193 8 10MB
English Pages 437 [440] Year 1992
Meaning and Grammar
Empirical Approaches to Language Typology 10 Editors Georg Bossong Bernard Comrie
Mouton de Gruyter Berlin · New York
Meaning and Grammar Cross-Linguistic Perspectives edited by Michel Kefer Johan van der Auwera
Mouton de Gruyter Berlin · New York
1992
Mouton de Gruyter (formerly Mouton, The Hague) is a Division of Walter de Gruyter & Co., Berlin.
© Printed on acid-free paper which falls within the guidelines of the A N S I to ensure permanence and durability.
Library of Congress Cataloging-in-Publication
Data
Meaning and Grammar : cross-linguistic perspectives / edited by Michel Kefer, Johan van der Auwera. p. cm. — (Empirical approaches to language typology ; 10) ISBN 3-11-012805-5 (alk. paper) 1. Universals (Linguistics) 2. Grammar, Comparative and general. 3. Semantics. I. Kefer, Michel. II. Auwera, Johan van der. III. Series. P204.G7 1991 91-45131 415 —dc20 CIP
Die Deutsche Bibliothek — Cataloging-in-Publication
Data
Meaning and Grammar : cross-linguistic perspectives / ed. by Michel Kefer ; Johan van der Auwera. — Berlin ; New York : Mouton de Gruyter, 1992 (Empirical approaches to language typology ; 10) ISBN 3-11-012805-5 NE: Kefer, Michel [Hrsg.]; G T
© Copyright 1992 by Walter de Gruyter & Co., D-1000 Berlin 30. All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Typesetting and Printing: Arthur Collignon GmbH, Berlin. Binding: Dieter Mikolai, Berlin. Printed in Germany.
Preface
This book arose in part from a symposium on Language Universals organized by the Linguistic Society of Belgium in December 1988. We selected some of the papers presented there, to which we added the papers by Dobrovol'skij, Haberland and Heltoft, Müller-Gotama, and Wierzbicka. Some of the less semantically oriented papers from the Symposium were published in an issue of the Belgian Journal of Linguistics.1 The first section of the volume starts off with a historical contribution by Bossong and then contrasts Joseph's ideas on the nature of linguistic universals with Haberland and Heltoft's programmatic claim that explanations of universals are necessarily pragmatic. The central part essentially deals with the syntax and semantics of the simple clause and especially its predicate. Thus, Stassen discusses how intransitive predicates are encoded; Wetzer and Hengeveld both deal with nonverbal predicates; Brus discusses voice alternations of transitive verbs; and Van de Walle double object constructions. Kañski deals with impersonal constructions, in a way that should interest both typologists and formal semanticists. Moreno presents a new typological parameter of "polarization". Müller-Gotama, finally, elaborates on Hawkins' so-called "Semantic Typology", distinguishing languages with semantically transparent constructions from languages with a high degree of grammaticalization. The four Dutch contributions (Stassen, Wetzer, Hengeveld, and Brus) are not only similar in focus but also in approach: they each use a diversified corpus of languages, make a clear distinction between independently defined semantic notions and their formal realizations, formulate typological (implicational or bidirectional) universals, and look for functional explanations. The third section is devoted to the lexicon and the meaning of grammatical morphemes. Wierzbicka presents some non-implicational substantive universals of the lexicon. Dobrovol'skij shows the possibility of making crosslinguistic sense of idioms. Gil, whose approach shares the characteristics of the Dutch contributions mentioned above, investigates "scopai quantifier" words such as English each and every, and formulates a dozen universals. Tobin pleads for the relevance of a notion of "semantic integrality" as the perception of one or more entities within a single continuous space. De Mulder's paper deals with the localist hypothesis of demonstratives, and the contribution by Vetters is a critical discussion of the distinction between
vi
Preface
foreground and background information, focusing on French tense-aspect morphemes. Of the organizations and people we are grateful to, we would like to single out the Belgian National Science Foundation for its financial help, Bernard Comrie for his editorial advice, and Marie-Louise Liebe-Harkort for a publisher's encouragement and patience.
Michel Kefer University of Liège
Johan van der Auwera University of Antwerp (UIA)
Note 1. U'niversals of language, edited by M. Kefer—J. van der Auwera. Brussels: Editions de l'Université de Bruxelles, 1989 ( = Belgian Journal of Linguistics 4). It contains the following papers: W. Abraham "Language Universals: The Chomskyan approach vs. Greenberg's typological approach"; G. Bossong "Morphemic marking of topic and focus"; B. Comrie "Translatability and language universals"; M. Dryer "Discourse-governed word order and word order typology"; R. Geluykens "The syntactization of interactional processes"; J. Greenberg "The internal and external syntax of numerical expressions"; R. Noske "Syllable geometry"; J.-O. Ostman "Testing iconicity"; H. Seiler "Iconicity in a functional perspective"; and W. Van Langendonck "Ergativity, markedness and prototypes".
Contents
Preface Contributors
ν ix
Part I General considerations Reflections on the history of the study of universals: the example of the partes orationis Georg Bossong
3
Universals, explanations and pragmatics Hartmut Η aberland and Lars Heltoft
17
Isomorphic, anisomorphic, and heteromorphic universals John E. Joseph
27
Part II Syntactic categories and constructions Towards a typology of voice Ineke Brus
45
Non-verbal predicability Kees Hengeveld
77
Impersonal constructions as a strategy for second-order predication Zbigniew Kanski
95
Polarization as a universal of linguistic organization Juan Carlos Moreno
123
Towards a semantic typology of language Fran% Müller-Gotama
137
A hierarchy of main predicate encoding Leon Stassen
179
vili
Contents
On explaining double object constructions Lieve Van de Walle
203
"Nouny" and "verby" adjectivals: a typology of predicative adjectival constructions Harrie Wetter
223
Part III Morphemes and lexical items Demonstratives and the localist hypothesis Walter De Mulder
265
Phraseological universalst theoretical and applied aspects Dmitrij Dobrovol'skij
279
Scopai quantifiers: some universals of lexical effability David Gil
303
Semantic integrality: a universal semiotic feature of language and perception Yishai Tobin
347
Foreground and background: Weinrich against Labov Carl Vetters
367
Lexical universals and universals of grammar Anna Wierybicka
383
Author index
417
Language index
423
Subject index
425
Contributors
Georg Bossong, University of Mannheim, Department of Romance Philology, Schloß, 6800 Mannheim 1, Federal Republic of Germany. Ineke Brus, University of Nijmegen, Department of General Linguistics and Dialectology, Erasmusplein 1, 6525 GG Nijmegen, The Netherlands. Walter De Mulder, University of Antwerp (UIA), Department of Romance Philology, Universiteitsplein 1, 2610 Wilrijk, Belgium. Dmitrij Dobrovol'skij, Linguistic University of Moscow, Ostoshenka 38, 119034 Moscow, USSR. David Gil, University of Haifa, Department of English, 31999 Haifa, Israel. Hartmut Haberland and Lars Heltoft, 260, 4000 Roskilde, Denmark.
Roskilde University Center, PO Box
Kees Hengeveld, University of Amsterdam, Department of Spanish, Spuistraat 134, 1012 VB Amsterdam, The Netherlands. John E. Joseph, University of Maryland, Department of French and Italian, College Park, MD 20742, USA. Zbigniew Kanski, University of Silesia, Institute of English and General Linguistics, Bando 10, 41-905 Sosnowiec, Poland. Juan Carlos Moreno, Autonomous University of Madrid, Faculty of Philosophy and Arts, Department of Linguistics, 28049 Madrid, Spain. Franta Müller-Gotama, California State University, Fullerton, Department of Linguistics, 800 N. State College Blvd., Fullerton, CA 92634, USA. Leon Stassen, University of Nijmegen, Department of General Linguistics and Dialectology, Erasmusplein 1, 6525 GG Nijmegen, The Netherlands. Yishai Tobin, Ben Gurion University of the Negev, Department of Foreign Literatures and Linguistics, PO Box 653, 84105 Be'er Sheva, Israel. Lieve Van de Walle, University of Antwerp (UIA), Department of Linguistics (GER), Universiteitsplein 1, 2610 Wilrijk, Belgium.
χ
Contributors
Carl Vetters, University of Antwerp (UIA), Department of Romance Philology, Universiteitsplein 1, 2610 Wilrijk, Belgium. Harrte Wetter, University of Nijmegen, Department of General Linguistics and Dialectology, Erasmusplein 1, 6525 GG Nijmegen, The Netherlands. Anna Wier^bicka, The Australian National University, Department of Linguistics, GPO Box 4, Camberra ACT 2601, Australia.
