Typology of Verbal Categories: Papers Presented to Vladimir Nedjalkov on the Occasion of his 70th Birthday [Reprint 2013 ed.] 9783110913750, 9783484303829

Citation preview

Linguistische Arbeiten

382

Herausgegeben von Hans Altmann, Peter Blumenthal, Herbert E. Brekle, Gerhard Heibig, Hans Jürgen Heringer, Heinz Vater und Richard Wiese

Typology of Verbal Categories Papers presented to Vladimir Nedjalkov on the occasion of his 70th birthday

Edited by Leonid Kulikov and Heinz Vater

Max Niemeyer Verlag Tübingen 1998

Die Deutsche Bibliothek - CIP-Èinheitsaufnahme Topology of verbal categories : papers presented to Vladimir Nedjalkov on the occasion of his 70th birthday / ed. by Leonid Kulikov and Heinz Vater. - Tübingen : Niemeyer, 1998 (Linguistische Arbeiten ; 382) ISBN 3-484-30382-4

ISSN 0344-6727

© Max Niemeyer Verlag GmbH & Co. KG, Tübingen 1998 Das Werk einschließlich aller seiner Teile ist urheberrechtlich geschützt. Jede Verwertung außerhalb der engen Grenzen des Urheberrechtsgesetzes ist ohne Zustimmung des Verlages unzulässig und strafbar. Das gilt insbesondere für Vervielfältigungen, Übersetzungen, Mikroverfilmungen und die Einspeicherung und Verarbeitung in elektronischen Systemen. Printed in Germany. Gedruckt auf alterungsbeständigem Papier. Druck: Weihert-Druck GmbH, Darmstadt Buchbinder: Industriebuchbinderei Hugo Nadele, Nehren

Contents Leonid Kulikov / Heinz Vater Introduction

1

I. ERGATIVITY AND TRANSITIVITY

Jean-Pierre Desclés / Zlatka Guentchéva Causalité, Causativité, Transitivité

7

YakovG. Testelec On Two Parameters of Transitivity

29

Alexandra Y. Aikhenvald Transitivity Increasing Operations in Tariana

47

Casper de Groot How to Butter a Sandwich. On 'spread' in Dutch and Hungarian from a Typological Perspective

61

Werner Drossard Labile Konstruktionen

73

Theodora Bynon Inverse Direction and Second Person in Dumi

85

Konstantin Kazenin On Patient Demotion in Lak

95

Π. VOICE, CAUSATIVE AND VALENCY

Masayoshi Shibatani Voice Parameters

117

Leonid I. Kulikov Passive, Anticausative and Classification of Verbs: the Case ofVedic

139

Heinrich Werner Zur Typologie der Passivkonstruktionen in den Jenissej-Sprachen

155

Juan C. Moreno Cabrera Allocutivity and Voice in the Basque Verb

169

vi

Contents

Alexander Κ. Ogloblin From Inert to Actional Causative

179

Jurij P. Knjazev Towards a Typology of Grammatical Polysemy: Reflexive Markers as Markers of Reciprocity

185

Simon van de Kerke Verb Formation in Leko: Causatives, Reflexives and Reciprocals

195

ΙΠ. TENSE AND MOOD

Jarmila Panevová / Petr Sgall Verbal Categories, Meaning and Typology

205

Wolfgang Klein / Heinz Vater The Perfect in English and German

215

Gilbert Lazard L'expression de l'irréel: essai de typologie

237

Elena Kordi Sur le paradigme complet de l'impératif

français

249

IV. VERBAL CATEGORIES AND LANGUAGE UNIVERSALS

Walter Bisang Verb Serialization and Attractor Positions: Constructions and Their Potential Impact on Language Change and Language Contact

255

Johan van der Auwera Defining Converbs

273

Bernard Comrie Translatability and Universale Revisited

283

Ronald Lötzsch Die Verneinung als funktional-semantische Kategorie (unter besonderer Berücksichtigung von Streckformen des Deutschen und Russischen)

287

List of Contributors

309

Introduction Leonid Kulikov / Heinz Vater

By the end of the 20th century, typological studies have achieved remarkable results. We dispose today of ample data on the structures of language categories and meanings, on the range of their variability and structural relations between them. Of course, a lot of work is still to be done and hundreds of languages wait for their researchers, but the perspectives of research are much clearer now than a few decades before. One of the central areas of research concerns the study of verbal morphology, the verb being the organizing centre of the clause, and, basically, the main actor in syntactic processes. Since the 60ies, typology of verbal categories has been one of the most successfully developing fields of linguistic investigations. Although the history of linguistic typology can be traced back to the beginning of 19th century and even earlier, its rapid development began in the mid of our century. The Eastern tradition can be associated, above all, with the school of grammatical semantics represented by a group of Moscow linguists (Ju.D. Apresjan, I.A. Mel'cuk, A.A. Zaliznjak, and others), and the famous Leningrad (now: St. Petersburg) typological school, founded by Alexandr Xolodovic and his younger colleagues Vladimir Nedjalkov, Viktor Xrakovskij and others.1 The interest for empirical data and the attempt to obtain as exhaustive data as possible, which is specific of the Lenigrad typological school, enables us to label it "taxonomy-oriented" as compared with the tradition coined in the USA and in West Europe. A deeper and more systematic knowledge of the achievements of the Leningrad typological school (LTS) has become possible for the first time by the anthology "Typology of Resultative Constructions" (1988) being an extended and enlarged English version of the original volume published in Russian under the guidance of Vladimir Nedjalkov and prepared for Western readers in cooperation with Bernard Comrie. This volume also marks the beginning of a successful cooperation between Western and Eastern traditions. The present volume unifies articles written by linguists belonging to both Western and Eastern traditions. It can be regarded as a further step to bridge the gaps between two traditions and to accumulate their efforts. It is therefore not accidental that this book is dedicated to Vladimir Nedjalkov, the person who contributed so much to the development of typological studies in and outside Russia. As one of the most brilliant typologists belonging to the Lenigrad school, Nedjalkov has worked on a number of popular issues related to verbal categories. He was one of the most active contributors to the Typology of causative constructions (XOLODOVIc 1969), the first collective monograph prepared by the LTS under the guidance of Xolodovic, which has remained until now, 30 years later, one of the best typological studies on verbal categories and a

1

A valuable sketch of the main achievements of the Leningrad Typological group written by Vladimir Nedjalkov and his colleague Viktor Litvinov was published in SHIBATANI & BYNON (1995).

2

Leonid Kulikov / Heinz Vater

classical specimen of the "Leningrad way" of linguistic description (c£, above all, NEDJALKOV / SIL'NICKJJ 1969a; 1969b). Nedjalkov's interest for causatives did not confine to typological studies (c£ also NEDJALKOV 1964; 1967; KULIKOV / NEDJALKOV 1992): w e o w e him a de-

tailed sketch of German causative constructions (NEDJALKOV 1971, German translation 1976), which belongs now to the classical studies on the German verb. Nedjalkov's interest for causatives continues in his studies on other valency-changing categories, such as passives (cf., in particular, a series of Nedjalkov's articles dealing with the relationships between resultatives and passives: NEDJALKOV 1980b; 1981a), reflexives (NEDJALKOV 1975; 1978; 1980a; NEDJALKOV / OTAJNA 1981; GENIUSENE / NEDJALKOV 1991, reciprocals (NEDJALKOV 1991).

In the before-mentioned study on resultatives, to which Nedjalkov contributed most actively both as editor and as (co)author of a few articles, his major achievement consisted in establishing the resukative as a distinct category, with specific morphological and syntactic features that set it off from such related categories as perfect and stative. The above-mentioned verbal categories are related, more or less directly, to the domain of verbal valency and argument structure, and this is, indeed, one of the main interests of Nedjalkov in typology. However, his studies on verbal categories are not confined to syntactic issues as voice and (in)transitivity but also encompass such semantic categories as tense and aspect, cf. his study on TAM-forms in Chukchee (NEDJALKOV 1993) and Evenki (NEDJALKOV, I.V. / NEDJALKOV, VP. 1988), as well as his sketches on mood (NEDJALKOV 1988a; 1990a). The meaning of perfect and perfective forms are discussed by Nedjalkov mostly in connection with the typology of resultative constructions (cf. NEDJALKOV 1980b; 1983/1988b). In the domain of aktionsarten, Nedjalkov has paid most attention to the meaning of inceptivity, which has resulted in a series of articles on typology of inchoative and inceptive verbs (NEDJALKOV 1984; 1986; 1987). We also owe Nedjalkov a comprehensive sketch of the typology of converbs (i.e. verbal forms used in adverbial function, Russ. deepricastie), which can serve as a good starting point for typological researches on the topic (NEDJALKOV 1990b, Engl. transL 1995). Finally, a number of articles deal with those aspects of the clause structure which are related to the verb and its functioning: nominalizations (NEDJALKOV 1979b; 1981b; 1981c) and ergativity (iNèNLlKèJ / NEDJALKOV 1967; NEDJALKOV 1979a; 1982). The years of the Iron Curtain and isolation of the Eastern (Soviet) linguistics were not easy for many Russian scientists. Many eminent linguists emigrated to the United States or to Western Europe. Nedjalkov remained and continued linguistic research, contributing much to the development of the Eastern tradition. Moreover, he is been a brillant organizer of scholarly activity. Due to his indomitable temperament, Nedjalkov has created around himself some kind of a "magnetic-linguistic" field and engaged in this atmosphere more and more most talented young researchers, helping them to develop their linguistic skills and to find an appropriate position within the scientific infrastructure. The contributions to this volume are oriented towards the areas of interests of Vladimir P. Nedjalkov, summarized here under the heading "typology of verbal categories"; they can, thus, give a representative view of current typological studies on verbal categories. Since the study of voice and valency (especially concerning problems of causatives and passives) has remained

Introduction

3

until now one of NedjalkoVs main interests (cf. his recent studies on reciprocal constructions), it is not accidental that more than half of the contributions to the volume relate to this subject. Part I, "Ergativity and transitivity", includes articles dealing with problems of transitivity and valency, subject-object relations and clause structure in general. J.-P. DESCLÉS and Z. GUENTCHÉVA ("Causalité, causativité, transitivité") discuss the relationships between transitivity, causativity and causation. They distinguish between causativity and causation (causality) demonstrating that the concept of causativity is rather intricate, covering two related but generally independent domains, transitivity and factitivity/permissivity. Y.G. TESTELEC ("On two parameters of transitivity") investigates the correlation between the approach to transitivity as a set of several interrelated parameters exemplified by HOPPER/THOMPSON (1980) and the more traditional view on transitivity as a feature of grammatical classification of verbs. The author shows that the two parameters are indispensable for a definition of transitivity as control and affectedness of the participants involved in the situation. A.Y. AIKHENVALD ("Transitivity increasing operations in Tariana") describes different types of causative constructions in Tariana (a North Arawak language from North West Amazonia) showing how manifold the distinctions between various formal types of causatives within a language can be. C. DE GROOT ("How to butter a sandwich. On 'spread' in Dutch and Hungarian from a typological perspective") discusses the opposition between two syntactic patterns in constructions with two objects in English, Dutch and Hungarian and demonstrates how languages differ in the application of these patterns. The author focuses on two parameters shown to be relevant for the opposition between these patterns. Three articles focus on ergativity and related problems. W. DROSSARD ("Labile Konstruktionen") discusses the behavior of "labile verbs" (i.e. verbs which can be employed in intransitive and transitive usage without any formal marker of the valency change, cf. Engl, break) in accusative and ergative languages. Special attention is paid to the relationship between lability and the feature "+/- oriented" (cf. patient-oriented break vs. agent-oriented eat), particularly in nominalized constructions. T. BYNON ("Inverse direction and second person in Dumi") demonstrates how both ergative and accusative patterns coexist in Dumi (the Kiranti subgroup of the Tibeto-Burman language family), which is consistently ergative, whereas verbal cross-referencing should be interpreted as accusative. The article focuses on the function of the prefix a-, which is shown to carry out two functions: marking of the so-called inverse voice and of second person agency. Κ KAZENIN ("On patient demotion in Lak") focuses on a specific subtype of antipassive constructions, the so-called biabsohitive constructions (the agent changes case-marking from the ergative to the absolutive, whereas the patient remains in the absolutive). The author describes this transformation attested in a Daghestanian language (Lak), and shows that it should be treated as a syntactic restructuring, which splits one clause into two, rather than as a canonical voice alternation. Part II, "Voice, causative, and valency", includes articles dealing with one of Nedjalkov's favourite topics, viz. voice and valency derivation. M. SHIBATANI ("Voice parameters") develops the semantic treatment of voice, which was neglected during the last decades in comparison to the predominantly syntactic studies on voice and valency. He shows which components of verbal meaning are especially sensitive to voice variation. In particular, the semantic value of

4

Leonid Kulikov / Heinz Vater

passive derivation is described in terms of the maximization of the contrast between two participants. L.I. KULIKOV ("Passive, anticausative and classification of verbs: the case of Vedic") focuses on the difference between two intransitivizing derivations: passivization and anticausativization. The latter has a more restricted scope of applicability than passivization, which can apply to practically all transitive verbs. The meaning of the entropy increase is shown to be one of the most relevant parameters which make anticausative derivation possible. H. WERNER ("Zur Typologie der Passivkonstruktionen in den Jenissej-Sprachen") describes the Ket forms which are traditionally treated as passives and shows that their actual status is rather far from the canonical passive. J.C. MORENO CABRERA ("Allocutivity and voice in the Basque verb") provides evidence from Basque to demonstrate that diathesis and voice, as determined in works of the Leningrad typological school, are independent concepts, and that, moreover, a situation is possible where voice variation does not presuppose changes in diathesis. A.K. OGLOBLIN ("From inert to actional causative") provides evidence from Austronesian languages to show how the causative meaning can arise and develop in non-causative morphemes. Ju.P. KNJAZEV'S article "Towards a typology of grammatical polysemy: reflexive markers as markers of reciprocity" is a typological sketch of polysemous morphological markers which can be used to express both reflexive and reciprocal meanings. The classification is based on a large language sample. S. VAN DE KERKE ("Verb formation in Leko: causatives, reflexives and reciprocals") draws attention to typologically most interesting peculiarities of valency changing in Leko, an almost unknown Amerindian language spoken in Bolivia. Part m , "Tense and mood", is opened with the contribution by J. PANEVOVÁ and P. "Verbal categories, meaning and typology". Using the Reichenbach framework - with the meaning of tenses derived from the relations between point of event (E), point of speech (S), and point of reference (S) - they postulate a "recursive principle of tense values" specifying the values of R for the verbs in different kinds of clauses in Czech. The two aspects "Complex" and "Processual" are in close interaction with tenses and aktionsarten. W. KLEIN and H. VATER investigate "The Perfect in English and German". Basing their study on the notion of "topic time" (TT) developed in KLEIN 1994, they show that the Engl, "present perfect" and Gm "Perfekt" both are complex, being composed of temporal and aspectual components; on the other hand, there are basic differences: whereas the Engl, present perfect is characterized by the relation "TT includes TU (time of utterance)", Gm Perfekt has to be interpreted as "TT shiftable with respect to TU". The contribution by G. LAZARD is called "L'expression de l'irréel: essai de typologie". The author investigates the expression of irreality in the past (cf. Si tu m'avais averti de la réunion d'hier, je serais venu) in about thirty languages. He comes to the conclusion that the preferred means of expressing past irreality in the apodosis of a sentence consists in using a verb form combining a morpheme of virtuality with a morpheme of past tense. In her article "Sur le paradigme complet de l'impératif français", E. KORDI, basing her description on functional grammar and general typology, shows that French (like several other languages) has a complete imperative paradigm including forms of the first person singular (que je parle) and the third person singular and plural (que NparlefntJ). SGALL

Part IV is devoted to "Verbal categories and language universals". W. BISANG investigates "Verb serialization and attractor positions: Constructions and their potential impact on

5

Introduction

language change and language contact". After investigating several languages and language groups, he comes to the conclusion that constructions with attractor positions can be a good explanation for abrupt processes of grammaticalization. J. VAN DER AUWERA, in "Defining converbs" in several languages, makes a difference between "converb" in a wide sense and a narrow sense. In both cases, the verb form is [+dependent, -argumentai, -adnominal]; converbs in a narrow sense are additionally [+embedded, -finite]. In "Translatability and universale revisited", B. COMRIE discusses the famous passage of Homer's Odyssey, where Polyphemus answers the other giants' question as to whether someone is killing him "Outis me kteinei..."; the ambiguity of outis (as a pronoun and a name) is preserved in English ("Nobody is killing me"), but not in Russian, where the pronoun nikto, different from the name, is used with double negation ("nobody is not killing me"). He concludes that translatability is not a suitable language universal. R LÖTZSCH shows in "Die Verneinung als funktional-semantiche Kategorie (unter besonderer Berücksichtigung von Streckformen des Deutschen und Russischen)" that sentence negation (like other verbal categories) can be expressed cross-linguistically and within a language by quite different morphological, lexical and lexico-syntactic means. Apart from the authors, a couple of people contributed to the completion of the anthology. We are especially grateful to Anja Werner and Christoph Β enden, who did an excellent job in formatting and producing the final version of this volume.

Leonid Kulikov, Leiden University

Heinz Vater, University of Cologne August, 1997

References Geniuïene, É.S., and V.P. Nedjalkov (1991) Tipologija refleksivnyx konstrukcij. In: A.V.Bondarko et al. (eds.) Teorijafimkcional'noj grommatiti. Personal'nost. Zalogovost'. St.-Peterburg: Nauka, 241-276. Inènlikèj, P.I., and V.P. Nedjalkov (1967) Iz nabljudenij nad èrgativnoj konstrukciej ν íukotskom jazyke. In: V.M. 2irmunskij et al. (eds.) Êrgativnaja konstrukcija predloíenija ν jazykax razliânyx tipov (Issledovanija i materialy). Leningrad: Nauka, 246-260. Kiefer, F. (ed.) (1973) Trends in Soviet theoretical linguistics. Dordrecht: Reidel. Kulikov, L.I., and V.P. Nedjalkov (1992) Questionnaire zur Kausativierung, Zeitschrift fir Phonetik, Sprachwissenschaft und Kommunikationsforschung 45 (2), 137-149. Litvinov, V.P., and V.P. Nedjalkov (1995) The St Petersburg/Leningrad typology group. In: Shibatani and Bynon (1995), 215-271. Nedjalkov, I.V., and V.P. Nedjalkov (1988) Meanings of tense forms in Evenki (Tungus), Lingua Posnaniensis, XXXI, 87-100. Nedjalkov, V.P. (1964) O syjazi kauzativnosti i passivnosti, Uâenye zapiski Baskirskogo universiteta XXI. Serija ftlologiöeskix nauk 9(13), 301-310. - (1967) Napravlenie derivacii i smysl, In: L.I. BazileviC et al. (eds.) Urovni jazyka i ix vzaimodejstvie. Moskva: Nauka, 112-115.

6 -

Leonid Kulikov / Heinz Vater

(1971) Kausativnye konstrukcii ν nemeckom jazyke: Analitiöeskij kauzativ. Leningrad: Nauka. (German transi.: Kausativkonstruktionen. Aus dem Russischen übersetzt von V. Kuchler und H. Vater. Tübingen: Narr, 1976 (Studien zur deutschen Grammatik, 4)). - (1975) Tipologija recessivnyx konstnikcij. Refleksivnye konstrukcii. In: Diatezy i zalogi. Leningrad: LO Institutajazykoznanija, 21-33. - (1978) Zametki po tipologii refleksivnyx deagentivnyx konstrukcij (opyt isòislenija). In: V S. Xrakovskij (ed.) Problemy teorii grammatiâeskogo zaloga. Leningrad: Nauka, 28-37. - (1979a) Degrees of Ergativity in Chukchee. In: F. Plank (ed.) Ergativity: towards a theory of grammatical relations. London etc.: Academic Press, 241-262. - (1979b) Zametki po tipologii dvupredikatnyx konstrukcij. Opyt isCislenija In: I.P. Susov et al. (eds.) Znaâenie i smysl reâevyx obrazovanij. Kalinin: Kalininskij gos. universitet, 35-47. - (1980a) Reflexive constructions: a functional typology. In: G. Brettschneider and Ch. Lehmann (eds.) Wege zur Universalienforschung. Festschrift fir H. Seiler. Tübingen: Nan, 222-228. - (1980b) Zametki po tipologii rezul'tativnyx konstrukcij (perfektiv, rezul'tativ, perfekt, passiv). In: I.P. Susov et al. (eds.) Kommunikativno-pragmatiíeskie i semantiöeskie funkcii reievyx edinstv. Kalinin: Kalininskij gos. universitet, 143-151. - (1981a) Κ tipologii sootnoSenija rezul'tativa i passiva: Na materiale nemeckogo jazyka. In: I.P. Susov et al. (eds.) Semantika i pragmatika sintaksiäeskix edinstv. Kalinin: Kalininskij gos. universitet, 27-40. - (1981b) O tipologii konstrukcij s predikatnymi aktantami. In: Semantika i sintaksis konstrukcij s predikatnymi aktantami. Leningrad: LO Institutajazykoznanija, 24-40. - (1981c) Tipologija dvupredikatnyx konstrukcij (isCislenie i materialy dlja ankety), Studia gramatyczne IV. (Prace institutu jçzyka polskiego, 38). Wroclaw etc.: Wydawnictwo Polskiej Akademii nauk, 83-106. - (1982) Êrgativnost' i nominativnost' ν tukotskom glagol'nom slovoobrazovanii. In: V.M. Solncev et al. (eds.) Teoretiöeskie problemy vostoënogo jazykoznanija, 2. Moskva: Nauka, 99-106 - (ed.) (1983) Tipologija rezul'tativnyx konstrukcij: rezul'tativ, Stativ, passiv, perfekt. Leningrad: Nauka. - (1984) Zametki po tipologii naCinatel'nyx konstrukcij. In: I.P. Susov et al. (eds.) Pragmatika i semantika sintaksiöeskix edinic. Kalinin: Kalininskij gos. universitet, 46-54. - (1986) Osnovnye tipy naCinatel'nyx glagolov: inxoativy, ingressivy, inceptivy. In: I.P. Susov et al. (eds.) Jazykovoe obSöenie i ego edinicy. Kalinin: Kalininskij gos. universitet, 124-134. - (1987) Naíinatel'nost' i sredstva ee vyraienija ν jazykax razliCnyx tipov. In: A.V. Bondarko et al. (eds.) Teorijafunkcional'nojgrammatiki. Personal'nost. Zalogovost'. Leningrad: Nauka, 180-195. - (1988a) Imperativ ν íukotskom jazyke. In: L. A. Biijulin et al. (eds.) Imperativ ν raznostruktumyx jazykax. Leningrad: LO Institutajazykoznanija, 89-91. - (ed.) (1988b) Typology of resultative constructions. Engl, transi, ed. by Β. Comrie. Amsterdam etc.: Benjamins. (Typological Studies in Language, 12). - (1990a) Kondicionalis ν íukotskom jazyke. In: L.A. Biijulin et al. (eds.) Funkcional'nye, tipologiceskie i lingvodidaktiâeskie aspekty issledovanija modal'nosti. Irkutsk: LO Institute jazykoznanija, Irkutskij pedagogiCeskij institut inostrannyx jazykov, 75-76. - (1990b) Osnovnye tipy deepriCastij. In: V.S. Xrakovskij et al. (eds.) Tipologija i grammatika. Moskva: Nauka, 36-59. (Engl, transi.: Some typological parameters of converbs. In: M. Haspelmath and E. König (eds.) Converbs in cross-linguistic perspective. Berlin etc.: Mouton, 1995, 97-136). - (1991) Tipologija vzaimnyx konstrukcij. In: A.V. Bondarko et al. (eds.) Teorija funkcional'noj grammatiki. Personal'nost. Zalogovost'. St.-Peterburg: Nauka, 276-312. - (1993) Tense-Aspect-Mood forms in Chukchi, Eurotyp Working Papers. Series VI, 4, 1-99. Nedjalkov, V.P., and G.A. Otaina (1981) Nivxskie refleksivnye glagoly i tipologija smyslovyx refleksivov. In: V.S. Xrakovskij (ed.) Zalogovye konstrukcii ν raznostruktumyx jazykax. Leningrad: Nauka, 185-220. Nedjalkov, V.P., and G.G. Sil'nickij (1969a) Tipologija kauzativnyx konstrukcij. In: XolodoviC (1969), 5-19. (Germ, transi.: Typologie der kausativen Konstruktionen, Folia Linguistica, VI (1973), 273-290). Nedjalkov, V.P., and G.G. Sil'nickij (1969b) Tipologija morfologiíeskogo i leksiCeskogo kauzativov. In: Xolodovi£ (1969), 20-50. (Engl, transi.: The Typology of Morphological and Lexical Causatives. In: Kiefer (1973), 1-32). Shibatani, M., andT. Bynon (1995) (eds.) Approaches to language typology. Oxford: Clarendon Press. XolodoviC, A.A. (ed.) (1969) Tipologija kauzativnyx konstrukcij: Morfologiâeskij kauzativ. Leningrad: Nauka.

Causalité, Causativité, Transitivité* Jean-Pierre Desclés / Zlatka Guentchéva

1. Introduction Avec ses deux articles fondamentaux et très riches en données de V.P. Nedjalkov et G.G. SiMckij ( 1969a,b), la monographie collective Tipologija kauzativnyx konstrukcij (Xolodovic (ed.) 1969) a inspiré, directement ou indirectement, de nombreuses études consacrées aux constructions qualifiées de "causatives". Nous disposons actuellement d'une abondante littérature sur le sujet dans des langues très diverses, mais les solutions avancées (syntaxiques ou sémantiques) ont conduit à des controverses qui montrent l'importance du problème tant du point de vue théorique que descriptif et typologique. Ainsi, par exemple, la décomposition lexicale d'un verbe comme "to kill" en "CAUSE (BECOME not alive)" a des implications théoriques1 non négligeables. En effet, dans les études typologiques, un certain nombre de marqueurs (des affixes verbaux) dans des langues aussi diverses que le turc, le hongrois, le géorgien, le japonais, le coréen, l'indonésien... sont interprétés implicitement du moins comme des traces morphologiques d'un opérateur abstrait, souvent noté CAUSE. Or, dans certains exemples, il nous semble que ces marqueurs sont plutôt des marqueurs de transitivisation (sémantique et, par extension, syntaxique, dans un sens précis qui sera rappelé plus loin). Quelle doit être l'interprétation qu'il convient de donner à cet opérateur abstrait CAUSE ? L'analyse des exemples fait apparaître clairement qu'il serait souhaitable de bien distinguer la notion de causalité de celles de causativité et de transitivité. Il ne nous semble pas qu'il y ait un opérateur CAUSE unique, avec une sémantique bien définie, qui serait constitutif à la fois des constructions qui expriment la causalité, la causativité et la transitivité. Dans cet article, nous n'apportons pas de nouvelles données, mais nous nous proposons de montrer que ces trois notions doivent être soigneusement définies. Ce que l'on appelle causativité ne s'identifie ni à la causalité qui met en relation deux situations (Nedjalkov et SiMckij 1969a), ni à l'agentivité qui, étant constitutive de la transitivité sémantique (Hopper & Thompson 1980), implique une certaine capacité de contrôle et d'effectuation d'un processus (Desclés 1990, 1997), de façon directe ou médiate. Plusieurs niveaux d'analyse sont nécessaires pour étudier des phénomènes tels les diathèses. • faut donc distinguer le niveau d'analyse conceptuelle ou cognitive du niveau des organisations linguistiques variables de langue à langue. Au niveau conceptuel, nous avons des

*

ι

Que L. Kulikov et H. Vater trouvent ici l'expression de nos remerciements pour la lecture attentive de cet article, pour leurs remarques et suggestions stimulantes. Voir à ce propos entre autres McCawley (1976), Shibatani (1976), Talmy (1976), Comrie (1976, 1981).

Jean-Pierre Desclés / Zlatka Guentchéva

8

représentations en termes de schèmes2 et de primitives. Ces schèmes sont encodes dans une langue par des configurations morphosyntaxiques spécifiques, ces dernières étant l'objet de découpages en morphèmes dont l'identification n'est pas toujours évidente. On doit noter qu'il n'y a pas de relations biunivoques entre les morphèmes et les opérateurs dont les morphèmes sont la trace puisque, dans un grand nombre de cas, un morphème grammatical est souvent polysémique. De plus, une configuration morphosyntaxique n'est pas toujours associée à un même schème puisque le contexte, comme nous le verrons dans beaucoup d'exemples phis loin, peut intervenir pour filtrer le schème adéquat.

2. Causalité Par causalité nous entendons un schème qui établit une "relation causale" entre deux situations Siti et Sit2, que ces dernières soient, du point de vue aspectuel, conceptualisées comme des événements, des états ou des processus. La "relation causale" signifie soit que Siti est irne condition pour que Sit2 puisse avoir lieu, soit que Siti est présentée comme étant une explication ou une justification de Sit2- Les deux situations Siti et Sit2 sont soit en concomitance temporelle, soit dans des rapports d'antériorité (Siti avant Sit2) ou de postériorité (Sit2 après Siti). Le schème de causalité (I) se présente donc comme suit (la, Ib) : (Ia)

Siti —

(Ib)

Sit2 *—— EST-CAUSÉ-PAR *

»

CAUSE

» Sit2 Siti

Cette caractérisation très générale de la notion de causalité rejoint celle qui a été donnée par Nedjalkov et Sil'nickij (1969:6) avec la notion de kauzacija3 : "Les composantes immédiates d'une situation causale sont au moins deux microsituations, reliées entre elles par une relation de causalité (otnoSenie kauzacii) ou de cause (priânenija)."

Illustrons des phénomènes de causalité par quelques exemples en français : (1) (2)

La fermentation [processus] se montre ainsi corrélative de la vie [état] La terre est un système pseudo-isolé [état] puisque c'est un référentiel galiléen [état]

2

Un schème est une représentation conceptuelle d'une signification souvent associée à un verbe ou à une préposition. Une instance de schème est construite en instanciant les variables du schème. Techniquement, un schème est représentable par une λ-expression typée (Desclés 1990).

3

Cette conceptualisation de la causalité est reprise par Kulikov et Sumbatova (1993:327) : "V. P. Nedjalkov & Sil'nickij (1969:6) defined causative situations (CS) as a complex situation which consists of two elementary situations (Sj and Sj) connected by the relationship of causation (causal copula

CAUS)..."

Causalité, Causativité, Transitivité

(3)

9

La croissance en épaisseur [...] [processus] est due à l'édification de la membrane secondaire [processus] Lorsqu'il pleut [événement], la chaussée est mouillée [état] Un court-circuit [événement] a provoqué un incendie [événement]

(4) (5)

La relation causale peut être exprimée par des relateurs (comme en fiançais parce que, puisque, est dû à, cause, est causé par...) entre propositions ou par des marques de ponctuation4 . Les moyens d'expressions de causalité ne sont pas toujours réalisés sous la forme de deux propositions complètes (Nedjalkov et Sil'nickij 1969:7), reliées par l'expression d'un relateur causal : (6)

Comme les conducteurs de camions bloquent les routes, la vie économique est complètement ralentie.

Les situations Siti et Sit2 peuvent être exprimées par des nominalisations (7): (7)

La construction de l'immeuble entraîne des perturbations dans le voisinage

La relation causale est souvent lexicalisée par des verbes comme provoquer, entraîner, contraindre, rendre possible, causer, avoir pour effet, impliquer...5 Remarquons que dans certains cas l'identification d'une relation causale sous-jacente est phis délicate comme dans : (8)

La fumée provoque la toux

ou dans les exemples suivants où l'on a des expressions nominales à valeur processuelle, événementielle ou encore stative : (9) D a pris un cachet : il avait mal à la tête (10) Q est tombé, on l'avait poussé violemment (11) Le PDG de l'Altex en prison : les juges ont été sévères En effet, dans (8), seul le verbe provoque est un indicateur fort de causalité, alors que ses arguments ne renvoient pas a priori à des situations distinctes. Cependant, dans le contexte de provoquer, ces arguments sont analysés comme des activités ou des localisations (au sens abstrait du terme). Ainsi, l'énoncé (8) peut, selon son insertion dans un contexte phis large, être paraphrasé par:

4

Voir les listes de relateurs exprimant la causalité en russe dans Nedjalkov et Sil'nickij (1969a).

5

Une étude des moyens d'expression des verbes causatifs en fiançais a été menée par E. Kordi (1988). De son côté, A. Jackiewicz (thèse en cours, Université Paris-Sorbonne) propose une organisation cognitive de la notion de causalité à partir d'une analyse de textes en vue d'un traitement informatique.

Jean-Pierre Desclés / Zlatka Guentchéva

10 (8) a. b.

L'activité de fumer [état d'activité] provoque l'activité de toux. La présence de fumée dans un lieu [localisation] provoque l'activité de toux pour les personnes dans ce lieu.

Dans l'exemple (9) ci-dessus, le marqueur de ponctuation ":" est un indice qui permet de rétablir une relation causale entre "la prise de cachet" (Sit2) et "le mal de tête" (Siti). Dans (10), l'indice de ponctuation "," accompagnés des indications temporelles permet de reconstruire une relation causale sous-jacente. En effet, si l'on change le temps (10') ou la ponctuation (10"), on obtient des suites textuelles phis ou moins acceptables (10') ou n'exprimant plus une relation causale ; la suite textuelle (10") s'interprète comme deux événements indépendants et successifs mais sans lien causal entre eux : (10') (10")

*Π tomba, on le poussa violemment ?Π tomba. On le poussa violemment

Dans l'exemple (11) ci-dessus, l'identification d'une relation causale est rendue possible grâce à la présence du marqueur de ponctuation ":" et à la configuration syntaxique ; en d'autres termes, le marqueur ":" acquiert la valeur d'un relateur de causalité lorsqu'il est inséré dans le contexte d'une phrase nominale locative et d'une phrase nominale verbale. Ce type d'énoncé pourrait apparaître dans un titre de journal. La relation causale sera expliquée et reconstruite dans le corps de l'article, mais des connaissances sur les relations entre "juge" et "prison" donnent des indices pragmatiques pour interpréter (11) sous forme d'une relation causale paraphrasable par (11'): (11')

Le PDG de l'Altex a été mis en prison car les juges ont été sévères

L'expression d'ime causalité entre deux situations est manifestée par des constructions plus ou moins grammaticalisées qui focalisent l'attention soit sur Siti (cause, explication, justification, raison), soit sur Sit2 (conséquence, effet). Des marqueurs spécifiques peuvent apparaître dans les verbalisations de ces situations focalisées, ce qui conduit au scheme (le) orienté vers l'effet ou au schème (Id) orienté vers la cause ou la justification : (le)

(Siti)

>

CAUSE

-•

>

Sit2

(Id)

siti

>

CAUSE

-*

>

(Sit2)

En général, l'orientation vers l'effet (schème le) est beaucoup mieux grammaticalisée et plus fréquente dans les langues que le schème de causalité orientée vers la cause (schéma Id). Sur ce point, on pourra consulter l'article de Nedjalkov et SiMckij (1969a), ainsi que celui de Kulikov et Sumbatova (1993:327-330). Π y a causalité dans une expression linguistique orientée vers l'effet ou vers la cause (Sit2) lorsqu'il est possible de reconstruire une autre situation (Siti) ou une classe de situations qui

Causalité, Causativité, Transitivité

11

sont soit une origine causale de Sit2, soit est une justification de Sit2- Ainsi, dans le schéma (le), Siti (cause ou raison) doit être reconstructible à partir de la verbalisation de Sit2 ; dans le schéma (Id), Sit2 (conséquence) est déduite de la verbalisation de Siti. Dans leur article Kulikov et Sumbatova (1993:337) proposent un tableau typologique des principaux modes d'encodage des situations de causalité orientées "vers le résultat" ou orientées "vers la cause". Il nous semble que cette typologie devrait phis nettement distinguer les schèmes qui relèvent de la causalité de ceux qui relèvent de la causativité. Nous allons expliquer ce qui peut être qualifié de causativité et comment cette notion s'oppose à celle de causalité. Cette distinction nous paraît essentielle pour toute étude typologique d'un même phénomène dans les langues.

3. Causativité et transitivité Beaucoup d'auteurs analysent sémantiquement certaines dérivations morphologiques obtenues à partir d'un verbe intransitif à l'aide d'une primitive sémantique CAUSE. Ainsi, par exemple, en turc la forme verbale öl-dür-dü est analysée comme "die-CAUSE-Past". Ce phénomène dérivationnel qui apparaît dans un grand nombre de langues a conduit certains linguistes à reconnaître "une cause" dans l'analyse sémantique de certains lexèmes verbaux qui impliquent un résultat et à proposer alors une décomposition lexicale de ces lexèmes à l'aide d'une primitive CAUSE (McCawley 1968 ; Shibatani 1976). On peut rappeler ici la fameuse analyse de to kill en anglais sous la forme suivante : CAUSE (BECOME (DO NOT ALIVE) (y)) (χ) c'est-à-dire : "χ CAUSE que y devient non vivant". Π est bien connu que des marqueurs spécifiques, appelés "causatifs", permettent dans certaines langues de dériver par des procédés réguliers et productifs des verbes "transitifs" à partir de verbes intransitifs6. C'est le cas bien connu du turc où öl-diir- , que l'on peut traduire en fiançais par "tuer" ou par "faire mourir", est un dérivé morphologique du verbe öl("mourir")7 : (12) a.

Hasan öl -dû Hasan mourir -PASSE "Hasan est mort"

6

Voir Nedjalkov et Sil'nickij 1969b.

7

Dans sa grammaire du turc, Bazin (1968) donne une liste de suffixes, qu'il appelle "modificateurs verbaux", qui permettent à partir «fune base verbale de former une autre base verbale. Ces modificateurs "transforment le sens fondamental du verbe pour lui donner une valeur réfléchie, contributive, factitive, passive, négative ou impossible" (p. 66). Prenons quelques exemples de ce que l'auteur appelle des "factitifs" : anla"comprendre" -> anla-t "faire comprendre, expliquer"/ "laisser comprendre", geber "crever" (en parlant d'un animal") —• geber-t "faire crever, laisser crever", ye- "manger" ye-dir- "faire manger", "laisser manger"...

12

Jean-Pierre Desclés / Zlatka Guentchéva

b.

Mehmet Hasan-i öl -dür -dû Mehmet Hasan-ACC mourir-CAUS -PASSÉ "Mehmet a tué Hasan"

c.

Mehmet Hasan-i Ali-ye öl -dür -t -dû Mehmet Hasan-ACC Ali-DATmourir-CAUS-CAUS -PASSÉ litt. Mehmet a-fait-tuer Hasan à Ali "Mehmet a feit tuer Hasan par Ali"

Le turc ne permet pas de rendre compte de la distinction sémantique entre "faire mourir" et "tuer" car il ne dispose pas de deux lexèmes pour exprimer cette différence8 . Lorsque le marqueur causatif est redoublé, on construit une relation avec deux agents et un patient9. Ainsi dans (12c), le premier agent (Mehmet) est l'instigateur, alors que le second agent (Ah) est l'effectueur mais, tout comme dans (12b), Hasan est le patient qui subit un changement d'état. S'il est vrai que le causatif simple ou redoublé (voire triplé) augmente la valence du point de vue syntaxique, il ne s'ensuit pas automatiquement que le marqueur causatif a exactement la même interprétation sémantique. Si l'on compare (12a) et (12b), le rapport entre les deux constructions est un rapport de transitivisation, tandis que le rapport entre (12b) et (12c) est un rapport non pas de transitivisation, mais de factitivité. Du point de vue strictement syntaxique, ces deux rapports sont homologues : augmentation de valence (Nedjalkov et Sil'nickij 1969b ; Comrie 1976, 1981) ; du point de vue sémantique, la factitivité diffère de la transitivité par certains aspects. On doit donc se demander si le morphème dérivationnel, appelé souvent "causatif', n'a pas une fonction plus profonde qui n'aurait rien à voir avec la causalité (au sens précis que nous lui avons donné plus haut) et qui le rapprocherait de l'agentivité (dans un sens qui est précisé plus loin). Nous reviendrons sur cette discussion. En français, il n'y a pas de marqueur morphologique dérivationnel équivalent. Cependant, un grand nombre d'items verbaux peuvent être employés de façon intransitive et transitive : (13)a. Le poulet cuit (Vjntr) b. Marie cuit (Vtr) le poulet ( 14) a. Le linge sèche au soleil b. Marie sèche le linge dans la machine c. Le soleil sèche le linge ( 15) a. La fumée monte au-dessus du toit b. Marie a monté les valises au premier étage c. Marie monte les escaliers

8

Voir à ce propos les commentaires sur la voix causative de Creissels (1995:286-296).

9

Par exemple: anla- "comprendre" -> anla-t "faire comprendre, expliquer" -> anla-t -tir "faire expliquer"

Causalité, Causativité,

13

Transitivité

En français, les énoncés (13b), (14b) et (15b) sont clairement transitifs. Si l'on considère maintenant (14c) et (15c), il convient de distinguer la transitivité sémantique de la transitivité syntaxique, la seconde étant une extension, selon certains processus assez réguliers, de la première (Givón 1984, 1990; Desclés 1997). La transitivité sémantique peut être plus au moins orientée vers un but (téléonomie) mais elle implique toujours un agent et un patient : l'agent a la capacité de contrôle, c'est-à-dire "la capacité de déclencher et d'interrompre un processus qui affecte le patient" ; de plus, l'agent possède "la capacité d'effectuer directement ou par l'intermédiaire d'un instrument ce processus". La transitivité sémantique est liée à un seul processus qui engendre un seul événement avec deux participants (un agent et un patient) et un état résultatif du patient. Certes, l'agent peut être considéré dans certains cas comme responsable de la situation qui affecte le patient et parfois comme étant à l'origine ou l'instigateur de cette situation mais nous avons un seul processus qui conduit à un état résultatif Lyons contribue à éclaircir ce point, mais sa formulation prête à confusion : "(...) on peut envisager ce qui vient de se passer comme un événement unique, comme un processus étendu (si peu que ce soit) dans le temps ou comme une séquence de deux ou plus de deux situations (événements, états ou processus). En utilisant le verbe 'tuer1 on peut décrire ce qui s'est passé comme un seul événement où se trouvent impliqués un agent (X) et un patient (Y). L'action de l'agent n'est pourtant décrite qu'en fonction de son effet sur le patient (...). (...) la situation que décrit X a tué Y (...) peut être analysée en termes de deux schémas de valence distincts. Selon un premier point de vue tuer est ce que nous appellerons verbe opératif : le fait de tuer est accompli aux dépens d'un patient et il l'affecte. Selon un autre point de vue, c'est ce que l'on appelle communément un verbe factitif : il dénote un processus ou un événement par lequel une cause produit un effet (ou résultat). Les deux schémas (...) sont : (1) (2)

AFFECTER (AGENT, PATIENT) (opératif) PRODUIRE (CAUSE, EFFET) (factitif)

En outre, en vertu du lien entre l'agentivité et la causalité, on a un troisième schéma possible combinant des éléments de (1) et de (2) à la fois. Soit : (3)

PRODUIRE (AGENT, EFFET)

(opératif-factitif)"

(Lyons 1980:123-124)

Il est clair que Lyons procède par analogie : l'agent est à la cause ce que le patient est à l'effet, d'où le schéma (3) déduit par analogie de (1) et (2). De phis, il assimile la relation d'affectation par l'agent à une relation de causalité, faisant ainsi jouer à l'agent le rôle d'une situation entière, alors que ce dernier n'est qu'un participant dans une situation unique. En effet, la relation de transitivité n'est pas établie directement entre un agent et un patient mais entre un agent et une situation (cinématique ou dynamique) de changement qui affecte un patient, que ce soit un changement de position spatiale ou un changement de propriété. Or, comme nous l'avons dit, la causalité est une relation entre deux situations et non pas une relation entre des participants. On ne peut donc pas assimiler l'agent à une cause et le patient à un effet : l'agent et le patient sont deux participants, la cause et l'effet sont deux situations. L'analogie de Lyons est donc contestable.

14

Jean-Pierre Desclés / Zlatka Guentchéva

Dans un scheme transitif un seul agent contrôle et effectue un processus en affectant complètement ou partiellement un autre participant qui, lui, assume le rôle de patient. Ainsi ( 16) a.

Le chasseur a tué un daim

ne s'analyse pas, selon nous, en termes d'une relation causale comme : ( 16)b.

Le chasseur CAUSE (le daim est mort)

mais par la construction plus complexe : ( 16) c.

Le chasseur CONTRÔLE et EFFECTUE le processus qui conduit le daim à l'état "le daim n'est plus en vie".

Nous voyons donc qu'une construction transitive ne s'analyse pas comme l'e^ression d'une relation de causalité entre deux situations. Certes, une construction transitive a souvent comme conséquence un état résultatif10 du patient comme dans les deux exemples suivants : (17) a. Le chasseur a tué le daim b. Le daim est mort (18) a. b.

Marie a cassé la tasse La tasse est cassée

» >

mais l'origine de cette affectation du patient ne peut pas être analysée comme une situation (qui dénoterait un événement) puisque l'agent n'est pas, à lui seul, un événement ; bien au contraire, il est entièrement intégré au processus d'affectation et ne fait nullement partie d'une quelconque situation qui serait la cause de la situation résultative du patient. Aussi l'agent peut-il être considéré comme étant à l'origine de l'affectation du patient mais la relation entre l'agent et l'état résultatif n'est pas une relation de causalité. Ainsi, dans (17), l'état "le daim est mort" ne permet pas de reconstruire une situation causale antérieure. Du reste, du point de vue linguistique, la situation "le daim est mort" n'est pas nécessairement interprétable comme une situation résultative d'un processus mis en œuvre par un agent. Cet état peut être purement descriptif et bon nombre de langues, dont le français, utilisent les mêmes procédés d'encodage morphologique pour exprimer l'état résultatif et l'état descriptif De même, l'état "la tasse est cassée" n'implique pas nécessairement un agent qui aurait effectué et contrôlé un processus ayant pour terme résultatif cet état : "la tasse est cassée" peut être simplement le constat d'un état descriptif de la tasse.

10

Ce test est largement utilisé dans les études sur les constructions causatives. Shibatani (1976:2), par exemple, s'en sert pour opposer verbes causatiis (open, melt...) et verbes transitifs (kick,push...).

Causalité, Causativíté, Transitivi té

15

Les constructions qui manifestent un schème de transitivité sémantique sont étroitement corrélées la notion d'agentivité, c'est- -dire aux notions de contrôle (au sens de "capacité de déclencher ou d'interrompre un processus") et d'effectuation directe ou médiate (par l'intermédiaire d'un instrument). Les schèmes de causalité et de transitivité sémantique sont donc disjoints. Le schème (I) établit une relation causale entre Siti et Sit2 ; le schème (Π) établit une relation entre un agent et une situation complexe qui, elle-même, exprime un changement ou un mouvement, c'est- dire une modification affectant un patient, lequel passe ainsi d'une situation stative Sit'i (patient) exprimant un état du patient une autre situation souvent résuhative Sit'2 (patient) qui exprime un autre état du même patient : (I)

Siti

»

CAUSE

>

Sit2

S i t ' 1 (PATIENT)

(Π)

CONTROLE

Agent

—» et EFFECTUATION

»

changement d'états Sit'2 (PATENT)

Le schème Π est celui de la transitivité sémantique qui s'exprime par des constructions transitives. Cependant, remarquons-le, une construction transitive ne renvoie pas obligatoirement un schème de transitivité sémantique. En effet, il existe des mécanismes d'extension assez réguliers, mais variables selon les langues, qui étendent syntaxiquement les constructions transitives comme dans : ( 19) a. Les colonnes supportent le toit b. Le vent a cassé les vitres du salon c. La maison abrite la famille Il est évident que, dans ces exemples, le terme en position sujet n'assume pas le rôle sémantique d'un agent qui contrôlerait un processus de changement (sur ce point, voir Desclés 1997).

4. La factitivité L. Tesnière (1966:259 sq) a remarqué qu'un verbe comme faire en français augmente la valence (tandis que le réflexif se est récessif). Ainsi, certains verbes intransitifs Vjntr se voient associer une locution verbale transitive formée au moyen de faire + infinitif : (20) a. b.

Marie tombe Jean fait tomber Marie

16

Jean-Pierre Desclés / Zlatka Guentchéva

Ainsi, trois cas se présentent en français : I o ) certains verbes (comme monter, toucher, avancer...) ont des emplois intransitifs et des emplois transitifs ; 2°) : à certains verbes intransitifs (comme tomber, aller...) sont associés des locutions verbales transitives; constituées au moyen de faire -, 3°) certains verbes intransitifs se voient associer des verbes transitifs qui, en tant qu'items lexicaux, sont différents (exemple : tuer, mourir). (21 ) a. La voiture avance a1. Pierre avance la voiture b. Le chien marche dans la rue b'. Jean fait marcher son chien dans la rue c. Le daim est mort c'. Le chasseur a tué le daim Tesnière a considéré que les énoncés (21a'), (21b') et (21c1) étaient des diathèses causatives avec des verbes causatifs qui sont soit lexicalisés de façon identique, soit dérivés par faire, soit encore lexicalisés différemment. Il en résulte alors les équivalences : (22) a. b. c.

avancertr = FAIRE + avancer^ {faire marcher)tr = FAIRE + marcher^ tuer\x = FAIRE + mourir^

Ces analyses entraînent plusieurs remarques. La première remarque concerne la nonéquivalence sémantique entre tuer et faire mourir. En effet, le verbe tuer exprime un schème transitif avec un seul processus et deux participants , faire mourir évoque plutôt un intermédiaire entre l'agent et le patient : (23) a. b.

Le chasseur a tué un daim Le dictateur a fait mourir de nombreux prisonniers dans ses camps.

La seconde remarque touche à la notion de "causativité". En effet, l'introduction d'un nouvel actant ne construit pas nécessairement une "relation de causalité", au sens précis qui a été donné plus haut. La "construction causative" avec faire augmente la valence sans que pour autant l'actant supplémentaire soit impliqué dans une relation entre deux événements dont l'un serait la cause de l'autre. La troisième remarque touche au statut même de l'actant introduit par faire. Cet actant, dans les exemples prototypiques, est un agent (c'est-à-dire qu'il a une capacité agentive de contrôle) mais il peut ne pas l'être. Prenons les exemples suivants :

Causalité, Causativité,

17

Transitivité

(24) a. b. c. d.

Pierre écrit une lettre Marie fait écrire une lettre à son fils (ambigu) L'hiver fait écrire plus souvent aux amis L'école fait écrire les élèves à leurs parents pour le Nouvel an.

(25)a. b.

Les tuiles sèchent au soleil Le soleil fait sécher les tuiles

Dans la première série d'exemples, le sujet Marie est un agent qui exerce sa capacité de contrôle (24b), ce qui n'est pas le cas du sujet dans (24c) et dans (24d). Dans (24c), le sujet est un locatif temporel d'où la paraphrase (24c') : En hiver, on écrit plus souvent à ses amis. Dans (24d), le sujet l'école peut être considéré également comme un simple heu ou encore comme un lieu qui a hérité des capacités agentives, d'où les paraphrases : (24d') A l'école, on fait écrire les élèves ... ou (24d") L'école, en tant qu'institution, oblige les élèves à écrire... Dans la deuxième série d'exemples, le soleil n'est nullement un agent qui exercerait une quelconque capacité de contrôle (25b). On peut considérer que le soleil est un locatif d'où la paraphrase (25b') : Au soleil, on fait sécher les tuiles. Nous ne pouvons donc analyser les phrases avec le schéma "faire +Vinf(initif)" comme étant toutes des phrases transitives, avec un agent et un patient, dérivées de phrases intransitives. Nous ne pouvons pas non plus les analyser comme des phrases qui encoderaient toujours des "relations de causalité". Cela nous amène donc à ne pas analyser faire comme un "verbe causatif'. Nous définirons la construction factitive comme une construction qui implique deux agents dont l'un exerce une certaine influence sur le second. Donnons quelques exemples : (26) a. b. c.

Le comte fait tuer les cerfs de son parc (par ses gardes) Mane fait repasser le linge (par sa femme de ménage) Le président fait écrire à tous les membres de son parti

Dans une construction factitive, un agent exerce sa capacité agentive par rapport à un autre agent qui peut être explicite ou non (par exemple en (26c) le second agent est implicite). Selon le rapport qui s'établit entre ces deux agents, le processus est exécuté ou non par le deuxième agent. Nous associons à la construction factitive le schème suivant : Sit'l (III)

(PATIENT)

CONTRÔLE +

Agenti

de Agenti — * sur Agent2

— » ACTION

Agent2 t

SÌt'2

(PATIENT)

18

Jean-Pierre Desclés / Zlatka Guentchéva

Dans le cas de la facthivité, nous avons un seul processus avec trois participants : 1°) un agent principal qui, ayant le contrôle sur le processus, exerce un rôle d'instigateur (il a le pouvoir de déclencher le processus) ; 2°) un agent secondaire doté d'un certain pouvoir de contrôle qui lui permet d'exécuter ou non le processus voulu par l'agent principal ; 3°) un patient affecté par le processus. Le schéma montre bien qu'il ne s'agit pas d'une relation de causalité entre deux situations. L'agent principal n'est qu'un instigateur ; il n'est pas impliqué dans un événement qui serait la cause de l'événement verbalisé. Nous sommes donc maintenant amenés à différencier explicitement les schèmes de causalité (schème I), de transitivité (schème Π) et de factitivité (scheme ΠΙ). Remarquons que la construction grammaticale du français "faire + Vjnf" n'exprime pas toujours un schème factitif11 : (27) a. b. c.

Piene feit mourir le daim Jean fait craquer ime allumette Jean fait marcher le toume-disque

Ces énoncés ne renvoient pas à des schèmes factitifs puisqu'un seul agent est impliqué et que cet agent contrôle et agit sur un patient qui, lui, n'a aucune activité agentive. Nous sommes donc dans des situations différentes de celles exprimées par les exemples (26). Ces constructions sont très proches des constructions transitives et on peut les comparer à : (28) a. b. c.

Pierre tue le daim Jean craque une allumette *Jean marche le tourne-disque

Prenons maintenant les exemples suivants : (29) a. b. c. d. e. f.

La fumée fait tousser La fumée fait tousser Jean Le choc a fait craquer les poutres qui soutenaient le plancher L'eau bouillante a fait éclater les verres Le vent faisait passer sur la ville une certaine douceur La sympathie fait éclore bien des qualités somnolentes

Dans ces exemples, le terme sujet n'est pas un agent et ces constructions ne sont analysables ni comme constructions transitives, ni comme constructions factitives. Ils renvoient ici à des schèmes de causalité où une situation initiale peut être reconstruite, cette dernière étant assimilée à une cause d'une autre situation. Ainsi, nous avons :

' 1 Sur le traitement des constructions avec faire (et laisser), dans le cadre du modèle typologique élaboré par l'École de Leningrad, voir les analyses et les distinctions introduites par E. E. Kordi (1988). Nous lui avons emprunté certains exemples, en les modifiant légèrement.

