+

+

+

+

+

+

medium

+

+

+

+

+

+

degree

+

+

+?

modal

+

+

+

motion

+

?

+

focus

+

chronology

+

reversative

+

Greenlandic

Table 5. Morphemes in affixal polysynthesis

+

+

+

+

+

+

+ +

+

+

+ +

+

+ +

+

+

Affixally polysynthetic languages share the lack of noun incorporation and verb root serialization by definition. As table 5 shows there is no hierarchy of domains. Morphemes of locality, classification, manner and phase, however, are manifest in almost all of them and jointly help to delimit the affixal type. These domains are also most characteristic of polysynthesis on the whole, especially in co-occurrence, as more analytic languages do not mark them on their verbs (except for phase, which tends to be encoded with auxiliaries).

The Ontology and diachrony of polysynthesis

335

As obvious from the individual sections above, Greenlandic (as a representative of the Eskimo languages) surprisingly does not exhibit local and classifier morphemes, i.e. the most common domains, but has manner and phase morphemes. Nevertheless, its general impression is clearly polysynthetic. A detailed look at the subtypes according to noun-verb distinction, internal structure and evolutionary type shows how heterogeneous the affixal class is: Lexical category distinction at least two distinct categories: Greenlandic, Klamath, Tiwi less-well delimited categories: Spokane (Kinkade 1983), Kwakwala (Boas 1947: 280), Tsimshian (Boas 1911d: 296) Internal structure scope-ordered: Greenlandic, Kwakwala templatic: Navaho, Tiwi, Washo, Tarasco, Hupa mixed: Klamath, Quileute, Spokane Evolutionary type onion: Navaho, Spokane, Tiwi, Hupa burdock: Klamath sandwich: Greenlandic, Washo, Tsimshian, Tarasco Languages with a less well-established distinction of lexical categories, e.g. nouns and verbs, have more non-root bound morphemes than those with a clear distinction. The internal structure and evolutionary type of the verb form, however, do not seem to correlate with the morpheme inventory. 4.2. Compositionally polysynthetic languages Compositionally polysynthetic languages share a lexical category distinction, as well as noun incorporation and verb root serialization by definition. The polysynthetic languages of Asia all belong to this type (ChukchoKamchadal, Ket, Ainu, arguably Nivkh, Munda languages) except for Eskimo languages. The Gunwinjguan languages of Australia and, in North America, Caddoan, Siouan, Iroquoian4, Tanoan and Uto-Aztecan languages belong here. ———–———————————————————————————— 4

As Sasse (1993a) shows for Cayuga, Iroquoian languages do not distinguish nouns and verbs, but two types of roots on the basis of their paradigm and behaviour with regard to incorporation.

336 Johanna Mattissen

Nez Perce

Chukchi

Cherokee

Lakhota

+

+

+

+

+

+

phase

+

+

+

+

+

+

+

?

manner

+

+

+

+

+

+

+

degree

+

+

+

+

chronology

+

+

artefact

+

+

+

classifier

+

+

+

focus

+

medium

+

body part

+

modal

Sora

Pano

+

Ket

Wichita

local

category

Ainu

Bininj Gun-wok

Table 6. Morphemes in compositional polysynthesis

+

+

+

+

+

+

+

+

+ + +

+

reversative

+

motion

+

+

+ +

+ +

The overall picture of the distribution of non-root bound morphemes is heterogeneous. There are no correlations visible to the subtypes of internal structure and evolutionary type. Internal structure scope: template: mixed: Evolutionary type onion: burdock: sandwich:

Chukchi, Pano Bininj Gun-wok, Wichita, Ket, Sora Lakhota, Nez Perce, Cherokee Wichita, Cherokee Ket Chukchi, Nez Perce

Local, phase and manner morphemes dominate among the compositionally polysynthetic languages. As the compositional languages have structural alternatives because of compositional mechanisms it is all the more suprising that Bininj Gun-wok and Wichita exhibit so many of the non-root bound domains.

The Ontology and diachrony of polysynthesis

337

In comparison to affixally polysynthetic languages, the domains of classifiers and body parts are underrepresented. Most probably body part morphemes are blocked by noun incorporation of body parts, which is a common feature. Nevertheless, Wichita and Bininj Gun-wok have both mechanisms, morphemes and incorporation. This is also true for Cherokee, but in this language the lexical category distinction is pretty weak (see King 1975: 38). Cherokee resembles more the affixally polysynthetic languages in its domains, and this is borne out by the language structure on the whole: Root incorporation is a residual mechanism operating on a small number of roots. The language furnishes good evidence for an evolution from compositional to affixal polysynthesis via the grammaticalization of erstwhile incorporated roots to non-root bound morphemes, as we saw in section 3.4. Within the compositional class as a whole, Ainu, Ket and Sora are very poorly equipped with non-root bound morphemes. Their delimitation from non-polysynthetic languages such as Swedish, Japanese or Laz hinges on the local, and the manner and degree morphemes, respectively (Mattissen 2003: ch. 10). I propose a diachronic dimension here as well: Whereas Swedish and Nivkh are on the verge of turning into a polysynthetic language (only the non-root bound morphemes are missing), Ainu and Ket may have only recently done so. Arranging the languages according to their stages on an evolutionary path, we get the following (“+” indicates the existence of non-root bound morphemes in that domain, “–” the absence of morphemes): Table 7. Evolutionary path Swedish

Nivkh

Ket

noun incorporation











verb root serialization











local

–

noun incorporation/ verb root serialization

+

+

+

verb root serialization

verb root serialization

verb root serialization

+

+

phase

–

+

–

+

+

classifier

–

–

–

–

+

manner

Chukchi Wichita

338 Johanna Mattissen In the transition from a language with compositional mechanisms (noun incorporation and verb root serialization) to a polysynthetic language, nonroot bound morphemes have to develop. In Nivkh, there are already modal affixes (for which verb root serialization can be shown to be the source construction; Mattissen 2003: ch. 6.4). Local and manner morphemes are not found, but local and manner information can be integrated into a complex verb form via noun incorporation and/or verb root serialization. A later grammaticalization to affixes is imaginable. The turning point then is between Nivkh and Ket due to the definition presented above, as Ket does have local non-root bound morphemes (some of which go back to nominal roots), but Nivkh does not. In Chukchi, there are already non-root bound morphemes in the local, manner, and phase domains (Kämpfe and Volodin 1995: 34–36) which exist alongside incorporation of nouns, verbs and adverbs. Wichita has more non-root bound morphemes than Chukchi, and has reduced its verb root serialization to only three verbs which can be the second verb (Rood 1976: 22), i.e. the mechanism grammaticalizes. This sketches a possible change into a polysynthetic language. Languages can, however, develop away from polysynthesis by exploiting extant analytic mechanisms and the fact that non-root bound morphemes are not always obligatory (see Mattissen 2003: ch. 3.9). Eskimo languages, for example, have a dummy base which, used in the position of a verbal or nominal root, hosts verbalizers (which are non-root bound morphemes) or inflectional morphemes, as is shown in the Greenlandic example below. (79) Greenlandic (Fortescue 1984: 83) a. ikinnguti-qar-puq friend-verbalizer.exist-IND.3s ‘s/he has a friend/friends’ b. ikinnguti-nik pi-qar-puq friend-INS dummy-verbalizer.exist-IND.3s ‘s/he has some friends’

(denominal verb)

(paraphrase)

Using such a dummy verb with a nominal undergoer (79b), as my Inuktitut consultant regularly did, circumvents the use of a denominal verb, formed from a nominal root with a verbalizer, as in (79a).

The Ontology and diachrony of polysynthesis

339

4.3. Polysynthetic languages with verb root serialization & without noun incorporation Another subtype of compositionally polysynthetic languages has verb root serialization as a compositional mechanism, but no noun incorporation. Interestingly, this is not due to a lack in lexical category distinction (a weak noun-verb distinction). In these languages, which are found in North and South America and in Papua New Guinea, the best represented domains are again local, modal and phasal non-root bound morphemes. Manner tends to be expressed via verb root serialization, e.g. in Yimas (Foley 1991), making verb root serialization the most probable source for manner morphemes. It is conspicuous that the categories of classifiers, body parts and other assumingly denominal material are practically absent, constituting an argument in favor of assuming that such morphemes develop from noun incorporation (which these languages do not have from the outset). Reversative morphemes (‘undoing’, see example (46a)) do not exist. Table 8. Morphemes in verb root serializing polysynthesis category

Awtuw

Maidu

Tunica

Pirahã

Yagua

Yimas

Tonkawa

local

+

+

+

+

+

+

+

modal

+

+

+

+

+

phase

+

+

+

+

+

manner

+

+

motion

+

+

degree

+

chronology

+

quantifier

+

+ + +

+ +

+

V+V

+

+ +

classifier

+

body part

+

The subtypes of internal structure and evolutionary type again seem irrelevant. Internal structure

scope: Yagua template: Tonkawa, Awtuw mixed: Yimas, Maidu

340 Johanna Mattissen Evolutionary type

onion: Awtuw, Tonkawa burdock: Yimas sandwich: Pirahã, Maidu

With the lack of noun incorporation and related domains the verb root serializing compositional type seems not to lie in the mainstream of the assumed development from compositional to affixal polysynthesis, but rather to be a side track or from a different source. 4.4. Polysynthetic languages with noun incorporation & without verb root serialization The mirror image of the preceding type is found in those languages which are compositionally polysynthetic because of noun incorporation and lack verb root serialization. This type is marginal, as is noun incorporation without other compositional features in general (see Mattissen 2003: 277), with only three languages spoken in North America in our sample. Therefore statements about it may not be representative. Sheer noun incorporation seems to be an unstable state for a language. Table 9. Morphemes in noun incorporating polysynthesis categories

Takelma

Blackfoot

Sarcee

local

+

+

+

phase

+

+

+

quantifier

+

+

modal

+

+

body part

+?

degree

+

These languages have a noun-verb distinction by definition. The most spread domains of non-root bound morphemes are locality and phase, followed by quantification and modal morphemes. Manner morphemes are lacking. This fits in well with the assumption that verb root serialization (absent from these languages) is the most probable source for manner morphemes. Again, reversative morphemes (‘undoing’) do not exist. The subtypes are represented as follows:

The Ontology and diachrony of polysynthesis

Internal structure scope: template: mixed: Evolutionary type onion: burdock: sandwich:

341

— Takelma, Sarcee Blackfoot Blackfoot, Sarcee Takelma —

Again, this type seems not to lie in the mainstream of the assumed development from compositional to affixal polysynthesis. 5. Conclusion Non-root bound morphemes have been singled out as a necessary and important parameter of polysynthesis in Mattissen (2003: ch. 10, 2004c). In order to try to concretize this parameter, we have studied their variety in polysynthetic languages and reached the conclusion that there is no homogeneous distribution of morphemes, nor any hierarchy among the different semantic domains. In absolute numbers, the largest class of non-root bound morphemes is found in the local domain, body parts are second, followed by classifiers, medium/natural force, and manner. Local non-root bound morphemes are the largest class in terms of number and variety of morphemes, both crosslinguistically and within the single languages, and in terms of dissemination. They need not be, however, the largest class in every language in which they exist. Tarasco, for instance, is best equipped in body part morphemes (cf. Friedrich 1971: 15–16). Among the languages with the overall greatest inventories of non-root bound morphemes, Kwakwala, Tshimshian and Klamath stand out for a most impressive set of local morphemes, whereas Greenlandic has none. Both individuocentric and geocentric local systems are manifest in polysynthetic languages, there is thus no characteristic system. A hierarchy of local relations with respect to the integration of non-root bound morphemes is not borne out. In terms of dissemination, the most important domains, however, are localities, classifiers, manner, phase and modals. They can be considered to jointly characterize polysynthesis, but this does not mean that they are extant in each and every polysynthetic language. These main domains correlate to a low degree with the word-formational subtypes of polysynthesis. Although affixal, fully compositional, verb-root-

342 Johanna Mattissen serializing compositional, and noun-incorporating compositional languages turned out to be fairly heterogeneous with respect to the semantic domains of their non-root bound morphemes, local morphemes are the strongest category in all of them. They are a fairly good indicator of polysynthesis, although not a necessary nor sufficient parameter, as they cannot be delimited from preverbs. Greenlandic, for instance, has no local morphemes. (80) Word-formational type affixal compositional verb root serializing noun incorporating

Semantic domains of non-root bound morphemes local, manner, classifier, phase local, (phase) local, modal, phase local, phase

Internal structure (scope-ordered, templatic or mixed, see Mattissen 2003, 2004c) and evolutionary type (onion, sandwich, burdock, see section 2), however, do not show any correlations with the semantic domains of the non-root bound morphemes. The integration of classifiers, manner, phase and modality in non-root form is not common enough among polysynthetic languages to define this structural type on the whole, and at least phase (e.g. in Nivkh, Mattissen 2003: 18) and modality (e.g. in Japanese, Mattissen 1995a) non-root bound morphemes are not exclusive to polysynthesis. It proves difficult to establish an evolutionary path between the different subtypes. A development seems most plausible between fully compositional and affixal polysynthesis via the grammaticalization of roots in noun incorporation and verb root serialization to non-root bound morphemes (see Mattissen 2004b for details). Whereas most polysynthetic languages exhibit morphemes with an ambiguous status or allow some reconstruction of sources, Greenlandic, with a presumably quite old polysynthesis, shows no traces of any evolution. Otherwise, there is a connection visible between body part and classifier morphemes to noun incorporation, which are absent from verb root serializing languages (without noun incorporation), and between manner, degree, phasal and modality morphemes to verb root serialization. Local morphemes may have both source constructions, which may be a reason for their multitude. Because of these diachronic links, a strict separation of affixal and compositional polysynthesis under two different headings (rather than under polysynthesis) is neither feasable nor does it make sense.

The Ontology and diachrony of polysynthesis

343

This means that a noun-verb distinction is an important precondition in the early development of polysynthesis. Indeed languages do exist which have noun incorporation and verb root serialization without non-root bound morphemes, e.g. Swedish or Yucatec Maya, and languages with both compositional mechanisms and non-root bound morphemes, e.g. Ket or Sora, which may be positioned before (Swedish, Maya) and after (Ket, Sora) the turning point to polysynthesis, respectively (see also Mattissen 2003, 2004c). In the later evolution, a lexical category distinction may vanish, with non-root bound morphemes “taking over”. Kwakwala, Spokane (and indeed Wakashan and Salishan languages generally), Tsimshian and Quileute do not distinguish two main lexical categories, but they dispose of an impressive inventory of non-root bound morphemes. Kwakwala, Tsimshian and Quileute are (besides Greenlandic) the polysynthetic languages with the most extensive set of suffixes. Quileute and Spokane are well-equipped in domains such as body parts, artefacts, human environment and even humans, i.e. classical “nominal” semantic domains. The exception is again Greenlandic (Eskimo), with a multitude of nonroot bound morphemes and a sound and stable noun-verb distinction. Greenlandic is different from most polysynthetic languages in lacking local, body part, classifier and other morphemes (suspected of developing from noun incorporation, which is absent from Greenlandic). However, Greenlandic does not have verb root serialization either, but disposes of manner, degree, phasal and modal morphemes which can be shown to derive from verb root serialization in other languages. Greenlandic polysynthesis therefore must be quite old. Iroquoian languages, on the contrary, seem to be at the turning point between compositional and affixal polysynthesis, their number of non-root bound morphemes, however, is small. These observations cover only one cycle of grammaticalization. Further detailed studies are needed to try to reconstruct the paths along which the states of affairs described here came about. I hope to have shown, however, that because of their heterogeneous designs a differentiating approach to polysynthetic languages is needed, and that lexical category distinction is intimately linked to polysynthesis.

344 Johanna Mattissen Abbreviations 1.A A ABS AGT AOR AP ASP BEN

C CHRON CISL CND COLL CPL CST CTRL

d DECL DEF DISTR DU DUR EMPH

epV ERG EVID

f FAC FOC FUT HAB HILI INDEF ICPL IMPF INCORP IND INF INS IRR

1st singular agent (etc.) affixal polysynthesis absolutive agent aorist antipassive aspect beneficiary compositional polysynthesis chronology (non-root bound m.) cislocative conditional collective completive causative control dual declarative definite distributive dual durative emphatic epenthetic vowel ergative evidential feminine factual focus (non-root bound m.) future habitual highlighting focus indefinite incompletive imperfective incorporee indicative infinitive instrumental irrealis

ITER ITR LOC LOCL

m MED

n NEG NI NML NSG

p PART PASS PAT PL PORT POSS POT PRF PROG PRS PRV PST PTCL PUNC QUO REC RED REFL REMPST REVERS

s S SG SEQ SRFL STAT SUB T/A

iterative intransitive locative local (non-root bound m.) masculine medium neuter negation noun incorporating polysynthesis nominalization non-singular plural particle passive patient plural portative possessor potential perfect progressive present preverb past participle punctual quotative recipient reduplication reflexive remote past reversative singular “subject” singular sequential semireflexive stative subordination tense/aspect

The Ontology and diachrony of polysynthesis TR TRSL TRV U

transitive translocative transitivizer undergoer

UDG VRS VOC

345

undergoer verb root serializing polysynthesis vocative

References Allen, Barbara J., Donna B. Gardiner and Donald G. Frantz 1984 Noun incorporation in Southern Tiwa. International Journal of American Linguistics 50: 292–311. Andrade, Manuel J. 1933 Quileute. Extract from Handbook of American Indian Languages III, 151–292. New York: Columbia University Press. Andrews, J. Richard 1975 Introduction to Classical Nahuatl. Austin: University of Texas Press. Aoki, Haruo 1970 Nez Perce Grammar. Berkeley: University of California Press. Asher, R. E. and J. M. Y. Simpson (eds.) 1994 The Encyclopedia of Languages and Linguistics. 10 vols. Oxford etc.: Pergamon Press. Baker, Mark C. 1996 The Polysynthesis Parameter. New York etc.: Oxford University Press. Barker, Muhammad A. R. 1964 Klamath Grammar. Berkeley: University of California Press. Bickel, Balthasar 1994 Spatial operations in deixis, cognition and culture: Where to orient oneself in Belhare. Working Paper of the Cognitive Anthropology Research Group No. 28. Nijmegen: MPI for Psycholinguistics. Bisang, Walter 1995 Verb serialization and converbs – differences and similarities. In Converbs in Cross-Linguistic Perspective, M. Haspelmath and E. König (eds.), 137–188. Berlin /New York: Mouton de Gruyter. Blair, Robert W. 1964 Yucatec Maya Noun and Verb Morpho-Syntax. Ann Arbor: UMI. Blankenship, Barbara 1997 Classificatory verbs in Cherokee. Anthropological Linguistics 39: 92–110. Boas, Franz (ed.) 1911/22 Handbook of American Indian Languages. 2 Vols. (= Bureau of American Ethnology Bulletin 40). Washington: Government Printing Office.

346 Johanna Mattissen Boas, Franz 1911a 1911b 1911c 1911d 1947

Introduction. In Boas (ed.). Part 1: 1–83. Tsimshian. In Boas (ed.). Part 1: 283–422. Kwakiutl. In Boas (ed.). Part 1: 423–557. Chinook. In Boas (ed.). Part 1: 559–677. Kwakiutl Grammar – with a Glossary of the Suffixes. New York: AMS Press. Boas, Franz and Ella Cara Deloria 1979 Dakota Grammar. Washington: National Academy of Science. Broadbent, Sylvia M. 1964 The Southern Sierra Miwok Language. Berkeley: University of California Press. Brown, Penelope and Stephen C. Levinson 1991 ‘Uphill’ and ‘Downhill’ in Tzeltal. Working Paper of the Cognitive Anthropology Research Group No. 7. Nijmegen: MPI for Psycholinguistics. Carlson, Barry F. 1972 A Grammar of Spokan: A Salish Language of Eastern Washington. Working Papers in Linguistics 4 (4). Honolulu: University of Hawaii. 1990 Compounding and lexical affixation in Spokane. Anthropological Linguistics 32: 69–82. Chappell, Hilary and William McGregor (eds.) 1996 The Grammar of Inalienability. Berlin/New York: Mouton de Gruyter. Cook, Eung-Do 1984 A Sarcee Grammar. Vancouver: University of British Columbia Press. DeLancey, Scott 1991 Chronological strata of suffix classes in the Klamath verb. International Journal of American Linguistics 57: 426–445. 1999 Lexical prefixes and the bipartite stem construction in Klamath. International Journal of American Linguistics 65: 56–83. Derbyshire, Desmond C. and Geoffrey K. Pullum (eds.) 1986–98 Handbook of Amazonian Languages. 4 vols. Berlin /New York: Mouton de Gruyter. De Reuse, Willem J. 1994 Noun incorporation in Lakhota (Siouan). International Journal of American Linguistics 60: 199–260. Dixon, R. M. W. and Alexandra Y. Aikhenvald (eds.) 1999 The Amazonian Languages. Cambridge: Cambridge University Press. Dixon, Roland B. 1911 Maidu. In Boas (ed.). Part 1: 679–734. Drossard, Werner 1992 Lokale Relationen: vom Einfachen (Topologischen) zum Komplexeren (Dimensionalen). Sprachliche Reflexe einer psycholinguistischen Erkenntnis. akup 86: 16–56.

The Ontology and diachrony of polysynthesis 1997

347

Polysynthesis and polysynthetic languages in comparative perspective. In Proceedings of Linguistics and Phonetics 1996, B. Palek (ed.), 251–264. Prague: Charles University Press. 2002 Paleosiberian Ket as a polysynthetic language (with special reference to complex verbs). In Evans and Sasse (eds.), 223–256. Egesdal, Steven M. 1981 Some ideas of the origin of Salish lexical suffixes. Working Papers in Linguistics (University of Hawaii) 13 (2): 3–19. Evans, Nicholas D. 1996 The syntax and semantics of body part incorporation in Mayali. In Chappell and McGregor (eds.), 65–109. 2002 The true status of grammatical object affixes: evidence from Bininj Gun-wok. In Evans and Sasse (eds.), 15–50. 2003 Bininj Gun-wok: A Pan-dialectal Grammar of Mayali, Kunwinjku and Kune. 2 vols. Canberra: Pacific Linguistics. Evans, Nicholas D. and Hans-Jürgen Sasse (eds.) 2002 Problems of Polysynthesis. Berlin: Akademie. Everett, Daniel L. 1986 Pirahã. In Derbyshire and Pullum (eds.), vol. 1: 200–325. Feldman, Harry 1986 A Grammar of Awtuw. Canberra: Australian National University. Foley, William A. 1991 The Yimas Language of New Guinea. Stanford: Stanford University Press. Fortescue, Michael 1983 A Comparative Manual of Affixes for the Inuit Dialects of Greenland, Canada, and Alaska. (Meddelelser om Grønland, Man and Society 4.) Kopenhagen: Nyt Nordisk Forlag. 1984 West Greenlandic. London: Croom Helm. 1990 Basic structures and processes in West Greenlandic. In Arctic Languages: An Awakening, D. R. F. Collis (ed.), 309–332. Paris: UNESCO Publication. 1994 Morphology, polysynthetic. In Asher and Simpson (eds.), vol. 5: 2600–2602. Foster, Mary LeCron 1969 The Tarascan Language. Ann Arbor: UMI. Friedrich, Paul 1971 The Tarascan Suffixes of Locative Space: Meaning and Morphotactics. The Hague: Mouton. Gatschet, Albert 1890 The Klamath Indians of Southwest Oregon. (Contribution to North American Ethnology, vol. 2). Washington DC: Government Printing Office.

348 Johanna Mattissen Gerdts, Donna B. 1996 Incorporation. In Handbook of Morphology, A. J. Spencer and A. M. Zwicky (eds.), 84–100. Oxford: Blackwell. Goddard, Ives (ed.) 1996 Handbook of North American Indians, vol. 17. Languages. Washington: Smithonian Institution. Goddard, Pliny Earle 1911 Athapascan (Hupa). In Boas (ed.). Part 1: 85–158. Golla, Victor 1996 Sketch of Hupa, an Athapaskan language. In Goddard (ed.), 364– 389. Haas, Mary R. 1941a Noun incorporation in the Muskogean languages. Language 17: 311– 315. 1941b Tunica. Extract from Handbook of American Indian Languages IV. New York: Augustin. Haeberlin, Hermann K. 1921–23 Notes on the composition of the verbal complex in Haida. International Journal of American Linguistics 2: 159–162. Haugen, Einar 1969 The semantics of Icelandic orientation. In Cognitive Anthropology, Stephen A. Tyler (ed.), 330–342. New York: Holt, Rinehart and Winston. Heath, Jeffrey 1978 Ngandi grammar, texts, and dictionary. Canberra: Australian Institute of Aboriginal Studies. Hercus, Luise A. 1994 A Grammar of the Arabana-Wangkangurru Language, Lake Eyre Basin, South Australia. Canberra: Australian National University. Hewitt, B. George 1989 Abkhaz. In The Indigenous Languages of the Caucasus 2, George Hewitt (ed.). Delmar: Caravan. Hill, Deborah 1993 Finding your way in Longgu. Working Paper of the Cognitive Anthropology Research Group, no. 21. Nijmegen: MPI for Psycholinguistics. Hoijer, Harry 1933 Tonkawa. Extract from Handbook of American Indian Languages III, 289–311. New York: Columbia University Press. Jacobsen, William H. 1980 Washo bipartite verb stems. In American Indian and Indoeuropean Studies. Papers in Honor of Madison S. Beeler, K. Klar, M. Langdon and S. Silver (eds.), 85–99. The Hague: Mouton.

The Ontology and diachrony of polysynthesis

349

Johnston, J. R. and D. J. Slobin 1979 The development of locative expressions in English, Italian, SerboCroatian, and Turkish. Journal of Child Language 6: 529–545. Kämpfe, Hans-Rainer and Alexander P. Volodin 1995 Abriß der Tschuktschischen Grammatik. Wiesbaden: Harrassowitz. King, Duane Harold 1975 A Grammar and Dictionary of the Cherokee Language. Ann Arbor: UMI. Kinkade, M. Dale 1983 Salish evidence against the universality of ‘noun’ and ‘verb’. Lingua 60: 25–39. Kozinsky, Isaac, Vladimir Nedjalkov and Maria Polinskaja 1988 Antipassive in Chukchee: oblique object, object incorporation, zero object. In Passive and Voice, M. Shibatani (ed.), 651–706. Amsterdam: John Benjamins. Krejnovi, E. A. 1982 Expression de l’orientation spatiale dans la langue Nivx. Études mongoles et sibériennes 13: 99–116. Kroeber, A. L. 1907 The Washo language of East Central California and Nevada. University of California Publications in American Archaeology and Ethnology 4 (5). Berkeley: University Press. Laughren, Mary 1978 Directional terminology in Warlpiri. Working Papers in Language and Linguistics 8. Launceston: Tasmanian College of Advanced Education. Levinson, Stephen C. 1992 Language and Cognition: The cognitive consequences of spatial description in Guugu Yimithirr. Working Paper of the Cognitive Anthropology Research Group No. 13. Nijmegen: MPI for Psycholinguistics. 1996 The Body in Space: Cultural differences in the use of body-schema for spatial thinking and gesture. Working Paper of the Cognitive Anthropology Research Group No. 39. Nijmegen: MPI for Psycholinguistics. Lewis, David 1976 Observations on route finding and spatial orientation among the Aboriginal peoples of the Western Desert region of Central Australia. Oceania 46: 249–282. Lichtenberk, Frantisek 1983 Grammar of Manam. Honolulu: University of Hawaii. Loos, Eugene E. 1999a Pano. In Dixon and Aikhenvald (eds.), 227–250. 1999b Grammatical Sketch of Capanawa. Ms., SIL.

350 Johanna Mattissen Lucassen, Wim 1985 On the verbal structure of Abkhaz. In Relational Typology, Frans Plank (ed.), 257–267. Berlin / New York: Mouton de Gruyter. Majewicz, Alfred F. 1980 Incorporation, polysynthesis, and typology. Lingua Posnaniensis 23: 137–144. Mattissen, Johanna 1995a Das Nomen im Japanischen: Abgrenzung und Subklassifizierung. Working Paper SFB 282 ‘Theorie des Lexikons’ No. 65, University of Düsseldorf. 1995b Verbmorphologie. In Das Mut‡afi-Lazische, S. Kutscher, J. Mattissen and A. Wodarg (eds.), 45–81. Arbeitspapier Nr. 24 (Neue Folge). Institut für Sprachwissenschaft der Universität zu Köln. 2002 Dependent-Head synthesis in Nivkh – with an outlook on polysynthesis in the Far Northeast. In Evans and Sasse (eds.), 136–166. 2003 Dependent-Head Synthesis in Nivkh. A Contribution to a Typology of Polysynthesis. Amsterdam: Benjamins. 2004a Structural types of complex noun forms and their evolution. In Dimensionen und Kontinua: Beiträge zu Hansjakob Seilers Universalienforschung, W. Premper (ed.), 35–56. Berlin: Akademie. 2004b The missing link between different types of polysynthetic languages. In Chicago Linguistic Society 38, vol. II: 386–399. 2004c A structural typology of polysynthesis. Word 55: 189–216. in prep. Vertikal und Quer – Lokalisation im Lasischen von Art‡aeni. To appear in Georgica. Mithun, Marianne 1984 The evolution of noun incorporation. Language 60: 847–893. 1988 System-defining structural properties in polysynthetic languages. Zeitschrift für Phonetik, Sprachwissenschaft und Kommunikationsforschung 41: 442–452. 1999 The Languages of Native North America. Cambridge: Cambridge University Press. Mithun, Marianne and Greville G. Corbett 1999 The effect of noun incorporation on argument structure. In Boundaries of Morphology and Syntax, L. Mereu (ed.), 49–71. Amsterdam: Benjamins. Newman, Stanley 1967 Classical Nahuatl. In Handbook of Middle American Indians. Vol. 5, N. A. McQuown (ed.), 179–199. Austin: University of Texas Press. Osborne, Charles R. 1974 The Tiwi Language. Canberra: Australian Institute of Aboriginal Studies.

The Ontology and diachrony of polysynthesis

351

Ozanne-Rivierre, Françoise 1987 L’expression linguistique de l’orientation dans l’espace: Quelques examples océaniens. Cahiers du LACITO 2: 129–155. Paris: LACITO-CNRS. 1997 Spatial references in New Caledonian languages. In Referring to Space – Studies in Austronesian and Papuan Languages, G. Senft (ed.), 83–100. Oxford: Clarendon Press. Panfilov, Vladimir Z. 1962/65 Grammatika nivxskogo jazyka [Nivkh Grammar]. 2 Vols. Moscow/ Leningrad: Nauka. Payne, Doris L. 1990 Morphological characteristics of Lowland South American languages. In Amazonian Linguistics. Studies in Lowland South American Language, D. Payne (ed.), 213–241. Austin: University of Texas Press. Payne, Doris L. and Thomas E. Payne 1990 Yagua. In Derbyshire and Pullum (eds.), vol. 2: 249–474. Pederson, Eric 1993 Geographic and manipulable space in two Tamil linguistic systems. In Spatial Information Theory, A.U. Frank and I. Campari (eds.), 294–311. Berlin: Springer. Pederson, Eric, Eve Danziger, Stephan Levinson, Sotaro Kita, Gunter Senft and David Wilkins 1998 Semantic typology and spatial conceptualization. Language 74: 557– 589. Pinnow, Heinz-Jürgen 1966 A comparative study of the verb in the Munda languages. In Studies in Comparative Austroasiatic Linguistics, N. H. Zide (ed.), 96–193. The Hague: Mouton. Rice, Keren D. 1989 Slave. Berlin /New York: Mouton de Gruyter. Rood, David S. 1976 Wichita Grammar. New York: Garland. 1996 Sketch of Wichita. In Goddard (ed.), 580–608. Sapir, Edward 1911 The problem of noun incorporation in American languages. American Anthropologist 13: 250–282 (Reprint in: W. Bright (ed.). 1990. The Collected Works of Edward Sapir V, 27–59. Berlin /New York: Mouton de Gruyter). 1922 The Takelma language of Southwest Oregon. In Boas (ed.), Part 2: 1–297. Sasse, Hans-Jürgen 1993a Das Nomen – eine universale Kategorie? Sprachtypologie und Universalienforschung 46: 187–221.

352 Johanna Mattissen 1993b

Syntactic categories and subcategories. In Syntax. An International Handbook of Contemporary Research, J. Jacobs, A. von Stechow, W. Sternefeld and T. Vennemann (eds.), 646–686. Berlin /New York: Mouton de Gruyter. 2002 Lexicological and lexicographic problems of word families in Cayuga. In: Evans and Sasse (eds.), 203–221. Seiler, Hansjakob 1977 Cahuilla Grammar. Banning: Malki Museum Press. Senft, Gunter 1992 Introduction. In Referring to Space – Studies in Austronesian and Papuan Languages, G. Senft (ed.), 1–38. Oxford: Clarendon Press. 2001 Frames of spatial reference in Kilivila. Studies in Language 25: 521– 555. Shibatani, Masayoshi 1990 The Languages of Japan. Chicago: Cambridge University Press. Skorik, Pjotr J. 1961/77 Grammatika ukotskogo jazyka [Grammar of Chukchi]. 2 Vols. Moscow: Izdatel’stvo Akademii Nauk. Spencer, Andrew 1995 Incorporation in Chukchi. Language 71: 439–489. Steever, Sanford B. 1986 Morphological convergence in the Khondmals: (Pro)nominal incorporation. In South Asian Languages. Structure, Convergence and Diglossia, Bh. Krishnamurti (ed.), 270–285. New Delhi: Motilal Banarsidass. Sullivan, Paul R. 1984 Noun incorporation in Yucatec Maya. Anthropological Linguistics 26: 138–160. Swanton, John R. 1911 Haida. In Boas (ed.), Part 1: 205–282. Talmy, Leonard 1985 Lexicalization patterns: Semantic structure in lexical forms. In Language Typology and Syntactic Description, T. Shopen (ed.), vol. III: 57–149. Cambridge: Cambridge University Press. Tamura Suzuko 2000 The Ainu Language. Tokyo: Sanseido. Taylor, Allan Ross 1969 A Grammar of Blackfoot. Ann Arbor: University of Michigan. Watkins, Laurel J. 1984 A Grammar of Kiowa. Lincoln: University of Nebraska Press. Werner, Heinrich 1997 Die ketische Sprache. Wiesbaden: Harrassowitz.