Part I General considerations
Reflections on the history of the study of universale: the example of the partes orationis Georg Bossong Vox diversa sonat, populorum est vox tamen una (Martialis as quoted by Beauzée)
Every man speaks his own language — there is only one human language, the language of mankind: this dialectic relationship between languages (langues) and language (langage) has often been formulated, for instance by Nicolas Beauzée (1767) quoting Marcus Valerius Martialis (see above), or by Joseph Vendryès (1923: 2 5 9 - 2 6 0 ) : Le langage est à la fois un et multiple; il est le même cbe\ tous les peuples et pourtant il se diversifie à l'infini dans tous les êtres qui parlent.... Chaque individu introduit dans le langage une certaine part d'innovation qui lui est propre. Il n'est donc pas si faux de prétendre qu'il y a autant de langages différents que d'individus. Mais il n'est pas faux de prétendre non plus qu'il n'existe qu'un langage humain, identique en son fond sous toutes les latitudes. This dialectic relationship of unity and diversity is fundamental for the linguistic sciences and for the philosophy of language. It is constitutive not only for present day linguistics, especially for linguistic typology in its relationship to universalist research, but also for linguistic thought in its historical development. In this short contribution the following point will be made. There are two different presuppositions for the scientific study of language universals, presuppositions which work independently of each other in the history of universalistic research: the first presupposition is the theoretical interest in language in general ( l a n g a g e ) instead of interest in particular languages (langues); the second presupposition is the concrete knowledge of linguistic diversity. With respect to the first of these two parameters, a dialectic alternation of periods oriented towards language and of periods oriented towards languages can be observed: the history of linguistic thought in Western Europe up to the early nineteenth century can be described, in its great lines, as a pendulum which swings from the universalistic to the particularistic approach, and back again (for details, see Bossong 1990).
4
Georg Bossong
The first great period was modistic scholasticism, characterized by a universalistic research into fundamental problems of syntax, semantics, and logic. Latin (the universalistic Latin of the Middle Ages which had but little in common with the language of Cicero and Virgil) served on the one hand as the medium of scientific thought; on the other hand it provided the examples which were to be submitted to logical and linguistic analysis. But Latin was not specifically focussed as an individual language: examples taken from this particular language stood for human language in general. We may call this period the first universalism in Western linguistics. This period was followed by Humanism and the Renaissance, when scholarly interest was centred upon the grammatical description of individual languages and their peaceful competition; the title of Du Bellay's Deffence et illustration de la langue françoise (1549) is paradigmatic for the whole period. The new national languages of Western Europe could now compete even with the classical languages, Greek and Latin, which were rediscovered, in their turn, as individual languages; Latin was no longer considered as the representative of the internal logic of human language in general, but as a vehicle of a great literary tradition with its inherent aesthetic values. This period may be called the first particularism. There followed a new period of universalistic research which had its peaks mainly in France and in England: the rationalistic thought of the seventeenth century laid the base for the linguistic research of the Enlightenment in the eighteenth century, a period when the genre of the grammaire générale flourished. Now it was mainly French which served as the new paradigm of human language in general. This period may be called the second universalism. From the early eighteenth century on, starting with Giambattista Vico, the pendulum swung back again to a new particularistic approach, the second particularism initiated by language philosophers such as Rousseau and Herder, and brought to its full development in the nineteenth century when the historical-comparative methodology came into being. With respect to the second of the aforementioned factors, the parameter of knowledge of linguistic diversity, there is no dialectic alternation of approaches, but a steady development: ideally, the increase of knowledge about the world's languages is a cumulative, linear process. In this respect, it does, in fact, make sense to speak of progress: the foundations upon which general statements about language rely become more and more solid since generalizations are made less and less from an insufficient empirical base. This is of course an open process which never comes to an end. Complete and perfect knowledge about the linguistic diversity of mankind is evidently a Utopian goal; nevertheless, it is also evident that the empirical knowledge
History: partes orationis
5
available today is incommensurably greater than the knowledge available in the days of Roger Bacon or Dante Alighieri. In the process of accumulation of knowledge in Western European linguistics, two main factors must be taken into consideration: the "discovery of the native tongue", as Karl Otto Apel (1963: passim) has put it; and the discovery of more distant languages by the multiple contacts Europeans had with non-European civilizations. We may call this self-discovery and discovery of the exotic, respectively. The process of self-discovery is intimately related with the emancipation from the authoritative model of Latin and the classical languages in general. From Dante's De vulgari eloquentia up to Gilliéron's dialect geography there is a steadily growing awareness of the intricacies of one's own native tongue and its internal variability. Spoken vernaculars are no longer considered as unworthy of scientific attention; not only Latin, but also French, Spanish, Italian, and even Lombardian, spoken Tuscan and Venetian have their own regularities, their own "grammar", a term which was formerly strictly limited to Latin and the Classical languages. The horizon of empirical knowledge grows internally insofar as insight into one's own language becomes deeper and richer. The process of discovery of the exotic depends immediately upon the European expansion over the world since the late Middle Ages. The first non-Indoeuropean languages with which European scholars came into closer contact were the Semitic tongues: Biblical Hebrew as one of the classical languages of the homo trilinguis·, and Arabic as a vehicle of a much admired and imitated civilization. Knowledge not only of the Hebrew and Arabic languages, but also of their respective grammatical theory and terminology can be detected, for instance, in the Spanish linguistic thought of the siglos de oro. Later on, the languages of the native Americas and of Asia became better and better known to European scholars. This process started with grammatical descriptions of languages such as Quechua, Tupi-Guaraní, and Japanese by Spanish and Portuguese Jesuits in the late sixteenth and early seventeenth century (see, e.g., Anchieta 1595, Rodriguez 1604, González Holguín 1607, Torres Rubio 1619, Montoya 1640 etc.). It continued throughout eighteenth century France when Chinese was held in high esteem as a kind of anti-world with respect to structures familiar from European languages, and it finally led to the first great language catalogues of Hervás, Pallas and Adèlung in the last decades of the eighteenth and the first decades of the nineteenth century, which were to form the base for the emerging new discipline of language typology.
6
Georg Bossong
In the development of linguistic thought, both these factors interact. Research on language universals is possible only if presuppositions from both these domains are made: there must be a truly universalistic approach, a true interest in human language in general, which transcends the interest in particular languages; and the empirical knowledge about linguistic diversity available at a given historical moment must be taken into account to as full an extent as humanly possible. This point will be illustrated here by a short discussion of the problem of the partes orationis, especially the verb-noun distinction, in the history of linguistic thought up to the early nineteenth century. Particular emphasis will be given to the historical moment when the newly awakened interest in the universalistic perspective was combined for the first time with a more than superficial knowledge of languages with radically "exotic" grammatical structures. This moment was the publication of Tommaso Campanella's Grammaticalia in 1638, a work which has received little attention from historians of linguistics, but which is of transcendental importance for the history of universals research. The starting point for the medieval discussion of the parts of speech is Priscianus. His own approach is formal and didactic. He does not make any universalist claim, but is exclusively oriented toward the grammar of Latin as an individual language. He naively postulates a simple equivalence of form and function: proprium est nominis substantiam et qualitatem significare ... nomen est pars orationis quae unicuique subjectorum corporum sen rerum communem vel propiam qualitatem distribuit. proprium est verbi actionem sive passionem significare ... verbum est pars orationis cum temporibus et modis, sine casu, agendi vel patiendi significativum. ['It is the property of the noun to denote substance and quality ... The noun is a part of speech which assigns a general or a specific quality to each underlying being or thing. / It is the property of the verb to denote actions or processes ... The verb is a part of speech with tenses and moods, without case, and with a mark for active and passive.'] (Priscianus, ca. 520, Institutiones grammaticae 2.4.18, 2.5.22, 8.1.1; apud Robins 1976: 63)
The modistic philosophers, of whom Thomas of Erfurt is one of the best representatives, have a genuine interest in linguistic universals. However, Latin still is the only empirical base (if it can be called "empirical": the approach is rather logico-deductive). The "grammatica", as Latin is called in the Middle Ages, is taken as a paradigm of human language in general. As for the partes orationis, an effort is made to define them not for Latin alone, but for human language in general. In this context, the notion of mode (modus significandt) represents a true epistemologica! progress. Nouns do not simply
History: partes orationis
7
denote substances, but they denote a part of reality in the mode of substance; correspondingly, verbs do not denote actions, but they denote some parts of extralinguistic reality in the mode of action. The naïve postulate of a one-toone-correspondence of form and content, which was still characteristic of late Classical Antiquity and the early Middle Ages, has now disappeared. There is an intermediate level between extralinguistic reality itself and its linguistic representation: the level of the modi significandi. Although they did not consider more than one natural language, namely Latin, the modistic philosophers aimed at human language in general: the form of this particular language served only as a paradigm of the form of the linguistic ability of man. The theoretical progress in comparison with Priscianus can be seen in the following quotations: nomen est pars orationis significans per modum entis vel determinatae apprehensionis. verbum est pars orationis significans per modum esse distantis a substantia. ['The noun is a part of speech which denotes something in the manner of entity, or in the manner of determinate apprehension. / The verb is a part of speech which denotes something in the manner of being, which is different from that of substance.'] (Thomasius de Erefordia, ca. 1300, Grammatica speculativa, sive Novi modi significandi, ed. 1902, § 25, 117; apud Robins 1976: 85)
The linguistic particularism of the Renaissance is well represented by Pietro Bembo, the author of the first comprehensive grammar of Italian written in the form of a philosophical dialogue: the famous Prose della volgar lingua, printed in 1525. This treatise shows clearly the exclusive interest of Renaissance linguistics in individual languages. As far as knowledge is concerned, there is a true progress in comparison to medieval grammatical theory: Italian is recognized as a literary language with its own autonomous grammar; many details of its morphology and syntax are treated here for the first time. What we have called here "self-discovery" has led to a comprehensive linguistic description which is a valuable contribution to our knowledge of standard literary Italian even from the perspective of modern linguistics. But there is no interest whatsoever in language in general, no attempt to define the notions of "noun" and "verb" independently from the idiosyncratic structure of Italian. When treating these parts of speech, Bembo immediately proceeds to the description of formal details and of literary usage. Although the empirical base for linguistic research was considerably enlarged in comparison with the Latin Middle Ages, universale of human language are not at all in focus. Compare the following quotations which show how Bembo (1525 [1955]: 100, 134) introduces the partes orationis. E per incominciar dal nome, dico che, si come nella maggior parte delle altre lingue dell' Italia, cosi eziandio in quella città mia, i nomi in alcuna delle vocali terminano e finiscono
8
Georg Bossong
sempre; sì come naturalmente fanno ancora tutte la toscane fuori. Ma passisi a dire del verbo, nel quale la licenza de' poeti lingua v'hanno più di malagevolezza portata, che mestier parole far chiaro; il qual verbo, tutto che in quattro maniere lingua come gli è nella latina.