19

Causalité, Causativité, Transitivité

(30)a\ Lorsqu'il y a de la iùmée dans un lieu, on tousse b'. Lorsqu'il y a de la fumée dans un lieu où est Jean, Jean tousse c'. Il s'est produit un événement, à savoir un choc et cet événement est la cause du craquement des poutres qui soutenaient le plancher d'. L'eau était bouillante et cela est la cause de l'éclatement des verres e'. Le vent, pour avoir soufflé, est la cause de la douceur qui a passé sur la ville f. Lorsque la sympathie se manifeste, les qualités somnolentes apparaissent Une construction en "faire + V¡nf" peut ainsi exprimer trois types de schèmes : a) la causalité ; b) la transitivité ; c) la factitivité. Dans certains cas, on peut hésiter entre ces trois schèmes. Lorsque le terme sujet n'a pas de capacité agentive et qu'il renvoie à ime classe d'événements ou à un processus, la construction exprime plutôt une causalité. Lorsque le terme sujet a une capacité agentive, la construction a tendance à exprimer une transitivité ou une factitivité selon que le second actant est analysé comme un patient ou comme un agent secondaire sous la dépendance d'un agent principal. Π est bien évident que la nature sémantique du verbe intervient dans l'identification du schème associé. Illustrons ce problème par les exemples suivants : (3 l)a. b.

La tempête fait découvrir aux enfants la force de la nature La force de la nature apparaît aux enfants comme étant la cause de la tempête [causalité]

(32)a. b.

Paul fait découvrir la mer à ses enfants Paul fait que ses enfants découvrent la mer

[factitivité]

(33)a. b.

Paul fait rencontrer Marie (à ses enfants ) Paul fait pleurer Marie (*à ses enfants )

[factitivité] [transitivité]

On pourrait rapprocher les constructions factitives (c'est-à-dire les constructions en "faire + Vinf ' qui s'analysent par un schème factitif) des constructions permissives ("laisser +Vjnf '') et des constructions avec certains verbes de perception + Vinf : (34) a. Paul laisse dormir Marie a'. Paul fait dormir Marie b. Pierre entend sortir Marie b'. Pierre fait sortir Marie c. Paul voit condamner Marie (par le juge) c'. Paul fait condamner Marie (par le juge) Cependant, il y a des différences syntaxiques et sémantiques entre ces constructions :

20

Jean-Pierre Desclés / Zlatka Guentchéva

(35)a. Paul voit le juge condamner Marie a'. Quant au jugei, Paul lei voit condamner Marie a". Quant à Mariej, Paul laj voit condamnée (par le juge) b. *Paul fait le juge condamner Marie b'. Quant au jugei, Paul lui¡ fait condamner Marie b". Quant à Mariej, Paul laj fait condamner (par le juge) Avec les verbes de perception, le sujet exerce plutôt sa capacité d'observateur (qui contrôle sa perception) que sa capacité agentive qui contrôle et effectue un processus affectant un patient 12 .

5. Causativité et typologie Π est devenu classique d'interpréter comme causatives certaines dérivations qui se produisent avec plus ou moins de régularité dans diverses langues et qui aboutissent dans certaines à une véritable grammaticalisation (Nedjalkov et Sil'nickij 1969b ; Inenlikej et al 1969). Ainsi, la dérivation par l'insertion d'un morphème dit causatif dans la forme verbale conduit à établir des paires régulières entre les formes de départ et les formes dérivées. C'est le cas souvent cité du hongrois (Hetzron 1976), du géorgien, du swahili (Combettes 1996). Cependant, dans certaines langues où l'on reconnaît une dérivation causative qui peut s'opérer à partir d'une construction aussi bien intransitive que transitive, le marqueur causatif ne semble pas toujours assumer le même rôle sémantique. Revenons à ce propos sur le turc en comparant (12a) Hasan öl-dü (Hasan mourir-PASSE) "Hasan est mort" et (12b) Mehmet Hasan-i öl-dür-dü (Mehmet Hasan-ACC mourir-CAUS-PASSÉ) "Mehmet a tué Hasan" à (36a) et (36b) respectivement13 : (36)a.

b.

Hasan kos -tu Hasan courir -PASSÉ "Hasan a couru" Ahmet Hasan-i kos Ahmet Hasan-ACC courir "Ahmet a fait courir Hasan"

-tur -du -CAUS -PASSÉ

Le morphème causatif en (36b) a une fonction comparable à celle en (12b) : il permet d'augmenter la valence en introduisant un nouvel actant en position sujet et en modifiant le statut syntaxique de l'actant de la structure de départ tout en respectant la hiérarchie des relations

13

Une étude sur les relations entre certains verbes de perception et les constructions avec faire, laisser, donner... devrait être systématiquement menée de façon à faire apparaître certaines organisations et gradients. A ce sujet, on pourra consulter, pour l'anglais, l'article de C. V. Chvany (1993). Certains exemples turcs ont été empruntés à Zimmer (1976:399-340) et parfois remaniés.

Causalité, Causativité,

Transitivité

21

grammaticales (Comrie 1976). Dans chaque structure dérivée il y a donc deux actants. Mais en tant que participant de la situation, le second actant Hasan en (36b) semble hériter de l'agentivité qu'il a en (36a). On peut alors être tenté de l'analyser comme un deuxième agent. Cependant, (36b) ne dit pas vraiment que Hasan a couru, comme le fait (36a), mais que l'événement 'Hasan a couru' a été effectué sous le contrôle de l'agent Ahmet et que cet événement concerne Hasan en l'affectant. Autrement dit, bien qu'ayant par lui-même ime capacité agentive, le participant Hasan ne l'exerce pas ; il n'est donc pas un deuxième agent, comme dans les factitives, mais mi participant dont le comportement est similaire au patient de (12b). Aussi peut-on considérer (36b) comme l'expression d'une transitivité sémantique et une instanciation du schéma (II). Mais le morphème causatif simple permet de construire également un rapport de factitivité qui implique deux agents même lorsque le deuxième agent n'est pas explicitement mentionné : (37)a.

b.

Hasan kutu-yu ac-di Hasan boîte-ACC ouvrir-PASSÉ "Hasan a ouvert la boîte" Mehmet kutu-yu [Hasan-a] ac Mehmet boîte-ACC [Hasan-DAT] ouvrir "Mehmet a fait ouvrir la boîte [à Hasan]"

-tir -di -CAUS -PASSÉ

Sa fonction est donc comparable à celle qui apparaît avec le double causatif dans (12c) Mehmet Hasan-i Ali-ye öl-dür-t-dü (Mehmet Hasan-ACC Ali-DAT mourir-CAUS-CAUSPASSE) "Mehmet a fait tuer Hasan par Ali" : Mehmet est un agent-instigateur du processus et Hasan est un deuxième agent-effectueur (voir le schéma (III)). Par conséquent, sur le plan sémantique, le morphème dérivationnel "causatif' ne joue donc pas toujours un même rôle. Nedjalkov et Sil'nickij (1969b:24) soulignent à juste titre que, dans la plupart des langues analysées, non seulement les moyens formels mis en jeu pour construire le causatif ont une très grande affinité à s'adjoindre à des verbes intransitifs, mais que ces procédés sont de plus polysémiques. Particulièrement intéressantes sont sur ce point les données de la langue chukchee (Inenlikej et al. 1969) où le dérivé causatif14 se construit au moyen d'un confixe, constitué par r(ah, lorsque ce dernier est en position initiale, et par un suffixe, qui peut se réaliser par -av/-cv, et/-at, -Qct/-ijat ou 0 : e'iketek "se marier" r-?elike-v-a-k "marier (quelqu'un)". Sur le plan syntaxique, le confixe augmente la valence du verbe auquel il s'adjoint; sur le plan sémantique, suivant les propriétés lexicales du verbe, il permet l'expression de valeurs très variées (comitative (38), agentive (39), instrumentale, reflexive, bénéfactive, intensive...). Prenons les deux exemples suivants (que les auteurs citent p. 262 et p. 267 respectivement) :

En chukchee (p. 266), il existe également un causatif analytique qui est formé au moyen du suffixe -jgo(t) et roink "avoir" : e 'likci-ak-jgut ralak "obliger quelqu'un à se marier (faire épouser)".

-jgu(t) /

22

Jean-Pierre Desclés / Zlatka Guentchéva

(38)a.

b.

(39) a.

b.

attag-sn pakir-g?i jara-k père-AB S arriver-3SG/Aor maìson-LOC "Le père est rentré à la maison" attag-e ra -pkir -en -nin père-ERG CAUS -arriver -avec-3SG:3SG/Aor litt. Le père est rentré avec le phoque à la maison "Le père a ramené le phoque à la maison"

memal jara-k phoque maison-LOC

qaat-t melek-v?et cerfs-AB S guérir-3PL "Les cerfs ont guéri / se sont rétablis/ sont rétablis" attag-e qaa-t ro-melev-nin-et père-ERG cerfs-ABS CAUS-guérir-3SG:3SG-PL(Aor) "Le père a guéri les cerfs"

Bien que l'article de Kozinsky et al. (1988) traite de la transitivité, les énoncés avec le morphème r(s)- ne sont pas traités. Π nous semble cependant que (38b) et (39b) devraient être analysés comme l'expression d'une transitivité sémantique. En effet, dans l'énoncé uniactanciel (38a), le participant (à l'absohitif) se déplace selon sa propre initiative d'un heu à m autre lieu (de l'extérieur de la maison à l'intérieur de la maison) et contrôle donc lui-même son déplacement ; dans (38b) où le confixe rfa)- + ... est obligatoire, ce même participant (à l'ergatif) effectue le même déplacement avec un deuxième participant. Le confixe participe ainsi à construire la relation qui s'instaure entre les deux participants où l'un "domine" l'autre : le père qui assume le rôle d'agent, est l'initiateur intentionnel du changement qu'il effectue et qui affecte en même temps le deuxième participant ; ce dernier se voit ainsi assigné un rôle de patient. En ce qui concerne (39b), l'introduction du père en tant que participant dans la situation permet d'introduire un agent dans le processus de changement qui affecte un autre participant en le faisant passer d'une situation stative à une autre situation stative. Ainsi, l'agent a le contrôle sur l'eflfectuation du processus 'guérir les cerfs' qui fait passer le patient d'un état à un autre état. On peut considérer que la fonction du marqueur r(g)-, constitutif d'un prédicat en chukchee, participe à encoder la notion de contrôle et celle d'effectuation : les énoncés du type (38b) et (39b) seront représentés par notre schéma (Π). Remarquons que les verbes de cette classe cités par les auteurs conduisent tous à cette même interprétation : racvsqsk "faire la course" -» r^-racvorj-avsk "faire la course avec quelqu'un". C'est par le biais de la transitivité sémantique que l'on pourrait expliquer aussi, à notre avis, l'opposition qui a été dégagée en nivkh (Nedjalkov et aL 1969:183) entre causatifs lexicalisés et causatifs morphologiques. En effet, les causatifs lexicalisés [formés au moyen du suffixe -u- , procédé non productif qui s'accompagne d'alternances bien précises en ce qui concerne la consonne initiale de la forme de départ] entrent dans des constructions qui, selon les auteurs, renvoient exclusivement à la "fàctitivité directe" [(40b) et (41b)]. En ce qui concerne les causatifs morphologiques [(40c) et (41c)] qui se construisent au moyen du suffixe -gu- ou après une

Causalité, Causativité, Transitivité

23

consonne sourde -ku- (procédé productif dans la langue), ils servent à exprimer le plus souvent la "factitivité distante et permissive" et, de façon plus restreinte, la "factitivité directe"15 : (40) a.

if pold· il tomber"11 est tombé"

b. if p'atik vol-u-d'· il frère-cadet tomber-CAUS"Π a fait tomber son petit frère / a renversé son petit frère" c.

(41)a.

b.

if p'atik axtr, jax pol-gu-d'· il frère-cadet a-poussé celui-ci tomber-CAUS"11 a poussé son petit frère et celui-ci est tombé" lep ce-d'· pain sécher"Le pain a séché" if lep se-n-d'· il pain sécher-CAUS"Π a grillé le pain "

c. if lep ce-gu-d'· il pain sécher-CAUS"11 a laissé sécher le pain (en ne le couvrant pas)" En utilisant la notion de contrôle et celle d'effectuation, nous proposons l'interprétation suivante : 1. La construction avec -u- implique que le changement subi par le patient est le résultat de l'action que l'agent a lui-même déclenchée et effectuée, directement dans (40b) ou par l'intermédiaire d'un instrument dans (41b)16 ; le contrôle opéré par l'agent est explicite dans (41b). Si la notion de contrôle paraît moins évidente dans (40b), la classe des verbes que les auteurs présentent nous conduit à penser qu'elle est constitutive du sémantisme du verbe : p-il-d' "être grand" —» v-il-u-d' "agrandir" ; k-on-d' "être gelé" —> γ-οη-ud' "congeler" ; n'-az-d' "se déshabiller" f-az-u-d' "déshabiller". Il serait très instructif d'examiner le type d'adverbes admis dans ces constructions.

15

Nous remercions L. Kulikov d'avoir attiré notre attention sur le fait que la notion de kontaktnaja kauzacija "causativité directe", introduite par Nedjalkov et Sil'nickij (1969b), correspond à la notion de transitivité sémantique. En ce qui concerne distantnaja kauzacija "causativité indirecte", elle englobe, à notre avis, aussi bien la notion de factitivité que celle de causalité.

16

Voir à ce propos la classification des constructions transitives donnée par Givón (1984:97).

24

Jean-Pierre Desclés / Zlatka Guentchéva

2. La construction avec -gu- présente deux cas : soit elle implique deux participants dont l'un peut exercer sa capacité de contrôle et effectuer le processus sur l'autre ou peut ne pas le faire par omission (41c), soit elle peut être constitutive d'une causalité (40c). Un cas intéressant se présente avec les langues athabaskanes (Kibrik 1993) où le causatif se construit à partir de bases verbales exclusivement intransitives qui sont essentiellement non agentives (irne exception signalée pour le navajo), comme dans l'exemple du slève (athabaskan): (42) a. b.

we-0-gq "The meat is dried" î^yq we-h-gq "S/he dried the meat"

Kibrik (1993:49) met en évidence que dans ces langues le verbe se présente sous forme d'une structure propositionnelle prototypique et qu'il se définit par un certain nombre de propriétés (valence, classe sémantique...). Du fait que cette structure peut être modifiée par l'addition ou la suppression d'un actant, par le changement de classe sémantique ou du rôle sémantique d'un argument..., l'auteur formule l'hypothèse que le morphème qui encode la causativité "factitive directe" dans l'ensemble des langues athabaskanes, est un indicateur de transitivité : "The Athabaskan system of signalling transitivity differs from those found in some other languages (see Hopper and Thompson 1982) in that it signals not degrees of transitivity but changes in transitivity - either increase or decrease. For example, the - 1 - TI [transitivity indicator] indicates not high degree of transitivity, but the process of transitivization, no matter what the starting and the final level of transitivity are." (Kibrik 1993:63)

Dans diverses langues le morphème "causatif1 participe le phis souvent à la construction d'un prédicat dont l'interprétation sémantique n'est pas automatique et dépend de plusieurs paramètres. Même dans le cas du double causati^ comme le signale Kulikov (1993:128-129) à propos du caraïbe (43) ou du tadjik (44), sa présence est dans certaines langues liée à la pluralité de certains participants qui sont impliqués dans la situation causative : Caraïbe: (43)a. [...] kaiku: si ?wa kisi: wopoi "Do not let him be killed by the jaguar" b.

kisi: wopo: poi kaiku: si ?wa "Do not let him be killed by all these jaguars"

Causalité, Causativité, Transitivité

25

Tadjik : (44) a. Ali vazifàro fahmirt Ali problem-ACC understood "Ali understood the problem" b.

mujsafed ba Ali vazifaro fahm-on-d old-man to Ali problem:ACC understand:CAUS "The old man explained the problem to Ah"

c.

mualimon ba Ah vazifaro fahm-on-on-dand teachersrPL to Ah problem:ACC understand:CAUS:CAUS "The teachers explained the problem to Ah"

Le yukaghir fournit un autre exemple très significatif (Maslova 1993:274) : les marqueurs utilisés pour signifier des oppositions "semelfactif / momentané" vs "multiplicatif / distributif ' coïncident partiellement avec le causatif. Étant donné la polysémie des marqueurs causatifs, il est légitime de se demander si un dérivé causatif doit être systématiquement interprété comme un prédicat complexe de type CAUSE(Verbe).

Conclusion Nous avons, comme beaucoup d'autres auteurs, employé les termes de "causatif', "structure causative", "verbe causatif', morphème "causatif'. Que signifie "causatif' ? Nous avons bien pris soin de distinguer la notion de "causalité", précisée par son schème (schème I) de ce que nous appelons causativité. Cette dernière notion n'est pas une notion pure car notre analyse, très rapide, de certains faits tirés du fiançais et de quelques langues, a montré qu'elle recouvre plusieurs notions plus élémentaires que l'on commence à mieux cerner. En effet, la causativité recouvre les notions de : (1) transitivité et processus de transitivisation ; (2) factitivité et permissivité ; (3) une classe de phénomènes plus hétérogènes impliquant la possession, la localisation, l'aspectuahté... (Kibrik 1993). Si l'on laisse de côté la troisième classe qui demande à être examinée plus en profondeur, il apparaît que derrière les notions (1) et (2) se cachent des paramètres sémantiques autres que ce que l'on trouve derrière la notion de causalité. L'analyse de la notion de transitivité fait apparaître les paramètres de contrôle et d'effectuation (schème Π). Certes, on peut avoir plusieurs degrés de transitivité (avec plus ou moins de contrôle, de plus ou moins d'effectuation), mais le paramètre même de contrôle est nécessaire pour appréhender l'agentivité et la transitivité. Remarquons que si l'agentivité est constitutive de la transitivité sémantique, il peut cependant y avoir agentivité, donc contrôle, indépendamment de la transitivité). Dans la factitivité et la permissivité, le paramètre de contrôle est lui aussi constitutif de cette notion mais s'y ajoute une relation inter-agents, dans laquelle soit un agent "domine" l'autre agent (factitivité), soit un agent "laisse" l'autre agent agir selon sa propre initiative. Dans certaines langues, la notion inter-agents peut être conceptualisée comme un contrôle d'un agent sur un autre agent. Π

Jean-Pierre Desclés / Zlatka Guentchéva

26

apparaît dans ces conditions que la causativité est fondamentalement liée au paramètre constitutif de contrôle par un agent et non pas à la causalité. Le terme même de causativité entraîne des glissements vers la causalité alors que ces deux notions ne sont pas de même nature. Certains schémas syntaxiques renvoient à plusieurs schèmes sémantiques. Par exemple, en français, le schéma "faire + Verbe infinitif' peut, selon les contextes où il s'insère et selon le verbe instancié, renvoyer au schème de transitivité analytique (45a), de factitivité (avec deux agents (45b)) ou de causalité (45c): (45) a. (45)b. (45) c.

Jean fait mousser le champagne Jean fait repasser le linge à Marie La fumée fait tousser

L'examen des données du turc (§ 3 ; § 5) ou d'autres langues fait apparaître que les schémas syntaxiques ne sont pas toujours assez transparents pour révéler les relations sémantiques sous-jacentes aux marqueurs causatifs. Si ces marqueurs augmentent la valence d'un verbe, il ne s'ensuit pas que, dans toutes les constructions où ils opèrent, la nature sémantique soit exactement la même. Il se peut qu'un invariant puisse être dégagé et que des valeurs phis spécifiques soient reliées à cet invariant. Les données des typologues laissent apparaître que la causativité est assez souvent une indication d'un certain contrôle agenti^ mais certaines de ces données (en particulier celles fournies par Kibrik) nous obligent à nuancer la tendance que nous venons de dégager.

Références bibliographiques Bazin, L. (1968) Introduction à l'étude pratique de la langue turque. Paris: Maisonneuve. Chvany, C.V. (1993) Bystander Voice in English. In Comrie & Polynsky (eds), 241-253. Combettes, B. (1996) Le causatif en swahili dans ses relations avec la forme stative et la forme applicative. In J. François (ed.) La sémantique des relations actancielles à travers les langues, Scolia 7. Strasbourg: UMR "LANDISCO" & Université de Strasbourg 2. Comrie, B. (1976) The syntax of causative constructions: Cross-language similarities and divergences. In M. Shibatani (ed.), 261-312. - (1985) "Causative verb formation and other verb-deriving morphology". In T. Shopen (ed.) Language Typology and Syntactic Description. Vol. 3: Grammatical Categories and the Lexicon. London/New York/ Melbourne: Cambridge University Press, 309-348. Comrie, B. & M. Polynsky (eds) (1993) Causatives and Transitivity. Amsterdam/Philadelphia: John Benjamins. Creissels, D. (1995) Éléments de syntaxe générale. Col. "Linguistique nouvelle". Paris: PUF. Desclés, J.-P. (1990) Langages applicatifs, langues naturelles et cognition. Paris: Hermès. - (1997, sous presse) Transitivité sémantique. Transitivité syntaxique. In A. Rousseau (ed.) La Transitivité, Lille: Presses du Septentrion. Givón, T. (1984) Syntax: a functional-typological introduction. Vol. I. Amsterdam/Philadelphia: J. Benjamins. - (1990) Syntax: a functional-typological introduction. Vol. II. Amsterdam/Philadelphia: J. Benjamins. Hetzron, R. (1976) On the Hungarian Causative Verb and Its Syntax. In M. Shibatani (ed.), 371-398. Hopper, P. J. & S. A. Thompson (1980) Transitivity in grammar and discourse. Language 56 (2), 251-299.

Causalité, Causativité, Transitivité

27

Inènlikej, P.I., V. P. Nedjalkov et A. XolodoviC (1969) "Kauzativ ν Cukotskom jazyke". In A. A. XolodoviC (ed.), 260-269. Kibrik, A. A. (1969) Transitivity increase in Athabaskan Languages". In Β. Coitine & M. Polynsky (eds), 4767. Kordi, E. E. (1988) Aíodal'nye i kauzativnye glagoly ν sovremennomfrancuzskomjazyke. Leningrad: Nauka. Kozinsky, I. S, V. P. Nedjalkov & M. S. Polinskaja (1988) Antipassive in Chukchee: oblique object, object incorporation, zero object. In M. Shibatani (ed.) Passive and Voice. Amsterdam/Philadelphia: John Benjamins, 651-706. Kulikov, L. I. (1993) The 'second causative': A typological sketch. In B. Comrie & M. Polynsky (eds), 121-154. Kulikov, L. I. & N. R. Sumbatova (1993) Through the looking-glass, and how causatives look there. In B. Comrie & M. Polynsky (eds), 327-341. Lyons J. (1978/1980) Semantics, Vol. 2. London/New York/Melbourne: Cambridge University Press. [Traduction française (J. Durand & D. Boulonnais) Sémantique linguistique. Paris: Larousse, 1980], Maslova, E. S. (1993) The causative in Yukaghir. In B. Comrie & M. Polynsky (eds), 271-285. McCawley, J. D. (1968) Lexical insertion in a transformational grammar without deep structure, Papers from the fourth regional meeting of the Chicago Linguistic Society, 71-80. - (1976) Remarks on What Can Cause What. In M. Shibatani (ed.), 126-129. Nedjalkov, V. P. et G. G. Sil'nickij (1969a) Tipologija kauzativnyx konstnikcij. In A. A. XolodoviC (ed.), 5-19. - (1969b) Tipologija morfologiCeskogo i leksiöeskogo kauzativov. In A. A. XolodoviC (ed), 20-50. Shibatani, M. (ed) (1976) The Grammar of Causative Constructions. (Syntax and Semantics, 6), New York: Academic Press. Shibatani, M. (1976) The Grammar of Causative Constructions. Conspectus. In M. Shibatani (ed.), 1-40. Talmy, L. (1976) Semantic Causative Types. In M. Shibatani (ed.), 43-116. Tesnière, L. (1966) Eléments de syntaxe structurale. Paris: Klincksieck. XolodoviC, A. A. (ed.) (1969) Tipologija kauzativnyx konstrukcij. Morfologiäeskij kauzativ. Leningrad: Nauka. Zimmer, Κ. E. (1976) Some Constraints on Turkish Causativization. In M. Shibatani (ed.), 399-412.

On Two Parameters of Transitivity* YakovG. Testelec

The purpose of this paper is to investigate the correlation between the approach to transitivity as a bundle of interrelated clausal parameters exemplified in Hopper and Thompson's (1980) work, and the more traditional view on transitivity as a feature of grammatical classification of verbs. As a result, some meaningful distinctive features of verbs and of clauses containing them will be proposed in terms of which some universal tendencies in grammatical classification of verbs can be characterized. I start [1-2] with reconsidering the semantic base for the "core" transitives. Further [3-4] Γ suggest two parameters of transitivity which enable us to compare the degree of transitivity in different classes of syntactic constructions, and to partly predict the contents of grammatical classes of verbs in languages of the world. Evidence will be drawn mainly from languages of the indigenous families of the Caucasus which show remarkable diversity in morphosyntactic classes of verbs.

1. Two approaches to transitivity Grammatical classifications based on formal properties of words with no or little semantic background like, e.g. inflectional classes tend not to coincide across languages. On the other hand, classifications based on meaningful features like animateness, show significant similarity. Classifications of lexical units based on their syntactic behaviour, like parts of speech, seem to overlap across languages to a less degree. The differences found among languages are great, but not random; rather they conform to some universal restrictions. Classifications based on the feature of (in)transitivity belong to this latter type. According to probably the most widespread traditional view, verbs which have a direct object valency are called transitive, those which have not are called intransitive. To avoid theoretical and descriptive problems connected with the notion of direct object, some typologists use semantically based universal definitions. The most elaborate definition I am aware of has been suggested in Kozinsky 1980. To put Kozinskys view shortly, a small semantic class of verbs, viz. verbs of destruction and creation, is assumed to be transitive in its basic voice in all languages. Further, any verb which requires the same constructions) as the verbs in the core class do, may be called transitive. Similarly, another small semantic class of verbs may be taken as the base for definition of intransitives. Therefore the core class of transitives must contain verbs that require two semantic (= thematic) roles which are typical, or "canonical", Agent and Patient; the core class for intransitives must contain veibs which have either a typical Agent, or * I would like to thank L. Kulikov and H. Vater for valuable comments on the draft of this paper. Discussions of the veib classes outlined here with K. Kazenin were also of much benefit. Of course all mistakes and shortcomings are mine.

30

Yakov G. Testelec

a typical Patient valency (resp. "unaccusative" and "unergative" verbs). Comrie (1978: 331) and Dixon (1979: 103) follow basically the same principle: in a given language, the class of verbs which contains verbs with canonical Agent and Patient as its semantic core is called transitive. We can therefore easily find out that, e.g., in English, to like is a transitive verb because it belongs to the same syntactic class as to make, but in Georgian moc'oneba 'to like' behaves differently from, e.g. aSeneba 'to build': (1)

mama-0 0-aáeneb-s saxl-s father-NOM 3SG.OBJ-build-3SG.SBJ house-DAT 'Father builds the house'

(2)

mama-s mo-s-c'on-s saxl-i father-DAT PRV-3SG.SBJ-like-3SG.OBJ house-NOM 'Father likes the house'

Comparing (1) and (2) it can be established that the verb moc'oneba 'to like' differs in its syntactic characteristics from aáeneba 'to build': first, it requires a different case marking pattern of subject and object and, moreover, it employs a different pattern of subject-object agreement. Similarly, to bring in English is transitive; in standard Arabic, however, the same meaning is expressed by intransitive verbs of motion accompanied by a prepositional phrase consisting of an instrumental preposition bi and an object in genitive: (3)

a. 'ata: l-ha:dim-u come.PRF DEF-servant-NOM. SG 'The servant came' b. 'ata: l-ha:dim-u bi finga:n-i qahwat-i-n come.PRF DEF-servant-NOM. SG PREP cup-GEN coffee-GEN-INDEF 'The servant brought a cup of coffee" (lit.: "the servant came with a cup of coffee'; Xrakovskij 1969: 89)

However, the way of defining transitivity based on a small semantic core of transitives does not seem not to be satisfactory for several reasons. One reason is that this approach to transitivity ignores correlations between languages outside the core class. Given the semantic core definition, one might assume that those coincidences are random and unpredictable. However, if we compile representative lists of transitives and intransitives for two or more genetically unrelated languages, we note that several large classes usually tend to coincide: the majority of one-place verbs are intransitive; among twoplace verbs, verbs with clausal objects ('say1, 'begin'), derived and lexical causatives Cboil smth', "bend smth'), verbs of controlled perception ('listen', 'look'), three-place verbs ('give', 'show"), verbs of pursuit ('search', "wait'), verbs denoting operations with property ('sell', 'choose', 'lose',

On Two Parameters of Transitivity

31

'gain') are transitive in most languages. Semantic core approach to transitivity therefore permits more differentiation among languages than it is really observed. Another problem with the semantic core approach is that there are intermediate classes which share both transitive and intransitive features, e.g. "middle" verbs in some West Caucasian (Klimov & Alekseev 1980: 13ff), Kartvelian, or Polynesian (Chung 1978: 48) languages. Chung characterises the "middle" class in Polynesian in terms that fit in altogether with its Caucasian analogues: "middle verbs describe events that do not affect the direct object immediately. Included among the middle verbs in most Polynesian languages are perception verbs Osee', 'listen to'), verbs of emotion and other psychological states Clove', 'want', 'understand'), verbs normally selecting animate direct objects, including some communication verbs ('meet with', 'help1, 'call'), and verbs such as 'follow1, 'wait for1 and 'visit' " (Chung 1978: 47).

The semantic core approach cannot capture the tripartite distinction, although the latter is also rather semantically consistent across languages. However, it is hard to establish any homogeneous semantic core of middles which would underlie that class in all tripartite systems. Therefore, it is not clear how a universal semantically based definition of middles might be designed. By contrast to the traditional notion, Hopper and Thompson (1980) regard transitivity as a multifactored notion, consisting of several parameters "whose aggregate resulted in the cardinal transitivity of a clause... we predicted that the grammatical or semantic markings of transitivity would covary in the clause in the same direction with respect to cardinal transitivity" (Hopper & Thompson 1982: 2).

The list of parameters includes: 1) (number of) Participants; 2) Kinesis (action vs. nonaction); 3) Aspect; 4) Punctuality; 5) Volitionality; 6) Affirmation; 7) Mode; 8) Agency; 9) Affectedness of O(bject); 10) Individuation of 0(bject). Hopper and Thompson hypothesize that whenever an obligatory pairing of two transitivity features occurs in the morphosyntax or semantics of a clause, the paired features are always on the same side of the high-low transitivity scale. For instance, if a highly individuated participant is marked with accusative marker only if it is the direct object, both features ('two participants' and Ό highly individuated') signal the high degree of transitivity. All or most non-volitional verbs denote non-actions, thus indicating the low degree of transitivity etc. Hopper and Thompson's parameters and their general discourse function ("backgrounding" vs. "foregrounding"), as well as some improvements suggested in Tsunoda 1985 deserve a more detailed discussion than I can afford here. However, it seems obvious that morphosyntactic classes of constructions in most languages do not seem to represent anything like a transitivity continuum with two polar points, as one might expect if Hopper and Thompson's hypothesis were true. What we really observe are rather scalar systems that include rarely more than several most frequent patterns of predicative constructions.

32

Yakov G. Testelec

Note that the advantages and the shortcomings of both approaches are complementary. The semantic core approach can account for the fact that observed grammatical classes of verbs are so large and heterogeneous: it is because most verbs assimilate their syntactic behaviour to one of the two semantic core classes. However, great similarities beyond the core classes cannot be accounted for. Hopper and Thompson's muhifactored approach, on the contrary, accounts for many crosslinguistic similarities but cannot answer the question why the resulting grammatical classes are so large. Below I will try to propose two parameters of transitivity, a semantic and a syntactic one, which seem to underlie both tendencies.

2. What is a "Patient"? Let us consider now some role properties of subjects and object which the semantic parameter has to generalize. Verbs are sometimes called "semantically transitive" if they have both Agent and Patient arguments. Semantic transitivity does not depend on the syntactic categories by which it is represented, e.g. nominalized NPs (e.g. his attack) may be semantically as transitive as a finite clause. Much work has been done to characterize the role of Agent explicitly; one can mention Lakoff 1977 and Polinskajal986, among others. Polinskaja puts it in the following way: "a participant which has the right (or ability) of a free choice to perform the beginning and/or the end of a given state of affairs" (Polinskaja 1986: 5). By contrast, I am aware of no convincing semantic definition of the role of Patient, i.e. of the most affected argument of a verb. Kibrik (1980) defines Patient as a "participant involved in a situation which it neither controls nor fulfils"; all non-Agent participants, however, conform this definition as well. Moreover, it seems reasonable to spread the notion of Patient also on some "affected" Agents (Saksena 1980), or, in other terms "Agent-Patients", such as arguments of verbs like 'swim' or 'dance', which display different characteristics from "non-affected" Agents. Patient is better defined in positive terms when it is regarded not as an element in the set of discrete roles but rather as a cluster of (sometimes mutually exclusive) prototypical characteristics opposed to a corresponding cluster of Agent's characteristics, e. g. Undergoer vs. Actor in Foley & Van Valin 1984, or Proto-Patient vs. Proto-Agent in Dowty 1991. Perhaps the most explicit view on Patient was presented in the early classical work by Fillmore (1968). Fillmore defined his Objective', i.e. Patient, as a semantic case which denotes the role in an action or in a state identified by the verb itself by its semantics. In other words, Agent, or Instrument, or Benefactive are semantic roles which are the same or similar with different verb predicates, whereas Patient semantics cannot be generalized but is rather a role installed individually by every particular verb. Therefore, the more verb predicates we find with a given semantic relation to the verb, the less "Patient-like" is this relation. Patient cannot therefore obtain any general semantic characteristics in positive terms; rather it may be defined negatively as a semantic role which is peculiar to a given verb and not shared by any significant number of other verbs.

On Two Parameters of Transitivity

33

Consider now some examples. Which arguments, intuitively, have the least "degree of patiency"? Most probably those with the semantic role of Agent, because they are semantically identical or close to each other and occur with the majority of verb predicates. Within the class of Agents, there are more and less "Patient-like" participants. The purest case of an Agent with no characteristics of a Patient is probably that participant of many-place predicates which is linked to them via the causative relation and bear no other relation of a more specific kind. The causative relation is sometimes supposed to be maybe the simplest one between two predicates (Mel'ëuk 1991: 53. If a causative Agent is employed, the first situation, the causing one, is left unexpressed (Nedjalkov & Sil'nickij 1969a: 7), but it is usually recoverable from the context. The most standard roles are Agents of causative situations, of which the situations that they cause can easily be separated. Agents of this sort occur with all causative verbs of spatial manipulation with objects which can move (or stay) independently from any Agent: 'move', 'lead', 'gather1, 'drive out', 'sit', 'he', 'open', 'leave'; verbs that denote switching on of processes which can, their resource being enough, go on independently: "burn', 'thaw1, 'fill', "boil' etc.; verbs of destruction denoting processes that can take place without Agent: 'put out (a fire)', 'switch off (the light)', 'break' etc. Another class of Agents has a more intricate semantic relationship with the verb predicate: the caused situation cannot be separated from its Agent, because the Agent referent participates in two or more elementary situations simultaneously. Here we find verbs of property relations: 'X buys' means approximately the conjunction of the following statements: (4)

a. b. c. d. e.

'X causes X to possess Y ; 'Z has possessed Y'; 'Z ceases to possess Y'; 'X causes Ζ to possess M'; 'M is money or another exchange equivalent for Y1.

Of the five elementary situations of which the situation of buying consists, only (4d ) can be decausativized: 'X gives W to Ζ 'Ζ has W'; although (4a) is a causative situation, the Agent is co-referent with the possessor, therefore the referent of X cannot be eliminated from the situation via decausativizing. Being an Agent by definition, the participant in (4) is still more verb-specific than the causative Agents mentioned above. Under "verb-specific" I mean that is not easy to find many verbs which have the same or a similar bundle of semantic relationships, that the referent of this Agent has with the verb 'buy*. Still there are some: 'choose', 'exchange', 'gain', 'put on' also have a reflexive proposition like (4a) as part of their meaning. Many agentive two- and three-place verbs can be organized in similar small groups with one or several semantic "additions" to the basic causative function. Such additions can involve instrumental manipulations ('beat', 'cut', 'scratch', 'wave', 'saw1, 'write'), spatial orientations of the action ('push', 'pull', 'throw"), or stages of which the action consists: ('tie', 'dig', 'thresh', 'spin', 'count'), etc. Cf. also the argument that to kill does not mean 'cause to die' (Fodor 1970, Wierzbicka 1980) but includes an additional part of meaning, viz. that the situation takes place at

34

Yakov G. Testelec

one single time and space. Hie difference between contact and distant causation also belong here: the former implies that Agent is more involved in the situation (Nedjalkov & Sil'nickij 1969b). Various kinds of objects can also differ in the degree of their patiency. Since Fillmore 1968 the basic opposition is stated between "Objective", or Patient, and other roles usually performed by syntactic objects, such as "Dative" (an animate participant affected by the action or state); "Factitive" (the result of the action or state: to become a doctor etc.), or Instrumental (an inanimate force or object causing the action or state). All those roles are, in general, semantically homogeneous and, given their standard formal devices (cases, adpositions etc.), one can easily predict which is the semantic relation that an argument marked with those devices bears with respect to the verbal predicate. Again, whereas semantic roles of oblique objects can be characterized as standard and defined in positive terms, "real" Patients are semantically nonstandard. The scale of patiency for objects was suggested by Fillmore (1977); he called it "saliency hierarchy". It included some ranking features like 'animate' vs. 'inanimate', 'changed' vs. 'unchanged', \vhole' vs. 'part', 'figure' vs. 'ground' etc. The feature of temporal change seems to be the most important: a participant is affected in a situation if it undergoes some relevant change in it, i.e. its physical, or spatial, or mental etc. characteristics are not the same as they would be if the situation did not take place. The change may be telic ('reach', "burn'), or cyclic ('dance', 'roll'). For some semantically related verbs it is possible to determine which of them has a more and which has a less affected object, cf. 'encounter1, 'attack', and 'defeat'. The feature of total vs. partial involvement is also significant, cf. "beat' and "hit": in many languages, the former is a canonical transitive whereas the latter is not. At first sight, any attempt to suggest a transitivity scale for two-place predicates has no prospect because of too many features involved in the choice of (in)transitivity marking devices and the intricacy of their relationships. Tsunoda (1985), however, did suggest a plausible, although somewhat rough, classification of two-place predicates. It really embraces many relevant cases (Tsunoda 1985: 388):

35

Ort Two Parameters of Transitivity

Table 1 1 Direct effect on Patient resultative: non-resultative: •kill1 •break1 •bend'

Knowledge 'know1 'understand' 'remember1 'forget'

•hit' 'shoot' •kick' 'eat'

Perception Patient more Patient less atattained: tained: •hear1 'find'

5 Feeling 'love' 'like* 'fond' 'afraid' 'angry*

'look'

Relationship 'possess* •have' 'lack' 'resemble' 'consist'

3 Pursuit

'search' •wait'

7 Ability 'capable' 'profiscient' 'good'

Tsunoda claims that "as we go down the scale, transitivity case frames are less likely to occur, and we tend to have some other case frames in addition to, or in place o£ them" (Tsunoda 1985: 390). The degree of affectedness of objects, presented in this or in a similar scale, obviously correlates with the feature of standardness: the more affected (Patient-like) is the object, the less standard is its semantic relationship with the verb.

3. Toward a semantic classification of two-place predicates We dispose therefore of at least two featurs of semantic transitivity outlined above: control and affectedness. The notion of affectedness, however, needs more clarification, especially in cases where more than one participant is affected. The feature of control may also have the positive value for more than one participant. The most obvious case is probably the causee role with causatives. The causee controls, however, only a part of the situation, whereas the causative Agent controls the situation as a whole. The same holds also for verbs like *beg', 'ask', 'threaten', 'flirt with', 'challenge' etc. which imply a possible reaction of a definite sort performed and controlled by the object participant. It should be emphasized here that the subject participant does not control that kind of reaction; cf. to the contrary 'frighten', 'make angry*, 'discharge' etc. where the object's reaction is under its control.2 2

An attempt to outline a formal model of Agent's control using structural domains in the semantic form see in (Wunderlich 1997).

36

ïakov G. Testelec

With all grammatical and lexical causatives, the distribution of both features is uniform: the causative Agent is [+control, -affected], whereas the causee is either [-control, +affected] or [-control, -affected]; the latter case occurs with trivalent causatives, where there is always another Patient participant marked with [-control, +affected] values. Saksena (1980) illustrates the formal difference between affected and non-affected causees in the following Hindi causative constructions (Saksena 1980: 812-813): (5)

a. raam-nee khaanaa khaa-yaa Ram-AG food eat-PAST 'Ram ate dinner1 b. màî-nee raam-koo/-*see I-AG Ram-ACC/-*INSTR Ί fed Ram'

(6)

khanaa khib-aa-yaa food eat-CAUS-PAST

a. raam-nee peer kaat-aa Ram-AG tree cut-PAST 'Ram cut the tree' b. mâï-nee raam-see/-*koo I-AG Ram-INSTR/-*ACC Ί made Ram cut the tree'

peer kaat-aa-yaa tree cut-CAUS-PAST

In (5b) the causee is affected, in (6b) it is not; this contrast determines the difference in case marking of the causee. Let us turn now to verbs that have lexical meanings other that semantic causatives, i.e. those that have no non-causative counterparts. Matrix verbs that require sentential arguments will also be excluded. Consider Table 2 where under 1 and 2 the participants of the verb predicate are listed, X>Y denotes that X has a higher degree of a given property than Y.

37

On Two Parameters of Transitivity

Table 2 CLASS

CONTROL

AFFECTEDNESS

NAME OF THE

EXAMPLES

I

1,2

none

II

1,2

1,2

symmetric non-Patient verbs symmetric Patient verbs solicitation verbs

'ask', 'threaten'

m

1>2

none

verbs of pursuit

'follow1, 'meet'

IV

1

2

verbs of reaction semantic transitives (with many subclasses, see below)

'agree', 'resist' 'make', 'write'

V

1

2>1

'pull', 'take'

VI

1

none or 1

inherent reflexive transitives semantic middles

VII

none

1,2

non-controlled processes

VIII

none

1

non-controlled relations

(SUB)CLASS

'speak to' 'fight with'

'praise', 'look', 'harm', 'search', 'enter1 'stick to' 'see', 'depend'

Class I verbs differ from Class II in that participants of the former do not undergo relevant changes in situations they are involved in. Like all symmetric verbs, they allow plural subjects. However, it is possible to discriminate between the two referents with the subject-object asymmetry. Class I includes verbs like 'play with', 'speak with', class II includes 'fight with', 'quarrel with' and the like. Class III consists of at least three groups. Solicitation verbs denote a controlled action which is an attempt to cause a human object to react: "beg', 'ask', 'threaten', 'solicit'. Verbs of pursuit include meanings like 'follow1, 'meet', 'pursue', 'ambush'. They imply that the object participant also fulfils some kind of controllable movement, although the main component of the action is beyond the scope of this control. The third group includes 'agree', 'resist', 'stand', 'help' and other kinds of human reactions. Here the object participant also controls a part of the situation, but in this case it is a situation which causes and temporally precedes the beginning of some other situation: X agrees with ¥ implies that V has made a proposition, X helps Y means that Y has started to act so that it has become possible for X to share the activity etc. First three classes belong entirely to middle verbs, e. g., in Kartvelian languages. Middle verbs differ from intransitives in that they have an obligatory indirect object valency in dative case which is cross-referenced in the verb by a special paradigm of agreement. On the other hand, they differ from transitives in that their subjects are marked with nominative case, and

38

Yakov G. Testelec

not with ergative/nominative/dative (the choice depending on the tense/aspect form of the verb predicate), as the transitive subjects are. Examples from Georgian: (7)

a. mama-0

áe-0-e-k'i-tx-a

giorgi-s

father-NOM PRV-3:OBJ-REL-ask-3:SG:AOR George-DAT 'Father asked George' b. mama-m mo-0-iqVa-na father-ERG PRV-3:OBJ-bring-3:SG:AOR 'Father brought George'

giorgi-0 George-NOM

Most Georgian middles contain with a so called "relative passive" prefix e- (marked as REL in glosses) denoting the indirect object valency: elap'arak'eba 'speak to', etamaseba 'play with', ebrjoleba 'struggle with', ejßbreba 'compete with', edaveba 'argue with', evedreba "beg1, emukreba 'threaten', ep'ranâeba 'flirt' etc. Class IV (semantic transitives) is the most numerous one and consists of many subclasses. We label each subclass with a typical representative in capitals: MAKE, DESTROY, DIG, SHAVE, BITE, SAY, HANG, BEND, THROW, WAVE, SWIM, BLESS, BOIL, DRAW, and, maybe, some others. They are transitive in most languages; BITE subclass, however, is often middle. MAKE verbs denote: 'X acts in order to cause Ζ to exist': 'make', ΐββτ", many verbs denoting production processes: 'write (a letter)', 'knit (a sock)', 'sew (a dress)', 'issue' etc. DESTROY verbs: 'X acts in order to cause Ζ to cease its existence': 'slaughter1, 'kill', 'melt', 'sink', 'rip', 'spend', 'put out (a fire)'. DIG subclass is very similar to MAKE, it also consists of telic verbs; the actions usually consist of several phases, and instruments are often needed to fulfil them DIG verbs imply that objects undergo serious changes: 'dig', 'chew1, 'cut', 'saw1, 'sharpen', 'reap', 'mow1, 'sift', 'mill' etc. Verbs of consumption like 'eat' and 'drink' belong here but, through a metaphorical shift, they may pattern with verbs of perception like 'see'. If a language has causative forms for intransitives only, those verbs may have causative forms despite their grammatical transitivity (Nedjalkov & Sil'nickij 1969b: 34-35). For example in Vedic Sanskrit verbs like pä 'drink', jus 'enjoy1, sru 'hear', unlike most transitives, have causative counterparts (cf. Kulikov 1989 where such verbs are called "quasi-intransitives"). SHAVE and BITE subclasses both denote actions which affect only a part of the object. BITE verbs are often middle, e. g. in many Caucasian languages, but SHAVE verbs are usually transitive. The latter is so probably because the action affects the front surface of the object, which is crucial for its recognition. Verbs denoting that action pattern therefore with cases of total involvement. BITE: 'lick', 'kiss', 'scratch', 'caress', 'touch', 'shove', 'pinch', "hit". In Kartvelian languages this subclass is middle. In Nakh-Dagestanian (East Caucasian) languages, however, Instruments are direct objects (often omitted). Cf. in Avar (Klimov & Alekseev 1980, p. 184):

39

On Two Parameters of Transitivity

(8)

di-c:a dos-da q'ab-una I-ERG he-LOC hit-AOR Ί hit him (with a stick)'

(t'il-0) stick-ABS

SHAVE class includes: 'wash', 'clean1, 'polish', 'cut (hair)', 'comb', 'paint' etc. SAY includes twoand three-place predicates with mental and textual objects: 'say1, think', 'whistle (a tune)', "believe', 'tell', 'ask smth', 'promise' etc. THROW and WAVE classes both denote spatial movements, the former consists of irreversible telic verbs, the latter denotes cyclic movements. THROW: 'shoot', "throw1. WAVE: "wave", 'turn', 'rock', 'knock', "beat'. SWIM class consists of verbs of movement or duration with sections of way or time as objects: 'walk (a mile)', 'swim (a river)', 'cross (a street)', 'live (two years)'. The object's involvement is not a real involvement, however, although its position relative to the point of view does change (a metaphor of "consumption"). BLESS verbs denote mystical or magical influences on the object: "bless', 'curse', 'damn', as well as verbs that denote assignment of a verbal characteristic: 'blame', 'name', 'call'. DRAW subclass consists of verbs denoting mental, e.g. intellectual, operations with objects, or physical operations accompanied with mental ones: 'draw1, 'measure', 'read', 'investigate' etc. In spite of that causative verbs are excluded from this classification, three sets of causatives deserve mention here that can be called HANG, BOIL and BEND subclasses. HANG subclass denotes reversible actions which imply that the location of the object, or of its part(s), is changed by the Agent. Many of them are paired, each pair denoting two possible reversible states of the object:, 'put', 'lift up', 'put on', 'take off, 'lock', 'rub into', 'squeeze' etc. There are semantic counterparts denoting spontaneous physical movements like 'shut' and 'open'. BOIL verbs denote situations which, if caused by Agents, can go on without their participation as long as the resourse is present. This class includes heat', 'raise', 'grow1, toil', 'fry1, "burn1, 'cool', 'pour1, 'drop' etc. BEND verbs express movements of the parts of an object rather than its movement relative to other objects: "bend", 'spread out', "break', 'repair1, 'roll up', 'fold', '(un)tie', '(un)zip' etc. Auxiliary or semi-auxiliary transitive verbs like 'do' and 'let' do not fit any subclass and are to be treated separately. Class V (inherent reflexive transitives) implies that the subject referent has at least one more role function in the situation. Two subclasses can be distinguished: PULL and TAKE. PULL implies that the object is directed toward or from the Agent in a special way or is situated at a particular place of it or near it: "pull', "swallow1, "bring', "hold1, 'drag' etc. TAKE verbs imply some change in the property relations of the Agent, see (4) above: 'chose', take', 'gain', 'lose', 'give', 'sell', 'buy', 'exchange', 'catch', 'make a present' etc. VI class consists of semantically middle verbs, i.e. of verbs with a "normal" Agent but an unaffected object which undergoes no relevant changes. They differ in various respects: verbs of controlled perception ('look", listen"), 'search' (maybe it is closer to the TAKE subclass), 'enter1, "harm", "praise" (if not of the BLESS type). VII class includes verbs of non-controlled processes like "stick to', 'reflect', 'infect', "undergo'.

40

Yakov G. Testelec

V m class includes verbs of non-controlled relationships, viz. verbs of perception 'see', •hear*, 'smell'; mental predicates: "want', 'know1, 'forget', 'remember', "be proud of etc.; abstract stable relations: 'depend', 'consist of, 'capable of. If we try to correlate all classes of two-place predicates with linguistic data we see that, as it was pointed out above, verbs with the most standard, i.e. causative, Agents are always transitive. Two-place verbs with Agent participants may be middle or even intransitive. It occurs if the second participant is non-affected (I, ΠΙ, VI). It is possible also if both participants are equally affected (Π), or if the Agent is partly affected (V), cf. Georgian PULL subclass (spatial transitives with inherent reflexive) which is middle: (9)

biò'-i mo-0-e-zid-a mor-eb-s boy-NOM PRV-3 : OB J-REL-drag-3 : SG: SB J: AOR log-PL-DAT "The boy has dragged logs'

(10) mc'q'ems-i mi-0-e-rek'-a shepherd-NOM PRV-3:OBJ-REL-send-3:SG:SUBJ:AOR 'The shepherd sent the flock to grass'

para-s flock-DAT

Non-controlled states and relationships are usually denoted by intransitives (VII, VIH). Vin class representatives, especially perception verbs, are transitive in many languages, however (e.g. in Russian).

4. Two parameters of transitivity From data presented above it can be seen that semantic roles can be subdivided into two classes: standard roles that are definable in positive semantic terms and identical with many verbs (Agent, Benefactive, Instrument, Recipient, Stimulus and some others), and the only non-standard role (Patient) which cannot be characterized by a set of necessary and sufficient positive semantic features but plays different roles with all or most of the verbal predicates. Standard roles like Agent may also be affected, i.e. bear some Patient-like characteristics. Verbs denoting situations that involve both a non-affected Agent and a Patient belong to the class of semantic transitives. One-place verbs that denote situations with Patients ('lie', 'stay1) or with Agent-Patients, or, in other terms, "affected Agents" ('work', 'dance'), are semantic intransitives. Verbs with two or more participants that have more than one affected or more than one controlling participant, are semantic middles. In all languages, semantic transitives belong to the class of grammatical transitives, semantic intransitives belong to grammatical intransitives. Semantic middles may be distributed between the two classes, or at least a part of them may form a separate grammatical class. The existence of large grammatical classes can be then easily explained: in every language lexical meanings of verbs which include only one obligatory participant, or those which have two participants, one being non-affected, and another affected, are an overwhelming majority. Indeed, most verbs

On Two Parameters of Transitivity

41

denote either one-place actions and states or human activities that include controllable operations with a non-controlling Patient. The distribution of semantic middles, or of some of them, between the two large grammatical classes, is not random. Since the degree of control and affectedness, as we have seen, is not a binary feature, a scale for semantic transitivity can be proposed. We suggest therefore the parameter of semantic transitivity: (11) The less similarity is there between the two major participants of the predicate in terms of control and affectedness, the more semantically transitive is the verb. In a given language, if a verb Vi belongs to the grammatical class of transitives, then any verb V2 which is more semantically transitive than Vi belongs to the same class. Conversely, if a verb Vi belongs to the grammatical class of intransitives, then any verb V2 which is less semantically transitive than Vi belongs to the same class. To illustrate ( 11), let us consider a large class of middle verbs in some Caucasian languages. The class is semantically heterogeneous, and the only feature common to all its members is that they all show less dissimilarity between the pair of participants than any verb belonging to the transitive class. Semantic groups of verbs that constitute grammatical middle classes in Caucasian languages include either two affected ('fight with', class Π), or two non-affected ("talk to', class I), or two controlling ('wait for1, class ΙΠ), or two non-controlling Cbe similar to', class Vm) main participants, thus showing low degree of dissimilarity of them both. For instance, middle verbs in Abkhaz-Adyghe (West Caucasian) languages have two arguments, which can be labelled with vague terms like "actor" and "recipient". Actors trigger the absohxtive, and not the ergative, agreement pattern, i. e. cross-reference morphemes in the verb, referring to them are the same as the only arguments of intransitive verbs. In those languages of the family that have cases, middle verbs have absohitive subject — ergative object construction, whereas transitives employ ergative subject and transitive object case marking. Recipients trigger the same pattern as they do with bitransitives like 'give' or 'show1. Cf in Kabardían (Kumaxov 1971: 194): (12) a. sa-wa-px-aá 1. SG-2.PL-look-PAST Ί was looking at you' (middle) b. wa-s-H-aS 2.PL-l.SG-bring-PAST Ί was bringing you1 (transitive) Personal markers are the same (phonetic changes apart), but their position is different: actorrecipient order in middles, object-subject order in transitives.