The Ontology and diachrony of polysynthesis

353

Widlok, Thomas 1996 Topographical gossip and the indexicality of Hai
om environmental knowledge. Working Paper of the Cognitive Anthropology Research Group No. 37. Nijmegen: MPI for Psycholinguistics. Wilkins, David P. 1989 Mparntwe Arrernte (Aranda). PhD thesis. Canberra: Australian National University. Young, Robert W. and William Morgan 1980 The Navajo Language: A Grammar and Colloquial Dictionary. Albuquerque: University of New Mexico Press.

Linguistic perspectives on morphological processing Harald Clahsen

1. Introduction What’s the use of theoretical linguistics for understanding human language processing? Looking at the vast majority of studies on morphological processing, the impression one will get is: not much. For example, casual inspection of the major international journals in experimental psycholinguistics (Cognition, Language and Cognitive Processes, Brain and Language, Journal of Memory and Language) reveals hardly any reference to notions and concepts from theoretical morphology. Instead, it is claimed that morphological distinctions have little effect on performance and that other (non-morphological) factors determine how a morphologically complex word is processed (see e.g. Raveh & Rueckl 2000). Connectionist models represent one of the most popular approaches, and most connectionist models are indeed ‘eliminative’ (Marcus 2001) in the sense that all inflected or derived words are said to be stored within a single associative system without any explicit representation of their morphological structure (see e.g. Bybee 1995; Rumelhart & McClelland 1986; Sereno & Jongman 1997; Joanisse & Seidenberg 1999; among others). Instead, these studies implement networks that represent the mapping relationship between the base of a word and its inflected or derived form through associatively linked orthographic, phonological and semantic codes. Whether morphological structure is indeed secondary, i.e. derivable from phonology and semantics, and as such dispensable is ultimately an empirical question. What is unfortunate, however, is how naïve some of the psycholinguistic work is from a linguistic perspective, with the result that data from otherwise carefully designed and empirically demanding studies are sometimes hard to interpret. A famous example is the phonological encoding system employed by Rumelhart & McClelland’s (1986) pattern-associator network of past-tense formation in English, in which phonological form is represented as a set of 3-character sequences, so-called ‘Wickelfeatures’ (after a scheme developed by Wickelgren 1969). The word strip, for example, takes the form {#st, str, tri, rip, ip#} in their notation. Pinker & Prince

356 Harald Clahsen (1988: 96ff.) show that this phonological encoding system is in principle incapable of handling phonological processes (e.g. reduplication) and appropriate phonological generalizations. In terms of Wickelfeatures it is, for example, as costly to change brid into something phonetically distant such as bald or blud as it is to turn brid to bird, the latter of which is the actual phonological change the word bird underwent in the history of English. On the other hand, there are no constraints or mechanisms in the trigram encoding system that would restrict phonological changes to those that are attested in natural languages. Pinker & Prince point out that the inadequacies of their encoding system have serious implications for Rumelhart & McClelland’s own simulations and the grand conclusions drawn from these simulations. Essentially, the pattern associator simply does not work without this particular encoding system, and if this system is ‘categorically the wrong thing’ (Pinker & Prince 1988: 101), then the simulations that depend on it cannot be right either. Consider another example of a psycholinguistic study in which ignoring basic linguistic properties of the morphological system under study has led to a dubious interpretation of the empirical data. Orsolini, Fanari & Bowles (1998) investigating past-tense and past participle formation in Italian children found that most errors were with 2nd conjugation forms, while 1st conjugation forms were rarely overapplied. For example, instead of the correct (but idiosyncratic) form cadde ‘fall-3sg. past’ for the 2nd conj. verb cadere, the children produced *cadè, which is the more common past-tense form of verbs of the 2nd conjugation, as for example in temere ‘to fear’ ~ temè ‘fear3sg. past’. Orsolini, Fanari & Bowles (1998) assumed that the 2nd conjugation, being non-productive, has inflected verb forms which are all listed, while 1st and 3rd conjugations, being productive, have rule-based inflected forms. Given these assumptions, it looks as if non-productive (listed) forms (i.e. 2nd conj. forms) rather than rule-based (default) forms are overapplied by Italian children. The problem in Orsolini et al’s treatment of Italian inflection is that it confuses the properties of conjugation membership with those of inflectional processes, even though in neither case is there a direct mapping between the two sets of properties in Italian. It is true that only the first conjugation shows unrestricted productivity in Italian and can accommodate any type of root, but verbs belonging to 2nd or 3rd conjugation classes may still be subject to rule-based inflectional processes in the same way as 1st conjugation verbs. Thus, contrary to Orsolini, Fanari & Bowles’ (1998) claims, an inflectional error in which a regular 2nd conjugation past-tense form is produced instead of the required irregular one (e.g. *cadè

Linguistic perspectives on morphological processing

357

instead of cadde) is a straightforward overregularization and does not represent an overextension of a non-productive form. Say & Clahsen (2002) reanalyzed their child language data and found that almost all of the errors were indeed overgeneralizations of regular stems and regular affixation rules to irregular verb forms. The fact that most errors were seen in 2nd conjugation simply reflects the fact that most of the verbs in Italian that have irregular past-tense forms are members of this class. It would not be difficult to come up with more examples of processing models that are based on oversimplified views of word structure and paradigm structure and claim that the recognition and production of inflected or derived words does not involve more than phonological (and semantic) pattern matching. Against this background, the purpose of the present chapter is to show that structured morphological representations from different morphological subsystems play a crucial role in language processing and are essential for understanding results from psycholinguistic experiments. Specifically, we argue that morphological processing is affected by structural representations at the level of (i) morphological types (roots, stems), (ii) inflectional versus derivational processes, and (iii) inflectional paradigms. 2. The Correspondence Hypothesis: A framework for linguistically informed psycholinguistic research Given the view that knowledge of language (and in particular grammatical knowledge) is represented in a cognitive system which is part of any normal human being’s mental or psychological structure, the question arises of how the mentally represented grammar is employed in language processing. The strongest and (to me) most interesting view concerning the grammar-processing relation is the correspondence hypothesis (originally proposed by Miller & Chomsky 1963) according to which the mental grammar is directly used in language processing. This means that grammatical rules and principles are mentally represented and that in recognition and production the language processor constructs such representations using the normal structures and operations of the grammar. The appeal of the correspondence hypothesis is that it provides a parsimonious and straightforward account of how grammatical knowledge and processing are related: The parser is said to make basically the same distinctions as the grammar (Jackendoff 1997; Phillips 1996). Connectionist models in the tradition of Rumelhart &

358 Harald Clahsen McClelland (1986) and other associative models of language (e.g. Bybee 1995) also claim psychological and even neurological plausibility for their accounts while, at the same time, denying any direct role of morphological structure for handling inflected or derived words. The controversial question then is whether the language processor constructs direct mappings involving orthographic, phonological and semantic codes or whether it also relies on structured morphological representations akin to those posited by linguists. In the domain of morphology, much psycholinguistic work has focused on the contrast between regular and irregular inflection. This research has employed different psycholinguistic methods and techniques and has led to a number of consistent and replicable experimental results; see Clahsen (1999), Marslen-Wilson & Tyler (1998) and Pinker (1999), for review. Consider, for example, results from lexical decision experiments. The subjects’ task is to discriminate between existing words (that have been encountered before) and nonce words (that have never been encountered before), e.g. house vs. nouse. The lexical decision task encourages subjects to rely on lexical entries stored in memory, and word-based effects have indeed consistently been obtained with these experiments. In particular, lexical decision times were found to be affected by word frequency: Subjects take less time to decide that high-frequency items are existing words than they do for low-frequency items (see Balota 1994 for review). This can be conceived of as an effect of lexical memory: As memory traces get stronger with additional exposure, high-frequency entries can be more readily accessed than low-frequency ones. Clahsen, Eisenbeiss & Sonnenstuhl (1997), for example, obtained processing differences between regularly and irregularly inflected word forms in visual lexical decision experiments. Shorter lexical decision times were found for irregular German participles with a high word-form frequency than for those with a low word-form frequency. For regular participles, however, there was no corresponding word-form frequency effect. These results suggest that only irregular participles access full-form entries in the mental lexicon; see section 3 for a more detailed discussion of these results. Another commonly used experimental task are priming experiments. In such studies, stimulus pairs are presented to subjects, and the researcher manipulates the relation between them. In a cross-modal immediate repetition priming experiment, subjects hear a spoken prime immediately followed by a visually presented target form for which they are required to make a word/non-word decision. It has been argued that cross-modal priming is particularly sensitive to the internal structure of a morphologically complex

Linguistic perspectives on morphological processing

359

word. Evidence for this comes from previous studies of regular and irregular inflection. The baseline condition is identical repetition, e.g. walk followed by walk. In this case, access to the lexical entry of the second presentation is facilitated, reflected in faster response times relative to a control condition in which walk is preceded by an unrelated item, e.g. smile. Sonnenstuhl, Eisenbeiss & Clahsen (1999) investigating priming effects in German past participles found an equivalent amount of facilitation when the prime was a regularly inflected participle and the target contained its corresponding root. That is, a regular participle form fully primes its corresponding root. Irregular participle forms, on the other hand, primed their stems less effectively than did the stem itself. Importantly, these regular/irregular priming differences were found despite the fact that the phonological and orthographical distance between the participle forms and their corresponding stem forms was identical for both regular and irregular verbs. For example, an irregular participle such as geschlafen ‘slept’ is as similar to its root (schlaf-) as a regular participle (e.g. geöffnet ‘opened’) is to its root (öffn-), and despite these formal similarities the regular form produced full priming, and the irregular one did not. The simple idea (see e.g. Rueckl et al. 1997) that phonological and/or orthographical overlap between the prime and the target accounts for the observed priming differences (which may apply to corresponding experiments on English (compare walked
walk vs. taught
teach) cannot explain the German priming these results. On the other hand, the findings on regular and irregular inflection and the contrast between built and frozen forms do not necessarily rule out a wholeword conception of the lexicon, provided that lexical entries are not simply encoded in terms of phonological (and semantic) properties, but incorporate morphologically structured representations for regulars, and not (or less so) for irregulars. The priming differences seen in Sonnenstuhl, Eisenbeiss & Clahsen’s (1999) experiments correspond to this contrast in morphological structure. Given that regular -t participles are morphologically decomposed forms (e.g. [ge-kauf-t] ‘bought’) they can directly access the root, hence the full priming effects in the experiments. If, on the other hand, irregularly inflected participle forms do not have such structured representations, they cannot directly access the root, and hence the reduced priming effects in the experiments. These results suggest that structural differences between built and frozen forms converge with the way they are segmented by the speaker-hearer during online production and comprehension. From the perspective of theoretical morphology, however, the linguistic distinction between regular and irregu-

360 Harald Clahsen lar inflection is a relatively simple one, and a general claim such as the correspondence hypothesis should not only rest on the set of phenomena that fall in the regular/irregular cluster. According to the correspondence hypothesis, the speaker-hearer’s internal representations of morphologically complex words are mediated by structures that are isomorphic to those that the grammatical formalism employs. If this is correct, we should not only see effects of the linguistic distinction between regulars and irregulars in psycholinguistic experiments, but also effects of other primitive morphological objects and operations. In this chapter, we will report three sets of experimental findings to show that morphological distinctions that go beyond the regular/irregular contrast are indeed crucial for understanding how the speaker/hearer processes inflected and derived words. 3. Beyond words and rules: Morphological types in the mental lexicon What are the basic morphological units and operations that the speaker/ hearer employs in processing inflected words? The answers to this question are controversial among psycholinguists. Some argue that words are the basic processing units and that each word form of a given lexeme has its own entry in the mental lexicon (Butterworth 1983; Manelis & Tharp 1977; see also Rumelhart & McClelland 1986). Others have argued that affixes and roots are the major units in morphological processing. Taft, Forster and their collaborators (Taft 1988; Taft, Hambly & Kinoshita 1986) posited parsing operations such as ‘prefix stripping’ by which roots are identified and matched against entries stored in the mental lexicon. The distinction between roots and affixes is also essential to Pinker’s (1999) ‘words-andrules’ model in which irregular inflection is tied to roots, and regular inflection is based on affixation, e.g. on rules such as ‘Add -ed’ which concatenates a bound past-tense morpheme to a verbal root (walk+ed). The fact that much psycholinguistic work on inflectional morphology has only employed a restricted set of largely morpheme-based concepts (words, roots, affixes) coincides with its strong focus on English for which these morphological types seem sufficient. Indeed, Pinker (1999) shows that his simple words-and-rules model can explain a range of findings from different kinds of processing experiments on the English past tense, but at the same time he readily admits that accounting for results from other languages may require more complex distinctions. To illustrate this point, we do not have to go very far. Consider, for example, the experimental results from Sonnenstuhl & Huth (2002) on -n plu-

Linguistic perspectives on morphological processing

361

rals in German. They found that -n plurals of feminine nouns that end in a schwa, e.g. (die) Tasche – Taschen ‘the pocket – pockets’ (labelled class I), yielded a word-form frequency effect in unprimed lexical decision, i.e. significantly shorter response times for high-frequency plural forms such as Bienen ‘bees’ than for low-frequency ones such as Gesellen ‘lads’ (518 vs. 568 ms), and a full root-priming effect in cross-modal priming, i.e. the plural form primed the root (e.g. Taschen
Tasche) as effectively as an identical prime (e.g. Tasche
Tasche). By contrast, -n plurals of non-feminine nouns without a final schwa, e.g. der Bauer – Bauern ‘the farmer – farmers’ (labelled class II) also produced a response-time advantage for high-frequency over low-frequency plural forms in unprimed lexical decision (535 vs. 594 ms), but in the cross-modal priming task it only yielded a reduced priming effect, i.e. response times that were shorter than for an unrelated prime (e.g. Glas
Bauer), but longer than for an identical prime. These results are hard to account for within the limited set of morphological types employed in the familiar psycholinguistic frameworks. The frequency effects may be taken to indicate that -n plurals have full-form representations. The full priming effect for class I -n plurals, however, is suggestive of a decomposed root+affix representation. In Pinker’s words-and-rules model, for example, one may say that class I -n plurals are regular, i.e. based on affixation, whereas class II -n plurals are irregular, i.e. roots. This provides an account of the different priming effects, but it requires additional considerations to explain why both types of -n plural yielded full-form frequency effects in unprimed lexical decision. An explanation for Sonnenstuhl & Huth’s findings will be proposed at the end of this section. In addition to words (or rather lexemes), roots, affixes (or exponents), many morphologists consider stems as an independent morphological type. Aronoff (1994), for example, points out that a root is morphologically unanalyzable and is defined with respect to a lexeme. A stem, by contrast, is defined with respect to a series of forms (not with respect to a lexeme), represents recurrent parts within such a series, and may be morphologically complex. To illustrate these notions, consider inflected verb forms of Spanish in which inflectional affixes are typically combined with stems, which in turn are combinations of theme vowels and roots. This is illustrated in (1) for a finite form of the verb cantar ‘to sing’. (1)

cant á ba mos sing theme vowel past imperfect, indicative 1plural ‘we sang’

362 Harald Clahsen Most verb forms in Spanish have (combinatorial) [root + theme vowel] stems which can be further divided into three conjugations, identified by the theme vowel, e.g. in the infinitive. The first conjugation is by far the largest (in terms of verb types), and almost all of the 1st conjugation verbs have combinatorial stems. The 1st conjugation is also the open class par excellence. For example, English to stress ‘to cause stress’ becomes Spanish estresar, English to film Spanish filmar, and so on. Approximately 900 verbs (almost all of which are 2nd and 3rd conjugation verbs) have marked stems, i.e. forms which are not simply [root+theme vowel] combinations. About 30 of these are highly irregular, including verbs such as estar ‘to be’, caber ‘to fit’, querer ‘to want’, poner ‘to put’, tener ‘to have’, ir ‘to go’, venir ‘to come’. This is illustrated in (2) for the past indicative of poner ‘to put’. (2)

puse pusiste puso pusimos pusisteis pusieron

‘I put’ ‘You put’ ‘she/he/it put’ ‘we put’ ‘you put’ ‘they put’

The paradigm illustrates that the marked stem pus- is used in past tense forms, sometimes without a theme vowel, as for example in puse, puso, and sometimes together with the 2nd conj. theme vowel (= -i), as for example in pus-i-ste. Another kind of marked stem form involves diphthongization, e.g. entender – entiendo ‘to understand – I understand’, rogar – ruego ‘to beg – I beg’, which is common in Spanish, not only in verbs. There are, however, many verbs that have (non-low) vowels which do not undergo diphthongization, e.g. notár-nóto ‘to note – I note’. Diphthongized stems are therefore opaque, in the sense that their alternating behaviour cannot be predicted from either surface alternant. The case of Spanish illustrates two general properties of stems. First, stems such as cant-a, pus- or duerm- (= the marked stem of dormir ‘to sleep’) preserve the inherent morphosyntactic and semantic properties of the root, and they form the base for the application of inflectional rules. These properties suggest that stems (along with roots) form lexical entries and may be members of an extended lexical stem set (Anderson 1992: 133). For example, for the verb with the root pon- in Spanish, one may define an indexed stem pus- which is selected by members of the past series. The second property of stems is that they fall into two broad classes, combinato-

Linguistic perspectives on morphological processing

363

rial and suppletive. Both types of stems can be defined in terms of realization rules (see Blevins 2003). Combinatorial stems, e.g. [cant-a], can be represented by a rule that introduces -a as the theme vowel of any 1st conj. root. Suppletive stems, e.g. [pus], can be described in terms of a realization rule that spells out the property [past] as a constant stem entry. Thus, the distinction between productive (combinatorial) stems and static (suppletive) stems can be recast as a contrast between rules that contain variables (for verbal roots of a given conj. class) and those that have a constant (lexemespecific) output. According to the correspondence hypothesis, we would expect to find effects of these different kinds of stem formation processes in morphological processing experiments. Given that stems form lexical entries, we would expect that in experimental tasks that tap lexical entries, all stems (combinatorial as well as suppletive) should produce lexicality effects, e.g. frequency effects in the lexical decision task. Moreover, in tasks that are sensitive to the internal morphological structure of an inflected word form, we would expect to find effects of the difference between combinatorial and suppletive stems. Combinatorial stems should produce the same experimental effects as regularly inflected word forms, since both processes are based on morphologically decomposed representations that contain variables. By contrast, suppletive stems should pattern with irregular inflection in such tasks. We have investigated stem formation processes in a number of languages using different experimental paradigms and other sources of psycholinguistic evidence (German: Clahsen et al. 2001a, 2002a; Spanish: Clahsen, Aveledo & Roca 2002b; Rodríguez-Fornells, Münte & Clahsen 2002; Catalan: Rodríguez-Fornells et al. 2001; Italian: Say & Clahsen 2002). Consider first findings from a series of experiments using Event-Related Brain Potentials (ERPs). Active neurons in the brain produce electrical activity which can be measured by electrodes placed on the scalp. Psycholinguists are concerned with isolating the electrical activity associated with a specific task – the event-related potentials – from background activity with the aim of identifying the electrical components associated with a given linguistic stimulus (for reviews see Osterhout & Holcomb 1995; Kutas & Van Petten 1994). While many of the techniques involved in psycholinguistic research are only able to measure the end-result of the task, this technique has the advantage of being able to measure linguistic processing as it actually occurs. Table 1 presents a summary of relevant ERP effects on inflected word forms.

364 Harald Clahsen Table 1. Summary of ERP effects on inflected word forms Representation

ERP effect

German Overregularized participles: *ge[lauf]-t (corr.: gelaufen ‘run’) Overregularized plurals: *[Vase]-s (corr.: Vasen ‘vases’)

LAN a LAN b

Catalan

Overregularized 1st conj. stems: *[dorm]-a]-t (corr.: dormit ‘slept’)

LAN c

English

Regular past-tense priming: [use]-d
use Irregular past-tense priming: [found]
find

reduced N400 d

Spanish

a

N400 d

Unmarked stem priming: [and]o
[and]-ar ‘(I) walk
(to)walk’

reduced N400 e

Marked stem priming: [duerm]o
[dorm]-ir ‘(I) sleep
(to) sleep’

N400 e

b

c

Sources: Penke et al. (1997); Weyerts et al. (1997); Rodríguez-Fornells et al. d e (2001); Münte et al. (1999); Rodríguez-Fornells, Münte & Clahsen (2002)

Using the ERP violation paradigm, Penke et al. (1997) compared ERPs to German participles with correct regular affixes (getanz-t ‘danced’) and correct irregulars (gelad-en ‘loaded’) with ERPs to incorrect participles, that is, regular verbs with the incorrect ending -n (*getanz-en) and irregular verbs with the regular affix -t (*gelad-et). The critical words were visually presented in three different versions to three different groups of subjects, as part of a simple sentence, in a word list, and embedded in a story; for each version separate ERPs were recorded. The task assigned to the subjects was to judge whether a probe sentence was or was not an exact repetition of one of the previous 10 sentences and to press one of two buttons accordingly. The probe sentence was shown in red letters and was either an exact repetition of one of the 10 sentences shown before or was slightly modified, by exchanging one word. The ERP responses were very consistent across the three versions of the experiment. Overregularized participles (e.g. *geladet) produced a negative waveform with a focal left anterior temporal distribution between 300 and 800ms after the onset of the critical word, an effect

Linguistic perspectives on morphological processing

365

that has been discovered in previous ERP studies and that has been called a L(eft) A(nterior) N(egativity); no such effect was observed for correct vs. incorrect regulars. The LAN effect was replicated by Weyerts et al. (1997) for overapplications of regular -s plurals (e.g. *Muskel-s instead of Muskeln ‘muscles’). In previous ERP studies, LAN effects have been obtained for violations of morpho-syntactic rules in a range of languages, for example, violations of subject-verb agreement (*you goes, see Münte, Heinze & Mangun 1993; Osterhout & Mobley 1995), number errors in Turkish (Münte et al. 1995), phrase structure violations in English (Neville et al. 1991) and in German (Friederici, Hahne & Mecklinger 1996) and ungrammatical wh questions (Kutas & Kluender 1994). These results suggest that morphological overregularizations such as *Muskel-s and *gelad-et and syntactic violations involve similar neural processes. Linguistically, misapplications of regular inflection are akin to the morpho-syntactic violations that were tested in the studies mentioned above, in that two legal components, i.e. the stem and the regular affix, appear in illegal combination. Irregulars, on the other hand, are not decomposed into stem and affix, and hence irregularizations do not produce an ERP effect, i.e. a LAN, that is typical of combinatorial violations. Against this background, Rodríguez-Fornells et al. (2001) investigated stem formation processes in Catalan, also using the ERP violation paradigm. The most interesting finding of this study was a left sided negativity for stem violations. LAN effects were seen for overapplications of the 1st conjugation theme vowel -a- to a verb form that requires a 2nd or 3rd conjugation form, e.g. *dorm-a-t instead of the correct dorm-i-t ‘slept’ or *tem-a-t instead of tem-u-t ‘feared’. This finding is parallel to our previous results on German participles and noun plurals. In line with the interpretation of the effects found in the German studies, the LAN effect for stem violations in Catalan can also be taken to reflect a combinatorial violation. Given that 1st conjugation stems are combinations of a verbal root and the theme vowel -a, the morphological anomalies in incorrect 2nd and 3rd conjugation forms such as *dormat and *temat represent violations of morphological structurebuilding in that two legal components (a 2nd or 3rd conjugation root and the 1st conj. theme vowel -a) appear in illegal combinations. Taken together, these findings show that LAN effects are not restricted to morphosyntactic violations, but that they can also be obtained for purely morphological violations, as in the case of Catalan. From a linguistic perspective, the fact that the brain produces similar responses for violations of regular inflection as well as for violations of productive stem formation processes confirms that these are based on the same process, namely on combinatorial rules.

366 Harald Clahsen ERPs have also been used to study brain activity in the repetition priming task. Several previous ERP experiments have shown that unexpected words elicit a so-called N400 component, i.e. a negativity at central electrode sites with a maximum at approximately 400ms after the onset of the stimulus word (for a recent review see Kutas & Schmitt 2003). The N400 is supposed to tap processes of lexical access. It has also been found that when written words are repeated within a list, the ERP to their second presentation is associated with an N400 that is reduced in amplitude relative to nonrepeated words. The reduced N400 is conceived of as a repetition priming effect in that lexical access is facilitated relative to unprimed words in the list. Münte et al. (1999) examined English past tense forms in an ERP priming study. They found that uninflected verb forms (walk) primed by regularly inflected past-tense forms (walked) showed a reduced N400 relative to unprimed verb forms (e.g. look). No such effect was observed either for irregular verbs or for other control conditions that exhibited the same degree of orthographic and phonological overlap to their targets as regularly inflected verbs (card
car). The reduced N400 for regular past tense forms indicates that these forms are morphologically decomposed, thereby making the unmarked stem/root (e.g. walk) available for priming. Irregular past-tense forms (drank), on the other hand, can only indirectly access the unmarked stem/root, and do therefore not lead to a modulation of the N400. These results are compatible with previous findings from behavioral priming experiments in which only regularly inflected word forms showed full priming effects (see e.g. Sonnenstuhl, Eisenbeiss & Clahsen 1999). Rodríguez-Fornells, Münte & Clahsen (2002) employed the same ERP repetition priming paradigm to examine the processing of different kinds of stem/root forms in Spanish. There were two experimental conditions; in the first condition, e.g. ando
andar ‘I walk-to walk’, the prime and the target shared the same root (= and-), whereas in the second condition, e.g. duermo
dormir ‘I sleep-to sleep’, the prime contained a marked stem/ root (= duerm-). A reduced N400 was found for the target forms in the first condition (ando
andar), but not in the second one (duermo
dormir). Control conditions using nonce words demonstrated that the surface form properties (i.e. the different degree of phonetic and orthographic overlap between primes and targets) do not explain the observed priming difference. As mentioned above, a reduced N400 is indicative of a priming effect, i.e. facilitated lexical access to the target form. This finding on Spanish is parallel to what was found for regularly inflected verb forms in English. In the case of Spanish, effective priming is possible for prime-target pairs

Linguistic perspectives on morphological processing

367

such as ando
andar because the prime and the target access the same entry, i.e. the root [and-], in the same way as for prime-target pairs such as used
use in English. By contrast, the priming effect is not present for marked stems (e.g. duerm-), parallel to irregular past-tense forms in English (e.g. found), indicating that these forms do not permit any direct access to the root and do therefore fail to produce an effective facilitatory effect on accessing the target. Consider finally some findings from unprimed visual lexical decision experiments investigating different kinds of German verb forms. Table 2 presents a summary from 4 experiments. Cases in which the bracketed forms shown in Table 2 yielded a significant frequency effect, i.e. shorter response times for high-frequency forms than for low-frequency ones, are indicated by an asterisk. Table 2. Summary of frequency effects in visual lexical experiments Representation

Advantage for high-frequency forms

irregular participles: [gelaufen] ‘run’ regular participles: ge-[kauf]-t ‘bought’

59ms* a –4ms a

A–B–B participles: regular participles:

[geschoben] ge-[kauf]-t ‘bought’

58ms* b –8ms b

Strong preterites:

[sprach]en ‘talked’

49ms* c

Preterite stems:

[log]en

56ms* d

a

‘lied’

b

Sources: Clahsen, Eisenbeiss & Sonnenstuhl (1997), experiment 2; Clahsen, Eisenc beiss & Sonnenstuhl (1997), experiment 3; Clahsen et al. (2001a), experiment 3; d Clahsen et al. (2001a), experiment 4

In experiment 2 of Clahsen, Eisenbeiss & Sonnenstuhl (1997), 20 -n participles of different subclasses of irregular verbs were compared with 20 -t participles of regular verbs. The -t and -n participles were divided into two subgroups each according to the (word-form) frequencies of the participle forms, a high-frequency group and a low-frequency group. The experiment revealed a significant (59ms.) advantage for high-frequency -n participles of irregular verbs, whereas for regular -t participles there was no such effect. In experiment 3 of Clahsen, Eisenbeiss & Sonnenstuhl (1997), 18 -n participles of the A–B–B subclass of irregular verbs which have the same

368 Harald Clahsen marked stem in the participle and in preterite forms (e.g. gleiten – glitt – geglitten ‘glide’) were compared with 18 -t participles of regular verbs. The -t and -n participles were again divided into two subgroups each according to the (word-form) frequencies of the participle forms. Moreover, the items were matched for stem frequency, and this was held constant across the high- and low-frequency conditions. The results are parallel to the previous one: Only irregular participles gave rise to a significant word-form frequency effect. Taken together, these results indicate lexicality effects for irregular participles suggesting that they access entries in the mental lexicon. In two further lexical decision experiment, we examined irregular preterite stem forms that were suffixed with regular person and number affixes, such as sangen ‘sang-1st/3rd pl.’. In experiment 3 of Clahsen et al. (2001a), 52 items from the three subclasses of irregular verbs were arranged in two conditions, one with low preterite stem and one with high preterite stem frequencies. To control for other frequency effects, the items in both conditions were matched for their mean word form and their mean verb frequencies. The experiment revealed a significant (49ms.) advantage for verb forms with high preterite stem frequencies indicating that irregular stems are lexically represented (rather than rule-based) and that preterite stems are represented separately from the inflectional affixes with which they may occur (rather than being stored as wholes). Experiment 4 of Clahsen et al. (2001a) investigated regularly inflected preterite forms of A-B-B verbs which exhibit the same stem change in preterite and participle forms. The experimental items were arranged pairwise in two conditions according to their stem frequencies in preterite forms, with a high and low frequency condition. In addition, the items in both conditions were matched for their mean B-stem frequencies, i.e. the mean frequency of all preterite and participle forms of each experimental item. The results are parallel to the previous experiment: Reaction times for A-B-B verbs with high preterite stem frequencies were significantly shorter than for those with low preterite stem frequencies indicating that preterite stems access entries in the mental lexicon. The results on German -n plurals mentioned at the beginning of this section can also be accounted for in these terms. Blevins (2000) argued (on independent linguistic grounds) that a second noun stem underlies all -n plurals as well as other German plural forms (except those in -s). Note, for example, that all kinds of plurals (except -s plurals) irrespective of their gender or stem properties are in principle available for further word-formation processes yielding forms such as taschen-los ‘pocket-less’, or Bauernschaft ‘farming community’. On the other hand, -n plurals do not represent

Linguistic perspectives on morphological processing

369

a homogeneous class in German. There are (at least) two types, with different degrees of productivity and predictability. While plural formation with -n is fully predictable for feminine nouns with a stem-final schwa, e.g. (die) Tasche – Taschen ‘the pocket – pockets’ (see class I above), plural formation with -n is not predictable for non-feminine nouns without a stem-final schwa, e.g. der Bauer – Bauern ‘the farmer – farmers’ (class II above). Given the fact that there is no single exception to -n plurals of class I, i.e., all feminine nouns with a stem-final schwa take the -n plural, we might hypothesize that class I -n plurals are based on combinatorial stems, whereas all other -n plurals are suppletive. The experimental results obtained by Sonnenstuhl & Huth (2002) correspond to these distinctions. Class I -n plurals form stems, hence the full-form frequency effect in the lexical decision task. On the other hand, these -n plurals are combinatorial, i.e. decomposable, hence the full priming effect found in the cross-modal priming task. Class II -n plurals are also stems, and consequently yield the same frequency effect as class I -n plurals, but unlike the latter, class II -n plurals are suppletive, i.e. not decomposable, and can therefore only indirectly access the root; the result is a reduced priming effect in the cross-modal priming task. Summarizing, the findings reported in this section were meant to show that distinguishing between different morphological types can be useful for better understanding how the speaker/hearer processes inflected word forms. The results mentioned specifically demonstrate the significance of stems in morphological processing. We found evidence for two essential linguistic properties of stems, firstly that they are associated with lexical entries, hence the lexicality effects in our experiments, and secondly that stems may be either combinatorial or suppletive, hence the decomposition effects in our experiments for combinatorial stems but not for suppletive ones. 4. Dissociating derivation and inflection in language processing Previous psycholinguistic studies of derived word forms have yielded a mixed set of results. Some studies argued that different mechanisms are involved in the processing of inflectional and derivational forms (see e.g. Stanners et al. 1979), while others claimed that non-morphological factors such as frequency and semantic transparency affect the processing of both inflected and derived word forms (Raveh & Rueckl 2000, among others). Here we argue that by considering relevant linguistic distinctions the psy-

370 Harald Clahsen cholinguist will get a clearer picture of the processing differences between derived and inflected word forms. Different morphological theories assume that derivational or lexeme formation processes are distinct from, and in some sense prior to, inflectional or paradigmatic processes (see Spencer 1991 for review). Anderson’s (1992) realization-based model, for example, recognizes a general class of Word Formation Rules but claims that this class is effectively partitioned into two discrete subclasses. Derivational rules ‘constitute sources for lexical stems’, whereas inflectional rules ‘introduce inflectional material into the surface forms of words’ (Anderson 1992: 184–185, emphasis added). Other realization-based approaches, such as Matthews (1991) and Stump (2001), establish a similar split between the rules that define derivational stem entries from those that define inflected word forms. The difference between the outputs of inflectional and derivational rules is what accounts for the relative ordering of derivation and inflection. Derivational rules map one stem entry, or feature-form pairing, onto another entry. This derived entry may then provide the input to subsequent derivational rules, or may provide the base for the application of inflectional rules. In contrast, inflectional rules are simple feature-form mappings that specify the form that ‘realizes’ or ‘spells out’ a particular set of features. As a consequence, regular inflectional rules do not define new entries of any kind and cannot, in principle, provide the input to derivation. On the other hand, derivation shares with inflectional processes the distinction between productive (‘built’) forms vs. nonproductive (‘frozen’) forms. Consider, for example, the contrast between productive derivational processes such as -ing nominalizations and frozen derivational forms such as refusal or strength in English. In realization-based approaches, this contrast can be recast in terms of the distinction between rules that contain variables and those that have a constant output, see e.g. Booij (2002) or Blevins (2003). The contrast is illustrated in (3), using a simplified version of Aronoff’s (1994) realization pair format. (3)

a. b. d.


d.