voci, da alcune pochissime in e la libertà medesima della non fa a doverlovi in poche si veda esser così nella nostra
['To begin with the noun, I say that, as in most dialects of Italy, in my town, too, nouns always end in a vowel, just like all other Tuscan words, with only a handful of exceptions. / Now let us talk about the verb, where poetic licence and the freedom of language itself have produced more infelicitous phenomena than could be made clear in a few words; which verb can clearly be seen to have the same four conjugation classes as in Latin.']
The second universalism in Western European linguistic thought starts with a famous grammatical treatise on Latin syntax: the Minerva·, seu de causis linguae Latinae of Sanctius ( = Francisco Sánchez de las Brozas), professor of classical languages at the university of Salamanca. This work appeared in 1587; it had a considerable influence all over Europe. Sanctius' approach is decidedly rationalistic. He attempts to reduce the apparent irregularities of Latin syntax, especially ellipsis and irregular word order, to a few underlying structures which are logically comprehensible, and therefore universal. His perspective is not yet fully universalistic: his main concern is still a better understanding of Latin grammar. But by postulating a "deeper" layer of syntax governed by logical rules, a layer which is hidden under a "surface" characterized by an often intriguing illogicality, he arrives at a level of abstractness where differences between individual languages begin to lose their importance. His claim that everything in "deep" syntax must be logically analyzable leads necessarily to a universalistic perspective, since reason is necessarily universal. As far as the problem of parts of speech is concerned, Sanctius excludes every category which does not stand up to logical scrutiny. In 1492, the first grammarian of Spanish, Antonio de Nebrija, had brought the number of parts of speech from eight to ten, in order to be able to treat some peculiarities of Spanish grammar which have no counterpart in Latin. Now, Sanctius reduces this number to three: noun, verb, and particle. Only these three seem to him logically necessary and essential for every human language. His definitions are formalized insofar as he uses the classical figurae of the genus proximum and the differentia specifica-, however, on the semantic side, these definitions are less sophisticated than what can be found in the modistic scholasticism. Moreover, it is of great interest to note that he explicitly refers to the Semitic grammatical tradition: in Hebrew and Arabic grammar, only the aforementioned three parts of speech are distinguished. In spite of the disastrous activities of the Holy Inquisition, some knowledge
History: partes oratiotiis
9
not only of the Hebrew and Arabic languages, but also of the corresponding linguistic theories had survived in Salamanca, which was then a main centre of Spanish intellectual life. In Sanctius' definitions (1587 [1664]: 9, 17, 38) the following points can be observed: rationalistic and universalistic reduction of the parts of speech to the logically and functionally necessary minimum; influence of the "exotic" model of the Semitic languages; and scholastic formalism: cum igitur oratio sit finis grammatici, excutiamus ex quibus haec oratio possit constituí, ita ut nihil sit quod per orationem non possimus enunciare, sunt autem haec tria, nomen, verbum, particulae: nam apud Hebraeos tres sunt partes orationis, nomen, verbum, et dictio consignificans. Arabes quoque has tantum tres orationis partes habent: Pheâl, verbum; Ismî, nomen; Herph, dictionem. { = f f l ¡ ?ismj harf] nominis /initio est vocem numeri casualis cum genere; quae definitio propria est perfecta: vox enim seu dictio, seu pars orationis genus proximum est; et casus est specialis differentia. verbum est vox particeps numeri personalis cum tempore, haec definitio vera est, et perfecta, reliquae omnes Grammaticorum ineptae. ['Since it is the aim of the grammarian [to clarify the constitution of] speech, let us examine the elements of which it can be constituted, so that there be nothing which we cannot express by speech. These elements are three: noun, verb, and particle; because in Hebrew there are three parts of speech, noun, verb, and co-significant word [Hebrew tevat finyan, "relational word"]. The Arabs, too, have only three parts of speech: verb, noun, and particle. / The noun is defined as a word with case, number, and gender. This definition is adequate and perfect: "word", or "part of speech" is the genus, and "case" is the specific difference. / The verb is defined as a word with person, number, and tense. This is the true and perfect definition, and all other definitions of the grammarians are worthless.'] Like Sanctius, Gondola Correas was a professor of the classical languages at the university of Salamanca. Today he is k n o w n mainly as the author of a comprehensive Spanish grammar where he tries to explain rather than simply describe the rules of his native tongue. He also tried to introduce into Spanish a reformed orthography. Although he failed in this attempt, his work is an important milestone in the history of universals research: he was the first linguist w h o applied the rationalistic methodology developed by Sanctius for Latin to the study of a Romance vernacular. As for the question of the partes orationis, he closely follows his predecessor Sanctius. He also claims that the number of parts of speech should be reduced to three, and when substantiating his claim, he refers not only to Hebrew, Arabic and Aramaic {caldeó), but also to other oriental and African languages, and even to "all languages of the world". This latter formulation has of course no real foundation whatsoever since at that historical moment the empirical base for such a claim was far
10
Georg Bossong
too weak. It should also be noted that the definitions of the parts of speech themselves are rather naïve. Nevertheless, it is remarkable for a work published in 1625 that its author starts from the idea of a "universal grammar", and that the postulate of not more than three parts of speech is explicitly stated as a language universal. la Gramatica en lo general es común à todas las lenguas, i una mesma en todas, allegóme ... a la rra^on que en esto es tan patente ... examine cada uno las palavras de qualquiera lengua, i verá que no halla ninguna que no sea o nonbre o verbo o partícula. Conocerá el nonbre en que tiene singular i plural, i demás deso en Latin i Griego en que se declina por casos: el verbo en que se conxuga i varia por personas i tienpos de singular i plural en todas las lenguas: i la partícula en que es invariable en todas. ... las palavras de una lengua sea cualquiera son tres, porque todos los vocablos son en tres maneras, i se dividen en tres partes ò montones, i se rredu^en à estos tres xeneros dichos nonbre, verbo, i partícula, como está llano i asentado en Hebreo, Caldeo i Aravigo, i todas las otras lenguas Orientales i de Africa, i todas las del Mundo convienen en esto. ['Grammar in general is common to all languages; it is one and the same in all of them. I I conform myself to reason, which is self-evident in this case. Let anyone analyze the words in any language, and he will see that no word can be found which is not noun, verb, or particle. He will recognize the noun by its having singular and plural, and moreover in Latin and Greek case declension; he will recognize the verb by its being conjugated and modified according to person, tense, and number in all languages; and he will recognize the particle by its being invariable in all languages ... The words of any language whatsoever are of three classes, because all words are of three kinds, and they are divided into three parts, and they are reduced to these three aforementioned classes noun, verb, and particle: as is clearly and firmly established in Hebrew, Aramaic, and Arabic, and in all other Oriental and African languages, and all those of the world agree in this point.'] (Gonzalo Correas Iñigo, 1625, Arte de la lengua española castellana, ed. 1954: 9, 132)
To the best of my knowledge, Tommaso Campanella is the first author in the history of European linguistics who based his discussion of the parts of speech on a serious comparison of European and East-Asian languages. This famous poet and philosopher, who spent many years of his life in the jails of the Holy Inquisition in Naples, is little known as a linguist. Nevertheless, his place in the history of our discipline can hardly be overestimated. His work on grammar is small, but highly important. Significantly enough, it appeared one year after Descartes' Discours de la méthode (1637). It had little influence, perhaps because Campanella buried it in a large philosophical treatise called Philosophia rationalis. If there is a work of the seventeenth century to be rediscovered by present-day linguistics, it is Campanella's Grammaticalia. Campanella is the first linguist who defines himself explicitly as a "grammairien philosophe", as the theoreticians of the following century would have called it. He claims to write a philosophical, and therefore