42

Yakov G. Testelec

Actors are genuine Agents in most cases; however, Patient characteristics are distributed between actors and recipients, thus displaying little dissimilarity in the degree of afiectedness. These cases fit in with the classification outlined above: ΒΓΓΕ subclass of semantic transitives: Abaza asra "hit1, mc'asra 'catch', chara Trite'; Kabardian jep'esk"m 'pinch', jep'estqm 'scratch', jebzejm 'lick', s'efsn 'suck', jefrq'en "bite1. Verbs of pursuit and solicitation: Abaza apqsra 'outrun', azpsra 'wait for1, aStagdara 'follow1; Kabardian jejen 'call', jexé°m 'ask'. Inherent reflexive transitives: Abaza c'ag°ara 'push', aqra 'pull' (data from: Klimov & Alekseev 1980: 13fi). Similar classes of middle verbs are found also in Svan (a Kartvelian language of Northwestern Georgia). Middles are subdivided in Svan into those with ergative-dative and nominative-dative case marking. BITE, BLESS, solicitation and controlled perception verbs have ergative-dative marking: xogik'i 'kick', xat'q'ci "beat1, xap'anc'kSve 'pinch', xomurjwi "help', xoywnäri 'listen', xecwa.l 'curse', xep'rebal 'praise': (13) ö'q'int'-d kot-silk'e-0 daòwir-s boy-ERG PRV:3:OBJ-push-3:SG:SUBJ sister-DAT "The boy pushed his sister* Verbs of non-controlled relations have nominative-dative marking: xemtkwi 'get used to', xecmi 'stick to', xencsde 'mix with': (14) zey-0 sgalo-x-emtkwe-0 ö'q'int'-s dog-NOM PRV-3 : OB J-get used-3 : SG: AOR boy-DAT "The dog got accustomed to the boy* Besides the features of semantic and grammatical (in)transitivity that characterizes verb lexemes, we need also the notion of syntactic (in)transitivity that characterizes clauses, and not verbs that are contained in them. Syntactic (in)transitivity is a feature which can determine, or subsume, at least some of the transitivity parameters definable at clause level and suggested by Hopper and Thompson. Again, the following scale of transitivity can be proposed: (15) Of two major NPs in a given clause, the more syntactic priority has that NP which is less affected, the more transitive is the clause. In other words, the NP that denotes the less affected participant has to be the subject, the NP denoting the more affected participant has to be the object in a transitive clause. By this definition, passive clauses are intransitive. The set of functional properties of subjecthood, i.e. the ability to control the form and/or position of other constituents, is language-specific, although taken basically from the universal set, which has been Usted by typologists (Keenan 1976, Kozinskij 1983, Mel'òuk 1988, among others). If the syntactic priority is given to the most affected participant, the construction may be called syntactically intransitive. In many cases, either the syntactic or afiectedness priority of one of the two participants can hardly be established. Those constructions can be called syn-

43

On Two Parameters of Transitivity

tactic middles. For example, in Svan, verbs of non-controlled perceptions and emotions have the dative-recipient and nominative-stimulus case frame. Since both roles are standard enough for this kind of verbs, it is not easy to decide what of the two is more affected: xalät' 'love', xat'k'leb "be greedy1, xak'u "want', xoxal "know1, xe.ádni 'forget', xesmi "hear1, xazäx "be proud' etc. Since the site of the. mental process is the recipient, it may be regarded as a participant which is more affected than the stimulus. The recipient NP controls the only functional subjecthood parameter relevant for Svan, Le. the binding of reflexives; therefore the construction may be labelled as a syntactically intransitive: (16) a. wano-s miôatxwim-0 John-DAT himself-NOM 'John loves himself b. *wano-0 John-NOM

x-alät'-0 3:SUBJ-love-3:OBJ:SG

micatxwim-s x-alät-0 himself-DAT 3:SUBJ-love-3:OBJ:SUBJ

In Nakh-Dagestanian languages, verbs that have the same, i.e. stimulus-recipient semantic role frame, show no or little contrast in subject properties of both participants. Cf data from Kryz language (Kibrik 1981: 30-31): (17) a. gada-r ugug îakaji boy-ERG himselfNOM push.AOR b. *gada ugug-ur Takaji boy.NOM himself-ERG push.AOR 'The boy pushed himself (18) a. gada ugug-us irqaji boy.NOM himself-DAT see.AOR b. gada-s ugug irqaji boy-DAT himselfNOM see.AOR 'The boy saw himself The object NP in (17b) cannot bind the subject reflexive whereas both NPs in (18) can bind each other. Therefore verbs with the stimulus-recipient frame not only have a weak semantic contrast in affectedness between both participants but require a middle syntactically construction.

Yakov G. Testeìec

44 5. Conclusion

To sum up, there are semantic correlates of transitivity: control and affectedness which, if applied to data of various languages, give satisfactory results close to formal classifications "transitive" vs. "intransitive" and "transitive" vs. "middle" vs. "intransitives". Although languages show great differences in distribution of verbs among those grammatical classes, this distribution conforms to a semantic parameter which restricts possible variation. Formal devices which express transitivity are sensitive not only to lexical properties of verbs but also to many other pragmatic, referential etc. properties of verbs, NPs and clauses that contain them It is enough to mention Yukuha language (Pama-Nyungan family, Australia), where transitivity marker is sensitive to direct/inverse combinations of personal pronouns, mood, negation and tense (Keen 1983: 236). Again, crosslinguistic variation in syntactic properties that correlate with grammatical and semantic transitivity of verbs, is restricted by a syntactic parameter that proposes a correlation between the degree of participants' affectedness and transitivity of the containing clause.

References Chung, S. (1978) Case Marking & Grammatical Relations in Polynesian. - Austin & London: University of Texas Press. Comrie, B. (1978) Ergativity. In: W.P. Lehmann (ed.) Syntactic Typology. Studies in the Phenomenology of Language, 329-394. Austin & London: University of Texas Press. Dixon, R.M.W. (1979) Ergativity. Language 55, 59-138. Dowty, D.R. (1991) Thematic proto-roles and argument selection. Language 67, 547-619. Fillmore, Ch. J. (1968) The case for case. In: E. Bach, R. Harms (eds.) Universals in Linguistic Theory, 1-89. New York: Holt. - (1977) Topics in lexical semantics. In: R.W. Cole (ed.) Current Issues in Linguistic Theory, 76-138. Bloomington & London: Indiana University Press. Fodor, J. A. (1970) Three reasons for not deriving 'kill· from 'cause to die' In: Linguistic Inquiry 1: 429-438. Foley, W.A., Van Valin R.D. jr. (1984) Functional syntax and universal grammar. Cambridge: Cambridge University Press. Hopper, P. J., Thompson, S. A. (1980) Transitivity in grammar and discourse. In: Language 56, 251-299. Hopper, P. J., Thompson, S. A. (eds.) (1982) Studies in Transitivity. New York: Academic Press. Keen, S. (1983) Yukulta. In: R.M.W. Dixon (ed.) Handbook ofAustralian Languages 3, 191-304 Canberra & Amsterdam: John Benjamins B.V. Keenan, E.L. (1976) Toward a universal definition of "subject". In: Ch. Ν. Li (ed.) Subject and Topic, 305333. New York: Academic Press. Kibrik, A. E. (1981): Materialy k tipologii ergativnosti. 17-20. In: Institut russkogo jazyka. Problemnaja gruppa po äcsperimental'noj i prikladnoj lingvistike. Predvaritel'nye publikacii, 141. Moskva: Institut russkogo jazyka. - (1990): Paciens. In: V. N. Jarceva (ed.): Lingvistiâeskij énciklopediâeskij slovar', 369. Moskva: Sovetskaja ènciklopedija. Klimov, G.A., Alekseev, M.E. (1986) Tipologija kavkazskixjazykov. Moskva: Nauka. Kozinskij, I.S. (1980) Nekotorye grommatiöeskie universalii ν podsistemax vyraienija sub'ektno-ob'ektnyx otnoienij. Avtoreferat kandidatskoj dissertacii. Moskva: Izdatel'stvo MGU. Kozinskij, I.S. (1983) O kategorii "podleiaSóee" ν russkom jazyke. - Institut russkogo jazyka. Problemnaja gruppa po äcsperimental'noj i prikladnoj lingvistike. Predvaritel'nye publikacii, 156. Moskva: Institut russkogo jazyka

Ort Two Parameters of Transitivity

45

Kulikov, L.I. (1989) Κ tipologii tranzitivnosti (o nekotoryx ograniíenijax na obrazovanie kauzativnyx glagolov ν vedijskom sanskrite). In: V.l. Podlesskaja, L.I. Kulikov (eds.) Problemy semantiäeskoj i sintaksiäeskoj tipologii. Moskva: Nauka, 31-51. Kumaxov, M. A. (1971) Slovoizmenenie adygskix jazykov. Moskva: Nauka. Lakoff, G. (1977) Linguistic Gestalts. Chicago Linguistic Society. Papers from the 13th regional meeting., 236-287. Chicago: Chicago University Press. Mel'Cuk, I. A. (1988) Dependency Syntax: Theory and Practice. Albany, New York: State University of New York Press. Mel'òuk, I. A. (1991) ESöe raz k voprosu ob èrgativnoj konstrukcii. Voprosy jazykoznanija, 1991, 4,46-88. Nedjalkov, V. P., Sil'mckij, G.G. (1969a) Tipologija kauzativnyx konstrukcij. In: A.A. Xolodovi£ (ed.) Tipologija kauzativnyx konstrukcij: morfologióeskij kauzativ, 5-19. Leningrad: Nauka. Nedjalkov, V.P., Sil'mckij, Georgij G. (1969b) Tipologija morfologiöeskogo i leksióeskogo kauzativa. - In: A.A. Xolodovií (ed.) Tipologija kauzativnyx konstrukcij: morfologióeskij kauzativ, 20-50. Leningrad: Nauka. Polinskaja, M. S. (1986) Diffuznye glagoly ν sintaksise érgativnyx jazykov. Avtoreferat kandidatskoj dissertaci). Moskva: Institut jazykoznanija AN SSSR. Saksena, A. (1980) The affected agent. Language 56, 812-825. Tsunoda, T. (1985) Remarks on Transitivity. Journal of Linguistics, 21, 385-396. Wierzbicka, A. (1975) Why 'kill' does not mean 'cause to die' - the semantics of action sentences. Foundations of language 13, 491-528. Wunderlich, D. (1997) Cause and the structure of verbs. Linguistic Inquiry 28, 27-68. Xrakovskij, V. S. (1969) Morfologiieskij i analitiëeskij kauzativy ν literaturnom arabskom jazyke. In: A.A. Xolodoviö (ed.) Tipologija kauzativnyx konstrukcij: morfologióeskij kauzativ, 78-97. Leningrad: Nauka.

Transitivity Increasing Operations in Tariana Alexandra Y. Aikhenvald

This paper is offered to Professor Vladimir P. Nedialkov in homage, and in appreciation of his manifold contributions to our understanding of the nature of language.

1. Introduction I investigate here different types of causative construction in Tariana, a North Arawak language from North West Amazonia. There is a morphological causative typically used with intransitive verbs, a serial causative construction, and two types of periphrastic causative. All these causative constructions are functionally and formally distinct, and show intricate interrelations as to the role of the causer, that of the causee, and the properties of predicate classes. The use of the morphological causative with transitive verbs correlates with the promotion of an oblique argument into the core. Thus, the causative morpheme, when added to a transitive verb, marks a kind of argument-adding derivation. The typological characteristics of Tariana are described in §2. The four different kinds of causative constructions are analyzed in §§3-5. Morphological causatives are considered in §3; serial causative constructions in §4, and periphrastic causatives in §5. These distinct causative mechanisms are compared in §6. The argument-adding derivation used with transitive verbs is discussed in §7. Transitivity increasing operations in Tariana are viewed in a typological perspective in §8. I use standard abbreviations for syntactic functions (Dixon 1994): A for transitive subject, O for transitive object and S for intransitive subject. In addition, S a is used for the subject of an active intransitive verb, and So for the subject of a stative intransitive verb, in Tariana's split-S system.

2. Typological characteristics of Tariana Typologically Tariana1 is predominantly head-marking with a few elements of dependentmarking.

1

Tariana is a North Arawak language spoken by around 100 people in the region of the river Vaupes, Upper Rio Negro (Brazil). It is the only Arawak language spoken in the multilingual context of the Vaupes linguistic area (see Sorensen 1967, Aikhenvald 1996), and it has suffered a heavy areal impact from Tucanoan languages. I am extremely grateful to my teachers of Tariana - Cándido, Olivia, Graciliano, Jovino and José, Brito. I owe thanks to R.M. W. Dixon for useful discussion and comment.

48

Alexandra Y. Aikhenvald

Tariana combines elements of morphological ergativity and accusativity, depending on discourse structure (Aikhenvald 1994a). Personal pronouns with animate referents distinguish subject case and non-subject case; the latter is used for any non-A/S constituent. Any topical NP in non-A/S function takes enclitic -miku. Focalized subjects are marked with a suffix -ne 'agentive' (homophonous with the instrumental marker) (see Aikhenvald 1994b on case marking in Tariana). Tariana inherited a morphological distinction between active and stative intransitive verbs from Proto-Arawak. Active intransitive and transitive verbs obligatorily take cross-referencing prefixes to mark the A/Sa constituent (examples 1 and 2). An important property of Tariana verbs is their transitivity value. In some languages every, or almost every verb is strictly transitive or intransitive; in other languages at least some verbs can have either transitivity value. These verbs are called ambitransitive, or labile (Dixon 1994:18; 54). All transitive verbs in Tariana are A=S ambitransitive. This means the object NP can always be optionally omitted, as in English 'eat' Che eats dinner* or lie eats'), or 'knit'2. Example (1) illustrates an ambitransitive verb, and (2) an active (S a -type) intransitive. (1)

(a:si) (pepper)

nu-hyà-ka lsg+A-eat-DECL

(2)

nu-ruku nu-a lsg+Sa-go down lsg+S a -go

Ί eat/am eating (pepper)'

Ί am going downstream'

Stative (or So-type) intransitive verbs do not take any cross-referencing markers: (3)

nuha I

keru-mha angry-NON VIS PRES

'lam angry"

Constituent order is free, with a strong tendency towards verb-final order 3 . Abbreviations used throughout the paper are: ANIM - animate; APPR - approximative; CAUS - causative; CL - classifier; DECL - declarative; DEM - demonstrative; DEP - dependency marker; f, FEM - feminine; FR - firustrative; INFR - inferred; INS - instrumental; m, MASC - masculine; NEG - negative; nf - nonfeminine; NOM - nominalizing; NON VIS - non-visual; OBJ - object case; PART - participle; PL - plural; REC - recent; REM - remote ; sg - singular; TOP NON A/S - topical non-subject. 2

S=0 ambitransitives, i.e. verbs similar to English break' (I broke the glass vs The glass broke) are rare. Some of them are definitely due to loan translations from Portuguese.

3

Tariana has a very complicated verb structure, which is typical for North Arawak languages. Simple predicate structure is as follows: Prefix + root (+ optional thematic syllable) +valency changing causative -ita + reciprocal kaka + modals and valency changing: negative -ka-de or purposive mood -hyu or generic objective -karu or -ni 'focus' or -kana 'passive' + verb-incorporated classifier

49

Transitivity Increasing Operations in Tariana

3. Morphological causatives Morphological causatives are regularly formed on active intransitive verbs (Sa type). They are marked with a suffix -ita which follows the thematic suffix. (4)

Ifinu-nuku dog-TOP NON A/S Ί drove the dog out'

nu-musu-ita-ka lsg-go out-CAUS-DECL

Younger speakers tend to replace morphological causatives of some intransitive active (Sa) verbs with periphrastic caustives with an obligatory dependency marker -ka. Thus, younger speakers rejected -wereta, a morphological causative of the verb -wera 'fall (of trees, fruit or hair); however, an older speaker insisted on using this. Morphological causatives can be formed on stative (S 0 ) verbs if the meaning of the verb presupposes a state changeable through the intervention of a causer. Verbs of physical state belong to this class, e.g. sakamu 'hike-warm', causative -sakamu-ita "to warm up'; hiwiri 'cool·, causative -hiwiriketa 'cool down (e.g. by stirring)'; makara 'dry1, causative -makareta 'to dry"; pusa 'wet', causative -pusita 'wetten, moisten'. So also do S 0 verbs which imply a transition from one state to another, e.g. kawhi 'awake, wake up', causative -kawheta "to wake somebody up'; hiku 'appear", causative -hikweta 'create, make come into being' 4 . Verbs which refer to an inherently unchangeable state cannot form a morphological causative. Thus, no morphological causative can be formed on verbs describing physical states such as colour: ha Je 'white', kada "black1, iri 'red'; physical properties: hamu "hot", hape 'cold'; size: hanu "big1, tsu 'small'; taste: hipisi "bitter1; value: mapa "good". Note that in Tariana stative verbs which denote such concepts as COOL and LUKE-WARM belong to a different system from COLD and HOT. Morphological causatives are also formed on few transitive verbs. Almost all of these refer to traditional actions performed during rituals. The only exception is -ira 'drink' with the corresponding causative -ireta 'make drink, make drunk' 3 . Other verbs are: -sita 'smoke a traditional cigar1, causative -siteta 'get one's partner to smoke in the cigar-smoking ritual'; -eme 'sniff snuff, causative -emeta 'get someone to sniff snuff; -ñapa "bless', causative -ñapeta 'get a shaman, or an older man, to bless someone'; -perù 'lick tobacco from partner's tongue in the

+ benefactive -pena + relativizing and nominalizing aflxes (relative and converb -Ji, etc). These are followed by aktionsart + tense-aspect-mood + complementizers + discourse enclitics (all 'optional'). Tariana also has productive verb serialization (Aikhenvald mss). 4

There are regular phonological processes on the boundary between the root and the suffix, e.g. a+ i > e.

5

In closely related languages the morphological causative is usually formed on intransitive verbs only. The only exception is 'drink' (Baniwa -ija. Bare -dia, Warekena -kulua). Cross-linguistically, verbs describing digestive processes often display abnormal behaviour with regard to their transitivity (Ken Hale, p.c.).

50

Alexandra Y. Aikhenvald

cigar-smoking ritual', causative -perita 'get someone to lick tobacco from partner's tongue1. (5) is an example of -erne 'sniff. (5)

yatu neme snuff 3pl+sniff 'They were sniffing snuff

(6) is an example of the corresponding causative -emeta. There is no morphological mechanism to distinguish between the 'old' O (yatu 'snuff) and the new O (nawiki 'man') in Tariana. (6)

diha kui-se-sina diha nawiki-nuku he Kui-CONTRAST-NON VIS REM PAST he person-TOP NON A/S yatu di-emeta snuff 3sgnf-sniff+CAUS 'Kui (the mythical 'Master of Shamans') made the man sniff the snuff

That morphological causatives are only used with verbs which describe highly ritualized traditional activities suggests that they are archaic. The same suffix, -ita, is used as a marker of argument-adding with transitive verbs. In §7, we will look at how different meanings of this morpheme interact.

4. Causative serial verb constructions Characteristics of serial verb constructions in Tariana are briefly described in §4.1. Then, in §4.2. causative serial constructions are analyzed as a subtype of serial constructions. 4.1 Properties of serial verb constructions in Tariana The following properties are used to define serial verb constructions (Foley and Olson 1985; Durie forthcoming). Í1 ) A serial construction has the properties of a single predicate: (a) it refers to a single event, or several events closely knit together; (b) it functions on a par with monoverbal clauses in discourse; (c) it has shared tense/aspect, modality and, often, polarity marking; (d) it has a single subject; (e) verbs in a serial construction often share other arguments. (2) A serial construction has the intonational properties of a monoverbal clause, and not of a sequence of clauses. (3) A serial construction can be distinguished from subordination or coordination, or complex predicates; it contains no markers of syntactic dependency. (4) A serial construction may consist of several morphological words. The whole serial construction occupies one core functional slot in the sentence or clause structure.

Transitivity Increasing Operations in Tariana

51

Serial constructions in Tariana are known to contain up to seven verbs. Each of these verbs is an independent phonological word, since it has an independent stress and enclitics can attach to it. Each of them receives a separate inflection for person/number/gender and a transitivity marker. Each can be used by itself and receives an independent stress, and can be considered an independent phonological and morphological word. The verbs in a serial construction must have the same subject. This implies identical crossreferencing on all the components. Serial constructions in Tariana are strictly contiguous. This means that no constituent can be inserted between the components of a serial verb construction. Serial verb constructions are widely used to express aspectual, directional, and modal meanings (for a further analysis of serial verb constructions in Tariana, see Aiklhenvald mss). (7) is an example of a serial verb construction which has a directional meaning. The construction is underlined. (7)

na-musu na-nu nema diha-pua-nuku 3pl-go out 3pl-come 3pl+stand he-CL:RIVER 6 -TOP NON A/S 'They (the ancestors of Tarianas) were coming out towards (lit. go out-come-stand) this (river)1

4.2 Causative serial verb constructions Causative constructions in serializing languages require 'that the object of one verb and the subject of another be preferential' (Foley and Olson 1985: 25; Crowley 1987: 38-39). They can be of two types. Some languages put independent inflection on the components of a serial construction to cross-reference the two different semantic subjects. Such a construction in Paamese (called 'switch-subject' serial constructions: Crowley 1987) is shown in (8). (8)

kaiko ko-muasi-nau nau-vaa netano 2sg 2sg-real-hit-lsg lsg-real-go down 'You hit me down (Ht. you hit-I M ) ' (Crowley 1987: 48, ex. 27)

The other technique, labelled 'concordant dependent inflection' (Durie forthcoming) involves putting the same subject marking inflection on all the components of a serial verb construction. Such a construction in Akan is shown in (9) (Schachter 1974: 258: 9): (9)

mede aburow migu msum I-take corn I-flow water-in Ί pour corn into water (ht. I take com I flow in water).'

In this case, the two predicates, 'take' and 'flow1, have different underlying subjects ('Γ and 'corn' respectively), but they receive the same surface subject marker which corresponds to the

6

Classifiers are given in caps.

52

Alexandra Y. Aikhenvald

subject of the verb of causation, in this case, take'. That is, in languages with concordant dependent inflection the components of a causative serial verb construction have different underlying semantic subjects which get marked as the same surface syntactic subject via crossreferencing on both verbs. This technique is rather rare in the languages of the world. Serial syntactic causatives in Tariana have concordant dependent inflection. The components of a serial syntactic causative construction have different underlying subjects, realized as the same subject in the surface form. The subject of the verb of causation is cross-referenced on the two verbs which form a causative serial verb construction. The semantic subject of the causativized verb, if expressed with an NP or a personal pronoun, is marked with the object case, e.g. i-na (2pl-OBJECT CASE) "you1 in (12). The serial syntactic causative can be used with transitive or active intransitive verbs and is in fact the only productive mechanim to causativize transitive verbs. Causative serial verb constructions with concordant dependent inflection can be formed with almost any transitive verb. The most frequent ones are -a 'let', as in (10); -warn 'call', as in (11); -ira 'order*, and -matara 'allow, give permission'. (10) ka:ru-ka nuha nn-a-mahka nu-hvä-niki fear-DEP I lse-let-NON VIS REC PAST 1 se-eat-COMPLETED piri-nuku 2sg+son-TOP NON A/S 'Being afraid, I let (the fish) eat (Ut. I let-I eat) your son' (11) na-wana na-wheta-pidana 3pl-call 3pl-stay+CAUS-INFR REM PAST 'They called (him) to stop (his canoe)' This list of verbs is by no means exaustive. (12) illustrates a transitive verb "bring' used in a causative serial verb construction. The same surface subject (T) is cross-referenced on the verb of causation ("bring') and the sequence of other verbs ('enter" - 'come'). Their underlying subjects are different: it is 'Γ for the verb of causation "bring", and "you' for the 'enter-come". A causative serial verb construction can contain aspectual serial verb. (12) contains an aspectual (habitual) serial verb construction with the verb -yä 'stay1. (12) i-na nhuta nhue 2pl-OBJ lsg+bring lsg+enter Ί usually get you to come in'

nu-nu lsg-come

nu-yâ-ka lsg-stay-DECL

Serial causative constructions satisfy the criteria used for serial verb constructions. They have the following properties of a single predicate: • They have shared tense/aspect, modality, evidentiality and polarity marking, i.e. verbs which form a serial construction cannot receive independent marking for these categories.

Transitivity Increasing Operations in Tariana

53

• They have the international properties of a monoverbal clause, and not of a sequence of clauses. • They can be distinguished from subordination or coordination, or complex predicates, since they contain no markers of syntactic dependency.

5. Periphrastic causatives There are two kinds of periphrastic causatives — those with an optional dependency marker (§5.1), and those with an obligatory dependency marker (§5.2). The reasons why periphrastic causatives are not serial verb constructions are given in §5.3. 5.1 Periphrastic causatives with an optional dependency marker Verbs which refer to an inherently unchangeable state (colour, physical properties such as size or taste, etc; see §3) can not form a morphological causative. Periphrastic causatives formed with the verb of causation -ni 'make' are used to causativize these stative verbs. The dependency marker -ka can be optionally used on the causativized stative verb, with no semantic difference. (13) is an example of a periphrastic causative without a dependency marker; (14) is an example with one. Both come from traditional stories. In each case periphrastic causatives are underlined. (13) di-na na-boleta-pidana thui-niki 3sgnf-OBJ 3pl-defeather-INFR REM PAST all-COMPLETED ne-ma-sale-da na-ni-pidana NEG-NEG-hair-CL:RQUND 3pl-make-INFR REM PAST 'They (people) took all his (eagle's) feathers from him, they made (his head) bald' (14) nawiki ka:da-ka-naka pi-ni person black-DEP-EYEWITNESS PRES 2sg-make 'You make people black' (said the Moon to the Sun)

phia you

5.2 Periphrastic causatives with an obligatory dependency marker Periphrastic causatives with the verb of causation -ni 'make' can also be formed on intransitive active and transitive verbs. These periphrastic causatives imply a special effort and/or unwillingness of the causee to bring about the activity. This is illustrated with (15). Here the subject (father) has to make a special effort to achieve the result. The dependency marker -ka on the causativized verb is obligatory. Its omission results in an ungrammatical sentence. (15) diha he

emi-peni-nuku nema-ka child-PL:ANIM-TOP NON A/S 3pl+close eyes/sleep-DEP

54

Alexandra Y. Aikhenvald

di-ni-mahka 3sgnf-do-NON VIS REC PAST 'He has made the children sleep (with special effort and actions)' In (16) the result is achieved naturally, with the help of the magic power of the cigar. The corresponding morphological causative of the verb -ima 'close eyes/sleep' is used here. (16) di-na mafia yema-ne nemeta 3sgnf-OBJ good cigar-INS 3pl+close eves/sleep+CAUS 'They made him really asleep with a cigar1 5.3 Why periphrastic causatives are not serial verb constructions Periphrastic causatives do not qualify as serial constructions for the following reasons. (i) Every verb in a periphrastic causative construction is marked for its surface and underlying subject independently, as in (15) and (17). In a causative serial verb construction the subject of the verb of causation is cross-referenced on both verbs. (ii) The two components of a periphrastic causative may have an independent tense/aspect/aktionsart value, and get independent marking. For example, in (17) -mha 'nonvisual present' goes on - Jena 'feel', and -yha 'approximative' characterizes the verb of causation -ni 'do, make': (17) nu-na ma: f i nu-J ena-ka-mha i-ni-vha lsg-OBJ bad lsg-feel-DEP-NON VIS PRESENT 2pl-do-APPR You made me feel miserable a little bit', i.e. "you did a little bit for me to feel miserable' (said the mother to the young man who had gone to live with snakes)' (iii) periphrastic causatives do not have the intonational properties of a monoverbal clause; a pausai marker can be inserted after every verb 7 . Periphrastic causatives are found more frequently in texts and other data from younger speakers, so some sort of influence from Portuguese can not be excluded.

6. Causative constructions in Tariana: a comparison The properties of the four types of causative constructions in Tariana are summarized in Table 1.

7

Unlike serial causative constructions, some periphrastic causatives are not one predicate, since they consist of -ni 'make' plus a complement clause with a dependency marker -ka. See (14) and (15).

55

Transitivity Increasing Operations in Tariana

TABLE 1. Properties of causative constructions in Tariana Properties

Morphologi-

Serial

cal causatives causative

Periphrastic

Periphrastic

causatives with an causatives with an

constructions

optional obligatory dependency marker dependency marker

1. Verb types transitive verbs

a few

yes

no

yes

Sa verbs

yes

yes

no

yes

S 0 verbs (changeable

yes

no

no

no

no

no

yes

nausee is in control

optional

yes

no

optional

causee does the action willingly

no

optional

no

optional

yes

optional

yes

yes

yes

optional

yes

yes

no

no

no

yes

properties) S 0 verbs (unchangeable properties) 2. Causee

3. Causer causer acts directly causer achieves the result intentionally action requires an effort The differences between four types of causative constructions in Tariana are: • Morphological causatives are productively formed on intransitive active (Sa) and stative (So) verbs which refer to changeable qualities and states. There are just a few instances of morphological causatives on transitive verbs describing traditional activities. • Serial causative constructions are used to form causatives of intransitive active and transitive verbs using a number of verbs of causation. • Periphrastic causatives formed with the verb of causation -ni 'make' with an optional dependency marker on the causativized verb, are used to causativize stative verbs which refer to unchangeable properties and states. • Periphrastic causatives with an obligatory dependency marker are use to form causatives of intransitive active and transitive verbs which imply a special effort on the part of the causer. • Unlike in other causative constructions, in serial causative constructions the causee is always in control; the causer may act indirectly; the causer may achieve the result unintentionally. Un-

56

Alexandra Y. Aikhenvald

like periphrastic causatives with an obligatory dependency marker, the action does not require an effort on the part of the causee.

7. Argument-adding derivation As described in §2, when -ita is added to intransitive verbs and a limited number of transitive verbs, it has causativizing effect. When added to the majority of transitive verbs, it has two, quite different effects. (i) It was mentioned in §2 that all transitive verbs in Tariana are A=S ambitransitives. When a transitive verb receives the transitivizer -ita, it becomes obligatorily transitive. (ii) -ita on a transitive verb also indicates that a peripheral constituent has to be obligatorily stated in the clause. This constituent may be locative, purpose, or comitative. Its choice depends on the semantics of a verb. The following examples illustrate this. The transitive verb -pala means 'get, put'; when -ita is added, -paleta means 'put in a particular location', as in (18). (18) ita-whya hi-nuku pi-paleta canoe-CL:CANOE DEM-TOP NON A/S 2sg-put+CAUS 'Put the canoe here'. The addition of -ita marks the obligatoriness of a purpose with other verbs. The transitive verb -wana means 'call', emit sound', as in (19). (19) ne di-sado-nuku di-wana-tha-pidana häwa then 3sgnf-spouse-FEM-TOP NON A/S 3sgnf-call-FR-INFR REM PAST eagle "Then the eagle called his wife('s name) (in vain).' With -ita, -waneta means 'call for something', as illustrated in (20). (20) na-waneta-pidana diha yakaJe-peni 3pl-call+CAUS-INFR REM PAST he village-PL:ANIM katu-nuku na-hña-kasu piraiba fish-TOP NON A/S 3pl-eat-INTENTIONAL 'The villagers called him to eat piraiba fish'.

diha he

The verb -wapa means 'wait'; and -wape(ta) means "wait with something, or for a particular purpose'. These meanings are illustrated in (21) and (22).

Transitivity Increasing Operations in Tariana

57

(21) wasä dapa wa-wapa di-a-pidana let's paca lpl-wait 3sgnf-say-INFR REM PAST 'Let' sit and wait for a paca (a large rodent) (to come along)'. nu-kesi-ni-nuku (22) nu-wapeta lsg-wait+CAUS 1 sg-friend-NOM-TOP NON A/S pa-ira-nipe pumeni-peri-ne IMPERSONAL-drmk-NOM sweet-COLLECTIVE-INS Ί am waiting for my friend with drinks and sweet things (as an offering)' Thus, the morpheme which marks the causative of intransitive verbs is used to mark a kind of argument-adding derivation when used with transitive verbs. In §3 we mentioned five transitive which allow formation of a morphological causative with the same formative -ita. The argument-adding or causative reading of -ita derivation in these cases depends on which constituent is present in the surface structure. A transitive verb -ñapa "bless1, without -ita, is illustrated in (23). The O is omitted. (23) diha-kuma khema-kaJi na-ñapa-sina he-CL:RITE REL+understand-PAST PART 3pl-bless-NON VIS REM PAST 'Those who understand the ritual, bless (do the blessing)' Ifjust an O constituent is present, the -ita derivation has a causative reading, as in (24). (24) waha haniri-nuku Kose we father-TOP NON A/S Concessäo du-ñapeta-sina duri-nuku 3sgf-bless+CAUS-NON VIS REM PAST 3sgf+son-TOP NON A/S 'Concessäo made our father bless her son (to cure him)' It has an argument-adding reading if there is an overt instrumental constituent (25). (25) maini-ne na-fiapeta tar-INS 3pl-bless+CAUS 'They blessed (the game) with tar* In everyday speech argument-adding readings (as in 25) are much more frequent than causative (as in 24).

58

Alexandra Y. Aikhenvald

8. Transitivity increasing operations in Tariana in typological perspective It is not unusual for a language to have more than one mechanism for forming causatives. Most often, languages have two mechanisms - one morphological, and the other periphrastic, or 'syntactic' (there are also lexical causatives which we are not discussing here). A typologically unusual property of Tariana is that there are four mechanisms for marking causatives: one morphological, two periphrastic, and a causative serial verb construction. It is not very unusual for a language to use different techniques of causative marking techniques depending on the transitivity value of a verb (see Dixon forthcoming). There is a cross-linguistic tendency among languages of the world for periphrastic causatives to express both (a) indirect, rather than direct, causation and (b) causation which involves a special effort on the part of a causer (Dixon forthcoming; cf. also Haiman 1983). Morphological causatives tend to express direct causation. Tariana conforms to this tendency. In many languages, if a causative derivation has another meaning, this is most likely to be applicative. Such is the case in the Australian language Yidiny (Dixon 1977: 215), and in a number of South American languages of the Arawak family (see Wise 1990). A very unusual property of Tariana is that the same morpheme (-ita) is used to mark causatives with intransitive and a few transitive verbs, and also to specify that an ambitransitive verb is obligatorily transitive and that a peripheral constituent must be stated. This argumentadding derivation is extremely rare cross-linguistically. There is one typological analogy to it. A few Campa languages, from the South Arawak subgroup, have a number of applicative derivations (see Dixon and Aikhenvald 1997) which affect the argument structure of the predicate by putting a benefactive, or an instrumental, or a 'presential' (meaning 'in the presence of) consituent into the O slot. One verb can have more than one derivational suffix of this sort. In this case one of the derivational suffixes just shows that a peripheral constituent is obligatory. See (26), from Pajonal Campa (Shaler 1971:45). This sentence contains an applicative derivation, -ako "benefactive; with reference to'; there is another suffix -imo 'in the presence of which just indicates that a peripheral constituent (in this case, Irene) has to be overtly expressed. (26) no-p-ako-ts-imo-ts»-ro-ri 1 sg-give-REFERENCE TO-EPENTHETIC-IN PRESENCE OF-ASPECT-3sgf-3sgm Irena Irocarto paño Irene Richard scarf Ί gave Richard the head scarf in Irene's presence' (Shaler 1971: 45) The argument-adding derivation in Tariana is a special subtype of valency increasing mechanisms distinct from both causatives and applicatives. This deserves more study with a typological perspective.

Transitivity Increasing Operations in Tariana

59

References Aikhenvald, A.Y. (1994a) Classifiers in Tariana, Anthropological Linguistics 36,405-465. Aikhenvald, Λ.Υ. (1994b) Grammatical relations in Tariana, Nordic Journal of Linguistics 17,201-218. Aikhenvald, A.Y. (1996) Areal diffusion in North-West Amazonia: the case of Tariana. Anthropological Linguistics 38, 73-116. Aikhenvald, A. Y. (mss) Verb serialization and verb compounding: the case of Tariana. Dixon, R.M.W. and Aikhenvald, A.Y. (1997) A typology of argument-determined constructions. In J. Hai man, J. Bybee and S. Thompson (eds): Essays on Language Function and Language Type. Amsterdam, Philadelphia: John Benjamins. Crowley, T. (1987) Serial verbs in Paamese, Studies in Language 11, 35-84. Dixon, R. M. W. (1977)A Grammar ofYidiny. Cambridge: Cambridge University Press. Dixon, R.M.W. (1994) Ergativity. Cambridge: Cambridge University Press. Dixon, R. M. W. (forthcoming) The semantics of causatives. To appear in Valency-changing Derivations. Durie, M. (forthcoming) Grammatical structures in verb serialization. In A. Alsina, J. Bresnan and P. Sells (eds.) Complex Predicates. Stanford: CSLI Press. Foley, W. and M. Olson (1985) Clausehood and verb serialization. In J. Nichols and A.C. Woodbury (eds.) Grammar Inside and Outside the Clause. Cambridge: Cambridge University Press, 17-60. Haiman, J. (1983) Iconic and economic motivation. Language 59: 781-819. Schachter, P. (1974) A non-transformational account of serial verbs. Studies in African Linguistics. Supplement 5,153-271. Shaler, D. (1971) Identification of clause types and participant roles in Pajonal Campa. Summer Institute of Linguistics, Peru. Wise, Mary Ruth (1990) Valence-changing affixes in Maipuran Arawakan languages. In Doris L. Payne, ed. Amazonian Linguistics: Studies in Lowland South American Languages. Austin: University of Texas Press, 89-116.

How to Butter a Sandwich. On 'spread' in Dutch and Hungarian from a Typological Perspective Casper de Groot

1. Introduction Since the early seventies there has been an ongoing interest in the semantic properties of verbs of filling and emptying (cf. Levin 1993: 49f). In many languages these types of verbs select two semantic patterns or argument structures, as for instance in the following examples from English: (1) a. b.

John spreads butter on the bread, John spreads the bread with butter.

Moravcsik (1978) discusses the relation between these two patterns from a typological point of view. On the basis of data from Dutch and Hungarian I will demonstrate in this paper that languages may show considerable differences in the application of the two patterns. For the discussion I will use the typological observations made by Moravcsik (1978). I will show that Moravcsik's typology needs some fine tuning.The data will be limited to the class of verbs of filling. The class of verbs of emptying will not be taken into consideration.

2. Moravcsik's view Moravcsik (1978: 254-261) discusses several types of alternations between arguments of predicates. One of the types consists of the accusative-prepositional alternation, as illustrated by the following examples from English: (2) a. b.

Mary filled wine into the bottle, Mary filled the bottle with wine.

(3) a. b.

Bill planted trees in the garden, Bill planted the garden with trees.

Moravcsik does not consider the accusative-adverbial alternation to correlate with a semantic case-functional alternation: i.e. each of the two noun phrases retains its semantic case function in spite of their variation in form. In contradistinction to Moravcsik, I will argue that the noun phrases have different semantic roles. Moravcsik recognizes a semantic difference between the alternative constructions. The semantic difference can be appreciated through the differential entailments of members of each

62

Casper de Groot

pair. Thus, for instance, spreading the bread with butter entails that something has been done to the bread; and spreading butter on the bread entails that something has been done with the bread. In the first instance the entailment is that the bread has been 'affected'; in the second case there is no such entailment. On the basis of data from Hungarian, I will argue that in the first case bread is the 'affected' argument, and that in the second case butter is the 'affected' argument. Moravcsik furthermore notices that planting the garden with trees entails that as a result the whole garden has trees in it, whereas planting trees in the garden does not entail this. Moravcsik suggests that in general the locative complement marked as accusative is asserted to be affected by the event in its full extent, whereas the locative marked as locative adverbial is not asserted to be so affected. Below, I will argue that in the latter case the accusative complement specified by trees in Hungarian can be affected by the event in its full extent. Moravcsik mentions that in English and Hungarian the class of verbs which allow this type of alternation behave in almost exactly the same way. There is, however, one significant difference between corresponding constructions in the two languages. Whereas the form of the verb in the alternative English constructions is in most cases the same, in Hungarian the two forms in most cases differ in that their verbal prefixes differ. Moravcsik concludes the following. The difference between the English and Hungarian constructions is that whereas in English completive verbal particles occur only with some verbs and only if the accusative is the semantically locative phrase, in Hungarian some verbal particles occur with any verb and with both kinds of accusative. In this article I will further elaborate on this distinction. Moravscik concludes the section on the accusative-adverbial alternation with the following generalization: if in a language the same verbal meaning is expressible either through a construction where a complement of a verb is in the accusative or through a construction where the same complement is in an adverbial case and there is a meaning difference between the two constructions, this semantic contrast will be either a contrast between a definite and an indefinite object, or a contrast between an object that is fully involved in the event and one that is partially involved, or a contrast between affected and not affected participants. I will return to these different aspects of the generalization in the sections below.

3. The semantic structure As noticed above, Moravcsik does not consider the accusative-adverbial alternation to correlate with a semantic case-fiinctional alternation: ie. each of the two noun phrases retains its semantic case function in spite of their variation in form However, on the basis of semantic and syntactic tests, it can be shown that the complements take different semantic roles and consequently that there are different predicate structures involved. In fact, I will argue that the vert) plant has the following two semantic structures: (4) a. plant b. plant

(John)AgCTt (John)Agent

(trees)««* (garden)^*«

(garden)^;™ (treesV™

How to Butter a Sandwich

63

The first test is based on the observation that gapping in coordination is only allowed if the NP that can be left out fulfils the same semantic role as its counterpart in the coordinated clause. Sentence (Sa) is such an example in which 'letter1 functions as the patient of both the verb "write1 and 'read'. Example (5b) is ungrammatical because "house' functions as the patient of "build1 but as the location of'live'. Consider: (5) a. Mary is writing and Peter is reading a letter. b. ""William is building and Susan is living in a house. Given the constraint on gapping, we cannot but conclude that in the following example 'garden' functions as the patient of "plant1, because 'design' selects this function for the object complement: (6)

Hockett designed and Wharfdale planted the garden.

However, when we have a look at example (7), it seems that trees' may also be used as a patient, because "buy1 selects a patient as its object: (7)

Carter bought and Bradley planted the trees.

Examples such as (6) and (7) show that 'plant' can either take 'garden' or trees' as its patient, which forces us to accept two predicate structures for 'plant'. A second test is based on the possibilities for passivization. If'butter' in the two patterns of 'spread' has just one and the same underlying semantic role, and "butter* in one pattern may function as the subject in a passive construction, we would expect that "butter1 in the other pattern may also function as the subject in a passive construction. This is not the case, as shown by the following examples: (8) a. John spread the butter on the bread, b. The butter was spread on the bread. (9) a. John spread the bread with butter. b. * The butter was spread the bread with. The fact that "bread' can function as the subject in the pattern in which "butter* cannot, and that in the other pattern "bread' cannot function as the subject, whereas "butter* can, reinforces the view that there are two predicate structures involved. Compare: ( 10) a. The bread was spread with butter, b. * The bread was spread the butter on.

64

Casper de Groot

When there are two patterns in which one of the alternating complements functions as the patient of the verb, the question arises what the semantic roles of the other complements are. It seems that in one pattern the semantic role of the third complement is the role of direction (cf. (2a)) or location (c£ (3a)), and in the other pattern the role of means, typically expressed with the preposition with. Different from Dik (1980:32), I do not consider this complement to have the function of instrument, because clauses with this pattern allow extension with an instrumental phrase. Clauses do normally not allow two different instrumental phrases. The following examples show that extension with an instrument is allowed: (1 l)a. b.

Hamlyn planted the garden with shrubs and trees with special tools and equipment, Hamlyn used special tools and equipment to get the garden planted with shrubs and trees.

I can now characterize the two patterns in the following fashion, where xl is the complement specified by the 'actor1, x2 by the 'substance' and x3 by the 'container'. Compare: (12) a. frame

1

b. frame 2

verb ( x l ) ^

( x 2 ) P a t i e n t ( X 3 ^Direction/Location

verb (xl)Agmt (x3)p,t¡ent (χ2)μ«™

In the remainder of this paper I will refer to these patterns as frame 1 and frame 2. 4. The data 4.1. Dutch In Dutch, both patterns occur. Unlike English, the Dutch verbs have different forms. The verb in frame 2 takes either the prefix be- or the adjective vol 'füll'. Consider: (13)a.

Jan smeert boter op het Jan spreads butter on the 'Jan spreads butter on the bread.'

brood. bread

frame 1

b.

Jan besmeert het brood met Jan spreads the bread with 'Jan spreads the bread with butter.'

boter. butter

frame 2

c.

Jan smeert het brood vol met Jan spreads the bread fìlli with 'Jan spreads the bread full with butter.'

boter. butter

frame 2

The verb smeren on the one hand and besmeren / volsmeren on the other cannot be used in the alternate frames. Thus the following patterns do not occur:

65

How to Butter a Sandwich

(14)a. "Jan besmeert boter op het brood. b. * Jan smeert boter vol op het brood. c. *Jan smeert het brood met boter. The relation between the two forms of the verb can be considered to be a relation of derivation. Dik (1980:34-36) shows that the two variants of frame 2 are derived from frame 1. The rule Dik proposes is as follows: (15)COMPLETIVE VERB FORMATION IN DUTCH input: Predv (xi)Ag (xî)ρ* tek« output: vol/be-Predy (xi)^ (x3)pat (x2)imn condition: "In order to serve as an input predicate-frame, a predicate-frame must indicate an Action by which something is applied to some surface in such a way that the surface gets covered with the something as a result of the Action." The form with be- is synchronically not productive, whereas the form with the adjective vol 'fixll' is. In most cases in which the be- form can be used, the form with vol is also allowed. The be- form and the form with vol are not fully exchangeable. In some cases, the form with behas an idiomatic meaning. I will return on the status of the adjective vol in section 6 below. 4.2. Hungarian Hungarian also uses the two patterns (cf. De Groot 1986, 1989). Hopper & Thompson (1980: 262fl) suggest that the two patterns have the following distribution of properties: (i) OV; subjective conjugation; imperfective aspect, versus (ii) VO; objective conjugation; perfective aspect. The data presented in this paper will show this suggestion to be wrong. The verb has the same form in both patterns. However, when the perfective aspect applies, the verb takes a different aspect marker. In the case of ken 'spread', the perfective marker in frame 1 is rá- and in frame 2 meg-. Frame 2 also allows the lexical form tele- 'full', which from a morpho- syntactic point of view behaves as meg-. (16) a.

János vaj-at ken a kenyér-re. János butter-ACC spreads the bread-SUBL 'János spreads(imperfective) butter on the bread.'

b.

János rá-keni a vaj-at a János PFV-spreads the butter-ACC the 'János spreads(perfective) butter on the bread.'

c.

János kenyer-et ken vaj-jal. János bread-ACC spreads butter-INSTR 'János spreads(imperfective) bread with butter.'

frame

kenyér-re. bread-SUBL

frame

frame

1

1

2

66

Casper de Groot

d.

János vaj-jal meg-keni a kenyer-et. János butter-INSTR PFV-spreads the bread-ACC 'János spreads(perfective) the bread with butter.'

frame

2

e.

János vaj-jal tele-keni a kenyer-et. János butter-INSTR M-spreads the bread-ACC 'János spreads(perfective) the bread full with butter.'

frame

2

The examples are basic declarative sentences which are unmarked with respect to word order and the use of the article of the object. The examples with the imperfective aspect, (16a) and (16c), also allow objects with the definite article. If the word order remains the same, the objects will have emphatic stress, Le. they will have the pragmatic function of focus. Consider the examples in (17). Also note that the finite verb takes the so-called objective conjugation in these cases. This type of conjugation is triggered by the presence of a definite object. (17)a.

b.

János a vaj-at keni a kenyér-re. János the butter-ACC spreads the bread-SUBL 'János spreads(imperfective) THE BUTTER on the bread.' János a kenyer-et keni vaj-jal. János the bread-ACC spreads butter-INSTR 'János spreads(imperfective) THE BREAD with butter.'

frame

frame

1

2

4.3. Dutch vs Hungarian The patterns found in Dutch and Hungarian can be summarized in the following way: (18)

Dutch

frame 1 frame 2 frame 2

smeren besmeren volsmeren

(19)

Hungarian frame 1 frame 2 frame 2

(rá)ken (meg)ken teleken

In the next three sections I will discuss this typology with respect to holistic interpretation, the overt expression of completeness, and the (in)definiteness of the objects.

5. Holistic interpretation As mentioned above, Moravcsik noticed that planting the garden with trees (= frame 2) entails that as a result the whole garden has trees in it, whereas planting trees in the garden (= frame

67

How to Butter a Sandwich

1) does not entail this. She suggests that in general the locative complement marked as accusative is asserted to be affected by the event in its full extent, whereas the locative marked as locative adverbial is not asserted to be so affected. The Dutch data confirm this observation. In the case of'spread' frame 2 entails that as a result the slice(s) of bread will be covered with butter, whereas frame 1 does not. In this respect, Dutch patterns along with English. Note, however, that Dutch uses two different forms: smeren (non-holistic) and besmeren (holistic) (cf (13) above). Hungarian behaves differently. The holistic entailment cannot be associated with a type of frame, but with the application of perfective aspect: when perfective aspect is used in Hungarian, the situation described by the verb and its complements is viewed as a whole (cf. Comrie 1976, and De Groot 1984). Because both frames allow perfective aspect, both frames can have a holistic meaning. Consider examples (16b) and (16d), here repeated as (20a) and (20b): (20) a.

b.

János rá-keni a vaj-at a János PFV-spreads the butter-ACC the 'János spreads(perfective) butter on the bread.'

kenyér-re. frame bread-SUBL

János vaj-jal meg-keni a kenyer-et. János butter-INSTR PFV-spreads the bread-ACC 'János spreads(perfective) the bread with butter.'

frame

1

2

Example (20a) entails that all the butter will be used, whereas example (20b) entails that all the bread will be spread with butter.

6. Overt expression of completeness In section 4 we saw examples of the overt expression of completeness. Dutch and Hungarian both use an adjective, respectively vol and tele 'full', in combination with the verb. Although these forms rather behave as particles which together with the verbs form a kind of complex expression with one specific meaning, the forms are also feh to function as predicative adjuncts in a kind of resultative construction in the following sense: "John spreads the bread with the result that the bread is covered with butter". Given the meaning of the adjectives, they can only be associated with the complement which specifies the "container". In a resultative construction of this type the "container" must function as the patient, or affected complement of the verb. That is why that the overt expression of completeness only occurs with frame 2 and not with frame 1. For instance: (21) a.

Jan smeerde het brood vol Jan spread the bread full 'Jan spread the bread full of butter.'

met with

boter. butter

68

Casper de Groot

b.

János vaj-jal tele-kente a János butter-INSTR full-spread the 'János spread the bread full of butter.'

kenyer-et. bread-ACC

The use of vol in Dutch and tele in Hungarian is in complementary distribution with be- and the perfective marker meg-, respectively. A difference between Dutch and Hungarian is that the application of the form with be- in Dutch is far more restricted than the form with meg- in Hungarian. Compare Dutch (22) and (24) with Hungarian (23) and (25): (22) a.

Karel schonk het glas vol Karel poured the glas full 'Karel poured the glas full of water.'

b. * Karel beschonk het Karel poured the

glas glas

met with

met with

water water

water water

(23)a.

Károly víz-zel tele-töltötte Károly water-INSTR full-poured 'Károly poured the glas full of water.'

a the

pohár-at. glas-ACC

b.

Károly víz-zel meg-töhötte Károly water-INSTR PFV-poured 'Károly poured the glas with water.'

a the

pohár-at. glas-ACC

(24) a.

Michiel laadde de wagen vol Michiel loaded the cart full 'Michiel loaded the cart full of hay.'

b. * Michiel Michiel (25)a.

b.

belaadde de loaded the

wagen met cart with

Mihály széná-val tele-rakta Mihály hay-INSTR full-loaded 'Mihály loaded the cart full of hay.'

a the

met with

hooi. hay

hooi. hay szeker-et. cart-ACC

Mihály széná-val meg-rakta a szeker-et. Mihály hay-INSTR PFV-loaded the cart-ACC 'Mihály loaded the cart with hay.' The distribution of préfixai marking in Dutch and Hungarian can be schematically summarized in the following way:

69

How to Butter a Sandwich

(26)

Dutch

holistic besmeren volvolschenken

non-holistic smeren (frame 1 ) schenken

(frame 1)

(frame 2) (frame 2)

Hungarian ken

(frame 2)

(frame 2)

tölt

(frame 2)

(frame 2)

The data in this section, in particular the examples with overt expression of completeness, have shown that the complement specified for the 'container1 may function as the patient of the verb. This reinforces the claim that the class of verbs under consideration have two predicate structures. The data also reinforces the view that the two patterns found in Dutch are related through derivation, whereas such an assumption is not motivated for Hungarian.

7. Definiteness of the object A third aspect relevant to the distinction of two patterns of the verbs is the definiteness of the object. Some languages, e.g. Eskimo, use the indefinite-definite opposition of the object for the distinction between partial or entire involvement in the event. Again, there is a difference between Dutch and Hungarian which can be related to the different ways in which these languages treat the two patterns. Dutch allows both definite and indefinite objects with the verbs which select frame 2. Consider the following examples with besmeren: (27) a.

b.

Jan besmeert het brood met Jan spread the bread with 'Jan spreads the bread with butter.' Jan besmeert brood met Jan spread bread with 'Jan spreads bread with butter.'

boter. butter

boter. butter

Example (27a) is specific in the sense that it is known which bread is involved and consequently also the amount of bread. The other example (27b) has a generic interpretation. Irrespective of the status of the object as definite or indefinite, the verb besmeren always entails a holistic interpretation. Still, there is a difference between these two examples. Example (27a) with the definite object has a twofold holistic interpretation in the sense that (i) either all the slices of bread will contain butter, or (ii) each slice of bread will be fully covered with butter. Example (27b), which is generic, entails just the second type of holistic

70

Casper de Groot

interpretation: each slice of bread (the number of slices is unknown) will as a result of the activity be fully covered with butter. We can thus conclude that to fidly cover (with butter)' belongs to the meaning of the verb besmeren. In Hungarian both frames allow definite and indefinite objects if imperfective aspect applies. In all such cases there is no holistic entailment, because that kind of entailment is due to the application of perfective aspect. A requirement for perfective aspect to apply is that the object is definite. Therefore, example (28) is ungrammatical: (28)

* János János

kenyer-et bread-ACC

meg-kent a PFV-spread the

vaj-jal. butter-INSTR

There is one case in which the object is indefinite and the so-called perfective markers are used. This kind of construction, which also involves a particular word order and intonation contour, however, counts as imperfective. In recent literature the construction has even been referred to as "progressive" (cf. Kiefer 1994). Consider: (29) a.

János kenyer-et kent meg a János bread-ACC spread PFV the 'János was spreading the bread with butter.1

vaj-jal. butter-INSTR

b.

János vaj-at kent rá a János butter-ACC spread PFV the 'János was spreading the butter on the bread.'

kenyér-re bread- SUBL

Examples such as (29) do not have a holistic entailment, because they are imperfective.

8. Conclusions The discussion of verbs of filling in Dutch and Hungarian has made clear that a typological characterization of this class of verbs based on one semantic structure, on "definiteness of the object", and on holistic entailment is not sufficient. Along these parameters, Dutch and Hungarian would ultimately differ, beyond any comparison. With respect to these verbs, definiteness does not play any role in Dutch, whereas definiteness in Hungarian plays a role with respect to perfective aspect. Holistic interpretation can be associated with one frame in Dutch, but with two frames in Hungarian. Holistic interpretation in Dutch can be considered to be the result of derivation, whereas holistic interpretation in Hungarian can be considered to be the result of an inflectional operation, namely the application of perfective aspect. No such problems arise if we take two semantic patterns as the parameter, in a similar way as has been proposed for the typology of the so-called teaching verbs (Workgroup on Functional Grammar 1981). A discussion of verbs of teaching in seven languages showed that semantic

71

How to Butter a Sandwich

differences between the verbs could be captured by the following two models of predicate structures: (30) a.

b.