Linguistic perspectives on morphological processing

371

The first element of each pair of the two inflectional rules (3a) and (3b) identifies the features to be realized, while the second element indicates the formal spell-out. The regular 3sg rule in (3a) spells out the bracketed features by adding the exponent ‘s’ to the base form represented by the variable ‘X’. Applied to the base of the regular verb WALK, this rule defines the regular 3sg form walks. The pair in (3b) realizes the 3sg present indicative features of the lexeme BE by the constant (irregular) form is. Likewise, there are two kinds of derivational rules, those that contain variables (e.g. 3c) for productive (‘regular’) derivational processes, such as deverbal -ing nominalizations in English, and those that have a constant output, e.g. for frozen (‘irregular’) forms such as refusal (3d). In contrast to the regular inflectional rule (3a), however, the derivational rule in (3c) represents a mapping from one realization pair to another. According to the correspondence hypothesis we would expect that differences and similarities between derived and inflected forms in their morphological representations yield corresponding effects in processing experiments. More specifically, we predict that if language processing is sensitive to the internal morphological structure of a derived or inflected word form, productive derivational processes should produce the same experimental effects as regularly inflected word forms, since both processes are based on morphologically decomposed representations that contain variables. By contrast, non-productive derivational processes should pattern with irregular inflection in such tasks. On the other hand, we predict that in experimental tasks that are sensitive to the lexical entry as a whole, derived forms (productive as well as frozen ones) and irregular inflected forms should produce lexicality effects. This is because all derived processes share with irregular inflected forms the fact that their outputs are lexical entries. Regular inflection, on the other hand, should not yield any lexicality effects, since the output of a regular inflectional rule does not define any new entries. To determine similarities and differences between derivational and inflectional processes in processing, we have studied a range of derived and inflected forms of German in cross-modal priming and visual lexical decision experiments. Table 3 presents a summary of some relevant results. The first column shows the assumed representation for the test items under study. Brackets indicate lexical entries, some of which have a decomposed structure. The second column shows whether or not cross-modal priming produced a full or a reduced priming effect for the forms under study. Full priming effects are determined by subtracting the response times to the tar-

372 Harald Clahsen get in the Identity condition (e.g. Waggon
Waggon) from the target response times in the morphological priming condition (e.g. Waggons
Waggon). Full priming is obtained in cases in which there are no significant differences between these conditions. Reduced priming is evident from significantly longer target response times (indicated by an asterisk) in the morphological priming condition than in the Identity condition. For the third column, we subtracted the lexical decision times of the high-frequency items from those of the low-frequency ones. Cases in which an inflected or derived form with a relatively high word-form frequency produced significantly shorter response times than an item with a relatively low word-form frequency are indicated by an asterisk. Table 3. A summary of experimental findings on German inflection and derivation Representation

Full or reduced Advantage for high Priming frequency word forms –6ms a 0ms a –19ms * a –32ms * a

–4ms b –4ms b 84ms * b 59ms * b

in- adjectives: [instabil] ‘unstable’ un- adjectives: [un[gesund]]‘unhealthy’

–27ms * c 5ms c

43ms * c 41ms * c

-ung nominalizations: [[stift]ung] ‘foundation’ diminutives: [[kind]chen] ‘small child’

3ms d 7ms d

30ms * d 41ms * d

-s plurals: -t participles: -er plurals: -n participles:

a

[waggon]s ge[kauf]t [kinder] [gelogen]

‘wagons’ ‘bought’ ‘children’ ‘lied’

b

Sources: Sonnenstuhl, Eisenbeiss & Clahsen (1999); Clahsen, Eisenbeiss & Sonnenc d stuhl (1997); Sonnenstuhl & Clahsen (in prep.); Clahsen, Sonnenstuhl & Blevins (2003)

Consider first inflected word forms. Regularly inflected word forms such as -t participles and -s plurals produced full priming and no word-form frequency effects. For irregularly inflected forms, such as -n participles, and -er plurals the opposite pattern of results was obtained, i.e. significant reductions in priming and a significant advantage for high-frequency word forms. These results correspond to the linguistic representations shown in Table 3 for these kinds of items. Irregular participles and noun plurals are undecomposed entries, hence the full-form frequency effect. They can only indirectly access their corresponding unmarked base and therefore produce

Linguistic perspectives on morphological processing

373

reduced priming effects. Regular participles and noun plurals, on the other hand, do not form lexical entries, and hence the lack of a full-form frequency effect. Instead, they are decomposable into an unmarked root plus the regular affix, and given this representation they can fully prime their corresponding base form. Nonproductive derivational processes such as in-adjective formation yielded results parallel to those of irregular inflection, i.e. significantly reduced priming and full-form frequency effects. This follows from assuming that such forms constitute undecomposed entries. By contrast, the productive derivational forms tested produced a pattern of experimental effects that differs from both the one for regular inflection and the one for irregulars. In terms of their morphological representations, productive derivational processes share with irregular inflection the fact that their output yields stems, i.e. lexical entries, and hence the significant advantage for high-frequency word forms in the lexical decision task. Their internal morphological structure, however, is parallel to regular inflection in that -ung nominalizations, -chen and -lein diminutives, and un-adjective formations are decomposed forms derived from rules that contain variables. Given that the recognition of forms such as un+gesund, Stift+ung, and Kind+chen involves the activation of the underived stem form in addition to the associated derivational suffix, the full priming effect obtained for such forms follows from the activation of the shared stem entry of prime and target. We conclude that our experimental results are compatible with models of morphology that treat productive inflection and derivation as the result of combinatorial operations and associate irregular inflection and nonproductive derivation with entries. 5. Morphological paradigms in language processing The question of how different regularly inflected word forms of the same lexeme (e.g. walk-s, walk-ed, walk-ing) are mentally represented and processed has received relatively little attention from psycholinguists. Some researchers have studied processing differences between what was considered to be ‘canonically inflected’ or base forms of a given lexeme and ‘non-canonically inflected’ forms of the same lexeme. In a set of studies on inflected Serbo-Croatian nouns, for example, it was found that nominative forms were processed faster than non-nominative forms (Lukatela, Carello & Turvey 1987; Feldman & Fowler 1987; Todorovic 1988; Kostic 1995). Similar results have been obtained for German. These results have been

374 Harald Clahsen taken to support a satellite-like representation of inflected word forms in which the inflectional variants of a given lexeme are connected to a nucleus (e.g. the nominative form). According to the satellite model, each lexeme has one designated nucleus, and if access is made via a satellite form, extra time is required, hence the longer response times for non-nucleus forms. This model is compatible with a traditional model of paradigm structure in which a particular form (or form set) has a distinguished status, based on their role in predicting other forms of the paradigm. On the other hand, however, it does not posit any representational difference between the various satellites of a nucleus, and hence there should not be any processing differences between the different inflected forms of any given lexeme. Recent experimental results to be reported below show that this prediction does not hold indicating that the satellite model only provides a partial account of morphological representation and processing. Linguists have developed the notion of a morphological paradigm to represent relationships between different inflected word forms of any given lexeme. An inflectional paradigm is a multi-dimensional, potentially recursive matrix which is defined by the morpho-syntactic features of word forms or affixes (see e.g. Aronoff 1994; Anderson 1992; Zwicky 1985; Carstairs 1987; Stump 1993). A paradigm contains a set of slots defined in terms of morpho-syntactic feature values and shows how each slot is to be filled. The result is that any lexeme that belongs to a particular syntactic category has inflected word forms defined by the paradigm. The formation of paradigms is constrained by general principles, such as Blocking and Specificity (Kiparsky 1982, 1998). Blocking and Specificity require that if two rules or affixes are in competition for one paradigm slot, the rule that is more specific in its application is preferred over the more general one. If morphologically related inflected word forms are indeed represented in paradigms, the correspondence hypothesis leads us to expect effects of paradigmatic representation in processing experiments, for example, processing differences between inflected forms with different paradigmatic specifications. To examine the role of morphological paradigms in language processing, we performed a series of experiments on two systems of German inflection for which inflectional paradigms have been proposed in the linguistic literature: (i) person and number marking on finite verbs, e.g. lach-st ‘laugh-2sg’, lach-t ‘laugh-3sg./2pl.; (ii) gender, number and case marking on attributive adjectives, e.g. wild-es ‘wild-nom. neut. sg.’, wild-em ‘wild-dat. masc. sg.’.

Linguistic perspectives on morphological processing

375

There are four overt person and number affixes that appear on finite verb forms in German, -e, -st, -t, and -n. In (4), the full paradigm of the regular verb lachen ‘to laugh’ is shown for illustration. (4)

Imperative

Present

Preterite/Past Subj.

Present Subj.

1sg. 2sg. 3sg.

lach(e)

lache lachst lacht

lachte lachtest lachte

lache lachest lache

1pl. 2pl. 3pl.

lacht

lachen lacht lachen

lachten lachtet lachten

lachen lachet lachen

In all tenses and moods, the 2sg. has the exponent -st and the 2pl. has the exponent -t, while the 1pl. and the 3pl. have the same exponent throughout (-n). Note also that the 1sg. and the 3sg. do not have overt person/number affixes in the preterite and the subjunctive. Only in the present tense indicative, the 1sg. and 3sg. have different exponents, namely -e and -t respectively. There are different ways of describing these person and number forms in terms of morphological paradigms. Blevins (2003) proposed an exponencebased analysis of the German conjugational system. This system involves the root (e.g. lach-), a second stem (e.g. lacht), which constitutes the basis for all preterite forms and for regular participle forms, and stem-formation and agreement rules. The two stem-formation rules add -e to the root or to the second stem to form the present tense subjunctive stem (e.g. lache) and the preterite stem (e.g. lacht-e). The three agreement rules derive the 2sg. forms ending in -st, the 2pl. forms ending in -t, and the 1/3pl. forms ending in -n. The contrast between 1sg. and 3sg. which is only distinctive for present indicatives is mediated in terms of different stems, i.e., 3sg. present forms in -t and 1sg. indicative and 1sg subjunctive forms are said to ‘select’ particular stems: 3sg. forms select the -t stem (e.g. lacht), while 1sg. forms select the secondary stem in -e. Wunderlich & Fabri (1995) and Bittner & Bittner (1990) presented alternative paradigm-based analyses of this inflectional system. What is common to these accounts is that they posit two different paradigmatic representations for the various person and number forms of German. The 2sg., the 2pl., and the 1/3pl. are based on general agreement rules in Blevins’ (2003) analysis, they form a general paradigm in Wunderlich & Fabri’s (1995) account, and they constitute the most general

376 Harald Clahsen paradigm structure constraint which applies to all verbs in Bittner & Bittner’s (1990) analysis. By contrast, the 1sg. and the 3sg. of the present tense are based on stem forms in Blevins’ account, they form a subparadigm in Wunderlich & Fabri’s model, and they are part of a specific paradigm constraint for Bittner and Bittner. Thus, there seems to be agreement among these accounts in that the paradigmatic representation of the 1sg. -e and the 3sg. -t is different from those of the other person and number forms. German attributive adjectives carry a portmanteau-affix that expresses the grammatical features gender, number, and case, as for example in kalt-em Wein ‘cold-masc.sg.dat. wine’. With respect to the morphological expression of these features, two declension classes are commonly distinguished, weak and strong declension as shown in (5). Strong declension applies to adjectives that are used without a determiner or a demonstrative or with an uninflected determiner, e.g. (ein) kalter Wein ‘(a) cold wine’, while adjectives that are combined with a strongly inflected determiner take an affix from the weak paradigm, e.g. der kalte Wein ‘the cold wine’, mit dem kalten Wein ‘with the cold wine’, mit einem kalten Wein ‘with a cold wine’. (5)

a. Strong adjective declension Singular

Nominative Accusative Dative Genitive

Masculine

Neuter

Feminine

-r -n -m -n

-s -s -m -n

-e -e -r -r

Plural -e -e -n -r

b. Weak adjective declension Singular

Nominative Accusative Dative Genitive

Masculine

Neuter

Feminine

-e -n -n -n

-e -e -n -n

-e -e -n -n

Plural -n -n -n -n

As is clear from (5), the affixes differ with respect to the degree of homonymity. Both, -e and -n occur in the strong and weak declension, while -s,

Linguistic perspectives on morphological processing

377

-r, and -m only occur in the strong declension. Several linguists (Zwicky 1986; Blevins 1995; 2000; Wunderlich 1997; Cahill & Gazdar 1997) have analyzed this system using morphological paradigms. Consider, for example, Blevins’ (2000) account of German adjective inflection in which the forms -e and -s have only negative feature specifications, while -m has two positive features, as shown in (6). Thus, in this analysis the -m affix comes out as the most specific form, being specified for [+OBLIQUE] and for [+DATIVE]. Moreover, as is clear from (6), -s is specified for more features than -e, indicating that -s is more specific in terms of its feature content than -e. (6)

Specifications of German adjective inflection (Blevins 2000) -e

-s

-m

— — —

[–PL] [–FEM] [–MASC] [–OBL] —

[–PL] [–FEM] — [+OBL] [+DAT]

[–OBL] —

Wunderlich’s (1997) account relies on specifications of the structural cases in terms of a hierarchy of Theta-roles. In this system, the -m affix of the strong declension paradigm is restricted to just one case, dative, whereas, for example, -s occurs in nominatives and accusatives. Moreover, dative requires two features for its specification, whereas nominative is completely unspecified and accusative has just one feature. Thus, despite various differences between the linguistic treatments of German adjective inflection, there seems to be agreement that the -m affix is paradigmatically more specified than any of the other adjective forms, and that the feature content of -s is more specific than that of -e. Given the correspondence hypothesis, the paradigmatic differences within these declensional and conjugational systems should affect the way inflected adjectives and verb forms are processed. More particularly, we would expect to find an effect of paradigmatic specificity. In a priming experiment, prime-target pairs in which the target contains unprimed features, i.e. features unavailable from the prime, should lead to reduced priming, whereas prime-target pairs with affixes whose features do not clash should produce more effective priming effects. Consider, for example, the various adjective forms. According to our predictions, adjectives inflected with -m

378 Harald Clahsen should be less effectively primable than adjectives inflected with -s, and adjectives with -s less effectively than those with -e. This is because forms affixed with -m require the processing of specific, i.e. unprimed features, which should lead to longer response times. Similarly, -s adjective forms contain features that cannot be primed by -e adjectives, whereas -e adjective forms do not have any features that could not be primed by -s. Therefore, if the morphological feature content of these adjective forms matters for processing, we would expect to find corresponding asymmetries between -s and -e adjective forms in priming. To determine potential effects of paradigmatic specificity in processing, we have studied sets of inflected adjectives and finite verb forms in three cross-modal priming experiments (Clahsen et al. 2001a, 2001b). Recall that in this experimental paradigm, subjects hear a spoken prime, e.g. kaltes, immediately followed by a visually presented target form (e.g. kaltem) for which they are required to make a lexical (word/non-word) decision. To determine the effectiveness of priming, we compared the response times to the visual targets preceded by morphologically related primes, e.g. kaltes
kaltem, with the response times to the same targets preceded by an identical prime, e.g. kaltem
kaltem. The smaller the differences between these response times, the more effective the priming. Table 4 presents a summary of the findings. The first column shows the prime-target pairs tested with the (auditory) prime shown on the left and the (visual) target on the right side of the arrow; the second column provides an example. The third column shows whether or not a particular form contains unprimed features, i.e. features unavailable from the prime. The fourth column provides the priming effect (shown in ms.) for each prime-target pair; the figures represent the difference between the baseline condition (= identical repetition) and the response times to the visual targets for each prime-target pair. These differences were also tested statistically using t-tests; an asterisk indicates a significant difference between the morphological and the identity priming conditions. Consider first the results on adjectives and compare the priming effects in Tab.4 with the feature specifications of the affixes shown in (6) above. Tab.4 shows that the prime-target pairs -e
-m and -s
-m produced the most significant reduction in priming relative to identical repetition (= -30ms. and -36ms.) indicating that adjectives with -m are more difficult to prime than any other adjective form. This is because (given the representations in (6)) the target -m contains positively specified case features that are not available from the primes. On the other hand, for the pair -s
-e

Linguistic perspectives on morphological processing

379

there are no unprimed features in the target form, as the [-OBL] feature of -e is available from the prime, and correspondingly, -s adjective forms fully prime -e target forms. The reverse prime target pair of -e
-s, however, has unprimed gender and number features in the target, and hence yields a significant reduction in priming. The same is true for -m
-s, due to the unprimed gender feature [-MASC] in the target (in addition to a case feature mismatch between prime and target), and for the pair -m
-e, due to a case feature mismatch ([+OBL] vs. [–OBL]). Table 4. A summary of the priming results on German adjective and finite verb inflection Prime
Target

Unprimed target feature(s)?

Full or reduced Priming

-s
-e -s
-m -e
-m -e
-s -m
-e -m
-s

(kaltes
kalte)

no yes yes yes yes yes

–7ms a –30ms * a –36ms * a –19ms * a –24ms * a –25ms * a

2sg. -st
-e 2pl. -t
-e 3sg. -t
-e

(du) lachst
lache

no no yes

–7ms b –8ms b –28ms *

yes yes yes

–22ms * –21ms * –22ms *

1sg. -e
-st 2pl. -t
-st 3sg. -t
-st a

b b b b

b

Sources: Clahsen et al. 2001a, experiment 2; Clahsen et al. 2001b

Turning to the experiments on finite verbs, note that in all conditions, the primes were presented together with appropriate personal pronouns in order to make it possible for subjects to distinguish 2pl. forms, e.g. (ihr) lackiert, from 3sg. forms, e.g. (er) lackiert, whereas the targets, e.g. lachst and lache were always presented without pronouns, as illustrated in Table 4. The results of experiment 2 from Clahsen et al. (2001b) show that target forms with the 2sg. -st yielded significant reductions in priming (relative to identical repetition) irrespective of which verb form was used as a prime (see Tab.4). This is due to the fact that -st contains a particular feature set ([+2, –Pl]) that is unavailable from any of the primes. On the other hand, -e

380 Harald Clahsen forms used as targets yielded a different pattern of results. Tab.4 shows that -e target forms are effectively primed by 2sg. and 2pl., but not by 3sg. forms. These differences correspond to their paradigmatic representation. In Blevins’ (2003) account, for example, the primes and targets of the pairs 2sg.
-e and 2pl.
-e come from different rule blocks. The target form with the exponent -e is based on a stem-formation rule, whereas the 2sg. and the 2pl. are pure agreement forms. Since the -e stem is compatible with these forms, these prime-target pairs yield effective priming. This is not the case for the prime-target pair 3sg.
-e, and hence the reduced priming effect observed in this case. Recall that in this condition, both prime and target are based on stem-formation rules, which are incompatible with each other: 3sg. forms are said to select the -t stem (e.g. lacht), while 1sg. forms select the secondary stem in -e. Summarizing, we found an effect of paradigmatic specificity in the morphological processing of inflected verbs and adjectives. The experiments show that exponents with highly specific feature sets, e.g. the affix -m in the declensional and the 2sg. affix -st in the conjugational paradigms are more difficult to prime than corresponding exponents with less specific feature sets. More generally, we found reduced priming when the target contains features that are incompatible with those of the prime. Without a paradigmatic analysis of the inflectional systems involved, these priming asymmetries are left unexplained. We therefore conclude that morphological paradigms are not only useful descriptive tools for linguists, but that they also contribute to a better understanding of how inflected word forms are used by the human language processor. 6. Concluding remarks We started out with the question of how the psycholinguist might benefit from theoretical morphology for the experimental study of morphological processing. The results reported in the previous sections lead us to two general conclusions on this issue, firstly that morphological structure does indeed play an important role in the organization of the mental lexicon, and secondly that morphological notions and concepts are not only useful descriptive tools for linguists, but also contribute to a better understanding of how the speaker/hearer processes inflected and derived word forms. A related question that we have not addressed in this chapter is how the morphologist might benefit from experimental psycholinguistic research. Per-

Linguistic perspectives on morphological processing

381

haps experimental findings favour one theory of morphology over its alternatives, and in this way help to resolve theoretical controversies. Consider the findings reported in the present chapter with respect to this issue. The first phenomenon under study concerned the processing of different kinds of stem forms. We found lexicality effects for all kinds of stem forms as well as decomposition effects for combinatorial stems, but not for suppletive ones. The lexicality effects reflect a linguistic property that stems share with roots, namely that both access lexical entries. The decomposition effects support accounts in which combinatorial stems are based on rules that contain variables yielding internally decomposable representations. These results are consistent with the Stem and Paradigm models proposed by Zwicky (1985), Anderson (1992), Aronoff (1994) and Stump (2001) in which stem formation rules are building devices for generating stems from items stored in the lexicon (i.e. roots and suppletive stems). Moreover, if one follows Blevins’ (2003) interpretation of these rules as constraints that permit the deduction of roots and stems from members of a morphological paradigm, our findings could also be made compatible with a Word and Paradigm model (see e.g. Matthews 1991) in which roots and stems are simply recurrent parts within a lexicon of whole words. On the other hand, the experimental effects seen for marked stem forms challenge an approach (developed in the tradition of Generative Phonology) in which stem allomorphy is derived by general phonological rules (see e.g. Halle & Mohanan 1985, Halle & Marantz 1993). Consider the results on Catalan and Spanish again. For marked stem forms, e.g. for diphtongized stems such as duerm- of the Spanish verb dormir ‘to sleep’, we found ERP responses (= N400) that are characteristic of lexical processing, whereas unmarked stem forms, e.g. overapplications of 1st conj. stem formation in Catalan (*dormat instead of dormit ‘slept’), elicited ERP signatures (= LAN) that index combinatorial (rule-based) processes. In addition, Linares, Rodríguez-Fornells, and Clahsen (2006) examining 3rd conj. verbs such as pedir – pido ‘to ask – I ask’ in an ERP violation paradigm found that marked stems elicit N400 responses (rather than LANs). These results do not support the idea that diphthongized stems (as well as the e~i alternation) in Spanish can be derived by general phonological rules (Harris 1985). Moreover, the contrast seen between regular and irrgular verb forms in German (see Table 1) in which, for example, irregular past-tense forms and irregular participles do not prime their corresponding verb roots whereas regular ones do, provides counterevidence to the ‘minor rules’ analysis in Halle & Marantz (1993: 128), in which irregular verb forms are derived by rules applied to verb roots.

382 Harald Clahsen Secondly, we saw that all derived forms (as well as irregularly inflected word forms) showed lexicality effects, but only productively defined derivational processes exhibited the same morphological decomposition effects as regularly inflected word forms. The lexicality effects indicate that (productive) derivation defines entries whereas (regular) inflection is purely form-defining without being cached out in lexical entries. The decomposition effects indicate that productive derivation as well as regular inflection are based on combinatorial rules. These results provide support for the idea that derivational or lexeme formation processes are distinct from, and in some sense prior to, inflectional processes, an idea that has been implemented into different theories of morphology (Kiparsky 1982; Di Sciullio & Williams 1987; among others). The third case we examined was the role of morphological paradigms in language processing. We found that inflectional affixes (or exponents) with highly specific feature sets, e.g. the affix -m in the declensional and the 2sg. affix -st in the conjugational paradigms are more difficult to prime than corresponding forms with less specific feature sets and argued that this provides support for a paradigmatic analysis of these inflectional systems. In morphological theory the status of inflectional paradigms is controversial. Lieber (1992), for example, has argued that paradigms are simply artefacts, parallel to lists of related sentences. Similarly, for Halle & Marantz (1993), paradigms do not have any theoretical status. Other frameworks consider inflectional paradigms as important representational devices (see e.g. Carstairs 1987; Zwicky 1985; Stump 1993; Blevins 2003; Wunderlich 1997; Aronoff 1994). Our results suggest that morphological paradigms are used by the human language processor. While our findings are compatible with both Blevins’ (2000, 2003) exponence-based as well as with Wunderlich & Fabri’s (1995) and Wunderlich’s (1997) affix-based accounts, models of the lexicon which try to do without morphological paradigms are not supported. In conclusion, psycholinguistic experimentation on morphological processing provides evidence that the morphologist may find useful (along with other sources of evidence) in developing descriptive and theoretical analyses for a given set of phenomena. In addition, as seen above, psycholinguistic results may even help to adjudicate between competing theoretical analyses. Clearly, however, it is rarely the case that experimental evidence uniquely favours one particular linguistic analysis and at the same time disconfirms all available alternatives.

Linguistic perspectives on morphological processing

383

Acknowledgements The research reported in this paper has been supported by the Deutsche Forschungsgemeinschaft (DFG) ‘German Research Council’ (grants SFB 282/B7 and C7). The collaboration with Thomas Münte and Antoni Rodríguez-Fornells has been supported by a NATO Collaborative Research Grant. I am indebted to many people who have collaborated with me over the years on studies of morphological processing, most particularly to Sonja Eisenbeiss, Meike Hadler, Ingrid Sonnenstuhl, and Helga Weyerts. I am also grateful to various colleagues for helpful discussions and comments on earlier versions, in particular to Jim Blevins, Claudia Felser, Iggy Roca, Tessa Say, Andrew Spencer, and Dieter Wunderlich. References Anderson, Stephen R. 1992 A-Morphous Morphology. Cambridge: Cambridge University Press. Aronoff, Mark 1994 Morphology by Itself. Cambridge, MA: MIT Press. Balota, David A. 1994 Visual word recognition: The journey from features to meaning. In Handbook of Psycholinguistics, Morton A. Gernsbacher (ed.), 303– 349. San Diego, CA: Academic Press. Bittner, Dagmar and Andreas Bittner 1990 Sind Flexionsklassen nur morphologische Klassen? In Wurzel(n) der Natürlichkeit, A. Bassarak, D. Bittner, A. Bittner and P. Thiele (eds.), 16–36. Berlin: Zentralinstitut für Sprachwissenschaft. Blevins, James P. 1995 Syncretism and paradigmatic opposition. Linguistics and Philosophy 18: 113–152. 2000 Markedness and blocking in German declensional paradigms. In Lexicon in Focus, B. Stiebels and D. Wunderlich (eds.), 83–103. Berlin: Akademie-Verlag. 2003 Stems and paradigms. Language 79: 737–767. Booij, Geert 2002 The Morphology of Dutch. Oxford: Oxford University Press. Butterworth, Brian 1983 Lexical representation. In Language Production. Vol. 2, B. Butterworth (ed.), 257–294. London: Academic Press.

384 Harald Clahsen Bybee, Joan L. 1995 Regular morphology and the lexicon. Language and Cognitive Processes 10: 425–455. Cahill, Lynne and Gerald Gazdar 1997 The inflectional phonology of German adjectives, determiners and pronouns. Linguistics 35: 211–245. Carstairs, Andrew 1987 Allomorphy in Inflexion. London: Croom Helm. Clahsen, Harald 1999 Lexical entries and rules of language: a multi-disciplinary study of German inflection. Behavioral and Brain Sciences 22: 991–1060. Clahsen, Harald, Sonja Eisenbeiss and Ingrid Sonnenstuhl 1997 Morphological structure and the processing of inflected words. Theoretical Linguistics 23: 201–249. Clahsen, Harald, Sonja Eisenbeiss, Meike Hadler and Ingrid Sonnenstuhl 2001a The mental representation of inflected words: an experimental study of adjectives and verbs in German. Language 77: 510–543. Clahsen, Harald, Ingrid Sonnenstuhl, Meike Hadler and Sonja Eisenbeiss 2001b Morphological paradigms in language processing and language disorders. Transactions of the Philological Society 99: 247–277. Clahsen, Harald, Peter Prüfert, Sonja Eisenbeiss and Joana Cholin 2002a Strong stems in the German mental lexicon: evidence from child language acquisition and adult processing. In More than Words: A Festschrift for Dieter Wunderlich, I. Kaufmann and B. Stiebels (eds.), 91–112. Berlin: Akademie-Verlag. Clahsen, Harald, Fraibet Aveledo and Iggy Roca 2002b The development of regular and irregular verb inflection in Spanish child language. Journal of Child Language 29: 591–622. Clahsen, Harald, Ingrid Sonnenstuhl and James P. Blevins 2003 Derivational morphology in the German mental lexicon: a dual mechanism account. In Morphological Structure in Language Processing, H. Baayen and R. Schreuder (eds.), 125–155. Berlin /New York: Mouton de Gruyter. Di Sciullo, Anna-Maria and Edwin Williams 1987 On the Definition of Word. Cambridge, MA: MIT Press. Feldman, Laurie B. and Carol A. Fowler 1987 The inflected noun system in Serbo-Croatian: Lexical representation of morphological structure. Memory and Cognition 15: 1–12. Friederici, Angela D., Anja Hahne and Axel Mecklinger 1996 Temporal structure of syntactic parsing: early and late event-related brain potential effects. Memory and Cognition 22: 1219–1248. Halle, Morris and K. P. Mohanan 1985 Segmental phonology of modern English. Linguistic Inquiry 16: 57– 116.

Linguistic perspectives on morphological processing

385

Halle, Morris and Alec Marantz 1993 Distributed Morphology and the pieces of inflection. In The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger, K. Hale and J. Keyser (eds.), 111–176. Cambridge, MA: MIT Press. Harris, James 1985 Spanish diphthongization and stress: a paradox resolved. Phonology Yearbook 2: 31–45. Jackendoff, Ray 1997 The Architecture of the Language Faculty. Cambridge, MA: MIT Press. Joanisse, Marc F. and Mark S. Seidenberg 1999 Impairments in verb morphology after brain injury: a connectionist model. Proceedings of the National Academy of Sciences 96: 7592– 7597. Kiparsky, Paul 1982 From cyclic phonology to lexical phonology. In The Structure of Phonological Representations, H. van der Hulst and N. Smith (eds.), Part 1, 131–175. Dordrecht: Foris. 1998 Paradigm effects and opacity. Ms., Stanford University. Kostic, Aleksandar 1995 Information load constraints on processing inflected morphology. In Morphological Aspects of Language Processing, L. B. Feldman (ed.), 317–344. Hillsdale, NJ: Erlbaum. Kutas, Marta and Robert Kluender 1994 What is who violating: A reconsideration of linguistic violations in light of event-related brain potentials. In Cognitive Electrophysiology, H.-J. Heinze, T. Münte and G. Mangun (eds.), 181–210. Boston: Birkhäuser. Kutas, Marta and Bernadette M. Schmitt 2003 Language in microvolts. In Mind, Brain and Language, M. T. Banich and M. Mack (eds.), Mahwah, NJ: Erlbaum. Kutas, Martha and C. K. Van Petten 1994 Psycholinguistics electrified. Event-related brain potential investigations. In Handbook of Psycholinguistics, Morton A. Gernsbacher (ed.), 83–143. San Diego: Academic Press. Lieber, Rochelle 1992. Deconstructing Morphology: Word Formation in Syntactic Theory, Chicago, IL: University of Chicago Press. Linares, Rafael, Antoni Rodríguez-Fornells and Harald Clahsen 2006 Stem allomorphy in the Spanish mental lexicon: Evidence from behavioral and ERP experiments. Brain and Language 97: 110–120. Lukatela, G., C. Carello and Michael T. Turvey 1987 Lexical representation of regular and irregular inflected nouns. Language and Cognitive Processes 2: 1–17.

386 Harald Clahsen Manelis, Leon and David A. Tharp 1977 The processing of affixed words. Memory and Cognition 5: 690– 695. Marcus, Gary F. 2001 The Algebraic Mind: Integrating Connectionism and Cognitive Science. Cambridge, MA: MIT Press. Marslen-Wilson, William D. and Lorraine K. Tyler 1998 Rules, representations and the English past tense. Trends in Cognitive Sciences 2: 428–435. Matthews, Peter H. 1991 Morphology. 2nd edition. Cambridge: Cambridge University Press. Miller, George A. and Noam Chomsky 1963 Finitary models of language users. In Handbook of Mathematical Psychology, vol. II, R. D. Luce, R. R. Bush and E. Galanter (eds.), 419–492. New York: Wiley. Münte, Thomas F., Hans-Jochen Heinze and Ron Mangun 1993 Dissociation of brain activity related to syntactic and semantic aspects of language. Journal of Cognitive Neuroscience 5: 335–344. Münte, Thomas F., E. Anvari, Marjori Matzke and S. Johannes 1995 Brain potential correlates of number errors in the Turkish language. Neuroscience Letters 199: 57–60. Münte, Thomas F., Tessa Say, Kolja Schiltz, Harald Clahsen and Marta Kutas 1999 Decomposition of morphologically complex words in English: Evidence from event-related brain potentials. Cognitive Brain Research 7: 241-253. Neville, Helen, Janet Nicol, A. Barss, Ken Forster and Merrill F. Garrett 1991 Syntactically based sentence processing classes: evidence from eventrelated brain potentials. Journal of Cognitive Neuroscience 3: 151– 165. Orsolini, Margherita, Rachele Fanari and Hugo Bowles 1998 Acquiring regular and irregular inflection in a language with verb classes. Language and Cognitive Processes 13: 425–464. Osterhout, Lee and Linda A. Mobley 1995 Event-related brain potentials elicited by failure to agree. Journal of Memory and Language 34: 739–773. Osterhout, Lee and Phillip J. Holcomb 1995 Event-related potentials and language comprehension. In Electrophysiology of Mind: Event-Related Brain Potentials and Cognition, M. Rugg and M. Coles (eds.), 175–215. Oxford University Press. Penke, Martina, Helga Weyerts, Matthias Gross, Elke Zander, Thomas F. Münte and Harald Clahsen 1997 How the brain processes complex words: an event-related potential study of German verb inflections. Cognitive Brain Research 6: 37–52.