History: partes orationis
11
scientific, grammar which has to be sharply distinguished from what he calls civil, or practical, grammar. In this philosophical grammar, Campanella attempts to establish an axiomatic system which enables him to define the partes orationis in a purely rational way. His approach is not inductive, he does not start from the observation of the facts of Greek and Latin, as all linguists preceding him had done. He deduces the properties of the parts of speech from a set of general categories which are postulated to exist independently of the structural properties of individual languages. Thus he is able to eliminate all idiosyncratic, and therefore accidental, features from the definition of the parts of speech. Moreover, it is exactly this deductive approach which enables him to take advantage of the empirical progress which had been accomplished in the meantime. In his prison, he had the opportunity to read the reports of a Jesuit missionary from East-Asia (Borri 1631). He discovered that in languages of the type which Wilhelm von Humboldt would call the "isolating" one, the partes orationis cannot be defined in the traditional way, simply because there are no morphological categories such as number, gender, and case declension, or personal, temporal, and modal conjugation. Consequently, Campanella distinguishes essential from accidental features in the definition of the parts of speech: the status of certain language specific morphological categories known from Greek and Latin school grammar can now be relativized; since these categories are lacking in other languages, they cannot be taken as language universale any longer. In Campanula's Grammaticalia (1638 [1954]) the universalistic perspective is, for the first time, empirically based upon a serious comparison of the inflectional type familiar to Europeans with the totally different isolating type found in East-Asian languages. The fully developed theoretical approach of the second universalism is combined not only with a thorough empirical knowledge of the classical languages, the Romance vernaculars, and the Semitic linguistic type, but also with the discovery of the extremely "exotic" type of Chinese and Vietnamese. Empirically and theoretically, this work is the first one which can be qualified as universalistic in the full sense of the word. Campanula's approach can be illustrated by the following quotations. Note that the language of Cochinchina is Annamite, or Vietnamese, a language which represents the "isolating" type even more markedly than Classical Chinese. It should also be noted that Campanella considers the simplicity and clarity of the isolating type as a model for an artificially created universal language (nova lingua)·. Duplex grammatica: alia civilis, alia philosophica ... civilis, peritia est, non scientia: constat enim ex autoritate usuque clarorum scriptorum ... philosophica vero ratione
12
Georg Bossong constat; et haec scientiam olet. Est enim investigantis intellectus et notantis investigata, copulantisque et dirimentis res, prout in natura reperiuntur, methodus. In lingua Graeca et casus et articuli clarificant essentiarum actualitates, in Latina soli casus, in vulgari Itala, Francica, Hispana, Hebraea et Arabica soli articuli: propterea diximus accidentia casum et declinationem esse nominibus. non enim ex hoc est verbum, quod habet modos et tempora, sed ex hoc, quod actum jluentem ab essentia ... in lingua Chinensium et Cocinchinensium verba non declinantur personis, nec temporibus variantur, sed notulis: ergo accidunt haec verbo, non essentiant verbum. quemadmodum in lingua Hebraica, Arabica, Hispana et Gallicana non dantur casus nominum, sed loco ips[o]rum ponuntur articuli, sic etiam in lingua Cocinchinorum et aliarum orientalium non dantur declinationes verborum aptandae personis, neque temporum varietates, neque varietates verborum aptandae temporibus, et ideo omne verbum est instar impersonalis vel infinitivi. Distinctio autem fit per adverbia temporalia, ut sì dicerem « nunc amo, imposterum amo, ante amo », ita quod non dantur concordantiae temporum, nec personarum, neque casuum, sed particulae adverbiales et agnominales totam orationem construunt et distinguunt, mirifica brevitate ac dicendi facilitate. Quapropter, qui novam linguam invenire studet, haec notabit. ['There are two kinds of grammar: civil, and philosophical. Civil grammar is a mere skill, not a science. Its essence lies in the authority and usage of famous authors. Philosophical grammar, on the other hand, is based on reason; therefore, it has a scientific smell. It is a method of the intellect which investigates and notes down what it has investigated; which establishes relationships and distinctions between things, such as they are found in nature. / In Greek, case endings and articles mark the actual states of the entities, in Latin, [the same function is fulfilled] only by case endings, in Italian, French, Spanish, Hebrew, and Arabic only by articles; therefore, we have said that case and declension of the noun are accidental. / The verb is not a verb because it has mood and tense, but because it denotes the act which emanates from the essence. In Chinese and Vietnamese, verbs have neither personal nor temporal conjugation, only particles; therefore, all these are accidental, not essential features of the verb. I Just as in Hebrew, Arabic, Spanish, and French, case is not expressed by endings, but articles are used instead, in the same way in Vietnamese and other Oriental languages there is no personal conjugation, no tense, and no temporal conjugation; this way, each verb is like an impersonal verb or an infinitive. The distinction is made by temporal adverbs, as if I would say "now I-love, later I-love, before I-love", so that there is no agreement of tense, person, or case, but only adverbial and adnominal particles which build up the whole sentence and mark all distinctions, with astonishing conciseness and ease. Therefore, whoever wants to invent a new [artificial] language should note this.'] (Tommaso Campanella, 1638, Philosophia rationalis. Grammaticalia, ed. 1954: 438, 498 f, 528, 699)
From the middle of the seventeenth century on, the tradition of general grammar is firmly established. The famous Grammaire générale et raisonnée (Arnauld—Lancelot 1660) written in the monastery of Port-Royal initiates a tradition which reaches until the first decades of the nineteenth century. The
History: partes orationis
13
mainstream of the second universalism is located in France. With respect to the empirical base of research in language universale, the work of Antoine Arnauld and Claude Lancelot does not represent a true progress; the base of typological comparison is constituted by the contrast between Latin and the Romance languages, especially French. More distant languages are not in focus. One has to wait until Beauzée's Grammaire générale of 1767 before the data of "exotic" languages are systematically compared with those found in Europe. The authors of the Port-Royal grammar define the parts of speech within the framework of a general linguistic theory: language is considered as universal insofar as it reflects the universal properties of human reason. The verb, for instance, has the basic function of marking the assertion which is assumed to be one of the most basic activities of the human mind. The definition as such does not depend upon the idiosyncratic properties of individual languages, such as inflection for person and tense. It depends on the universals of the human mind; the basic features of noun and verb are directly derived from the basic activities of the mind: to conceive entities, i.e., the objects of our thoughts; and to judge about these entities, i.e., to make statements and utterances. The fundamental categories of every conceivable human language are but the reflex of the fundamental categories of thought: On ne peut bien comprendre les diverses sortes de signification qui sont enfermées dans les mots, qu'on n'ait bien compris auparavant ce qui se passe dans nos pensées, puisque les mots n'ont été inventés que pour les faire connaître. Il y a cette différence entre les