GIVING MODEL teach (John)Agent

(French^eni

OPERATING MODEL teach (Johnsen, (students)?^

(students)^^™,

(French) Com picmcnt

The Giving Model has been called like this because it presents the teaching relation as if the teacher presents the thing-taught to the learner as a gift. In the Operating Model, on the other hand, the teacher is presented as performing some direct operation on the learner by means οζ or with respect to, the thing-taught. The paper shows that languages may vary in the way they make use of the two frames. Some languages use both, others just one, either (30a) or (30b). The paper also shows that these two semantic frames are compatible with morpho-syntactic rules of the languages investigated, i.e. case marking, word order, passivisation etc. In this paper I proposed the following two frames as the alternative predicate structures for verbs of filling: (31) a. b.

verb ( J o h n ) ^ verb (John)Agcnt

(substance^«« (container^m,

(container ) L „c» t ¡ OT /Di re ct¡o„ (substance^«™

These two patterns neatly account for the syntactic and morphological properties of the verbs of filling in Dutch and Hungarian. Although many languages have verbs which correspond to the frames in (31), the usages made of the opposition are not necessarily the same, and thus require detailed research in each particular language. The comparison of Dutch and Hungarian illustrates this point. Another type of example comes from Latin. Bolkestein (1985) argues that the choice of one of the two patterns in Latin is conditioned by pragmatic factors. Consider for instance: (32) a.

urbem fossa circumdedit city-ACC moat-ABL have-surrounded 'He surrounds the citadel with a moat.'

b.

fossam urbi circumdare ditch-ACC city-DAT have-surrounded 'He surrounds a moat around the citadel'

If there are more contextual strings attached to the citadel (e.g. because it is the topic of the story), then urbem will have a strong preference for the accusative (object) position; if there are more contextual strings attached to the moat, fossa will much more often qualify for the accusative position.

72

Casper de Groot

In conclusion I claim that typological research intoclasses of verbs, in particular those verbs which have alternates, can be best carried out on the basis of different predicate structures. These predicate structures are not deep abstract semantic networks or conceptual frames, but rather representations of the verbs with their arguments, the semantic roles of the arguments, and the selection restrictions on the argument positions. An advantage of this approach is that the predicate structures may also serve as part of the grammar of languages, i.e. frames to which syntactic and morphological operations can be applied.

References Bolkestein, A. M. (1985) 'Cohesiveness ans syntactic variation: quantitative vs. qualitative grammar1. In: A.M. Bolkestein, C. de Groot & J.L. Mackenzie (eds.) Syntax and Pragmatics in Functional Grammar, 1-14. Dordrecht: Foris. Comrie, B. (1976) Aspect. Cambridge: CUP. Dik, S.C. (1980) Studies in Functional Grammar. London: Academic Press. Groot, C. de (1984) 'Totally affected. Aspect and three-place predicates in Hungarian'. In: C. de Groot & H. Tommola eds. Aspect Bound. A voyage into the realm of Germanic, Slavonic and Finno-Ugric Aspectology, 133-151. Dordrecht: Foris. - (1986) 'Nederlanders en Hongaren smeren niet hetzelfde'. In: E. Mollay (ed.) Németalfildi-Magyar kontrasztiv filológiai tanulmányok, 4-18. Budapest: ELTE. - (1989) Predicate Structure in a Funcational Grammar of Hungarian. Dordrecht: Foris. Hopper, P.J. & S.A. Thompson (1980) 'Transitivity in grammar and discourse1. Language 56, 251-299. Kiefer, F. (1994) 'Aspect and syntactic structure'. In: F. Kiefer ed. The syntactic structure of Hungarian. Syntax and Semantics, Vol. 27,415-464. London & New York: Academic Press. Levin, B. (1993) English verb classes and alternations. Chicago & London: Chicago University Press. Moravcsik, E. A. (1978) 'On case marking of objects'. In: J. Greenberg (ed.) Universals of human language, vol 4, 249-289. Stanford: Stanford University Press. Workgroup on Functional Grammar (1981) 'On the Functional Grammar of teaching verbs'. In: T. Hoekstra, H. van der Hulst & M. Moortgat (eds.) Perspectives on Functional Grammar, 203-231. Dordrecht: Foris.

Labile Konstruktionen Werner Drossard

1. Zwei Aufsätze von V. P. Nedjalkov zur anstehenden Thematik 1967 haben P.I. Inènlikej und V.P. Nedjalkov in dem von V.M. 2irmunskij herausgegebenen Sammelband "Érgativnaja konstrukcija predlozenija ν jazykax razlicnyx tipov" einen Aufsatz zu labilen Konstruktionen im Tschuktschischen veröffentlicht, der über die behandelte Einzelsprache hinaus sehr gut als Anregung für breiter angelegte komparatistische Untersuchungen dienen kann, da die Konzeption der Labilität hier nicht auf den "klassischen" (d.h. im engeren Sinne kaukasistischen) Fall festgelegt ist, bei dem zu einem intransitiven Patiens (im Absolutas) im Rahmen einer transitiven Variante ein Agens (im Ergativ) hinzukommen kann, sondern generell alle Fälle als labil aufgefaßt werden, bei denen ohne zusätzliche Verbmorphologie Varianten in der Valenz auftreten. Neben den eher nur für das Tschuktschische relevanten Fällen sind im Hinblick auf den Sprachenvergleich herauszuheben: (1)

ABS fur PAT (intr.) vs. ERG fur AG + ABS fur PAT (tr.)

(la) v9cgottot-0 Brett-ABS

caqat-g'e zerbrech-PRÄT 1/3 sA

(lb) snpsnacg-a caqan-nen v9cgottot-0 Ahe-ERG zerbrech-PRÄT I/3sE:3sA Brett-ABS

"Das Brett zerbrach"

"Der Alte zerbrach das Brett"

Zum zweiten ist ein Fall mit fakultativem Patiens zu nennen: (2)

ABS fur AG (ohne PAT, intr.) vs. ERG fur AG, ABS für PAT (tr.)

(2a) anpanacg-an r'ejipa-rkan Ahe-AB S bohr-PRÄS 1/3 sA

"Der Alte ist beim Bohren / mit Bohren beschäftigt"

(2b) snpsnacg-a Ahe-ERG

"Der Alte bohrt die Kufen"

r'ejips-rkan-in pagtslgsn-0 bohr-PRÄS I/3sE:3sA Kufen-ABS

In beiden Beispielpaaren werden jeweils eine intransitive und eine transitive Variante miteinander konfrontiert. Aufgrund der für Ergativsprachen eigenen Kodierungsstrategie werden das intr. PAT (la) und das intr. AG (2a) im ABSohitus kodiert. Die Labilität der hier aufgeführten Verben liegt in der Oszillation zwischen den intransitiven vs. transitiven Versionen begründet. Wir werden weiter unten sehen, daß sich Labilität jedoch nicht nur im Rahmen der Valenz

74

Werner Drossard

feststellen läßt. Wenn nämlich für bestimmte Ergativsprachen grundsätzlich gilt, daß alle zweiwertigen Verben im Prinzip syntaktisch nur einen ABS-Aktanten verlangen (das gilt für das Tschuktschische mit seiner polypersonalen mtransrtiv-vs-transkiv-Konjugation allerdings nicht!), Hann können Fälle wie bohren hinsichtlich ihrer Orientierung, d.h. der im Lexem angelegten Grundausrichtung auf Agentivität oder Patientivität im Rahmen eines intransitiven Verb+ABS-Schemas labil sein. So gibt es in einigen Sprachen die Möglichkeit, daß für ein Lexem wie essen im Rahmen einer intransitiven Version die Interpretationen essen vs. gegessen •werden vorliegen. Dies impliziert jedoch, daß ein Lexem wie zerbrechen im intransitiven Kontext einer Ergativsprache nicht labil ist. Dazu weiteres unten zu einem Fallbeispiel im Tonganischen. 1969 hat V.P. Nedjalkov in dem von I.F. Vardul' herausgegebenen Band Jazykovye universalii i lingvisticeskaja tipologija den Aufsatz "Nekotorye verojatnostnye universalii ν glagol'nom slovoobrazovanii" publiziert. In diesem Beitrag wird der Status von vier Lexemen (laugh, boil, burn und break) hinsichtlich ihres Verhaltens in intransitiven und kausativ-transitiven Kontexten (also z.B. kochen (intr.) vs. kochen (tr. = "zum Kochen bringen")) in 60 Sprachen untersucht. Es geht dabei vor allem um die Ableitungsrichtung von Simplex zu deriviertem Verb, um die Neutralität von Lexemen und Kausativität vs. Antikausativität. Die drei Lexeme boil, burn und break gehören zum Grundbestand der sogenannten labilen Verben in den Daghestan-Sprachen und sind damit in die oben zum Tschuktschischen entworfene Kategorie 1 einzuordnen. Nedjalkov (1969:112-114) kennzeichnet in seinem Korpus die drei Daghestan-Sprachen Awarisch, Lakkisch und Lezgisch wie folgt: boil

bum

break

Avar

Κ

Ν

Ν

Lak

Κ

Ν

Ν

Lezgian

Ν

Ν

Ν

(Κ = kausativ, d.h.zum Simplex kommt im tr. Fall kausative Morphologie hinzu, bei Ν (= neutral) zeigt ein- und dieselbe Form die intr. und tr. Variante an.)

In allen drei Sprachen sind zumindest burn und break durchgängig als neutral gekennzeichnet, im Lezgischen sogar alle drei Verben. Realiter empfiehlt es sich, im Einklang mit der gängigen kaukasistischen Terminologie, die N-Fälle Nedjalkovs als labil zu bezeichnen. Haspelmath (1993a:289) nennt dementsprechend die drei Lexeme unter den elf labilen Verben des Lezgischen, für die allgemein gilt: /A,ERG, T, ABS/"transitive" vs. /Τ, ABS/ "intransitive"

75

Labile Konstruktionen

Das bedeutet, daß in dieser Ergativsprache, wie auch in den beiden anderen von Nedjalkov (1969) genannten daghestanischen Ergativ-Sprachen, prinzipiell in beiden Fällen, intransitiv wie transitiv, eine Patiens-Orientierung vorliegt, die wir hier mit "Theme" (Τ) gleichsetzen wollen. Anders ausgedrückt: zu einem intransitiven patiens-orientierten Nukleus, bei dem der einzige Partizip ant im Absohitus kodiert wird, tritt in der transitiven Version fakultativ ein Agens im Ergativ hinzu. Dies läßt sich an Haspelmaths Beispielen veranschaulichen: (3a) Get'e-0 pot-ABS

xa-na break-AOR

(3b) Ajal.-di get'e-0 child-ERG pot-ABS

"The pot broke"

xa-na break-AOR

"The child broke the pot"

In diesem, und allen anderen zehn Fällen von Labilität bleibt die Orientierung auf das PATIENS (=THEME) erhalten, nur die Valenz oszilliert zwischen ein- und zweistellig, wobei die Agens-Phrase als fakultative Valenz bezeichnet werden muß. Andererseits hat Haspelmath (1987:4) auch die englischen Lexeme break (intr.) vs. break (tr.) als labil bezeichnet, was im übrigen wiederum bei Nedjalkov (1969) der Kennzeichnung durch Ν (= neutral) entspricht. Das Englische ist trotz fehlender Morphologie für die Rollen AG und PAT akkusativisch, wie die von Nedjalkov zitierten Sprachen Tamil, Japanisch, Ungarisch etc. (alle mit N-Kennzeichnung bei break). Bei einem Vergleich von Lezgisch und Englisch in 3a,b) wird aber klar, daß sich bei engl break nicht nur die Valenz labil verhält, sondern auch die Orientierung, d.h. im transitiven Satz hegt, im Unterschied zum Lezgischen, keine Patiensorientierung mehr vor, sondern eine Agensorientierung. Deutlicher wird dies, wenn man eine akkusativische Sprache mit Kasusmarkierung wählt. Dazu vergleiche man das deutsche Paar: (4a) Der Stock zerbrach.

vs.

(4b) Der Junge zerbrach den Stock.

Natürlich sind Nominativ sprachen dadurch ausgezeichnet, daß auch sie (wie der ABS in Ergativsprachen) in Gestalt eines intr. NOMinativs intr. PAT- und AG-Rollen gleich kodieren können. So kann dann neben 4a) mit einem intr. PAT auch 4c) mit einem intr. AG stehen: (4c) Das Kind lief Wählt man ein anderes Verb aus der Nedjalkov-Gruppe, das Verb kochen, so ergibt sich für: (5)

Der Missionar kochte

neben

(5a) Der Missionar kochte (etwas)

auch die aus Witzen bekannte makabere Version (5b) Der Missionar wurde gekocht (um von den Pygmäen verspeist zu werden)

76

Werner Drossard

Es bleibt festzuhaken, daß Verben vom Typ break/boil in NOM-Sprachen nicht nur valenzlabil sind, sondern auch potentiell orientienmgslabil sind, besonders wenn sie, wie in S), in intransitiven Kontexten auftreten können. In NOM-Sprachen, wie dem Engl, oder Deutschen, ist jedoch ein weiterer valenzlabiler Fall wie essen bekannt, der gleichzeitig nicht orientierungslabil ist. Im Deutschen und Englischen z.B. haben diese Verben eine fakultative Objekt-Valenz, verhalten sich also spiegelbildlich zur labilen Variante von break in einer Ergativsprache wie Lezgisch: (6a) Das Mädchen aß.

vs.

(6b) Das Mädchen aß einen Apfel.

Nun zeigt ein Blick in verschiedene ERGATIV-Sprachen, deren ERG-Partizipanten sich wie Adjunkte verhalten, bzw. bei denen das tr. AG nicht im Verb pronominalisiert ist (das gilt, wie bereits gesagt, offensichtlich nicht fur das Tschuktschische), daß eine durchaus mögliche intransitive Verb+ABSohitus-Verbindung für essen eine Orientierungslabilität ergeben kann. Diese hegt in Gestalt des tonganischen Beispiels vor: (7)

Na'e Temp

kai essen

(7a) Der Fisch

fraß.

'ae ika' ABS.DEF Fisch

bedeutet:

(7b) Der Fisch wurde gefressen/gegessen.

Damit ist der a)-Version eine AG-, der b)-Version eine PAT-Ausrichtung zuzuschreiben. Andererseits zeichnet sich ab, daß im Falle der eat-Verben in NOM-Sprachen keine Orientierungs-Labilität vorliegt. Tabellarisch läßt sich das bisher Gesagte so zusammenfassen: + orientierungslabil

- orientierungslabil

NOM - Sprachen

ERG - Sprachen

boil

intr. NOM fur AG und PAT

ABS immer fur PAT

eat

ERG - Sprachen

NOM - Sprachen

intr. ABS fur AG und PAT

NOM immer für AG

77

Labile Konstruktionen

Somit gilt unter Berücksichtigung der tabellarisch aufgeführten Fälle: - Im Fall des όο/7-Lexems liegt im Deutschen bei der Konfiguration kochen +NOM im intransitiven Fall eine Labilität in der Orientierung vor. Bei eat-Verben (essen, trinken, bohren) weist der NOM sowohl in ein- wie auch in zweiwertiger Verwendungsweise eine AG-Orientierung vor, so daß nur eine Valenzlabilität vorliegt. - In ERG-Sprachen zeigt bei Verben des break-Typs der ABS in ein- wie in zweiwertiger Verwendungweise eine PAT-Orientierung an. Die in Frage kommenden Verben sind nur valenzlabil. Bei den Verben des eat-Typs kann der zentrale Kasus, der ABS, in einigen Sprachen im einwertigen Schema beide Orientierungen anzeigen. Letztendlich ergeben sich spiegelbildliche Verhältnisse. Was hat es damit auf sich und wie ist das zu erklären?

2. Zu den Orientierungsverhältnissen in NOM- und ERG-Sprachen In seinem Aufsatz "Orientierung" hat Serzisko (1991:273-308) gezeigt, daß in der linguistischen Literatur der letzten 15 Jahre Einigkeit über die Verhältnisse in NOM- und ERG-Sprachen herrscht. So werden zitiert: Comrie (1981): "Active-accusative implies A-orientation, whereas the passive-ergative implies P-orientatiori". Mallinson / Blake (1981): "A number of linguists over a long period of time have noted that ergative-languages are patient-oriented and that the ergative is peripheral relative to the patient". Sasse (1982): "Ergativ-Sprachen sind Patiens-thematisierend, (Nominativ-) Akkusativsprachen Agens-thematisierend". Charakterisiert man Verben der break- Gruppe als prototypisch PAT-orientiert oder, im Sinne von Perlmutter (1978:157ff), als "unaccusative", die der ea/-Gruppe als prototypisch AGorientiert bzw. als "unergative", dann ergibt sich: AG-Orientierung (NOM - Sprachen)

PAT-Orientierung (ERG - Sprachen)

AG-or. Verben

- orientierungslabil

+ orientierungslabil

PAT-or. Verben

+ orientierungslabil

- orientierungslabil

Es liegt auf der Hand, daß es bei prototypisch AG-orientierten Verben in AG-orientierten Sprachen keine "Probleme" gibt, ebenso wie bei prototypisch PAT-orientierten Verben in PAT-orientierten Sprachen. Die Kompatibilität ist ideaL Wenn aber prototypisch PAT-orientierte Verben mit potentiell zweistelliger Valenz in NOMSprachen in Erscheinung treten, dann ist auch, im Rahmen einer Labilität, eine AG-orientierte

78

Werner Drossard

Version möglich, entsprechend der Grundorientierung des Sprachtyps, und wenn prototypisch AG-orientierte Verben mit potentiell zweistelliger Valenz in ERG-Sprachen in Erscheinung treten, dann ist auch, im Rahmen einer Labilität, eine PAT-orientierte Version möglich, entsprechend der Grundorientierung des Sprachtyps.

3. Labilität im Kontext der Nominalisierung Ein immer wiederkehrendes Thema in der Linguistik — und dies theorienübergreifend — ist die Diskussion um die "genitivus subjectivus"- vs. "genitivus objectivus"-Lesarten bei Nominalisierungen, sei es, in der lateinischen Grammatik und Grammatiktradition der Fall "amor dei" oder in der Generativen Syntax der Fall "the shooting of the hunters" (Chomsky 1970). Neben weiterer Literatur aus dem Bereich der Generativen Grammatik sind in diesem Zusammenhang auch einige Beiträge aus der Cognitive Grammar (besonders Taylor 1994) zu nennen. Ungeachtet der verschiedenen Erklärungsansätze und Termini ist dennoch festzustellen, daß zur allgemeinen Argumentation Überlegungen zur Orientiertheit der Verben gehören.

3.1 Labilität zwischen subjektiver und objektiver Lesart in Nominativ-Sprachen Es zeichnet sich zunächst ab, daß, trotz der Präferenz einer objektiven Interpretation im Falle PAT-orientierter Verben in NOM-Sprachen auch AG-orientierte Fälle möglich sind. So fuhrt Taylor (1994: 215ff) fur das Englische u.a. an: (8a) The city's destruction continued for three days vs. (8b) The enemy's destruction will take many months to repair. In 8a) hegt der "Normalfall" vor, in dem bei der Nominalisierung des PAT-orientierten Verbs destroy der "prenominai possessor" (oder "preposed possessor") objektiv interpretiert wird, während in 8b) die agentiv-subjektive Interpretation vorliegt. Damit wäre dann letztlich eine zunächst aus dem Kontext gerissene Nominalphrase wie (8c) the enemy's destruction ambig und in unserem Sinne labil (der Feind kann zerstören und zerstört werden), wobei die Labilität wieder durch den Textzusammenhang disambiguiert bzw. aufgehoben werden kann. Chomskys legendäres Beispiel "the shooting of the hunters" ist bekanntlich auch orientierungslabil: "the hunters" kann agentiv wie patientiv verstanden werden. Betrachtet man im Kontrast dazu einige deutsche Beispiele, so fallt auf daß die für das Englische angeführte Restriktion, daß PAT-or. Interpretationen nur mit der im Lexem angelegten "afFectedness" konform sind (vgl. M. Anderson 1984), im Deutschen nicht generell zutrifft.

Labile Konstruktionen

79

Es ist bekannt, daß bei der Nominalisierung in vielen Fällen eine "Labilisierung" der Orientierung eintritt, die im Ausgangslexem nicht angelegt ist, so etwa bei einem einfachen AG-orientierten Lexem wie fragen, das bei einer Nominalisierung (und, wie wir sehen, nicht nur unbedingt in einer Genitivphrase) durchaus ein Oszillieren zwischen AG- und PATorientierter Interpretation an den Tag legt: (9) (9a) (9b)

Bei der Frage nach Zeugen zögerte er. kann heißen: Als er nach möglichen Zeugen fragte (fragen wollte) , zögerte er. vs. Als er nach möglichen Zeugen gefragt wurde , zögerte er.

Andererseits wird die Orientierungs-Labilität von patientiven Verben (zerbrechen, verbrennen etc.) im Deutschen in die Nominalisierung mit hineingenommen. Erkennbar wird die Labilität der nominalisierten Formen wiederum bei satzartigen ¿¿/-Konstruktionen. So läßt 10) zwei Interpretationen zu: (10) Beim Zerbrechen des Glases verletzte er sich. ( 10a) Als das Glas zerbrach (PAT-or), verletzte er sich. vs. (10b) Als er das Glas zerbrach (AG-or), verletzte er sich.

3.2

Zur Situation in Ergativ-Sprachen

In seinem Aufsatz von 1976 hat B. Comrie die Syntax von "action nomináis" in Sprachen verschiedenen Typs untersucht. Nach wie vor ist es mit einiger Mühe verbunden, zu unserer speziellen Fragestellung genügend Informationen zu bekommen, doch Comries Erörterungen fähren zu einer für unser Projekt interessanten Generalisierung. Im allgemeinen, d.h. zunächst ohne direkten Bezug auf NOM- vs. ERG-Sprachen, kristallisiert sich heraus: a) wie die Fälle Deutsch und Englisch (aber auch, so Comrie, das Arabische) zeigen, wird in bestimmten Sprachen durch die Neutralisierung der subjektiven und objektiven Lesart in einer Genitiv-Konstruktion eine Orientierungslabilität erzeugt. b) in einigen Fällen jedoch werden diese Neutralisierung und Labilisierung vermieden, indem entweder die Kasusrektion des Verbs mit in die Nominalisation übernommen wird, so daß AG und PAT-Kodierung eindeutig sind, oder im Verbalnomen wird eine AGPAT-Opposition markiert (Usbekisch). Innerhalb der Ergativsprachen zeichnet sich in den meisten eruierbaren Fällen ab, daß die verbale Kasusrektion des transitiven Satzes, das ERG-ABS-Schema fiir AG- und PAT-Relationen, mit in die Nominalisierung übernommen wird. Comrie (1976:181) gibt zwei Beispiele aus ostkaukasischen Sprachen, von denen wir das Awarische wiederholen:

80

Werner Drossard

"You are reading the book'.11

(Ila) Du-ca tex çalula 2sg-ERG book read-PRÂS Mit dieser imiten Konstruktion ist zu vergleichen: (IIb) Du-ca tex çal-i 2sg-ERG book read-VN

bugo çaq likab is is very good thing

ir

'Your reading the book is a very good thing"

Neben das von Connie zitierte zweite Beispiel aus dem Tabassaranischen können wir als dritte Sprache aus dem Daghestan das Legische stellen. Haspelmaths Ausführungen (1993a: 153) zur Syntax des Verbalnomens (Masdar) in dieser Sprache lassen eindeutig Parallelen zu den beiden anderen Sprachen erkennen. So wird bei der Verbindung von Partizipant + Masdar die vom imiten Verb geforderte Kasusmarkierung beibehalten: (12)

muhman-ar atu-n guest-ABS:PL come-MSD

"the guests'arrival"

Im Falle einer Transformation einer transitiven Konstruktion in eine nominalisierte Fassung bleibt die Kasusrektion derfiniten Version ebenfalls erhalten: (13)

gada.-di ktab-0 kxi-n boy-ERG book-AB S write-VN

"the boy's writing of a book"

Auch im nicht-kaukasischen Kontext ist diese Erscheinung der Transposition der verbalen in die nominale Rektion zu beobachten, so etwa im Baskischen. In dieser Sprache wird durch das Suffix -tel-tze nominalisiert: (14)

etor-tze-a come-NMS-DEF:sg:ABS

"the coming"

Einem englischen Satz wie "That you tell me lies surprises me" entspricht im Baskischen: (15)

zu-k ni-ri gezurr-a esa-te-ak 2sg-ERG lsg-DAT lie-sg:ABS tell-NMS-sg:ERG harri-tzen η — a — u — 0 suiprise-hab lsg~pres~aux~3sg ABS ERG

81

Labile Konstruktionen

Hier wird das Agens zur Nominalisienmg esateak im ERG beibehalten: zuk (2sg:ERG). Nun zeigen alle diese Beispiele zwar, daß die verbale in die "verbalnominale" Kodierung übernommen wird, ambige und wirklich labile Fälle aber werden dort nicht erörtert. Eine Übertragung des Chomsky-Beispiels ins Baskische schließlich erläutert den Sachverhalt besser: (16)

ehiztari-en hunter-GEN:PL

tiro egi-te-a shot do-NMS-ABS:DEF

"the shooting of the hunters" (agentiv)

Hier wird zwar aus der entsprechenden finit en Verbalversion kein Ergativ (für das AG) mit übernommen, aber die Konstruktion aus Genitiv und Verbalnomen ist eindeutig aktivisch, d.h. AG-orientiert. Eine PAT-orientierte Version, die die Jäger als Ziel des Schießens ausweist, ergibt sich durch den Wechsel des Genitivs zum Dativ: (17)

ehiztari-ari hunter-DAT:PL

tiro egite-a shot do-NMS-ABS:DEF

"the shooting of the hunters" (patientiv)

Das schönste Minimalpaar aber ergibt sich im Samoanischen. Hier sind die beiden Possessivklassifikatoren a vs. o bekannt, deren semantische Distribution sich grob in "kontrolliert, agentiv, etabliert" einerseits vs. "nichtkontrolliert, patientiv, inhärent" andererseits umschreiben läßt. (18a) 'o se NMP INDF

tali a answer PKL1

Matautu

"an answer given by Matuatu" (AG-or.)

(18b) 'o se NMP INDF

tali o answer PKL2

Matuatu

"an answer given to Matuatu"(PAT-or.) (Breidbach 1983:17)

Erklärungsversuche fur die Übernahme der verbalen Rektion in die des nommalisierten Pendants bzw. die strikte Trennung der Orientierungsopposition können mithilfe der "Role and Reference Grammar" (Foley/van Valin) unternommen werden. Wenn man die Nominativsprachen Deutsch, Englisch (und andere) bezüglich des Status ihrer grammatischen Kategorien untersucht, so weiß man, daß diese Sprachen im Sinne von Foley und van Valin (1980, 1984) bzw. van Valin (1977, 1980) über einen "referential peak" bzw. "pragmatic peak" verfugen, der die konträren Rollen AG und PAT integriert und syntaktisch gleich, d.h. als Subjekt, behandelt. Dies zeigt sich besonders bei Koordinationen von agentiven und patientiven Satzreihen, wie sie z.B. in Foley/van Valin's Beispiel deutlich wird: (19) John went to work, and promotion (by his boss).

talked to his boss and

was given a

82

Werner Drossard

Her kann der nicht-agentive, im Passiv stehende letzte Teilsatz problemlos angeschlossen werden, da das "reference tracking" bzw. das Konstanthalten des Hauptpartizip anten John über ein emheitiches Subjekt gewährleistet ist. Der Diathese-Wechsel ("switch function") garantiert, ungeachtet der Agentivität oder Nicht-Agentivität der Partizipanten (im letzten Teilsatz ist boss agentiv und nicht mehr John), daß kontinuierlich John als Topic figurieren kann. Ein (neben einigen anderen) entscheidendes Kriterium für "Subjektheit" liegt somit in der Integration konträrer Rollen in eine syntaktische Größe begründet. Während nun die referenzdominierten Sprachen Englisch und Deutsch über diese Art von Subjekten verfügen, die als "targets" von Passivierungen auch die Topikalität von PAT-Rollen gestatten, weisen anderseits "rollendominierte" Sprachen, die nur über "semantic pivots" verfugen, gerade bei koordinativen Strukturen erhebliche Diskrepanzen auf) weil ein einheitliches Subjekt fehlt. In "rollendominierten" Sprachen z.B., die über "switch reference'-Mechanismen verfugen, müßte hier ohnehin mit dem Wechsel der Agentivität von einem auf einen anderen Partizip anten (z.B. von John auf the boss) eine "different subject"-Markierung erfolgen. Neben dem Subjekt als integrativer Größe für AG und PAT (in Aktiv- und Passivsätzen) ist es als zweite Kategorie der Genitiv (im Engl der "preposed possessor" mit sächsischem Genitiv oder die postponierte o/-Phrase, im Deutschen ein Genitiv oder eine von-Phrase), der die konträre AG vs. PAT-Semantik integriert. Der Genitiv ist damit neben einem Subjekt, das im Deutschen im Nominativ steht und damit greifbarer ist als im Englischen, eine weitere typische rollennivellierende Kategorie. Betrachtet man aber die oben erwähnten Ergativsprachen, so weiß man aus der zur Verfügung stehenden Literatur, daß die zitierten Sprachen über einen problematischen Subjektbegriff verfugen. Am klarsten hat Haspelmath (1993a: 15.2 und 16ff) für das Lezgische eine Rollendominanz postuliert, so daß wir, auf der Basis dieser sehr gut dokumentierten Sprache, mit der Projektion der im Lezgischen zu beobachtenden Verhältnisse auf die anderen ostkaukasischen Sprachen und unter Einbezug der Erkenntnisse zum Baskischen (vgL Brettschneider 1979) und Samoanischen (Mosel/Hovdhaugen 1992:772fF.) die Tendenz formulieren können, daß gerade in all diesen rollendominierten Sprachen, analog zur Unmöglichkeit, daß in einem Subjekt konträre AG- und PAT-Semantik untergebracht werden können, auch die Unmöglichkeit festzustellen ist, daß AG- und PAT-Semantik in einem Genitiv integrierbar sind. Somit korreliert mit der Rollendominanz der genannten Ergativsprachen deren "semantikgetreue" Kodierung in nominatisierten Kontexten, so daß mit der Abwesenheit einer nivellierenden Genitivkonstruktion (oder ähnlich nivellierenden Verfahren) die Abwesenheit der Labilität von nominalisierten Verben gekoppelt ist.

Labile Konstruktionen

83

Abkürzungen A=Absohitus am Verb, ABS=Absohitus, AG=Agens, AG-or-=agens-orientiert, AKK=Akkusativ, AOR=Aorist, aux=auxiliar, DAT=Dativ, DEF=definit, E=Ergativ (am Verb), ERG=Ergativ, hab=habitualis, INDF=indefinit, intr.=intransitiv, MSD=Masdar, NK=Nominalklasse, NOM=Nominativ, NMP=Nominalphrasenmarkierer, NMS=Nomina- lisierung, PAT=Patiens, PAT-or.=patiens-orientiert, PKL=Possessivklassifikator, PRÄS=Präsens, PRÄT=Präteritum, s=Singular, TEMP=Tempus, tr.=transitiv, VN=Verbal- nomen (verbal noun)

Literatur Anderson, M. (1984) Prenominai genitive NPs, The Linguistic Review 3, 1-24. Breidbach, W. (1983) Zur Possession im Samoanischen. Köln: Inst, für Sprachwissenschaft. (Arbeitspapier 42 (alte Folge)). Brettschneider, G. (1979) Typological characteristics of Basque, in: F. Plank (ed), Ergativity. London etc.: Academic Press, 371-384. Chomsky, N. (1970) Remarks on nominalization, in: R. A. Jacobs and P. S. Rosenbaum (eds), Readings in English Transformational Grammar, Waltham MA: Ginn & Co., 184-221. Comrie, B. (1976) The syntax of action nomináis: A cross-linguistic study, Lingua 40, 117-201. - (1981) Aspect and Voice: Some Reflections on Perfect and Passive, in: P. J. Tedesci and A. Zaenen (eds), Syntax and Semantics 14, Tense and Aspect. New York etc.: Academic Press. Drossard, W. (1991) Veibklassen, in: H.Seiler/W.Premper (eds), Partizipation, Das sprachliche Erfassen von Sachverhalten. Tübingen: Gunter Narr, 150-182. Foley,W.A./ van Valin jr. R.D. (1984) Functional Syntax and Universal Grammar. Cambridge: Cambridge UP. Haspelmath, M. (1987) Transitivity Alternations of the Anticausative Type. Köln: Inst, für Sprachwissenschaft. (Arbeitspapier 5 (neue Folge)). - ( 1993a) A Grammar of Lezgian. Berlin: Mouton-de Gruyter. - (1993b) More on the typology of inchoative/causative verb alternations, in: B. Comrie and M. Polinsky (eds), Causatives and Transitivity. Amsterdam - Philadelphia: Benjamins. Inènlikej, P.I./ Nedjalkov, V.P. (1967) Iz nabljudenij nad èrgativnoj konstrukciej ν Èukotskom jazyke in: V.M. ¿irmunskij et al. (eds), Êrgativnaja konstrukcija predloïenija ν jazykach razlicnych tipov. Leningrad: Nauka, 246-260. Keenan, E. (1976) Toward a Universal Definition of , in: C. Li (ed), Subject and Topic. New York etc.: Academic Press, 303-333. Lazard, G. (1994) L'actance. Paris: Presses Universitaires de France. Mallinson, G./ Blake, B.J. (1981) Language typology, Amsterdam etc.: North Holland. Mosel, U./ Hovdhaugen, E. (1992) Samoan Reference Grammar. Oslo: Scandinavian University Press. Nedjalkov,V.P. (1969) Nekotorye verojatnostnye universal» ν glagol'nom slovoobrazovanii, in: I.F. Vardul , I.F. (ed), Jazykovye universalii i lingvistiíeskaja tipologija. Moskva: Nauka, 106-114. Perlmutter, D. M. (1978) Impersonal passives and the unaccusative hypothesis, BLS 4, 157-189. Saltarelli, M. (1988) Basque. London etc.: Croom Helm. Sasse, H.J. (1982) Subjektprominenz, in: Festschrift Stimm, Fakten und Theorien. Tübingen: Gunter Narr, 267286. Serzisko, F. (1991) Orientierung, in: H. Seiler und W. Premper (eds), PARTIZIPATION. Tübingen: Gunter Narr, 273-308.

84

Werner Drossard

Taylor, J. (1994) Subjective and objective readings of possessor nomináis, Cognitive Linguistics 5 (3), 201-242. Van Valin jr., R.D. (1977) Ergativity and the Universality of Subjects, CLS 13, 689-705. -

(1980) On the Distribution of Passive and Antipassive in Universal Grammar, Lingua SO, 303-327.

Van Valin jr. and W.A. Foley (1980) Role and Reference Grammar, in: E. Moravcsik and E. Wirth (eds), Syntax and Semantics 13. New York etc.: Academic Press, 329-350.

Inverse Direction and Second Person in Dumi Theodora Bynon

1. Introduction In his recently published Grammar of Dumi, a dying Tibeto-Burman language of the Himalayas used today by only a dwindling minority of elderly speakers in a small area to the south of Mount Everest, van Driem (1993:123) divides the inflected forms of the simple verb into two sets, those which are formally unmarked and those which are marked by the prefix a-.1 The distribution of the members of each of these two sets in intransitive and transitive clauses is as follows (the direction of the action being indicated by an "acting upon" arrow which runs from the agent on the left to the patient on the right): Unmarked (no prefix)

Marked (a- prefix)

Intransitive verb, Subject:

1,3

2

Transitive verb, Agent -> Patient:

1 -» 2 1 -> 3 3 —» 3 2 -» 3

2 1 3 —> 1 3 —» 2

This pattern, van Driem says, reflects "a pronominal markedness hierarchy of pragmatically more vs. less obvious scenarios, at least to the mind of the author if not in the mind of the native speakers of Dumi" (1993:123). Ever since I first came to grips with the complexities of Dumi when I was reviewing van Driem1 s Grammar (Bynon 1997), I have been interested in further exploring this distribution. The markedness hierarchy to which he makes allusion is undoubtedly the so-called "animacy hierarchy" which was first proposed by Silverstein and which arranges referential categories in a continuum, ranging from first and second person pronouns on the left to inanimate common nouns on the right (Dixon 1994:85): 1st person > 2nd person > 3rd person > proper noun > common noun (human > animate > inanimate) This animacy hierarchy is the determining factor in a variety of split ergative systems within the grammars of certain American Indian, Australian and Siberian languages (Dixon 1994:83-94, 1

I am grateful to Leonid Kulikov, George van Driem and James Bynon for their comments on earlier versions of this paper. Naturally they are not responsible for any misrepresentations.

86

Theodora Bynort

Nedjalkov 1979), although the precise location of the split varies from language to language. It is also at the root of various kinds of direction marking systems in certain American Indian and Sino-Tibetan languages which distinguish a "direct" and an "inverse" category (Thompson 1994, Givón 1994a,b) depending upon whether the agent-patient constellation of a transitive verb conforms to the above animacy hierarchy or goes against it. In the Dumi categorisation represented above the semantic principle held responsible for direct-inverse systems appears not to hold in the case of the 2 - » 3 constellation which, although semantically direct, figures in the inverse column on account of its prefix. One therefore wonders whether there are other factors involved in the distribution of the prefixed a-. One possible clue is found in Table 1, which shows that second and third person are not formally differentiated when they interact with first person and only minimally so when they interact with one another. The basic system of verbal inflection is thus two-term, opposing first to second and third person. Another clue may be more clearly seen from Table 2, which points to an association of prefixed a- not only with inverse direction but also with second person quite irrespective of transitivity.

2. The direct-inverse opposition and prefixed a2.1 Basics of Dumi morphosyntax Dumi has predominantly nominative-accusative syntax but it has ergative-absolutive case marking on nouns and independent pronouns. Whether or not an independent pronoun is in fact present in any given clause is a matter of discourse pragmatics, the person and number affixes of the verb being themselves referential, and for this reason the term "cross-referencing" would perhaps be preferable to "verb agreement". The grammar distinguishes three persons (first, second and third), three numbers (singular, dual and plural) and, in the non-singular forms of the first person, inclusion versus exclusion of the addressee. These three sets of oppositions are reflected in the independent personal pronouns as well as in the morphology of the verb.2 The intransitive verb agrees with its single actant (S, subject), the transitive verb agrees in principle with both actants, agent (A) and patient/object (O), but there are various constraints such that non-singular number for instance can be encoded only once in any one form The contrast between second and third person is largely neutralised when the other actant is first person, so that the fundamental opposition is thus between first and non-first person. There are accordingly only two sets of dual and plural suffixes, first person having -/', -/', -ki, 2

All the following sentences, including their translations and in most instances their grammatical glosses, are taken from van Driem (1993); references are to page and example number. Additions in square brackets and occasional changes to the grammatical glosses are my own and are indicated in the text. I alone am responsible for any possible misanalyses and mistakes. An arrow is employed to represent the direction of the action from A to O of a transitive clause. The tense system distinguishes preterite (unmarked) and nonpreterite (marked by an infixed -/-). ABS = absolutive case (unmarked), ERG = ergative case, Pt = preterite, NPt = non-preterite; 1,2,3 = first, second and third person; du/in = dual inclusive; INV = inverse.

87

Inverse Direction and Second Person in Dumi

-ka, second and third persons having dual intransitive -i, dual transitive -si, and plural -ini. A distinct second person dual suffix -i is found only in those cases when second and third person interact. The singular suffix -a is common to both second and third person in the non-preterite. The possible agent-patient constellations, extracted from van Driem's paradigms, are given below in Table 1. Table 1: Possible agent-patient constellations Patient

• 2sg

2du

2pl

3sg

3du

3pl

lsg

-n-a

-n-si

-n-ni

-a/u

a/i-si

a/i-ni

ldu/in

-i

-i

-i

ldu/ex

-i -i

-i

-i

-i

-i -i

-i -i

lpl/in

-ki

-ki

-ki

-ki

-ki

-ki

lpl/ex

-ka

-ka

-ka

-ka

-ka

-ka

lsg

Agent •

ldu/in

ldu/ex lpl/in

lpl/ex

2sg

a-a

a-i

a-si

a-ni

a-a-si

a-i

a-ka a-ka

a-a/i

2du

a-i

a-i

a-i

2pl

a-a-ni

a-i

a-ka

a-ni

a-ni

a-ni

3sg

a-e

a-i

a-i

a-ki

a-ka

a-a

a-i

a-ni

-a/i

-si

-ni

3 du

a-a-si

a-i

a-i

a-ki

a-ka

a-a

a-i

a-ni

-si

-si

-ni

3pl

a-3-ni a-i

a-i

a-ki

a-i

a-ni

-ni

-ni

-ni

a-ka

a-a

2.2 First and third person In the following examples, (1) versus (2), and (3) versus (4) illustrate the direct-inverse opposition. The prefix is present when the patient outranks the agent (van Driem's "marked scenario"), as in (2) and (4), where it indicates an inverse constellation. It is absent in the direct construction, when the agent outranks the patient. This marking system thus implies that the speaker is the natural subject of the clause whereas a third person agent acting upon a first person patient is a marked event indicated by prefixed a-. From the perspective of the verbal suffixes, the prefix marks a reversal of the semantic roles of their referents. The morphologically simplest verb forms include a third person singular actant, which lacks an overt exponent. For a third person singular patient this yields a nominative-accusative agreement pattern, as in (3), and for a third person singular agent an ergative agreement pattern, as in (4). These latter regularities are displayed in Table 2.

Theodora Bynon

88

(141:87)

(1)

lsg->3pl:

him-t-3-m search-NPt-1 sg-2/3pl Ί am looking for them'

(2)

3pl

lsg:

tsi:mo-mil-?a aq a-ka:ts-a-ni deity-pl-ERG I-ABS INV-bite-lsg-2/3pl "They say the deities have afflicted me'

(3)

lsg

3sg:

aq-a im de:?e phot-t-a I-ERG he+ABS summon call-NPt-lsg Ί will call him'

(65:23)

(4)

3sg —> lsg:

im-a aq a-ka:ts-a he-ERG I+ABS INV-bite-lsg 'He bit me'

(81:97)

?e (141:88) REPORT

Table 2: Verb affixes of the intransitive paradigm and of transitive forms where one of the actants is 3sg Intransivitive

S

ldu/in

Transitive 3sgA

All persons A

with all persons O

with 3sgO

a—a a—i

—o / -u

—i

ldu/ex

—ι

a—i

lpl/in

—k-i

a—k-i

—k-i

lpl/ex

—k-a

a—k-a

—k-a

a—a a—i

a—a a—i

a—a/-i

2du 2pl

a—ini

a—ini

a—ini

3sg

—a —i

—a/-i

—a/-i

3du

—si

—si

3pl

—ini

—ini

—ini

ham—a

a—i

89

Inverse Direction and Second Person in Dumi

2.3 Second person We have seen that van Driem allocates all forms marked by the prefix a- to the inverse category and that this category includes all instances of second person agency irrespective of the direction of the action. This latter association of the prefix is clearly seen from Table 2 while examples (5) to (8) illustrate the various constellations involving second person actants. (5a) 2sg -> 3sg:

an-a im a-ka:d-i you-ERG he+ABS 2-bhe-(2/3 You bit him'

(81:100) 3sgPt)

(5b) 3sg -> 2sg:

simar-?a an a-luph-a? illness-ERG you+ABS INV-seize+Pt-2/3sg 'Have you taken ill?

(228:17)

(5c) 3sg—>3sg:

im-a bili tsa:m-i he-ERG money+ABS lose-(2/3-»3sgPt) 'He lost the money1

(144:104)

(6)

hempa a-khus-t-ini? where 2-go-NPt-2/3pl "Where are you (pi) off to7

(117:4)

2plS:

(7a) 2sg

3sg:

an-a a-ram a-òri you-ERG your-body+ABS 2-wash+(2/3 -» 3sgPt) •You took a bath'

(85:paradigm)

(7b) 3sg

3sg:

im-a i-ram siri he-ERG his-body+ABS wash+(2/3 -> 3sgPt) 'He took a bath'

(85:paradigm)

2sg -> lsg:

[an-a] arj da:p a-mits-a. [you-ERG] I+ABS feel INV-do-lsg You made me feel it.1

(224:3)

lsg —»· 2sg:

[aq-a] an ya da:p mit-n-t-a I-ERG you+ABS too feel do-( l s g 2 ) - N P t - 2 / 3 s g 'Now I'm going to make you feel it too.'

(8)

Examples (5a) and (5b) show a reversal of the semantic roles of the referents of agent and patient not unlike that in (1) vs. (2) and (3) vs. (4), but it is to be noted that both verbs carry an α-prefix; their suffixes are furthermore identical in the non-preterite. The question is whether in

90

Theodora Bynon

(Sa) the prefix should be interpreted not as a marker of the inverse but rather of second person, as it is in what is semantically a direct constellation, and that only in semantically inverse constellations is the prefix an instance of inverse marking. The fact that the two prefixes are homophonous would then be purely coincidental. Further evidence for a possible second person prefix comes from examples (6) and (7a). Example (6) is intransitive and the direct-inverse opposition would seem inapplicable. (7a) is both structurally and semantically parallel to (7b): the agent and possessor are coreferential in both, neither clause is semantically inverse, and the only feature which distinguishes them is the presence or absence of the prefix. In example (8), on the other hand, the direction of the 2 1 constellation is semantically inverse and the verb has the same first person ergative agreement pattern as 3sg lsg in (4), a fact which I had previously failed to notice (Bynon 1996). All 3 -> 3 constellations, as for instance (5c) and (7b), are systematically prefixless. The crucial constellation is 2 -> 3 which, although semantically direct, is prefixed by a-. It can be seen from Table 1 that both 2 3 and 3 -> 2 have second person dual agreement marked by the second person specific suffix -/. The distribution of this latter suffix is significant, since it marks the agent in 2 -» 3 but the patient in 3 -> 2, implying a nominative-accusative agreement pattern for the former and an ergative agreement pattern for the latter. We would therefore appear to find here traces of a morphosyntactic contrast between these two constellations. It would seem that, at some point in time, the prefix marking second person has been conflated with that marking inverse direction. The presence of prefixed a- in second person intransitives as well as in the semantically direct 2 3 would be inexplicable if there were only an inverse prefix. Table 2 effectively displays the association of the a- prefix with inverse direction on the one hand and with second person participation on the other. The frequent non-differentiation of second and third person is to some extent compensated for by a number of constellation specific markers: -u and -/' mark lsg —> 3Pt and 2/3 3sgPt respectively, as in example (9); lsg —> 3 has, under certain specific phonological conditions, as for instance in example (10), an infixed-/;- (van Driem 1993:133-4) which is etymologically a first person pronomi; lsg—> 2 has a special marker -«- related to the second person pronoun an (as is perhaps agreement/possessive second person a-?) The use of first and second person pronouns in marking the latter two constellations is found in related languages (DeLancey 1981, Ebert 1987). (9)

2sg -> 3sg:

itsi η» a-tsikt-i ? truly EMPH 2-understand-(2/3 -> 3sgPt ) 'Did you really understand?

lsg —> 3sg:

tsikt-u understand lsg -> 3Pt Ί did.'

(149:124)

91

Inverse Direction and Second Person in Dumi

(IOa) lsg -»• 3:

dza-q-t-a eat-(lsg -)· 3)-NPt-2/3sg TU eat it'

(133:49)

(10b)3 -»· lsg:

a-be:-so-q-3-ni INV-give-dispatch-( 1 sg -> 3)-lsg-2/3pl 'They gave it away to me'

(133:51)

We conclude that a case can thus be made for differentiating second person a- from inverse marking a- on semantic grounds, thereby removing second person intransitive forms together with the 2 -> 3 constellation from the inverse category. But this is a costly solution, as it involves the setting up of two homophonous prefixes. Cross-linguistic studies (Givón 1994a,b; Croft 1996) do not appear to report any problems with second person so that the morphological connection between inverse and second person appears to be specific to Dumi (and probably other Tibeto-Burman languages; cf. DeLancey 1987:808).

3. Function and categorisation of the inverse forms We have not so far discussed the place occupied by the direct-inverse opposition in the grammar of Dumi. DeLancey (1981,1987) says that such direction marking systems are found in a number of Tibeto-Burman languages and that they represent an interaction between what he terms attention flow and view-point, Le. between the linear structure of the clause and the natural propensity of a noun phrase for subject role. More recent work by Givón and others (Givón 1994a, Thompson 1994, Payne et al. 1994, Croft 1996) incorporates such direction marking systems into the domain of grammatical voice. Their typology of voice systems comprises both direct-inverse and active-passive systems. The relationship of the inverse vis-à-vis the direct construction is compared with that of the passive vis-à-vis the active in terms of markedness and, perhaps rather more surprisingly, also of function. From the perspective of Dumi such a comparison appears at first sight somewhat far fetched since the two types of system show fundamental differences. In the first place the prototypical passive is intransitive and agentless whereas the Dumi inverse is fully transitive and "agentfiil". Secondly, whereas in the proto-typical passive the patient is promoted to the status of grammatical subject, it is to all appearances the agent which has subject status in the Dumi inverse, just as it has in the direct construction. Thirdly and perhaps most importantly, whereas selection of passive voice tends in general to be driven by considerations of discourse structure, selection of the Dumi inverse is absolute because it is grammatically controlled. Givón's arguments for the existence of structural and semantic affinities between the two types of voice system derive in the first instance from discourse pragmatics. What unites directinverse with active-passive systems is, he claims, the "relative topicality" of agent and patient (1994a:9ff ). This is a quantitative judgement and the degree of topicality is arrived at by measuring the ability of the respective actants to hold on to the topic role, or to resume this

92

Theodora Bynon

role, in the course of a text. In those systems which have a "pragmatic inverse", as is the case in certain Algonquian languages, the inverse construction is preferred when of two third person actants the patient is considered to be more topical than the agent. This pragmatic inverse is certainly closer to a traditional passive than is, at least on the face of it, the obligatory "semantic inverse" controlled by the person hierarchy which we find in Dumi. Givón argues that pragmatic and semantic inverses are, nevertheless, related because there are languages, again within the Algonquian family in particular, which employ one and the same morphological form to encode both these functions (Givón 1994a:22-3). He identifies the factor they share as "norm reversal" in relation to the expected topicality of agent and patient: the pragmatic inverse is claimed to involve norm reversal in the feature of agentivity (manifest in the hierarchy agent > dative >patient), the semantic inverse in the person hierarchy. Thus, on the one hand, I was cheated (by him) has a patient subject by comparison with I cheated him, which has the expected subject, much as Dumi a-ka:ts-3 'He bit me' by comparison with ka:ts-a Ί bit him'. What the pragmatic and semantic inverses have in common is thus the fact that their selection is controlled by the animacy hierarchy, albeit by different dimensions of it. For Croft (1996) the constraints governing the selection of one or other member of the voice category form a cline which runs from obligatory constraints laid down in the grammar to relatively free discourse driven choices. He notes, as others have done, that even in English the selection of active and passive voice is not entirely free, and that constraints relating not only to topic continuity but also to animacy and person are relevant (DeLancey 1981:638, Croft 1996:3). Thus, while first person I is a natural subject in both active and passive clauses, the first person agentive phrase by me is generally much less acceptable. No such constraints, however, apply to the third person. Givón tentatively subsumes the entire range of passives and inverses under the heading of "de-transhive voice". We have, however, seen that the "semantic inverse" of Dumi can be considered as less transitive than the direct only to the extent that a subject lower on the person hierarchy is inherently less topical than one situated higher up. The characterization as detransitive applies rather to the Dumi reflexive, which shares with simple intransitives the nontransitive third person plural prefix ham- (see Table 2). The stative variants of the reflexive furthermore have their subject in the absohitive case, just as simple intransitives do, whereas the "active" reflexive has its subject in the ergative case. (lla)hammil ham-kam-sti they+ABS 3plS-get.dressed-REFL+NPt+2/3sg "They are dressed up'

(231:27)

(llb)intsi-?a intsi-ho:p qo kam-si-sti (231:26) we [du/in]-ERG our [du/in]-self EMPH get.dressed-REFL·REFL+NPt+ldu/m 'Let's get ourselves dressed'

93

Inverse Direction and Second Person in Dumi (12) hammil

khab i p

they+ABS all

ham-ho.-t-a

(64:16)

EMPH 3plS-come-NPt-2/3sg

'They will all come' The inverse is formally and semantically distinct from these de-transitive constructions and its origins in direction marking can still be clearly seen (c£ DeLancey 1981 on direction systems elsewhere in Tibeto-Burman). There is certainly no trace so far of the patient of the inverse clause being promoted to subject. Dumi thus gives us a unique insight into the genesis of a new voice category.

References Bynon, Theodora (1996). How Ergative is Dumi? SOAS Working Papers in Linguistics and Phonetics 6, 37184. Bynon, Theodora (1997). Review of van Driem (1993), Bulletin of the School of Oriental and African Studies 60, 171-3. Comrie, Bernard (1979). Degrees of Ergativity: Some Chukchee Evidence. In Plank 1979, 219-40. Croft, William (ms). The Voice Continuum. (Hand-out to Paper presented to the Philological Society, London, October 1996). DeLancey, Scott (1981). An Interpretation of Split Ergativity and Related Matters, Language 57, 626-57. DeLancey, Scott (1987). Sino-Tibetan Languages. In Bernard Comrie (ed.), The World's Major Languages. London & Sydney. Croom Helm. Dixon, R. M. W. (1994). Ergativity (Cambridge Studies in Linguistics, 69) Cambridge: Cambridge University Press. Driem, George van (1993). A Grammar of Dumi (Mouton Grammar Library, 10) Berlin & New York: Mouton de Gruyter. Ebeit, Karen (1987). Grammatical Marking of Speech Act Participants in Tibeto-Burman, Journal ofPragmatics 11,473-82. Fox, Barbara, and Paul J. Hopper (eds.) (1994). Voice: Form and Function (Typological Studies in Language, 27) Amsterdam/Philadelphia: John Benjamins. Givón, Talmy (1994a). The Pragmatics of De-transitive Voice: Functional and Typological Aspects of Inversion. In Givón 1994b, 33-44. Givón, Talmy (1994b). Voice and Inversion (Typological Studies in Language, 28) Amsterdam & Philadelphia: John Benjamins. Hockett, Charles F. (1966). What Algonquian is Really Like, 1JAL 32, 59-73. Nedjalkov, Vladimir P. (1979). Degrees of Ergativity in Chukchee. In Plank 1979,241-62. Payne, Doris (1994). The Tupí-Guaraní Inverse. In Fox and Hopper 1994, 31-40. Payne, Doris, Mitsuyo Hamaya and Peter Jacobs (1994). Active, Inverse and Passive in Maasai. In Givón 1994b, 283-316. Plank, Frans (ed.) (1979). Ergativity: Towards a Theory of Grammatical Relations. London & New York: Academic Press. Thompson, Chad (1994). Passive and Inverse Constructions. In Givón 1994b, 47-63.

On Patient Demotion in Lak* Konstantin Kazenin

0. Introduction. Among voice alternations attested in ergative languages, antipassive surely is one of the most widely known. Cooreman (1994:50) sums up the accepted understanding of antipassive in the following way: "While the A and the O in an ergative clause are marked as ergative and absolutive respectively, the A in an antipassive is typically coded as an absolutive NP, and the O (if present) appears in a case other than the absolutive."