Linguistic perspectives on morphological processing

387

Phillips, Colin 1996 Order and structure. PhD diss., MIT. Pinker, Steven 1999 Words and Rules. New York: Basic Books. Pinker, Steven and Alan Prince 1988 On language and connectionism: analysis of a parallel distributed processing model of language acquisition. Cognition 28: 73–193. Raveh, Michael and Jay G. Rueckl 2000 Equivalent effects of inflected and derived primes: Long-term morphological priming in fragment completion and lexical decision. Journal of Memory and Language 42: 103–119. Rodríguez-Fornells, Antoni, Harald Clahsen, C. Conxita Lleo, Wanda Zaake and Thomas F. Münte 2001 Event-related brain responses to morphological violations in Catalan. Cognitive Brain Research 11: 47–58. Rodríguez-Fornells, Antoni, Thomas F. Münte and Harald Clahsen 2002 Morphological priming in Spanish verb forms: An ERP repetition priming study. Journal of Cognitive Neuroscience 14: 443–454. Rueckl, Jay G., Michelle Mikolinski, Michael Raveh, Caroline S. Miner and F. Mars 1997 Morphological priming, fragment completion, and connectionist networks. Journal of Memory and Language 36: 382–405. Rumelhart, David E. and James L. McClelland 1986 On learning the past tenses of English verbs. In Parallel Distributed Processing, vol. 2, J. L. McClelland, D. E. Rumelhart and the PDP Research Group (eds.), 216–271. Cambridge, MA: MIT Press. Say, Tessa and Harald Clahsen 2002 Words, rules and stems in the Italian mental lexicon. In Storage and Computation in the Language Faculty, S. Nooteboom, F. Weerman and F. Wijnen (eds.), 93–122. Dordrecht: Kluwer. Sereno, Joan A. and Allard Jongman 1997 Processing of English inflectional morphology. Memory and Cognition 25: 425–437. Sonnenstuhl, Ingrid, Sonja Eisenbeiss and Harald Clahsen 1999 Morphological priming in the mental lexicon: evidence from German. Cognition 72: 203–236. Sonnenstuhl, Ingrid and Axel Huth 2002 Processing and representation of German -n plurals: A dual mechanism approach. Brain and Language 81: 276–290. Sonnenstuhl, Ingrid and Harald Clahsen in prep. Prefixed word forms in the German mental lexicon. Spencer, Andrew 1991 Morphological Theory. Oxford: Blackwell.

388 Harald Clahsen Stanners, Robert F., James J. Neiser, William P. Hernon and Roger Hall 1979 Memory representation for morphologically related words. Journal of Verbal Learning and Verbal Behavior 18: 399–412. Stump, Gregory T. 1993 On rules of referral. Language 69: 449–479. 2001 Inflectional Morphology: A Theory of Paradigm Structure. Cambridge University Press. Taft, Marcus 1988 A morphological decomposition model of lexical representation. Linguistics 26: 657–667. Taft, Marcus, G. Hambly and S. Kinoshita 1986 Visual and auditory recognition of prefixed words. The Quarterly Journal of Experimental Psychology 38: 357–366. Todorovic, Dejan 1988 Hemispheric differences in case processing. Brain and Language 33: 365–389. Weyerts, Helga, Martina Penke, Ulrike Dohrn, Harald Clahsen and Thomas F. Münte 1997 Brain potentials indicate differences between regular and irregular German plurals. NeuroReport 8: 957–962. Wickelgren, Wayne A. 1969 Context-sensitive coding, associative memory, and serial order in (speech) behavior. Psychological Review 76: 1–15. Wunderlich, Dieter 1997 Der unterspezifizierte Artikel. In Sprache im Fokus, C. Dürscheid, K.-H. Ramers and M. Schwarz (eds.), 47–55. Tübingen: Niemeyer. Wunderlich, Dieter and Ray Fabri 1995 Minimalist Morphology: An approach to inflection. Zeitschrift für Sprachwissenschaft 14: 236–294. Zwicky, Arnold M. 1985 How to describe inflection. Berkeley Linguistics Society 11: 372–386. 1986 The general case: Basic form versus default form. Berkeley Linguistics Society 12: 305–315.

The representation of inflectional morphology in the mental lexicon: An overview on psycho- and neurolinguistic methods and results Martina Penke

In many languages and inflectional systems regular and irregular inflected forms exist side by side. Consider for instance English past tense formation, where we find regular forms inflected with -ed (laugh – laughed) and irregular ones like went that are idiosyncratic and largely unpredictable. According to an influential view in linguistics and psycholinguistics – most prominently defended by Pinker (1999) – regular inflected forms are built by application of a mental symbolic rule (add -ed), whereas irregular forms are stored in the mental lexicon. The distinction between regular and irregular inflected forms has been argued to exemplify the distinction between a mental lexicon, where words are stored together with learned idiosyncratic information, and a mental grammar component that contains the rules to generate composite structures such as sentences and complex words out of the stored elements in the mental lexicon (cf. Pinker 1999; Clahsen 1999; Ullman 2001). The dualistic view, which states that regular and irregular inflection are based on qualitatively different representations and mental operations, has not gone undisputed. Based on modelling data that come from connectionist network models, unitary connectionist approaches to inflection have denied that regular inflection is based on a mental operation that combines a verbal stem and an affix. Instead they assume that regular as well as irregular inflected forms are stored as fully inflected forms in an associative network structure (Rumelhart & McClelland 1986, for an overview on such models confer Elman et al. 1996; Marcus 2001). The debate between unitary and dualistic approaches to inflection has been a central topic in psycho- and neurolinguistic work for the last 20 years. During this time a number of different experimental effects and techniques have been used to test the different predictions that derive from these two approaches. The rationale behind such psycho- and neurolinguistic investigations is that they make use of specific experimental effects that have been established in methodological work. Once an effect such as the frequency effect, the priming effect or the ungrammaticality effect has been

390 Martina Penke established, it can then be used to test certain predictions derived from the conflicting theories under investigation. The presence, absence, or strength of the tested effect in the experimental data is then interpreted as evidence for or against a given theoretical model. An overview of the extensive research that has been carried out on the representation and computation of inflectional morphology until now is beyond the scope of this article (see Clahsen 1999; Pinker 1999; Penke 2006 for overview). Instead, my aim is to introduce some of the relevant experimental effects that can be used to explore representations and mechanisms involved in inflectional morphology and to present some of the major results obtained in our research groups while using these methods. 1.

Distinction between regular and irregular inflected forms

A central tenet of dualistic approaches to inflection such as Pinker’s DualMechanism Model (Pinker & Prince 1988; Pinker 1999) or the theoretical model of Minimalist Morphology (Wunderlich & Fabri 1995; Wunderlich 1996a,b) is that the representations and mechanisms involved in the processing of regular and irregular inflectional forms are fundamentally different. Unitary approaches to inflection, in contrast, claim that all forms, regular and irregular ones, are represented in an associative, non-compositional network structure. Hence proponents of the dualistic view of inflection have tried to provide experimental evidence that the mechanisms and representations involved in regular and irregular inflection are qualitatively distinct. 1.1. Selective deficits – indicative of qualitatively distinct and independent modules A first prediction of the dualistic view of inflection is the following: If the dualistic view holds and qualitatively distinct mental representations and operations underlie regular and irregular inflection, then we should find language deficits that selectively affect only one type of inflection, but not the other. This prediction rests on four assumptions on the nature of the human language capacity and its breakdown following brain lesions. According to the assumptions of autonomy and modularity (Fodor 1983), the human language faculty is autonomous from other cognitive domains or modules (i.e. it is domain-specific) and consists of task-specific and independent sub-

The representation of inflectional morphology in the mental lexicon

391

modules carrying out different types of computations on different types of representations. Under this view, regular and irregular inflection are dependent on two independent modules of the human language faculty: a computational component, where affixation is carried out, and a component in which stored information is retrieved from the mental lexicon. According to the fractionation assumption (Caramazza 1984), brain damage can result in the selective impairment of specific submodules of the language capacity, such as the ones underlying regular and irregular inflection. The transparency assumption (Caramazza 1984) states that the components unaffected by the brain lesion will continue to function as normal such that the output of the language system will directly reflect the impairment of the affected component. Under this view, the impaired language system can be conceived of as the intact system minus the affected submodule (see Kosslyn & Van Kleek 1990; Caramazza 1992; Penke 2006 for discussion). Given that regular and irregular inflection are subserved by two autonomous submodules of the language system, the assumptions of fractionation and transparency predict that we should find language disorders that either selectively affect the regular inflectional component with the irregular inflectional module spared or, vice versa, selectively affect the irregular inflectional component leaving the regular submodule unimpaired. A failure in finding such a selective deficit would, on the other hand, weaken the dualistic view on inflection. Whether or not language disorders can be found that selectively affect specific language submodules, thus, constitutes a test for the dualistic approach to inflection. Agrammatic Broca’s aphasia, a language impairment predominantly caused by strokes that affect left frontal brain regions, is a good candidate to examine a selective disorder of one type of inflection because it is characterised by omissions and substitutions of bound inflectional morphology (e.g. Friederici 1984; Menn & Obler 1990: 14, Obler & Gjerlow 1999). Particularly relevant is the investigation of inflectional deficits in such languages where both regular and irregular inflected forms are marked by overt inflectional endings. Consider English, where regular inflected forms have inflectional endings, whereas irregular forms often only display a change of the stem vowel (e.g. sing – sang). This constitutes a confound: a deficit affecting only regular inflection cannot be distinguished from a deficit affecting inflectional endings in general, since the only inflectional endings are regular affixes. We, therefore, investigated participle formation in German, where regular as well as irregular forms are marked by different inflectional endings.

392 Martina Penke 1.1.1. Language background – German participle formation German past participles are built by a phonologically conditioned prefix ge-, a verb stem, and the endings /t/ or /n/. Regular inflected (weak) verbs do not show stem changes in the participle form and have the ending /t/ (1a-b). In contrast, irregular inflected (strong) verbs often show a modification of the stem vowel in the participle stem (Ablaut) and take the participle ending /n/ (1c). Neither the stem’s vowel nor the phonological shape of the stem predict whether a verb is regular or irregular. Consider for example the verbs blinken ‘flash’ and trinken ‘drink’: whereas trinken (1c) has the irregular participle form getrunken, the phonologically similar verb blinken (1b) has the regular participle form geblinkt. (1)

Infinitive a. tanz-en b. blink-en c. trink-en

Participle ge-tanz-t ge-blink-t ge-trunk-en

Gloss ‘dance’ ‘flash’ ‘drink’

According to the dualistic approach to inflection, two qualitatively different cognitive mechanisms are responsible for the inflection of regular and irregular participles: whereas regular participles such as getanzt ‘danced’ (1a) are built on the fly by combining the basic verb stem tanz- with the participle affix -t, irregular participles such as getrunken ‘drunk’ (1c) are stored as fully inflected forms in the mental lexicon. A theoretical spell-out of the dualistic view is given in Minimalist Morphology (Wunderlich & Fabri 1995; Wunderlich 1996a). In this theory, irregular participles are stored in structured lexical entries that spread from an unmarked base node (2a). In contrast, entries for regular verbs consist only of the base node (2b) and participles are built by regular affixation of the affix -t to the base. (2)

trink- [+ VERB] ‘drink’ qp trunken [+ PARTICIPLE] trank- [+ PRETERITE] ! tränk- [+ SUBJUNCTIVE] a.

b. tanz- [+ VERB] ‘dance’ Note that according to Minimalist Morphology there is a qualitative distinction between the participle endings /n/ and /t/. Whereas the irregular ending

The representation of inflectional morphology in the mental lexicon

393

/n/ is stored together with information about the changes in the verb’s stem in the participle node of a structured lexical entry of a strong verb, the regular ending /t/ is an affix with its own lexical entry (see §3 on affix entries).1 1.1.2. Example case: Regular and irregular participle inflection in German Broca’s aphasia We elicited participle forms for 39 regular and 39 irregular verbs, carefully matched for word and participle frequency, from 13 German individuals with agrammatic Broca’s aphasia. The subjects had to complete a sentence by transforming a given 1st person singular present tense verb form that was introduced in a short sentence (e.g. Ich schreibe. ‘I am writing.’) into a past participle (Ich habe __ ?__ [geschrieben] ‘I have written.’). The elicitation of forms or constructions under interest is often to be preferred to a collection of spontaneous speech data. For one, elicited data will contain more of the relevant constructions or forms while influencing factors, such as frequency of the word or inflected form, can be controlled for. Also, comparisons between elicited and spontaneous speech data have shown that constructions or forms whose production poses a problem for a language impaired speaker are often avoided in spontaneous speech production and only show up when the context provided in the elicitation task requires the formation of a specific form or construction (Kolk & Heeschen 1992; Penke 2006). A comparison of the correctness scores for regular and irregular participles reveals that regular inflection is largely spared in the 13 agrammatic subjects (see fig. 1). On average, only 8.8% of the regular inflected participles are not produced correctly. In contrast, with an average of 33% error rates for irregular participles are significantly higher (t(12) = .057, p = .002). As can be seen from the individual data in figure 1, ten of the 13 aphasic subjects have problems in participle inflection that selectively affect irregular participles. Only three of the 13 subjects (subjects GB to FW) display comparable problems with regular and irregular participle formation (for a detailed ———–———————————————————————————— 1

An affix entry for the irregular /n/ ending is rejected on the grounds of economy considerations. An affix entry for the /n/ ending would have to contain the information that it applies on the participle stem of a strong verb and marks the resulting verb form as a participle form. Affixation that adds the morphosyntactic information [+PARTICIPLE] to a stem already marked for this information is, however, simply redundant (cf. Wunderlich & Fabri 1995).

394 Martina Penke description of procedure and data and for similar results on Dutch Broca’s aphasia see Penke, Janssen & Krause 1999, Penke & Westermann in press).2

Figure 1. Error rates for the production of regular and irregular German participles in 13 subjects with Broca’s aphasia.3

Deficits that selectively affect only regular respectively only irregular inflection have meanwhile been observed across languages in a number of acquired and developmental language disorders such as agrammatic Broca’s aphasia (Ullman et al. 1997; Penke, Janssen & Krause 1999), progressive semantic aphasia (Cholewa & de Bleser 1995; Patterson et al. 2001), Specific Language Impairment (SLI) (Gopnik & Crago 1991; van der Lely & ———–———————————————————————————— 2

3

Note that it is not necessary to find a selective deficit in each individual subject of the group under study. Brain lesions associated with language disorders such as agrammatic Broca’s aphasia always show some variation across individuals. Therefore, a brain lesion might not only affect the irregular inflectional component, but also the regular inflectional submodule. Error rates were calculated as follows:
incorrectly inflected regular (respectively irregular) participles
analysable regular (respectively irregular) participles Forms that were not overtly marked as participles, i.e. forms which were neither marked with the prefix ge- nor with the participle endings /t/ or /n/ nor with an ablauting stem, were not included in this analysis. A produced participle form was counted as incorrect if a wrong ending was used or if the ending was omitted.

The representation of inflectional morphology in the mental lexicon

395

Ullman 2001) or Williams syndrome (Clahsen & Almazan 1998, 2001, Penke & Krause 2004). Whereas the existence of such deficits is no longer disputed, the controversy now centres on the question how to interpret selective deficits of regular respectively irregular inflected forms. Proponents of dualistic accounts to inflection argue that selective deficits provide evidence against unitary accounts to inflection, since it seems hard to explain how a unitary mechanism can be distorted in such a way as to only affect one type of inflected forms, but not the other (e.g. Pinker 1999; Ullman et al. 1997). This reasoning, however, does not necessarily hold true as selective deficits can in principle also result from non-modular representations, for instance from a deficit that only affects part of a system where information is represented on a continuum (cf. Shallice 1988). Accordingly, a number of connectionist models have been proposed that attempt to simulate selective deficits within unitary associative representations of inflected forms (e.g. Marchman 1993; Bullinaria & Chater 1995; Joanisse & Seidenberg 1999; Juola & Plunkett 2000; Westermann 2000; Penke & Westermann in press). Thus, whereas the observation of deficits that selectively affect only regular respectively only irregular inflected forms confirms a prediction of the dualistic approach to inflection, it is, in principle, compatible with both – a modular dualistic view and a unitary account of inflection (see Penke 2006 for discussion). 1.2. ERP components: Different components are indicative of different cognitive processes Another neurolinguistic method that can be used to investigate whether or not the representations and computations involved in regular and irregular inflected forms are qualitatively distinct makes use of so-called event-related potentials (ERPs). In ERP studies the electrophysiological activity of a pool of neurons that is correlated in time with performing a cognitive task is recorded from electrodes placed on standard positions on the scalp. The electrical activity that is correlated with a cognitive function has to be extracted from the surrounding electrophysiological activity of the brain that is not correlated with this function (the background EEG activity). This is done by averaging over those segments of the EEG signal that are timelocked to the processing of items of a specific experimental condition. An ERP waveform is typically obtained from the averaged data of 20 to 50 trials. It reflects the neural activity time-locked to the processing of the items of a given experimental condition.

396 Martina Penke According to the invariance assumption commonly accepted in functional neuroimaging, the mapping between a cognitive operation and its neural substrate is invariant, such that a specific cognitive function will be instantiated in the brain in only one way (Rugg 1999). Two experimental conditions that are correlated with two qualitatively different ERP patterns, therefore provide evidence that the different ERP patterns reflect qualitatively different cognitive operations. In this way, ERPs can be used to test whether or not the processing of different types of stimuli involves qualitatively distinct cognitive operations. ERP experiments often make use of the so-called violation paradigm where ERPs associated to a grammatically correct and semantically appropriate stimulus condition are compared to ERPs related to an inappropriate or incorrect stimulus condition. In their seminal work, Kutas & Hillyard (1980) used this paradigm to compare ERPs related to semantically appropriate sentences (e.g. The pizza was too hot to eat.) to ERPs associated with semantically inappropriate sentences (e.g. The pizza was too hot to cry). In comparison to the semantically appropriate sentences, semantically inappropriate sentences led to a more negative waveform, reaching its maximum amplitude at about 400 ms. An ERP waveform that displays a specific shape with respect to parameters such as amplitude, polarity, latency, and topographical distribution of the recorded electrophysiological activity and that is typically related to a specific type of experimental manipulation is called an ERP component. Such components are labelled according to certain characteristics of their waveform. A typical ERP component related to semantic violations is the N400 mentioned above that leads to a negativity with a maximum amplitude after 400 ms. Other language-related components discussed in the ERP literature are LAN, ELAN, and P600 that are associated with grammatical violations (see e.g. Hagoort, Brown & Osterhout 1999 for overview). A more thorough introduction of the ERP method can be found, for instance, in Kutas & Van Petten (1994). We made use of the ERP technique to investigate whether or not the processing of regular and irregular inflected forms is associated with distinct ERP patterns (cf. Penke et al. 1997). Under the invariance assumption, distinct ERP patterns would provide evidence that the processing of regular and irregular inflected forms is subserved by different cognitive operations, as assumed in dualistic approaches to inflection. Unitary approaches to inflection that assume similar representations and computations for regular and irregular inflected forms would, in contrast, predict similar ERP responses for the two types of inflected forms.

The representation of inflectional morphology in the mental lexicon

397

1.2.1. Example case: Neural correlates of regular and irregular German participle formation We measured ERPs from 20 right-handed German students (11 women, 9 men, mean age 24.8 years) while subjects silently read a set of unrelated sentences each including a participle. The task assigned to the subjects was to judge whether a ‘test-sentence’ presented every ten sentences was or was not an exact repetition of one of the previous ten sentences. Participles differed in type (regular vs. irregular) and ending (correct vs. incorrect ending), resulting in four different experimental conditions, see (3, 4): (3)

a. regular participle with correct /t/ ending: Sie haben die ganze Nacht durchgetanzt. ‘They have danced the whole night through.’ b. regular participle with incorrect /n/ ending: *Sie haben die ganze Nacht durchgetanzen.

(4)

a. irregular participle with correct /n/ ending: Sie haben die Möbel schon aufgeladen. ‘They have already loaded the furniture.’ b. irregular participle with incorrect /t/ ending: *Sie haben die Möbel schon aufgeladet.

Fifty sentences per condition were tested. All test sentences were randomized and presented one word at a time on a computer screen. To restrict the differences between correct and incorrect participles to the ending, we selected only irregular verbs which do not show verbal Ablaut in the participle form (e.g. lad- – geladen ‘load – loaded’). To control for frequency effects, we only presented participles with low frequency of participle form. To sample fifty irregular verbs with low frequency and non-ablauting participle form, we made use of prefix verbs which combine a prefix such as auf ‘on’ or durch ‘through’ with a base verb. The same participles were presented with a correct and an incorrect ending in an otherwise identical sentence. Each sentence consisted of six words including the participle which was placed constantly at the end of the sentence. The EEG was recorded from 19 standard points on the scalp. ERPs averaged over all subjects and over the fifty trials of each experimental condition were obtained separately for each electrode site and each stimulus condition. Figure 2 shows the ERP waves recorded on the left frontal electrode F7, for which the maximal effects were observed.

398 Martina Penke For regular participles, waveforms for participles with the incorrect /n/ ending are not significantly different from waveforms for participles with the correct /t/ ending. The slight difference between the two waveforms that can be seen in figure 2 proved to be inconsistent in two follow-up studies where participles were presented in texts and word lists (cf. Penke et al. 1997). A consistent and significant effect was, however, found for irregular participles. Compared to correct irregular participles, irregular participles with the incorrect /t/ ending are associated with a long-lasting more negative waveform, starting at 200 ms after stimulus presentation and reaching a maximum after 250 to 500 ms after presentation. Latency and topography of this negativity suggest that it is an instance of an ERP component called LAN, due to the left anterior distribution of this negativity (cf. e.g. Hagoort, Brown & Osterhout 1999).

Figure 2. ERPs for correct and incorrect regular and irregular participles on left anterior electrode site F7

According to the invariance assumption, the differences in the waveforms observed for regular and irregular participles support the view that two different cognitive processes are involved in the processing of regular and irregular participles. Note, that the incorrect participles of both, regular and irregular verbs are equally ungrammatical and readily rejected by native speakers of German. Yet they are associated with different ERP effects: whereas incorrect irregular participles are associated with a LAN, incorrect regular participles do not elicit consistent and significant ERP effects. Based on an additional study on German noun plurals (Weyerts et al. 1997), we suggest that the LAN, we observe for incorrect irregular participles, can be interpreted as a reflex of incorrect application of regular affixation.

The representation of inflectional morphology in the mental lexicon

399

Summarising, the different ERP waveforms correlated with the processing of regular and irregular inflected participles are not compatible with unitary approaches to inflection that assume similar representations and computations for regular and irregular inflected forms. Given the invariance assumption, the different waveforms provide evidence that two different cognitive operations are involved in regular and irregular inflection – as proposed in dualistic approaches to inflection. The finding that regular and irregular inflection can be selectively affected in neurological disorders further supports the dualistic view, since it confirms the dualistic model’s prediction that the independent modules involved in inflection should fractionate in certain brain lesions, leading to selective deficits. 2.

The nature of regular and irregular inflection

According to dualistic approaches to inflection, two qualitatively different cognitive mechanisms are involved in regular and irregular inflection: whereas regular inflected forms such as the regular participle getanzt ‘danced’ are combined via productive affixation of the basic verb stem and the participle affix -t, irregular inflected forms such as the participle getrunken ‘drunk’ are stored as fully inflected forms in the mental lexicon (cf. Wunderlich & Fabri 1995; Wunderlich 1996a). Evidence that regular inflected forms are composed out of a stem and an affix, whereas irregular inflected forms are not, comes from psycholinguistic studies making use of the priming and the frequency effect. 2.1. The priming effect: Cross-modal priming yields evidence for morphological decomposition The priming effect occurs, for instance, in lexical-decision tasks where subjects have to decide as quickly as possible whether a visually or auditorily presented word is an existing word or not. The time necessary to make this decision is measured. Is the same word presented again during the experiment, the lexical-decision time for this word is shorter compared to its first presentation. The difference in decision time observed for the first and the second presentation of a given word is called the priming effect. Priming is explained as follows: The activation of a lexical entry lowers the activation threshold for this entry. Therefore, lexical-access times are quicker when

400 Martina Penke the word is presented again (for a thorough discussion of the priming effect see Balota 1994). A shorter reaction time is also observed when inflected forms such as the German verb form lache ‘laugh[1SG]’ are presented for a second time in a lexical-decision experiment. The reaction time for the second identical presentation of the inflected form gives a baseline for the magnitude of the priming effect (identity condition).

Figure 3. Priming effect in a cross-modal priming task

The priming effect has been used to investigate whether regular and irregular inflected forms are decomposed into stem and affix. The method most often used in such experiments is the cross-modal priming paradigm. In a cross-modal priming task, the reaction times to a visually presented target word are evaluated in relation to different types of auditorily presented prime words. Such experiments usually contain an identity condition, where prime and target words are identical, and a control condition, where prime and target words are semantically, morphologically, and phonologically un-

The representation of inflectional morphology in the mental lexicon

401

related (see fig. 3 for an example of the set-up of a priming experiment). The reaction time obtained for a given target (e.g. Zeh ‘toe’) is quicker if the target is preceded by an identical prime word, compared to a condition (tested with a different group of subjects) where the target Zeh is preceded by an unrelated prime (e.g. Kuh ‘cow’). The cross-modal presentation of prime and target ensures that phonological or orthographical similarity of prime and target do not influence reaction times, as would be possible if both were presented auditorily or visually. 2.1.1. Example case: Cross-modal priming for regular and irregular German participles Sonnenstuhl, Eisenbeiss & Clahsen (1999) used the cross-modal priming paradigm to investigate whether or not regular and irregular inflected German participles are decomposed into stem and affix. In addition to the identity and control condition discussed above, Sonnenstuhl, Eisenbeiss & Clahsen also tested a morphological condition. In this condition, they presented a regular inflected participle such as geplant ‘planned’ as prime and a regular inflected 1st singular present tense verb form such as plane ‘plan[1SG]’ as the target. In their experiment, they compared the lexical-decision times obtained for regular inflected 1st person singular verb forms in three experimental conditions: – identity condition: prime plane ‘plan[1SG]’, target plane ‘plan[1SG]’; – control condition: prime schätze ‘estimate[1SG]’, target plane; – morphological condition: prime geplant ‘plan[PART]’, target plane. In addition to 21 weak verbs such as plan- ‘plan’ which have a regular inflected participle form (e.g. geplant), Sonnenstuhl, Eisenbeiss & Clahsen also tested 21 strong verbs such as schlaf- ‘sleep’. Whereas the 1st person singular form of these verbs is regularly inflected (schlafe ‘sleep[1SG]’), their participle form is irregular (geschlafen ‘slept’). To keep the regular and irregular morphological condition as similar as possible, Sonnenstuhl, Eisenbeiss & Clahsen only tested irregular participles that – like the verb schlafen – do not show stem changes in the participle form. Figure 4 presents the results for regular and irregular inflected participles in the three relevant experimental conditions.

402 Martina Penke

Figure 4. Priming results for regular and irregular participles according to Sonnenstuhl, Eisenbeiss & Clahsen (1999)

Consider first the data for regular participles. The mean lexical-decision time for a target such as plane does not differ between the morphological condition (prime geplant, target plane: 578 ms) and the identity condition (prime plane, target plane: 575 ms) (t-test comparison, p = .956). Reaction times in these two conditions are however significantly shorter than in the control condition (identity vs. control condition t-test, p = .002, morphological vs. control condition t-test, p = .000). This finding indicates that the regular participle form (e.g. geplant) is as good a prime for the inflected form plane as is the form plane itself. For irregular participles, in contrast, the mean lexical-decision time in the morphological condition (prime geschlafen, target schlafe) is significantly longer compared to the identity condition where schlafe is presented twice (t-test, p = .001). These results indicate that the irregular participle geschlafen does not prime the inflected form schlafe as effectively as the regular participle geplant primes the inflected form plane. Under the dualistic view of inflection, the explanation for these findings is straightforward. A regular participle such as geplant is decomposed into its constituent parts, the stem plan- and the participle affix -t, during lexical access. The stem plan- then activates its entry in the mental lexicon. Due to this activation, the activation threshold for this entry is lowered. When the regularly inflected form plane is presented subsequently, this form, too, is decomposed into the stem plan- and the 1st person singular affix -e. Since

The representation of inflectional morphology in the mental lexicon

403

the stem plan- has already been activated before and its activation threshold has been lowered, the access to this stem’s entry is now faster – a priming effect occurs. Irregular inflected participles such as geschlafen, in contrast, are not decomposed into stem and affix, since they are stored as whole forms in the mental lexicon. Thus, the presentation of the irregular participle geschlafen does not lead to a direct activation of the stem entry schlaf-. Subsequent presentation of the regular inflected form schlafe that is decomposed into stem and affix does not lead to a comparable priming effect, since the stem schlaf- has not been directly activated by the participle presented before.4 Summarising, the priming effect observed for regular, but not irregular participles indicates that regular inflected forms are decomposed into stem and affix, as proposed in dualistic approaches to inflection. 2.2. The frequency effect: The lexical-decision task as a window to lexical storage The priming experiment on German participles provides evidence that regular and irregular inflection are qualitatively different. Regular but not irregular forms are decomposed into stem and affix. Dualistic models of inflection such as Minimalist Morphology (Wunderlich & Fabri 1995; Wunderlich 1996a) or Pinker’s Dual-Mechanism Model (Pinker & Prince 1988; Pinker 1999) assume that the representations and mechanisms involved in the pro———–————————————————————————————

4

A reviewer of this paper commented that basing claims on experimental data as provided, for instance, by the priming experiment discussed above requires a detailed “explanation on the mechanism of morphological comprehension that is operative in the experiments”. However, the focus of this paper is not to provide a model of language comprehension, but to present what the available data might tell us about the representation of regular and irregular inflected forms. Proceeding in this way, our interest is in whether an established effect such as the priming effect will occur in two distinct experimental conditions, such as a condition testing regular inflected forms and a condition testing irregular inflected forms. The priming task has shown that irregular participles do not prime inflected forms as effectively as regular participles. This difference asks for an explanation. The common explanation for morphological priming is that an inflected form is decomposed into stem and affix (Balota 1994). That regular participles display a priming effect indicative of morphological decomposition, whereas irregulars do not display a similar priming effect is compatible with the dualistic view on inflection according to which regular, but not irregular inflected forms consist of two independent components – the base stem and an affix.

404 Martina Penke cessing of regular and irregular inflectional forms are fundamentally different. Whereas regular inflected forms consist of stem and affix that are combined by productive affixation, irregular forms are stored as fully inflected forms in the mental lexicon. This allows for another prediction that can be experimentally tested. If irregular forms are stored in the mental lexicon, we should find evidence for this full-form storage. An experimental effect that is generally seen as indicative of lexical storage is the frequency effect that can, for instance, be measured in a lexical-decision experiment. In a lexical-decision experiment, subjects have to decide as quickly and accurately as possible whether a presented item is an existing word or not. The reaction time required to carry out this word-nonword discrimination task is measured (see fig. 5 for an example of the set-up of a lexical-decision experiment). Lexical-decision times are affected by frequency (see Balota 1994; Lively, Pisoni & Goldinger 1994 for review), i.e. subjects take less time to decide that a frequent item (such as Katze ‘cat’ in fig. 5) is an existing word than they take for infrequent words (such as the phonologically similar word Tatze ‘paw’). This effect reflects the assumption that memory traces get stronger with each exposure, making frequent forms easier accessible than infrequent ones.

Figure 5. Frequency effect in a lexical-decision experiment

The representation of inflectional morphology in the mental lexicon

405

The lexical-decision technique has been used to provide evidence for a dualistic distinction between regular and irregular inflected forms. According to dualistic approaches, irregular forms should show effects of word-form frequency in a lexical-decision experiment, since they are stored as fully inflected forms in the mental lexicon. Regularly inflected forms, in contrast, are not stored but are formed by a process of affixation and should, therefore, not be subject to word-form frequency effects in lexical-decision experiments. We used this effect to explore the German noun-plural system (Penke & Krause 2002). 2.2.1. Example case: Frequency effects in German noun plural inflection Plural nouns in German are marked by four different endings: the marker -e (Tisch-Tische ‘tables’), the infrequent marker -er (Kind-Kinder ‘children’), the most frequent marker -n (Blume-Blumen ‘flowers’), the infrequent marker -s (Auto-Autos ‘cars’), or they can remain unmarked (Ritter-Ritter ‘knights’). In contrast to the other plural markers, -s is special in several ways. It is very productive and freely occurs with all sorts of nouns which do not have a lexical entry, such as derived nouns (die Wenns ‘the ifs’) or neologisms. Moreover, it is the only marker that allows for monosyllabic plural forms such as Parks (‘parks’), is relatively free in its phonological distribution, and occurs with nouns of any gender (see Marcus et al. 1995). All other plural forms are restricted to particular morphophonological contexts, show a dependency on gender, and are subject to a prosodic constraint that requires the plural form to end in a reduced syllable, i.e. an unstressed syllable with Schwa or a syllabic sonorant (Neef 1998; Golston & Wiese 1996; Wiese 1996). Due to these differences in distribution and productivity, the general consensus between researchers is that -s is the default plural affix in German (e.g. Bornschein & Butt 1987; Marcus et al. 1995; Dressler 1999). A focus of debate, however, is whether -s is the only regular plural affix in German and whether all other German plural forms are stored irregular forms, as suggested in work by Clahsen and colleagues (cf. Marcus et al. 1995, Clahsen, Eisenbeiss & Sonnenstuhl 1997, Sonnenstuhl, Eisenbeiss & Clahsen 1999; Clahsen 1999). A critical case is the German -n-plural. Whereas the -n-plural is completely predictable for the rather large group of -n-plurals on non-umlauting feminine nouns that end in Schwa in the singular (die Blume – Blumen ‘flower[FEM]’, henceforth -nfem-plurals), it is largely unpredictable which masculine or neuter nouns will or will not take the -n-ending. It has therefore

406 Martina Penke been suggested that -nfem-plurals are regular too (e.g. Bittner 1994; Wiese 1996; Dressler 1999; Wunderlich 1999), while -n-plurals for masculine or neuter nouns that do not end in Schwa in the singular (das Hemd – Hemden ‘shirt[NEUT]’, henceforth called -nnonfem-plurals) are irregular forms that are stored as fully inflected forms in the mental lexicon (e.g. Marcus et al. 1995; Wiese 1996; Wunderlich 1999). To investigate the question which German plural forms are stored irregular or affixed regular forms, we conducted an auditory lexical-decision experiment (see Penke & Krause 2002). We tested four German plural markings: the default -s, the irregular marker -er, and the critical -nfem and -nnonfem markers. For each marker, we selected 20 pairs of plural nouns from the CELEX-database (Baayen, Piepenbrock & van Rijn 1993). The plural forms of each pair were matched for the number of syllables and for their word frequency. The word-form frequency of the plural forms however differed: whereas one noun of the pair had a low plural-form frequency, the other had a high plural-form frequency. Consider, for example, the two nouns of the pair Nasen ‘noses’ and Flaschen ‘bottles’. Both have a similar word frequency: 211 for Nase, 216 for Flasche. They differ however in their respective plural-form frequency: while the frequency for the plural form Nasen is low (= 15), the frequency of the plural form Flaschen is about 4 times higher (= 62). This pairing allows to directly test effects of plural-form frequency while controlling for other possible effects such as effects of word frequency. Table 1 gives the relevant information on plural-form frequency and word frequency for the four different sets of items. Table 1. Distribution of word and plural frequencies in the test material Marking

 frequency of infrequent plurals

 frequency of frequent plurals

plural frequency word frequency plural frequency word frequency -s -er

12

108

48

112

43

386

141

419

fem

24

187

93

163

-nnonfem

22

99

78

101

-n

In addition to the test items, we constructed 160 nonce words with different plural endings. The whole lexical-decision experiment consisted of 320 items that were presented auditorily. We tested 16 unimpaired, right-handed native speakers of German without hearing deficits.