choses et les substances, et la manière des choses ou des accidents, que les substances subsistent par elles-mêmes, au lieu que les accidents ne sont que par les substances. C'est ce qui fait la principale différence entre les mots qui signifient les objets des pensées: car ceux qui signifient les substances ont été appelés noms substantifs; et ceux qui signifient les accidents, en marquant le sujet auquel ces accidents conviennent, noms adjectifs. Les hommes n'ont pas eu moins de besoin d'inventer des mots qui marquassent /'affirmation, qui est la principale manière de notre pensée, que d'en inventer qui marquassent les objets de notre pensée. Et c'est proprement ce que c'est que le verbe, un mot dont le principal usage est de signifier l'affirmation, c'est-à-dire, de marquer que le discours où ce mot est employé, est le discours d'un homme qui ne conçoit pas seulement les choses, mais qui en juge et qui les affirme ... à ne considérer simplement que ce qui est essentiel au verbe, sa seule vraie définition est: vox significane affirmationem; un mot qui signifie l'affirmation. (Arnauld —Lancelot 1660 [1969]: 55) Although Arnauld and Lancelot do not take into consideration linguistic structures other than Latin, Greek, French, and occasionally Hebrew, their definition of the partes orationis are decidedly universalistic. From a purely theoretical point of view, they arrive at the postulate of a language where
14
Georg Bossong
verbs lack all the accidental properties usually found in European languages, such as tense, person and so on, but where nonetheless they constitute a distinct word class. Although the authors of the Port-Royal grammar had no direct empirical evidence for such a language, they constructed it as a theoretical possibility in order to show that all these properties are in fact accidental. In contrast to Campanella, they had no idea about the structure of East-Asian languages; but they demonstrated that assertiveness is the only essential property of the verb by postulating the structural features of the isolating language type all by themselves: Si on avait inventé un mot, comme serait est, qui marquât toujours l'affirmation, sans avoir aucune difference ni de personne ni de temps, de sorte que la diversité des personnes se marquât seulement par les noms et les pronoms, et la diversité des temps par les adverbes, il ne laisserait pas d'être un vrai verbe. (Antoine Arnauld — Claude Lancelot, 1660, Grammaire générale et raisonnée, ed. 1969: 66, 70 — 71)
A word like the Classical Chinese sentence final particle j è would correspond exactly to this theoretical postulate. Let us conclude with a quotation of Wilhelm von Humboldt. H e continues
the tradition of logical and universal grammar established in French rationalism and Enlightenment, and he reformulates the argument of the Grammaire générale et raisonnée in the terms of German romanticism. It goes without
saying that in Humboldt's work, the theoretical interest in language universale is definitely based upon a solid empirical foundation: his statements rely upon deepened research in as many genetically, areally and typologically unrelated languages as could possibly be expected in the first decades of the nineteenth century. Here is his definition of the verb as the basic part of speech in all human languages: Das Verbum unterscheidet sich vom Nomen und den anderen, möglicherweise im einfachen Sat^e vorkommenden Redetheilen mit schneidender Bestimmtheit dadurch, dass ihm allein der Act des synthetischen Setzens als grammatische Function beigegeben ist. ... Alle übrigen Wörter des Satzes sind gleichsam todt daliegender, ^u verbindender Stoff, das Verbum allein ist der, Leben enthaltende und Leben verbreitende Mittelpunkt. Durch ein und denselben synthetischen Act knüpft es durch das Seyn das Praedicat mit dem Subjecte zusammen. ... Der Gedanke, wenn man sich so sinnlich ausdrücken darf, verlässt durch das Verbum seine innre Wohnstätte und tritt in die Wirklichkeit über.
It is also Wilhelm von Humboldt who has formulated the ultimate goal of universalistic research with unsurpassable clarity: Da die Naturanlage %ur Sprache eine allgemeine des Menschen ist und Alle den Schlüssel %um Verständniss aller Sprachen in sich tragen müssen, so folgt von selbst, dass die Form aller Sprachen sich im Wesentlichen gleich seyn und immer den allgemeinen Zweck erreichen muss. Die Verschiedenheit kann nur in den Mitteln und nur innerhalb der Grän^en
History: partes orationis
15
liegen, welche die Erreichung des Zweckes verstattet. (Wilhelm von Humboldt, ca. 1832, Ueber die Verschiedenheit des menschlichen Sprachbaues und ihren Einfluss auf die geistige Entwicklung des Menschengeschlechts, ed. 1963: 608 f., 651) If w e w a n t t o learn f r o m h i s t o r y , the c o n c l u s i o n that can be d r a w n is simple: empirical research w i t h o u t t h e o r y is blind; t h e o r y w i t h o u t an empirical base is e m p t y . D e t a i l e d k n o w l e d g e o f t y p o l o g i c a l d i v e r s i t y is a necessary p r e r e q u i s i t e o f universalistic t h e o r y ; and o n the o t h e r hand,
typological
research o n l y m a k e s sense if it aims at p o s t u l a t i n g u n i v e r s a l s o f
human
l a n g u a g e . L i n g u i s t i c s is o r i e n t e d t o w a r d s a better u n d e r s t a n d i n g o f
language·,
b u t the o n l y w a y t o c o m e closer t o this goal passes t h r o u g h a d e e p e n e d knowledge of
languages.
References A. Early
works
Adelung, Johan-Christoph 1806 ff. Mitbridates, oder allgemeine Sprachenkunde. Bearbeitet von f . f . Vater. Berlin (ed. Hildesheim: Olms, 1970). Anchieta, José de 1595 Arte de Gramática da Lingua mais usada na Costa de Brasil. Coimbra (edited by Julio Platzmann, Leipzig, 1874). Arnauld, Antoine—Claude Lancelot 1660 Grammaire générale et raisonnée contenant les fondemens de l'art de parler expliqués d'une manière claire et naturelle: les raisons de ce qui est commun à toutes les langues, et des principales différences qui s'y rencontrent, et plusieurs remarques nouvelles sur la langue françoise. Paris (edited by Michel Foucault, Paris: Paulet, 1969). Beauzée, Nicolas 1767 Grammaire générale ou exposition raisonnée des éléments nécessaires du langage, pour servir de fondement à l'étude de toutes les langues. Paris (edited by Β. E. Bartlett, Stuttgart-Bad Cannstatt: Frommann-Holzboog, 1974). Bembo, Pietro 1525 Prose della volgar lingua. Venezia (edited by Mario Marti, Padova: Liviana, 1955). Borri, Cristoforo 1631 Relatione della nuova missione del Padre della Compagnia di Giesù al Regno della Cocincina. Roma. Campanella, Tommaso 1638 Philosophiae rationalispars prima, continens Grammaticalium libros tres. Paris (edited by Luigi Firpo, Milano: Mondadori, 1954). Correas Iñigo, Gonzalo 1625 Arte de la lengua española castellana. Salamanca (edited by Emilio Alarcos Garcia, Madrid: Concejo Superior de Investigaciones Científicas [ = RFE Añejo LVI], 1954).
16
Georg Bossong
González Holguín, P. Diego 1607 Gramática y arte nueva de la lengua general de todo el Perú llamada lenguage qquichua o lenga del Inca. Lima (edited by R. Porras Barrenechea, Lima 1952). Hervás y Panduro, Lorenzo n. d. Catálogo de las lenguas de las naciones conocidas, y numeración, división, y clases de éstas según la diversidad de sus idiomas y dialectos. Madrid. Humboldt, Wilhelm von 1830 — 1835 Ueher die Verschiedenheit des menschlichen Sprachbaues und ihren Einfluss auf die geistige Entwicklung des Menschengeschlechts. MS. Berlin (edited by Andreas Flitner & Klaus Giel (Werke in fünf Bänden, Band III), Darmstadt: Wissenschaftliche Buchgesellschaft, 1963, 3 6 8 - 7 5 6 ) . Montoya, Antonio Ruiz de 1640 Arte, bocabulario, Tesoro y catecismo de la lengua guaraní. Madrid (edited by Julio Platzmann, Leipzig, 1876). Pallas, Peter Simon 1786 — 1789 Linguarum totius orbis vocabularia comparativa. Sankt Petersburg (ed. Hamburg: Helmut Buske, 1976). Priscianus Caesariensis before 526 Institutio de arte grammatica. Constantinople (edited by M. Hertz [ = Grammatici Latini 2 & 3], Leipzig 1855-1859; reprint Hildesheim: Olms, 1961). Rodriguez, Joâo 1604 Arte da lingoa de fapam, composta pello padre foam RodrigueNagasaki. Sanctius, Franciscus (Francisco Sánchez de las Brozas) 1587 Minerva: seu de causis linguae Latinae. Salamanca (edited by G. Scioppius, Amsterdam: Judocus Pluymer, 1664). Thomasius de Erfordia ca. 1300 Grammatica speculativa (edited by Geoffrey L. Bursill-Hall, London: Longman, 1972). Torres Rubio, Diego de 1619 Arte de la lengua quichua. Lima. Β. Contemporary works Apel, Karl Otto 1963 Bossong, Georg 1990
Die Idee der Sprache in der Tradition des Humanismus von Dante bis Vico. Bonn: Bouvier.