Consider ergative (la) and antipassive (lb) clauses in Chamorro (Cooreman 1988:578): (1)

a. un-patek i ga'lago ERG.2SG-kick the dogABS You kicked the dog.' (Agent - Erg, Patient - Abs) b. ma-matek hao gi ga'lago APASS-kick 2sg.ABS LOC dog You kicked at the dog.' (Agent - Abs, Patient - Loc)

Extensive research on antipassive in various languages has revealed as one of its basic functions "backgrounding" of the Patient, or marking of its demoted pragmatic status (see Heath 1976, Givón 1984, Cooreman 1994). Backgrounding of the Patient was reported as the central function of antipassive in a considerable number of genetically diverse ergative languages, including Mayan languages (England 1988), Chukchee (Kozinsky, Nedjalkov, Polinskaya 1988), Chamorro (Cooreman 1988), Nez Perce (Rude 1988), Groenlandie Eskimo (Bittner 1987) and some others. Another function of antipassive - promotion of the Agent into the syntactically most privileged position marked with absolutive case - appears to be the central function of antipassive only in certain "syntactically ergative" languages and will not be of interest for the present paper1. Being the most widespread means to background the Patient, antipassive demonstrates a clear relation between form and function, because the backgrounded participant receives an oblique case (if not deleted). In this light, it looks especially strange that there is a construc-

* I would like to thank A.E.Kibrik, L.I.Kulikov and Tom Payne for helpful comments on an earlier draft of this paper. All remaining mistakes are my own, of course. 1

But see Cooreman 1994:75-81 for an attempt to discover a functional relation between antipassive of this type and antipassive used for demotion of the Patient.

96

Konstantin Kazenin

tion attested across languages which also backgrounds the Patient, but does not resemble antipassive from the formal perspective. This construction changes the case marking of the Agent from ergative to absolutive, but does not assign the Patient an oblique case. This results in socalled "biabsohitive" construction, where both the Agent and the Patient bear absolutive case marking. Biabsolutive constructions have not been studied as extensively as the "canonical" antipassive, probably because of their relatively low frequency. However, they consistently occur in languages of the Nakho-Daghestanian family of the Caucasus (see e.g. Kazenin 1996a for Godoberi), and also in Basque (Rebuschi 1989). If biabsolutive constructions function like antipassives, the question will immediately arise as to why a language may choose to "background" the Patient by changing the case of the Agent. The present paper attempts to answer this question for one language of the NakhoDaghestanian family - Lak. In Lak the biabsolutive construction (BC) is remarkably more frequent than among its relatives, so that probably Lak is the "champion" in biabsolutive case marking among languages of the world. Section 1 suggests an overview of the biabsolutive construction in Lak, considered against the background of the ergative construction (EC) in that language. It will be argued that, in fact, the biabsohitive construction may not be treated as a normal intraclausal voice alternation - rather it is the result of syntactic restructuring which splits a clause into two "layers", with one absolutive NP per each layer: the Patient absolutive NP is governed by the main verb, and the Agent absolutive NP is governed by the auxiliary. Section 2 demonstrates that the BC indeed serves for backgrounding of the Patient (except those verbal forms with which the BC has become obligatory). Finally, section 3 attempts to find out typological "preconditions" for the BC, i.e. reasons why Lak makes use of this construction in cases where other ergative languages employ antipassives.

1. Ergative and biabsolutive constructions in Lak 1.1. The ergative construction As the other languages of the Nakho-Daghestanian family, Lak employs the ergative construction as basic both for independent and subordinate sentences (for a comprehensive outline of Lak morphology and for some basics in Lak syntax see Zirkov 1955, Murkelinskij 1971; for a deeper insight into verbal morphology see Friedman 1984, and for a more detailed sketch of Lak syntax see Kibrik 1978). The verb in an EC agrees in class with the Patient (four nominal classes are distinguished): (2)

2

a.

rasul-lul qata buw-ni Rasul-ERG(lcl) house.ABS(3cl) 3cl.make:PAST-3sg 'Rasul built a house'.2

Here and below, choice of tenses in English translations is to a large extent arbitrary and does not convey certain semantic peculiarities of Lak tenses and constructions, including sometimes even those discussed in the paper.

97

On Patient Demotion in Lak

b.

rasul-M surat dur-ni Rasul-ERG(lcl) picture.ABS(4cl) 4cl.make:PAST-3sg 'Rasul made a picture.'

Apart from synthetic verbal forms, analytic verbal complexes are widely used, where a converb or participle is followed by an auxiliary. In the EC, both components of analytic forms agree in class with the Patient:

(3)

rasul-lul qata buw-nu b-u-r Rasul-ERG(lcl) house.ABS(3cl) 3cl.build-CON.PAST 3cl-AUX-3sg 'Rasul has built the house.'

Another relevant feature of Lak is that person agreement is obligatory in all, and only in, finite clauses (for detail on person agreement in Lak see Burculadze 1979, Helmbrecht 1996). The most widespread set of person agreement markers is represented in Table 1:

1 2 3

SG -ra -ra -r(i)

PL -ru -ru -r(i)

TABLE 1: PERSON AGREEMENT MARKERS

(The third person marker is -r after vowels and -ri after consonants; the consonant of agreement morpheme undergoes assimilation after /, m, n.) Prototypically, person agreement is triggered by the Patient absohitive NP in the EC3 :

(4)

3

a.

nit-il ulrc mother-ERG(2cl) child.ABS(lcl) 'Mother has caught the boy.'

b.

nit-il na mother-ERG(2cI) I(lcl) 'Mother has caught me.'

uwh-u-n-ni lcl.catch-PAST-A0R-3sg

uwh-u-n-na lcl.catch-PAST-AOR-lsg

An exception is represented by a special set of synthetic verbal forms which are used only with 1-2 person Agent - these forms, sometimes termed "categorical forms", agree in person with the Agent (see Kibrik (1979) for detail). I will not treat them in the present paper.

98

Konstantin Kazenin

In the analytic EC, person agreement, which is also triggered by the Patient, is expressed on the auxiliary :

(5)

a.

nit-il ulrc uwh-u-nu u -r mother-ERG(2cl) chMABS(lcl) lcl.catch-PAST-CON lcl.AUX-3sg "Mother caught the boy.1

b.

nit-il na mother-ERG(2cl) I(lcl) 'Mother caught me.'

uwh-u-nu lcl.catch-PAST-CON

u-ra AUX-lsg

1.2. The analytic biabsolutive construction There is another option for analytic forms: instead of putting the Agent in the ergative case, both the Agent and the Patient may be in the absohitive case. When so, the converb agrees in class with the Patient, but auxiliary agrees in class and in person with the Agent:

(6)

a.

rasul cu buwh-u-nu u -r Rasul.ABS(lcl) horse.ABS(3cl) 3cl.catch-PAST-CON lcl.AUX-3sg 'Rasul has caught the horse.'

b.

na cu buwh-u-nu 1(1 cl) horse.ABS(3cl) 3cl.catch-PAST-CON Ί have caught the horse.'

u-ra lcl.AUX-3sg

This will be called the analytic BC. This option, rather marginal in the perfective aspect (6), is very common in the durative aspect. Thus, (7b), with the analytic BC, is much more natural than (7a), where the durative analytic form is used in an EC:

(7)

a.

nit-il na uh-l-ej u-ra mother:OBL-ERG(2cl) I(lcl) lcl.catch-DUR-CON.PRES lcl.AUX-lsg 'Mother is catching me.'

99

On Patient Demotion in Lak

b.

ninu na mother.ABS(2cl) I(lcl) 'Mother is catching me.'

uh-l-ej du-r lcl.catch-DUR-CON.PRES 2cl.AUX-3sg

Ergative and Absohitive of 1-2 person pronouns coincide in one form, so that case marking does not help to distinguish between the EC and the BC when the Agent is 1 or 2 person. Agreement becomes crucial for distinguishing between the two constructions for that matter. Consider the sentences in (8), with an analytic form of the perfective aspect. Analytic forms of this aspect allow both agreement patterns with an 1-2 person Agent: either the auxiliary agrees with the Patient (8a), as it always does in the EC, or it agrees with the Agent (8b), as it always does in the BC :

(8)

a.

na qata buw-nu bu-r 1(1 cl) house.ABS(3cl) 3cl.build-C0N.PAST 3cl.AUX-3sg Ί have built the house.'

b.

na qata buw-nu u-ra 1(1 cl) house.ABS(3cl) 3cl.build-C0N.PAST lcl.AUX-lsg Ί have built the house.'

Crucial for the further discussion, with 1-2 person Agent under the durative aspect the BC becomes obligatory, so that among the two agreement patterns in (9), only the second remains grammatical when aspect switches from the perfective to the durative:

(9)

a.

na qata bull-ej u-ra 1(1 cl) house.ABS(3cl) 3cl.build:DUR-CON.PRES lcl.AUX-lsg Ί am building the house.'

b. * na qata bull-ej bu-r 1(1 cl) house.ABS(3cl) 3cl.build:DUR-CON.PRES 3cl.AUX-3sg Now let us turn to some further structural characteristics of the analytic BC. Unlike the EC, which freely allows various word order permutations, the analytic BC with Patient discontinuous from the verb is marginal:

100

Konstantin Kazenin

(10) a.

rasul qata bull-ej Rasul,AB S house,ABS 3cLbuild:DUR-CON.PRES 'Rasul is building the house.' b. ? qata rasul bullej ur

u-r lcl.AUX-3sg

The auxiliary, which may follow the Agent, hardly may intervene between the Patient and the verb: (11) a. rasul b. ? rasul

ur qata bullej qata ur bullej

These word order restrictions suggest that the Patient and the main verb (converb) of the analytic BC form a special constituent which does not include the Agent. This constituent, as well as the Agent, is governed by the auxiliary, thus yielding the following structure: (12) The biabsolutive construction S

NPpATENT

V

The EC, where word order asymmetries discovered for the BC are not attested, does not give evidence for treating the complex "Patient NP + main verb" as a separate constituent excluding the Agent NP. The free word order of the EC rather suggests that both core NPs occur at one level in its structure: (13) The ergative construction S

NPagent

ΝΡραήεντ

V

(Aux)

This solution is also supported by agreement facts. As mentioned earlier, in the analytic BC the main verb agrees in class with the Patient absolutive NP, but the auxiliary agrees in class with the Agent absolutive NP. The structure in (12) accounts for this asymmetry in a straightforward manner, if only it is assumed (with large support from other domains of Lak syntax) that class agreement is always triggered by an absolutive NP most proximate to the agreeing element, where "proximity" is understood in terms of constituency rather than in terms of linear order.

101

On Patient Demotion in Lak

So far, I have argued that the two absohitive NPs of the analytic BC belong to different levels of constituent structure. One position for absolutive case is governed by the main verb, and the other by the auxiliary. This is what allows two absolutives to coexist within one clause. 1.3. The synthetic biabsolutive construction The biabsolutive case marking may also occur with a restricted set of synthetic verbal forms. These are those durative verbal forms which may be decomposed morphologically into participle phis person agreement marker. With third person Agent such BC is possible as a marked option:

(14) a.

rasul-hil qata Rasul-ERG(lcl) house.ABS(3cl) 'Rasul is building the house.'

bullali-sa-r 3cl.build:DUR-PART-3sg

b.

rasul qata Rasul.ABS(lcl) house.ABS(3cl) 'Rasul is building the house.'

bullali-sa -r 3cl.build:DUR-PART-3sg

If the Agent is 1-2 person, then the BC with these verbal forms is obligatory. This is manifested by obligatory person agreement with the Agent:

(15) a.

na rasul uhlahi-sa-ra I Rasul.ABS lcl.catch:DUR-PART-lsg Ί am catching Rasul.'

b. * na I

rasul Rasul ABS

uhlahi-sa- r lcl.catch:DUR-PART-3sg

The alleged syntheticity of these forms is in feet only superficial. Note that person agreement markers may act as copulas in Lak nominal predicates, as in (16): (16) wa tul usu-r this I: GEN brother-3sg 'This is my brother.'

Konstantin Kazenin

102

Here the person agreement marker functions as a "light" copula, cMcised onto the nominal part of the predicate. This possibility may be extended to verbal predicates as well: in verbal forms like those of (14)-(15), participles may be treated as main verbs, and person agreement markers - as (light) auxiliaries cliticised onto them This yields the following structure for (14b):

qata

bulalisa-

r

The structure in (17) is quite parallel to the one that has been assigned to the analytic BC in (12). This parallelism is very desirable because of the observed similarities between the two BCs - common pattern of person agreement and equal obligatoriness of the two BCs in the case of 1-2 person Agent. It should be noted that, given this analysis, the label "synthetic" for this BC is not that obvious. A more appropriate name would be "the BC with the cliticised auxiliary". However, I shall continue to call this type of the BC "synthetic" dew to relative shortness of this label.

2. The BC as a functional correlate of antipassive 2.1. Patient demotion in the synthetic BC The synthetic BC is a phenomenon of special interest, because, unlike the analytic BC, it is not attested in most of the other languages of Daghestan. For Lak, it has been studied from the morphological perspective in Burculadze 1979. Some remarks on the meaning of this construction, largely based on its parallelism with the analytical BC, are found in Helmbrecht 1996. Here I would like to present some rather unstandard evidence for Patient demotion in this construction. As mentioned in 1.1, person agreement marking is obligatorily present in Lak finite clause. However, the verb is not the only possible locus for person agreement. When the verb refers to presupposition, but one of participants is focused, a person agreement marker is cliticised onto that participant; below this will be called focus construction (for detail see Xajdakov 1986; for an attempt of syntactic analysis of these constructions see Kazenin 1996b):

On Patient Demotion in Lak

(18) a.

b.

103

rasul-hil qata buw-sa-r Rasul-ERG house.ABS 3cLbuild:PAST-PART-3sg Rasili is building the house.' rasul-lul-li qata buw-sa Rasul-ERG-3sg house.ABS 3cLbuild:PAST-PART It is Rasul who is building the house.

Crucially, the trigger for person agreement is not changed when such "floating" of person agreement takes place in the EC:

(19) a.

na qata buw-sa-r I house.ABS 3cLbuild:PAST-PART-3sg Ί have built the house.' n * H

b.

na -ri/*-ra Qata buw-Sa I-3sg/Msg house. AB S 3cLbuild:PAST-PART 'It is I who built the house.' ^

c.

na qata -r/ *-ra buw-Sa I house.ABS-3sg/*-lsg 3cLbuild:PAST-PART 'It is a house that I have built.'

(20) a.

b.

ιπ ι

rasul-hxl ina uwhu-sa-ra Rasul-ERG you lcLcatch.PAST-PART-2sg "Rasul has caught you.'

rasul-hil- la/* -lì ina uwhu-§a Rasul-ERG-2sg/*-3sg you lcLcatch.PAST-PART 'It is Rasul who has caught you.'

104

Konstantin Kazenin

^Χτ c.

Π

rasul-lul ina-ra/*-ri uwhu-sa Rasul-ERG you-2sg/*-3sg lcLcatch.PAST-PART 'It is you who Rasul has caught.'

Here agreement is triggered by the Patient without focusing ((19a), (20a)), and with focusing of the Agent or the Patient. Things differ, however, in the synthetic BC. Remember that person agreement in the BC (with an agreement marker on the verb) is triggered by the Agent absolutive NP. Also the BC is obligatory with 1-2 person in the durative aspect - cf (15) repeated here: (15) na rasul uhlahi-sa-ra/*r I Rasul. AB S lcl.catch:DUR-PART-lsg/*3sg Ί am catching Rasul.' When the Agent is focused in this sentence, the trigger for person agreement, as expected, is not changed:

Π

Γ*Ί

(21) na- ra/*-r rasul uhlahi-sa I-lsg/*-3sg Rasul.ABS lcl.catch:DUR-PART 'It is I who is catching Rasul.' However, when the Patient is focused, movement of agreement marker to the Patient is accompanied by an optional change of trigger - person agreement in (22) may be triggered either by the Patient or by the Agent (moreover, for some speakers this change is obligatory, and thus (22a) unacceptable):

(22) a.

na rasul-la uhlahi-sa I Rasul. ABS-lsg 1 cL catch:DUR-PART 'It is Rasul whom I am catching.'

b.

na rasul -Ii uhlahi-sa I Rasul. AB S-3sg lcl.catch:DUR-PART 'It is Rasul whom I am catching.'

Π

This phenomenon is not restricted to 3rd person Patients - the same asymmetry is observed in the durative aspect when both the Agent and the Patient are 1 or 2 person. Thus in (23), again, person agreement is triggered by the Agent when the marker is on the verb or on the

On Patient Demotion in Lak

105

focused Agent (23a-b). But when the Patient is focused, it can control person agreement instead of the Agent (23c)4 :

(23) a.

b.

c.

zu na uhlahi-sa-ru/*-ra you I lcl.catch:DUR-PART-2pl/*-lsg *You are catching me.'

zu- ru/*-ra na uhlahi-sa you-2pl/*-lsg I lcl.catch:DUR-PART 'That is you who are catching me.' I I zu na-ra/-ru uhlahi-sa you I-lsg/-2pl lcl.catch:DUR-PART 'It is I whom you are catching.'

The freedom in choice of agreement trigger is very striking because, as we have seen in (19)(20), this freedom does not accompany focusing in the EC, where person agreement is uniformly triggered by the Patient. At first glance, the freedom of agreement could be explained in purely structural terms, based on the syntactic structure of the BC. As argued in 1.2-3, in the BC, unlike the EC, the Patient absohitive NP and the mam verb form a constituent which excludes the Agent NP. The agreement pattern could depend upon whether the agreement marker is inside or outside that constituent. In (15) it is outside, occupying the position of Aux as in the structure (17). Therefore, agreement can be triggered only by the Agent absohitive NP, which is also outside the constituent "Patient NP + main verb". In the same way, in (21) the agreement marker is outside that constituent, and agreement again is triggered by the Agent. On the contrary, in (22b) the agreement marker is attached to the Patient, that is, it is inside the constituent formed by the Patient and the main verb. In this case, the Patient absohitive NP contained within that constituent is also allowed to trigger person agreement. The general rule accounting for this could be as follows: person agreement is triggered either by the highest absohitive NP, or by the absohitive NP most proximate to the position of a person agreement marker. When a person agreement marker in a BC is on the verb, or is attached to the Agent, the two possibilities coincide in one, hence person agreement is uniformly triggered by the Agent in (15), (21), (23a-b). But when a person agreement marker is attached to the Patient, the two possibilities do not coincide in one, and the agreement can be triggered either by the Agent or by the Patient (for those speakers for whom the switch of agreement from the Agent to the Patient is

4

Again, for some speakers person agreement with the Patient is not only possible, but obligatory in (23c).

106

Konstantin Kazenin

obligatory, person agreement, under this interpretation, obligatorily would be triggered by the most proximate, but not by the highest, absohitive NP). However, this simple structural account is very problematic for one reason. As we have seen in 1.2, the analytic BC also involves a separate constituent formed by the Patient and the main verb. Given that, if a phenomenon observed in the synthetic BC is accounted for on the basis of its constituent structure, then one should expect that the same phenomenon is observed in the analytic BC as well. However, this expectation is not borne out for person agreement. As (24)-(26) show, person agreement in the analytic BC is uniformly triggered by the Agent, whether the agreement marker is on the verb (24), on the Agent (25) or on the Patient (26):

(24) a.

na qata bull-ej u-sa- ra 1 house. AB S 3cl.build:DUR-CON.PRES lcl.AUX-PART-lsg Ί am building the house.'

b. * na I

qata bull-ej house.ABS 3cLbuild:DUR-CON.PRES

u-sa- r lcl.AUX-PART-3sg

(25) na-ra/*-r qata bull-ej I-lsg/*-3sg house.ABS 3cl.build:DUR-CON.PRES 'It is I who is building the house.' (26) na qata-ra/*-r bull-ej I house. AB S-lsg/*-3sg 3cl.build:DUR-CON.PRES 'It is the house what I am building.'

u-sa lcl.AUX-PART

u-sa lcl.AUX-PART

Of course in (26), in the same way as in (22), person agreement marking is to be found within the constituent "Patient NP + main verb". However, in this case that does not lead to any variability of agreement patterns, although the structural configuration of (26) is virtually the same as that of (22). This makes invalid the above explanation of the person agreement in (22) (as well as in (23c)) by allowing person agreement with the most proximate absohitive NP, since that would require additional stipulations in order to explain why this agreement pattern is not materialised in the analytic BC. A more promising explanation of the agreement facts observed above does not treat the variability of agreement patterns observed with focusing as variability of agreement in the synthetic BC. Instead, it states that whenever person agreement is triggered by the Agent, we have an instance of the BC, but whenever person agreement is triggered by the Patient, we have an instance of the EC. This means that, in the observed examples with the durative aspect, we can have only the BC when person agreement is expressed on the verb or on the Agent. But when person agreement is expressed on the Patient, we can have either the BC or

On Patient Demotion in Lak

107

the EC (note, again, that person agreement is the only way to distinguish between the two construction here, because the ergative and absohitive cases of 1-2 pronouns coincide in one form): (27) The biabsohitive construction a. na rasul uhlahi-sa-ra I Rasul. ABS lcLcatch:DUR-PART-lsg Ί am catching RasuL 1 b. (=22a) na rasul-la uhlahi-sa I Rasul. ABS-lsg lcLcatch:DUR-PART 'It is Rasul whom I am catching.' (28) The ergative construction a. * na rasul uhlahi-sa-r I Rasul. AB S lcl.catch:DUR-PART-3sg Ί am catching Rasul. ' b(=(22b)). na rasul-li uhlahi-sa I RasuL AB S-3sg lcLcatch:DUR-PART 'It is Rasul whom I am catching.' This account does not incorrectly predict that in (26) repeated here both agreement patterns be grammatical: (26) na £jata-ra/*-r bull-ej u-sa I house.ABS-lsg/*-3sg 3cl.build:DUR-CON.PRES lcl.AUX-PART 'It is the house what I am building.' For the person agreement to be triggered by the Patient, it is necessary that the construction be ergative. However, in (26) the auxiliary agrees in class with the Agent, and this is an essential feature of the BC, but not of the EC (cf (7)). Therefore although person agreement in the ungrammatical variant of (26) is triggered by the Patient, this sentence cannot be treated as an instance of the EC. This means that it does not parallel (22b) and its ungrammatically should not constitute a problem against the background of (22b), under the proposed account. At the same time, this account makes a number of correct predictions. First, it predicts that if the Patient is focused in the synthetic BC with a third person Agent, person agreement may be triggered only by the Agent (here, unlike the above sentences, the BC can be distinguished from the EC not only by person agreement, but also by case marking of the Agent, since third person ergative and absohitive do not coincide in one form). This prediction can be checked only when the Patient is first or second person. It turns out that, in general, the synthetic BC is not perfect with this combination of persons (see below for an explanation). However, whereas (29a), where person agreement triggered by the Agent, is marginal, (29b), with person agreement triggered by the Patient, is sharply ungrammatical:

108

Konstantin Kazenin

(29) a. ? rasul na-ri uhlahì-Sa RasuLABS I-3sg lcLcatch:DUR-PART 'It is I whom Rasul is catching. '

b. * rasul ιψΧ-ja RasuLABS I -lsg

uhlahi-Sa lcLcatch:DUR-PART

Note that if (22) or (23c) were accounted for via variability of person agreement in the BC, there would not be any good reason for (29b) to be worse than (29a). Both these sentences would represent various agreement patterns in the BC, exactly like (22a) and (22b) would. On the contrary, the proposed uniform necessity of the EC for person agreement triggered by the Patient predicts ungrammatically of (29b), as well as perfect grammaticality of (29b1), where the Agent is expressed by the ergative case:

(29) b'. rasul-hil na-ra uhlahi-sa Rasul-ERG I-lsg lcLcatch:DUR-PART "Rasul is catching me.' Another correct prediction of the proposed analysis concerns class agreement in the negative auxiliary Cl=ainstrument >institution >generic human >specific human >3rd person >SAPs The first agent parameter (46a) is a familiar one, which has received focused attention in the past. The English-type, in which an agentive phrase of a personal passive is optional, is most commonly found. Siewierska (1984:35) states that an agent always occurs in passive clauses of Indonesian, Palauan, and the Dravidian language Kota. Among these, Indonesian clearly

135

Voice Parameters

permits omission of an agent in its passive clauses. The status of the Palauan passive is controversial (see De Wolf 1988). Kota shows an agreement between the instrumental nominal and the verb in what looks like a passive clause. However, other than the occurrence of the agent in the instrumental form, the clause in question shows no passive morphology in the verb or elsewhere. Impersonale limit the expression of an agent more severely than personal passives. Turkish, for example, permits an agent in personal passives (rather grudgingly) but prohibits its appearance in impersonal passives. Less recognized in the general discussion regarding the agentive phrase in passives is the second agent parameter (46b), which has a very important bearing on the diachronic development of passive constructions. There are, however, language-specific studies of passives that include relevant discussions. For example, Maldonado (1992) recognizes the difference in acceptability of Mexican Spanish reflexive passives depending on the nature of an agent (and the tense), illustrating his point with the examples below: (47) Spanish (Maldonado 1992:251, 255) a. La puenta se cerró con/por viento / 77/*por Juan. "The door was closed with/by the wind/ by John.1 b. La taza se rompió con/por la pelota / 71/*por Juan. 'The cup was broken with/by the ball/ by John.' c. Esos problemas se resuelven por autoridades competentes. 'Those problems are solved by competent authorities.' d. ??Esos problemas se resolvieron por autoridades competentes. 'Those problems were solved by competent authorities.' e.*Esos problemas se resovieron por Juan. "Those problems were solved by John.' The Arizona Tewa suffixal passive, which is said to prohibit an overt expression of an agent, permits an instrumental or non-human "agent," as seen below: (48) Arizona Tewa suffixal passive (Kroskrity 1985: 310, 317) a.*hëi sen-di hêi tú na-c'á:la-t:í. that

man-OBL

that

meat 3SG.STA-cut-PASS

"The meat was cut by that man.' b. na:bí ciyó-d hêi tú na-c'á:la-tí. I-gen knife-OBL that

meat 3SG.STA-cut-PASS

'The meat was/has been cut with my knife.' c. nan-di phé-mele na-khá:be-n. sand-OBL

stick-vessel

3SG.ST-break.PASS

'The crate was crushed by sand.'

Masayoshi Shibatani

136

A similar observation can be made with the reflexive-passives in Modern Greek (Warburton 1970: 70), Russian, and Italian, in which the agent is easier to express overtly in the order of the reverse empathy hierarchy given in (46b). Southern Tiwa, as described by Allen and Franz (1983), allows in its passive clauses only a third person agent to the exclusion of the speech act participants, as is the case with certain passive forms in Mayan languages (see Dayley 1983 and England 1988). For example, whereas (49a) is possible, (49b,c) are ungrammatical. (49) Southern Tiwa (Allen and Franz 1983:305) a. seuanide-ba te-mu-che-ban. man-iNST

lSG-see-PASS-PAST

Ί was seen by the man.' b.*te-muche-ban 'i-ba. lSG-see-PASS-PAST 2-INSTR

Ί was seen by you.' c.*a-mu-che-ban na-ba. 2SG-see-PASS-PAST 1-INSTR

"You were seen by me.' The agent parameter in (46b), in conjunction with the general development pattern of passives, makes an interesting prediction concerning the maturity of passive constructions. But this topic requires a brief discussion of the development of passives. One major source of passives is the middle voice forms, which in turn may arise from reflexive expressions as in Romance and Slavic languages (see Geniusienè 1987). It appears that the middle voice category is instable with a tendency to develop (further) into the passive category, as in a number of Mayan languages, where the passive morphology is related to a middle formant that is no longer productive (see Dayley 1983). This diachronic tendency is again an expression of the Principle of Maximization of Contrast. The contrast between active and middle is not as maximal as that between active and passive, as discussed in section 2, and as the term "middle" suggests itself. The development of a passive from a middle form, as in many other historical changes, appears to be gradual not only in the replacement of the function, but also in the development of the passive meaning. The particular middle form that gives rise to a passive form expresses a spontaneous event (as in Spanish La taza se rompió 'The cup broke'), and thus the first passive meaning that emerges is the one more compatible with this kind of spontaneous meaning, i.e. where the external force is construable as something that naturally brings about a change; hence the preference for natural forces, general or collective agency, and for routinized activities in the generic tense. That the passive form from other sources may also follow the predicted pattern is shown by the English gei-passive, which does not easily combine with a specific human agent. These considerations make the following prediction. If a language contains two or more passive constructions, the one allowing an overt agent high in the Empathy Hierarchy (or low in the Reverse Empathy Hierarchy in (46b)) is a more mature or older construction. This

Voice Parameters

137

prediction is borne out in Romance and Slavic languages, where the ¿^-passives, which permit all kinds of agent, are older than the reflexive passive, which permit only a restricted range of agent, as seen in (47). Between the English ôe-passive and the gei-passive, the former is the older form and permits an expression of an agent more liberally than the latter. In addition to the diachronic reason for the late development of the passive agents low in the Reverse Empathy Hierarchy, there is another force at work that discourages these to occur in the peripheral adjunct position. That is, the speech act participants (SAFs) and those high on the Empathy Hierarchy are most likely to be placed in the syntactically most prominent position, namely subject position, because the speaker tends to empathize with them and to treat them as central entities in the described event. This centrality principle, together with the diachronic pressure for guaranteeing a gradual change, has a blocking effect on the Principle of Maximization of Contrast in such a way that, while the latter pressures the language to include an explicitly stated agent high on the Empathy Hierarchy, the former leads to a tendency to avoid such a structure. The universal tendency for a passive structure not to have an overt agent seems to indicate that such a structure is the best compromise meeting these two opposing forces.

References Âfarli, T. (1992) The Syntax of Norwegian Passive Constructions. Amsterdam: John Benjamins. Allen, B. & Franz, D. (1983) Advancements and verb agreement in Southern Tiwa. In Perlmutter, D. (ed.) Studies in Relational Grammar 1. Chicago: University of Chicago Press, 303-314. Aze, F.R. (1973) Clause patterns in Parengi-Gorum. In Trail, R.L. (ed.) Patterns in Clause, Sentence, and Discourse in Selected Languages of India and Nepal. Norman: Summer Institute of Linguistics, 235-312. Bandhu, C. (1973) Clause patterns in Nepali. In Hale, A. & Watters, D. (eds.) Clause, Sentence, and Discourse Patterns in Selected Languages of Nepal. Norman: Summer Institute of Linguistics, 1-80. Barber, E. (1975) Voice beyond the passive, Proceedings of the First Annual Meeting of the Berkeley Linguistics Society, 16-24. Biktimir, T. (1986) Impersonal passives and -ArAk construction in Turkish. In Slobin, D.I. & Zimmer, Κ. (eds.) Studies in Turkish Linguistics. Amsterdam: John Benjamins, 53-75. Chomsky, N. (1981) Lectures on Government and Binding. Dordrecht: Foris. Dayley, J. (1983) Voice and ergativity in Mayan languages. In Schlicter, Α., Chafe, W., & Hinton, L. (eds.) Studies in Mesoamerican Linguistics, Survey of California and Other Indian Languages Report No. 4. Berkeley: Survey of California and Other Indian Languages, 5-119. De Wolf, C. (1988) Voice in Austronesian languages of Philippine type: passive, ergative, or neither?. In Shibatani, M. (ed.) Passive and Voice. Amsterdam: John Benjamins, 143-193. Dixon, R. (1979) Ergativity, Language, 55, 59-138. England, N. (1988) Mam voice. In Shibatani, M. (ed.) Passive and Voice. Amsterdam: John Benjamins, 525546. Frajzyngier, Z. (1982) Indefinite agent, passive and impersonal passive: A functional study. Lingua 58, 267290. Gabelentz, H. (1861) Über das Passivum. Abhandlungen der philologisch-historischen Klasse der königlich sächsischen Gesellschaft der Wissenschaften 3. Leipzig, 449-546. GeniuSiené, E. (1987) The Typology of Reflexives. Berlin: Mouton de Gruyter. Givón, T. (1988) Tale of two passives in Ute. In Shibatani, M. (ed.) Passive and Voice. Amsterdam: John Benjamins, 417-440.

138

Masayoshi Shibatani

Keenan, E. (1975) Some universels of passive in relational grammar, Papers from the Eleventh Regional Meeting of the Chicago Linguistic Society, 340-352. Kemmer, S. (1993) The Middle Voice. Amsterdam: John Benjamins. Kroskrity, P. (1985) A holistic understanding of Arizona Tewa passives, Language 61, 306-328. Kruisinga, E. (1925) A Handbook of Present-day English. Part 3. Utrecht: Kemink. Langacker, R. W. (1976) Non-distinct Arguments in Uto-Aztecan, Vol. 82. Berkeley: University of California Press. Maldonado, R. (1992) Middle Voice: The Case of Spanish se. Unpublished Ph.D. Thesis. University of California, San Diego. Marlett, S. (1984) Personal and impersonal passives in Seri. In Perlmutter, D. & Rosen, C. (eds.) Studies in Relational Grammar 2. Chicago: University of Chicago Press, 217-239. Mel'Cuk, I. (1993) Inflectional category of voice: Towards a more rigorous definition. In Comrie, B. & Polinsky, M. (eds.) Causatives and Transitivity. Amsterdam: John Benjamins, 1-46. Mithun, M. (1988) The 'passive' in an active language. In Redden, J. (ed.) Papers from the 1987 Hokan-Penutian Languages Workshop and Friends of Uto-Aztecan Workshop. Carbondale: Department of Linguistics, Southern Illinois University, 9-45. Moorcroñ, R. (1985) The role of semantic restrictions in German passive formation. In Faarlund, J. (ed.) Germanic Linguistics: Papers from a Symposium at the University of Chicago April 24, 1985, Bloomington: Indiana University Linguistics Club, 157-170. Noonan, M. (1994) A tale of two passives in Irish. In Fox, B. & Hopper, P. (eds.) Voice: Form and Function. Amsterdam: John Benjamins, 279-312. Pandharipande, R. (1981) Syntax and Semantics of the Passive Construction in Selected South Asian Languages. Unpublished Ph.D. Thesis. University of Illinois, Urbana-Champaign. Perlmutter, D. (1978) Impersonal passives and the unaccusative hypothesis, Proceedings of the Fourth Annual Meeting of the Berkeley Linguistics Society, 157-189. Perlmutter, D. (ed.) (1983) Studies in Relational Grammar 1. Chicago: University of Chicago Press. Perlmutter, D. & Postal, P. (1983) Toward a universal characterization of passivization. In Perlmutter, D. (ed.) Studies in Relational Grammar 1. Chicago: University of Chicago Press, 3-29. Perlmutter, D. & Rosen, C. G. (eds.) (1984) Studies in Relational Grammar 2. Chicago: University of Chicago Press. Pokharel, M. (1996) Passivization in Nepali. In Contributions to Nepalese Studies. Center for Nepal and Asian Studies, Kirtipur: Tribhuvan University, 451-472. Shibatani, M. (ed.) (1988) Passive and Voice. Amsterdam: John Benjamins. Siewierska, A. (1984) The Passive: A Comparative Linguistic Analysis. London: Croom Helm. Sohn, H-m. (1975) Woleaian Reference Grammar. Honolulu: The University of Hawaii Press. Stein, G. (1979) Studies in the Function of the Passive. Tübingen: Narr. Sullivan, T. (1988) Compendium of Náhuatl Grammar. Salt Lake City: University of Utah Press. Warburton, I. (1970) On the Verb in Modern Greek. Bloomington: Indiana University.

Passive, Anticausative and Classification of Verbs: The Case of Vedic* Leonid Kulikov

1. Passive and anticausative: introductory remarks This paper deals with some problems of distinguishing between passive and anticausative, two closely related verbal categories. By anticausative I mean the intransitive (non-causative) counterpart ( lb) in verbal pairs like ( 1 ): (1)

a. John opens the door b. The door opens

Although passive and anticausative share a number of features (both being, above all, valency decreasing categories) and use the same or similar morphological marking in many languages, their position within the system of verbal categories and their status in a linguistic description differ crucially. Passive is a well-established and thoroughly investigated verbal category, which has been widely discussed in both typological literature and grammatical descriptions of individual languages; the relevant bibliography is vast, if not to say unsurveyable, and includes works written in the last century or even earlier, before the beginning of systematic typological investigations. In contrast, the term 'anticausative' is less than 30 years old. It has been introduced by Nedjalkov and Sil'nickij (1969) in the collective monograph on causatives which belongs now to the classical specimens of the typological literature. Since the 70ies, this term has appeared in both descriptive studies (cf. Masica 1976, Babby 1983, Winford 1988) and typological works dealing with verbal categories, cf. esp. Comrie 1981: 161; Comrie 1985: 325-328; Siewierska 1984: 77-79; Miller 1993: 179-183. There are a few typological studies focusing on anticausatives, Moreno 1984, Moreno 1985 and Haspelmath 19871, the latter being, to my knowledge, the only monographic treatment of the problem and the most comprehensive bibliographical source of the relevant literature until the mid-80ies. Thus far, however, studies on this category are still few in number. This position of an orphan among other verbal categories (such as passive, causative, reflexive, etc.) can be illustrated by the very fact that the term 'anticausative' is not widely accepted yet2 and even using * I would like to thank H. Vater, A. Lubotsky, V.A. Plungian, E.V. Rakhilina and A.Y. Aikhenvald for valuable comments on earlier drafts of this paper. 1

Cf. also Haspelmath 1993.

2

For other possible terms, like 'decausative1, 'inchoative', 'middle', see Haspelmath 1987: 9-10; in the IndoEuropean scholarship we also find terms 'eventive' (cf. Gonda 1951) and, more recently, 'fientive'. It is worth mentioning that an important contribution to the description of non-passive intransitive verbs is made by

140

Leonid Kulikov

this term in a descriptive work generally requires comments and some kind of apologies. Symptomatically, the list of scholars who widely use this term, consists, almost exhisively, of those who can read Russian, so that this label remains till now, in a sense, "East-oriented" and usually means a good knowledge of the works of the Leningrad/St.Petersburg typological group, above all of most comprehensive typological studies on causatives by V.P. Nedjalkov and his colleagues.

2. Passive vs. anticausative: problems of distingushing The unprivileged position of anticausative can be easily explained by some of its features. Here some properties of both categories under discussion should be recalled. A canonical passivization entails promotion of the underlying direct object (which is usually a patient, or an undergoer, in semantic terms) to the subject position, whereas the underlying subject moves down the hierarchy of grammatical relations and either becomes an oblique or remains unexpressed in the passive clause: (1)

a. John opens the door c. The door is opened [by John]

An anticausative derivation also entails advancement of the direct object, and this is the most important feature shared by passive and anticausative, which accounts for their similar morphological marking in many languages, such as Russian, Spanish, Swahili, etc. (cf. Comrie 1985: 328-330; Haspelmath 1987: 29-31). There is, however, a crucial semantic difference between these categories. Let me quote Comrie's (1985: 326) definition here: "Passive and anticausative differ in that, even where the former has no agentive phrase, the existence of some person or thing bringing about the situation is implied, whereas the anticausative is consistent with the situation coming about spontaneously."

Cf also Haspelmath 1987: 6-7, Siewierska 1984: 78. This semantic distinction implies the following syntactic feature of passive: constructing with an agentive nominal, usually in the instrumental (in languages with case-marking), is unambiguous evidence for the passive interpretation. Hie opposite is not always true, however, so that this feature is, in a sense, a "one-way" criterion: the lack of an overtly expressed agentive nominal cannot serve as sufficient evidence for an anticausative interpretation (cf Siewierska 1984: 78). Moreover, it seems that in many cases the question whether the agentless intransitive should be treated as anticausative or passive is simply irrelevant for the speaker. studies on unaccusatives (other terms are: 'ergative', 'inactive intransitive'); for the most exhaustive treatment of the problem, cf. Levin and Rapoport 1995. The notions 'anticausative' and 'unaccusative' overlap in many important ways but are not identical and, most importantly, belong to different frameworks.

Passives, Anticausatives and Classification of Verbs: the Case ofVedic

141

Furthermore, this is not the only difficulty concerned with the "Instrumental criterion". In some languages, the instrumental case can cover non-agentive usages, being also employed for marking such semantic roles as Tool or Means, so that it is often no easy task to identify whether we have to do with an agentive instrumental or not, c£: (2)

Thousands of people were killed by this earthquake

On the other hand, Agent can be expressed otherwise than in the instrumental, cf. in German: (3)

Verletzte wurde zwischen zwei Sanitätern zum Krankenwagen geleitet.3

The morphological criteria often do not help much either. As already mentioned, some languages use the same morphological marking for both passives and anticausatives. Others exhibit different markers for these two categories, without drawing a clear-cut distinction between passive and anticausative usages, however, so that we can only note a more or less strict tendency to use, for instance, marker A for anticausatives and marker Ρ for passives, whereby "A-passives" and "P-anticausatives" are possible, too.4 This is probably the case with Russian verbs in -sja generally employed as anticausatives and opposed to constructions with passive participles in -n, -t. As a rule, -s/a-verbs cannot be constructed with an agentive instrumental, cf.: (4)

a. Palka byla slomana (Petrom) stick:NOM be:PAST break:PASS.PRT (PeterlNS) 'The stick was broken (by Peter)' b. Palka slomala-s' stick:NOM break.PFV:PAST-ANTICAUS/PASS 'The stick broke.' c. 'Palka slomalas' Petrom

Nevertheless, -ΐ/α-passives (which usually look more artificial though) can also be formed, if the verb is put in the imperfective aspect,5 cf.: (5)

Dom stroil-sja zakljucennymi house:NOM build:PAST-ANTICAUS/PASS prisoners:INS 'The house was built by the prisoners'

3

For a discussion and possible interpretations of such cases, see Höhle 1978: 147-161.

4

Cf. for instance, the statement made by the author of one of the most insightful and comprehensive typological studies on Asian languages: "Anticausative" is not the same as "passive", to be sure, being more akin to "middle", but certain forms in some languages partake of the nature of both, and there is an area of overlap and uncertainty [bold is mine - LK.]" (Masica 1976: 60).

5

As shown by Gerritsen (1988: 132-136, 163-168), this aspectual constraint is due to that only 'non-actual' readings are possible with -.^'α-passives.

Leonid Kulikov

142

Even in languages with a clear-cut distinction between passives and non-passive intransitives we come across some instances of an "unexpected" morphological marking, as in the following English example with the non-agentive instrumental nominal: (6)

John was killed by the Californian earthquake6

Interestingly, the literal Russian translation is almost unacceptable: (7)

a. "John

byl be:PAST

ubit Kalifomijskim kffl:PASS.PRT Californian: INS

zemletijaseniem earthquake:INS

In order to render (6) in Russian, we have to replace the instrumental with an adverbial phrase 'during the Californian earthquake' and to use another (non-passive) verb: (7)

b. John pogib pri Kalifomijskom zemletijasenii

Note that the same sentence with the passive verbal form byl ubit instead of the non-derived intransitive pogib can only mean: 'John was killed by somebody (a criminal or whoever) during the earthquake*. Of course I will make no attempt to elaborate a universal criterion for dealing with the passive/anticausative distinction. The aim of this paper is more limited: to draw attention to an important aspect of the problem, viz. to the interrelation between classification of verbs (predicates) and constraints on the passive/anticausative derivation. There is no need to argue that some transitive verbs can be both passivized and anticausativized (decausativized) (cf. (la-c)), while some other verbs (which can be labeled inherently transitive) only allow for passivization, cf.: (8)

The table is made by John

We hardly can figure out a situation when a thing (human artifact) is made by itself without any external agent; correspondingly, this sentence has no anticausative or reflexive counterpart. This means that the anticausative derivation has a more limited scope of applicability than the passivization. Incidentally, this may also be one of the reasons why anticausatives usually occupy a less privileged position within the system of verbal categories. Note that many languages use the same morphological marking for several intransitive derivations (anticausative, reflexive, potential passive, etc.), all of which are different from the canonical passive, so that in case when an anticausative interpretation is impossible for semantic reasons, the same form can be usurped by another category, such as potential passive ("The table is easy to make'), reflexive or something else. 6

Passive marking in the non-passive usages, as exemplified by (6), has been noticed for a number of languages, e.g. for German by Höhle (1978: 139, fil. 1), who treats such cases as idiomatic.

Passives, Anticausatives and Classification of Verbs: the Case ofVedic

143

The question arises how to define the class of transitive verbs which can undergo anticausativization. An effective criterion for classification of verbs in accordance with whether or not they can be anticausativized would be a helpful tool for disambiguating some of the usages without the instrumental agent. At first sight, the solution of the problem is simple and, as a matter of fact, can be deduced immediately from the definition of anticausatives (as opposed to passives) as verbs denoting spontaneous processes (c£ Comrie, op.cit.; Haspelmath 1987: IS). Thus, if an action is conceivable as coming about spontaneously, the corresponding intransitive verb can be treated as anticausative; otherwise only a passive interpretation is possible. Unfortunately, this criterion is rather vague and abstract, so that the limits of the class of verbs which can be anticausativized (hereafter, AC-class) remain unclear. Some authors suggest extra features for defining the AC-class. Haspelmath (1993: 92fF.) formulates the most distinctive feature as "the absence of agent-oriented meaning components".7 Haspelmath adduces a few minimal pairs like tear ~ cut: the former verb has an anticausative (inchoative in Haspelmath's terms) counterpart due to the absence of the agent-oriented meaning components, whereas the latter has the semantic component "by means of a sharp instrument' and, in virtue of that, can only be passivized but not anticausativized. Likewise, execute is said to have agent-oriented semantic components, while kill does not. First of all, I am afraid, the notion of agent-oriented semantic components is rather difficult to be accurately defined. As Song (1995: 213) in his review rightly points out, many of Haspelmath's examples of verbs without agent-oriented meaning elements are dubious: killing can be performed by means of some instrument, e.g. a shotgun, etc. Furthermore, in my opinion, the explanatory force of this criterion is illusory, since this is but a reformulation of the feature of spontaneity. In fact, an action can be conceived of as spontaneous if and only if an agentive external force is unnecessary, which (almost) exactly corresponds to the absence of the agent-oriented meaning components. A similar approach to distinguishing verbs which can be anticausativized from those which cannot has been developed by Levin and Rapoport (1994: 61-66; 1995: 102-106). The authors state that "the transitive causative verbs that detransitivize [ = anticausativize - LK] are those in which the eventuality can come about spontaneously without the volitional intervention of an agent" (Levin and Rapoport 1994: 61 = 1995: 102). They also notice that the classification of verbs according to whether or not they describe an eventuality that can occur spontaneously overlaps with the classification according to whether or not they describe an externally caused eventuality (Levin and Rapoport 1995: 105). Again, this is one more reformulation of the feature of spontaneity. I am less optimistic about formulating a universal non-trivial criterion (i.e. a criterion which is not just one of the facets of spontaneity) for identifying the AC-class. Moreover, this class of verbs seems to be language-specific, so that the same transitive verb can be only passivized in 7

This seems to be a generalization of the earlier formulation, as suggested by Haspelmath (1987: 15): "For a change in the undergoer to come about spontaneously, the change may not be effected with too specific means".

144

Leonid Kulikov

some language whereas the corresponding verb with (nearly) the same meaning in another language may have both passive and anticausative counterparts (often merged within the same form; cf. Russian byl ubit as opposed to English was Mied, as exemplified by (6)); for a discussion on some subtleties of such kind, cf. Levin and Rapoport 199S: 98-101. There is an approach, however, which may help to make a step further and to contribute to the solution of the problem. It might be advisable to locate the most important subclasses of the AC-class which belong to its core and have a less abstract semantic definition.

3. The case of Vedic In what follows I will discuss a specific group of verbs in Vedic Sanskrit. It will be argued that this group can be interpreted in terms of a non-trivial semantic feature, which seems to me highly relevant for dealing with the AC-class. Most interesting is that this group can also be defined independently on formal grounds. 3.1. Passive and anticausative in Vedic The Vedic verbal system displays a morphologically marked opposition between passives and non-passive intransitives. The system of present, which I am concerned with, is the most elaborated among other tense systems, as to the opposition of valency-changing categories. Several transitive verbs exhibit a tripartite opposition: transitive forms (with active endings 1SG -mi, 2SG -si, 3SG -ti, etc.) are opposed to both passives (built with the accented suffix -yá- and middle endings: 1SG -e, 2SG -se, 3SG -te, etc.) and middle forms which can be derived either from the same stem as active forms or from some other stem Middle forms often have more than one meaning. These usually include anticausative, reflexive and "transitive-affective" ('do smth. for oneself). Thus, although anticausatives are not always consistently distinguished from other non-passive intransitive categories, they are, at any rate, more or less consistently opposed to passives. A good illustration of such tripartite opposition is the verb pü : punati 'purifies' (transitive) ~ pass, puyóte 'is purified [by smb.]' ~ pcrvate "becomes clean, purifies (oneself)'. 3.2. Vedic ->>a-formations 3.2.1. Present formations with the suffix -ya- and fluctuating accentuation Vedic present formations built with the suffix -ya- (hereafter, -^α-presents) can be subdivided into two main classes, depending on the accent placement. As mentioned earlier, -^α-presents with suffix accentuation are passives; they are opposed to -^α-presents with root accentuation (the so-called class IV within the Old Indian tradition), most of which are non-passive intransitives: pádyate 'fells', ríyate 'flows', etc. Any verbal root for which -^α-presents are attested has either only suffix-accented -_ya-passives or only root-accented non-passive -^α-presents. This rule perfectly holds in Vedic

Passives, Anticausatives and Classification of Verbs: the Case ofVedic

145

and is only rarely violated in late Vedic texts, where a few verbal roots can form both -^α-passives and -^α-presents with a non-passive meaning (asyati "throws' ~ asyate 'is thrown', etc.).

There is, however, a small group of less than 20 verbs which is, at first áght, a counter-example to the rule above, viz. verbs for which both suffix-accented and root-accented forms occur: mucyáte/múcyate 'is released; becomesfree';ksyáte/ksíyate 'perishes; is destroyed', etc. (hereafter labeled -^α-presents with fluctuating accentuation). No regularity has been found in this accent fluctuation, and the only attempt to relate it to the passive/non-passive distinction made by Gonda (1951: 98-99) proved unsuccessful. In the Vedic scholarship such verbs are usually mentioned as counter-evidence against the aboveformulated correlation between the place of stress and the passive/non-passive distinction. Below I will argue that the accentual patterning of these verbs does not actually violate the rule. On closer examination, it turns out that there is no semantic feature correlating with accent fluctuation in -^α-presents.8 Forms differing in accentuation can be found in very similar contexts; let me give a few examples. The following two occurrences in the TaittirTya-Samhitä and áatapatha-Brühmana describe the waning of the moon: (9)

ami'im apaksiyamänam ánv ápa ksïyeta (TS 3.5.1.3-4) 'He would waste away as this waning [moon]'

(10) yá evapüryáte 'rdhamásáh sádeva, yò 'pakslyáte sápitaráh (SB 2.1.3.1) 'That half-moon which increases [represents] the gods, that which decreases [represents] the fathers' The two passages below are even more similar: (11) yáda hi pramíyaté 'themam upavasyáti (TS 2.6.9.6) '... because when [the sacrificer] dies he has recourse to this [earth]' (12) yó vai pramlyáte 'gnim tásya sárTram gácchati sómam rásah (MS 2.3.5:32.7-8) 'Verily, who dies, his body goes to Agni (= fire), [his] essence to Soma' Cf. also the following two passagesfromritual texts: (13) yádajyam ucchísyate téna samídho *bhyájyadadháti (TB 1.1.9.3) 'Whatever butter is left, [the priest] puts fuel, having anointed [it] with this [butter]'

8

The only exception, mentioned by many grammars, is the verb pac: pacyáte is passive ('is cooked'), while pácyate is clear instance of an anticausative, 'ripens'.

146

Leonid Kulikov

(14) yád ajyam ucchisyáte téna rasanam abhyájyadatte (SB 13.1.1.1) 'Having greased a rope with the butter which is left [after greasing dishes of cooked rice], [the priest] takes it' Moreover, we even find parallel passages which differ only in the place of stress: (15) ná yásya hanyáte sákha nájíyate káda caná (RV 10.152.1) = ná yásya hanyáte sákha ná jíyáte káda caná (AV 1.20.4) '... whose friend is not killed, nor ever suffers loss' In this situation we have to take a closer look at accented verbal forms in all accentuated Vedic texts and to compare the number of accented occurrences. Once we look at the account (for details, c£ Kulikov (forthcoming)), we immediately observe that forms with accent on the root as opposed to forms with accent on the suffix are distributed in the texts as follows: In the Rgveda and TaittirTya-Samhitä, -ya-presents most often are root-accented, while in the Atharvaveda, Maitrâyanï Samhitä and áatapatha-Brahmana, -^α-presents are mostly suffix-accented. This rule is not valid for two cases only: namely, for pac 'cook; ripen' and tap "heat, suffer1, and holds for the other fifteen -^α-presents with fluctuating accentuation (for a complete list, see below). Exceptions are very few (no more than 10-12% of the total number of accented occurrences). This means that for the -^α-presents with fluctuating accentuation (except pac and, perhaps, tap) it is useless to search for any semantic (functional) correlation with the place of stress. 3.2.2. A tentative semantic interpretation Most surprising is that a great deal of these -^α-presents, defined in purely formal terms (fluctuating of the accent) exhibit a striking semantic affinity and share a number of independent features. From the semantic point of view, -jw-presents can be subdivided into two major sub-classes. The first sub-class comprises verbs of destruction referring to breaking (especially sacrificial vessels), splitting, as well as to destruction in general: (i)

chid

dr bhid sr ksi ml

lup

chidyate 'is cut off) breaks', dayate 'splits, breaks', bhidyate 'splits, breaks', soyate 'crushes, breaks, collapses' ksyate 'wastes, disappears', myate 'is damaged, perishes', lupyate 'tears, is omitted'

Intuitively, -^ά-presents of another sub-class, such as mucyate "becomes free', sfyate 'falls', 'remains', can be said to belong to nearly the same semantic area, although their com-

sisyate

Passives, Anticausatives and Classification of Verbs: the Case ofVedic

147

mon semantic denominator is rather difficult to describe. One may argue that they all denote a process when an element ceases to be included in any system or structure (for instance, is lost from the system), i.e. a process which results in the destruction of a system Thus, again, like in group (i), we have to do with some kind of destruction. For this group I propose a tentative label "verbs of destructuring". The following -ya-presents belong here: (ii) jiyate 'suffers loss'; mucyate 'becomes free, is released' (i.e. 'ceases to be bound, included into a bound system'); ssyate 'is left over* (i.e. "becomes remainder of a structure'); h/yate 'id.'; sfyate 'falls (out)' (i.e. 'ceases to be included into a structure through falling out of it'); Two more verbs require more detailed comments. ricyate, when used without preverbs, has the meaning 'is emptied', thus denoting destruction of a structure consisting of a container and its content. This verb, however, occurs mostly with the preverbs áti and prá and then acquires the meaning 'surpasses, is redundant', that is 'goes beyond the scope of some structure', which suits the notion of destructuring still better. The meaning of rdhyate, 'thrives, is successful' does not belong to the domain of destructuring. However, when used with the preverb vi, it acquires the meaning 'is deprived of sth., loses sth.', i.e. 'ceases to be connected with some (structural) elements', which also suggests destroying a structure. In my opinion, it is possible to define a still more abstract semantic feature shared by both subgroups (i) (destruction) and (ii) (destructuring). Both destruction and losing from a structure (destructuring) are spontaneous processes which result in destroying some natural or artificial system or organism and, to say it in more general and abstract terms, in the increase of entropy. The concept of entropy was proposed in 1850 by R. Clausius, a German physicist, and is sometimes presented as the second law of thermodynamics. According to this law, entropy increases during irreversible processes such as the spontaneous mixing of hot and cold gases, the uncontrolled expansion of a gas into a vacuum, and the combustion of a fuel. Very soon the notion of entropy has been extrapolated to many other domains, in particular, to a number of mechanical phenomena, so that many spontaneous processes, such as destruction or simplification of a system, death of a living being, etc., are often referred to as entropy increasing processes.9 Note that any event concerned with (spontaneous) falling down of an object leads to decrease of energy of this object and, hence, to decrease of total energy of the system consisting of this object and the source of gravitation (normally, the earth), which implies increase of entropy. Furthermore, the falling of an element out of a system simplifies this system, so that its total energy decreases while entropy increases, again. Finally, here belong a number of phenomena concerned with some natural chemical processes, such as burning of fuel, souring of milk, etc. 9

For the notion of entropy and its developments, cf. e.g. Chambadal 1963.