The representation of inflectional morphology in the mental lexicon

407

Figure 6 presents the mean reaction times for the frequent (light bars) and the infrequent (dark bars) noun plurals in each of the four tested groups of plural markers. The figure shows a strong frequency effect for -er plurals and for -nnonfem-plurals: reaction times for frequent plural nouns with these markers are significantly shorter than those for infrequent plural forms (ttest comparisons, p < .05 for -er and -nnonfem-plurals). In contrast, -s-plurals and -nfem-plurals take similar reaction times irrespective of whether the plural form is frequent or infrequent and, thus, show no effects of word-form frequency (t-test, p > .05 for -s- and -nfem-plurals). The frequency-effect observed for -er- and -nnonfem-plurals provides evidence that these plural forms are indeed forms stored in the mental lexicon. The lack of a frequency effect for -s-plurals and -nfem-plurals, in contrast, suggests that these plural forms are not stored as fully inflected forms in the mental lexicon, but are composed out of stem and affix via affixation.

Figure 6. Mean lexical-decision times for plurals on -er,-s, -nnonfem, and -nfem

Our findings indicate that noun plurals on -s and -nfem are not stored, but composed by productive regular affixation. Nevertheless, the -nfem-plural is restricted to a particular class of German nouns, i.e. feminine nouns that do not show Umlaut, whereas the default -s-plural is not restricted to a particular noun class. Thus, our data provide evidence that regular inflection and default inflection are not necessarily equivalent, as is, for instance, proposed in the Dual-Mechanism Model (e.g. Pinker & Prince 1988; Pinker 1999; Clahsen 1999). Specifically, our data suggest a division of regular inflection into default inflection and regular inflection that requires a specific input and thus is input restricted. This is captured by the analysis in Wunderlich (1999) where plurals on -nfem are built by regular affixation of a suffix

408 Martina Penke which requires a non-umlauting feminine noun and adds the information [+ PLURAL] (for similar suggestions along these lines see Dressler 1999; Indefrey 1999; Wiese 1999). All information required for the input restriction is provided by substantial properties of the stem, such as gender or phonological shape, so that no arbitrary class features have to be postulated. Specificity conditions ensure that regular default -s-inflection only applies when the lexical look-up of stored irregular inflected forms fails and the input conditions of input restricted regular inflection are not fulfilled. 3.

Affixes in the mental lexicon

In contrast to unitary approaches to inflectional morphology where regular inflected forms are stored in the mental lexicon, affix-based dualistic approaches to inflection such as Minimalist Morphology (Wunderlich & Fabri 1995; Wunderlich 1996a) or Pinker’s Dual-Mechanism Model (Pinker 1999) assume that regular inflected forms are built by combining a stem and an affix. The assumption that stems and regular inflectional affixes have independent entries in the mental lexicon that can be productively combined via affixation is the hallmark of strong lexicalist approaches to inflection (cf. e.g. Lieber 1980, 1992; Jensen 1990; Wunderlich & Fabri 1995; Wunderlich 1996a, 1996b). This section presents some evidence in favour of this strong lexicalist assumption. Moreover, experimental data making use of the ungrammaticality effect indicate that the morphosyntactic information represented in affix entries is underspecified. 3.1. What to learn from errors: Speech production in language acquisition and language disorders The study of the mental representations and operations involved in the human ability to produce and comprehend language does not necessarily require expensive technical equipment. Erroneous forms produced during language production also constitute an interesting and very valuable data source. The value of erroneous forms was probably first noticed in speech-error research (Freud 1916; Fromkin 1971, 1973). A linguistic investigation of speech errors quickly revealed that such errors were not random, but constrained by the architecture of the language system. According to the Unit Similarity Constraint (Shattuck-Hufnagel 1979) the elements involved in a speech error are of the same level of representation and category. Consider a speech

The representation of inflectional morphology in the mental lexicon

409

error such as glear plue sky for the target phrase clear blue sky (Fromkin 1971). In this speech error the [+ voiced] feature value of the consonant [b] and the [– voiced] feature value of the consonant [k] exchange places. Both elements involved in the exchange are of the same category, i.e. the phonetic feature [± voiced]. Their exchange provides evidence for a level in language processing where this feature is represented and computed. Speech errors, thus, reflect the structural organisation of our language faculty. A similar logic can be applied to erroneous forms produced in child language or aphasic speech. Comparable to speech errors, errors in child and aphasic language production can be used to draw inferences about the structure of the language system generating the erroneous form. Consider as an example the so-called overregularisation errors in language acquisition data where a regular inflectional pattern is overapplied to stems that have irregular inflected forms in the target language. Such an overregularisation would, for example, result in the participle form *getrinkt instead of the correct irregular form getrunken (stem trink- ‘drink’). The occurrence of such forms has been taken as evidence that the child has acquired a mental operation that adds an inflectional affix (the participle affix -t) to a stem. Once learned, this operation can freely apply to any instance of the category verb for which no irregular participle form has been stored in the mental lexicon yet (cf. Marcus et al. 1992; Clahsen & Rothweiler 1993; Weyerts & Clahsen 1994; Weyerts 1997). Speech errors in unimpaired language and erroneous forms produced by children and aphasic speakers first and foremost differ with respect to the number of errors occurring, with aphasic speakers and children producing more errors to analyse. An interesting case relevant to the controversy between storage-based and affix-based approaches to regular inflection comes from the investigation of -t inflected verb forms in German child language (see Grijzenhout & Penke 2005). 3.1.1. Example case: -t-inflected verb forms in German child language Under an affix-based view of inflection, a regular participle form such as getanzt ‘danced’ is built by adding the participle affix -t to a verbal stem. But now consider a verb like heften (‘to staple’). Suffixing the verbal stem heft- with the participle affix -t would result in a participle form geheftt where two identical segments, the stem final [t] and the affix [t], follow each other. Sequences of two adjacent identical elements are, however, generally avoided in languages (e.g. Leben 1973; Goldsmith 1979; McCarthy 1986;

410 Martina Penke Yip 1998). To avoid such a sequence, in German the inflectional suffix -t is realised as [
t] when the stem ends in a coronal stop /t/ or /d/. Thus, the stem heft- is followed by [
t] in inflectional forms such as the participle geheftet. Another option is chosen in Dutch, for instance, where regular past participles are formed by adding a suffix -d to a verbal stem. When the stem ends in a coronal stop /t/ or /d/, as in the Dutch verb land ‘land’, only one of the two stops is realised (land + -d
geland). In an investigation of German child data (cf. Grijzenhout & Penke 2005), we found that German children go through a stage in language acquisition where they have considerable problems in producing -t inflected forms, such as regular participles or 3rd person singular present tense verb forms, when the verbal stem already ends in a coronal stop /t/ or /d/ (e.g. heft-et ‘staple[3SG]’, geheft-et ‘staple[PART.]’). These problems are exceptionally clear in the data of Naomi, a monolingual German child acquiring standard German. In Naomi’s speech, the first -t inflected forms for verbal stems ending in a coronal stop [t] appear at 1;9.18 (year;month.day). From that age onwards until the last of the weekly speech recordings available to us at age 2;1.20, Naomi varies between the three possible realisations of -t-inflected forms for verbs with stem final coronal stop: (i) she omits [t] (5a, d) in the majority of cases (71.4%, 65 out of 91 contexts for [t] inflection in 3rd person singular and participle forms), (ii) she produces a sequence of two identical elements (5c, f) in a few cases (11%, 10 of 91) and (iii) she produces correct target forms with [t] (5b, e) in very few instances (6.6%, 6 of 91): (5)

Naomi’s realisation of German 3rd singular verb forms and past participles Spelling Adult form Child’s form a. schneid-et [SnAId´t] [nAIt] (2;1;03) b. [nAId´t] (2;1;03) c. geschnitten [g´SnIt´n] [nAItt] (2;1;03) d. angeguckt [ang´kUkt] [tUt]6 (1;11;24) e. [tUt´t] (1;9;18) f. [an´tUtt] (2;1;03)

Gloss ‘cut[3SG]’ ‘cut[PART]’5 ‘look at[PART]’

———–————————————————————————————

5

6

Note that this form is an overregularisation where the regular participle affix -t is overapplied to the basic verb stem of the strong verb schneid- ‘cut’ (correct participle geschnitten). Naomi pronounces velar [g] and [k] as [t], thus the verb guck- ‘look’ is pronounced as [tUt] by Naomi.

The representation of inflectional morphology in the mental lexicon

411

This distribution is in marked contrast to the inflected forms Naomi produces for verbs not ending in stem final /t/ or /d/. For these verbs, the -t inflection is correctly realised in 82.1% (465 of 566) of the relevant contexts, compared to 6.6% of correct [´t] inflection after stem final /t/ or /d/. Also, whereas the omission rate of [´t] inflection after stem final /t/ or /d/ is 71.4% in 3rd person singular and participle contexts, the omission rate for the affix -t in verbs without stem final /t/ or /d/ is only 6.4% (36 of 566). This difference between correctly and incorrectly produced forms for verbs with and without stem final /t/ or /d/ in 3rd person singular and participle contexts is highly significant (chi2 = 218.838 Yates corrected, p < .000). The data exemplified under (5) show how Naomi is engaged in finding out the German solution for this morphophonological problem which results from the composition of a stem ending in a coronal stop /t/ or /d/ and an affix -t. If there were no composition of stem and affix, as suggested in unitary approaches to inflection where regular inflected forms are stored in the mental lexicon, then there would be no boundary between stem and affix and, hence, no reason for this morphophonological boundary problem to occur. In this case, the production of non-standard forms like tutt or tut – which the child has never heard in the input – cannot be explained. Naomi’s data suggest that affixes such as the 3rd person singular marker -t or the regular participle suffix -t are independent units of the language system that are added to a stem by an affixation process. Additional evidence for this assumption comes from a comparison of error rates aphasic speakers obtain in elicitation tasks on regular inflection. 3.1.2. Example-case: Effects of affix frequency in regular inflected forms in Broca’s aphasia Our investigation of participle inflection in German Broca’s aphasia revealed that aphasic speakers display a selective deficit with irregular participle formation (cf. fig. 1 and §1.1). A closer analysis of the data showed that errors with irregular participles, but not errors with regular participles, were dependent on the frequency of occurrence of the respective participle form. The experimental material included both rather frequent participles (participle frequency of each verb >100 in the 6-million-token CELEX database [Baayen, Piepenbrock & van Rijn 1993]) and very infrequent participles (participle frequency of each verb > list_of_immediate_subtypes’. (1)

A simple type hierarchy pers >> 1, 2, 3. num >> sg, pl. 1 >> 1sg, 1pl. 2 >> 2sg, 2pl. 3 >> 3sg, 3pl. sg >> 1sg, 2sg, 3sg. pl >> 1pl, 2pl, 3pl.

num

pers

1

1sg

1pl

2

3

sg

pl

2sg

2pl

3sg

3pl

Inheritance-based models of the lexicon

435

The relation ‚subtype’ is defined with the reflexive transitive closure of the graph induced by the relation ‚immeditate subtype’; it consequently consists of every type paired with itself as well as every pair of types which are connected by a directed path in the graph. Type unification is the binary operation on types
: T
T  T, where t1
t2 = t3 is defined when t3 exists as the most general subtype of t1 and t2. The type t3 is called the most general unifier (mgu). In (1) we have 1
pl = 1pl, but sg
pl is undefined. In contrast to other formalisms (e.g., ALE; cf. Carpenter & Penn 1999) QType does not require the type system to form a bounded complete partial order (bcpo) and thus a semi-lattice. As a consequence, type unification is nondeterministic, i.e., for a pair of types there can be more than one most general unifier. (2)

A type hierarchy violating the bcpo condition a b a >> c. a >> d. b >> c. c b >> d. d

In (2) there is no single most general unifier of a and b, but rather two: c and d. Our implementation of the type system is based on a bit-vector encoding of the types (cf. Aït-Kaci et al. 1989, Bernhardt 2001), which opens an extensional perspective since the bit vector for a given type represents all the subsumed minimal types that can be regarded as the extension of that type. All types have bit vectors whose length is identical with the total number of minimal types, where each minimal type has a defined position in the vector. In the bit vector of a given type, all the bits of the subsumed minimal types are set to 1, and all others to 0. Accordingly, every minimal type has a bit vector in which exactly one bit is set to 1, whereas for the most general type top all bits are set to 1. In the case of unary branchings and more complex redundancies in graphs like examples (1) and (2) this encoding leads to the problem that types can no longer be distinguished by their bit vectors because the sets of subsumed minimal types are identical. We solve this problem through the addition of further, so-called ‘dummy’ types as the following example shows:

436 James Kilbury, Wiebke Petersen, Christof Rumpf (1a) Bit-vector encoding of the type hierarchy in (1) 0111111

1111111

0000011 0001100 0110000 0010101 0101010 1000000

0000001 0000010 0000100 0001000 0010000 0100000

The type with bit vector 1000000 in (1a) is a dummy type which is necessary to distinguish the two most general types num and pers. Without the dummy type these would have identical bit-vector representations since they would cover the same set of minimal types. It is furthermore necessary to add an additional dummy type as an immediate subtype of pers to allow a common supertype of num and pers that has a bit vector distinct from that of num, but we have avoided this in (1a) for the sake of simplicity. The bitvector coding of types permits simple and efficient computation of operations on types through logical operations on bit vectors. (3)

Let bi be the bit-vector representation of type ti ; then the following holds: (3.1) unification t1
t2  b1 AND b2 (3.2) generalization t1  t2  b1 OR b2 (3.3) complement ¬t2  NOT b (3.4) subsumption t1  t2  b2 =? (b1 AND b2)

The unification of two types is normally computed with the AND operation on the corresponding bit vectors, where only the bits remain which are set to 1 in the same positions of both vectors. This corresponds to the intersection of the sets of minimal types that constitute the extensions of the types to be unified. If the intersection is empty, a bit vector results which contains only zeros, and this is equivalent to failure of unification. For generalization an OR operation is computed, where all those bits remain which are set to 1 in a given position in one of the vectors. This corresponds to the union of the sets of minimal types. The complement of a type is built simply by switching all the bits: 0 is replaced with 1 and vice versa. If the re-

Inheritance-based models of the lexicon

437

sult of unifying one type with another is identical to the second type, then the former subsumes the latter and is more general. The operator ’=?’ tests the identity of two bit vectors. It has already been mentioned that unification in the type system of QType is nondeterministic. Nevertheless, the AND operation always produces a unique result. From this it follows that the operations on bit vectors may produce vectors that do not correspond to any defined type of the signature. We call such vectors ‘virtual types’ (Bernhardt 2001). They can be resolved in a disjunction of defined types by identifying the minimal sets of most general types that cover all the subsumed minimal types. The same phenomenon appears in the case of the complement, where the same solution strategy is employed. 2.2.2. Appropriateness conditions The set of corresponding attribute-value pairs (henceforth ‘AV pairs’), also called feature-value pairs, for a class of feature structures of a given type is defined by the partial function ‘Appropriateness’: Types
Attributes  Types, which is introduced in the QType type signature with the symbol ‘::’. (4)

Description of a class of feature structures and its most general AVmatrix agr :: num:numerus, pers:person, gen:genus.
agr   num : numerus  numerus >> sg, pl.  pers : person  person >> 1st, 2nd, 3rd.   genus >> masc, fem, neut.  gen : genus 

In (4) a class of feature structures of type agr is introduced for which the three AV pairs are appropriate, e.g., agr
num  num. Possible combinations give 2
3
3 = 18 different instances for this class, but there can be classes with an enumerably infinite number of instances: (5)

Description of a class list and an instance of length two list >> elist, nelist.
nelist   head :1  nelist :: head:top, tail:list.  
nelist   tail :  head : 2      tail : elist   

438 James Kilbury, Wiebke Petersen, Christof Rumpf This recursive definition of the type list allows lists of arbitrary length and with elements of arbitrary type to be defined, including the type list itself. The possible instances for this class correspond to at least the set N of natural numbers. Thus, the type signature defines a finite number of classes of feature structures, but these potentially may have indefinitely many instances. 2.2.3. Inheritance The AV pairs of a type are inherited by its subtypes, which for their part may introduce further AV pairs or may sharpen the inherited values. A specificity hierarchy thus arises in which supertypes subsume their subtypes. In a framework with monotonic inheritance the AV pairs of subtypes must be consistent with those which are inherited; otherwise, one speaks of ‘type clashes’. In QType type clashes are allowed, but (more) specific information takes precedence over inherited information. This point will be discussed in §2.7, but first the feature logic of QType in the context of monotonic inheritance will be introduced. 2.3. Feature logic A feature logic is a description language for feature structures that has a syntax and a compositional semantics. It can be used to create instances for the classes which are defined in the type signature. The feature logic of QType has the following descriptive elements, where Desc denotes the set of descriptors: (6)

Syntax of QType’s feature logic (6.1) variables Vars  Desc (6.2) types Types  Desc f  Feat, di  Desc: (6.3) feature-value pairs (6.4) conjunction (6.5) disjunction (6.6) complement (6.7) relational constraints

f : d  Desc d1 & d2  Desc d1 ; d2  Desc –(d)  Desc @r(d1,…,dn)  Desc

Inheritance-based models of the lexicon

439

The interpretation of this feature logic is given by a function mgu : Desc  2FS. Because of disjunction and negation the function returns values from the power set 2FS of those feature structures which are possible instances of the classes defined in the type signature. (7)

Semantics of QType’s feature logic v  Vars, t  Types: (7.1) (7.2)

mgu(v) = {top} mgu(t) is given directly by the definition of t in the type signature.

f  Feat, ti  Types, di  Desc: (7.3) (7.4) (7.5) (7.6) (7.7) (7.8) (7.9) (7.10) (7.11) (7.12)

mgu( f : d) = {FS1 | FS1  mgu(t) where t  lub-pf( f : d)}
Set {[ f : FS2 ] | FS2  mgu(d)} mgu(d1 & d2) = mgu(d1)
Set mgu(d2) mgu(d1 ; d2) = mgu(d1)
mgu(d2) mgu(-v) = mgu(-t) = max( U mgu(ti)) for all ti with t
ti = mgu((-f) : d) = mgu(f : -d) mgu(-(f : d)) = mgu(f : -d) mgu(-(d1 & d2)) = mgu(-d1 ; -d2) mgu(-(d1 ; d2)) = mgu(-d1 & -d2) mgu(@r(d1, …,dn)) = mgu(def1(r(d1, …,dn)))  …  mgu(defm(r(d1, …,dn)))

The interpretation of descriptions consisting of a variable (7.1) gives a set of feature structures with the special type top as its only element. Descriptions consisting of a type name give the feature structure of the type according to its definition in the type hierarchy. (8)

Interpretation of descriptions consisting of a single type 
agr  

  num : numerus  mgu(agr ) = 

 pers : person    gen : genus  

440 James Kilbury, Wiebke Petersen, Christof Rumpf Feature-value pairs occur in descriptions as attribute-description pairs, whose interpretation (7.3) requires that the most general types be determined for which the pair is appropriate. The latter is achieved with the function lub-pf (least upper bound for polyfeatures). QType allows polymorphic features (‘polyfeatures’) such that a given feature can be appropriate for several disjoint types, possibly with different values. Other frameworks (e.g., ALE) require unique ‘feature introduction’. The value of the feature f of the candidates ti that have been found must be compatible with the description d. This is guaranteed by the operation
Set, which unifies the feature structures of the resulting set pairwise. Conjunctions of descriptions (7.4) return for each member of the conjunction a set of feature structures whose elements must be unified pairwise. Our implementation uses an incremental strategy for this, which – according to our experience – leads to early failure in cases of inconsistency and thus results in greater efficiency. The incremental procedure is based on the principle that, for a series of conjunctions d1 & …& dn each interpretation mgu(di) must be unified with the unifications of the interpretations (mgu(d1)
Set …)
Set mgu(di-1), i.e., intermediary results are accumulated and available as constraints for following computations. Disjunctions of descriptions (7.5) lead to union of the sets of feature structures given by the interpretations of the members of the disjunction. The semantics of negated descriptions must be specified separately for each kind of description. The negation of a variable (7.6) corresponds to the complement of top, i.e., the empty set. The negation of a type t (7.7) consists of the maximal elements with respect to subsumption in the set of feature structures of all types ti that are inconsistent with t and thus have no subtype in common with t. In feature-description pairs (7.8, 7.9) the negation is always shifted to the description. This excludes an interpretation according to which, for negated features, the set of all types is given which do not have this feature. The negation of complex descriptions (7.10, 7.11) is resolved by shifting the negation inward according to DeMorgan’s laws. The semantics of relational constraints is the union of the feature structures corresponding to their definitions, where the definitions are nothing other than the descriptions just defined in (6) and (7) (cf. §2.8 for a more detailed discussion). Aside from the differences in the type system, the interpretation of descriptions in QType in a way is like that in ALE since most general unifiers are computed. In CUF (cf. Dörre & Dorna 1993) and ConTroll (cf. Götz et al. 1997), in contrast, unifiers are computed which always are minimal types. A nonminimal type accordingly represents the disjunction of all minimal types that it covers, i.e., the most specific unifiers (msu). Pollard & Sag (1994) and Penn (2000) call these representations ‘sort resolved’, where

Inheritance-based models of the lexicon

441

‘sort’ refers to types. Such differing definitions have grave consequences for grammar development, as the following example is intended to show. The interpretation of mgu(agr) results in a set with a single element (see (8)), whereas the interpretation of msu(agr) results in a set with 18 elements, because every value of every feature has to be a minimal type. (9)

Interpretation of types with the function most specific unifier 
agr 
agr 
agr 
agr   





num : pl   num : sg num : pl num : sg

msu(agr ) =  , , ,..., pers : 3rd   pers :1st pers :1st pers : 2nd







  gen : masc   gen : masc   gen : masc   gen : neut  

2.4. Type constraints The framework presented thus far consists of a type signature and an interpretation function for descriptions. See Rumpf (forthcoming) for an implementation of finite-state automata within this framework. Since variables are not used in the description language for the type signature, it follows that co-indices are excluded, although they are required for modelling HPSG schemata and subcategorization in the type signature. To solve this problem we extend the type signature with type constraints. The latter are partial functions Desc  Types for which we use the notation t ::= d in QType. With type constraints the classes of feature structures defined in the type signature can be refined to any desired degree and in such a way that the additional information of a type is inherited by its subtypes. However, the refinement must remain within bounds set by appropriateness conditions, i.e., no new feature-value pairs may be introduced. (10) Simplified grammar fragment modelling the head feature convention of HPSG headed-phrase :: HEAD:sign, DTRS:dtrs. headed-phrase :: = HEAD:$1 & DTRS:HEAD-DTR:HEAD:$1.  headed
phrase     HEAD : 1     DTRS : HEAD
DTR : HEAD : 1 

442 James Kilbury, Wiebke Petersen, Christof Rumpf The symbol ‘$’ in (10) is used to distinguish variable names from types with the same name. The coindexation is made possible only through the use of type constraints with which one can use expressions in feature logic to address constituents embedded at arbitrary depth in feature structures; this would not be possible with a simple extension of the description language for type signatures with variables. Type constraints are comparable to implicational constraints like those in ConTroll (cf. Götz et al. 1997) but slightly weaker since the left-hand side of implicational constraints consists of arbitrary descriptions and every feature structure subsumed by it must be consistent with the arbitrary description on the right-hand side of the implicational constraint. Furthermore, type constraints provide a new means for representing nondeterminism since expressions in feature logic may contain disjunction and negation. Cyclic feature structures can be defined as well. 2.5. Relational constraints Relational constraints (RCs) extend a constraint-based grammar-development environment with a general constraint-logic programming (CLP) language for typed feature structures. In other frameworks RCs are also called definite clauses (ALE, cf. Carpenter & Penn 1999) or sorts (CUF, cf. Dörre & Dorna 1993). An RC r(d1,…,dn)  Desc is a relation r over feature-logic descriptions d1,…,dn which, like all descriptions, are mapped onto the set 2FS. In contrast to other descriptive elements of our feature logic which refer exclusively to classes defined in the type signature, RCs are defined outside the type signature (although they conform to it) and thus constitute an independent system in which further inheritance relations can be defined. So one must distinguish between the definition and the use of a relational constraint as a feature-logic expression, where they are preceded by the symbol ‘@’ to distinguish them from types with the same name. The syntax for definitions of RCs appears as follows: (11) Syntax of relational constraint definitions with di  Descr, r  String, n
0 r(d1,…,dn) #> d0. There can be more than one definition for a relation r with arity n, and these definitions are then interpreted disjunctively (see (7.12)). The practical use of the parameters d1,…,dn is manifold. They can trigger the matching of a specific constraint definition and transport information between a feature-logic expression, where the constraint appears, and its definition.

Inheritance-based models of the lexicon

443

A subset of the RCs is comparable with the templates of PATR-II (cf. Shieber 1986), which may be regarded as abbreviations for complex feature structures. Since templates can themselves be used in the definition of templates, they provide a means – indeed, in the case of PATR-II the only means – of defining inheritance hierarchies. (12) shows the translation into QType of an example from Shieber (1986: 57): (12) Relational constraints defined as templates verb #> cat:v. finite #> @verb & head:form:finite. The constraint finite is defined with the constraint verb. In contrast to PATR-II templates, RCs can be parametrized with descriptions. Coupled with recursive data structures like lists, they thereby achieve a status that goes far beyond the abbreviatory character of templates. A constraint-logic programming language arises which is comparable with „Pure Prolog“ and with which programs involving infinite sets of feature structures can be defined, e.g., in order to implement Turing machines. This does not mean that relational constraints are essential in order to implement Turing machines with feature structures; they just make it simpler. Keller (1993) describes how „Negated Regular Rounds-Kasper Logic“ (NRRK Logic) can be used as an extension of PATR-II to do the latter. NRRK Logic is an extension of Rounds-Kasper Logic (Kasper & Rounds 1986), one of the first studies to provide a feature logic for untyped feature structures with a denotational semantics. Johnson (1988) shows an implementation of Turing machines with unary syntax rules as state transitions and lists as tapes. Nevertheless, relational constraints can be used to define arbitrary relations over lists: (13) A relational constraint defined for lists of arbitrary length append(elist, $L) #> $L. append((first:$H & rest:$T), $L) #>first:$H & rest:@append($T, $L). Recall that ‘$’ distinguishes variables from other descriptions. The interpretation of the description append($Prefix, $Suffix) & first:1 & rest: (first:2 & rest: (first:3 & rest:elist)) produces the disjunction of all decompositions of the list into prefixes and suffixes as instances of the parameters $Prefix and $Suffix. Type constraints can be used to embed RCs in the inheritance hierarchy of the type signature, e.g., in order to permit constraintbased parsing as the following example shows:

444 James Kilbury, Wiebke Petersen, Christof Rumpf (14) A type constraint using a relational constraint sign >> phrasal_sign, lexical_sign :: phon:phonlist. phrasal_sign >> left:phonlist & right:phonlist. phrasal_sign ::= phon:@append($L, $R) & left:$L & right:$R. The type phrasal_sign is treated here as a binary tree, where the type constraint phrasal_sign ensures that the value of the feature phon is always the concatenation of the values of the features left and right. Since undirected constraint-based parsing cannot compete for efficiency with established parsing strategies, QType provides the possibility of specifying a grammar with context-free syntax rules and a lexicon, which can then be processed using a left-corner parser with linking. 2.6. Phrase-structure rules and lexical entries As mentioned above, the phrase-structure rules and lexical entries of QType are provided to allow for efficient parsing. Phrase-structure rules are written as context-free phrase-structure rules which define sets of trees whose nodes are associated with feature structures. (15) Syntax of phrase-structure rules in QType with di  Descr d0 => d1 ^ … ^ dn. The symbol ‘=>’ denotes immediate dominance of the mother node d0 over the daughter nodes d1 ^ … ^ dn, while ‚^’ denotes concatenation and therefore linear precedence among the daughters. Since the rules are processed using a left-corner parser with linking (cf. Covington 1993), left-recursive rules do not present a problem. The linking relation is computed automatically during compilation of the grammar. The following example shows the adaptation of a phrase-structure rule for sentences S  NP VP from Shieber (1986: 25):

Inheritance-based models of the lexicon

445

(16) A phrase-structure rule for main clauses in QType and the corresponding syntax tree s & HEAD:$H => np & $NP ^ vp & HEAD:($H & SUBJECT:$NP & FORM:finite).
s     HEAD : 1 

2 np


vp 
FORM : finite  HEAD : 1    SUBJECT : 2 

     

The Head-Feature Principle of HPSG (cf. Pollard & Sag 1987, 1994) has been explicitly stated in the rule here. The application of rules follows the left-corner alogrithm: starting with a top-down prediction, a bottom-up step involving lexical access is triggered by a linking relation that determines which constituents are useful beginnings of a phrase dominated by the topdown prediction. So, parsing involves matching the strings of lexical items and unification of the feature structures at the nodes of trees that are adjoined. We decided to distinguish syntax rules and lexical entries to make the parser more efficient and to facilitate the application of lexical rules to lexical entries. The format for lexical entries pairs strings with descriptions of feature structures and appears as follows: (17) Syntax of lexical entries in QType s –> d. where s  String, d  Descr Two concrete examples of lexical entries are given here: (18) Two lexical entries and a relational constraint used as a template man –> cnoun. sleep –> @vbase. vbase #> v & HEAD:FORM:base. The word man has the type cnoun for common nouns in its description, while the description of the word sleep contains a relational constraint which

446 James Kilbury, Wiebke Petersen, Christof Rumpf functions here as an abbreviatory template and specifies the type v as well as the value base for the feature path HEAD:FORM. In order to take advantage of the power of inheritance in the type signature, good practice would be to define almost all information of lexical classes down to specific lexical entries in the type signature. This would lead to lexical entries in the format (17) that pair a string with a single type. A further device in QType consists of lexical rules, which we have incorporated merely for historical reasons, since they can be eliminated with suitable techniques (cf. Bouma et al. 2000). They facilitate a compact description of the lexicon and define binary relations between lexical entries by generating new lexical entries as output from previously-defined entries as input to cover lexical processes like inflection and derivation. The generation process involves asymmetric nonmonotonic unification to copy all parts of the input into the output that are compatible with the output specification of the lexical rule. See Briscoe & Copestake (1999) for a general survey and Rumpf (forthcoming) for a detailed description of the implementation of lexical rules in QType. 2.7. Nonmonotonic inheritance There are two reasons for employing nonmonotonic inheritance of information in object hierarchies: on the one hand it helps to avoid redundancy and thus to achieve more compact representations; on the other hand it provides a way to model rule exceptions explicitly. Precisely the last point can be a decisive criterion for the adaquacy of a theory, but with monotonic means it remains elusive. Although the majority of linguists working within the framework of constraint-based methods have emphasized the advantages of monotonicity, approaches involving nonmonotonic processing of feature structures have a long tradition. Most previous attempts to integrate nonmonotonicity into constraint-based approaches use variants of nonmonotonic unification (e.g., Bouma 1990, Carpenter 1993), which is neither associative nor commutative, however. This leads to serious problems for the semantics of constraint-based frameworks since they thereby cease to be declarative. With YADU Lascarides & Copestake (1999) present a variant of default unification whose result is independent of the order of the computations and which therefore preserves the declarative character of the formalism. For this gain they pay a high price, however: so-called ‘tails’ store the history of unifications, and the computation of symmetric unifica-

Inheritance-based models of the lexicon

447

tions is factorial in the worst case (Carpenter 1993) and thus may be intractable. We therefore propose an alternative which preserves declarativity in the framework of QType but is also tractable: nonmonotonic inheritance in the type signature. The basic idea is to allow conflicting information in type signatures; conflicts in the top-down inheritance are then resolved according to the principle that more specific information has precedence, whereby the subtype relation is suspended. A default hierarchy can be converted offline into a corresponding monotonic hierarchy so that only the monotonic hierarchy is needed at runtime. (19) Conversion of a nonmonotonic to a monotonic hierarchy


verb  PAST : 1 + ed      PASTP : 1     PASSP : 1

 pst
t
verb  PAST : +t 

verb / pst
t
verb   PAST : 1     PASTP : 1     PASSP : 1


verb   PAST : +ed 

 pst
t
verb  PAST : +t 

In (19) we see an example involving verb inflection in English (cf. Lascarides & Copestake 1999). Regular verbs like push have the ending +ed in the forms for past, past participle, and passive participle. Exceptionally, a group of verbs like burn has the ending +t for all three forms. In the nonmonotonic link shown to the left of the arrow, the type pst-t-verb inherits the structure sharing of the type verb, and the inheritance of +ed is blocked. The nonmonotonic inheritance can be converted to monotonic inheritance by computing the generalization of the types verb and pst-t-verb, which produces the abstract type verb/pst-t-verb. The latter inherits the compatible information from verb and pst-t-verb monotonically. As the example clearly shows, the default relation between verb und pstt-verb is lost since the types are disjoint and equally specific. In our system, however, a default connection can be reconstructed by referring back to the nonmonotonic inheritance hierarchy. For this purpose a mapping relation between the types of the two signatures is computed, with the help of which reference is made only to the nonmonotonic signature in output. Accord-

448 James Kilbury, Wiebke Petersen, Christof Rumpf ingly, a user of QType does not even notice that the computation takes place with a signature containing types that he has not defined. Although the example just presented involves structure sharing, it still assumes a simplified scenario since only local conflicts at the level of the relation ‘immediate subtype’ need to be solved, whereas in general conflicts occur within the transitive closure of this relation. The supertype t0, which introduces a feature-value pair into the hierarchy via an appropriateness specification, and the subtype tn containing a conflicting value may be arbitrarily far apart. All conflict-free types ti along the path to t0 should, of course, remain in the monotonic inheritance relation to the supertype t0. This requires unfolding the signature at t0 by inserting a common immediate supertype t0-tn for t0 and tn which represents their generalization. However, this also suspends the inheritance relation between ti and tn, which is unacceptable since the types ti can introduce additional feature-value pairs which are inherited by tn. Consequently, a complete procedure must also compute the generalizations of all ti with tn and insert the corresponding supertypes ti-tn. This raises the question of where all the new inserted types should be attached. Clearly, t0-tn must be an upper bound for the new generalizations. The procedure just described is always applied to a single conflict; in order to resolve several conflicts it must be iterated until all the conflicts are removed. It remains to be seen whether the order of conflict resolution leads to different monotonic signatures and whether a strategy can be found that is most efficient. (20) The Nixon diamond

We see no plausible solution for conflicts arising from multiple inheritance. The Nixon diamond (cf. Touretsky 1986) represents such a scenario: The

Inheritance-based models of the lexicon

449

type nixon immediately inherits the property pacifist:yes from quaker as well as pacifist:no from republican. Which property has precedence can only be decided with the help of a given specificity weighting. We are inclined to reject solutions with nixon as a nondeterministic type. The procedure is nevertheless applicable to signatures with multiple inheritance, but conflicts may arise in the way just described; otherwise, due arbitrarily to the order of computation, the value would survive which is computed last. Signatures are consequently tested for such inconsistencies before the procedure is applied. A further problem involves embedded structures. In the previous examples, all values occurred at the same depth, which in general is a gross simplification of the possibilities, unless one first considers type signatures without type constraints. Here the situation is in fact relatively simple since the appropriateness specifications always have the same basic form Types
Attributes  Types and, thus, neither embedded nor shared structures can be addressed. Besides these extensions, type constraints also introduce disjunction and relational constraints, however. QType requires type constraints to be locally consistent, i.e. no type constraint may conflict with the type for which it is defined. On the other hand, the information introduced by type constraints is inherited by all subtypes. Since the conversion of nonmonotonic signatures into monotonic ones takes place offline, the type constraints should also be computed offline so that conflicts introduced by them arising through inheritance may be taken into account. Unfortunately, this is not always possible since there is a class of relational constraints that cannot be computed offline, namely, recursive relational constraints, which are defined for partial data structures like lists of indefinite length. In such cases the solution set is nonfinite, so we have chosen partial evalutation (cf. Pereira & Shieber 1987) of type constraints as a remedy: whenever possible, the computation is carried out offline, but it is delayed till runtime for recursive relational constraints. As a consequence, no conflicts may be introduced into the hierarchy through recursive relational constraints in connection with type constraints, since they can be resolved neither offline nor during parsing. In practice we are aware of no cases where this is important. The model of nonmonotonic inheritance presented here is significantly more restrictive than that suggested by Lascarides & Copestake (1999). Nevertheless, we believe that it entails no essential limitations on the specification of default relations in grammar development. In the following section, we present a method by which type hierarchies of the kind we have discussed can be induced automatically using Formal Concept Analysis.