Sprachwissenschaft und Sprachphilosophie in der Romania. Von den Anfängen bis August Wilhelm Schlegel. Tübingen: Narr. Robins, Robert Henry 1976 Brève histoire de la linguistique. De Platon à Chomsky, translated by Maurice Borell. Paris: Editions du Seuil. Vendryès, Joseph 1923 Le langage. Introduction linguistique à l'histoire. Paris: La Renaissance du Livre (edited by Paul Chalus, Paris: Albin Michel, 1968).
Universals, explanations and pragmatics1 Hartmut Haberland and Lars Heltoft
Martin Joos suggested that languages differ in all conceivable ways and to all possible degrees (Philippaki-Warburton 1981: 181; Joos 1957: 96). The universaliste' claim is that they do not but rather that they vary "... only within certain, not necessarily logically determined or obvious limits" (Coulmas et al. 1983: 1235). The linguist's task then is "to state those limits" (ibid.). This is correct but it is only half the task. The other half is to explain those limits. Not all linguists who study universale would subscribe to the latter claim. There are different concepts of universality in natural language among linguists and the emphasis on explanation will vary depending on what concept one adheres to. The Greenberg type of universal is based on empirical generalization (Greenberg 1966). There are purely statistical universals (e.g., 90% of the languages recorded have the feature X), and there are implicational universals (if a language has X, it will also have Z). Neither of these are universals in any strict sense (features necessarily part of any human language). Statistical "universals" are normally tendencies and since even a 100% score may be accidental, they are not universals in the strict sense. Implicational universals do not express necessary features of human language either but are generalizations about language types. In Greenberg's concept of implicational universals, universality is mixed up with the problem of typological classification (cf. Lehmann 1980). This is different from the second concept of universality, the Chomsky type. Here the very task of linguistics does not lie in the correct sampling of data or their classification. Rather, empirical data have to be explained by highly abstract, universal restrictions on possible syntactic structures. In the self-understanding of that school, abstract principles are thought of as analogues to natural laws in physics, and grammar is conceived of as a nomothetic natural science. These principles are not necessarily self-evident — in a sense, they can be justified because they hold across a number of different languages in spite of their surprising specificity. "Holding" means here that they make it possible to predict grammatical phenomena in languages not yet investigated. As long as "explanation" is understood as "predictive potential", this
18
Hartmut Haberland and Lars
Heltoft
is what the principles derive their explanatory force from. Thus, the explanans is (along with the data) the grammatical principles — they are always syntactic in nature — and the explananda are both individual grammars and the range of variation between languages. The latter is also called the range of possible values for grammatical parameters. No explanation is offered for the grammatical principles themselves; they are only justified by their predictive power. There is a third concept of universality which goes beyond the type of explanations found in Chomskyan grammars. In this view, universale are conceptually necessary elements of human language. From this vantage point universale are not, and could not be, empirical generalizations. Rather, they are principles of rationality. Examples of this view are to be found in Karl Bühler's Sprachtheorie (1934) and in Jürgen Habermas' Universal pragmatics (1969, 1971). Grice's maxims of conversation (1975) may be interpreted as similar principles of rationality in communicative behaviour. Such principles are found through conceptual analysis, and this again means that they must be substantiated not via empirical generalization, but through reflexive analyses in the sense of Strawson (1959). Universals are features of human language that cannot be "thought away", i. e., they resist any attempt to deny their validity, in that the speaker denying them would eo ipso presuppose their validity. Explaining such norms and universal distinctions means showing their inevitability. We think that proper explanations of universal features of human languages can only be derived from universale of the third type. By "proper explanation" we aim at an explanation that does not require an explanation itself. Neither should it exclusively be justified by its predictive power: it should make sense by itself and not just in connection with what it is to explain. We do not want to claim that pragmatic principles are the core of linguistics. As principles of rationality they remain relatively uninteresting, precisely because of their highly general and self-evident nature. Neither do they tell us anything about particular languages. But the value of the pragmatic principles lies precisely in the light they enable us to throw on empirical phenomena (pragmatic and other) in particular languages. Seemingly ad-hoc grammatical restrictions may very well make sense once they are related to pragmatic principles. Take existential sentences and the so-called indefiniteness restriction as an example. Lars Heltoft has argued elsewhere (1987) that this restriction should be committed to the waste-paper basket. Semantically, there are two types of existential sentences: categorials and presentati ves. Categorial existentials classify an object NP as a member of a class or set. 2 They occur freely with both
Universals, explanations
and pragmatics
19
definite and indefinite NPs. (The following are authentic Danish examples with definite NPs, taken from Hansen 1967.) (1) (2)
Der er for Exempel Koleraen — den medferer Krampeanfald. "There is for example [the] 3 cholera — it causes convulsions." J eg har sä travlt. Der er stuerne, der skal holdes i orden, og bernene, der skal passes. "I am so busy. There are the rooms which must be kept in order, and the children who must be looked after."
The set or class will normally have been mentioned in the context, or at least will be recoverable from it. In (1) the class may be thought of as the class of terminal infectional diseases, in (2) the set is explicitly mentioned as the set of things that keep the speaker busy. Presentative existentials describe dynamic or static spatio-temporal situations. They are more frequent than categorial existentials, and actual occurrences of presentatives very often disallow definite NPs. The following are classical. (3) (4)
Der sad en fugl pä taget. "There was a bird sitting on the roof." *Der sad fuglen pa taget. "There was the bird sitting on the roof."
Presentatives introduce a situation and in order to introduce a situation an informative and hence indefinite NP is necessary. An utterance cannot both introduce a situation and identify a presupposed (definite) NP as one of the entities of that situation. In categorials, there are no such restrictions, and crucially, presentatives may occur with definite NPs if they can be interpreted as categorials. (5) (6)
I œsken er der PENGENE OG DE UBETALTE REGNINGER. "In the box there is THE MONEY A N D THE UNPAID BILLS." Nu sad der KUN STORKEHUNNEN tilbage pä taget. "Now there was ONLY THE FEMALE STORK left sitting on the rooftop."
These examples are reinterpretations of presentatives as categorials. They are presentatives insofar as they describe a situation, but they are categorials insofar as they specify the members of sets (of the set of things in the box, of the set of living beings sitting on the roof). The reinterpretation results in a meaningful (informative) focus in each example. No syntactic principles or restrictions are necessary to account for this, neither indefiniteness restrictions nor binding theories. Definite NPs are
20
Hartmut Haberland and Lars Heltoft
allowed whenever the result is an acceptable information structure. The pragmatic principle behind the acceptable message structure is a version of the Gricean maxim of informativity: "Sensible messages must be informative", combined with a universal about information structures and definite NPs. The use of definite NPs presupposes that the speaker assumes that their referents have already been identified by the hearer. Hence in information structures, referents of definite NPs cannot be focussed (new). Rather, the information centre is on the relation of the definite NP focus to the rest of the message. It is related to some given background. 4 In presentatives, however, there is no candidate for the background (since they are subjectless), unless the verb is able to take on this role (which implies interpretation as a categorial). An example like (4), then, is deviant because it contains two inconsistent information structures: one in which the verb is background (the definite NP insists on this), and one in which the whole predicate (VP) is informative. The point: those sentences that are blocked are those that are pragmatically deviant; those allowed are those that comply with the pragmatic principles. To take another example: one of the better-known parameters in Government and Binding theory is the pro-drop parameter which allows complete sentences without overt subject in languages like Italian (cf. (7 a, b)) but not in German (cf. 8 a, b)): (7) a. b. (8) a. b.
Claudia mangia una ρί^χα. "Claudia is eating a pizza." Mangia una ρί^χα. "[She] is eating a pizza." Claudia ißt eine Ρί^χα. "Claudia is eating a pizza." *Ißt eine Ρί^χα. "Is eating a pizza."