148

Leonid Kulikov All spontaneous processes of such kind are irreversible; hence, it has been said that the entropy of the universe is increasing. That is, more and more energy becomes unavailable for conversion into mechanical work, and because of this the universe is said to be "running down". According to the same law, everything in the universe is irrevocably moving in the direction of random chaos and waste. This notion has not only a physical significance but has also been largely conceptualized within human culture and world view (cf. Rifkin 1989), which, eventually, means that the concept of entropy and entropy increase is relevant within a system of language meanings. Of course, it would be inadequate to look for an exact correspondance to the strict physical notion of entropy (as well as many other concepts of modern physics, biology and other natural sciences) in a natural language. Nevertheless, entropy, as defined by physicists, seems to be a good generalization and formalisation of some concepts, such as destruction, decay, destructuring, which are relevant for natural languages and, thus, may be relevant for a linguistic description.

One more verb can be appended to the above-described group, pr 'fill1. Although it does not refer to any 'entropy increase' process, in compounds with the preverb a it becomes antonymous (or, to be more exact, reversive) to one of the 'entropy increase' verbs: a-pr 'wax' and άρα-ksi 'wane' refer to opposite changes of the half-moon. Cf. examples (10) and (16), where both verbs co-occur in the same passage: (16) yá evapüryáte tám deva upayan, yò 'pakslyáte tám ásuráh (áB 1.7.2.22) 'The gods entered upon the [half-moon] which waxes, the Asuras [upon] the one which wanes' The co-occurrence in texts and semantic parallelism may be a sufficient reason for assimilating some formal properties of pr (in particular, its accentual behaviour) to those of ksi and, eventually, for putting it into the 'entropy increase' class. Two -ja-presents whose accentual behaviour clearly differs from that of the 'entropy increase' verbs, pacyate and tapyate, do not belong here semantically either, which, no doubt, is not accidental. 3.2.3. Paradigmatic similarity There is one more remarkable feature shared by a great deal of the -^α-presents with fluctuating accentuation (except for pacyate and tapyate): they are opposed to transitive-causative presents with nasal affixes, viz. with the suffixes -ria-/-ni-, -nó-/-nu- and with the infix -ná-/-n-: rdh

rnáddhi (RV+), rdhnáti (RV1 +) 'makes thrive';

ksi

ksinati ( R V +), ksinái

chid

chinòtti (RV +) 'splits';

JÎ

jinati (RV +) 'deprives of property1;

df

-drnati ( A V P -drna, SB -drn'tyai)

pr bhid

prnati

mi

minati (RV +) 'diminishes; violates';

( R V +), prnati

( A V + ) 'destroys';

'splits';

( R V + ) 'fills';

bhinátti (RV +) 'splits';

Passives, Anticausatives and Classification of Verbs: the Case ofVedic

lup

rinákti (RV +) 'leaves'; lumpáti (AV +) freaks, tears, omits';

ss

ánásti (Br. +), ämsati (Br. +) 'leaves';

sr

srnati (RV +) "breaks'.

rte

149

The above-listed common features prove close affinity of the -^α-presents with fluctuating accentuation. I hypothesize that their similarity in accentual patterning is not accidental but results from their semantic affinity. Very often, the question whether death, destruction or ruin come about spontaneously or due to any external force (agent) cannot be answered, being merely irrelevant for the speaker / the author of the text. For instance, it makes probably no sense to try to identify usages exemplified by (15) as passive or anticausative. Most likely, it is just irrelevant for the author of the passage in question whether the main participant of the situation described in (15) suffers loss spontaneously (which would entail an anticausative interpretation) or is deprived of property due to the intervention of an agent, which would imply a passive reading. Assuming that the Rg-Veda, as the most ancient Vedic text, has the older placement of the accent, one may suppose that originally the -^ά-presents were nonpassive (anticausative) class IV '-^α-presents and had root accentuation. Since these verbs can be considered in some contexts either as passives or as non-passive intransitives (anticausatives), and in virtue of the increasing productivity of the suffix-accented -jw-passives, verbs of the type chídyate, ksfyate, múcyate, etc. may have been reinterpreted in some postRgvedic dialects as passives, due to the following semantic developments: "breaks' -> 'is broken [by accident]' —> 'is broken [by smb.]', "becomes free' -> 'is released', etc. These developments have led, eventually, to the accent replacement in certain Vedic texts. Thus, suffix accentuation of '-yá-presents in the Atharva-Veda, MaitrâyanT Samhitâ and áatapatha-Brahmana may represent an interesting isogloss shared by the corresponding Vedic dialects. 3.2.4. Some erroneous accentuations and their tentative explanation Now, when the class of -^α-presents with fluctuating accentuation is well-defined, semantically, syntactically, paradigmatically and morphologically, and its linguistic reality is beyond any doubt, it would be interesting to take a look at some other -^α-presents under this new view-point. Thus far I have been concerned only with the -^α-presents with middle inflexion, which are attested with both accentuations. The active -^α-presents can only be root-accented. Exceptions are very few in number and are generally treated as mere manuscript errors. Not counting some erroneous forms in late and corrupt texts, such as Taittirïya-Âranyaka, we find only two clear exceptions of such kind. )f 'waste away, grow old, decay1 This verb builds root-accented active -^α-presents with non-passive meaning, cf.:

150

Leonid Kulikov

(17) jíryanti ha vai júhvato ydjamänasyägnäyah (SB 11.7.1.1) 'For, while he is offering, the Sacrificeos fires become worn out' The only suffix-accented form occurs in the MaiträyanT Samhitä: (18) sám hi jíryátah (MS 3.7.3:78.6) '... because they both grow old' The accent placement is exceptional, so that Mittwede (1986: 126) has conjectured

*jfryatah.

mrit 'decay, crumble away, die' This -ya-present is poorly attested (4 times in the àatapatha-Brâhmana); only forms with active endings occur. The only accented occurrence bears stress on the suffix, cf : (19) yáthamapátrám udaká äsikte vimrityéd evám haivá té vímrityeyuh (áB 12.1.3.23) 'They assuredly would crumble away, as ajar of unbaked clay would crumble away, if water were poured into it' Of course, one might treat both of the above-discussed forms as merely erroneously accentuated and irregular. However, these errors may be not accidental, being due to a certain tendency. B o t h ) f 'waste away, grow old, decay* and mrit 'decay, die' are typical representatives of the 'entropy increase' class. I hypothesize that the accentual patterning in these ->w-presents, albeit exceptional, results from their semantic affinity with middle -ja-presents with fluctuating accentuation. Note that both of the above-discussed exceptions are attested in the MaiträyanT Samhitä and áatapatha-Brahmana, that is in the texts which prefer suffix accentuation (cf. section 3.2.1). One may assume that the corresponding Vedic dialects have had a weak tendency to group some active -^α-presents referring to 'entropy increase' processes together with middle -ja-presents with fluctuating accentuation, in virtue of semantic affinity and notwithstanding the difference in the diathesis (active/middle). Eventually, this may have given rise to their abnormal accentuation, which appears sporadically though.

4. The relevance of the notion of entropy 4.1. Entropy in a linguistic description Vedic -^α-presents with fluctuating accentuation represent a typologically interesting verbal class, since they belong to the semantic area ('entropy increase') which poses great difficulties for anyone concerned with the distinguishing passives from anticausatives. On the other hand, this semantic feature is of special importance for locating the core of the AC-class of verbs. 'Entropy increase' seems to be one of the most relevant and non-trivial semantic sub-components of spontaneity, and thus can be used in order to identify anticausatives as opposed to passives.

Passives, Anticausatives and Classification of Verbs: the Case of Vedi c

151

It is interesting to note that the notion of entropy may also be relevant for the describing anticausatives in other languages. For instance, almost 2/3 of Gothic anticausatives in -nan-10 quoted by Haspelmath (1987: 16-17) refer to 'entropy increase' processes (fralusnan 'get lost', fraqistnan 'perish', etc.). Haspelmath draws special attention to one of these verbs, andbudnan "be unbound, get loose', the anticausative counterpart of the transitive verb andbindan "bind', pointing out that only the prefixed verb andbindan "bind' can be anticausativized, whereas the corresponding simplex bindan "bind' cannot. Haspelmath attempts to account for this asymmetry in terms of the feature of "unspecificity" (in order a causative/anticausative alternation can be allowed, "the change may not be effected with too specific means"). It is rather difficult to argue, however, that binding is a more specific process that untying. In my opinion, this discrepancy may receive a simpler and more natural explanation in terms of entropy increase. Getting loose, unlike getting bound, is a process leading to the destruction of a system and to an increase of entropy (cf. section 3.2.2 (ii) on Vedic muc). 4.2. Entropy and other spontaneous processes Of course, not all of the spontaneous processes entail entropy increase.11 A number of natural processes which can come about spontaneously even imply decrease of entropy. Note, however, that the set of elementary processes of such kind is rather limited, mostly being due to some natural sources of energy, such as, above all, sun, fire, etc. In particular, these are processes concerned with heating Cheat', "become dry", 'shine', "boil'),12 growing, etc. It is beyond any doubt that processes leading to increase of entropy occupy an important position among natural phenomena. The universe is running down, everything will be ruined sooner or later, so that most of the natural processes contribute to this total ruin - the conclusion is unavoidably pessimistic.13 No wonder that the group of predicates referring to 'entropy increase1 processes belongs to the core of the class of spontaneous predicates. Yet, it would be probably premature to claim that the class of 'entropy increase' predicates is exactly the main core of the class of spontaneous predicates (= AC-class). For the reason of space, I only mention here one of the most interesting AC-features, first formulated by Otto Jespersen in his "English Grammar" (Jespersen 1927: 332-337; cf. also Smith 1978). Jespersen points out that many verbs referring to motion and/or change of state undergo causative alternation, Le. can be employed both transitively and intransitively, and labels this important class of verbs Move and Change-class (move, turn, boil, improve, etc.). The 'entropy increase' and 'move and change' classes overlap a little (cf fall, perish) but of course are not identical. There 10

For the function of Gothic -ηαη-verbs and problems of distinguishing between anticausative and passive interpretations, see Jansen 1988.

11

A sketch of semantic subclasses of spontaneous events can be found in Kemmer 1993: 142-147.

12

Cf., incidentally, two Vedic -yo-presents with fluctuating accentuation which do not belong to the 'entropy increase' class, pacyate 'is cooked, ripens' and tapyate "becomes hot, is heated'.

13

Recently it has been suggested (Brodes and Wiley 1986) that biological evolution also entails entropy increase, although in quite a specific sense of the word.

152

Leonid Kulikov

is no need to argue that the structure of the AC-class is rather intricate and, probably, language· specific. 4.3. Entropy vs. destruction It is worth emphasizing that not all of the processes concerned, in some ways, with destruction of things lead to increase of entropy. A destruction may pursue certain goal, be planned and brought about by an external agent and, eventually, lead to creation of a new object, instead of the destroyed one, or, in a sense, to a new order of things. Slicing vegetables in order to prepare a salad is not just a mere destruction of the ingredients; it gives rise to a new object. Likewise, slaying a sacrificial animal amounts to more than its death and is a part of the ritual, which should contribute to the right order of things, from the point of view of the sacrificer. The same is true, mutatis mutandis, for some specific (implying "agent-oriented semantic components", in Haspelmath's terms) kinds of killing, such as executing etc. This may account for that many processes of destruction are often treated as inherently transitive or, in other terms, as unmarked causatives (cf. Croft 1990: 61). Thus, entropy increase and destruction proper do not coincide, although they may overlap and share semantic components. The relevance of such semantic features as 'entropy increase' may be different for different languages. The case of Vedic is of special interest, since here this feature seems to acquire a "privileged" position. As a result, the 'entropy increase' verbs share a number of features which belong to different layers of the language structure (accentual behaviour, paradigmatic properties). No doubt, similarity of these formations as observed throughout the semantics, morphology and paradigmatic properties, cannot be accidental. This means that the -^α-presents with fluctuating accentuation are not a random group but a linguistically well-established verbal class. Thus, 'entropy increase' becomes, in a sense, a "morphologically influential" feature, being the main parameter organizing the above-listed -ja-presents to a structural unity and influencing the properties of the corresponding verbs. 14 A search for other semantic features, such as 'entropy increase', may be of particular importance for identifying the AC-class as well as for the description of the system of verbal classes on the whole.

Abbreviations AV = Atharva-Veda, AVP = Atharva-Veda, Paippaläda recension, Br. = Brähmanas, JB = Jaiminîya-Brâhmana, MS = MaiträyanT Samhitä, RV = Rg-Veda, SB = áatapatha-Brahmana, TB = TaittirTya-Brähmana, TS = TaittirTya-Samhitä

14

In particular, this feature may have been the main reason for building the secondary nasal presents ánásti, âmsati 'leaves' as the transitive counterpart of äsyate 'is left'.

Passives, Anticausatives and Classification of Verbs: the Case ofVedic

153

References Babby, L.H. (1983) The relation between causative and voice: Russian vs. Turkish, Wiener Slawistischer Almanach 11,61-88. Brooks, D.R., and E.O. Wiley (1986) Evolution as entropy: Toward a unified theory of biology. Chicago: Chicago University Press. Chambadal, P. (1963) Évolution et applications du concept d'entropie. Paris: Dunod. Comrie, B. (1981) Language universals and linguistic typology. Oxford: Basil Blackwell. - (1985) Causative verb-formation and other veit>-deriving morphology. In: Th. Shopen (ed.) Language typology and syntactic description, Vol. 3: Grammatical categories and the lexicon, Cambridge: Cambridge University Press, 301-348. Croft, W. (1990) Possible verbs and the structure of events. In: S.L. Tsohatzidis (ed.) Meanings and prototypes: Studies in lingustic categorization. London/New York: Routledge, 48-73. Gerritsen, N. (1988) How passive is 'passive' -sjal In: A.A. Barentsen et al. (eds.) Dutch studies in Russian linguistics. Amsterdam: Rodopi, 97-179. (Studies in Slavic and general linguistics, 11). Gonda, J. (1951) Remarks on the Sanskrit passive. Leiden: Brill. Haspelmath, M. (1987) Transitivity alternations of the anticausative type. Köln: Institut fiir Sprachwissenschaft. (Arbeitspapier 5 (N.F.)) - (1993) More on the typology of inchoative/causative verb alternation. In: B. Comrie and M. Polinsky (eds.) Causatives and transitivity. Amsterdam/Philadelphia: Benjamins, 87-120. Höhle, T.N. (1978) Lexikalistische Syntax: die Aktiv-Passiv-Relation und andere Infinitkonstruktionen im Deutschen. Tübingen: Niemeyer. (Linguistische Arbeiten, 67). Jansen, O. (1988) Zur Funktion der gotischen -nan-Verben. In: J.O. Askedal et al. (eds.) Gedenkschrifi fir Ingerid Dal. Tübingen: Niemeyer, 54-62. Jespersen, O. (1927) A modern English grammar on historical principles. Pt. III. Syntax. Second volume. Heidelberg Carl Winter. Kemmer, S. (1993) The middle voice. Amsterdam/Philadelphia: Benjamins. (Typological studies in language, 23). Kulikov, L.I. (forthcoming) Vedic -_ya-presents: semantics and the place of stress, Akten der 10. Fachtagung der Indogermanischen Gesellschaft. Innsbruck. Levin, B., and M. Rapoport (1994) A preliminary analysis of causative verbs in English, Lingua 92, 35-77. - (1995) Unaccusativity: at the syntax-lexical semantics interface. Cambridge, Mass.: ΜΓΓ Press. Masica, C.P. (1976) Defining a linguistic area: South Asia. Chicago: University of Chicago Press. Miller, J. Gary (1993) Complex verb formation. Amsterdam/Philadelphia: Benjamins. Mittwede, M. (1986) Textkritische Bemerkungen zur Maiträyarfi Sam hi ta. Stuttgart: Steiner. (Alt- und Neu-Indische Studien, 31). Moreno Cabrera, J.C. (1984) La diátesis anticausativa. Ensayo de sintaxis general. RSEL 14 (1), 21-43. - (1985) Anticausatives. A typological sketch, Chicago Linguistic Society 21 (2), 172-181. Nedjalkov, V.P., and G.G. Sil'nickij (1969) Tipologija morfologiCeskogo i leksiíeskogo kauzativa. In: A.A. Xolodoviò (ed.) Tipologija kauzativnyx konstrukcij: Morfologiöeskij kauzativ. Leningrad: Nauka, 20-50. Rifldn, J. (1989) Entropy: into the greenhouse world. New York etc.: Bantam books. Siewierska, A. (1984) The passive: A comparative linguistic analysis. London: Croom Helm. Smith, C.S. (1978) Jespersen's 'move and change' class and causative verbs in English. In: M.A. Jazayery et al. (eds.) Linguistic and literary studies in honor of Archibald A. Hill. Vol. II. Descriptive linguistics. The Hague etc.: Mouton, 101-109. Song, J.J. (1995) Review of: B. Comrie, M. Polinsky (eds.): Causatives and transitivity. Amsterdam/Philadelphia: Benjamins, 1993, Lingua 97, 211-224. Winford, D. (1988) Stativity and other aspects of the Creole passive, Lingua 76, 271-297.

Zur Typologie der Passivkonstruktionen in den Jenissej-Sprachen Heinrich Werner

Das Problem der Passivformen in den Jenissej-Sprachen kann zu den umstrittensten gerechnet werden, und dies hat seine guten Gründe, denn bisher werden von den Ketologen als Passiv ganz unterschiedliche Verbalformen betrachtet (vgl. insbesondere die unterschiedliche Auffassung der Passivformen in Dulson 1968 und Krejnovic 1968). Eine passive Umwandlung, bei der ein Patiens die linke Terminalposition im Satz, die Position des pragmatischen Gipfels, einnehmen kann, scheint auf den ersten Blick ganz problemlos zu sein, vgl. (1)

ket. (ketisch) ke?t irjGus' dub-bet 'der Mensch baut (wörtl. macht) ein Haus' > irjGus' ker'as' bimba-vet 'das Haus ist vom Menschen gebaut (wörtl. gemacht)'; ke?t iijGus' dbil'-bet 'der Mensch baute (wörtl. machte) ein Haus' > irjGus' ker'as' bil'a-vet 'das Haus war vom Menschen gebaut (wörtl. gemacht)'.

Man fragt sich jedoch, ob es sich in solchen Fällen tatsächlich um eine passive Umwandlung handelt, oder ob man es bloß mit einer Beseitigung des Agens aus der linken Terminalposition im Satz zu tun hat, wobei, wie in Van Valin/Foley 1980 behauptet wird, keine passive Umwandlung vorliegt. Als E. A. Krejnovic ketische Beispiele wie (2)

tipdajoj 'der Hund haart', s'i?avoj 'das Rentierfell haart' und bu un'äq dsvun' 'er stellt das Fischnetz?, un'äij avun' 'das Fischnetz ist gestellt'

miteinander verglich, ist er zur Schlußfolgerung gekommen, daß die zwei ersten Beispiele ihrer Struktur nach identisch sind, die zwei anderen aber nicht, denn hier tritt das Nomen unti g 'Fischnetz1 in zwei verschiedenen Funktionen auf: als Objekt und als Subjekt; im letzteren Fall bezeichnet die Verbalform avun' den infolge einer Handlung eingetretenen Zustand. Solche Verbalformen bezeichnete E. A. Krejnovic als intransitiv-passiv mit resultativer Bedeutung (Krejnovic 1968: 249). Diese Deutung ist dann in Polenova 1986 bestätigt worden, und sie entspricht völlig der von V. P. Nedjalkov entwickelten Theorie über die Verbalformen des Passivs-Resuhativs (siehe Nedjalkov 1981: 153-161; 1983: 5-52). Interessant ist in diesem Zusammenhang die Beobachtung von B. A. Uspenskij, daß die ketischen transitiven Verbalformen in solchen Fällen genauso wie in den ergativischen Sprachen wie intransitive gestaltet werden (Uspenskij 1964; 1968: 204-205). Auf diese Konzeption der Intransitivität stützt sich auch E. A. Krejnovic' Auffassung der Passivität im Ketischen (Krejnovic 1968: 244-260); während aber B. A. Uspenskij in diesem Fall eher eine Analogie zwischen dem Ketischen und den ergativischen Sprachen sieht und folglich den passiven Charakter der entsprechenden ketischen Verbalformen in Zweifel zieht (Uspenskij 1968: 205), findet E. A. Krejnovic ketische transitive, intransitive und passive Verbalformen, die sich formal voneinander unterscheiden, und den

156

Heinrich Werner

Autor veranlassen, von einer in dieser Sprache entstehenden Passivkategorie zu sprechen (Krejnovic 1968: 257-259). Sollte man sich in dieser Frage der Ansicht von B. A. Uspenskij anschließen, so müßte man die oben unter (1) angeführten Beispiele ket. irjGus' keras' bim-bavet, iifius' ker'as' bil'avet entsprechend wie 'das Haus baut sich (entsteht) durch den Menschen', 'das Haus baute sich (entstand) durch den Menschen' übersetzen, und die Keten würden diese Übersetzung vollständig akzeptieren, genauso wie sie die oben angeführte Übersetzung akzeptieren. Merkwürdigerweise stimmen mit bimba vet ketische Verbalformen wie folgende überein, die aber immer wie übliche Intransitiva übersetzt werden: (3)

ket. ture biade bimba-ray 'dieses zerknüllt sich selbst', Λ: binde bimba-l'okq 'das Baumblatt zittert von selbst', i's' bimba-raqg 'das Fleisch brät' (wird durch das Braten gar), i s' bimb-aq < *bimba-aq 'das Fleisch faultVdas Fleisch wird faul' usw. (vermutlich läßt sich das tón-Element in diesen Gegenwartsformen auf bin 'selbst' zurückfuhren).

Aufgrund solcher Beispiele läßt sich allerdings in formaler Hinsicht keine bestimmte Grenze zwischen aktiven intransitiven und passiven resultativen Verbalformen ziehen. Dies ist aber nur der Fall, wenn sich die Aktivformen der Transitiva auf ein unbelebtes Objekt beziehen. Eigentlich hat man es dabei mit einem Fall wie im Russischen zu tun, vgl. (4)

ket. i's' binde bimb-a-raqrj i's' degas' bimb-a-raqg

russ.

Maco caMO acaHT-ca MUCO

JiioflbMH acapHT-c«.

Man denke hier je nach dem Kontext an homonyme Formen wie bimbaraij 'es zerknüllt sich' (Medium) - bimbaraq 'es ist [von jemandem] zerknüllt' (Passiv-Resultativ), wie es auf ähnliche Art Nedjalkov (1981: 154) für die deutschen Beispiele die Äpfel sind (allmälich) verfault (Perfekt) - die Äpfel sind (ganz) verfault (Resuhativ) vermutet. Jedoch können sich die entsprechenden russischen Passivformen im Unterschied zu den ketischen auch auf belebte Objekte beziehen; außerdem sind die russischen Formen übliche Passivformen (Formen des Vorgangspassivs), während sich die ketischen nur als Resuhativformen übersetzen lassen. Das Problem ist also in den Jenissej-Sprachen viel komplizierter, und davon zeugen volle Paradigmata wie das folgende jugische:

157

Zur Typologie der Passivkonstruktionen in den Jenissej-Sprachen

(5)

Aktivformen du-di-bet' du-gi-bef du-ja-bet' du:-bef du-b-bet' du-dai}-bet' du-garj-bet' du-jaif-bet'

er macht mich er macht dich er macht ihn er macht sie (Sg.) er macht es er macht uns er macht euch er macht sie (PI.)

Passivformen ba-jar bet' ku-ja-bet' arjarbet' J-ja-bet' bimb-a-bet' dAij-a-bet' kAQ-a-bet' aq-a-bet'

ich bin gemacht du bist gemacht er ist gemacht sie ist gemacht es ist gemacht wir sind gemacht ihr seid gemacht sie sind gemacht1

Die passive Umwandlung wird also in diesem Paradigma durch folgende Mittel erreicht: 1) Weglassen des Subjektaffixes du-, welches in den Aktivformen auftritt; 2) Umwandlung der Objektaffixe der Aktivformen in Subjektaffixe der Passivformen (im angeführten Paradigma werden dabei die Personalaffixe der Gruppe D durch die Personalaffixe der Gruppe Β ersetzt, was aber nicht immer der Fall ist); 3) in die Passivformen wird zusätzlich das a-Affix infigiert. Nach Ε. A. Krejnovic (1968: 244 ff) weist dieses a-Affix auf die Intransitivität und Passivität der entsprechenden Verbalformen hin; dies ist aber nicht der Fall, und zwar aus folgenden zwei Gründen: zum einen kommt dieses a-Affix in intransitiven und transitiven Verbalformen vor, die mit der Passivität nichts zu tun haben, und zum anderen werden die Passivformen vieler Verben ganz anders gebildet und enthalten überhaupt kein a-Affix. Im allgemeinen kann man auf folgende drei Möglichkeiten der Bildung von Passivformen in den Jenissej-Sprachen hinweisen: 1. Passivformen, die einfach durch Weglassen der Subjektaffixe der Aktivformen gebildet werden: (6)

1

jug. (jugisch) ad do?n d-abi? 'ich wetze das Messer1 > do?n a~bi? 'das Messer ist gewetzt'; ad cb?n d-obini? 'ich habe das Messer gewetzt' > do^n obini? 'das Messer war gewetzt'; Passivformen ket. Aktivformen a-js' er ist angezogen d-a-js' er zieht ihn an J-js· sie ist angezogen d-i-js' er zieht sie (Sg.) an o-l's' er war angezogen d-o-l's' er zog ihn an i-rul's' sie war angezogen usw. d-i-rul's' er zog sie (Sg.) an

Z.B. ket. es'kAJibag dbi I'bet haj doqol'bel bil'de defy es'as' ayavet, haj at es'as' bajaveL 'Der Gott schuf (wörtl. machte) die Welt und schuf (wörtl. machte) die Menschen. Alle Menschen sind vom Gott geschaffen (wörtl. gemacht), auch ich bin vom Gott geschaffen (wörtl. gemacht)'.

158

Heinrich Werner

2. Passivformen, die durch Weglassen der Subjektaffixe der Aktivformen und Infigierung des a-Affixes gebildet werden: (7)

jug. at tuda d-utcap-tuj 'ich hänge dieses auf > tuda utcab-a-tuj 'dieses ist aufgehängt'; at tuda d-utcobin-tuj 'ich habe dieses aufgehängt' > tuda utcobin-a-tuj 'dieses war aufgehängt'; ket. at tur'e d-es'ab-daq 'ich lege dieses hin' > tur'e es'av-a-raq 'dieses ist hingelegt'; bu dil' d-es'a-mq 'er legt das Kind hin' > dil' es'a-j-a-raq 'das Kind ist hingelegt' usw.

3. Passivformen, die durch Weglassen der Subjektaffixe der Aktivformen und den Wechsel von Wurzelmorphemen in der Nullposition2 innerhalb der entsprechenden Verbalformen gebildet werden: (8)

jug. at si? di-xeil'gib-it 'ich breite das Rentierfell aus' > si?xeil'-gib-ot < xeil'-gib-xot 'das Rentierfell ist ausgebreitet'; ket. at ture t-qavi-t'ich melde es' > tur'e qavi-ta 'dieses ist gemeldet'; bu qu?s' da-hap-to 'sie stellt ein Birkenrindezelt auf > qu?s' hap-ta 'das Birkenrindezelt ist aufgestellt'.

Nach E. A. Krejnovic' Auffassung sollte man bei der Bildung der Passivformen im Ketischen und Jugischen dagegen vor allem von Affixen der Intransitivität und Passivität ausgehen, zu welchen er folgende gerechnet hat: 1) das schon erwähnte a-Infix und 2) folgende Suffixe: j, -ij, -ej, -aj, -Rut'-ut Meines Erachtens lassen diese Elemente eine ganz andere Deutung zu. Was die Elemente -j, -ij, -ej, -aj, -Rut'-ut anbetrifft, die E. A. Krejnovic als Suffixe betrachtet hat, so nehmen sie in den Verbalformen immer nur die Position eines Wurzelmorphems ein und können nicht als Suffixe gedeutet werden. Es handelt sich in den entsprechenden Fällen um einen Wechsel von Wurzelmorphemen, der nicht mit der Passivität, sondern mit der Transitivität vs. Intransitivität verbunden ist; vgl folgende zwei Paradigmata aus dem Ketischen:

2

Wie in Werner 1995: 52-68 gezeigt, liegt einer Verbalform in den Jenissej-Sprachen ein durch Wurzelmorpheme (Rj, R2, R3) gebildeter Rahmen zugrunde: (R3) + R2 + ... + Ri. Es gibt Verbalformen, in denen die Wurzelmorpheme R2 und R3 fehlen; dagegen kann das Wurzelmorphem Ri, welches die Nullposition bzw. die Ausgangsposition innerhalb der Verbalform einnimmt, nie fehlen. Bei der Bildung der Passivformen mancher Verben wird das Wurzelmorphem R j in der entsprechenden Position durch ein anderes ersetzt.

159

Zur Typologie der Passivkonstruktionen in den Jenissej-Sprachen

(9)

Transitivum (neutrale Version) sie versetzt mir einen Hieb (mit der Axt) da-ka-yu-Ro sie versetzt dir... da-k-a-j-Ro sie versetzt ihm... da-k-i-j-Ro sie versetzt ihr... da-ka- v-Ro sie versetzt ihm (einem Gegenstand)... da-ka-farf-Go sie versetzt uns... da-ka-yarj-Go sie versetzt euch... da-k-aij-a-Ro sie versetzt ihnen... da-kon-di-Ro sie versetzte mir einen Hieb (mit der Axt) da-kon '-gu-Ro sie versetzte dir... da-k-o-n-Go sie versetzte ihm... da-k-i-tnu-Ro sie versetzte ihr... da-ko-v-in-Go sie versetzte ihm (einem Gegenstand)... da-kon-daq-Go sie versetzte uns... da-kon '-gai)-Go sie versetzte euch... da-k-otj-on-Go sie versetzte ihnen... da-ka-fi-Ro

Intransitivum (Subjektversion) t-ka-r-Roj (k)-ka-yu-Roj t-ka-j-a-Roj da-ka-j-a-Roj karV-a-Roj t-ka-r'arj-Goj (kJ-ka-yay-Goj t-ka-j-ay-Goj t-kon-di-Roj (k)-kon

'-gu-Roj

t-kon-a-Roj da-kon-a-Rqj ko-v-in-a-Roj t-kon-darj-Goj (k)-kon

'-garj-Gqj

t-kon-aij-Goj

ich versetze mir einen Hieb (mit der Axt) du versetzt dir ... er versetzt sich... sie versetzt sich... es ist abgehauen/zerhauen (unbel. Klasse) wir versetzen uns... ihr versetzt euch... sie versetzen sich... ich versetzte mir einen Hieb (mit der Axt) du versetztest dir... er versetzte sich... sie versetzte sich... es war abgehauen/zerhauen (unbel. Klasse) wir versetzten uns... ihr versetztet euch... sie versetzten sich...

Man könnte vermuten, daß in diesem Fall das Roj -Morphem von Ra abgeleitet sei; jedoch berücksichtige man, daß das Verbum in den Jenissej-Sprachen keine derartigen Derivationsaffixe kennt, folglich hat man es bei Roj mit einem historischen Kompositum zu tun, welches aber infolge der phonetischen Veränderungen heute nicht mehr als solches wahrgenommen werden kann, und man gehe deshalb einfach vom Wechsel der Wurzelmorpheme Ro : Roj aus3 . In Beispielen wie (10) ket. a?q qot aRas-arj-aj 'das Holz ist schon gesägt', na?n' qot ha's'a-j-aj'das Brot ist schon geschnitten1 (vgl. ket. ar' a?q d-aRas-a 'ich säge Holz1, ar a?q d-aRasl'-a 'ich sägte Holz1; atnafo' t-has'-a 'ich schneide Brot', at nafa' t-hal'-a 'ich schnitt Brot') vermutete E. A. Krejnovic das Passivaffix -aj im Auslaut, welches durch das /Interfix vom Wurzelmorphem -a- getrennt sei (Krejnovic 1968: 256); in der Tat ist -aj das Wurzelmorphem der Passivform, welches mit dem Wurzelmorphem -a der Aktivform wechselt, und das letztere 3

Infolge der historischen phonetischen Veränderungen ist das Wurzelmorphem in einigen Fällen sogar ganz verschwunden, was aber an und für sich kein Grund dazu ist, das Vorhandensein des Wurzelmorphems in der Ausgangsposition (Werner 1995: 52-531Γ.) der jenissejischen Verben überhaupt abzulehnen; vgl. mittelketisch dib-a, südketisch dip 'ich esse es', wo in der letzteren Verbalform das Wurzelmorphem -a verschwunden ist.

160

Heinrich Werner

darf mit dem infigierten a-Affix in den Passivformen aRas-a-j-aj und has'-a-j-aj nicht verwechselt werden. Auch im Kottischen hat das /-Element im Auslaut, welches dem ketischen auslautenden jElement entspricht (Starostin 1995), mit dem Passiv nichts zu tun, denn es kommt nicht in allen Passivformen vor (wogegen das in die kottischen Passivformen infigierte a-Aifix unvermeidlich ist). Also sollte man auch bei der Deutung der entsprechenden kottischen Verbalformen nicht vom Passivaffix -i, sondern vom Wechsel der Wurzelmorpheme ausgehen. Dies läßt sich mit kottischen Beispielen wie folgende veranschaulichen, in denen es sich in bestimmten Fällen eben um einen Wechsel der Wurzelmorpheme in der Null- bzw. Ausgangsposition und um kein /-Affix handelt: (11) Aktivformen ich stellte auf ich steckte hinein bala-fuj-ag ich blies bala-paj-ai] ich machte o:na-fu:j-a¡η ich flocht o:la-fi-ar¡ ich löschte o.na-si-aq ich kämmte o:na-sal-ai) ich schliff agola-ki: t-ar) ich schnitzte oga?a: la- co-ay ich tötete ihn a:la-pe:j-aq ich jagte acantholo:-k-r¡ ich jagte Tiere a:na-thij-aq ich schlug aca:?a:l-aj-aq ich rief ihn aspo: rai'-ic-aij ich hob auf a: ra: thala-paj-aq ich betrügte althu:r-taj-aq ich nahm heraus o:na-fu:l-arj ich drehte

Passivformen ist aufgestellt ist hineingesteckt -fu bal-a:-fu ist (auf)geblasen -pe bal-a:-pe ist gemacht fui o:n-a:-fui ist geflochten -fi o:l-a:-fi ist gelöscht -si οα-a:-s i ist gekämmt -sal on-a:-sal ist geschliffen -ki:t agol-a:-ki:t ist geschnitzt -cex ogal-a:-cex ist getötet -pei al-a:-pei ist gejagt -ki acanthol-a: u-ki auf der Jagd erworben -Λ an-a:-thi ist geschlagen -gi aca:?a:l-a-gi ist gerufen -gas aspor-a:-gas ist gehoben h -pi a:rut al-a:-pi ist betrogen -tei altho:r-a:-tei ist herausgenommen -ful an-a:-ful ist gedreht4

-tek-

akfui-tek-rj

-teki

-tek-

wan-tek-η

-tek

-fuj-paj-fu:j-fi-si-saJ-ki:t-ce-pe:j-k-thij-aj-ic-paj-taj-fu:J-

akfuij-a-:

teki

u:an-a:-tek

Eine ganz besondere Erscheinung stellt das in die Passivformen infigierte a-Affix dar. Wie bereits betont, kann man diesem Affix keine passive Bedeutung zuschreiben, wie E. A. Krejnovic glaubte, obwohl es aufs engste mit den Passivformen verbunden ist. Meines Erachtens gehört es zu den Versionsaffixen, und man sollte auf diese Frage tiefer eingehen, um die Natur der jenissejischen Passivformen zu verstehen.

4

In Castren 1858: 139 werden die Passivformen mit Perfekt- und Plusquamperfektformen übersetzt, z.B. ola:kiti 'es ist/war beschmiert'.

161

Zur Typologie der Passivkonstruktionen in den Jenissej-Sprachen

Im Ketischen und Jugischen lassen sich zwei Reihen von Personalaffixen unterscheiden, die zum Ausdruck der Subjektversion dienen: Versionsaffixe D l.P.Sg. 2.P.Sg. 3.P. Sg. 1.P.P1. 2.P.P1. 3.P.P1.

di! ri! dl

t

ku / gu lk Ig / γ

Versionsaffixe Β ba, bo ku / gu

a

bu

dag / rag

dag / rag

kag / gag / yag

kag I gag / yag

ag

bu

Mit der Intransitivität haben diese Affixe nichts zu tun, denn sie können in transitiven und intransitiven Paradigmen auftreten. Man wollte ihnen immer wieder reflexive Bedeutung zuschreiben, da diese Bedeutung mitinbegriffen sein kann, vgl. (13) ket. du-j-a-s' 'er zieht sich an', dA-j-a-s' 'sie zieht sich an', t-tu/un'-bu-tayit sich', da- tuyun '- bu- tayit 'sie kämmt sich' usw.

'er kämmt

Vielen Verbalformen dieser Art läßt sich aber keine reflexive Bedeutung zuschreiben: (14) t-ta-J-a-raq 'er fällt um', darta-j-a-mq 'sie fallt um', jug. ta-b-a-dax 'es fallt'; ket. d-bu1 tsaq 'er läuft mal hin , da-bu-tsaq 'sie läuft mal hin'; t-til'-a-Ro 'er lud es (das Gewehr)', da-til'-a-Ro 'sie lud es (das Gewehr)' usw.

Im Grunde genommen werden die Reflexivformen im Rahmen der entsprechenden transitiven Paradigmata ohne die angeführten Versionsaffixe D und Β gebildet, wobei oft vor den betreffenden Verbalformen zusätzlich das Pronomen bin 'selbst' erscheint: (15) jug. durd'-a-j-gfig'ich

wasche ihn', durd'-i-J-gi^g 'ich wasche sie',aber: durd'a-di-gi^g / bindi durd'a-di-gfig 'ich wasche michVich wasche mich selbst'; ket. di-y-a-tos' 'ich erziehe ihn', di:-tos' 'ich erziehe sie', d-o-l'tos' 'ich erzog ihn', d-i-rul'tos' 'ich erzog sie', aber: di-t-tos' / bindi di-t-tos' 'ich erziehe michVich erziehe mich selbst', diJ'-di-tos' / bindi dil'-di-tos''ich erzog michVich erzog mich selbst' usw.

Es können auch solche Reflexivformen innerhalb der transitiven Paradigmata der Subjektversion gebildet werden: 'ich verkaufe ihn1, d-u-g-di-χϊρ 'ich verkaufe sie', aber: / bindi di-bo-g-di-χί-ρ 'ich verkaufe mich' / 'ich verkaufe mich selbst'.

(16) jug. d-o-g-di-χί'ρ

dj-bo-g-di-χίρ

Heinrich Werner

162

Der Unterschied zwischen den Versionsaffixen der D- und B-Reihe bleibt bislang unklar; der Bedeutung nach sind die entsprechenden Verbalformen identisch. Es lassen sich, wie bemerkt, intransitive und transitive Paradigmata mit diesen Affixen beobachten: Verbalformen mit den Versionsaffixen D (17)

Neutrale Version jug- dura-gify kura-gi?η dura-gi?η daura-gif η dura-gÍQ-in kura-giq-in durargiij-in

(18)

Intrans. Versionsparadigma ket. di-rj-Ra ku-yu-Ra du-j-a-Ra dA-j-a-Ra dj-r'aη-Ga ku-yaq-Ga du-j-aq-Ga diJ'-dj-Ra kil'-gu-Ra dil'-a-Ra dail'-a-Ra dίl'-daη-Ga kίl'-gaη-Ga dil'-aq-Ga

5

ich wasche du wäscht er wäscht sie wäscht wir waschen ihr wascht se waschen

Subjektversion dura-d-gi7η kura-k-kPrj dura-j-a-gi7η daura-j-a-gi?q dura- daq-giq-in kura-gaη-gíη-ín dura-j-aq-giq-in

ich wasche mich du wäscht dich er wäscht sich sie wäscht sich wir waschen uns ihr wascht euch sie waschen sich5

Trans. Versionsparadigma

ich verkaufe du verkaufst er verkauft sie verkauft wir verkaufen ihr verkauft sie verkaufen

d-o-g-di-Ra k-o-k-ku-Ra d-o-k-s-a-Ra da-o-k-s-a-Ra d-o-g-dag-Ga k-D-k-kaq-Ga d-o-k-s-ag-Ga

ich verkaufe ihn du verkaufst ihn er verkauft ihn sie verkauft ihn wir verkaufen ihn ihr verkauft ihn sie verkaufen ihn

usw.

ich verkaufte du verkauftest er verkaufte sie verkaufte wir verkauften ihr verkauftet sie verkauften

d-o-l'-di-Ra k-o-l'-gu-Ra d-o-l'-a-Ra da-o-l'-a-Ra d-o-l'-daq-Ga k-o-l'-gaq-Ga d-o-I'-ag-Ga

ich verkaufte ihn du verkauftest ihn er verkaufte ihn sie verkaufte ihn wir verkauften ihn ihr verkauftet ihn sie verkauften ihn

usw.

In den Formen der neutralen Version fehlt die Reihe der Versionsaffixe. Die Formen der Subjektversion werden genauso wie die üblichen Reflexivformen übersetzt; die letzteren werden aber nach dem Muster des transitiven Paradigmas gebildet. Dies ist aber kein Grund dazu, um beide Paradigmata zu verwechseln (vgl. mit den Formen der Subjektversion folgende transitive Formen, unter denen es auch übliche Reflexivformen gibt: dwd'-a-j-gt?!]'ich wasche ihn', dwd'-t-j-gtfq'ich wasche sie', aber: dwd'a-di-gi?q'ich wasche mich').

163

Zur Typologie der Passivkonstruktionen in den Jenissej-Sprachen

Verbalformen mit den Versionsaffixen Β (19)

di-j-aq ku-j-aq du-j-aq dA-j-aq di-j-aR-an ku-j-aR-an du-j-aR-an

(20)

Subjektversion

Neutrale Version ich gehe mal hin du gehst mal hin er geht mal hin sie geht mal hin wir gehen mal hin ihr geht mal hin sie gehen mal hin

Intrans. Versionsparadigma d-ba-qluRo (k)-ku-rjIuRo d-bu-rjluRo da-bu-qluRo (φ-dAíj-luRon (k)-kAQ-luRon d-bu-rjJuRon

ich schaute du schautest er schaute sie schaute wir schauten ihr schautet sie schauten

d-ba-tsaq (k)-ku-tsaq d-bu-tsaq da-bu-tsaq (d}-dAQ-st-aR-an (k)-kAQ-st-aR-an d-bu-ts-aR-an

ich laufe mal hin du läufst mal hin er läuft mal hin sie läuft mal hin wir laufen mal hin ihr lauft mal hin sie laufen mal hin

Trans. Versionsparadigma sie zog mich heraus sie zog dich heraus dü-bu-t-o-n-daq sie zog ihn heraus' da-bu-dd-i-tn-daq sie zog sie heraus da-bu-ton-daq-daq sie zog mis heraus da- bu- ton-garf-daq sie zog euch heraus da- bu- t-otj-on- daq sie zog sie (PI.) heraus usw. da- bu- ton- di-raq da-bu-ton-gu-raq

Die Grundbedeutung der Versionsformen ist also viel breiter als die der üblichen Reflexivformen; sie bezeichnen eine Handlung, die aufs engste mit dem Subjekt verbunden ist, und in der Regel im Interesse oder zugunsten der handelnden Person ausgeführt wird 6 . Eben deshalb ist die reflexive Bedeutung in solchen Fällen meist mitinbegrifien. Merkwürdig ist aber, daß auch bei einem unbelebten Subjekt intransitive Verben das Versionsaffix der 3.P.Sg. -a- annehmen können 7 . Solche Verbalformen haben in der Regel mediale Bedeutung:

6

Diese Bestimmung der Grundbedeutung der Versionsformen ist nur unter bestimmtem Vorbehalt anzunehmen, denn im Grunde genommen wird in solchen Fällen die Betroffenheit der handelnden Person durch die entsprechende Handlung auf eine besondere Art hervorgehoben.

7

Das iu-Affix läßt sich nur selten in Beispielen wie ket. bej at da-bu-g-d-it 'der Wind treibt mich' beobachten; sie fallen völlig mit Verbalformen zusammen, in welchen das Subjekt der weiblichen Klasse zum Ausdruck kommt.

164

Heinrich Werner

(21) jug. boksej

bimb-a-χο.Γ

'die W u n d e heilt', samma

iimò-a-p/n

zeitig', cfis tab-ard.αχ 'der Stein fallt', ket. Λ: η bimb-a-]'okrj

'die Beeren werden

'das Laub zittert', nah'

kab-

a-tij 'das Brot geht zur Neige', da:π ab-artij 'das Gras wächst' usw.; eben diese Bedeutung konnte in Verbalformen wie bimb-a- vet 'es ist gemacht', rs' bimb-a-raqq 'das Fleisch ist gebraten', s'ujat bimb-a-r'ag 'das Kleid zerknüllt sich' / 'das Kleid ist zerknüllt', ta·η av-a-ro 'das Haar ist geschnitten' usw. als passiv umgedeutet werden (vgl. die analogen russischen Formen in PEKH HanojmaioTca BOJIOH [CAMH] — cocy^bi HanoJiHHIOTCH BOFLOH [κβΜ-το], njiaTbe MHÖTC« [caMo] — njiaTbe MHCTCH [κβΜ-το], 3TO FLEJIAETCJI [CÜMO] — 3To flejiaeTCH [κβΜ-το].) Dieses Entwicklungsschema erinnert an die Entstehung des Mediopassivs und Passivs aus dem Medium in den indogermanischen Sprachen (siehe hierzu insbesondere Savcenko 1974: 294-310). Merkwürdigerweise bezeichneten die Mediumformen in manchen indogermanischen Sprachen, z.B. im Indo-Iranischen, eine Handlung, die der handelnden Person zugunsten ausgeführt wird, und A. Meillet hielt diese Bedeutung für die Grundbedeutung des indogermanischen Mediums (Meillet 1938: 257). Am produktivsten sind von den jenissejischen passiven Konstruktionen die Konstruktionen mit den Passivformen auf -bet I - vet (< südket. bei", mket. be:da, jug. beh: t' 'machen'), die im Ketischen als haben-Konstruktionen gebraucht werden. Diese Passivformen werden nach folgendem Muster gebildet (von bogdom 'Flinte'): (22)

at bogdom-ba-j-ar u bogdom-ku-j-abu bogdom-a-j-abu bogdom-i-j-a-

vet vet vet vet

'ich habe eine Flinte', wörtl. 'ich bin beflintet'; 'du hast eine Flinte', wörtl. 'du bist beflintet'; 'er hat eine Flinte', wörtl. 'er ist beflintet'; 'sie hat eine Flinte', wörtl. 'sie ist beflintet'.

In den Pluralformen kann das Versionsaffix -a- fehlen: atn bogdom-dAQ-a-vet / atn bogdom-dAg-bet okq bogdom-kAQ-a-vet/

akq bogdom-kAij-bet

bu i7 bogdom-aiya- vet / bu r) bogdom-arf-bet

'wir haben eine Flinte', wörtl. 'wir sind beflintet'; 'ihr habt eine Flinte', wörtl. 'ihr seid beflintet'; 'sie haben eine Flinte', wörtl. 'sie sind beflintet';

Analoge Passivformen entstehen von sehr vielen ketischen Nomina: dil'gat 'Kinder', dil'tij 'kleiner Kahn', as'1'εηεη

'Boot', mnul'

'Brötchen', il'dup 'Angel', tes'ig 'Filzstiefel', tA/jgat 'Ofen'

usw. Unerwartet erscheinen neben diesen passiven Verbalformen auch solche, in denen das Versionsaffix -a- überhaupt fehlt. Im Unterschied zu den angeführten Formen mit bogdom 'Flinte', in denen die Personalaffixe Β der Reihe ba, ku, a, i usw. auftreten, werden in den entsprechenden Passivformen ohne das Versionsaffix -a- die Personalaffixe D der Reihe di, ku, a, i usw. gebraucht. Als Beispiel folgt hier das Paradigma solcher Passivformen vom Nomen do?n 'Messer1:

Zur Typologie der Passivkonstruktionen in den Jenissej-Sprachen (23) at don-di-vet u don'-gu-vet bu don'-a-j-bet bu don'-i-j-bet ato don-darbet okij don'-gaij-bet bu g don'-aq-bet

165

'ich habe ein Messer1, wörtl. 'ich bin bemessert'; 'du hast ein Messer1, wörtl. 'du bist bemessert'; 'er hat ein Messel4, wörtl. 'er ist bemessert'; 'sie hat ein Messer*, wörtl. 'sie ist bemessert'; 'wir haben ein Messer*, wörtl. 'wir sind bemessert'; 'ihr habt ein Messer1, wörtl. 'ihr seid bemessert'; 'sie haben ein Messer1, wörtl. 'sie sind bemessert'.

Auch in diesem Fall können analoge Formen von sehr vielen ketischen Nomina gebildet werden: ej'Zunge', din' 'Tanne', a m 'Mutter*, im 'Zedernüsse', i?n "Nadel', qu?s' 'Birkenrindezelt', tok'Axt', l'a?m'Brett' usw. Solche Verbalformen, die nach zwei unterschiedlichen Modellen gebildet werden (mit dem a-Affix wie in bogdom-baj-a- vet 'ich bin beflintet' und ohne das a-Affix wie in don-di- vet 'ich bin bemessert'), sind nur iiir das Ketische kennzeichnend. Die Kriterien der Einteilung der Nomina in zwei Gruppen je nach den zwei Bildungsmodellen der entsprechenden Verbalformen bleiben bislang unklar; vermutlich handelt es sich um ein morphonologisches Merkmal: nach dem ersten Modell werden Verbalformen von mehrsilbigen und nach dem zweiten von einsilbigen Nomina gebildet. Dabei werden Verbalformen von einsilbigen Singularformen und mehrsilbigen Pluralformen entsprechender Nomina nach unterschiedlichen Modellen gebildet. Merkwürdigerweise gehören einsilbige Nomina zur Gruppe der mehrsilbigen, wenn sie historisch auf zweisilbige zurückgeführt werden können: (24) ket. s'u.7''Polarschlitten' < *s'uyal' (vgl. kott. cogar): s'u:l'-ba-j-a-vet'ich habe einen Polarschlitten' (wörtL 'ich bin beschlittet'); ket. er'Zobel' < *ετ'ί (vgl. mket. ε: da, nket. ε: ri, kott. ija): er'-ba-j-a-vet 'ich habe einen Zobel' (wörtl. 'ich bin bezobelt'); ket. i?n, PL m'aij "Nadel1, vgl.: in-di- vet 'ich habe eine Nadel' (wörtl. 'ich bin benadelt') m'aq- ba-j-a-vet 'ich habe Nadeln' (wörtl. 'ich bin benadelt'). Wie schon erwähnt, hat E. A. Krejnovic als erster entdeckt, daß es in manchen Fällen einen formalen Unterschied zwischen entsprechenden transitziven, intransitiven und passiven Verbalformen gibt, und unsere Materialien bestätigen diese Beobachtung. Vgl. folgende ketische Beispiele:

166

Heinrich Werner

ich biege den Baum der Baum biegt sich der Baum ist geoks' hal's'aRut bogen t- toRojiqqadda er trocknet mich ab ttaRjjiqqaddij ich trockne mich ab toRojiijqis'qut ich bin abgetrocknet ar' u'I' da: tfiabda ich erwärme das Wasser uI'a:taRan das Wasser wird warm ul' a: qqavafij das Wasser ist erwärmt

(25) at

oks'

ich bog den Baum der Baum bog sich oks' hal'l'aRut der Baum war gebogen t-toRojiqqol'dida er trocknete mich ab t- toRDjiqqol'didij ich trocknete mich ab toRjjirjqil'Gut ich war abgetrocknet ar' ul' da: lfiovil'da ich erwärmte das Wasser ul· a.toRon das Wasser wurde warm ul' a: qqovil'ar'ij das Wasser war erwärmt

oks't-hal'dajit

at oks'

t-hal's'atn

oks'

t-hal'dona

t-hal'l'atn

Eben solche Fälle haben E. A. Krejnovic zur Schlußfolgerung bewogen, daß im Ketischen die Kategorie des Passivs entsteht. Dieser Ansicht kann man meines Erachtens beistimmen. Die angeführten Materialien zeigen, daß die Bildung der Passivformen in den Jenissej-Sprachen vor allem mit zwei Erscheinungen verbunden ist, und zwar: (1) dem Weglassen der Subjektaffixe und (2) dem Erscheinen des a-Affixes der Subjektversion. Die Umwandlung der aktiven transitiven Verbalformen in intransitive passive durch Weglassen der Subjektaffixe, wobei die Objektaffixe die Funktion der Subjektaffixe übernehmen, scheint in den Jenissej-Sprachen das wichtigste Mittel der Passivbildung zu sein. Die auf diese Art gebildeten Passivformen sind wie aktive intransitive Verbalformen gestaltet und unterscheiden sich von den letzteren bei einem unbelebten Subjekt nur dadurch, daß sie einen infolge einer bestimmten Handlung eingetretenen Zustand bezeichnen, vgl. die unter (2) angeführten Beispiele, und zwar: (a) ket. s'i? avoj'das Rentierfell haart' (Aktivform), (b) ket. un'äij avun' 'das Fischnetz ist gestellt' (Passiv-Resultativ). Oft sind die Verbalformen dieser Art zweideutig: der entsprechende Zustand konnte von sich selbst oder auch infolge einer bestimmten Handlung eingetreten sein: (26) ket. s'ujat un'avil'aron 'das Hemd ist zerrissen' [(a) ist von selbst zerrissen, (b) ist von jemanden zerrissen worden], vgl. at s'ujat dun'avil'don 'ich habe das Hemd zerrissen'; bo?k un'i 'das Feuer ist ausgegangenTdas Feuer ist gelöscht', vgl. at bo?k dun'i'ich habe das Feuer gelöscht'. Im Grunde genommen sind auch die Verbalformen wie in ket. qu?s' hapta 'das Birkenrindezelt stehtTdas Birkenrindezelt ist aufgestellt' als passiv zu betrachten, da man solch einen Zustand in der Regel immer nur als Resultat einer Handlung wahrnimmt, auch wenn es nicht so eindeutig wie in ket. kavaRoj'ts ist zerhauen' (nicht aber: 'es hat sich zerhauen') scheint. Eindeutig aktiv bleiben dagegen jene Verbalformen, die ihre Subjektaffixe (d-, k-, d-, da-) bei der Inkorporierung eines Nomens bewahren. Vgl. folgende ketische Verbalformen, in die bogdom 'Flinte' inkorporiert ist:

167

Zur Typologie der Passivkonstruktionen in den Jenissej-Sprachen (27)

'ich habe nun eine Flinte', wörtl. 'ich habe mich beflintet' 'du hast nun eine Flinte', wörtl. 'du hast dich beflintet' d-bogdom-bu-gdoja 'er hat nun eine Flinte', wörtl. 'er hat sich beflintet' da-bogdom-bu-gdoja 'sie hat nun eine Flinte', wörtl. 'sie hat sich beflintet' d- bogdom- 0Λ ij-dojan 'wir haben nun eine Flinte', wörtl. 'wir haben uns beflintet' k- bogdom-JG\ η-do jan 'ihr habt nun eine Flinte', wörtl. 'ihr habt euch beflintet' d- bogdombu-gdojan 'sie haben nun eine Flinte', wörtl. 'sie haben sich beflintet'. d-bogdom-bo-gdoja

k-

bogdom-ku-gdoja

Das Vorhandensein oder Fehlen in den entsprechenden Sätzen des Instruments bzw. des Agens spielt dabei keine Rolle und kann an und für sich nicht als Beweis für den passiven oder aktiven Charakter des Satzes betrachtet werden, vgl. (28) ket. ul' boyas' a: toRon 'das Wasser wurde durch das Feuer warm' (Aktiv) - ul' boyas' a:qqovil'arij 'das Wasser war mit dem Feuer erwärmt' (Passiv)8 . Das Erscheinen in den meisten Passivformen des a-Affixes der 3.P.Sg. aus der Reihe der Versionsaffixe in den Paradigmata der Subjektversion entspricht typologisch dem c h-Element in den entsprechenden russischen Konstruktionen: (29) ket. d-ayiv-raq 'ich erschieße ihn' > ayav-a-raq 'er ist erschossen', d-ab-bet 'ich wische es

ab' > a v-a- vet 'es ist abgewischt', bu is' da-Uta-yjt'ÚQ salzt die Fische' > rs' Uta-j-a-yit 'die Fische sind gesalzt', d-il'uks'i-vet 'ich zerbreche es1 > il'uks'a-j-a-vet 'es ist zerbrochen', d-ul'tab-daq 'ich binde es los' > ul'tav-a-raq 'es ist losgebunden'. Für den Status der passiven Konstruktionen in den Jenissej-Sprachen sprechen jedoch vor allem solche Fälle, in denen sich die aktiven und passiven Verbalformen mehr oder weniger deutlich gegenüberstehen: (30) ket. fin hal'bil'atn 'die Nadel bog sich' - fin hal'bilGut 'die Nadel war gebogen', qa?t aggipta 'der Mantel hängt' - qa?t arjgisut < aqgisqut 'der Mantel ist aufgehängt', bu ttoRojiqqol'ar'ij 'er trocknete sich ab' - bu toRojiqqil'Gut 'er war abgetrocknet' usw.