450 James Kilbury, Wiebke Petersen, Christof Rumpf 3.

Induction of inheritance hierarchies

3.1. General remarks Technical progress in recent decades has led to an enormous growth in the quantity of data that can be obtained through corpus-based empirical studies. It is no longer possible to evaluate such data manually for the construction of lexica in the form required by modern grammar formalisms. The use of automatic methods is therefore unavoidable. In principle there are two different approaches for obtaining appropriate inheritance-based lexica from collected linguistic data on words and their properties: Incremental and dynamic procedures imitate human behavior (cf. Barg 1996). A relatively small data set is used to produce an initial inheritance hierarchy, which then, on the basis of additional data, is refined. A ‘pruning’ at all intermediate stages ensures that the resulting hierarchy is not too detailed. The ultimate result is a compact representation of the complete data set in an inheritance hierarchy which depends to a great degree, however, on the initial data, the order of the subsequently processed data, and the pruning algorithm employed. With this procedure it normally is only possible to obtain results that are locally optimal. The alternative procedure is to obtain a single hierarchy in one step; this hierarchy logically corresponds to the data in a defined form and may be extremely large. In a second step it can be reduced to an acceptable size through statistical or semi-automatic pruning. The disadvantage of this procedure is that the insertion of new words constitutes an independent problem, the solution of which is not given directly by the induction procedure. The method presented here, based on Formal Concept Analysis, although belonging to the second group of approaches, results however in hierarchies which are so detailed that new, unknown words normally can directly inherit appropriate information from a node already present in the hierarchy. If such a node does not exist, the new, extended hierarchy can be computed from the current one without starting over again from scratch. 3.2. Formal Concept Analysis Formal Concept Analysis (FCA) is a mathematical theory that models the notion of the ‘concept’ set theoretically and that orders concepts in concept lattices. These concept lattices can be regarded as inheritance hierarchies. Ganter & Wille (1998) characterize FCA as follows:

Inheritance-based models of the lexicon

451

“The sophisticated name of Formal Concept Analysis needs to be explained. The method is mainly used for the analysis of data, i.e. for investigating and processing explicitly given information. Such data will be structured into units which are formal abstractions of concepts of human thought allowing meaningful and comprehensible interpretation. We use the prefix formal to emphasize that these formal concepts are mathematical entities and must not be identified with concepts of the mind. The same prefix indicates that the basic data format, that of a formal context, is merely a formalization that encodes only a small portion of what is usually referred to as a context.”

A detailed introduction of Formal Concept Analysis is given by Ganter & Wille (1999).

3.2.1. Formal contexts In order to be analyzed with FCA, data must be structured as a formal context. A formal context (FC) is a triple (G,M,I) consisting of a set of objects G, a set of attributes M, and a binary incidence relation I GM between the two sets. If an object g is in a relation I with an attribute m, we write (g,m)  I, and read it as “the object g has the attribute m”. A formal context can — at least in finite cases — be represented in a table like that in (21). (21) Example of a formal context. The context covers the inflectional paradigms and the gender information of the following seven German nouns: Herr ‘mister’, Friede ‘peace’, Staat ‘state’, Hemd ‘shirt’, Farbe ‘color’, Onkel ‘uncle’, Ufer ‘bank/shore’. In the attribute names ‘*’ stands for the stem of the noun.

452 James Kilbury, Wiebke Petersen, Christof Rumpf Since “lexical entries have evolved from simple pairings of phonological forms with grammatical categories into elaborate information structures, in which phonological forms are now paired with more articulated feature structure descriptions” (Sag &Wasow 1999: 173), data must be collected in empirical studies that fill these feature-structure descriptions. Such data build ternary relations between phonological forms, attributes, and values of the attributes. The data must be put in the form of a suitable FC, i.e. a binary relation, through the choice of a suitable scaling for each manyvalued attribute. It often is enough just to regard each attribute-value pair as an independent attribute, as has been done in (21). This transformation process is called “scaling with the nominal scale”. Sometimes, however, e.g., in the analysis of subject-verb agreement in English, a more differentiated scaling is better suited. (22a) shows a small example of data about English inflected verb forms. In order to permit explicit reference to non-third values the scale in (22b) can be chosen, which introduces the values local and nonlocal, to refer to the status of the subject. The resulting formal context is shown in (22c). The choice of suitable scales is normally not a problem in linguistic contexts since only discrete values appear; these can always be scaled without information loss using the nominal scale. (22) Possible scaling of the attribute PERS

(a)

(b)

(c)

3.2.2. Formal concepts In FCA concepts are conceived in terms of sets as in the classical definition of concepts. They are defined through their extension as well as their intension. A formal concept of a context is thus a pair consisting of a set of objects, its extent, and a set of attributes, its intent. The extent consists exactly of the objects in the context for which all the attributes of the intent apply; the intent consists correspondingly of all attributes of the context which the

Inheritance-based models of the lexicon

453

objects from the extent have in common. In order to formally express the strong connection between the extent and the intent of a formal concept given by the binary relation I, two derivational operators are defined between the set of objects G and the attribute set M of a formal context: If A

G is a set of objects, then the set of the common attributes of A is A' := {mM |
gA: (g,m)I}, and if, analogously, B M is a set of attributes, then the set of objects that have B in common is B' := {gG |
mB: (g,m)I)}. A formal concept is thus a pair (A,B) with A G, B M where A = B' and B = A'. Furthermore, the definition of the derivational operators guarantees that (A'',A') is a formal concept for any set of objects A G and (B',B'') is a formal concept for any set of attributes B M. Formal concepts from (21) are, e.g., ({Staat, Farbe}' ' , {Staat, Farbe}' ) = ({Hemd, Staat, Farbe},{sing nom:*, sing dat:*, sing acc:*, plur nom:*_n, plur gen:*_n, plur dat:*_n, plur acc:*_n}),

and ({sing acc:*}', {sing acc:*}'') = ({Ufer, Onkel, Hemd, Staat, Farbe},{sing nom:*, sing dat:*, sing acc:*, plur dat:*_n}).

3.2.3. Concept lattices The subconcept-superconcept relation on the set of all formal concepts of a context defines a partial order: (A1,B1)(A2,B2)  A1 A2  B1 B2. This order relation corresponds to our intuitive notion of super- and subconcepts. Superconcepts are more general, encompass more objects, and are characterized by fewer attributes. It can be shown that the set of all formal concepts of a formal context ordered with respect to the subconcept-superconcept relation constitutes a complete lattice, which is called the concept lattice of the formal context.1 (23) shows the concept lattice corresponding to the formal context in (21). Concept lattices are normally represented by Hasse diagrams as in (23). Superconcepts are located above subconcepts and are a linked to them by a ———–———————————————————————————— 1

Furthermore, the Basic Theorem on Concept Lattices states that every complete lattice is the concept lattice of a formal context.

454 James Kilbury, Wiebke Petersen, Christof Rumpf path. The definition of a concept lattice allows an especially economical labelling: instead of labelling every concept with its complete extent and intent, only the object and attribute concepts are labelled. The object concept g of an object g is the smallest concept whose extent includes g, i.e. g = (g'',g'). The attribute concept μm of an attribute m is analogously the largest concept whose intent includes m, i.e. μm = (m',m''). For example, the attribute concept of the attribute sing acc:* from (21) is ({sing acc:*}', {sing acc:*}'') = ({Ufer, Onkel, Hemd, Staat, Farbe}, {sing nom:*, sing dat:*, sing acc:*, plur dat:*_n}). In this way a concept lattice becomes an inheritance hierarchy: Any concept of the lattice inherits all objects which are labelled with subconcepts as its extent, and it inherits all attributes with which superconcepts are labelled as its intent. Inheritance hierarchies based on concept lattices are completely nonredundant, i.e. each attribute and each object appears exactly once in the hierarchy. Concept lattices visualize structural connections between the data of a flat data list. Ganter & Wille (1998) stress: “It is our belief that the potential of Formal Concept Analysis as a branch of Applied Mathematics is just beginning to show. A typical task that concept lattices are useful for is to unfold given data, making their conceptual structure visible and accessible, in order to find patterns, regularities, exceptions, etc.” (23) Concept lattice from the context in (21)

Inheritance-based models of the lexicon

455

3.2.4. Complete theories and concept lattices The mutual dependencies between the data of a context can also be shown in another way, namely, with a complete theory that describes the data. Such a complete theory is a set of clauses over the attributes of the context which are compatible with the data of the context and from which all the compatible clauses can be derived. The clauses are of the form
xG : , where the set of attributes are treated as one-place atomic predicates and  and  are terms built inductively from and and the atomic predicates.2 Quantifiers and predicates no longer need to be specified explicitly; instead, we write PERS:first  PERS:nonthird rather than
x : PERS:first(x)  PERS:nonthird(x). As has been shown in Osswald & Petersen (2003), it frequently makes sense to restrict the form of clauses, e.g. by dispensing with the disjunctive or conjunctive operators. Three principal classes of complete theories result: theories with neither disjunctive nor conjunctive operators are termed simple inheritance theories, theories without disjunctive operators are Horn theories, general theories without restrictions are termed observational theories (for a thorough discussion see Osswald & Petersen 2002 and 2003). Every theory determines a hierarchical classification, the information domain of the theory. Its classes are the attribute sets maximally consistent with the theory and are ordered by the subset relation. The set of intents of all formal concepts of a context ordered by the subset relation correspond precisely to the information domain of a complete Horn theory of the context. Due to the convention in FCA we call Horn clauses (attribute) implications. The implications can be read off directly from the concept lattice: An implication m1  m2 … mk  m holds exactly when the greatest lower bound of the attribute concepts μm1,… ,μmk is a subconcept of the attribute concept μm. Hence, the implication sing dat:*  gender: masc  sing gen:*_s holds in (21), since in (23), the greatest lower bound of the nodes labelled with sing dat:* and gender: masc is a lower neighbor of the node labelled with sing gen:*_s. Since the union of every premise and the maximal corresponding conclusion forms an intent of one of the concepts of the context, the structure of the concept lattice is determined by the set of valid attribute implications up to isomorphism. FCA thus occupies an intermediate position with respect to the complexity of the underlying theories. In the following section we will show that ———–———————————————————————————— 2

Standard transformations can be used to show that a negation operator would introduce no additional expressivity.

456 James Kilbury, Wiebke Petersen, Christof Rumpf precisely this intermediate position makes FCA an outstanding tool for the induction of lexical hierarchies. More complex theories (those with disjunctive operators) tend to overfit, the induced hierarchies are too flat, and overgeneralizations are made. On the other hand, simpler theories (those without conjunctive operators) fail to capture some interesting linguistic generalizations, which can be seen from the fact that many linguistic formalisms include implications with conjunctions. It is noteworthy that many linguistic formalisms can be situated in this intermediate position. For example, feature-structure descriptions in the framework of HPSG are based on conjunction, whereas, e.g., the semantics of disjunction of types is a widely discussed problem. 3.3.

Examples using FCA

3.3.1. Strategies for avoiding redundancy in the lexicon: feature co-occurrence restrictions in GPSG versus hierarchical type signatures in HPSG Lexicalist linguistic theories encode a great deal of grammatical information in the lexicon and employ various strategies in order to avoid unnecessary redundancy. This section will introduce two of these strategies and show how FCA can be used to translate lexica built according to one strategy into corresponding ones for another. GPSG uses so-called feature co-occurrence restrictions (FCRs) to restrict the distribution of features and their values. A pair consisting of a feature and a feature value is called a feature specification. Whereas GPSG features are atomic symbols, feature values are either atomic symbols or categories,3 i.e., set of feature specifications (cf. Gazdar et al. 1985: 22). The FCRs are part of a grammatical theory and restrict the set of possible categories and their extensions in the theory. A typical FCR is [+INV]  [+AUX, FIN], which is []  [, ] when written out fully.4 The condition stated here is that in English the feature specification implies and : if a verb occurs initially in a sentence containing a subject, then this verb must be a finite auxiliary. ———–———————————————————————————— 3 4

Categories of GPSG correspond to the untyped features structures of other unification-based formalisms. The feature specification marks sentence-initial verbs, marks auxiliary verbs and specifies, that the verb is finite.

Inheritance-based models of the lexicon

457

As noted above, HPSG adopts no obvious counterpart for the FCRs of GPSG and instead employs inheritance hierarchies, whose informational domain consists of typed feature structures. HPSG further reduces redundancy by ordering the types in a type hierarchy, the so-called type signature, in which appropriateness conditions are inherited (for more details see §2). HPSG thus replaced the FCRs of GPSG with inheritance hierarchies of types, but the relations between these formal devices were not well understood. With FCA a general framework is now available which allows the equivalence of the devices to be explained in a transparent and declarative fashion. (24) Lexemes classified with respect to their feature specifications in Gazdar et al. (1985) The feature vform distinguishes parts of the verb paradigm (bse, base-form; fin, finite; pas, passive participle), and nform analogously distinguishes the special expletive pronoun it from normal nouns (norm).

The main idea for showing the convertability of FCRs and type signatures is to construct a suitable formal context in order to employ the methods of FCA. (24) shows a lexical fragment consisting of 10 lexemes classified with respect to some of the features proposed in Gazdar et al. (1985). The chosen features are exactly those which play a role in the first 4 FCRs given in Gazdar et al. (1985); these FCRs regulate the distribution of the features v, n, vform, nform, pform, inv, aux, and their values (see (25)). (24) is formed by taking each feature-value pair as an independent attribute of the context.

458 James Kilbury, Wiebke Petersen, Christof Rumpf GPSG also allows FCRs of the form [VFORM]  [+V, -N] (see FCR 2), which encode not only restrictions on feature-value pairs but also on the admissibility of certain features in the first place. Because of this, further attributes of the form feature:VAL have been added in (24), where such an attribute applies to a word if there is some value value such that feature:value is an feature specification of the word. Feature structures with embedded structures can be represented in a formal context by flattening them and viewing the path-value pairs as attributes (cf. Sporleder 2003; Petersen 2004b, Petersen forthcoming). We call such a formal context obtained from feature structures a feature structures context. (25) The first 4 FCRs from Gazdar et al. (1985)

To illustrate the relationship between FCRs and type signatures we will proceed as follows: First we show how a type signature can be derived from a feature structure context, and then we do the same for a system of FCRs. Finally, we demonstrate how a feature structure context can be constructed from either a type signature or a set of FCRs. (A detailed description can be found in Petersen & Kilbury 2005.) (26) shows the concept lattice for the feature structure context in (24), which can be directly interpreted as a type signature of HPSG if a unique type is assigned to each node of the lattice. The feature labels then encode the appropriateness conditions associated with each type, and the subconcept relation corresponds to the subtype relation in the type signature. Only the lowest node of the concept lattice has no direct counterpart in the type signature; it expresses the failure of unification and often is represented with , the symbol for bottom. If one reads the hierarchy in (26) as a type signature it is apparent that the hierarchy makes extensive use of multiple inheritance, which is possible to only a limited degree in some varieties of HPSG. Due to the definition of formal concepts, however, it can never be the case that a type inherits incompatible information from its upper neighbors.5 ———–———————————————————————————— 5

The principle of unique feature introduction is never violated because the features are introduced on their own, unique feature concepts.

Inheritance-based models of the lexicon

459

(26) The concept lattice for the feature structure context of (24), which can be regarded as a type signature6

The hierarchy in (26) encodes so much information about the distribution of atomic values that it is sufficient to pair phonological forms with types in the lexicon in order to obtain adequate feature structures. For realistic lexica such a procedure leads to enormous type signatures since every lexical feature structure must have its own type. Moreover, it conflicts with the idea of types when they, e.g. in the case of with, cover only a single object. On this issue Pollard & Sag (1987: 192) write: “lexical information is organized on the basis of relatively few – perhaps several dozen – word types arranged in cross-cutting hierarchies which serve to classify all words on the basis of shared syntactic, semantic, and morphological properties. By factoring out information about words which can be predicted from their membership in types (whose properties can be stated in a single place once and for all), the amount of idiosyncratic information that needs to be stipulated in individual lexical signs is dramatically reduced.”

———–———————————————————————————— 6

This figure was created using the program ConExp (http://www.sourceforge.net/ projects/conexp).

460 James Kilbury, Wiebke Petersen, Christof Rumpf Consequently, type signatures are frequently constructed in such a way that they primarily encode information about the general structure of feature structures. This is accomplished with appropriateness conditions which regulate the distribution of the feature, but restrict their values only to general types. (27) shows such a type signature which was generated with our system FCALING. The restricted distribution of the atomic values from the FCRs in (25) can be enforced with the introduction of additional type constraints. (27) A typical type signature for the data in (24)

The data of our example are rather atypical for an HPSG analysis since they are based on feature structures containing paths with no more than one feature. In the following we show that FCA is able to induce type signatures from sets of more deeply embedded feature structures of the sort shown in (28). A module of our system FCALING computes such inductions (details are given in Petersen 2004b, Petersen forthcoming). The following approach was chosen for the implementation: In an initial step the feature structures are translated into a feature structure context; in the course of this translation all atomic values are replaced by the marker ‘av_’, however. In addition to the large matrix feature structures we also take all the embedded structures and encode them in the same feature structure context. In a procedure reverse to that of Penn’s unfolding (cf. Penn 2000) this concept lattice is then folded, which results in a type signature that encompasses all structural aspects of the feature structures. Finally, information about the atomic values is added and the appropriateness conditions are sharpened correspondingly. When applied to the data in (28) the type signature of (29)7 results. Type signatures induced with FCALING can be used directly as a starting point for developing grammars in QType (see §2). ———–———————————————————————————— 7

The graphic output was realized with the help of the program GraphViz (http:// www.graphviz.org/).

Inheritance-based models of the lexicon

461

(28) A small example lexicon with untyped feature structures from Shieber (1986)

(29) Type signature induced from the untyped feature structures in (28)

Sporleder (2003) proposes another approach for the induction of type signatures. Here a lattice like that of (26) is defined as the search space for an induction task. A maximum entropy model is applied to this lattice which classifies nodes into plausible and implausible generalizations. This maximum entropy model is the result of a machine learning process which was trained on plausible manually constructed hierarchies. Having constructed type signatures from formal concept lattices, we will now show how to derive FCRs from a feature structure context as given in Table (24). The FCRs of GPSG are nothing else but implications that are

462 James Kilbury, Wiebke Petersen, Christof Rumpf compatible with the data of Table (24). In order to restrict the set of admissible categories sufficiently, the FCRs must reflect the mutual dependencies between the data of the context. Hence, the data of the context must respect the FCRs and the object concepts of the context must be derivable from the FCRs. The latter ensures that the FCRs license no feature structure with features which do not co-occur in the input structures. (30) shows a basis of the complete Horn theory of (24);8 some abbreviations have been used: A  B denotes the two implications A  B and B  A and  is the (inconsistent) set of all attributes of the context in (24). Because of the eqivalence (A  C)(B  C)  AB  C, we summarize implications with equal conclusions in one clause (e.g. clause 17). If one compares the implications in (30) with the corresponding FCRs from Gazdar et al. (1985) in (25), it is apparent that all FCRs can be derived from the implications in the basis of the Horn theory: FCR 1 corresponds to implication 8; FCR 2 is contained in equivalence 5, FCR 3 in equivalence 4, and FCR 4 in equivalence 2. (30) Horn theory for the feature structure context in (24)

———–————————————————————————————

8

Minimal bases of complete Horn theories for formal contexts can be automatically calculated with the help of the program ConImp (http://www.mathematik.tudarmstadt.de/ags/ag1/Software/DOSProgramme/Welcome_de.html).

Inheritance-based models of the lexicon

463

However, (30) includes further implications, some of which bear no real information in the sense of GPSG since they either result from the sparse input data (cf. equivalence 3), from the special role of the feature value VAL (cf. implications 12–16), or from the fact that no knowledge about the exclusivity of features is implemented in FCA (cf. implication 17). Implication 9 follows from FCR 1 on the condition that, first, whenever inv is specified then aux is also specified and vice versa (implication 7) and, second, inv is restricted to the values + and –; implication 11 follows analogously from FCR 1. The implications 1, 6, and 7 are missing in the FCRs and moreover only one direction of the equivalences 2, 4, and 5 is stated in the FCRs. These implications regulate when categories necessarily must be specified with respect to certain features without saying anything about the concrete feature values. A surprising gap in the list of FCRs is evident in implication 10, according to which passive verbs are not auxiliaries. This fact cannot be derived from the FCRs in (25). The GPSG grammar given in Gazdar et al. (1985) thus allows the feature-specification bundle {[PAS], [+AUX], [–INV]}, which encodes a non-inverted auxiliary in passive. This demonstrates that the automatically extracted complete Horn theory is more explicit than the manually formulated FCRs, which arise from linguistic intuition and miss statements which were probably too obvious for the investigators. Hence, FCA can be a powerful tool for tracking down gaps in theories that experts have constructed manually. However, the FCRs of GPSG are not restricted to attribute implications (Horn clauses). Some of the FCRs in Gazdar et al. (1985: 246) make use of negation and disjunction. The following consideration shows that such FCRs are implicitly encoded in the concept lattice too. As we saw before, a clause with a disjunctive premise and conjunctive conclusion can be directly transformed into a set of attribute implications. Hence, we can focus on the derivation of clauses with disjunctive conclusions from concept lattices. The information domain of a complete observational theory of a formal context consists exactly of the intents of the object concepts. These considerations open a way to derive a complete observational theory and hence a complete set of FCRs from a concept lattice: Each formal concept (A,B) of the lattice which is not an object concept corresponds to an observational statement whose premise is the conjunction of the elements of the intent B and whose conclusion is the disjunction of the conjunctions of its subconcept intents minus B. Adding the corresponding statement of a concept to the theory amounts to removing the concept

464 James Kilbury, Wiebke Petersen, Christof Rumpf intent from the informational domain of the theory. For example, the statement corresponding to the attribute concept of nform:val is n:VAL
v:VAL
v:-
n:+
nform:VAL
nform:norm  nform:it which can be simplified by (30) to nform:VAL
nform:norm  nform:it This states that if an object bears the feature nform, then the latter must be specified for the value norm or it.9 The necessary FCRs can be generated fully automatically with FCA. The task of the grammar developer is thus restricted to the selection of suitable features and the construction of a feature-structure context like that of (24). In this section we have shown so far how a lexicon in pure list form, as in (24), can be used to produce either GPSG FCRs or an HPSG type signature. It remains to show how feature structure contexts can be derived either from a type signature or from FCRs. Given a type signature, the adequate feature structure context can be directly constructed from the set of totally well-typed and sort-resolved feature structures. A detailed description of the construction method (even for type signatures with co-references) is given in Petersen (forthcoming). Given a set of FCRs, the corresponding formal context is equivalent to the information domain of the FCRs (see Osswald & Petersen, 2003). Hence, FCA enables us to switch directly from the format of a type signature to FCRs and back. 3.3.2. Inherent lexical features versus inflectional classes This section will present a further example for the applicability of FCA in linguistics, and, especially, in the lexicon. In particular we wish to show how different approaches to the analysis of German noun inflection can be reconciled with each other. Classical approaches to inflectional morphology as seen, e.g., in Wahrig (1966) assign word stems an inflectional class in the lexicon; the inflec———–———————————————————————————— 9

Ganter & Krause (1999) present an alternative procedure for systematically constructing a complete observational theory of a formal context.

Inheritance-based models of the lexicon

465

tional class itself determines the inflectional paradigm of a stem. Minimalist Morphology, however, argues that “the membership in an inflectional subclass should not be arbitrarily assigned, but rather follow from features that can be memorized: either on the basis of a substantial property of the stem itself (such as gender or phonological shape), or on the basis of an additional lexical entry.” (Wunderlich 1999). Different inflectional paradigms typically reveal numerous similarities so that it is appropriate to model hierarchical relations between the classes in order to avoid unnecessary redundancy. Kilbury (2001) analyzes the inflectional classes of German nouns and proposes the hierarchical ordering shown in (31).10 From the class NS of strong nouns without umlaut in plural the class NU with umlaut in plural inherits all the singular forms as well as the fact that the plural forms end in a schwa-syllable. According to Minimalist Morphology it is necessary to analyze such relations in terms of lexically inherent, memorizable features that determine the class membership of individual nouns. (31) Analysis of the inflection classes of German nouns and their hierarchical relations in Kilbury (2001) NA 3 NS NM ! ! NU NWN ! ! NR NWS

NA

regular nouns (i.e., with s-plural)

NS

strong nouns (i.e., with a schwa-syllable in plural)

NU

strong nouns with umlaut in plural

NR

strong nouns with r-plural

NM

mixed nouns (with n-plural)

NWN

weak nouns without s in gen sg

NWS

weak nouns with s in gen sg

(32) shows a list of representative German nouns together with their inflectional classes and the features proposed in Kilbury (2001) as determining the class membership; clearly, this is a formal context. In (33) we see the corresponding concept lattice for this context, in which the gender features have been omitted for better legibility. The nodes in the hierarchy, which correspond to an inflectional class, are labelled accordingly.

———–———————————————————————————— 10

Further examples of hierarchical analyses of the nominal inflection classes are given in Cahill & Gazdar (1999).

466 James Kilbury, Wiebke Petersen, Christof Rumpf (32) German nouns together with their inflectional classes and the features that determine these classes according to Kilbury (2001)*

*The following abbreviations for features are used: nt f m schwa inan RFS_pl

typical noun feminine masculine stem-final -e (schwa) inanimate reduced final syllable in plural

uml_pl r_pl n_pl n_obl

umlaut in plural suffix -r in plural suffix -n in plural suffix -n in singular nonnominative (i.e., oblique) ns_gen -n-s in genitive singular

Inheritance-based models of the lexicon

467

(33) Concept lattice from the context in (32)

It is not necessary to know all the features of a noun in order to inflect it since, normally, some follow from others. For each noun, Kilbury (2001) therefore distinguishes between the features that are distinctive, redundant (i.e. those that follow from others), or irrelevant for inflection. The previously unsolved problem of (semi-)automatically separating the distinctive and redundant features is one for which the method we have proposed lends

468 James Kilbury, Wiebke Petersen, Christof Rumpf itself. In particular, the representation of feature distributions in the form of feature implications makes it possible to characterize the interdependencies between the features explicitly and to depict them visually. On the basis of (32) in the following we will show a procedure with which the membership of nouns in inflectional classes can be determined. To do this we first augment the context with features for the inflectional classes. We then consider the set of valid feature implications for the dichotomous context, i.e., the context, which also contains the negation of each feature. We obtain seven independent implications in which a class feature constitutes the conclusion, as shown in (34). (34) Dependency of the inflectional classes on the features of (32)

(35) Hierarchical representation of the dependency relationship of the inflectional classes from features of (32)

Inheritance-based models of the lexicon

469

Minimalist Morphology assumes a principle of underspecification according to which unspecified features are interpreted as being negatively specified. Given this assumption the seven implications can be depicted in a hierarchy as in (35a). Edges are labelled with the distinctive features that determine class membership, and these features are percolated down along the edges. The features in (32) obviously differ with respect to their memorizability. The first five are more easily memorized and thus more plausible than the last six features. In order to take this into account we have recalculated the valid implications of the context disregarding the last six features. As shown in (36), this results in three new implications with a class feature which are not in (34). (36) Additional dependencies of the inflectional classes on the features of (32) which are easier to memorize

When these implications are added, the hierarchy in (35b) results. Note that dotted edges, in contrast to normal ones, do not denote inheritance. This hierarchy strongly resembles the manually constructed hierarchy of Kilbury (2001). 3.4. General remarks on Formal Concept Analysis as a tool for automatic induction “Most other techniques of data analysis have the aim to drastically reduce the given information and to obtain a few significant parameters. By contrast, a concept lattice does not reduce complexity since it contains all the details of the data represented by the formal context. […] Nevertheless it may be exponential in size, compared to the formal context. Complexity therefore is, of course, a problem, even though there are efficient algorithms and advanced program systems. A formal context of moderate size may have more concepts than one would like to see individually.”

(Ganter & Wille 1998) This quotation expresses the particular strength of concept lattices as well as the problems that emerge when they are employed for the induction of hierarchies. FCA is superior to many alternative approaches to induction inasmuch as it is neutral with respect to the analyzed data and that it describes them completely. The relations between formal contexts and the correspond-

470 James Kilbury, Wiebke Petersen, Christof Rumpf ing concept lattices is absolutely transparent since there is exactly one of the latter for each context. When examination of the valid attribute implications reveals strange statements, this points either to new discoveries11 or else to an incomplete or incorrect context. A semi-automatic attribute exploration can be carried out in order to complete a context (cf. Stumme 1996; Ganter & Wille 1999).12 We saw that FCA provides three different ways to display data: in tabular form, as a hierarchy, and as a set of implications. Each of these representations can be useful as a tool for various tasks in an analysis. Data frequently are available in tabular form, which facilitates input. Information about a single object can most easily be read from the context, while the hierarchical concept lattice shows information about connections which are only implicit in the context. Furthermore, the concept lattice gives a more compact and less redundant representation of the data in comparision with the context. Finally, attribute implications express dependencies between the attributes. In the following we will address two problems involving the application of FCA and possible solutions. 3.4.1. Pruning the hierarchies Concept lattices explicitly include all possible generalizations since each set of common attributes induces a corresponding node, so the induced hierarchies can become very large. In the case of the lexical data base CELEX13 the concept lattice corresponding to the context that describes German lexemes and their derivational potential contains more than 72,000 concepts. The basic context, which likewise was extracted from CELEX using our system FCALING, has 9,567 objects and 2,032 attributes. One possibility for avoiding such large hierarchies is, as Sporleder (2003) discusses, to take the highly differentiated hierarchies as a point of departure for pruning; here one may have to accept a high cost for the computation of large hierarchies, but the approach has the advantage that the pruning algorithms can be specially tailored to the individual task. Sporle———–———————————————————————————— 11 12 13

Remember implication 10 in (30). The program ConImp can be used for feature explorations. CELEX, which is compiled by the Dutch Center for Lexical Information, consists of three large electronic databases and provides users with detailed English, German, and Dutch lexical data.