(7 b) and (8 b) have an empty NP of the type pro in Subject position (so they are properly pro mangia una ρί^χα and pro ißt eine Ρί^χα). In German, a nonpro-àtop language, this blocks the derivation of *(8b), 5 while in Italian, pro can be dropped, which results in (7 b) which is grammatical. Now it is interesting that the term pro-drop as it occurs in this framework makes in itself no reference to subjects at all. Nevertheless it is usually understood that it is subject pros that are dropped. The standard explanation in Government and Binding theory is roughly the following: an abstract principle, the Empty Category Principle (ECP), states that empty categories
Universals, explanations and pragmatics
21
(like pro) have to be properly governed. In the case of pro subjects, a candidate for the proper governor is the element AGR in (9). (9)
S COMP
S
Now only in pro-drop languages with a rich verbal morphology does AGR contain enough features for number, person, and, possibly, gender to guarantee that pro is interpretable. That pro usually does not occur as an object NP is explained by the lack of object agreement features in the verb in most languages. 6 This standard argumentation makes use both of a universal principle (the ECP) and empirical observations about the "richness" of agreement phenomena. It is interesting that the latter (which are empirically testable) are the weak part of the argument. The claim that all pro-drop languages should have a rich AGR system does not hold water: Japanese lacks verbal agreement almost totally and is still a language that can drop not only ^»ro-Subjects but apparently almost anything else. We want to suggest an alternative analysis for consideration, an analysis which does not refer to syntactic or morphological features at all. This alternative explanation draws on a hierarchy of topic-worthiness. What has the semantic role of agent is more topic-worthy than what has the semantic role of a theme. 7 It is the agent which most typically receives the status of a topic (or topic expression) and subject. On the other hand, what refers to the topics is the most given part of the sentence and is therefore easily omitted. So it makes sense that subjects are easier to get by without. The apparent weakness of this pragmatic approach is that it only answers the question why pro-àtoç> is possible and why it makes sense that subject pros are dropped rather than others. No attempt is made to tell us why some languages are pro-drop and why others are not. This kind of question is sometimes tempting to ask. Why do some languages have present progressives? We can explain why perfective present tenses do not occur and why a perfective vs. imperfective opposition between two future tenses is highly
22
Hartmut Haberland and Lars
Heltojt
marked, but why do the languages that have this opposition have it? We have to realize two things here. First, pragmatic principles will not help us here, since they are languageindependent. 8 Therefore, they cannot explain why languages are different, i. e., why, within the range of possible options compatible with the pragmatic principles, one language manifests opposition A vs. B, while another language manifests A' vs. C. The pragmatic principles only delimit the range of possible options. Second, explanations based on other concepts of universale do not help us either since nothing prevents them from being circular. In our sense of "explain" there is nothing to explain here. One qualification might be appropriate at this place. Implicational univers a l may of course be related to pragmatic maxims and functional tendencies. Such examples are abundant in the literature (e.g., Comrie 1981; Hawkins 1983; Kefer 1986). If claimed to be universal, they might be thought to have the structure: (10)
Necessarily (if A, then B)
where the operator "necessarily" is substantiated through pragmatic considerations. Are such implications really candidates for the status as universals? We believe not. (a) In conditionals like (10), A is of course contingent, and for this reason alone such implicationals are not universals in the sense of 'necessary property of all languages'. If A does not apply to language L, (10) is of course true about L as well, but only in a trivial, spurious sense. (b) In our view, the operator "necessary" will be justified only if the pragmatic rule invoked is a pragmatic principle. If the substantiation behind is an empirical backing (in Toulmin's sense, Toulmin 1958), e. g., a widely attested tendency, only a weaker operator like "almost always" can be inserted. We are claiming that what is the case in a particular language cannot be explained. What we can hope to show is that the choice of an option makes sense, not to show why this option is preferred. To explain a choice in a particular language, we would have to exclude all other structural possibilities. Consider the following example from Kefer (1986: 57 — 60), which we shall interpret from our own vantage point. Kefer reflects on Greenberg's universal 25: "If the pronominal object follows the verb, so does the nominal object" (Greenberg 1966: 91). He derives this implication from the interaction of two more general tendencies which he argues are functionally motivated: a) light constituents go left, heavy constituents go right; and b) uniformity is preferred to non-uniformity. Where the relative order of object pronouns and object
Universals, explanations and pragmatics
23
nomináis is concerned, the interaction of these two tendencies will result in four theoretical combinations. I.
object pronoun >verb verb > nominal object (Tendency of lightness followed, uniformity violated)
II.
verb > object pronoun verb > nominal object (Tendency of uniformity followed, lightness violated)
III.
object pronoun > verb nominal object > verb (Tendency of uniformity followed, lightness violated)
IV.
verb > object pronoun nominal object > verb (Both tendencies violated)
The fourth option is excluded since it goes against both tendencies. Seen through our eyes, it is easy to sympathize with Kefer's interpretation of universal 25, but we are not dealing with a universal, but with an explanation of a widespread functional tendency. Kefer's explanation presupposes a third principle which we have reconstructed as: "Structural options contrary to all relevant functional tendencies are universally excluded", but how, in its turn, could this rule be substantiated? What could prevent the existence of a dysfunctional language in which (IV) was realized? Secondly, Greenberg's universal 25 is deceptive because of its positive formulation. As Kefer shows, it is really a statement about facts which we do not find, and it is the negative options that we can try to explain. As far as the three "possible" combinations (I —III) are concerned, what Kefer really shows is — much in the same style as our interpretation of subject pronoun dropping — that these three options make sense: they comply with at least one functional tendency. But there is no way to explain the actual choice between them. To explain the choice in, say, English, we would have to exclude not only case (IV) above, but also (I) and (III). This could only be done by showing that, in the context of the grammar of English as a whole, these options would violate pragmatic principles (or functional tendencies, if explanation is taken in Kefer's weaker sense).
24
Hartmut Haberland and Lars
Heltoft
P r o p e r e x p l a n a t i o n s in linguistics w i l l a l w a y s h a v e a p r a g m a t i c character. T h i s is because any o t h e r k i n d o f e x p l a n a t i o n w o u l d still ask f o r the t u r t l e beneath: o n l y p r a g m a t i c e x p l a n a t i o n s m a k e sense w i t h o u t r e f e r r i n g t o o t h e r principles. E x p l a n a t i o n s in the p r a g m a t i c sense are n o t m e r e l y causal explan a t i o n s o r regularities, b u t r a t h e r teleological explanations. L a n g u a g e is a f u n c t i o n a l sign system, and linguistic features are explained t h r o u g h their ( o f t e n v e r y c o m p l e x ) relations t o p r a g m a t i c p r i n c i p l e s . 9 L i n g u i s t i c features ( w h e t h e r o b s e r v a t i o n s a n d empirical generalizations, o r the e v e r y d a y
rule
k n o w l e d g e o f l a n g u a g e users) m a k e sense w h e n u n d e r s t o o d as p a r t s o f a basically n o r m a t i v e system w h i c h enables speakers t o a c h i e v e c o m m u n i c a t i v e ends. In this sense, o u r e n t e r p r i s e does n o t r e s e m b l e the n o m o t h e t i c sciences, but rather hermeneutic understanding. Writing grammars and
explaining
G r e e n b e r g u n i v e r s a l s means f u n c t i o n a l systematization o f p e o p l e ' s k n o w l e d g e o f h o w to act w i t h w o r d s .
Notes 1. Some of the ideas expressed in this contribution came up in discussions with Ole Nedergaard Thomsen, which we gratefully acknowledge. Another source of inspiration were discussions with Tassos Christidis which Hartmut Haberland had in Thessaloniki in November 1988. 2. In a certain sense they are converses of ordinary predicative constructions. Predicatives classify a subject NP ( D y n a s t y is on ABC), while categorials take the set for granted (What has ever been on ABC that was worth watching?) and specify the object NP as a member of that set {Well, there has been Dynasty). 3. In English, the NP has no definite article, as it has in Danish, but it is clear that this NP still is not indefinite. 4. In one interpretation of John hit Mary, Mary is the only focus, the rest being background. The background includes an unspecified referential χ (the object slot of the verb), and Mary is related to this background as the identification of that x. 5. Under certain circumstances, a sentence like (8 b) can occur, e. g., as an answer to the question was macht Claudia?. That this phenomenon of ellipsis is different from pro-drop proper is clear from the fact that it is restricted to clause-initial (probably even utterance-initial) positions; (i) is not acceptable in any context: (i) *Weil eine Ρί^χα ißt. "Because is eating a pizza." On the other hand, in Italian (ii) is fully acceptable: (ii) Perché mangia una pizza. 6. Horrocks (1987: 147) mentions the possibility to analyze verbal clitics as object agreement markers, in which case pro could occur in object position. In this non-standard analysis, we would have a dropped object pro in sentences like (i), since la only is an agreement marker on the verb: (i) Claudia la mangia. "Claudia is eating it." Many but by no means all languages with verbal clitics are pro-drop (for Subject pro).
Universal·s, explanations
and pragmatics
25
7. This is argued, e. g., in Givón (1984), but Givón is not alone in his claim. 8. For example, one has tried to argue that the present tense in languages like German or Modern Greek which can also have future-time reference is a real present tense (not a nonpast tense) and that the future-reference reading can be derived by application of the Maxim of Quality. This attempt is bound to fail, since, if this approach works for these languages, why do not all languages allow for such a reading of the present tense? Pragmatic principles cannot explain why languages are different. 9. See Heltoft (1981) for an attempt to relate epistemic verbs to universal pragmatics.