8

In beiden Sätzen tritt boyas' 'mit dem Feuer* (Instrumentalform von ixfik Verbalform a:jgqovil'arijdes zweiten Satzes kann als passiv gedeutet werden.

'Feuer1) auf, aber nur die

168

Heinrich Werner

Literatur Belimov, E. I. (1991) Ketskij sintaksis. Situacija, propozicija, predlozenie. Novosibirsk. Castren, Α. M. (1858) Versuch einer jenissei-ostfakischen und kottischen Sprachlehre nebst Wörterverzeichnissen aus den genannten Sprachen. St. Petersburg. Dul'zon, Α. P. (1968) Ketskij jazyk. Tomsk. Krejnovic, Ε. A. (1968) Glagol ketskogo jazyka. Leningrad. Meillet, A. (1903) Introduction à l'étude comparative des langues indo-européennes. Paris (niss. Übers.: Vvedenie ν sravnitel'noe izucenie indoevropejskich jazykov. Moskva/Leningrad 1938). Nedjalkov, V. P. (1981) Subjektnyj rezul'tativ i perfekt ν nemeckom jazyke (Stativ ot intranzitivov). Lingvisticeskie issledovanija 1981. Grammaticeskaja i leksiceskaja semantika. Moskva, 153-162. Nedjalkov, V. P. (Hrsg.) (1983) Tipologija rezul'tativnych konstrukcij. Leningrad. Polenova, G. T. (1986) Ketskij rezul'tativ. - Issledovanija po grammatike i leksike enisejskich jazykov. Novosibirsk, 31-38. Savcenko, Α. N. (1974) Sravnitel'naja gramma tika indoevropejskich jazykov. Moskva. Starostin, G. S. (1995) Morfologija kottskogo glagola i rekonstnikcija praenisejskoj glagol'noj sistemy. Ketskij Sbomik. Lingvistika. Moskva, 122-175. Uspenskij, Β. A. (1964) Zamecanija po tipologii ketskogo jazyka. - Voprosy struktury jazyka. Moskva, 144156. Uspenskij, Β. A. (1968) O sisteme ketskogo glagola - Ketskij sbomik. Lingvistika. Moskva, 196-228. Vali, Μ. Ν. / Kanakin, I. Α. (1988) Kategorii glagola ketskogo jazyka. Novosibirsk. Vail, M. N. / Kanakin, I. A. (1990) Ocerk fonologii i grammatiki ketskogo jazyka. Novosibirsk. Van Valin, Robert D. / Foley, William A. (1980) Role and Reference Grammar. In: E. Moravcsik (ed.) Syntax and Semantics, vol. 13, Current Aproaches to Syntax. New York: Academic Press. Werner, Heinrich (1995): Zur Typologie der Jenissej-Sprachen. Wiesbaden (= Veröffentlichungen der Societas Uralo-Altaica, Bd. 45).

Allocutivity and Voice in the Basque Verb Juan C. Moreno Cabrera

1. Introduction A demonstration of the interlinguistic relevance of the diathesis/voice distinction can be considered one of the major contributions of the Leningrad /Sankt Petersburg school of language typology (see V. P. Nedjalkov and V. P. Litvinov (1995) for an overview). Some of its most important publications explore the typological consequences of such a distinction (Xolodovic (ed.) (1974), Uspenskij (1977), Xrakovskij (ed.) (1978), Xrakovskij (1981), Xrakovskij (1991)). Moreover, the diathesis/voice opposition can also be employed to describe important aspects of the syntax of one language as is evident in the emblematic study of Chukchee diathesis by V. Nedjalkov (Nedjalkov (1976)). These concepts are characterized by Xrakovskij (1991:142-143) in the following terms. First, here is X's definition of diathesis. "The novelty of the Leningrad conception lies in the fact that for the first time the concept of diathesis is introduced in the scientific field as distinct from the concept of voice. The diathesis of a given verbal form in a given sentence can be defined as the correspondence between a set of elements of conceptual structure characterizing a lexeme appearing as a given verbal form and a set of elements of the syntactic structure, appearing around that verbal form"

This scholar characterizes grammatical voice in the following terms. "It is easy to proceed from the concept of diathesis to the concept of voice and to characterize it as the diathetical marking of the verb. In other words, we can speak of voice in cases in which in a language we find a sufficient number of verbal lexemes presenting diverse synthetic and analytic forms connected to certain diatheses"

Diathesis is, therefore, determined by the correspondences between the semantic and syntactic elements associated to a sentence. Voice is identified with the verbal enconding of diathesis. When different diatheses are conveyed by different analytic or synthetic verbal forms, voice distinctions are called for. All the work of the Leningrad /Sankt Petersburg school concerning this subject is devoted to detecting generalizations in the diathesis/voice co-variation. It is assumed that a diathetical change can entail a voice-change and viceversa. Languages differ in this respect. Consider for example the sentences in (1).

Juan C. Moreno Cabrera

170 (1)

Diathetical Variation in Spanish and English a) English The vase broke b) Spanish El jarrón se rompió ART vase MED broke 'The vase broke'

The diathesis of both sentences is exactly the same. The protagonist of the event referred to is the undergoer, not the agent or experiencer. This type of diathesis is sometimes called anticausative. The two languages differ only in voice. In English, the verb break is lexically assigned two alternative diathetical associations: subject / agent or subject /undergoer. In the given example the second has been selected. As a consequence, the verb shows up in the active voice, since this diathetical association is given as such in the lexicon. In Spanish, the subject of the corresponding verb romper, is not lexically assigned the undergoer semantic role (if it could, the expression *el jarrón rompió would be grammatical). So the required subject/ undergoer association must be syntactically introduced through a valencychanging rule; as a consequence, the verb is put in the middle se voice. In this paper I will provide evidence to demonstrate that diathesis and voice are in fact independent concepts. I will do so by showing how a language can present voice changes not conveying diathetical changes. This is indeed a very uncommon phenomenon, since as said before, languages abide in general by diverse diathesis/voice co-variations as investigated by the members of the Leningrad /Sankt Petersburg school. Nevertheless, the possibility of having in some special cases voice variations without a corresponding diathetical change, demonstrates that diathesis and voice are indeed independent concepts. The case of a diathetical change without a corresponding voice change is fairly common. Witness the following English examples. (2)

Diathetical variation in English a) The vase broke b) John broke the vase

It is very easy to see that the diatheses of both sentences differ. In the first sentence there is a correspondence subject/undergoer; in the second, a correspondence subject/agent and object/undergoer is evident. In spite of this, the verb in both sentences is in the active voice. This demonstrates that diathesis is an independent concept that cannot be identified with voice. In this paper a case of voice variation without diathetical change will be presented. This will prove that voice and diathesis are independent from each other.

171

Allocutivity and Voice in the Basque Verb

2. Voice distinctions in Basque Basque has a rich diathetical morphology. The majority of the Basque verbs are analytically conjugated using an auxiliar verb. The auxiliary conveys tense, person, number and in certain cases to be dicussed below, gender. There are two auxiliaries in Basque: the be and the have auxiliaries. Many verbs can be conjugated using both auxiliaries depending on their being transitive or intransitive. An example of each one of the possibilities follows. (3)

Example of a Basque intransitive sentence Jon etoni da John come is 'John has come'

(4)

Example of a Basque transitive sentence Jon-ek ogi-a jäten du John-ERG bread-ART eaten has 'John has eaten the bread'

The auxiliary agrees not only with the subject, but also with the direct and indirect objects. In example (3), the auxiliary da agrees with the only participant in the conveyed event. In this sentence type there is no place for a direct object, but an indirect object could be included; in that case, the verb shows an agreement morpheme for that new participant. An example of this possibility is given below. (5)

Basque Jon etorri za-it John come is-IPS 'John has come to me'

The verbal form da adopts the form zait when there is a first person indirect object in the sentence. Concerning the auxiliary du of sentence (4), we have a form conveying a third person object and a third person subject (d- is the marker for the third person object and the third person subject marker is in this case the zero morpheme; so this form analyzes as d-u-0). As in the preceding example, we could also introduce an indirect object in the corresponding sentence. In that case, the auxiliary presents an additional morpheme referring to the new participant. Let me illustrate this possibility with the following sentence. (6)

Basque Jon-ek liburu-a Andoni-ri John-ERG book-ART Anthony-DAT 'John has given the book to Anthony*

eman given

di-o has-3SG

Juan C. Moreno Cabrera

172

The forni dio of the auxiliary has an -o suffix referring to the indirect object of the sentence, and a zero affix for the subject / agent (this form should be therefore analyzed as di-o-0). Let me illustrate now how the alternance of the be and have auxiliaries can convey a difference in voice. Consider the following sentences: (7)

(8)

A Basque active sentece Jon-ek telesail-a ikusi John-ERG TV series-ART seen 'John has watched the TV series'

du has

A Basque passive sentence Telesail-a ikusi da TV series-ART seen is 'The TV series is watched'

In the first sentence there is a subject and an object and the auxiliary is ukan "have1 (the one used for transitive verbs). In the second sentence there is only one participant, the undergoer, and the auxiliary is the intransitive verb izan to be'. We could even introduce the agent in sentence (8) in the ergative case and obtain the following sentence: (9)

Basque Telesail-a Jon-ek ikusi-a da TV series-ART John-ERG seen-ART is 'The TV series has been watched by John'

There is a slight difference between this sentence and the preceding one. The first is a passiveimpersonal sentence and the second is a passive personal sentence. In the second sentence the participle ikusi is provided with the Basque article a. In his grammar of literary Navarro-Labourdin, Laffitte (1979: 434), cites sentences such as aitak igorria dia 'it is sent by his father1 in which igorria is the participle igorri 'sent' provided with the a article and aitak is the ergative case form for aita 'father1. As mentioned earlier, many Basque verbs can be conjugated with the two auxiliaries depending on their use as intransitive or transitive verbs. Let me illustrate this possibility with a pair of examples. (10) Basque a) Jon-ek seme-a etxe-ra-tu du John-ERG son-ART house-ADL-PART has 'John has led his son home' b) Jon etxe-ra-tu da John house-ADL-PART is 'John has gone home'

173

Allocutivity and Voice in the Basque Verb

In these examples we can see a transitive and an intransitive use of the verb etxeratu 'gone home'. In the first case we have the transitive use and Jonek is the agent and semea is the patient; in the second case, Jon is at the same time the agent and the patient. He is the patient because he undergoes the movement and he is also the agent since he controls that movement. The idea that the same participant can be assigned two different sematic roles has been proposed and developed by R. JackendofF (Jackendoff (1990)). From the point of view of case marking, what is important here is that John is the entity undergoing the movement denoted by the verb, and therefore is put in the absohitive or patient case (not in the ergative or agentive case). We find the same case in semea of the first sentence, which is also the entity undergoing the movement. We can think of the difference between the two auxiliaries as a diathetical difference, as shown in the following schema: (11) Auxiliary change as diathetical change a) Diathetical association for izan 'to be': ΡΑΊΊΕΝΤ/ΑΒ SOLUTIVE b) Diathetical associations for ukan 'to have': AGENT/ERGATIVE + PATIENT/AB SOLUTIVE Up to now, it has been shown that the alternance between the two auxiliaries conveys in Basque a diathetical difference. In fact, Lafitte (1979:194) says that intransitive verbs have a nominative and a dative voice depending on their reference to the subject or to the subject and indirect object. On the other hand, transitive verbs have an active and an active-dative voice depending on their reference to the agent and to the undergoer only (active form) or additionally to the recipient or beneficiary (active-dative voice). Let me illustrate Lafitte's distinction with the following table. AUXILIARIES INTRANSITIVE TRANSITIVE

NOMINATIVE / ACTIVE VOICE DA l e is' DU "he has it'

DATIVE / ACTIVE-DATIVE VOICE ZAIT Tie is to me' DIT "he has it to me'

Table 1. Voice in Basque following Lafitte (1979: 194)

3. The intimate second person in Basque In Basque, the pronoun hi 'thou' is only used between close relatives and intimate friends; as a personal pronoun, it has no gender distinction. But when the addressee of speech is referred to in a verbal form, the affixes -k and -n are used in this type of treatment; the first affix is used to refer to a male and the second is employed to refer to a female. Let me illustrate this point through the following examples.

174

Juan C. Moreno Cabrera

(12) Second person as goal a) Hi-ri eman di-k Thou-DAT given has-2PSM 'He has given it to you' b)

Hi-ri eman di-n Thou-DAT given has-2PSF Ήε has given it to you'

Lafitte (1979: 194) dubs these forms family voice and considers them as contextual variants of the voice forms included in table 1. It is important to note that if the hi pronoun is used, these k/n forms must be employed. These verbal forms constitute what is known in current Basque grammars as the allocutive conjugation.

4. Allocutive verbal forms In Basque Until now all the morphological variation we have seen has to do with diathetical changes; that is, it reflects different association-types between semantic and syntactic functions. We have seen earlier in this paper that it is possible for a diathetical variation not to be conveyed by a corresponding voice distinction. Recall the two uses of the English verb break illustrated in the two sentences of (2). If we pay attention to the translations of these sentences into Basque it will be evident that this diathetical difference is obligatorily conveyed in Basque by a change in the auxiliary; that is, a change in voice: (13) Basque a) Lihontzi-a apurtu vase-ART broken "The vase broke' b) Jon-ek liliontzi-a John-ERG vase-ART 'John broke the vase'

zen was apurtu broken

zuen had

In the first sentence, the auxiliary zen is the past form for da and in the second sentence, the auxiliary zuen is the past form for du. A form such as duk means "you have it' when we address our male partner with the hi second person intimate pronoun, as in the following sentence: (14) Basque Hi-k liburu-a irakurri you-ERG book-ART read You have read the book'

du-k has-2PSM

175

Allocutivity and Voice in the Basque Verb

If we are addressing a female, the dun form must be used: (15) Basque Hi-k liburu-a irakurrri you-ERG book-ART read You have read the book1

du-n has-2PSF

It is thus evident that duk/dun are transitive auxiliaries used when the association pairs SUBJECT/AGENT, OBJECT/PATIENT occur. Now consider the following intransitive sentence: (16) Basque Gizon bat etorri man one come Ά man has come'

da is

In this sentence, the intransitive auxiliary da is used and it conveys the single diathetical association SUB JECT/UNDERGOER-AGENT. But we must use the following versions of it when talking to a person we address using the intimate second prononun hi and is a male or a female respectively. See Euskaltzaindia (1987: 387)): (17) Two allocutive versions of the preceding intransitive sentence a) Gizon bat etorri du-k Man one come has-2PSM Ά man has come' (when addressing a man) b) Gizon bat etorri du-n Man one come has-2PSF Ά man has come' (when addressing a woman) It is very important to realize that these verbal forms are transitive auxiliaries (duk, dun) in euskera batua or standard Basque and must therefore be translated as "you (masc.) have it' and 'you (fem.) have it'. These auxiliaries, in their normal non-allocutive use are assigned the two diathetical pairs SUBJECT/AGENT and OBJECT/PATIENT. But the sentences in (17) do not convey this double diathetical association; instead, they must be related to a single diathetical association. So the second person referred to by the suffix (k/n) is not a participant in the event conveyed by the entire sentence; it is precisely the addressee of the statement made by that sentence. What is important here is that in order to refer to the addressee of speech, a transitive auxiliary has been chosen. Therefore, the da/duk variation is a case of voice variation without a corresponding diathetical variation, since as we have seen etorri da and etorri duk/n have exactly the same diathetical structures.

176

Juan C. Moreno Cabrera

Let me now consider a version of the above transitive sentences in which the agent is a third person noun. (18) Basque Jon-ek liburu-a John-ERG book-ART 'John has read the book'

irakurrri read

du has

If in these cases we are addressing a person to whom we refer as hi we must also employ an allocutive verbal form. The allocutive forms corresponding to du are dik for the masculine addressee and din for the feminine addressee: (19) Basque a) Jon-ek John-ERG 'John has read b) Jon-ek John-ERG 'John has read

liburu-a irakurrri di-k book-ART read have-2PSM the book' (when addressing a man) liburu-a irakurrri di-n book-ART read have-2PSF the book' (when addressing a woman)

Recall that dik and din are bitransitive auxiliaries conveying the triple diathetical association SUBJECT/AGENT, OBJECT/PATIENT and INDIRECT OBJECT/ GOAL. But in the preceding examples these forms are used in a transitive sentence in which there is no INDIRECT OBJECT/ GOAL diathetical association. So we get a voice variation without a corresponding diathetical variation if we compare the sentence of (18) with those of (19). Let me consider now a third person indirect object version of the sentences in ( 12). (20) Bitransitive sentence with a third person goal Koldo-ri eman di-o Koldo-DAT given has-3PS 'He has given it to Koldo' The form dio refers to the three participants in the action conveyed by this sentence. It analyzes as d-i-o-0, where d- refers to a third person singular participant as the patient, -o- refers to a third person singular participant as the goal and 0 refers to a third person singular agent participant; if this participant were plural the suffix -te would appear giving as a result the form diote. Even in these cases, an allocutive verbal form is called for in contexts in which the hi pronoun would be used. So, the preceding sentence has the following two allocutive variants.

177

Allocutivity and Voice in the Basque Verb

(21) Allocutive versions of a bitransitive clause a) Koldo-ri eman zi-o-k Koldo-DAT given has-3PS-2PSM 'He has given it to Koldo' (when addressing a man) b) Koldo-ri eman zi-o-n Koldo-DAT given had-3PS-2PSF 'He has given it to Koldo' (when addressing a woman) The auxiliaries ziok and zion are derived from dio by spirantizing the initial d- into z- and by adding the corresponding allocutive suffix -k/n. So a form like ziok refers to four persons: three of them are the participants in the conveyed event and the fourth is the addressee of speech. They analyze as zi-o-0-k and zi-o-0-n. We can have in certain cases a non-zero morpheme for each one of the four persons referred to in these allocutive forms. Consider the following sentence together with one of its allocutive variants. (22) Basque a) Ni-k Koldo-ri eman I-ERG Koldo-DAT given Ί have given it to Koldo' b) Ni-k Koldo-ri eman I-ERG Koldo-DAT given Ί have given it to Koldo' (when

d-i-o-t 3PS-had-3PS-lPS z-i-o-n-at 3PS-had-3PS-2PSF-lPS addressing a woman)

In these cases we get tritransitive verbal auxiliaries with no corresponding four-pair diathetical association. This shows that the morfological mechanism of voice in Basque is autonomous and can be employed even when no diathetical change has taken place.

5. Conclusion The theoretical distinction between diathesis and voice and the demonstration of its relevance to language typology is in my opinion one of the major contributions of the Leningrad/ Sankt Petersburg school of language typology. This distinction is not usually made in the current relevant literature since voice differences usually are accompanied by corresponding diathetical differences. An important step towards that distinction is the recognition that in many cases diathetical variation does not produce voice differences (as in John broke the vase and the vase broke). In this paper I have introduced a clear case in which a voice distinction does not cause a corresponding diathetical change. I think that what all this shows is that diathesis and voice must be carefully distinguished in typological research, as different ahhought intimately related phenomena.

178

Juan C. Moreno Cabrera

Abbreviations ADL = Adlative case morpheme, ART = Article, DAT = Dative case morpheme, ERG = Ergative case morpheme, MED = Middle voice clitic, PART = Participle suffix, IPS = First Person Singular, 2PSM = Second Person Singular Masculine, 2PSF = Second Person Singular Feminine, 3PS = Third Person Singular

References Euskaltzaindia (1987) Euskal Gramatika. Lehen Urratsak-11. Bilbo. JackendoS; R. (1990) Multiple Thematic Roles for a Single NP, R. Jackendoff (1980) Semantic Structures, Cambridge: The ΜΓΓ Press, 1990, 59-70. Lafitte, P. (1979) Grammaire Basque (Navarro-Labourdin Littéraire), Donostia: Elkar. Nedjalkov, V. P. (1976) Diathesen und Satzstruktur im Tschuktschischen, In: R. Lötzsch and R. Rû2i£ka (eds.) Satzstruktur und Genus Verbi, Berlin: Akademie Verlag, 181-211. Nedjalkov, V. P. and V. P. Litvinov (1995) The St Petersburg/Leningrad Typology Group, In: M. Shibatani and Th. Bynon (eds.) Approaches to Language Typology, London: Clarendon Press, 215-272. Uspenskij, V. A. (1977) Κ ponjatiju diatezy.[Concerning the concept diathesis], In: V. S. Xrakovskij (ed.) Problemy lingvisticeskoj tipologii i struktury jazyka. [Problems of linguistic typology and language structure], Leningrad: Nauka, 65-83. Xolodovid, A. A. (ed.) (1974) Tipologija passivnyx konstrukcij. Diatezy i Zalogi. [Typology of Passive Constructions. Diatheses and Voices], Leningrad: Nauka. Xrakovskij, V. S. (ed.) (1978) Problemy teorii grammatiâeskogo zaloga. [Theoretical Problems of grammatical voice], Leningrad: Nauka. Xrakovskij, V. S. (1981) Diateza i referentnost' (K voprosu o sootnoSenii aktivnyx, passivnyx, refleksivnyx ι reciproknyx konstrukcij). [Diathesis and Reference (concerning the question of the relationship between passive, reflexive and reciprocal constructions)], In: V. S. Xrakovskij (ed.) Zalogovye konstrukcii ν raznostrukturnyxjazykax [Voice constructions in languages of different structure], Leningrad: Nauka, 5-38. Xrakovskij, V. S. (1991) Passivnye konstrukcii [Passive Constructions], In: V. Bondarko (ed.) Teorija Fmkcional'noj Grammatiki. Personal'nost'. Zalogovost'. [The theory of functional Grammar. Person. Voice]. Sankt-Peterburg: Nauka, 141-180.

From Inert to Actional Causative Alexander K. Ogloblin

The present paper deals with the semantic history of causative in Malay and in closely related languages, particularly in Madurese. Causative affixes in these languages are polysemous. Besides more or less comprehensive descriptions of this polysemy, there are also reconstructions of the original form and meaning of some affixes in question. Here we shall suggest a possible explanation of how the causative meaning could emerge from a different semantic field. The main difference between causative and non-causative lies in the relationship of both kinds of derived verbs to the corresponding underlying verbs from which the former are derived. Both kinds of derived verbs take new participants (semantic actants). While the causative derivation implies the new agent (the causer), the non-causative one involves a certain less active participant, viz. location, instrument/medium, beneficiary, information (the contents of mental activity) or something else. Semantically, these latter may be quite new, that is not presupposed by the underlying verb, or they are just in a more subtle or intricate connection with the meaning of the derived verb. Syntactically, in languages in question the derived verbs of both kinds are transitive. In the active causative construction the new participant (the causing agent) takes the position of a subject, while in the non-causative one the new participant has the function of a direct object. For instance, the Malay verb duduk 'sit (down)' does not require any object, while the derived verbs -duduk-kan 'make sit (down)' and -duduk-i 'sit on (smth)' both require direct objects. In the causative construction this object corresponds to the underlying subject, while in the non-causative one it is the location of the state 'sit' which was irrelevant in the underlying construction. The latter function of the suffix may be labelled applicative, or deabsohitive. In Malay the suffix -kan is the most productive causative marker. Cf. (lab). (1)

a. Aku meng-gigil (Mihardja 1958) I ACT-tremble Ί trembled' b. Puput angin sejuk meng-gigil-kan badan-ku (Notosusanto 1958) gust wind cool ACT-shake-CAUS body-my Ά gust of cool wind made my body tremble'.

The prefix meN- is the marker of activity on the part of the subject, used in transitive and sometimes in intransitive sentences. It has variants mem-, men-, and others, depending on the first phoneme of the verbal base, replacing certain initial consonants of the latter. The same suffix can be also used as applicative. In (2ab) the "contents" of a question is the prepositional object with the basic and the direct object with the derived verb.

180

Alexander Κ. Ogloblin

(2)

a. Nabi Sulaiman ber-tanya kepada puteri itu tentang singgasana itu prophet Salomo ask to princess that about throne that 'Prophet Salomo asked the princess about that throne'. (Usman 1954) b. Beliau me-nanya-kan daku (Toer 1959) he ACT-ask-APPL me 'He asked about me'.

The prefix ber- has a purely derivational function without any definite semantic function. It is dropped when other derivational process apply. Considering verbs with an "abstract directional object", T.Givón (1984: 112) remarks: "In many instances, the object is clearly the cause of the subject's mental state", e.g. in Engl. "She mourned his death". The meaning of something producing a causing effect by its mere existence may be labelled inert causative, as opposed to the actional causative which implies a controlled exertion by the causing agent. In Malay we have similar examples with the applicative suffix -kan : (3)

Orangi tua2-ku me-rindu-kan abang-ku parentsi^-my ACT-long-APPL elder.brother-my 'My parents are longing for my elder brother1.

Such sentences have been interpreted by some Indonesian scholars as having an "inert" causative sense: "my elder brother makes my parents to be longing" (Munaf 1951, vol.1: 207; Lubis 1952:94). Historically, the suffix -kan is not an actional causative. In Old Malay inscriptions from the 7th century it has the form -akan which coincides with the preposition akan 'to, on, towards' still used in contemporary, but mostly characteristic for medieval Malay and denoting direction of an action to patient (goal).1 The emergence of actional meaning may be explained by a switch in interpretation from inert causative. The latter may have had other meanings than that of "mental state" in the earlier stages of language evolution. It could encompass also instruments and means used in action, addressees, beneficiaries and locations. Cf. the interpretation of a Sundanese sentence by a native speaker and a grammarian Ardiwinata (1984: 82) early in this century: "the pupil is the cause of teacher's action" in (4): (4)

Guru nga-deukeut-an ka murid teacher ACT-near-APPL to pupil 'The teacher approaches the pupil'.

1 It may have been phonologically aken (with second schwa vowel). See the discussion in Adelaar 1985 and 1994.

From Inert to Actional Causative

181

(The prefix nga- is a parallel to Malay meN-.) Such interpretation may seem odd and forced. However, there is some evidence in favor of a broader conception of inert causative. In Madurese it has a distinct morphological expression (although its marker is also polysemous). Inert causative verbs in this language are derived by adding the prefix ka- to the verbal root and usually appear in passive sentences. The passive marker is prefix e-:2 (5)

a. Fatimah sake' Fatimah ill 'Fatimah is ill'. b. Hal jareya e-ka-sake' bi' Fatimah thing this PASS-ICAUS-ill by Fatimah 'Because of this Fatimah became ill'.

As common in the passive sentences of such structure, 'Fatimah' in (Sb) is the agent whose action is, however, rather unusual: the agent lets or allows him/herself to be involved in a mental ot physiological state by some cause. The action may be also physical as in (6b): (6)

a. Fatimah nanges (=N-tanges; N- is equivalent to Malay meN-) Fatimah ACT-cry 'Fatimah cries'. b. En-maenan se pera' cara jareya e-ka-tanges bi' Fatimah toy REL only kind this PASS-ICAUS-cry by Fatimah 'Because of such an (insignificant) toy Fatimah cries'.

The passive construction exemplified by (5b-6b) may be historically early. It has a counterpart in Tagalog, which is not too closely related to Madurese genetically, cf. (7): (7)

I-ka-ga-galak ko ang iyo-ng anyaya PASS-ICAUS-CONT-joy by.me DEF your invitation Ί was glad because of your invitation'.

(Note: reduplication of the first syllable of the root marks continuous aspect; ng is a ligature to the following head noun.) This example from Rackov 1981: 141 is quoted by the author for Tagalog causal passive (there are also other semantic types of passive, e.g. instrumental or locative). Cf (Wolff et al. 1991: 547, quoted in Skarban 1995: 137). This correlation between Madurese and Tagalog may indicate relatively ancient origin of the inert causative marker ka-.

2

The Madurese examples are mostly from Muntaha/Ogloblin 1984 and Ogloblin 1986. Cf. Stevens 1968.

182

Alexander Κ. Ogloblin

In Madurese, however, some faz-verbs are synonymous with verbs built with transitive suffixes -e and -aghi denoting the relation of the action to patients and other non-active participants (the suffix -aghi functionally corresponds to Malay -kan). Cf.(8-9): (8)

a. Bini-na se mate e-ka-pekker bi' Ah wife-his REL dead PASS-ICAUS-think by Ah 'His deceased wife is being thought about by Ali'. b. Ta'-kanta' badha se e-pekker-e as.if be REL PASS-think-APPL 'As if there was something thought about [by her]'.

(9)

a. Jareya la... segghut e-ka-careta this already often PASS-ICAUS-tell 'This was often told [by him] to his friends'.

da' ca-kanca-na to friends-his

b. Neghara jareya ella e-careta-aghi bi' Hosen ka sengko' state this already PASS-tell-APPL by Hosen to I 'About this country was already told by Hosen to me'. The prefix ka- accounts for interpretation of the contents of what is thought or told about as a cause of actions 'think' resp. 'tell'. The suffixes -e and -aghi, on the other hand, indicate a correlation of this 'information goal' with other sorts of referentially non-identical non-causative ("applicative") participants. As a typological parallel one might remember the "causative and telling" case of nouns in Itelmen (Volodin 1976: 151,155). Indeed, there are some grounds for assuming the inert causation to be a rather broad semantic unit in some languages, maybe, also reflecting an archaic mode of categorisation of extra-linguistic data.3 However, this does not rule out the existence of an actional causative at the early stage of evolution. A great many of Austronesian languages have reflexes of a causative proto-prefix reconstructed as *pa- which has the same form in Madurese pa-, e.g. -pa-toron 'lower" from toron 'descend' (cf. Dahl 1977: 119, also Brandstetter 1916: 174-175). In Malay the causative prefix per- produces regularly derivatives from adjectives, e.g. -per-besar 'enlarge' from besar 'large'. The development of the suffix -kan from inert to actional causative results in a competition with the older marker of actional causative prefix per-. The same trend can be observed in Madurese, where the suffix -aghi in some functions is synonymous with the prefix pa-, cf. toron-naghi 'lower1 besides -pa-toron. In Javanese the prefixation in the verbal base, which was rather productive in Old Javanese, later was completely replaced by suffixation. Thus, we can observe formal evolution alongside semantic change, both of which result in the elimina-

3

The reverse historical movement, i.e. from actional to inert causative, can also be traced in Malayo-Javanic languages.

From Inert to Actional Causative

183

tion of inert causative as a separate grammatical category and its merger with other "applicative" meanings.

References Adelaar, K.A. (1994J Bahasa Melayik Purba. Rekonstruksi Fonologi dan Sebagian dari Leksikon dan Morfologi. Jakarta. Adelaar, K.A. (1985) Proto-Malayic. The reconstruction of its phonology and parts of its lexicon and morphology. Alblasserdam. Ardiwinata, D.K. (1984) Tata bahasa Sunda. Jakarta. Brandstetter, R. (1916) An introduction to Indonesian linguistics. London. Dahl, O.Ch. (1977) Proto-Austronesian. Lund. Givón, T. (1984) Syntax. A functional-typological Introduction. Vol.1. Amsterdam. Lubis, M. (1952) Paramasastera landjut. Amsterdam/Djakarta. Mihardja A.K. Atheis. Roman. Djakarta, 1958. Munaf, H. (1951) Tatabahasa Indonesia. Djakarta. Muntaha, S. and Ogloblin, A.K. (1984) Polisemija kauziruju&ego ob'ekta (madurskij glagol'nyj prefiks ka-). In Vostokovedenie 10. Leningrad (University Press). Notosusanto N. Hudjan kepagian. Djakarta, 1958. Ogloblin, A.K. (1986) Madurskij jazyk i lingvistiâeskaja tipologija. Leningrad. RaCkov, G.Ye. (1981) Vvedenie ν morfologiju sovremennogo tagal'skogo jazyka. Leningrad. Skarban, L.I. (1995) Grommati ieskij stroj tagal'skogo jazyka. Moscow. Stevens, A.M. (1968) Madurese phonology and morphology. New Haven. Toer P.A. Bukan pasar malam. Djakarta, 1959. Usman Z. Kesusasteraan lama Indonesia. Djakarta, 1954. Volodin, A.P. (1976) Itel'menskij jazyk. Leningrad. Wölfl; J.U., Senteno, M.T.C., Rau, D.-H.V. 0991) Philipino through Self-Instruction. Ithaca

Abbreviations ACT - active meaning, in transitives also active voice APPL - applicative DEF - article CAUS - actional causative ICAUS - inert causative PAS - passive voice REL- relative marker

Towards a Typology of Grammatical Polysemy: Reflexive Markers as Markers of Reciprocity1 Jurij P. Knjazev

1. Introduction It seems to be generally accepted now that polysemy should be treated as a necessary property of human language, not only in the lexical domain, but also in grammar. At the same time as it has been shown, among others, in a number of V.P.NedjalkoVs papers concerning typology of causatives, reflexives and reciprocals2, the scope of divergent senses of a grammatical marker tends to remain within limited semantic boundaries. The reflexive/reciprocal polysemy is a well-known and widely attested phenomenon of this sort; cf German3 : (1)

Sie kennen sich "They know themselves' (reflexive) or 'They know each other* (reciprocal)

A significant amount of available data provides sufficient (though, to be sure, by no means exhaustive) grounds for possible generalizations. Besides, it seems reasonable to take a somewhat wider range of linguistic facts than mere polysemy into conáderation. For the purposes of this study, a reciprocal construction is regarded as containing two main components: a reciprocal marker and a verbal stem Facts from a large number of languages show that, when a reflexive marker is used to express the reciprocal sense, both the marker and the stem may either remain unchanged or undergo some alteration. Leaving aside the trivial case of absence of a common part either in markers or in stems, there are four types of logically possible combinations of formal relations between reflexive and reciprocal constructions4 : 1) formal identity both of the markers and stems; 2) formal alteration of the markers accompanied by formal identity of stems; 3) formal identity of the markers accompanied by formal alteration of stems; 4) formal alteration both of the markers and stems. All these four types are attested in natural languages, though with different frequency. 1

The present study is the updated version of my paper presented at the Conference on linguistic typology (Moscow, 1990). It appeared in a shortened version in Knjazev (1990).

2

See, for instance, Kedjalkov (1978, 1980, 1991) and Geniusiené, Nedjalkov (1991).

3

Examples without attribution are either mine or taken from Geniusiené (1983, 1985) and the papers mentioned in footnote 2. Translations of examples cited from the papers, which are listed in the References, are by their authors.

4

This classification is partly influenced by the Mel'Cuk's (1973) calculus of possible formal semantic relations between linguistic signs.

186

Jurij P. Knjazev

2. A tentative summarizing of linguistic data 2.1. Type 1 Formal identity of the markers accompanied by formal identity of stems can be illustrated by cases of the reflexive/reciprocal polysemy such as example (1). Before examining the details of this marking pattern, it is worth noting that although this type of reflexive/reciprocal polysemy seems to be the most widespread one across the languages of the world, pure instances of this kind are not found very frequently because of their semantic vagueness. There may be two marking strategies here that are not mutually exclusive. First, in the vast majority of languages there are unambiguous lexical items rendering a reciprocal meaning similar either to the English each other (e.g. German einander, French l'un l'autre, Russian drug gruga) or to the Czech navzájem 'reciprocally, mutually. They may cooccur (though not always freely) with ambiguous reflexive/reciprocal markers in case it is necessary to emphasize reciprocal semantics. Secondly, it often turns out that sets of verbs with the polysémie marker referring to reciprocal or reflexive situations are very limited in number and may have no (or almost no) common items. Consequently, the polysemy of a marker does not necessarily entail the polysemy of verbs or sentences with this marker. This holds for Russian verbs with the postfix -sja (-s' after vowel stems). In its true reflexive use, this marker appears mostly with verbs indicating body care actions of the type myt'sja Svash oneself or odevat'sja 'dress oneself. On the other hand, Russian verbs in -sja may indicate reciprocal situations as well, but with a very small class of verbs denoting either affectionate actions such as obnimat'sja 'embrace each other1, celovat'sja tiss each other1 or antagonistic actions undesirable to humans such as tolkat'sja 'push each other1, rugat'sja 'abuse each other*. There seems to be only one point of contact between the two underlying verbal classes, namely, the verb oblivat' 'pour over1 referring to an event that may be treated either as an unfavourable action or as a body care action. Consequently, its reflexive counterpart oblivat'sja, as an exception, combines both the reciprocal and the reflexive meanings. The problem of actual lexical scope of the reflexive/reciprocal polysemy in a given language lies, however, beyond the goals of this paper and is disregarded in what follows. Given these reservations, I will turn to illustrations of Type 1, with some attention to genetic and areal affiliation of languages. 2.1.1. Type 1 of formal relations between reflexives and reciprocals occurs in many IndoEuropean languages. It can be exemplified, for instance, by Latin verbal forms in -r, Ancient Greek medial forms and their successors in modern Greek, forms of the reflexive passive voice in Albanian, verbs in -v- in Armenian, a set of etymologically related reflexive pronouns, particles and affixes in *s- in Slavic, Baltic, Romance and Germanic (excepting English) languages such as Russian -sja, Lithuanian -si, French se, German sich, Icelandic -sk. South and West Slavic languages (with the exception of Serbian, Croatian and Polish) differ from many other languages in that they distinguish between a direct (accusative) and an

Towards a Typology of Grammatical Polysemy: Reflexive Markers as Markers ofReciprocity

187

indirect (dative) forms of the clitic ("light"3) reflexive / reciprocal marker depending on whether the subject is co-referential with a direct object as in Czech milovat se love oneself / each other1 or with an indirect object as in Czech fikat si 'speak to oneself/to each other1). Besides, in the West Slavic languages (including Polish), the autonomous ("heavy") reflexive marker can have reciprocal semantics as well. It is used to mark co-reference with a prepositional object or (in Polish) with an indirect object; cf

(2)

Czech Sédeli jsme proti sobé "We were sitting opposite each other1

(3)

Polish X i Y piszq do siebie 'X and Y write to each other"

(4)

X i Y daj3 sobie ksiqzki 'X i Y give books to each other*

In other Indo-European languages, one marker is used for all these varieties; cf

(5)

French a. Ils s'aiment 'They love each other" b. Ils se disent tout "They tell each other everything'

2.1.2. The same Type 1 is also attested in many Finno-Ugric languages. The reflexive/reciprocal polysemy is characteristic of the suffixes -s'/-z' in Komi, -xat-l-axt- in Mansi (Vogul), -alt in Mari, -s'k in Udmurt; cf.

(6)

Mansi (Rombandeeva 1973: 148-152) lowtunkwe "wash' low-hat-unkwe "wash oneself tittunkwe 'feed' titt-hat-unkwe "feed each other*

In Hungarian, this type of polysemy is shared by some affixes with the consonant components -d- and -z- (Majtinskaja 1959: 106-107); cf.

5

The terms "light" and "heavy" were introduced in (Haiman 1983) to distinguish between concurrent grammatical markers with regard to their relative physical length and degree of boundedness to neighboring morphemes.

Jury P. Knjazev

188 (7)

a. ruházni'dress' verni "beat' b. oheni 'dress' csokolni 'kiss' c. borotválni'shave'ölelni 'embrace' -

ruház-kod-ni 'dress oneself vere-ked-ni 'fight' ok-oz-ni 'dress oneself csokol-oz-ni "kiss each other* borotvál-koz-ni 'shave oneself ölel-kez-ni 'embrace each other1

In the Mordvin, Finnish and Estonian, however, the marking patterns of reflexives do not coincide with those of reciprocals. 2.1.3. As for languages of the Afro-Asian region, formal coincidence of reflexives and reciprocals may be illustrated by the Tamili reflexive/reciprocal expressions consisting of the verb kol take' and a converb of the main verb (Rudin 1975). It also occurs in some Austronesian languages; cf.

(8)

Tagalog (Krus, Skarban 1966: 54-55) d-um-amit 'dress' mag-damit 'dress oneself b-um-ati 'greet mag-bati 'greet each other"

(9)

Bahasa Indonesia (Ogloblin 1981: 136, 143) -cukur 'shave' ber-cukur 'shave oneself -pehikan 'embrace' ber-pehikan 'embrace each other*

A similar relation between reflexives and reciprocals is also found in certain Nilo-Sah aran languages, e.g. in Kanuri (Kemmer 1988: 144) and Logo (Andersen, Goyvaerts 1986: 299-300) as well as in a few Bantu languages, mostly in the northern frontier area of their spreading; cf. Bolia verbs in -ka- (Aksenova, Toporova 1990: 176, 181): (10) -ka-liaka "kill oneself -ka-amba love each other* 2.1.4. Finally, this marking pattern is used in many aboriginal languages of Australia, Polynesia and America. According to B.Blake (1987: 57), "it is common in Australian languages for derived intransitives to express the notions of reflexive and reciprocal. Usually the same marker covers both notions". A combination of the reflexive and the reciprocal senses is characteristic of he-lse in Klamath, in- in Jebero, -at- in Apalai, naa- in Hopi, na- in Southern Paiute, -si in Aymara, -wisl-ms in Yokuts; cf. Hixkaryana (Derbyshire 1979: 62) (11) oso-xnatxhe self-wish be.3p.NonPast "They love themselves/each other1

Towards a Typology of Grammatical Polysemy: Reflexive Markers as Markers ofReciprocity

189

In West Groenlandie Eskimo both these meanings are indicated by the allative case of the reflexive/reciprocal pronoun immi- together with the replacement of transitive subject-object agreement by phiral intransitive (subject-only) agreement on the verb (Fortescue 1984: 160164); cf. (12) immi-η -nut tuqup-put self -PL-ALL kill -3pIND 'They killed themselves/each other1 This detransitivizmg process by itself may (without immi-) produce reflexives and reciprocals (as well as anticausatives and antipassives). A somewhat similar situation occurs in English where some unmarked intransitive verbs of certain restricted semantic classes may express either reciprocal situations, as in They kissed, or reflexive situations, as in He shaved. 2.2. Type 2 Formal alteration of the markers accompanied by formal identity of stems has two main variants: reduplication of a reflexive marker and its insertion into a reciprocal marker6. 2.2.1. The first subtype can be exemplified, for instance, by the reflexive and reciprocal markers in Korafe (New Guinea), where the reciprocal marker tofo tofo is a reduplicated form of the singular reflexive marker tofo, the tatter's partial reduplication functioning as the plural reflexive marker totofo (Farr, Fan 1975: 738). In Twi, as S.Kemmer (1988: 138-139) mentions, the reciprocal construction [pronoun]+/io is formed by reduplication of the reflexive marker ho (ht. "body"). Some of the ambiguous reflexive/reciprocal markers mentioned in 2.1.4. may be reduplicated as well. Thus, in Southern Paiute, the reduplicated form nana- of the marker na- "is frequently employed where emphasis on reciprocity, as distinct from reflexive activity, is desired" (Sapir 1931: 109). Similarly, in Apalai, reduplication of the préfixai ambiguous marker -at"always results in reciprocal meaning" (Koehn, Koehn 1986: 43). 2.2.2. Formal insertion of a reflexive marker into a reciprocal marker occurs in pairs of the type -ep- and -epew- in Yurok (Robins 1958: 74), -ku- and -naku- in Quechua (Weber 1989: 169-170), -ri and -dbari in Thargari (Dixon 1972: 93). The origin of these pairs is sometimes easy to trace. As it is pointed out in Longacker (1977: 69-70), the Tubatulabal (a Uto-Aztecan language) reciprocal pronoun ?omohic, related to the reflexive pronoun ?omohi, originally consisted of the latter and the postposition -c, which may have meant something like 'between' or 'across' but was eventually lost as an independent postposition. 6

One may, probably, assume the existence of the third subtype, which would cover pairs of reflexive and reciprocal markers such as Guarani ye and yo or Kabardian zy and ze. These markers have a part in common, but they also have distinctive components. This case is left aside because it is hardly possible to reach a definitive decision as to whether these markers are really morphologically related to each other or not.

Jurij P. Knjazev

190

Another example is provided by Russian verbs of joining and separation derived from intransitive verbs with a complex marker (circumfix), comprised of the prefixes s- or raz-/ras- and the reflexive postfix -sja. This complex marker, which (in its primary meaning) is used for collective situations like (13a), may also appear in verbs of reciprocal semantics such as (13b): (13) a. polzti'crawl' b. zvonit' 'ring'

-

spolztis"crawl together from different points' raspolztis1 'crawl from one point in many directions' sozvonit'sja 'get in touch over the telephone'

We may assume that the reflexive component of these complex markers conveys the idea of collective or reciprocal action while the prefixes (in accordance with their concrete spatial meanings) determine the "direction" of an action. 2.3. Type 3 Formal identity of the markers accompanied by formal alteration of stems has two variants: reduplication of a stem and partial overlapping of stems. 2.3.1. Reduplicating the root is a condition of reciprocal reading of the polysémie reflexive/reciprocal markers ju- in Urubu-Kaapor (Kakumasu 1986: 320-321) and -bariy in Dyirbal (Dixon 1972: 92-93) where verb reduplication involves repeating the first two syllables. In Bahasa Indonesia, the ambiguous prefix ber- (see 2.1.3.), being used in the reciprocal meaning, is often attached to a reduplicated root (Ogloblin 1981: 143); cf. (14) -kasih'love' -kejal 'chase'

-

berkasih-kasihan'love each other" kejal-berkejan 'chase each other"

In Guugu Yimidhirr, root reduplication, which is used optionally in order to overcome semantic ambiguity of the reflexive/reciprocal marker -dhi (Dixon 1980: 448), is in this respect functionally quite parallel to reduplication of ambiguous markers in some other languages (see 2.2.1. above). It is noteworthy that stem reduplication by itself without any additional markers, may indicate reciprocity; e.g. in Maori (Krupa 1967: 36) and Amele (Roberts 1987:132). 2.3.2. Another sort of stem modification is characteristic of the South Semitic languages. Thus, in Tigre, there is a tendency to attach the ambiguous marker tä- to distinct verbal themes (Forms) depending on whether it is used in the reflexive or in the reciprocal meaning. To form a reflexive, it combines with the basic Form, whereas the reciprocal meaning is signalled by its combination either with Form Π (often labelled 'intensive') marked by geminating the second radical of the root or with Form ΙΠ (labelled 'conative') marked by lengthening the first vowel of the root (Gankin 1982: 16-18). A similar situation occurs in Arabic, the difference being that reflexive and reciprocal meanings are marked with combinations of ta- with Forms Π and III respectively (Grande 1972: 220-221).

Towards a Typology of Grammatical Polysemy: Reflexive Markers as Markers ofReciprocity

191

2.4. Type 4 An example of formal alteration of both the markers and stems may be seen in Russian multiplicative reflexive reciprocals like perekidyvat'sja 'throw sth to each other (repeatedly)' derived from the verb kidat' 'throw* or peregljadyvat'sja 'glance at each other (repeatedly)' derived from gljadet' 'look'. They are formed by means of a complex marker composed, apart from the reflexive postfix -sja, of the prefix pere- and the imperfective-iterative suffix -va-. 7 These verbs designate events consisting of an indefinite number of uniform (micro)actions associated with (at least) two participants, each playing two roles in the event: each participant is a subject of the (mirco)action and also its adressee (Knjazev 1989). The suffix -va- is, in principle, a replaceable constituent part of this complex formant and is often interchangeable with the semelfactive suffix -ra/-; cf (15) peremigivat'sja 'wink at one another (repeatedly)' peremignut'sja 'wink at one another (once)' For this reason, derivation of these verbs may be seen as a two-stage process: the combination pere...sja contributes the reciprocal component of their meaning whereas the suffix (-να- or nu-) adds either the multiplicative or semelfactive aspectual component.

3. Conclusion Having considered the evidence above, we may conclude that this type of polysemy, though widespread, is distributed among the world's languages rather unevenly. Most languages sharing this property (excepting the Amerindian) form a kind of a wide "stripe" which extends from the shore of continental Europe in the south-east direction via the Middle East and South India to the islands of South East Asia, Australia and Polynesia. It should be noted that languages within this "stripe" are far from being uniform. Thus, we could draw an imaginary line from Czech through Lithuanian and Latvian to Russian indicating the gradual increase of restrictions on the derivation of reflexive reciprocals: Russian reciprocals in -sja of Type 1 (see 2.1.) do not number more than two dozens, Latvian and Lithuanian, according to Geniusienè (1987: 80, 89-93), possess about 80 and 160 reflexive reciprocals respectively, whereas in Czech there seem to be no appreciable restrictions on the reciprocal use of the reflexive marker se* It is remarkable that Czech and Russian, which form the extremes of this line, belong to the same Slavic group of languages.

7

The suffix -va- is used either to imperfectivize perfective verbs such as dat' - davat' 'give' or to derive iterative verbs such as byt' 'be' - byvat' 'be (habitually)'. It occurs, besides, in some types of complex verbal derivational formants (see Knjazev 1989).

8

It should be added that, as regards the relative frequency of the true reflexive uses of the reflexive markers, these four languages are aligned along the same scale.

192

Jurij P. Knjazev

As to the semantic links between reciprocity and the scope of meanings subsumed under a reflexive marker, semantic contiguity of the reciprocal meaning and the true reflexive meaning seems to be the most evident. In both instances, "the set designated by the subject is the same as (co-referencial to) the set designated by the object" (Langacker 1977: 69), the distinction being that the reciprocal relation pairs different individuals, while in the reflexive relation the two roles are assigned to the same entity. Russian demonstrates another possibility, namely, the direct semantic relationship between reciprocals and so-called "absolutive" ("object-deletion" or "anti-passive") uses of reflexives referring to situations with two participants, agent and patient, the latter being, however, only implied. It appears that almost all Russian reflexive reciprocals allow the absolutive use as well;9 cf. (16) a. Posmotri, dve korovy bodajutsja 'Look, two cows are butting each other" (reciprocal) b. Bud' ostorozen, korovy bodajutsja 'Be careful, cows butt' (absolutive) The patient in absolutive uses of type (16b) is understood as generic or indefinite and, in any case, as being pragmatically non-prominent. Semantic continguity of reciprocal and absolutive uses seems to be due to the following fact: if the patients of an action are of no importance to its agents (as it is in the case of the "absolutive" use), it may easily happen that the set of patients is the same as the set of agents (as it is with the reciprocal proper). Finally, it turns out that the reciprocal constructions, which are morphologically related to the reflexive constructions, are never (or, at least, far less often) "lighter" then the latter; they are either "heavier" or the same in degree of complexity. The reason for this may be sought in iconic reflection of the fact that the reciprocals are semantically more complex than the reflexives.

9

As was mentioned in section 2.1., the lexical reflexive/reciprocal polysemy is practically non-existent in Russian; see for minor details (Knjazev, Nedjalkov 1985).

Towards a Typology of Grammatical Polysemy: Reflexive Markers as Markers ofReciprocity

193

References Aksenova, I.S., and I.N. Toporova (1990) Vvedenie ν bantuistiku. Moscow: Nauka. Andersen, T., and D.L. Goyvaerts (1986) Reflexivity and logofbricity in Moru-Madi, Folia Linguistica, XX, 3-4. Blake, B. (1987) Australian aboriginal grammar. London etc.: Croom Helm. Derbyshire, D.C. (1979) Hixkaryana. Amsterdam: North Holland. (Lingua descriptive studies, 1). Dixon, R.M.W. (1972) TheDyirbal language of North Queensland. Cambridge. - (1980) The languages of A ustralia. Cambridge. Farr, J., and C. Farr (1975) Some features of Korafe morphology. Studies in languages of Central and SouthEastem Papua. Canberra, 731-769. (Pacific linguistics. Series C. Ν 29). Fortescue, M. (1984) West Greenlandic. Beckenham, Kent: Croom Helm. Gankin, L.E. (1982) Glagol'naja sistema tigrajskogo jazyka. Moscow. Geniusiené, E. (1983) Refleksivnye glagoly ν baltijskix jazykax i tipologija refleksivov. Vilnius. - (1985) The typology of reflexives. Berlin etc.: Mouton de Gruyter. Geniusiené, E., and V.P. Nedjalkov (1991) Tipologija refleksivnyx konstrukcij, Teorija fmkcional'noj grammatiki. Personal'nost'. Zalogovost'. Saint-Petersburg: Nauka, 241-276. Grande, E.M. (1972) Vvedenie ν sravnitel'noe izuóenie semitskix jazykov. Moscow: Nauka. Haiman, J. (1983) Iconic and economic motivation, Language, 59, 781-819. Kakumasu J. (1986) Urubu-Kaapor, Handbook of Amazonian languages, v.l. Berlin etc. Kemmer, S. E. (1988) The middle voice: a typological and diachronic study. Doctoral dissertation. Department of Linguistics. Stanford University. Knjazev, Ju. P. (1989) Vyraienie povtorjaemosti dejstvija ν russkom i drugix slavjanskix jazykax, Tipologija iterativnyx konstrukcij. Leningrad: Nauka, 132-145. - (1990) Κ tipologii reciproka: sovpadenie pokazatelej vzaimnosti i vozvralnosti, Vsesojuznaja konferencija po lingvistiieskoj tipologii (tezisy dokladov). Moscow: Nauka, 71-72.. Knjazev, Ju. P., and V.P. Nedjalkov (1985) Refleksivnye glagoly ν slavjanskix jazykax, Refleksivnye glagoly ν indoevropejskixjazykax. Kalinin, 20-37. Koehn, E., and S. Koehn (1986) Apalai, Handbook of Amazonian languages, v. 1. Berlin etc. Krus M., and L.I. Skarban (1966) Tagal'skij jazyk. Moscow: Nauka. Krupa, V. (1967) Jazyk maori. Moscow: Nauka. Langacker, R.W. (1977) Syntactic Reanalysis, Mechanisms of syntactic change. Austin, London: Univ. of Texas Press, 58-139. Majtinskaja, K.E. (1959) Vengerskij jazyk, v. 2. Moscow: Nauka. Mel'Cuk, I. (1973) The structure of lingustic signs and formal-semantic relations between them. Recherches sur les systèmes signifiants. The Hague, Paris: Mouton, 103-135. Nedjalkov, V. P. (1978) Zametki po tipologii refleksivnyx deagentivnyx konstrukcij, Problemy teorii grammatiôeskogo zaloga. Leningrad: Nauka, 28-37. - (1980) Reflexive constructions: a functional typology, Wege zur Universalienforschung. Tübingen: Narr, 222-228 (Tübingen Beiträge zur Linguistik, 145). - (1991) Tipologija vzaimnyx konstrukcij, Teorija funkcional'noj grammatiki. Personal'nost. Zalogovost'. Saint-Petersburg: Nauka, 276-312. Ogloblin, A.K. (1981) Refleksiv i reciprok ν indonezijskom jazyke, Zalogovye konstrukcii ν raznostruktumyx jazykax. Leningrad: Nauka, 130-146. Roberts, J.R. (1987) Amele. London et al.: Croom Helm. Robins, R.H. (1958) TheJurok language. Grammar. Texts. Lexicon. Berkeley, Los Angeles. Rombandeeva, E.I. (1973) Mansijskij (vogul'skij) jazyk. Moscow: Nauka. Rudin, S.G. (1975) Specializirovannye glagol'nye soietanija tamil'skogo jazyka, Voprosy jazykoznanija, 4, 6476. Sapir, E. (1931) Southern Paiute, a Shohonean language. Los Angeles (Proceedings of the American academy of art and sciences. Vol. 65, Ν 1). Weber, D. J. ( 1989) A grammar of Huallaga (Huanuko) Quechua. Berkeley etc. : Univ. of California Press.