Inheritance-based models of the lexicon

471

der achieves this by training her pruning algorithm on manually constructed hierarchies. The resulting hierarchies, however, are dependent on the learned pruning parameters. In Sporleder’s approach this is intended, of course, since the pruning parameters reflect linguistic expert knowledge and are aimed at inducing hierarchies which are linguistically plausible. Here we present a possibility for reducing the size of hierarchies that has already been described in Petersen (2004a). Instead of starting with the concept lattice of a context we take just the partially ordered set (poset) of object and attribute concepts and then add unique maximal and minimal elements; in this way one obtains a hierarchy, the AOC-poset (attribute-object-poset), which normally is much more compact than the lattice.14 In the case of the CELEX context mentioned above the hierarchy is reduced to one with less than 4,000 nodes. The AOC-poset, just like the concept lattice, is directly defined by the context. If the AOC-poset is labelled as an inheritance hierarchy as before, there again results a redundancy-free inheritance hierarchy, from which the context can be reconstructed.15 The shift to AOC-posets also reduces memory requirements. In computing an AOC-poset only the attribute and object concepts must be computed, which can be done independently of the others, so that the procedure is very efficient. Thus, in comparision to concept lattices AOC-posets offer a very simple method for the induction of redundancy-free inheritance hierarchies from large data sets. Inference tasks, however, are better supported by concept lattices, due to the explicit representation of shared attributes. Concept lattices are also better when the obtained hierarchies must be visualized, since AOC-posets normally are less legible due to their numerous crossing edges. Furthermore, valid attribute implications can be read off directly only in the lattice. 3.4.2. The problem of nonmonotonic hierarchies In the preceding sections we have seen how FCA can be used to induce monotonic inheritance hierarchies. The relationship between a context and the corresponding monotonic hierarchy is completely transparent. ———–———————————————————————————— 14 15

If a context consists of g objects and m features, then the maximum number of concepts is 2min{g,m}, while that of nodes in an AOC-poset is at most g + m + 2. In section 3.3.1 we presented a method to induce non-Horn clauses from a formal context, which can be applied to construct the complete theory describing the AOC-poset.

472 James Kilbury, Wiebke Petersen, Christof Rumpf In contrast, induction of nonmonotonic hierarchies introduces special problems since there can be no such transparency between the contexts and hierarchies. Information in nonmonotonic hierarchies can be overwritten arbitrarily often, so there is an unlimited number of possibilities for representing data in such hierarchies. The most important decision in constructing a nonmonotonic hierarchy is about what information should be regarded as regular. We will now show how monotonic lattices can be employed as a point of departure for the transformation into nonmonotonic hierarchies. For a given context we first find the set of attributes which describe a standard object of the context. Three conditions are placed on such a default set of attributes (cf. Petersen 2004a): it must be internally consistent, its use must result in a more compact representation of the data, and it must be closed with respect to the valid attribute implications of the context. Such sets are just the intents of concepts with nonempty extents, and this connection allows such possible sets to be computed efficiently. (37) Example of a nonmonotonic inheritance network for the context in (32)

Since the choice of a plausible default attribute set is only possible with the aid of expert knowledge, we propose a semi-automatic procedure. Our system FCALING presents the user with all theoretically possible default attribute sets ordered according to the size of the corresponding extents. After the selection of a set, FCALING calculates the corresponding nonmonotonic hierarchy. This procedure can be applied interactively in order to represent

Inheritance-based models of the lexicon

473

subregularities as well. (37) shows a nonmonotonic inheritance network constructed with our method for the data from the context in (32). Note that defeasible attributes appear in parentheses. The hierarchy is based on the default attribute set {gender: masc, sing nom:*, sing gen:*_s, sing dat:*, sing acc:*, plur nom:*_n, plur gen:*_n, plur dat:*_n, plur acc:_n}. 4. Conclusion It is clear that the lexicon has become a major focus of linguistic investigation in the past quarter century. Research in computational linguistics has increasingly been aimed at modelling hierarchical relations within the lexicon and their formal representation. In particular, much recent work has concentrated on inheritance-based formalisms for typed feature structures, the choice between monotonic and nonmonotonic versions of these formalisms, and the automatic induction of classifications in general. Nonmonotonic inheritance is important as a means of capturing the continuum of regularity, subregularity, and irregularity in the lexicon. These topics have been at the center of our attention in the present paper. We have presented our formalism QType, which combines major elements of typed unification-based formalisms with nonmonotonic inheritance hierarchies. Crucially, we confine nonmonotonicity to the static type hierarchy and compile the latter into a strictly monotonic counterpart for efficient processing. We have described a technique using Formal Concept Analysis that aids in the construction of type signatures such as those of QType and other related formalisms; see Petersen (forthcoming) for an elaboration of this technique. At the same time we have shown how FCA provides an adequate formal basis for explicating the relationship between the type signatures of HPSG and the feature co-occurrence restrictions of GPSG. Future work will undoubtedly seek to clarify further the empirical linguistic motivation for nonmonotonic type hierarchies as well as the formal constraints governing their automatic induction. References Aït-Kaci, Hassan, Robert Boyer, Patrick Lincoln and Roger Nasr 1989 Efficient implementation of lattice operations. Programming Languages and Systems 11: 115–146.

474 James Kilbury, Wiebke Petersen, Christof Rumpf Barg, Petra 1996

Automatischer Erwerb von linguistischem Wissen. Ein Ansatz zur Inferenz von DATR-Theorien. Tübingen: Niemeyer. Bernhardt, Wolfram 2001 Möglichkeiten der Effizienzsteigerung bei der Verarbeitung getypter Merkmalstrukturen. Master’s thesis, Universität Düsseldorf. Boguraev, Branimir and James Pustejovsky (eds.) 1996 Corpus Processing for Lexical Acquisition. Cambridge, MA: MIT Press. Bouma, Gosse, Frank Van Eynde and Dan Flickinger 2000 Constraint-based lexica. In Van Eynde and Gibbon (2000), 43–71. Bouma, Gosse 1990 Defaults in unification grammar. Proceedings of the Annual Meeting of the Association for Computational Linguistics 28: 165–173. Bresnan, Joan (ed.) 1982 The Mental Representation of Grammatical Relations. Cambridge, MA: MIT Press. Briscoe, Ted, Valeria de Paiva and Ann Copestake (eds.) 1993 Inheritance, Defaults, and the Lexicon. Cambridge University Press. Briscoe, Ted and Ann Copestake 1999 Lexical rules in constraint-based grammars. Computational Linguistics 25: 487–526. Cahill, Lynne and Gerald Gazdar 1999 German noun inflection. Journal of Linguistics 35: 1–42. Carpenter, Bob and Gerald Penn 1999 ALE: The Attribute Logic Engine User's Guide. Murray Hill, NJ: Bell Laboratories, Lucent Technologies. Carpenter, Bob 1993 Sceptical and credulous default unification with application to templates and inheritance. In Briscoe et al. (1993), 13–37. Copestake, Ann 2002 Implementing Typed Feature Structure Grammars. Stanford: CSLI. Covington, Michael 1993 Natural Language Processing for Prolog Programmers. Englewood Cliffs, NJ: Prentice Hall. Daelemans, Walter and Gert Durieux 2000 Inductive lexica. In Van Eynde and Gibbon (2000), 115–136. De Smedt, Koenraad 1984 Using object-oriented knowledge representation techniques in morphology and syntax programming. Proceedings of the European Conference on Artificial Intelligence 1984: 181–184. Dörre, Jochen and Michael Dorna 1993 CUF – A Formalism for Linguistic Knowledge Representation. DYANA 2 deliverable R.1.2A, Universität Stuttgart.

Inheritance-based models of the lexicon

475

Evans, Roger and Gerald Gazdar 1989 Inference in DATR. Proceedings of the Conference of the European Chapter of the Association for Computational Linguistics 4: 66–71. 1996 DATR: a language for lexical knowledge representation. Computational Linguistics 22: 167–216. Flickinger, Dan 1987 Lexical rules in the hierarchical lexicon. PhD thesis, Stanford University. Ganter, Bernhard and Rudolf Wille 1998 Applied lattice theory: Formal concept analysis. In General Lattice Theory, G. Grätzer (ed.), 591–605. Basel: Birkhäuser Verlag. 1999 Formal Concept Analysis. Mathematical Foundations. Berlin: Springer. Gazdar, Gerald 1987 Linguistic applications of default inheritance mechanisms. In Linguistic Theory and Computer Applications, P. Whitelock (ed.), 37– 67. London: Academic Press. Gazdar, Gerald and Geoffrey Pullum 1982 Generalized Phrase Structure Grammar: A Theoretical Synopsis. Bloomington, Indiana: Indiana University Linguistics Club. Gazdar, Gerald, Ewan Klein, Geoffrey Pullum and Ivan Sag 1985 Generalized Phrase Structure Grammar. Oxford: Blackwell. Gerdemann, Dale and Paul John King 1993 Typed feature structures for expressing and computationally implementing feature cooccurence restrictions. Proceedings of 4. Fachtagung der Sektion Computerlinguistik der Deutschen Gesellschaft für Sprachwissenschaft: 33–39. 1994 The correct and efficient implementation of appropriateness specifications for typed feature structures. Proceedings of the 15th Conference on Computational Linguistics: 956–960. Götz, Thilo, Detmar Meurers and Dale Gerdemann 1997 The ConTroll Manual. Ms., Universität Tübingen. Hopcroft, John and Jeffrey Ullman 1979 Introduction to Automata Theory, Languages and Computation. Reading, MA: Addison-Wesley. Kaplan, Ronald and Joan Bresnan 1982 Lexical-functional grammar: a formal system for grammatical representation. In Bresnan (1982), 173–281. Karttunen, Lauri 1986 Radical Lexicalism (= Report No. CSLI-86-68). Stanford: CSLI. Keller, Bill 1993 Feature Logics, Infinitary Descriptions and Grammar. Stanford: CSLI.

476 James Kilbury, Wiebke Petersen, Christof Rumpf Kilbury, James 2001 German noun inflection revisited. Journal of Linguistics 37: 339–353. Kilbury, James, Petra Naerger [Barg] and Ingrid Renz 1991 DATR as a lexical component for PATR. Proceedings of the Conference of the European Chapter of the Association for Computational Linguistics 5: 137–142. Koenig, Jean-Pierre 1999 Lexical Relations. Stanford: CSLI. Krieger, Hans-Ulrich and John Nerbonne 1993 Feature-based inheritance networks for computational lexicons. In Briscoe et al. (1993), 90–136. Krieger Hans-Ulrich, Hannes Pirker and John Nerbonne 1993 Feature-based allomorphy. Proceedings of the Annual Meeting of the Association for Computational Linguistics 31: 140–147. Lascarides, Alex and Ann Copestake 1999 Default representation in constraint-based frameworks. Computational Linguistics 25: 55–106. Osswald, Rainer and Wiebke Petersen 2002 Induction of classifications from linguistic data. Proceedings of the ECAI-Workshop on Advances in Formal Concept Analysis for Knowledge Discovery in Databases. 2003 A logical approach to data-driven classification. Lecture Notes in Computer Science 2821: 267–281. Penn, Gerald 2000 The algebraic structure of attributed type signatures. PhD thesis, School of Computer Science, Carnegie Mellon University. Pereira, Fernando and Stuart Shieber 1987 Prolog and Natural Language Analysis. Stanford: CSLI. Petersen, Wiebke 2004a A set-theoretic approach for the induction of inheritance-hierachies. In Proceedings of the Joint Conference on Formal Grammar and Mathematics of Language (FG/MOL-01), Electronic Notes in Theoretical Computer Science 53: 296–308. 2004b Automatic induction of type signatures. Unpublished manuscript. forthc. Induktion lexikalischer Vererbungshierarchien mit Mitteln der formalen Begriffsanalyse. PhD thesis (in prep.), University of Düsseldorf. Petersen, Wiebke and James Kilbury 2005 What feature co-occurrence restrictions have to do with type signatures. Proceedings of the Joint Conference on Formal Grammar and Mathematics of Language (FG/MOL-05). Pollard, Carl and Ivan Sag 1987 Information-based Syntax and Semantics, Vol. 1. Stanford: CSLI. 1994 Head-driven Phrase Structure Grammar. Stanford: CSLI.

Inheritance-based models of the lexicon

477

Riehemann, Susanne 1998 Type-based derivational morphology. Journal of Comparative Germanic Linguistics 2: 49–77. Rumpf, Christof forthc. Default inheritance in constraint-based frameworks. PhD thesis (in prep.), University of Düsseldorf. Russell, Graham, John Carroll and Susan Warwick-Armstrong 1991 Multiple default inheritance in a unification-based lexicon. Proceedings of the Annual Meeting of the Association for Computational Linguistics 29: 215–221. Russell, Graham, Afzal Ballim, John Carroll and Susan Warwick-Armstrong 1992 A practical approach to multiple default inheritance for unificationbased lexicons. Computational Linguistics 18: 311–337. Sag, Ivan and Thomas Wasow 1999 Syntactic Theory: A Formal Introduction. Stanford: CSLI. Shieber, Stuart 1986 An Introduction to Unification-Based Approaches to Grammar. Stanford: CSLI. Shieber, Stuart, Hans Uszkoreit, Fernando Pereira, Jane Robinson and Mabry Tyson 1983 The formalism and implementation of PATR-II. In Research on Interactive Acquisition and Use of Knowledge, B. J. Grosz and M. Stickel (eds.), 39–79. Menlo Park, CA: SRI. Sporleder, Caroline 2003 Discovering lexical generalisations. A supervised machine learning approach to inheritance hierarchy construction. PhD thesis, Institute for Communicationg and Collaborative Systems, University of Edinburgh. Stumme, Gerd 1996 Attribute exploration with background implications and exceptions. In Data Analysis and Information Systems. Statistical and Conceptual Approaches, H.-H. Bock and W. Polasek (eds.), 457–569. Berlin: Springer. Touretsky, David 1986 The Mathematics of Inheritance Systems. London: Pitman. Van Eynde, Frank and Dafydd Gibbon (eds.) 2000 Lexicon Development for Speech and Language Processing. Dordrecht et al.: Kluwer. Wahrig, Gerhard 1966 Das Große Deutsche Wörterbuch. Gütersloh: Bertelsmann LexikonVerlag. Wunderlich, Dieter 1999 German noun plural reconsidered. Behavioral and Brain Sciences 22: 1044 –1045.

Appendix 1: List of subprojects in the SFB 282

Clahsen, Harald (Düsseldorf/Essex) 1991–1993 Lernbarkeitstheorie und lexikalisches Lernen [Learnability theory and lexical learning] 1994–1999 Verarbeitung und Repräsentation von Flexionselementen im Lexikon [Processing and representation of inflectional elements in the lexicon] 2000–2002 Erwerb und Verarbeitung von Flexions- und Derivationselementen des Deutschen [Acquisition and processing of inflectional and derivational elements in German] Beeh, Volker (Düsseldorf) 1991–1993 Feldstrukturen im Lexikon [Field structures in the lexicon] Grijzenhout, Janet (Düsseldorf) 1998–1999 Output-Beschränkungen in der Lexikalischen Phonologie [Output restriction in the lexical phonology] 2000–2002 Lexikalische Phänomene in der OT-Korrespondenztheorie [Lexical phenomena in the OT correspondence theory] Jacobs, Joachim (Wuppertal) 1991–1999 Valenz im Lexikon [Valency in the lexicon] 2000–2002 Nominalisierte Infinitive [Nominalized infinitives] 1998–1999 Schnittstelle Lexikon/Syntax: Prädikative Nominalphrasen [Interface lexicon/syntax: predicative nominal phrases] Kilbury, James (Düsseldorf) 1991–1996 Simulation lexikalischen Erwerbs [Simulation of lexical acquisition] 1997–1999 Dynamische Erweiterung des Lexikons [Dynamic extensions of the lexicon] 2000–2002 Modellierung von Subregularität im Lexikon [Modelling of subregularities in the lexicon] Lang, Ewald (Wuppertal) 1991–1995 Duale Operatoren [Dual operators] Löbner, Sebastian (Düsseldorf) 1991–2002 Verbbedeutungen [Verb meanings] Lenerz, Jürgen (Köln) 1991–1996 Derivation und lexikalische Semantik [Derivation and lexical semantics] 1997–1999 Syntaktische Prinzipien in der Wortbildung [Syntactic principles of word formation]

480 Appendix 1: List of subprojects in the SFB 282 Penke, Martina (Düsseldorf) 1997–2002 Neurolinguistische Untersuchungen zur Flexions- und Derivationsmorphologie [Neurolinguistic investigations of inflectional and derivational morphology] Primus, Beatrice (Köln) 2000–2002 Synchrone und diachrone Konstruktionsvariation bei psychischen Verben [Synchronic and diachronic variation of constructions with psych verbs] Rauh, Gisa (Wuppertal) 1991–1996 Syntaktische Eigenschaften englischer Präpositionen, Konjunktionen und Adverbien im Lexikon [Syntactic properties of English prepositions, conjunctions and adverbs] Sasse, Hans-Jürgen (Köln) 1991–1998 Das Nomen im Lexikon [The noun in the lexicon] Sasse, Hans-Jürgen & Werner Drossard (Köln) 2000–2002 Lexikalische Kategorien und Argumentlinking in polysynthetischen Sprachen [Lexical categories and argument linking in polysynthetic languages] Scherfer, Peter (Wuppertal) 1993–1996 Struktur und Genese des mentalen Lexikons bei Fremdsprachenlernern [Structure and genesis of the mental lexicon of foreign language learners] Vater, Heinz (Köln) & Richard Wiese (Düsseldorf) 1994–1999 Prosodische Morphologie [Prosodic morphology] Wiese, Richard (Düsseldorf) 1991–1993 Ebenen im Lexikon [Levels in the lexicon] 1994–1996 Lexikalische Phonologie: Zur Theorie phonologischer Regeln [Lexical phonology: On the theory of phonological rules] Wunderlich, Dieter (Düsseldorf) 1991–1996 Lexikalische Fundierung der Kongruenz [Lexical foundations of agreement] 1997–2002 Kategorien und Systeme der Flexion [Categories and systems of inflection] 1994–1996 Verbstrukturen [Verb structures] Wunderlich, Dieter & Barbara Stiebels (Düsseldorf) 1997–2002 Verbstrukturen: Komplexe Prädikate und Argumentlinking [Verb structures: Complex predicates and argument linking] Wunderlich, Dieter & Janet Grijzenhout (Düsseldorf) 2000–2002 Nichtkonkatenative Morphologie [Nonconcatenative morphology]

Appendix 2: List of major publications from SFB 282

Monographs Adone, Dany 1994 The Acquisition of Mauritian Creole. Amsterdam: Benjamins. 2002 A Cognitive Theory of Creole Genesis. Habilitationsschrift Düsseldorf Barg, Petra 1996 Automatischer Erwerb von linguistischem Wissen: Ein Ansatz zur Inferenz von DATR-Theorien. Tübingen: Niemeyer. Bartke, Susanne 1998 Experimentelle Studien zur Flexion und Wortbildung. Tübingen: Niemeyer. Blume, Kerstin 2000 Markierte Valenzen im Sprachvergleich: Lizenzierungs- und Linkingbedingungen. Tübingen: Niemeyer. 2004 Nominalisierte Infinitive. Eine empirisch basierte Studie zum Deutschen. Tübingen: Niemeyer. Clahsen, Harald 1991 Child Language and Developmental Dysphasia. Amsterdam: Benjamins. Clahsen, Harald (ed.) 1996 Generative Perspectives on Language Acquisition. Amsterdam: Benjamins. Eisenbeiß, Sonja 2002 Merkmalsgesteuerter Grammatikerwerb: Eine Untersuchung zum Erwerb der Struktur und Flexion von Nominalphrasen. PhD diss., Düsseldorf Engelberg, Stefan 2000 Verben, Ereignisse und das Lexikon. Tübingen: Niemeyer. Evans, Nicholas and Hans-Jürgen Sasse (eds.) 2002 Problems of Polysynthesis. Berlin: Akademie Verlag. Fabri, Ray 1993 Kongruenz und die Grammatik des Maltesischen. Tübingen: Niemeyer. Fabri, Ray, Albert Ortmann and Teresa Parodi (eds.) 1998 Models of Inflection. Tübingen: Niemeyer.

482 Appendix 2: List of major publications from SFB 282 Gamerschlag, Thomas 2003 Komposition und Argumentstruktur komplexer Verben. Eine lexikalische Analyse von Verb-Verb-Komposita und Serialverbkonstruktionen. Berlin: Akademie Verlag. Gerlach, Birgit and Janet Grijzenhout (eds.) 2000 Clitics in Phonology, Morphology and Syntax. Amsterdam: Benjamins. Gerlach, Birgit 2002 Clitics between Syntax and Lexicon. Amsterdam: Benjamins. Grijzenhout, Janet 2001 West-Germanic Sound Structure: A Contrastive Study. Habilitationsschrift Düsseldorf. Hong, Upyong 1995 Null-Subjekte im Erst- und Zweitspracherwerb des Deutschen. Tübingen: Narr. Indefrey, Peter 2002 Listen und Regeln. Erwerb und Repräsentation der schwachen Substantivdeklination im Deutschen. PhD diss. Düsseldorf Jacobs, Joachim 1994 Kontra Valenz. Trier: Wissenschaftlicher Verlag. Janssen, Ulrike 2003 Wortakzent im Deutschen und Niederländischen: Empirische Untersuchungen und theoretische Analysen. PhD diss. Düsseldorf. Joppen-Hellwig, Sandra 2001 Verbklassen und Argumentlinking. Nicht-kanonische Argumente, Expletiva und vierstellige Kausativa in Ergativ- versus Akkusativsprachen. Tübingen: Niemeyer. Kaufmann, Ingrid 1995 Konzeptuelle Grundlagen semantischer Dekompositionsstrukturen. Die Kombinatorik lokaler Verben und prädikativer Komplemente. Tübingen: Niemeyer. 2004 Medium und Reflexiv. Eine Studie zur Verbsemantik. Tübingen: Niemeyer. Kaufmann, Ingrid and Barbara Stiebels (eds.) 2002 More than Words: A Festschrift for Dieter Wunderlich. Berlin: Akademie Verlag. Kehrein, Wolfgang and Richard Wiese (eds.) 1998 Phonology and Morphology of the Germanic Languages. Tübingen: Niemeyer. Krämer, Martin 2003 Vowel Harmony and Correspondence Theory. Berlin /New York: Mouton de Gruyter. Lakämper, Renate 2000 Objekt- und Pluralmarkierung in Quechua. PhD diss., Düsseldorf.

Appendix 2: List of major publications from SFB 282 483 Lenz, Barbara 1995 un-Affigierung: unrealisierbare Argumente, unausweichliche Fragen, nicht unplausible Antworten. Tübingen: Narr. Löbel, Elisabeth and Gisa Rauh (eds.) 1997 Lexikalische Kategorien und Merkmale, Tübingen: Niemeyer. Löbner, Sebastian 2002 Understanding Semantics. London: Arnold. 2003 Semantik. Berlin /New York: Mouton de Gruyter. Mattissen, Johanna 2003 Dependent-Head Synthesis in Nivkh. A Contribution to a Typology of Polysynthesis. Amsterdam: Benjamins. Mori, Yoshiko 1998 Aspekt und Verbsemantik. PhD diss. Düsseldorf. Naumann, Ralf 1995 Aspectual Composition. Habilitationsschrift Düsseldorf, Neef, Martin 1996 Wortdesign. Eine deklarative Analyse der deutschen Verbflexion. Tübingen: Stauffenburg. Ortmann, Albert 2002 Kategorien des Nomens: Schnittstellen und Ökonomie. Tübingen: Niemeyer. Parodi, Teresa 1998 Der Erwerb funktionaler Kategorien im Deutschen. Tübingen: Narr. Penke, Martina 1998 Die Grammatik des Agrammatismus. Eine linguistische Untersuchung zu Wortstellung und Flexion bei Broca-Aphasie. Tübingen: Niemeyer. 2006 Flexion im mentalen Lexikon. Tübingen: Niemeyer. Popescu Alexandra 2003 Morphophonologische Phänomene des Rumänischen. PhD diss. Düsseldorf Renz, Ingrid 1993 Adverbiale im Deutschen: Ein Vorschlag zu ihrer Klassifikation und unifikationsbasierten Repräsentation. Tübingen: Niemeyer. Siebert, Susann 1999 Wortbildung und Grammatik. Syntaktische Restriktionen in der Struktur komplexer Wörter. Tübingen: Niemeyer. Sonnenstuhl, Ingrid 2003 Deutsche Plurale im mentalen Lexikon. Tübingen: Niemeyer Stiebels, Barbara 1996 Lexikalische Argumente und Adjunkte. Berlin: Akademie Verlag. 2002 Typologie des Argumentlinkings: Ökonomie und Expressivität. Berlin: Akademie Verlag. Stiebels, Barbara and Dieter Wunderlich (eds.) 2000 Lexicon in Focus. Berlin: Akademie Verlag.

484 Appendix 2: List of major publications from SFB 282 Walther, Markus 1999 Deklarative prosodische Morphologie: Constraint-basierte Analysen und Computermodelle zum Finnischen und Tigrinya. Tübingen: Niemeyer. Weyerts, Helga 1997 Reguläre und irreguläre Flexion: Psycholinguistische und neurophysiologische Ergebnisse zu Erwerb, Verarbeitung und mentaler Repräsentation. PhD diss. Düsseldorf. Wiese, Richard 1996 The Phonology of German. (2nd ed. 2000). Oxford: University Press.

Articles Barg, Petra 1996

Automatic inference of DATR theories. In Data Analysis and Information Systems: Statistical and Conceptual Approaches, Hans-Herrmann Bock and Wolfgang Polasek (eds.), 506–515. Berlin: Springer. Barg, Petra and Markus Walther 1998 Processing unknown words in HPSG. Proceedings of COLINGACL98, 91–95. Montréal. Bartke, Susanne, Gary F. Marcus and Harald Clahsen 1995 Acquiring German noun plurals. In Proceedings of the 19 th Annual Boston University Conference on Language Development, Dawn MacLaughlin and Susan McEwen (eds.), 60–69. Boston: Cascadilla. Blume, Kerstin 1998 A contrastive analysis of interaction verbs with dative complements. Linguistics 36: 253–280. Broschart, Jürgen 1997 Why Tongan does it differently: Categorial distinctions in a language without nouns and verbs. Linguistic Typology 1: 123–165. Clahsen, Harald 1999 Lexical entries and rules of language: A multi-disciplinary study of German inflection. Behavioral and Brain Sciences 22: 991–1013. Clahsen, Harald and Martina Penke 1992 The acquisition of agreement morphology and its syntactic consequences. In The Acquisition of Verb Placement, Jürgen M. Meisel (ed.), 181–224. Dordrecht: Kluwer. Clahsen, Harald, Monika Rothweiler, Andreas Woest and Gary F. Marcus 1992 Regular and irregular inflection in the acquisition of German noun plurals. Cognition 45: 225–255.

Appendix 2: List of major publications from SFB 282 485 Clahsen, Harald, Martina Penke and Teresa Parodi 1993 Functional categories in early child German. Language Acquisition 3: 395–429. Clahsen, Harald and Monika Rothweiler 1993 Inflectional rules in children's grammars: Evidence from German participles. Yearbook of Morphology 1992: 1–34. Clahsen, Harald, Sonja Eisenbeiß and Anne Vainikka 1994 The seeds of structure. A syntactic analysis of the acquisition of case marking. In Language Acquisition Studies in Generative Grammar, Teun Hoekstra and Bonnie D. Schwartz (eds.), 85–118. Amsterdam: Benjamins. Clahsen, Harald and Upyong Hong 1995 Agreement and null-subjects in German L2 development. New evidence from reaction-time experiments. Second Language Research 11: 57–87. Clahsen, Harald, Upyong Hong and Ingrid Sonnenstuhl-Henning 1995 Grammatical constraints in syntactic processing: Sentence-matching experiments on German. The Linguistic Review 12: 5–33. Clahsen, Harald, Sonja Eisenbeiß and Martina Penke 1996 Lexical learning in early syntactic development. In Clahsen (ed.), 129–159. Clahsen, Harald, Gary F. Marcus, Susanne Bartke and Richard Wiese 1996 Compounding and inflection in German child language. Yearbook of Morphology 1995: 115–142. Clahsen, Harald, Sonja Eisenbeiß and Ingrid Sonnenstuhl-Henning 1997 Morphological structure and the processing of inflected words. Theoretical Linguistics 23: 201–249. Clahsen, Harald, Sonja Eisenbeiß, Meike Hadler and Ingrid Sonnenstuhl 2001a The mental representation of inflected words: An experimental study of adjectives and verbs in German. Language 77: 510–543. 2001b Morphological paradigms in language processing and language disorders. Transactions of the Philological Society 99: 247–283. Clahsen, Harald, Fraibet Aveledo and Iggy Roca 2002 The development of regular and irregular verb inflection in Spanish child language. Journal of Child Language 29: 591–622. Clahsen, Harald, Peter Prüfert, Sonja Eisenbeiß and Joana Cholin 2002 Strong stems in the German mental lexicon: Evidence from child language acquisition and adult processing. In Kaufmann and Stiebels (eds.), 91–112. Clahsen, Harald, Ingrid Sonnenstuhl and James P. Blevins 2003 Derivational morphology in the German mental lexicon: A dualmechanism account. In Morphological Structure in Language Processing, Harald Baayen and Robert Schreuder (eds.), 115–155. Berlin / New York: Mouton de Gruyter.

486 Appendix 2: List of major publications from SFB 282 Eisenbeiß, Sonja 2000 The acquisition of the determiner phrase in German child language. In The Acquisition of Syntax, Marc-Aurel Friedemann and Luigi Rizzi (eds.), 26–62. London: Longman. Engelberg, Stefan 1999 The magic of the moment – What it means to be a punctual verb. In Proceedings of the 25th Annual Meeting of the Berkeley Linguistics Society, Steve S. Chang et al. (eds.), 109–121. Berkeley: BLS. 2000 Verb meaning as event structure. In The Lexicon, Alan K. Melby and Arle R. Lommel (eds.), 257–268. Fullerton, CA: LACUS. 2002 Intransitive accomplishments and the lexicon: The role of implicit arguments, definiteness, and reflexivity in aspectual composition. Journal of Semantics 19: 369–416. 2004 Lexical event structures for verb semantics. Journal of Language and Linguistics 3: 62–108. Fabri, Ray 1996 The inverse morphology of Plains Cree. Yearbook of Morphology 1995: 17–41. Fabri, Ray, Martina Urbas and Markus Walther 1996 A computational model of Minimalist Morphology. Theoretical Linguistics 22: 231–279. Fischbach, Christian and James Kilbury 1999 Realisierung paradigmenbasierter Derivationsmorphologie in finitestate-Umgebungen. In Multilinguale Corpora: Codierung, Strukturierung, Analyse, Jost Gippert and Peter Olivier (eds.), 245–252. Prag: Enigma Corporation. Gamerschlag, Thomas 2002 Composition and argument structure formation of Japanese V-V compounds. In Japanese/Korean Linguistics 10, Noriko Akatsuka, Susan Strauss and Bernard Comrie (eds.), 533–545. Stanford: CSLI. Golston, Chris 1996a Direct optimality theory: Representation as pure markedness. Language 72: 713–748. 1996b Prosodic constraints on roots and words. In Interfaces in Phonology, Ursula Kleinhenz (ed.), 172–192. Berlin: Akademie Verlag. Golston, Chris and Richard Wiese 1996 Zero morphology and constraint interaction: subtraction and epenthesis in German dialects. Yearbook of Morphology 1995: 143–159. 1998 The structure of the German root. In Kehrein and Wiese (eds.), 165– 185. Graf, Dafna and Adam Ussishkin 2003 Emergent Iambs: Stress in Modern Hebrew. Lingua 113: 239–270.

Appendix 2: List of major publications from SFB 282 487 Grijzenhout, Janet 1998 The role of coronal specification in German and Dutch phonology and morphology. In Kehrein and Wiese (eds.), 27–50. 2001a Representing nasality in consonants. In Studies on Distinctive Feature Theory, T. Alan Hall (ed.), 177–210. Berlin /New York: Mouton de Gruyter. 2001b Devoicing and voicing assimilation in German, English and Dutch: a case of constraint interaction. In Constraints and Preferences, Katarzyna Dziubalska-Kolaczyk (ed.), 201–228. Berlin/New York: Mouton de Gruyter. 2002 Stress at the phonology-morphology interface. In Kaufmann and Stiebels (eds.), 15–45. Grijzenhout, Janet and Sandra Joppen-Helwig 2002 The lack of onsets in German child phonology. In The Process of Language Acquisition, Ingeborg Lasser (ed.), 319–339. Frankfurt am Main: Lang. Grijzenhout, Janet and Martin Krämer 2000 Final devoicing and voicing assimilation in Dutch derivation and cliticization. In Stiebels and Wunderlich (eds.), 55–82. Grijzenhout, Janet and Martina Penke 2005 On the interaction of phonology and morphology in language acquisition and German and Dutch Broca’s aphasia: the case of inflected verbs. Yearbook of Morphology 2004: 49–81. Jacobs, Joachim 1992 Bewegung als Valenzvererbung. Linguistische Berichte 138: 85–122. 1993 Integration. In Wortstellung und Informationsstruktur, Marga Reis (ed.), 63–116. Tübingen: Niemeyer. 1994 Das lexikalische Fundament der Unterscheidung von fakultativen und obligatorischen Ergänzungen. Zeitschrift für germanistische Linguistik 22: 284–319. 2003 Die Problematik der Valenzebenen. In Dependency and Valency. Vol. 1, Vilmos Ágel et al. (eds.), 378–399. Berlin/New York: Mouton de Gruyter. Janssen, Ulrike and Martina Penke 2000 The organization of agreement affixes in the mental lexicon: Evidence from German Broca’s aphasia. Brain and Language 74: 507– 509. 2002 How are inflectional affixes organized in the mental lexicon? Evidence from the investigation of agreement errors in agrammatic aphasics. Brain and Language 81: 180–191. Joppen, Sandra and Dieter Wunderlich 1995 Argument linking in Basque. Lingua 97: 123–169.

488 Appendix 2: List of major publications from SFB 282 Kaufmann, Ingrid 1995b What is an (im-)possible verb? Restrictions on Semantic Form and their consequences for argument structure. Folia Linguistica 29: 67– 103. 1995c O- and D-predicates: A semantic approach to the unaccusative-unergative distinction. Journal of Semantics 12, 377–427. 2002 Die Interaktion von Ereignis- und Argumentstruktur in Aktiv/Medium-Systemen am Beispiel des Fula. Linguistische Berichte 191: 299–243. 2003 Reflexive Verben im Deutschen. In Arbeiten zur Reflexivierung, Lutz Gunkel, Gereon Müller and Gisela Zifonun (eds.), 135–155. Tübingen: Niemeyer. Kaufmann, Ingrid and Dieter Wunderlich 1998 Cross-linguistic patterns of resultatives. Working Papers ‘Theorie des Lexikons’, 109. University of Düsseldorf. Kilbury, James 1992 Paradigm-based derivational morphology. In Proceedings of KONVENS 92, G. Görz (ed.), 159–168. Berlin: Springer. 1996 Top-down predictive linking and complex-feature-based formalisms. Proceedings of the 16th International Conference on Computational Linguistics, Vol. 2, 658–663. 2001 German noun inflection revisited. Journal of Linguistics 37: 339–353. Kilbury, James, Petra Barg [Naerger] and Ingrid Renz 1991 DATR as a lexical component for PATR. Proceedings of the 5th Conference of the EACL, 137–142. 1994 Simulation lexikalischen Erwerbs. In Kognitive Linguistik: Repräsentation und Prozesse, Sascha W. Felix et al. (eds.), 251–271. Opladen: Westdeutscher Verlag. Klein, Katarina and Silvia Kutscher 2002 Psych-verbs and lexical economy. Working Papers ‘Theorie des Lexikons’, 122. University of Düsseldorf. Krämer, Martin and Dieter Wunderlich 1999 Transitivity alternations in Yucatec, and the correlation between aspect and argument roles. Linguistics 37: 431–479. Krämer, Martin 1999 On the role of prosodic and morphological categories in vowel harmony and disharmony. A correspondence approach. In Proceedings of ConSOLE-7, T. Cambier-Langeveld et al., 183–199. Leiden: SOLE. 2000 Voicing alternations and underlying representations: The case of Breton. Lingua 110: 639–663. 2001a Yucatec Maya vowel alternations. Arguments for syntagmatic identity. Zeitschrift für Sprachwissenschaft 20: 175–217.

Appendix 2: List of major publications from SFB 282 489 2001b

On obstruent voicing in Breton, German, and Italian. In Proceedings of the 18th Scandinavian Conference of Linguistics. Vol. 1, Arthur Holmer, Jan Olof Svantesson and Åke Viberg (eds.), 39–55. Lund University. Lakämper, Renate and Dieter Wunderlich 1998 Person marking in Quechua – a constraint-based minimalist analysis. Lingua 105: 113–148. Lang, Ewald 1995 Das Spektrum der Antonymie. Semantische und konzeptionelle Strukturen im Lexikon und ihre Darstellung im Wörterbuch. In Die Ordnung der Wörter, Gisela Harras (ed.), 30–98. Berlin /New York: Mouton de Gruyter. 1996 Lexikalisierung und Wortfeldstruktur – typologisch betrachtet. In Deutsch – typologisch, Ewald Lang and Gisela Zifonun (eds.), 312– 355. Berlin /New York: Mouton de Gruyter. Latrouite, Anja 2001 Linking in Tagalog: Argument encoding determined by the semantic properties of arguments. Yearbook of Morphology 2000: 123–154. Lenz, Barbara 1998 Objektvariation bei Genitiv-Verben. Papiere zur Linguistik 58: 3–34. Löbel, Elisabeth 1998 Classifiers vs. genders and noun classes. A case study in Vietnamese. In Gender in Grammar and Cognition, Barbara Unterbeck (ed.), 245– 302. Berlin /New York: Mouton de Gruyter. Löbner, Sebastian 1999 Why German schon and noch are still duals: A reply to van der Auwera. Linguistics and Philosophy 22: 45–107. 2000 Polarity in natural language: predication, quantification and negation in particular and characterizing sentences. Linguistics and Philosophy 23: 213–308. Marcus, Gary, Ursula Brinkmann, Harald Clahsen, Richard Wiese and Steven Pinker 1995 German inflection: The exception that proves the rule. Cognitive Psychology 29: 189–256. Mattissen, Johanna 2002 Dependent-head synthesis in Nivkh – with an outlook on polysynthesis in the Far Northeast. In Evans and Sasse (eds.), 136–166. 2004a A structural typology of polysynthesis. Word 55: 189–216. 2004b The missing link between different types of polysynthetic languages. Chicago Linguistic Society 38, Vol. II: 386–399. 2004c Structural types of complex noun forms and their evolution. In Dimensionen und Kontinua: Beiträge zu Hansjakob Seilers Universalienforschung, Waldfried Premper (ed.), 35–56. Berlin: Akademie.

490 Appendix 2: List of major publications from SFB 282 Mori, Yoshiko, Sebastian Löbner and Katharina Micha 1992 Aspektuelle Verbklassen im Japanischen. Zeitschrift für Sprachwissenschaft 11: 189–215. Morimoto, Yukiko 2003 Markedness hierarchies and optimality in Bantu. In Perspectives in Linguistics: Papers in Honor of P. J. Mistry, Ritva Laury et al. (eds.), 261–280. New Delhi: Indian Institute of Language Studies. Naumann, Ralf 1996 Aspectual composition in dynamic logic. In Proceedings of the 10th Amsterdam Colloquium, Martin Stokhof et al. (eds.), 567–586. Amsterdam: ILLC. 1999 A dynamic approach to the present perfect in English, Theoretical Linguistics 24: 185–217. 2000 A dynamic temporal logic for aspectual phenomena in natural language. In Advances in Temporal Logic, Howard Barringer et al. (eds.), 223–254. Dordrecht: Kluwer. 2001 Aspects of changes – A dynamic event semantics. Journal of Semantics 18: 27–81. Naumann, Ralf and Chris Piñón 1997 Decomposing the progressive, Proceedings of the 11th Amsterdam Colloquium, ITLI, Amsterdam. Neef, Martin 1998a A case study in declarative morphology: German case inflection. In Kehrein and Wiese (eds.), 219–240. 1998b The reduced syllable plural in German. In Fabri, Ortmann and Parodi (eds.), 244–265. 1999 A declarative approach to conversion into verbs in German. Yearbook of Morphology 1998: 199–224. 2000 Phonologische, morphologische und syntaktische Konditionierung. In Morphologie, Geert Booij, Christian Lehmann and Joachim Mugdan (eds.), 463–484. Berlin /New York: Mouton de Gruyter. 2001 A non-universal approach to English inflectional morphology. Linguistics 39: 1149–1170. Ortmann, Albert 1999 Affix repetition and non-redundancy in Inflectional Morphology. Zeitschrift für Sprachwissenschaft 18: 76–120. 2000 Where plural refuses to agree: feature unification and morphological economy. Acta Linguistica Hungarica 47: 249–288. 2002 Economy-based splits, constraints and lexical representations. In Kaufmann and Stiebels (eds.), 147–177. 2004 A factorial typology of number marking in noun phrases: the tension of economy and faithfulness. In Explorations in Nominal Inflection, Gereon Müller, Lutz Gunkel and Gisela Zifonun (eds.), 229–267. Berlin /New York: Mouton de Gruyter.

Appendix 2: List of major publications from SFB 282 491 Penke, Martina, Helga Weyerts, Matthias Gross, Elke Zander, Thomas F. Münte and Harald Clahsen 1997 How the brain processes complex words: An Event-Related Potential study of German verb inflections. Cognitive Brain Research 6: 37–52. Penke, Martina, Marion Krause and Ulrike Janssen 1999 The Representation of inflectional morphology: Evidence from Broca’s aphasia. Brain and Language 68: 225–232. Penke, Martina and Marion Krause 1999 Broca's aphasia and German plural formation. Brain and Language 69: 311–313. 2002 German noun plurals – a challenge to the Dual-Mechanism Model. Brain and Language 81: 303–311. 2004 Regular and irregular inflectional morphology in German Williams syndrome. In Williams Syndrome across Languages, Susanne Niedeggen-Bartke and Julia Siegmüller (eds.), 245–270. Amsterdam: Benjamins. Penke, Martina, U. Janssen and Sonja Eisenbeiß 2004 Psycholinguistic evidence for the underspecification of morphosyntactic features. Brain and Language 90: 423–433. Piñón, Chris 2001 A finer look at the causative-inchoative alternation. In Proceedings of Semantics and Linguistic Theory XI, Rachel Hastings et al. (eds.). Ithaca, NY: CLC Publications,. to appear The existential tense in Hungarian. In The Nature of the Word: Essays in Honor of Paul Kiparsky, Kristin Hanson and Sharon Inkelas (eds.). MIT Press. Popescu, Alexandra 2000 The morphophonology of the Romanian clitic sequence. Lingua 110: 773–799. Pustet, Regina 1995 Obviation and subjectivization: The same phenomenon? A study of participant marking in Blackfoot. Studies in Language 19: 37–72. Raffelsiefen, Renate 1996 Gaps in word formation. In Interfaces in Phonology, Ursula Kleinhenz (ed.), 193–208. Berlin: Akademie Verlag. Rauh, Gisa 1993 Kasus und Präpositionen im Englischen. Indogermanische Forschungen 98: 252–292. 1995 Präpositionen und Rollen. Sprachwissenschaft 20: 123–167. Sasse, Hans-Jürgen 1993a Das Nomen – eine universale Kategorie? Sprachtypologie und Universalienforschung 46: 187–221.

492 Appendix 2: List of major publications from SFB 282 1993b

Syntactic categories and subcategories. In Syntax. An International Handbook of Contemporary Research, Joachim Jacobs, Arnim von Stechow and Theo Vennemann (eds.), 646–686. Berlin /New York: Mouton de Gruyter. 1999 Scales between Nouniness and Verbiness. In Language Typology and Language Universals, Martin Haspelmath et al. (eds.), 495–509. Berlin /New York: Mouton de Gruyter. 2002 Lexicological and lexicographic problems of word families in Cayuga. In Evans and Sasse (eds.), 203–221. Say, Tessa and Harald Clahsen 2002 Words, rules and stems in the Italian mental lexicon. In Storage and Computation in the Language Faculty, Sieb G. Nooteboom, Fred Weerman and Frank Wijnen (eds.) 93–129. Dordrecht: Kluwer. Sonnenstuhl, Ingrid, Sonja Eisenbeiß and Harald Clahsen 1999 Morphological priming in the German mental lexicon. Cognition 72: 203–236. Sonnenstuhl, Ingrid and Axel Huth 2001 Processing and representation of German -n plurals: A dual mechanism approach. Brain and Language 81: 276–290. Steins, Carsten 1998 Against arbitrary features in inflection. In Kehrein and Wiese (eds.), 241–265. 2000 How to account for non-concatenative phenomena in a morphemebased theory. In Stiebels and Wunderlich (eds.), 105–122. Stiebels, Barbara 1998 Complex denominal verbs in German and the morphology-semantics interface. Yearbook of Morphology 1997: 265–302. 1999 Noun-verb symmetries in Nahuatl nominalizations. Natural Language and Linguistic Theory 17: 783–836. 2000 Linker inventories, linking splits and lexical economy. In Stiebels and Wunderlich (eds.), 211–245. 2003 Transparent, restricted and opaque affix orders. In Syntactic Structures and Morphological Information, Uwe Junghanns and Luka Szucsich (eds.), 283–315. Berlin /New York: Mouton de Gruyter. Stiebels, Barbara and Dieter Wunderlich 1994 Morphology feeds syntax: the case of particle verbs. Linguistics 32: 913–968. 1999 Second stems in Hungarian nouns. The Linguistic Review 16: 253– 294. Vainikka, Anne and Martha Young-Scholten 1994 Direct access to X'-theory: Evidence from Korean and Turkish adults learning German. In Language Acquisition Studies in Generative Grammar, Teun Hoekstra and Bonnie D. Schwartz (ed.), 265–316. Amsterdam: Benjamins.

Appendix 2: List of major publications from SFB 282 493 Walther, Markus and Richard Wiese 1995 Deklarative versus prozedurale Modellierung von Konsonantenalternationen im Deutschen. Linguistische Berichte 157: 175–185. Walther, Markus 1998 Computing declarative prosodic morphology. In Proceedings of SIGPHON ’98, Workshop on the Computation of Phonological Constraints, T. Mark Ellison (ed.), 11–20. Montreal. Westermann, Gert, David Willshaw and Martina Penke 1999 A constructivist neural network model of German verb inflection in agrammatic aphasia. Proceedings of the 9th International Conference on Artificial Neural Networks, 916–921. Weyerts, Helga and Harald Clahsen 1994 Netzwerke und symbolische Regeln im Spracherwerb: Experimentelle Ergebnisse zur Entwicklung der Flexionsmorphologie. Linguistische Berichte 154: 430–460. Weyerts, Helga, Martina Penke, Ulrike Dohrn, Harald Clahsen and Thomas F. Münte 1997 Brain potentials indicate differences between regular and irregular German plurals. Neuroreport 8: 957–962. Wiese Richard 1996a Morphological vs. phonological rules: on German umlaut and ablaut. Journal of Linguistics 32: 113–135. 1996b Phrasal compounds and the theory of word syntax. Linguistic Inquiry 27: 183–193. 1996c Prosodic alternations in English morphophonology: a constraint-based account of morpheme integrity. In Current Trends in Phonology: Models and Methods, Vol. 2, Jacques Durand and Bernard Laks (eds.), 731–756. University of Salford: European Studies Research Institute. 1997 Underspecification and the description of Chinese vowels. In Studies in Chinese Phonology, Wang Jialing and Norval Smith (eds.), 219– 249. Berlin /New York: Mouton de Gruyter. 2001a The phonology of /r/. In Distinctive Feature Theory, T. Alan Hall (ed.), 335–368. Berlin /New York: Mouton de Gruyter. 2001b Regular morphology vs. prosodic 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. Wunderlich, Dieter 1994 Towards a lexicon-based theory of agreement. Theoretical Linguistics 20: 1–35. 1996a Lexical Categories. Theoretical Linguistics 22: 1–48. 1996b Models of lexical decomposition. In Lexical Structures and Language Use, Edda Weigand and Franz Hundsnurscher (eds.), 169–183. Tübingen: Niemeyer.

494 Appendix 2: List of major publications from SFB 282 1996c

A minimalist model of inflectional morphology. In The Role of Economy Principles in Linguistic Theory, Chris Wilder, Manfred Bierwisch and Hans-Martin Gärtner (eds.), 267–298. Berlin: Akademie Verlag. 1996d Minimalist morphology: the role of paradigms. Yearbook of Morphology 1995: 93–114. 1997a Cause and the structure of verbs. Linguistic Inquiry 28: 27–68. 1997b Argument extension by lexical adjunction. Journal of Semantics 14: 95–142. 2000 Predicate composition and argument extension as general options. In Stiebels and Wunderlich (eds.), 247–270. 2001a A correspondence-theoretic analysis of Dalabon transitive paradigms. Yearbook of Morphology 2000: 233–252. 2001b How gaps and substitutions can become optimal: the pronominal affix paradigms of Yimas. Transactions of the Philological Society 99: 315–366. 2002a Argument linking types – approached from the perspective of LDG. In Report of the Special Research-Project for the Typological Investigation of Languages and Cultures of the East and West 2001, Part II, H. Suzuki (ed.), 777–799. University of Tsukuba, Japan. 2002b On the nature of dative in Hungarian. In Papers from the Budapest Conference (Approaches to Hungarian 8), Istvan Kenesei and Peter Siptár (eds.), 163–184. Budapest: Akadémiai Kiadó. 2003 Optimal case patterns: German and Icelandic compared. In New Perspectives on Case Theory, Ellen Brandner and Heike Zinsmeister (eds.), 331–367. Stanford: CSLI. 2004 Is there any need for the concept of directional syncretism? In Explorations in Nominal Inflection, Gereon Müller, Lutz Gunkel and Gisela Zifonun (eds.), 373–395. Berlin/New York: Mouton de Gruyter. to appear The force of lexical case: German and Icelandic compared. In Kristin Hanson and Sharon Inkelas (eds.) The Nature of the Word: Essays in honor of Paul Kiparsky. MIT Press. Wunderlich, Dieter and Ray Fabri 1995 Minimalist Morphology: An approach to inflection. Zeitschrift für Sprachwissenschaft 14: 236–294. Wunderlich, Dieter and Renate Lakämper 2001 On the interaction of structural and semantic case. Lingua 111: 377– 418.

Subject index

adjective, declension (German), 376–377, 379 formation (German), 372–373 adverbials, ambiguites in, 247–248, 267 agent noun, 175–176 agent, prototypical, 256 agentivity, 69–72, 256, 267–270 agreement, 171, 189, 191–192 alienable, see possessor anticausative, 5, 73, 76 antipassive, 73, 173, 179–180, 184, 213–214 antipossessive, 180–182 applicative, 173, 175, 183, 184, 193 appropriateness conditions, 430, 437 argument extension, 173, 177, 183 argument hierarchy, 65, 120–122, 136, 152–156, 188 tests for, 67–68, 153, 156–157 argument linking, 168, 187, 236–239, 244–246, 253–262 types, 18 active type, 77–82, 129, 190–192, 204 generalized case type, 97–105, 188– 190 incorporation type, 128, 200–202 inverse type, 83–90, 130, 195–197 portmanteau type, 197–199 positional type, 94–97, 133–137 salience type, 90–91, 194–195 voice type, 90–94 argument reduction, 173, 183 roles 64–65 saturation, 210, 217 structure, 7, 19, 169, 170–171, 188 swapping 210, 218 transfer 219–222

argument, referential, 170, 171, 177, 188 structural, 122, 170 autonomy assumption, 390 Broca’s aphasia, 391, 393–395, 411–413 case, 189–190 case hierarchy, 142, 189, 255 case split, four-way, 100–101 case, abstract, 65, 105 accusative, 66–67, 78, 97, 105, 131– 132, 189, 203 ~, double, 132, 261–262 ~, sporadic, 87, 97 accusative-ergative (mixed) systems, 99–101 dative, 104–105, 106–109, 117, 140– 142, 189–190, 205, 215, 255–261 ergative, 66–67, 97, 105, 189–190, 203, 213, 259 generalized, 97–98, 105 genitive, 189, 203, 206, 207, 261–262 nominative, 189 proprietive, 139 structural, 142 case-pattern alternations, 100–101, 107– 108, 260–261 case-patterns of ditransitive verbs, 144, 146, 148 categorial restriction, 239, 248–249 causation, 238, 252 causative, 118, 122, 128, 141, 143, 173– 174, 183 child language data, 356–357, 409–410 cognitive moment, 277–278 concept lattice, 453–454, 459, 467 conceptual shift, 182 constraint ranking, 219

496 Subject index constraints for lexical decomposition, 120–121 constraints, relational (RCs), 442–444 contextual scales, 102 correspondence hypothesis, 20, 357– 360, 363, 371 cross-modal priming, see priming declarative linguistics, 48–50 deep structure, 36–38, 40–43 deictic system, 304–305 dependent-marking, 98–99 derivation (German), 369–373, 382 deverbal noun, 177 deverbalization hierarchy, 209 diathesis, 173 dictionary, 28–30 dictionary, extended, 41–42 direct/inverse marker, 85–86 ditransitive, see verbs double accusative, 132, 261–262 double object construction, 133–136, 150–157 double-(in)active, 81 double-give, 125 doubling, internal/external, 317–319 Dual-Mechanism Model, 390, 403, 408 Dynamic Event Semantics, 15 event nominal, 205–222 ontology, 262–263 presupposition, 240, 242, 248 semantics, 13, 15, 235–279 structure, 19, 235–279 trigger, neural, 272 event, control of, 246–247, 256–257, 269, 273 duration of, 275–279 granularity of, 262–263 iterative, 248–249 perception of, 264–265 punctual, 245, 248–249, 275–279 eventive roles, 64–65

event-related brain potentials (ERPs), 363–366, 395–399 evidential, 147 evolution of language, 17 Extended Standard Model, 42 feature co-occurrence restrictions (FCRs), 429–430, 456, 458, 462 feature logic, 438–441 feature specification, 457 features, +active, 78–80 +affected, 80 +control, 70–71 +hp, 85 +hr, 65, 104–105, 136, 188–189 +lr, 65, 104–105, 136, 142, 188 +ls, 195 +telic, 70–71 features, relational, 65, 105 field-affix, 291 focus, 223–225 Formal Concept Analysis (FCA), 431, 450–473 fractionation assumption, 391 frequency effect, 358, 361, 367–369, 372–373, 403–407, 411–413 functional application, 218 categories, 3–4 composition, 193, 217 generative grammar, models of, 36, 40, 42 glottochronology, 14 Government and Binding Theory, 42 harmonic alignment, 102–104 head-marking, 98–99 Horn theory, 455, 462 hyponomy, 250–251 inalienable, see possessor, 192–193, 204 incorporation, fossilized, 314, 322

Subject index

497

inflection, regular/irregular, 356–357, 359, 364, 367, 390–403 inflectional affixes, 375–379, 408–419 classes, 466–469 morphology, 20, 389–421 paradigm, 8, 362, 365–366, 370, 373–382 information structure, 223–225 inheritance hierarchy, 20, 430–432 interface systems, 12–13 invariance assumption, 396 irregularities, 32–33, 47, 49, 51

mental, 28, 31, 50, 389–421 structuralist, 32–33, 47, 51 linguistic relativity, 16 linker, default, 189 linking principle, universal, 258 linking split, 100–101, 192, 202–204 case-agreement, 103 linking, nominal, 167–226 subdomain, 211, 216–217 LOC, 113, 151–152 locative alternation, 114 locative incorporation, 127

lexeme, 28 lexical affix, 291 asymmetry of transitive verbs, 65, 87 borrowing, 10 categories, 3–4, 167 decision experiments, 358, 361, 363, 367–373, 399, 404 decomposition, 74, 77, 113, 115, 120–121, 238 Lexical Decomposition Grammar, 16, 18, 120–121, 169–170 lexical entries, 3–5, 29, 37–38, 43–44, 237, 241–243, 247, 249–252, 254, 257–260, 275 event structure, 19, 235–279 formative, 29, 37 information, 3, 9 insertion, 36–37, 42 integrity, 8 marking of verbs, 105–111, 255–262 morphology, 7–8, 40 phonology, 28, 44–46 lexicalist hypothesis, 38–41 lexicology, 28 lexicon simulation, 17 lexicon, architecture of the, 5–9 as list of exceptions, 47 concepts of the, 1–2, 27–51 generative, 39, 47, 51

markedness hierarchy, 102 meaning postulates, 242–243 Minimalist Morphology, 16, 390, 392, 403, 408, 412, 421 Minimalist Program, 12, 33, 46 modularity assumption, 390 morpheme, 32–33, 46 morphological complexity, 7–8 decomposition, 370–372, 399 processing, 355–383 morphology, 10, 35, 39–40, 46–49 morphology, modulatory, 12 morphology-syntax duality, 11 multifunctional morphemes, 17 nominalization, 19, 38–39, 172, 177, 184–187, 206–222 nominalization, ACC-POSS type, 207, 212, 214, 216 agent, 177–178, 186 event, 186, 205, 213 instrument, 177, 178, 186 locative, 177, 186 nominal type, 207–208, 211 POSS-OBL type, 208–209, 211, 213 verbal type, 208, 212 nonmonotonic inheritance, 431–432, 446–449, 472 non-root bound morpheme, 291

498 Subject index non-root bound morphemes, ontological domains, 297–333 animal, 332 artefact, 331 body part, 310–315 chronology, 328–329 classification, 305–306, 315–322 degree, 326–328 focus, 328–330 generic, 332–333 human, 331–332 instrumental, 312–313 location, 298–299, 301–309 manner, 323–326 medium, 322–323 motion, 309–310 natural force, 322–323 quantification, 329 noun incorporation, 125–128, 184, 193, 201–202, 292–294, 316–318, 321, 331–332, 340–34, 342 noun plural (German), 361, 364–365, 368–369, 372, 405–408 noun-phrase internal agreement (German), 416–417 noun-verb distinction, 57–58, 168, 288 object marking, differential, 101–103, 203 object, cognate, 73 direct/indirect, 129, 136 double, 133–136, 150–157 primary/secondary, 129, 136 objects, symmetric/asymmetric, 134– 135 obviative, 85, 87 one-prefix restriction, 91 optimality theory, 13, 48 overregularization, 357, 364–365, 409, 412 paradigm, inflectional, 8, 362, 365–366, 370, 373–382

paradigmatic specificity, 377 participle formation (German), 45, 358– 359, 364–365, 367–368, 372–373, 381, 392–403 passive, 75, 135, 145, 149, 173, 179, 182, 214–215 passive of intransitive verbs, 267–271, 274–275 person hierarchy, 81, 84, 87, 102, 195 phrasal coherence, 216 phrase structure rule, 35–36, 40 polymorphemic, 291 polypersonalism, 291 polysemy, 17 polysynthesis, 19, 287–343 polysynthesis, evolutionary paths, 319–321, 337, 342–343 polysynthesis, evolutionary types (burdock, onion, sandwich), 294– 296, 335–336, 340–341 polysynthesis, types of internal structure (scope-ordered, templatic), 293–294, 335–336, 339, 341 polysynthesis, word-formational types (affixal, compositional), 292, 334– 338, 342 portmanteau affix, 61–62, 197–200 POSS, 113, 151–152 possessor extension, 178–179, 183 raising 117, 140, 193, 225, 318 reduction, 180–182, 183 split, 192, 204 possessor, alienable/inalienable, 117, 138, 192–193, 204 postlexical phonology, 46, 48, 51 prepositional object construction, 134, 150–155 present participle, 215, 219 pre-syntactic, 39, 47, 51 priming experiments, 358–359, 361, 364, 366–367, 369, 371–373, 378– 380, 399–403 Principles and Parameters Theory, 42 processing models, 355–357

Subject index progressive, 248–249, 252 projection principle, 43–44 psycholinguistic fallacy, 30 QType, 433 quantifier floating, 89 reflexive, 205 reflexives, inherent, 110–111 relations, interlexematic, 239, 250–251 resultative, 122, 174 reversative morpheme, 318, 334, 336, 339 roles, macro-, 64–65 proto-, 64–65, 241, 256–262 thematic, 64–65 root vs. stem, 360–363, 365–367, 375, 381 root incorporation (see also noun incorp.), 292, 321 salience hierarchy, 102, 194–195, 196 relation 194–195, 199 selectional restriction, 239, 246–248 selective deficits, 390 semantic representation, 19 sentence matching experiments, 415–419 serial verb construction, 95–96, 121, 123–125, 155–157 serialization of verb roots, 292–294, 325–326, 339–340, 342 spatial orientation, 16 spatial orientation, individuocentric/ geocentric, 299, 301–309 Standard Model, 36, 41 stem, 361–364, 366, 368, 375, 381, 392, 399, 402–403, 407–408 structural linguistics, 31–33, 37 structure-building potential, 9 subcategorization, 43 subevent, 240–243, 250–252 subject, different, 82 surface structure, 36–38, 40–43 SVO, 95, 133

499

syncretism, dative-accusative, 144–150 telicity, 69–72 topic, 223–225 transformationalist hypothesis, 39 transitivity, semantic, 255–262 transparency assumption, 391 tripartite architecture of grammar, 10, 13 type constraints, 441–442 hierarchy, 434–436 signature, 20, 434, 460–461 unaccusatives, 70–71 underspecification of features, 4, 8, 413–415 unergatives, 70–71 ungrammaticality effect, 415–419 uniqueness, 218 unspecified object deletion, 180 valency alternation, 100–101, 107–108, 245, 260–261, 278–279 valency, marked, 105–111, 255–262 multidimensional theory of, 15, 253–254 verb root serialization, 292–294, 325– 326, 339–340, 342 verbs, ambitransitive = hybrid, 76 change of location, 114, 127 change of possession, 113, 127 decomposition of ditransitive, 113, 115 transitive, 74, 77 denominal, 116, 172 ditransitive, 18, 94, 96, 111–157 ~ canonical, 70, 118 ~ derived, 90, 117–120, 123, 143–144, 148–149 experiencer, 107, 110–111 give, 94, 96, 113, 119, 141–144 intransitive, 68–76 intransitive, passive of, 267–271, 274–275

500 Subject index verbs, lexically marked, 105–111, 255–262 light, 220–221 movement, 264–267 serial, 95–96, 121, 123–125, 155–157 suffixal, 11, 144 transitive animate, 84 canonical/prototypical, 63–64,72 impersonal, 110–111 noncanonical, 105–109

violation paradigm, 396 vocabulary, 28–29, 50 voice, 90–94, 173, 184–187, 202 volitionality, 267–275 weak cross-over, 88–89 word structure rule, 41–42 word-formational types, 292, 342 word-store, 29–30

Language index

Abkhaz (Northwest Caucasian), 209 Ainu (Paleosiberian), 327 Akan (Kwa), 155–156 Algonquian, 59, 83–90 Alutor (Chukotko-Kamchadal), 128 Arabic (Semitic), 216 Atayal, Mayrinax (Austronesian), 185–186 Atsugewi (Palaihnihan), 315 Awtuw (Papuan), 310 Bantu, 134–135 Basque [Isolate Europe], 66, 75–76, 99, 141, 208, 259 Caddo (Caddoan), 319, 320 Catalan (Romance), 365, 381 Cayuga (Iroquoian), 321 Cherokee (Iroquoian), 322 Chinese (Sino-Tibetan), 95–97, 99, 116, 125 Choctaw (Muskogean), 78, 108–109, 140, 314 Chukchi (Chukcho-Kamchadal), 292, 295, 317, 318, 326, 329 Cree, Plains (Algonquian), 83–89, 197 Creek (Muskogean), 170 Dakota (Siouan), 78–79 Dutch (Germanic), 268 Dyirbal (Pama-Nyungan), 100, 103, 139 Edo (Benue-Congo), 121, 124 English (Germanic), 11, 122, 133–134, 150–155, 249, 252, 457 Fongbe (Kwa), 157 Fox (Algonquian), 130 French (Romance), 10, 14, 250–251

Georgian (South Caucasian), 147–149 German (Germanic), 4–5, 17, 39, 45, 71–75, 106–107, 110–111, 115–117, 152, 175, 182, 205, 215, 240–243, 245, 247, 250, 254, 257, 261, 267– 271, 275, 279, 358–359, 361, 363– 365, 367–369, 371–377, 381, 383, 392, 397, 416, 451–452, 465–468 Greenlandic (Eskimo-Aleut), 287, 292, 295, 300, 338 Gunwinjgu (Non-Pama-Nyungan), 319, 320 Haida (Athapascan), 308, 312, 323, 325 Hindi (Indo-Aryan), 101, 103, 145–146, 203 Huastec (Mayan), 179, 181, 183, 199 Hungarian (Finno-Ugric), 61, 118, 220, 223, 260 Hupa (Athapascan), 307 Icelandic (Germanic), 261–262, 268 Ijo (Atlantic-Congo), 124 Indonesian (Austronesian), 118–119 Inuit (Eskimo-Aleut), 11, 143–144 Italian (Romance), 356 Itzaj Maya (Mayan), 204 Japanese, 5, 99, 141, 145, 220–221 Kayardild (Pama-Nyungan), 139 Ket (Yeniseian), 293, 299 Kham (Tibetan), 131 Khasi (Mon-Khmer), 131 K'iche' (Mayan), 174, 175, 180 Kikuyu (Bantu), 135 Kinyarwanda (Bantu), 193 Kiowa (Kiowa-Tanoan),61–62,198,310

502 Language index Klamath [Isolate North America], 295, 325 Koasati (Muskogean), 191–92 Korean, 17, 267 Kwakwala (Wakashan), 305, 323 Lakhota (Siouan), 69, 80–82, 129 Latin (Romance), 11, 59 Laz (Kartvelian), 299, 307 Maltese (Semitic), 5 Mam (Mayan), 176, 177, 179, 181, 223– 24 Miwok (Miwokan), 293 Mohawk (Iroquoian), 305, 318, 319, 320 Nahuatl, Classical (Uto-Aztecan), 171, 176, 178, 180, 184, 193 Nahuatl, Guerrero (Uto-Aztecan), 196 Navaho (Athapascan), 293 Nez Perce (Sahaptian), 324, 328 Nivkh (Paleosiberian), 127, 201–202, 308, 329–330 Oaxaca Chontal (Hokan), 125 Ojibwe (Algonquian), 84, 130–131 Paiwan (Austronesian), 187 polysynthetic languages, sample, 288– 290 polysynthetic languages, tables, 294, 302–304, 334, 336, 339–340 Pomo, Central (Hokan), 81 Potawatomi (Algonquian), 195–96 Quechua, 61 Russian (Slavic), 151 Sesotho (Bantu), 134–135 Shuswap (Salishan), 208 Slave (Athapascan), 313

Spanish (Romance), 17, 267, 361–363, 366, 381 Spokane (Salishan), 311, 315, 331–332 Swedish (Germanic), 268 Tagalog (Philippine), 93–94 Tahitian (Polynesian), 133 Takelma [Isolate North America], 298, 302, 307, 312 Tamazight (Berber), 138 Tarasco [Isolate Mexico], 305 Tewa, Arizona (Kiowa-Tanoan), 90–92, 126 Tiwa, Southern (Kiowa-Tanoan), 317, 319 Tiwi (Non-Pama-Nyungan), 314, 331 Tongan (Polynesian), 138, 260 Tonkawa [Isolate North America], 324 Tsimshian (Tsimshianic), 307, 308 Tukang Besi (Austronesian), 172, 207 Tunica (Gulf), 327 Turkish (Turkic), 9, 207, 214 Tyvan (Turkic), 203 Tzotzil (Mayan), 119–120 Udi (Northeast Caucasian), 100, 103 Warlpiri (Pama-Nyungan), 103 Washo [Isolate North America], 298, 314 West Greenlandic (Eskimo-Aleut), 213 Wichita (Caddoan), 294, 302, 322, 327 Yana (Hokan), 319 Yaqui (Uto-Aztecan), 132 Yidiny (Pama-Nyungan), 17 Yimas (Papuan), 99, 141, 296, 328 Yindjibarndi (Pama-Nyungan), 132 Yoruba (Benue-Congo), 123 Yucatec (Mayan), 68–69, 126, 184, 316, 318 Yupik (Eskimo-Aleut), 190, 198