Verb Formation in Leko: Causatives, Reflexives and Reciprocals Simon van de Kerke

1. Introduction This paper presents a first view of verbal derivation in Leko, a virtually unknown language spoken in the Andean eastern slopes to the north of La Paz, Bolivia.1 The major source on the Leko language, apart from the small word lists in Lázaro de Ribera (Palau & Saíz 1989), Wedeil (1853), Cardús (1886), and Montaño Aragón (1987), is a Christian doctrine created by a father Andrés Herrero in the beginning of the nineteenth century, published in Lafone Quevedo (1905). In Grimes (1988) the Leko language, classified as an isolate 2 , was reported to be dead, but Montaño Aragón (1987) reported a number of speakers in the region of Atén and Apolo in the province of Franz Tamayo and along the river Mapiri in the province of Larecaja. Responding to an appeal in Adelaar (1991) to enquire the possibility that speakers of the Leko language might still survive, in 1994 a fieldwork trip to Bolivia was undertaken. A thorough search in the region of Atén and Apolo was in vain, but I encountered some elder men and women along the Mapiri river, who still spoke the Leko language. Subsequent visits in 1995 and 1996 produced enough language data to check and enlarge the first grammatical sketch of the language by Lafone Quevedo (1905). However, gathering the data proceeded slowly since the informants, most of them male and over 50 years of age, had not spoken the language for a long time. The data that are presented here are all from the only couple3 who address each other from time to time in Leko. Although I checked the data various times, I wish to warn the reader that considerable care has to be taken since I have not been able to check their information with other speakers.

1 My first personal contact with Vladimir Nedjalkov was when he invited me to contribute an article on the southern Quechua dialects to a typological work on reciprocal formation. I was very much impressed by his very precise criticism of the draft version of my text and I did not dare to consent when he asked me to write a short note on reciprocal formation in Leko. I am glad that I could collect enough data on this construction during my last stay in Bolivia which enables me to offer him a first view of reciprocal formation in Leko for his 70th birthday. Readers acquainted with Quechua complex verb formation - for those that are not. Quechua examples are given in footnotes - will be struck, as much as I was, by the parellellism between Leko and Quechua reciprocal formation. I would like to thank the Netherlands Foundation for the Advancement of Tropical Research (WOTRO), a subdivision of the Netherlands for Scientific Research (NWO), for financial support for the field work trips of 1994 and 1996.1 thank L. Kulikov, W. Adelaar, and P. Muysken for their comments on a draft version of this article. 2

In Greenberg (1987), Leko is classified as pertaining to the Northern subdivison of Andean. While the other members of this group, Catacao, Cholon, Culli, and Sechura, show a number of interesting lexical correspondences, in the case of Leko the evidence for this classification is very weak.

3

Tristan Figueredo (born in Uyapi, around 60 years of age) and his wife Trinidad (born in K'arura, around 65 years of age) who live in Mina Blanca, in the vicinity of K'arura.

196

Simon van de Kerke

2. Verb formation in Leko Leko is an agglutinative language with prefixes and suffixes. In (1) we find noq-, prefixed to the verbal root soq- "wait', that seems to add the semantic feature that the action expressed by the verb holds for a relatively large group of people. The suffix -ku is used to form first person plural exhortatives of the type 'let us ..'. Further we find the case marker -ra, realized as -da after liquids and nasals (the same process applies to the inceptive marker -ri 'to go' in (3b-c)). It may have a locational interpretation as in hora here', and a temporal interpretation as in toi wisonda 'in two days'. Reciprocals are formed by adding the suffix -mo to a verbal root as in mi-mo- 'see each other1. Then we find the future tense marker -ra, and finally the first person plural marker -tean is attached4 : (1)

noqsoq -ku ho -ra toi wison -da mi PLUR wait EXH this LOC two day LOC see 'Let's wait here, we see each other in two days.'

-mo -ra -tean REC FUT 1PL

3. Subject and object marking Leko marks person of subject and object on the verb. In some cases subject marking fuses in, as yet not fully understood ways, with the marking of tense, but in most of the paradigms clearly recognizable subject markers can be found, cf. (2) which presents the durative (expressed by means of -cha) present: (2)

era chilchi -cha -no -to: iya chilchi -cha -no -ten kìbi chilchi -cha -no -te chera chilchi -cha -no -tean heka chilchi -cha -no -te -noq kìbi -aya chilchi -cha -no -aya -te

Ί am dancing' You are dancing' 'He is dancing' "We are dancing' You are dancing' 'They are dancing'

Object marking is expressed by means of a prefix on the verb. What exactly counts as an object is not yet entirely clear, and it may be the case that object marking, in the presence of a full referential NP, is optional. The object marker consists of a vowel, in most cases a copy of the first vowel of the root, preceded by /y/ in the case of first and by /d1 or /k/ in the case of third

4

The following glosses are used in the text: INF FUT PAST PRES EXH

Infinitive Future Past Present

Exhortative

PLUR PL SG OB ABL

Pluralizer Plural Singular Object

Ablative

LOC AC INSTR COM

Locative Accusative Instrument Comitative

INC DUR CAUS NEG

Inceptive Durative Causative Negation

197

Verb Formation in Leko: Causatives, Reflexives and Reciprocals

person singular objects. Second person singular objects are marked by the mere realization of the vowel, cf (3a-c): (3) a.

era -ki yo- sobon I AC 1SG visit 'He visited me.'

-aka PAST

-te 3SG

b.

iya -ki osobon -di -eh da -no you AC 2SG visit INC INF want PRES Ί want to go to visit you.'

c.

do- sobon -di -ku cho- bora -ki 3SG visit INC PLUR 1PL friend AC 'Let's go visit our friend.' (-di = -ri Ίο go')

-to: 1SG

In the plural we find /ch/ for first person and /h/ for second and third person. The same process is used to derive the possessive forms of inalienable nouns (cf. 3c, 6-10).

4. Reflexive and reciprocal formation Reflexive and reciprocal formation may be characterized as a valency decreasing process. With the exception of inalienable possession constructions5, to be treated in 4.3., this holds for Leko as well Like many other agglutinative languages Leko does not have reflexive and reciprocal pronouns. While reciprocal marking is expressed by means of a bound morpheme as we have seen in (1), the language appears to be exceptional in not marking the reflexive morphologically.

4.1 Reflexive formation Given the fact that Leko has a full system for object marking, cf. (3a-c), it may be understood that verb forms without an object marker (and without an NP object in the sentence) may be interpreted as reflexives, compare (4a) and (4b):

5

There are two possessive constructions that distinguish alienable and inalienable possessed items. Family, body parts, but also won 'house' are inalienably possessed. The possessor is marked by means of a person marker that is realized on the noun, cf. (3c, 6-10). This is impossible with alienably possessed items. Possession is expressed by means of a construction with -moki on which the person marker is realized: yo-moki walpa 'my chicken'. Person marking on the noun is impossible: *ya-walpa.

198

Simon van de Kerke

(4) a.

b.

era I Ί cut era I Ί cut

cuchillu -ra ber -ki de- wei -ate knife INSTR one AC 3SG cut 1SGPAST someone else with a knife.' cuchillu -ra wei -ate knife INSTR cut 1SGPAST myself with a knife.'

I do not have definite negative judgements on forms like ye-wei-ate with both a subject and an object marker refering to first person singular. However, the fact that the informants never supplied examples of this type suggest that they do not exist.

4.2 Reciprocal formation A reciprocal verb is not marked for object and the affix -mu is added. The pluralizer noq-, already presented in (1) as a suffix for plural marking, enables the speaker to mark explicitly that the subject of the reciprocal verb does not refer to two, but to a larger number of a people: (5)

chera noq- kuqchu -mo -tean we PL kiss REC 1PL *We (various) kiss each other.'

4.3 Reflexive and reciprocal possessives Leko allows double object constructions in which the second object refers to an inalienable possessed item like a body part as in (6) 6 : (6)

era iya -ki owo usuti -no I you AC 2SG back 2SG wash PRES Ί am washing you (with respect to) your back.'

-to: 1SG

The same holds for reflexive and reciprocal constructions:

6

Quechua also has this type of possessive double object construction: (i) for (6) and (ii) for (8): (i) uya -y -ta maylla -wa -n face 1SG AC wash 10B 3SG 'He washes me (with respect to) my face.' (ii) uya -nku -ta maylla -na -ku -nku face 3PL AC wash REC REFL 3PL 'They wash each other (with respect to) their face '

199

Verb Formation in Leko: Causatives, Reflexives and Reciprocals

(7)

era yo- wo suti -cha -no -to: I 1SG back wash DUR PRES ÎSG Ί am washing me (with respect to) my back.'

(8)

suti -mu -cha -no -tean on toi cho- wo wash REC DUR PRES ÎPL this two 1PL back We two, we are washing each other (with respect to) our back.'

5. Causative formation Causative verbs are formed by adding the suffix -ki, which also may occur as an independent verb 'make', to a verb root. The arguments of a transitive verb embedded under the causative verb are both viable targets for object marking: (9) a.

ya -achki ye -kel -ki -aka 1SG father 1SG hit CAUS PAST 'My father made me slap my brother.'

-te yo- noko -ki 3SG 1SG brother AC

b.

ya -achki ye -kel -ki -aka ÍSG father 1SG hit CAUS PAST 'My father made my brother slap me.'

-te yo- noko -i 3SG 1SG brother COM

The two clauses are distinguished by the difference in case marking. In (9a) the object marker refers to the embedded subject (the Causee) and the embedded object is realized with accusative case -ki. In (9b) we find the reverse, the object marker refers to the embedded object, and the Causee is realized with the combative case marker -i?

5.1 Causative reflexives The absence of an embedded object or an object marker triggers the reflexive interpretation in causative-reflexive interactions:

7

This typologically exceptional distribution is also found in Quechua (i) tata -y wawqe -y -ta fether 1SG brother 1SG AC 'My father made me slap my brother.'

maqa-chi -wa hit CAUS 10B

-rqa 3SG.PAST

(ii) tata -y wawqe -y -wan maqa -chi -wa father 1SG brother 1SG COM hit CAUS lOB 'My father made my brother slap me.'

-rqa 3SG.PAST

200

Simon van de Kerke

(10) on atspai da -in -te suti -ki -ch ko- yoki -i that child want NEG 3SG wash CAUS INF 3SG mother COM 'That child does not want to let his mother wash him (self).'

5.2 Causative reciprocals When causative reciprocals are derived the order of the causative and the reciprocal suffix is strictly causative-reciprocal: (11) kibi -aya yates -ki -mo -no -aya -te he PL know CAUS REC PRES PL 3SG 'They are teaching each other (cause each other to know).1 The order causative-reciprocal as given in (11) is expected, since here the reciprocal relation holds between the surface subject and the subject of the embedded verb. However, in (12) the reciprocal relation holds between the arguments of the embedded verb, and still the order of the causative and the reciprocal marker must be as given8 : (12) paya -ki walpa-ki teq -ki -mo -cha -no -to: duck AC chick AC split CAUS REC DUR PRES ÎSG mis ber -tawan wilhi -ki -mo -ra -no -to: tomorrow one after unite CAUS REC FUT PRES ÎSG Ί am separating the ducks from the chickens, tomorrow I will unite them again.'

6. Lexicalized reciprocals and affix raising Now I will turn to an interesting adjacency constraint: the interaction between lexicalization and affix raising. In some cases derivational affixes combine with a verbal root and form a lexicalized or 'frozen' combination with a specialized, although semantically related, meaning. Affix raising is found in complex verbal expressions with a modal verb. In such cases an affix, that 8

In this respect Leko differs from Quechua, since the reciprocal marker may be realized at either side of the causative affix. In (i), comparable to (11), it is also realized after, but in (ii), the translation of (12), it is realized before the causative affix: (i) wawa -kuna yacha -chi -na -ku child PL know CAUS REC REFL 'The children are teaching each other.'

-nku 3PL

(ii) pili -s -ta wallpa -s -ta raki -na -chi -ni duck PL AC chick PL AC separate REC CAUS lsg q'aya tanta -na -chi -saq tomorrow unite REC CAUS 1SG.FUT Ί separate the ducks and the chickens, tomorrow I will unite them again.'

201

Verb Formation in Leko: Causatives, Reflexives and Reciprocals

one would expect to be realized on the embedded verb as far as semantic interpretation is concerned, is realized on the modal verb. It will be shown that lexicalized reciprocal derivations do not allow the reciprocal marker to be raised to the modal verb.

6.1 Lexicalization With certain verbs the reciprocal marker may have a lexicalized meaning, leading to related but specialized meanings. In (13) hepkamoch, derived from the base verb hepkach 'grasp' may both be interpreted as 'grasp each other* and 'give each other a hand': (13) on yobas -aya hepka -mo -cha these man PL grasp REC DUR "These men are grasping each other.' 'These men are giving each other a hand.'

-no PRES

-aya -te PL 3SG

A similar case is the verb kelich *hit' with its reciprocal derivation kelmoch 'fight': (14) hino yobas that man

ye -kel -aka 10B hit PAST

-te on -dep era 3SG that ABL I

de -kel -ate kibi -ki kel -mo -cha -no -tean 30B hit 1SGPAST he AC hit REC DUR PRES ÎPL 'That man has hit me, after that I have hit him, we are fighting.' Such lexicalized reciprocals allow reciprocal constructions with a singular subject. The reciprocal derivation of min- 'see', min-mo may be interpreted as 'see each other*, but may also have the specialized interpretation 'meet' 9 : (15) era min -mo -in -to: on yobas I see REC NEG 1SG this man Ί have not met this man'

9

-ki AC

Some Quechua verbs show a similar semantic specialization. The reciprocal derivation of maqay 'hit' may both be understood as 'hit each other1 and 'fight'. Quechua also allows reciprocal constructions with a singular subject: (i) noqa -yku maqa -na -ku -yku I 1PL hit REC REFL 1PL 'We slap each other.' / 'We fight.' (ii) wawqe -y -wan maqa -na -ku brother 1SG COM hit REC REFL Ί have fought with my brother.'

-rqa PAST

-ni 1SG

202

Simon van de Kerke

6.2 Affix raising With modal verbs like dach "want' and puidisich 'can' that select a sentential complement and that do not have a full thematic grid of their own, the reciprocal marker may either be realized on the lexical verb or on the modal verb 10 : (16) a.

b.

Pedru

Maria

paus -mo -ch puidis forget REC INF can 'Peter and Mary cannot forget each other.' Pedru Maria paus -ich puidis -mo forget INF can REC 'Peter and Mary cannot forget each other.'

-in -aya -te NEG PL 3SG -in -aya -te NEG PL 3SG

6.3 Lexicalization and affix raising Given the possibility of affix raising, treated in the last section, it is interesting that for a lexicalized meaning to hold for a derived reciprocal verb like hepkamoch 'to give each other a hand', the reciprocal marker may not be realized on the modal verb: (17) a.

b.

yobas -aya hepka -mo -ch da -in -aya -te man PL grasp REC INF want NEG PL 3SG 'The men do not want to grasp each other.' "The men do not want to give each other a hand.' yobas -aya hepka -ch da -mo -no -aya -te man PL grasp INF want REC PRES PL 3SG 'The men want to grasp each other.' not: 'The men want to give each other a hand.'

7. Conclusion This article has given a first view of complex verb formation in Leko, a barely known and even less studied Amerindian language of the Andean eastern slopes in Bolivia. The correspondes with complex verb formation in Quechua are striking, as the relevant examples in the footnotes have shown. This could be the outcome of a prolonged contact of Leko with Quechua, but one

Quechua presents the same type of affix raising: (i) Pedru

-wan Maria much'a -na -ku COM kiss REC REFL 'Peter and Mary want to kiss each other.'

(ii) Pedru -wan Maria much'a -y -ta COM kiss INF AC 'Peter and Mary want to kiss each other.'

-y INF muña want

-ta muña -nku AC want 3 PL -na REC

-ku REFL

-nku 3 PL

Verb Formation in Leko: Catisatives, Reflexives and Reciprocals

203

of the prospects for further research will be the question whether other neighbouring lowland languages show similar derivational and interpretational possibilities, which could be suggestive of an opposite influence, or similar typological features.

References Adelaar, W.F.H. (1991) 'The Endangered Languages Problem: South America', in: R.H. Robins & E M. Uhlenbeck (eds.) Endagered Languages, Collection Diogène, Berg Publishers, Oxford, pp. 45-92. Brinton, D.G. (1946) The American Race, Biblioteca Americanista, Editorial Nova, Imprenta López, Buenos Aires. Cardús, Fr. José OFM (1886) Las Misiones Fransiscanas entre los Infleles de Bolivia: descripción del estado de ellas en 1883 y 1884, Librería de la Inmaculada Concepción, Barcelona. Greenberg, J.H. (1987) Languages in the Americas, Stanford University Press. Grimes, B.F. (1988) Ethnologue: Languages of the World, 11th edn. Dallas, Summer Institute of Linguistics. Lafone Quevedo, S.J. (1905) La Lengua Leca, Anales de la Sociedad Científica Argentina tomo 60. Kerke, S. van de (1996) Affix Order and Interpretation in Bolivian Quechua, unp. diss. University of Amsterdam. - (to appear) 'Reciprocal Formation in Bolivian Quechua' in: V. Nedjalkov & Z. Guentcheva (eds.) Typology of Reciprocal Constructions, Lincom Europa, München. Montaño Aragón, M. (1987) Guia Etnográfica Lingüística de Bolivia, Don Bosco, La Paz (Bolivia). Palau, M. & Β. Saíz (1989) Moxos, descripciones exactas e historia fiel de los indios, animales y plantas de la provincia de Moxos en el virreinato del Perú por Lázaro de Ribera 1786-1794, Ministerio de Agricultura y Alimentación, Ediciones El Viso, La Paz (Bolivia). Wedell, H.A. (1853) Voyage dans le nord de Bolivie et dans les parties voisines de Pérou.

Verbal Categories, Meaning and Typology Jarmila Panevová/Petr

Sgall

1. Approaches based on European structural linguistics prototypically understand language as a system organized in several levels coming from de Saussure's 'signifiant' (phonemics) to his 'signifié' (underlying structure); the latter level constitutes a language specific patterning of cognitive content and thus can be viewed upon as an interface between language and cognition. In the Praguian Functional Generative Description, morphological categories and their values are classified as underlying (tectogrammatical) items and are described in their relationships to surface (or morphemic) items expressing them The verb with its valency is the centre of the sentence, and among its morphological categories there are those of modality, tense and aspect, whose values semantically pertain directly to the interpretation of the sentence as a whole. The set of verb categories usually is much richer than that of the noun or of other word classes. Thus it is not surprising that the subsystem of the verb proper to a single language often is not confined to a single type of language. Viewing language types with V. Skalicka as 'ideal extremes' not fully attained in existing languages,1 we find properties of various types in the conjugation of many languages in which the declension is much more restricted to a single type: in Latin the verb exhibits highly agglutinative forms such as canta-vi-sse-n-t [they would have sung], in Czech there are analytic forms such as byl by ses mohl smâ [you would have been able to smile], and so on, although the noun of the two languages is inflectional, with a single ending (or alternation) for both case and number. In this paper we discuss tense and aspect, which, in Czech, exhibit patterns a comparison of which to those of other languages may be of interest. Altogether, we distinguish the following morphological categories of the Czech verb with their values: tense (Simultaneous, Anterior, Posterior), extension (Extended, Immediate, which can be illustrated by Cz. koufi in its two meanings: [smokes = is a smoker] and [is smoking - in a specific time interval]), aspect (Processual, Complex, Resuhative), iterativeness (Iterative, Non-Iterative), and at least two layers of modalities (the motivation for this classification of the categories and their values can be found in Sgall et al. 1986). In several recent publications, also by V.P. Nedjalkov, the terms 'Simultaneous, Anterior, Posterior' are used in connection with the category of taxis, rather than for tense. We will see in Section 2 that the mentioned three values can be understood as relevant for the meaning of

1

Differing from Skalifka, we regard the clusters that constitute the individual types as based on an asymmetric relation of favourability, rather than conceiving the relevant properties as 'mutually favourable'; this modification allows us to make use of a hierarchical ordering of the properties, see Sgall ( 1995).

206

Jarmìla Panevová / Petr Sgall

morphemic tense in the general case, the 'absolute' tenses representing a specific case of this category. 2. The limited inventory of grammatical tenses can render the rich variety of cognitive temporal relations as relations between individual events following each other, overlapping or simultaneous on the time axis. As is well known from classical writings on Slavonic and other languages, absolute (basic) tense meanings refer to the relationship of the event expressed by the given verb to the time point of speech, of the utterance (present, past and future), whereas the relative values concern relationships between two events (simultaneous, anterior, posterior). Thus, in (1) and (2), the verbs feel, assures and told express absolute tense; on the other hand, the verbs would do and would need express the temporal relationship to another event to that of knew and of told, respectively. (1) (2)

I feel worried, since Jim assures me that he didn't know what Mike would do. Jim told us that his wife would need a new coat.

However, the following two major questions still seem often to be answered without any precision: (a) Under what conditions does a verb occur with an absolute, respectively relative tense meaning? (b) If relative tense is present, which time point is taken as the point of reference? Question (a) usually is being answered so that relative tense occurs with the verb of a content clause.2 As for question (b), usually relative tense values are claimed to specify the event in its temporal relationship to the event of its (immediately) governing clause, as in (1), or to that of the main clause of the complex sentence, as in (2). As example (3) shows, answers of this kind are not satisfactory. (3)

Jim told us a year ago that John thought that his wife would need a new coat in a month, since the fashion would change before the autumn.

In (3) the tense of would change is relative, although this verb occurs in a causal (not content) clause. Moreover, this tense relates neither to the main clause, nor to the immediately governing one. It expresses posteriority towards thought. A more adequate solution to questions (a) and (b) has been offered, although it still has not been widely discussed. Approaching this issue (see Hajicová et aL 1971), we, following Reichenbach (1947), work with the point of speech S, the points of individual events (E¡ for different clauses in a sentence) and the points of reference of individual verbs (R,).3 However, we prefer to specify R in such a way that the relationship between E¡ and R always can be

2 The prototypical case of a content clause is an indirect speech in the position of the Objective of say, teli, assert, claim·, other content clauses are those after know, hope, suspect. 3 This implies a simplification, because not only time points are concerned. We are aware of the necessity also to work with intervals, since the events may be repeated, continuous, and so on.

Verbal Categories, Meaning and Typoiogy

207

understood as determined by the tense of the verb V¡ - as simultaneous, anterior, posterior.4 The opposition of absolute and relative tense can then be seen as reversed: the general case is reflected by the just mentioned values of relative tense, whereas absolute tense is present if the point R is determined as identical to S on the basis of general rules of the sentence structure. These rules can be captured by the following principle, valid (with certain exceptions, mentioned below) for every declarative sentence: The recursive principle of tense values (RP): (i) if the verb V¡ is the main verb of the sentence, then Ri is identical to the point of speech, S; (ii) if Vj is the verb of a content clause C, then R, is identical to the time point Ek, Le. to the point of event of Vk, this verb being the verb of the clause governing C in the sense of dependency syntax;5 (iii) in all other cases, Ri is identical to Rk (where Vk again is the verb of the governing clause). It is then possible to state the following two theorems (the proofs of which, in a slightly different wording, can be found in Hajicová et al. 1971); Vn denotes here a verb reporting V¡, i.e. a verb governing a content clause C such that V¡ is subordinated to the main verb of C (with 'subordinated to' meaning the transitive closure of'dependent on', with which the identity of Vj with the main verb of C is not excluded); if there is no content clause Κ different from C such that Vi is subordinated to the main verb of Κ and the main verb of Κ is subordinated to Vri, we say that Vri immediately reports V¡. Theorem 1: If there is no V„, then R¡ is identical to S; otherwise Rj is identical to Ek, where Vk immediately reports V¡. (In the former case we say that V¡ has an absolute tense, in the latter case the tense of V¡ is relative.) Theorem 2: The order ofEjto S on the temporal axis (i.e. Έ, < S' if E¡precedes S, Έ,= S' if the two time points coincide, Έ; > S' if E¡ follows S) can be inferred uniquely from the values present in the tenses of the sentence, iff at least one of the two tenses 'anterior1 and 'posterior1 is absent from V¡ and from any V„; at least one of the following points then holds: (a) every Vri has the value 'simultaneous'; (b) Vi has not the value 'simultaneous', and with every Vri the value of tense is either 'simultaneous' or is equal to the value of the tense of V¡; (c) Vi has the value 'simultaneous' and either 'anterior* or 'posterior1 is absent from every Vri; iff (a) holds, then Ri can be inferred to equal S; iff (b) holds, then E¡ can be inferred to be simultaneous, anterior or posterior to S iff the value of V; is simultaneous, anterior or posterior, respectively; iff (c) holds and (a) does not hold, then E¡ can be inferred to be anterior

4

We do not deal here with tense values such as 'immediately after (before)' (present e.g. in I am going to do, or in the French je viens de faire, respectively), 'regular repetition' (/ used to do) or 'remote past'.

5

In this sense, e.g. assures governs did not know, which itself governs would do in (1). The dependent clause always occupies a syntactic position (that of object, subject, adverbial, etc.) within its governing clause.

Jarmila Panevová / Petr Sgalì

208

(posterior) to S iff the value of the tense of Vm is anterior (posterior), where Vm is a Vn whose tense differsfrom'simultaneous'. Let us present further illustrating examples corresponding to the conditions specified in Theorem 2: (A) Point (a) is met for existed in (4) and for will do in (5), so that their points of reference can be inferred to equal S: (4)

John doubts whether anybody among his schoolmates maintains that s/he knows whether Homer existed. R(existed) = E(knows) according to (ii) in (RP); R(knows) = E(maintains) according to (ii); R(maintains) = E( doubts) according to (ii); R(doubts) = S according to (i); since T(doubts), Le. the tense value of doubts, is 'simultaneous', E(doubts) = R( doubts) = S;

since T(maintains) is 'simultaneous', E(maintains) = R(maintains) = E(doubts) = S; since T(knows) is 'simultaneous', E(knows) = R(knows) = E(maintains) = S; since T(existed) = 'anterior1, E(existed) < R(existed) and, since R(existed) = E(knows) = S, E(existed) < S. (5)

I hope that you know whether John will do it. R(do) = E(know) according to (ii); R(know) = E(hope) according to (ii); R(hope) = S according to (i); since T(hope) is 'simultaneous', E(hope) = R(hope) = S; since T(know) is 'simultaneous', E(know) = R(know) = E(hope) = S; since T(do) = 'posterior1, E(do) > R(do) and, since R(do) = S, E(do) > S.

(B) Condition (b) is met in (6) and (7): (6)

John doubted whether anybody among his schoolmates maintains that s/he had found out whether Homer had existed. E(had existed) can be inferred to precede S.

(7)

I hope that you will know whether John will do it. E (will do) > S.

(C) Condition (c), not (a), is met in (8): (8)

I hope that you will know at noon whether John is taking lunch already. E(is taking) > S.

Verbal Categories, Meaning and Typology

209

(D) None of the conditions (a) - (c) is met in (9): (9)

John doubted whether his schoolmates would maintain, after two days of drill, that they still enjoy it. The temporal relationship between E(enjoy) and S is not determined.

Although neither the rules of the sequence of tenses in English, nor the interplay of secondary tenses such as Pluperfect or the Second Future with would, can be discussed in the present paper, it can be seen that even in languages with very different systems of verb tenses we may find the opposition of absolute and relative tense. The former is present in cases in which the identity of R¡ and S is given by the absence of any Vn, see Theorem 1 above, and the latter is present in all other cases, Le. with the verbs of content clauses and of clauses subordinated to these. In Czech, the primary tenses, Present, Preterite and Future, when having the relative value, express simultaneity, anteriority and posteriority of E¡to Ri, respectively: (10) ftekl mi, ze cte/cetl/pfecte tu knihu. - [lit. He-told me that he-reads/read/will-read that book.] In instances of'the speaker's comment', especially in relative clauses, an absolute value of tense may occur, although the given clause is subordinated to a content clause: (11) Pozoroval(Pret), ze bude(Fut) nucen vejít ν diskusi, která mu byla(Pret) protivná. - [ht. He-saw that he-will have to-enter into (the) discussion, which for-him was nasty.] Although such exceptions exist, the recursive principle appears to be an important means for making the distinction between absolute and relative tense values explicit. 3. Aspect often has been characterized as a "subjective" category (see e.g. Kopecny 1962: 12, note 13 and the writings quoted there, more recently, cf. esp. Comrie 1976). However, this category is relevant not only for the "speaker's viewpoint", but also for semantics itself (truth conditions), cf. the analysis presented in Panevová and Sgall (1973) and Sgall et al (1986). The basic aspectual opposition in Czech, Russian and most other Slavonic languages can be seen in the pair of the meanings Complex and Processual, prototypically expressed by the morphemic forms of Perfective and Imperfective, respectively. (We have introduced the term 'Complex1, since the meaning of Perfective cannot in general be understood as 'completed', although with many action verbs this would correspond to the expressed content). As (12) and (13) illustrate, the Complex aspect specifies the event as finished, closed, and with Processual aspect the event is viewed upon as proceeding, without a delimitation.

210

Jarmila Panevová / Petr Sgoli

(12) Prepisovaljsem dopis. [I was rewriting a letter.]6 (13) Prepsal jsem dopis. [I rewrote a letter.] A third value of the category of aspect in Czech can be seen in the value of Resuhative, which is expressed either by the Czech Perfect forms (e.g. mam tu knihu pfeäenou [I have read the book]) or by the Preterit Perfective forms (which thus are supposed to be ambiguous). The Czech Resultative is not combined with the value of Processual, so that it is possible to include the three units of aspectual meaning among the values of a single category. Along with the mentioned prototypical relationships between tectogrammatics and morphemics, there are also asymmetries in the ways of expressing aspect. One of them concerns the use of Imperfective forms in the sense of Complex aspect with the 'Historical Present1, cf. such often discussed examples as: (14) V roce 1848 Marx prijízdí do Parize. [In the year 1848 Marx came to Paris.] (Isacenko 1960: 450) In such a case the combination of Present and Imperfective expresses that of the meanings of Anterior and Complex, i.e. tense and aspect share a single morpheme. Another case of asymmetry is that of the mentioned ambiguity of Perfective forms, which express Complex as well as Resultative aspect (often differing in stylistic values from the Czech Perfect in the latter case). Preterite Perfective forms are then also ambiguous in expressing both Simultaneous and Anterior Resultativeness. It is also possible to find Imperfective forms expressing Resultativeness in certain marginal cases, esp. with verbs of apperception (Jak to slysel, sel ti to fíct [As he heard this, he went to tell you], and also Ceti to [He has read it]). A similar kind of asymmetry can then be seen in the simple (Present) forms of Perfective aspect expressing Processual Posteriority, e.g. ponese [he will be carrying], pojede [he will be driving/riding]. We thus classify these morphemic items similarly as the regular Present Complex forms, e.g. pfinese [he will bring], which clearly express Complex Posteriority (this well known case of asymmetry is of a larger range than the others). An often discussed question is whether aspectual pairs (forms differing only in that one expresses the Complex, and the other one the Processual aspect) constitute two lexical units or are just different grammatical forms of a single lexical item (though differentiated by morphemic means that regularly serve for lexical derivation). If morphemic properties of lexemes (conjugation clases, etc.) are in the foreground, then the pairs may well appear as pairs of lexical units. However, it is difficult to imagine a definition of lexical unit that would distinguish between two units sharing their lexical meanings and differing just in grammatical values. Thus, if e.g. dà - dávat [give - be giving] or vant - uvant [be cooking - cook] are understood to differ just in aspect, the difference between them belongs to the same layer as that between e.g. dává - dával - bude dávat [he is giving - he was giving - he will be giving], i.e. it is a

6

In (13) the action is characterized as having been completed, which is not the case in (12).

Verbal Categories, Meaning and Typology

211

difference of grammatical forms. Therefore we regard every pure aspectual pair as belonging to the same lexical unit in tectogrammatics, and constituting two verbs at the level of morphemics. The so-called bi-aspectual verbs (Cz. 'obojvidová slovesa') are then classified as ambiguous, cf vénovat [donate] or informovat [inform]: their Imperfective forms express either Processual or Complex. The (numerous) verbs called by J. S. Maslov (1963) and others 'imperfectiva tantum1, e.g. mit [have] or bydlet [stay] exhibit a restriction on the values of aspect: both Complex and Resultative are excluded with them. On the other hand, the 'perfectiva tantum' are not really numerous, if pure aspectual pairs are seen not only in cases of suffixation, but also in some cases with prefixes, e.g. psät - napsat [be writing - write]. Pure aspectual prefixation is found there where the prefixed form does not serve as a basis for aspectual suffixation (there is no *napisovat), or, else, if the resulting suffixed form is synonymous with the basic one, as e.g. with úóastnit se - zúcastnit se - zúcastnovat se [participate] or blíiit se piibliiit se - piibliiovat se [approach] (with the basic Imperfective form slowly getting obsolete). Lexical ambiguity has then to be admitted (a) with Imperfective forms, e.g.: krajetj - nakrájet [slice] : krájet2 - ukrojit [cut up] blednouti - zblednout [pale] : blednout2 - vyblednout [fade] däit\ - rozdäit [distribute] : délit2 - vydélit [divide], and (b) with Perfective forms, e.g.: rozdäit\ - délit [distribute] : rozdäit2 - rozdélovat [part] In several cases, such as dédit - podédit, -dédit [inherit], or ρtat se - zeptat se, optât se [ask] the Imperfective form corresponds to two synonymous Perfective ones. Although the opposition in linguistic meaning of Complex and Processual can be perceived as general, individual groups of verbs display various oppositions in the cognitive layer. The prototypical case is that of the Complex aspect specifying the event as having achieved its terminal state (e.g. psät - napsat [be writing - write]); this is not the case with libat - polibit [kiss] (where the lexical meaning determines the activity as always achieving its termination: if it is true that Bill was kissing Mary at a certin timepoint, then it also holds that he kissed her at that timepoint). Other shades of the opposition can be found as determined by the lexical meanings e.g. in dávat - dát [give], zrazovat - zradit [betray], tcsit - potest [delight], etc. Usually the Imperfective form is ambiguous in that it also expresses the Iterative meaning. Distinguishing between aspect as a grammatical category and 'Aktionsart' as a set of lexical derivations (as has regularly been done in Slavonic studies), we should ask whether Iterativeness can be understood as systematic, as a matter of grammar, even though it often lacks an overt morphemic expression. In the quoted writings, we offer arguments for regarding the opposition of Iterative - Non-Iterative as grammatical. Pairs such as jit - chodit [go] witness that the opposition in meaning is linguistically patterned in Czech, and the presence of this opposition in dát - dávat and other pairs just quoted (with one of the readings of the Imperfect members) supports the view that Iterativeness is systematic, general enough to be viewed as belonging to grammar.

212

Jarmila Panevová / Petr Sgall

On the other hand, oppositions such as those between psát [be writing] and pfepsat [rewrite], dopsat [finish writing], zapsat [inscribe], pñpsat [add], etc., concern not only aspect, but also lexical derivation: the derived Perfective forms have their specific lexical meanings and their secondary Imperfective counterparts (prepisovat, dopisovat, etc.). Rules for the transduction of the tectogrammatical units of aspect (in part together with those of tense) into morphemic items were presented in Panevová and Sgall (1973). The situation of aspect in Russian is very similar, but certain differences have to be taken into account: (a) in tectogrammatics, Resultative seems to be absent from the repertoire of units in Russian; (b) the combinability of Iterative and Complex is more restricted in Russian than in Czech, or at least the morphemic expression of this combination differs: cf e.g. Cz. Kazdydeti potkám listonóse ηα rohu ulice [Every day I meet the postman at the corner of the street], where the Perfective Present exceptionally (in the context of Extended) expresses Simultaneity (Cz. potkám is the Perfective form, which primarily expresses the values of Posterior and Complex); (c) the linguistic patterning of the cognitive content may be understood to differ, if Czech Non-Iterative is understood to be cognitively equivalent to Russian Iterative in examples such as: (15) Cz. Vcera sel do kina. - R Vcera on xodil ν kino. [Yesterday he went to the cinema.] However, as the presence of the Perfective sxodit' suggests, R xodit' can perhaps be understood to have also a Non-Iterative meaning. In any case, the differences between the two languages are just marginal and the rules proposed for Czech can serve as a starting point for the description of Russian aspect (with Resuhativeness being excluded).

Verbal Categories, Meaning and Typology

213

References Comrie B. (1976) Aspect: An introduction to the study of verbal aspect and related problems. Cambridge: Cambridge University Press. Hajifová E., Panevová J. and P. Sgall (1971) The meaning of tense and its recursive properties. Philologica Pragensia 14, 1-15, 57-64. Isa£enko A.V. (1960) Grommatiéeskij stroj russkogo jazyka ν sopostavlenii so slovackim. II. Bratislava: SAV. Kopeiny F. (1962) Slovesny vid ν CeStiné. [Verbal aspect in Czech]. Prague: ÓSAV. Maslov Ju.S. (1963) Morfologija glagol'nogo vida ν sovremennom bolgarskom literaturnom jazyke. Moskva/Leningrad. Panevová J., BeneSová E. and P. Sgall (1971) Cas a modali ta ν ôeStinê [Tense and modality in Czech], Acta Universitatis Carolinae - Monographie XXXIV. Panevová J. and P. Sgall (1973) ¿as a vid éeského a ruského slovesa [The tense and aspect of the Czech and the Russian verb], Slavia 42,16-24. Reichenbach H. (1947) Elements of symbolic logic. New York. Sgall P. (1995) Prague School typology. In: M. Shibatani and Th. Bynon (eds.) Approaches to language typology. Oxford: Clarendon Press, 49-84. Sgall P., Hajiôovà E. and J. Panevová (1986). In: J. L. Mey (ed.) The meaning of the sentence in its semantic and pragmatic aspects. Dordrecht: Reidel - Prague: Academia. Skalifka V. (1979) Typologische Studien. Braunschweig: Vieweg.

The Perfect in English and German Wolfgang Klein / Heinz Vater

1. "Perfect" - a universal or language specific category? 1.0 Preliminary remarks What beyond mere tradition entitles the grammarian to assign a particular form to a grammatical category? What is the rationale behind the fact that we infallibly call hominis a genitive, perepisyvat' an imperfective, and vidi, (ich) habe gesehen, (I) have seen and (¡') ai vu a perfect? This immediately raises the more fundamental question of what constitutes the unity of a category within a particular language and across languages. What, for example, defines "perfect" in German, "perfect" in Latin, and "perfect" in general? A grammatical category is always a mapping between particular formal means and particular meanings (or functions). In the ideal case, unity of formal marking and of specific meaning component should go together in the definition of a category. Apparently, this case is rarely found in human language. In fact, the situation is even worse. Even within a particular language, a grammatical category is often neither uniform with respect to form nor with respect to meaning. Latin genitive, for example, has no fixed meaning, and its morphological marking is highly variable. We cannot say that it is always marked by a -i or by -is or -ae, nor can we say that it is defined by the fact that it expresses "possession" (or whatever other semantic feature). Still, we all assume that that there is a grammatical category "genitive" in Latin, and we also assume that other (though perhaps not all other) languages have 'this' category, as well. Does this solid faith, so basic to linguistic theory and and even more to linguistic practice, have a fundamentum in re; if so, what is it? In this paper, we shall address a small instance of this general problem - what is "perfect" in German, what is "perfect" in English, and how are these two instantiations of the cross-linguistic category "perfect" - if there is such a category at all - related to each other. This is not the first attempt in this regard. ANDERSON ( 1 9 8 2 : 2 2 7 ) begins his study with the following question: "A grammatical category, such as the "Perfect", will not have exactly the same range of uses in one language as it does in another. Given that this is true, why are linguists willing to use the same name for the category in two different languages? How can we sensitively compare languages?"'

He solves the problem in looking for common traits of meaning among the corresponding forms in different languages and drawing "maps" that show the meaning areas of each form.

1

"Meaning" is to be understood in the sense of "literal meaning", whereas "use" is determined by the literal meaning plus meaning components furnished by the context and the construction.

216

Wolfgang Klein/ Heinz Vater

IÇ for example, a form in language Li (e.g. the perfect) has the variants A, Β and C, whereas a corresponding form in L2 has the variants A, Β und D, then a map is drawn on which the two languages overlap in A und Β.

"We can determine 'similarity* of meaning typologically. If two particular meanings are often expressed by the same surface form (across languages), then we can assume that the two meanings are 'similar1 to the human mind." (ANDERSON 1 9 8 2 : 2 2 7 )

On the whole, this inductive procedure2 has considerable plausibility. Most linguists probably proceed like this unconsciously in comparing categories across languages. However, Anderson's procedure has to be supplemented in two respects: First, the so-called "surface forms" have to be submitted to a comparative analysis. Usually, different languages do not use the same morphosyntactic means for representing a category. And secondly, one of a number of different meaning variants (or uses) of a category may be more central as the other ones, as COMRIE (1976:11) notices: "Where a form is said to have more than one meaning, it is often the case that one of these meanings seems more central, more typical than the others. In such cases, it is usual to speak of this central meaning as the basic meaning."

"Central meaning" should, however, not be equated with "basic meaning"; basic meaning can be obtained by abstraction from individual meanings. Thus, REIS 1980 holds that every lexical or grammatical expression has a literal meaning being the sum of all meaning components that have turned out to be invariable throughout all possible contexts: There is reason to assume that the "basic meaning" of an expression is, on the whole, equivalent to its literal meaning. The central meaning is a component of the basic (literal) meaning of an expression.

1.1 Formal and functional divergence What is called "perfect" in different languages, is neither uniform in meaning nor in form; but there is a number of similarities in both regards. Typical meaning components will be discussed in the next section.

2 "We can develop grammar/meaning spaces inductively, finding a 'map' which works consistently for many languages..." (ANDERSON 1 9 8 2 : 2 7 ) .

217

The Perfect in English and German

As for form, it is common to distinguish two different modes of construction: the synthetic (morphological) and the analytic (syntactic) perfect. The synthetic construction allows for several different procedures: (l-02)a Suffixation: b root modification: c Reduplication: d Prefixation:

lat. amavit "he has loved";3 lat. fecit "he has made " (vs. pres. facit); gr. pepaideuke "he has educated " (vs. pres. paideuei)·, Swahili amefika "he has come" (COMRIE 1976:57). 4

Analytically the perfect is formed by means of auxiliary verbs in many languages ; thus, in all Germanic and Romance languages, a finite form of the auxiliary is combined with the participle of the main verb (cf. (l-03)a); this is not the only possibility; in Bulgarian, for example, the finite form of the auxiliary is combined with the finite preterit tense form of the main verb.3 In forming the progressive perfect in English, the finite auxiliary is combined with the participle of the auxiliary and the -wig-form of the main verb: ( l-03)a b c

Aux + participle: Aux + preterit: Aux. + part. +\ing:

Engl, he has arrived·, Gm., er ist gekommen, Fr. il est venu; Bulg. Ivan e dosai (COMRŒ 1976:107); Engl, he has been waiting.

It is not at all clear, however, whether the two types of construction indeed reflect the same notional category. Latin perfect, for example, - that is, in the language which gave its name to the category - is typically used to describe singular, completed actions in the past, that is, essentially it serves the function of French passé simple (in written language) and English simple past, rather than French passé composé and English present perfect; next to this "perfectum historicum" usage, it can also serve as a "perfectum praesens" in some cases; but this function is clearly marginal. French, English or German translations of say De bello gallico normally render Caesar's perfect forms by passé simple, simple past and preterit, respectively, hardly ever by passé composé, present perfect or Perfekt. This divergence in form as well as in function is somewhat less salient if we only consider the Romance and Germanic languages in which the "perfect forms" are composed of a present tense auxiliary and past participle. This construction naturally invites a compositional interpretation of the perfect meaning. COMRIE (1976:107) surely gives expression to a wide-spread view when he writes: "the present auxiliary conveys the present meaning, while the past participle conveys that of past action." 3

It seems that -v- has to be analyzed as perfect suffix, to which the personal suffixes (e.g.. -it in the 3.pers.) are added; the l.Ps.Sg. remains unmarked. The suffixes -υ- (cf. tenuif) und -s (cf. rexit) can be assumed to be phonological or morphological variants of the suffix -v-.

* According to COMRIE ( 1976:57), the Perfect prefix me- is preceded by the subject prefix a-. 3

Originally, this preterit form used to be a participle. Now, it can be considered as a form that is unmarked for the contrast 'finite' vs. 'infinite' (analogously to the infinitive construction in Bulgarian).

218

Wolfgang Klein / Heinz Vater

This is in agreement with the feet that quite often the perfect while describing an event that took place in the past relates in a special way to the present. But quite correctly, COMRIE (1976:107) also points out that sometimes "...there is a discrepancy between form (which includes both present and past) and meaning (which is often just past)". In fact, the apparent parallelity in construction may be quite deceptive. A brief comparison betwen English Peter has left London, German Peter hat London verlassen and French Pierre a quitté Londres will illustrate the point. They all describe an event in the past; they do not say when exactly this event took place; all that is said is that the time of the event, here Peter's leaving London, precedes the time of utterance. It could have been yesterday, for example. In German and french, the precise event time can easily be made explicit by an adverbial, whereas this is impossible in English: ( 1-04) a b c

* Yesterday, Peter has left London, Gestern hat Peter London verlassen, Hier, Pierre a quitté Londres.

The only way to be more specific with the precise time in English would be to replace the present perfect with the simple past: Yesterday, John left London. This and similar observations have given rise to the assumption that despite their formal composition German Perfekt and French passé composé are semantically not compound in the same way as the English present perfect (i.e., consisting of a "present component" and a "past component") but rather function like a simple past tense. But it is easy to see that there are other contexts in which English and French go hand in hand, whereas German is different. Suppose someone wants to state a simple fact in the past, e.g. the fact that the Colossus of Rhodes weighed 100 tons. In this case, neither English nor French can use their "perfect", whereas German can: (l-05)a b c

*The Colossus of Rhodes has weighed 100 tons, *Le colosse de Rhode a pesé 100 tonnes, Der Koloss von Rhodos hat 100 Tonnen gewogen.

Here, English requires the simple past, and French requires the imparfait; in German, the preterit is also possible but probably less common. Hence, the mere idea that the "perfect" functions sometimes like a composition of present and past meaning components and sometimes simply like a past tense as well as the connected idea that this difference reflects a typical historical development is perhaps not false but surely a gross oversimplification. In order to understand what "perfect" in a particular language really means, we must have a much more careful look at its various uses in this language. This is an extremely difficult task. The following discussion will essentially be confined to English and German. From now on, we shall use the term "present perfect" for the English present perfect, as in John has left London, the term "Perfekt" for its German counterpart in Peter hat London verlassen, and the term "perfect" for the category in general.

The Perfect in English and German

219

1.2 Range of uses of the perfect There are two salient differences in the way in which Perfekt, on the one hand, and present perfect, on the other, can be used. One of those has been mentioned above: German Perfekt oñen functions like the English simple past. In this function, it competes with the preterit. It has often been argued that these two forms share the same temporal meaning, the difference being more a question of style, register or perhaps dialect. This is not correct, though; we shall return to this point below. Second, the German Perfekt can easily relate to the future. It cannot only be combined with past time adverbials, as in Gestern um zehn hat er London verlassen Yesterday at ten, he left London' but also with future time adverbials, as in Morgen um zehn hat er London verlassen 'Tomorrow at ten, he will have left London'. Here, the temporal adverbial does not refer to the time of event itself) his leaving, but a time at which this event is over. Any comparative account of Perfekt and present perfect has to explain these two facts. Other differences in the range of use, partly connected to those, partly different, are less obvious and hence much harder to grasp. There are numerous descriptive accounts on how the perfect can be used in English (see for example MCCOARD 1978, FENN 1987). Consider, for example, the list of "major uses" that ANDERSON ( 1 9 8 2 : 2 2 8 ) offers: (l-06)a b c d e

"experiential": "current relevance of anterior": "new situation" ("hot news"): "result-state": "continuous":

Have you (ever) been to Japan? He has studied the whole book, (so he can help) The Etna has just erupted! He has gone, (or) He is gone, (is not here) I have been standing here for three hours.

A sixth use, "anterior", is illustrated by the example John thought Mary had left. But this is a pluperfect, which should remain out of consideration here. Concerning e, it can be assumed that the "continuous" meaning has to be ascribed to the progressive aspect (expressed by be + -ing) rather than to the perfect. In the interlingual comparison, it is interesting to note that the meaning of the continuous perfect is expressed by the present tense in German (and, incidentally, in French). The uses a - d can be compared in a direct way with the corresponding uses in German; d can be expressed in German - as in English - only by the Perfekt, whereas a, b, and c also allow for the preterit - which often is preferred - (cf. (1-08)).6

6 The situation in German is complicated: In those areas, where preterit and Perfekt overlap, it is oñen said that a complementary distribution has been obtained: with certain groups of verbs (e.g. auxiliary and modal veibs), the preterit is predominant, with other groups the Perfekt. Besides, some uses of the preterit seem to be idiomatic and, thus, obligatory, as e.g. (1-08)0' shows.

220

Wolfgang Klein / Heinz Vater

(l-07)a ?Sind Sie schon (einmal) in Japan gewesen? a1 Waren Sie schon (einmal) in Japan? b Paul hat sich mit Biologie befaßt. (P. kennt sich in Biologie aus.) c Eben hat es geblitzt, d Fritz ist weggegangen. (Fritz ist weg.) ( 1-08) a ?Gerade ist der Briefträger dagewesen. a' Gerade war der Briefträger da. b *Soeben haben Sie die vierte Symphonie von Anton Bruckner gehört, b' Soeben hörten Sie die vierte Symphonie von Anton Bruckner. In the northern and central areas of German, the Perfekt of sein (as well as the Perfekt of haben, the modal verbs and some frequent strong verbs like kommen and gehen) is extremely infrequent; (l-07)a' and (l-08)a' with preterit sound more natural than the corresponding sentences (l-07)a and (l-08)a with Perfekt. (l-08)b* is a normal broadcast announcement, whereas (l-08)b - which one would expect - is never used. The emerging picture is not a very homogeneous one. In the sense of Andersons's diagram (1-01), we get some overlap but also some clear divergences. Rather than following these up in detail, we shall now turn to the question whether there is a uniform meaning behind the different range of uses in the two languaes. We will then discuss whether this uniform meaning can explain the two salient differences as well as some of the subtler ones.

1.3 Perfect and temporal relations Most recent attempts to assign a uniform meaning to the present perfect as well as to the Perfekt use the framework which REICHENBACH 1947 originally proposed for English. He noted, as others did before him (see EHRICH 1992:65), that the meaning of tense forms in general cannot just be described in terms of temporal relations between the "point of speech (S)" and the "point of the event (E)", as the traditional notion of tense suggests but that at least in many cases, a third point in time - which he calls "point of reference (R)" is needed.7 This idea was taken up by many authors. There is some disagreement as to whether R is necessary for all tenses. COMRIE, for example, argues that present or past tense in English are adequately represented ifjust the relationship between E and S is taken into account (cf COMRIE: 1985:36FF): 8

7

Reichenbach's temi "event" comprises states, processes and actions, i.e. everything that can be expressed by a verb. In more recent works (e.g. EHRICH 1992), the term "situation" is used as a cover term. This is also done here.

8

For COMRIE (1985:56ft) the point of reference seems only to be relevant for the interpretation of "relative tenses" such as perfect and pluperfect.

The Perfect in English and German

(1-09) (1-10)

I see John. I saw John.

221

(E,S; i.e.: E simultaneous with S). (E unvermeidbar bzw. unvermeidlich), auf die hier nicht näher eingegangen werden soll.

Die Verneinung als funktional-semantische

Kategorie

297

Probleme, die sich für die semantiche Interpretation im Verhältnis von bleiben und sein hinsichtlich des Ausdrucks von Prozessualität und Statualität ergeben, sind die gleichen wie im Abschnitt 4.1.1.2.1., brauchen hier also nicht noch einmal behandelt zu werden. 4.1.1.2.2.1 VS mit Präposition ohne Der am häufigsten gebrauchte Typ sind Verbindungen mit dem FV bleiben. Da die lexikalischen Restriktionen hinsichtlich der Bildbarkeit und Verwendbarkeit hier offenkundig größer sind als in den SF mit FA, sollen die jeweiligen Frequenzen, die sich anhand des Belegmaterials ermitteln ließen, angegeben werden. Für mehrfach vorkommende Belege eines VS steht die Zahl in Klammern dahinter. In SF dieses Typs kommen folgende nomina actionis vor: Antwort (7), Aufklärung, Ausgleich, Bildung, Einschränkung, Gliederung, Hilfe (2), Kontrolle (2), Nominierung, Resonanz, Schutz, Strafe (2), Unterstützung, Widerspruch (3), Widerstand, Zerstörung. Zu diesen direkt von Verben abgeleiteten VS kommt das Kompositum Berufsausbildung. Außerdem drei sementiseli mit VS korrespondierende Konkreta, nämlich Blessur, Wunde, Unterschrift, in Kontexten, in denen sie sich als Synonyme von VS interpretieren lassen. Vgl.: w o keine der 300000 Seelen ohne Blessur geblieben ist (