References Bühler, Karl 1934 Sprachtheorie. Jena: Fischer. Comrie, Bernard 1981 Language universals and linguistic typology. Oxford: Blackwell. Coulmas, Florian — Hartmut Haberland — Jacob Mey 1983 "Universals are unicorns in an impossible world, Report from Working Group 4, Universals of linguistic action", in: Shirò Hattori et al. (eds.), Proceedings of the 13th International Congress of Linguists, Tokyo 1982. Tokyo: Gakushûin University, 1232-1238. Givón, Talmy 1984 Syntax, a functional-typological introduction, Volume I. Amsterdam and Philadelphia: Benjamins. Greenberg, Joseph H. 1966 "Some universals of grammar with particular reference to the order of meaningful elements", in: Joseph H. Greenberg (ed.), Universals of language. 2nd edition. Cambridge, MA: MIT Press, 73 — 113. Grice, H. P. 1975 "Logic and conversation", in: P. Cole and J. Morgan (eds.), Syntax and Semantics, 3. New York: Academic Press, 41 — 58. Habermas, Jürgen 1969 "What is universal pragmatics?", in: J. Habermas, Communication and the Evolution of Society. Boston: Beacon, 1—68. 1971 "Vorbereitende Bemerkungen zu einer Theorie der kommunikativen Kompetenz", in: Jürgen Habermas —Niklas Luhmann, Theorie der Gesellschaft oder So^ialtechnologie: was leistet die Systemforschung? Frankfurt am Main: Suhrkamp, 101-141. Hansen, Aage 1967 Moderne dansk I—III. Kobenhavn: Grafisk. Hawkins, John A. 1983 Word order universals. New York and London: Academic Press. Heltoft, Lars 1981 "Grammar and the intersubjectivity of language." Nordic Journal of Linguistics 4: 9 1 - 1 0 7 .
26
Hartmut Haberland and Lars 1987
Heltoft
"The pragmatic syntax of Danish ¿«/--constructions", in: Pirkko Lilius —Mirja Saari (eds.), The Nordic languages and modern linguistics 6. Proceedings of the 6th International Congress of Nordic and General Linguistics. Helsinki: Helsinki University Press, 177 — 192. Horrocks, Geoffrey 1987 Generative grammar. London: Longman. Joos, Martin (ed.) 1957 Readings in linguistics I. Chicago: Chicago University Press. Kefer, Michel 1986 "On functional explanations of language change, universals and rules." Belgian fournal of Linguistics 1: 53 — 66. Lehmann, Christian 1980 "Aufbau einer Grammatik zwischen Sprachtypologie und Universalistik", in: Gunter Brettschneider—Christian Lehmann (eds.), Wege \ur Universalienforschung. Sprachwissenschaftliche Beiträge %um 60. Geburtstag von Hansjakob Seiler. Tübingen: Gunter Narr, 29 — 37. Philippaki-Warburton, Irene 1981 "Katholikoi periorismoi kai neoelliniki sintaksi" [Universal constraints and Modern Greek syntax], in: Studies in Greek linguistics. Proceedings of the 2nd Annual Meeting of the Department of Linguistics, Faculty of Philosophy, Aristotelian University of Thessaloniki, 4 - 6 May 1981, 1 7 9 - 2 1 2 . Strawson, Peter F. 1959 Individuals. An essay in descriptive metaphysics. London: Methuen. Toulmin, Stephen 1958 The uses of argument. Cambridge: Cambridge University Press.
Isomorphic, anisomorphic, and heteromorphic universale1 John E. Joseph
1. Isomorphic universale The title of this paper identifies the three logically possible types of language universale insofar as motivation is concerned, where we define motivation or isomorphism as the existence of a convincing parallel in the relation between (1) linguistic signifiers and (2) something else, either (2a) linguistic (i.e., signifieds) or (2 b) extralinguistic but implicated in language production (e. g., objects, perception). 2 A classic isomorphic universal is universal 35 from Greenberg (1966 a: 94): "There is no language in which the plural does not have some nonzero allomorphs, whereas there are languages in which the singular is expressed only by zero." Expressions of plurality are grosso modo longer than those of singularity (familiar exceptions include certain Old French nominative and Russian genitive nouns, and English present indicative regular verbs), and the motivation for this is patent: plurality is conceptually "larger" than singularity (on these and related issues, see further Joseph 1985 and 1987 a). A bit less obvious is the explanation for Universal 36 (Greenberg 1966 a: 95): "If a language has the category of gender, it always has the category of number". This is isomorphic because, conceptually, a much larger class of things have count than have gender. Apart from animate nouns, gender attaches to things by metaphor and cultural accretion, and although this process is a powerful one (sometimes overriding natural gender), number is closer than gender to the "inherent" end of the continuum between inherent and derived attributes. As for verbs, Bybee (1985: 23) has noted that number distinctions are much more often lexicalized than are person and gender categories, because, she explains, the latter "seem to have little effect on the meaning of a verb", whereas The number of participants in a situation, whether agents or recipients of an action, can affect the situation profoundly. Thus lexicalized distinctions based at least in part [upon] the number of participants do exist, e. g., English run vs. stampede, murder vs. massacre.
28
John E. Joseph
Only 16% of the languages in her forty-language sample inflect verbs for gender, while 54% inflect for number and a further 12% express number through derivation (Bybee 1985: 3 0 - 3 1 ) . Modern theories of isomorphism differ crucially from those offered prior to the work of Jakobson. The oldest recorded version of isomorphism dates at least from Heraclitus (c. 545 —c. 485 B. C. E.), and may be formulated as in (1): (1)
WORD
REAL-WORLD DESIGNATUM
Here the focus was on the bond between language and the physical world, and whether a, a morpheme (usually interpreted as a word) had a motivated ("natural") or arbitrary ("conventional") relationship to a", the real-world thing it designated (see further Joseph 1990). The formulation in (2) indicates how Aristotle (384—322 B. C. E.) radically reframed the issue by positing a', the conceptual level, as a full member in the semiotic chain: (2)
a WORD (SIGNIFIER)
=
a' CONCEPT (SIGNIFIED)
=
a" REAL-WORLD CORRELATES
Centuries later, Ferdinand de Saussure (1857 — 1913) would effectuate another radical reformation by designating a as a part of the linguistic system. Yet right up through Saussure neither theory (1) nor theory (2) admitted of a linguistic solution, and positions taken had more to do with general ideology than with actual evidence. The assertion of a — a" had more to do with belief in the harmony of the universe and the natural logic of the human mind than with the existence of onomatopoetic words or analogical patterns in morphology. Denial that a = λ" was often based as much on belief in free will (see Taylor 1988) as on the observation that different peoples called the same a" by a different a. The contribution of Roman Jakobson (1896 — 1982) was to reframe the comparison using multiple rather than single units of the system, in the spirit of the Saussurean assertion that not units themselves, but the relations among them, constitute the values of the system. Beginning with his work on "marked" and "unmarked" elements in Jakobson 1932, and culminating with his consideration of isomorphism in Jakobson 1966, he demonstrated the degree to which the relationships among signifiers actually reproduce the relationships among signifieds and among the corresponding segments of extralinguistic reality (see Joseph 1987 b). He suggested that this kind of
Isomorphic, anisomorphic, and heteromorphic universals
29
reproduction is an important source of economy in linguistic structure, since it renders the acquisition of language immeasurably easier than if the relations of signifiers were purely arbitrary relative to the other two levels. We can formulate this isomorphic situation as in (3): (3)
ab > ac
=
a'b' > a'c'
=
a"b" > a"c"
where again abc are signifying elements of the system and a' b' c' the concepts signified, but with Λ" b" c" defined more broadly as the significant extralinguistic correlates of the other levels; where we are able to establish differential values for b and c relative to a, such that the relationship ab is different from the relationship ac — thus ab > ac or vice versa, where the symbol ' > ' suggests a stronger or closer relationship in terms of linguistic make-up (place and manner of articulation, morphological structure, etc.), complexity (simple vs. derived), stability (proneness vs. resistance to assimilation), and so on. Jakobson's question is whether the difference between ab and ac can be shown to equate with (or, if one wishes, to reproduce) a corresponding difference between a'b' and a'c', or between a"b" and a"c", or both. In Universal 36, a, the "ground", is any morpheme capable of bearing number or gender inflection; b = number inflection; c = gender inflection; and the observation is that ab > ac. a b' c' — the language-specific concepts signified by abc. As discussed above, a'b' > a'c' by the greater direct semantic applicability of b' over c'. In this instance, there is a direct correlation as well to a"b" > a'c" with the real-world correlates, though this is not always the case; and as we shall see below, many examples of ab > ac have been provided with an a"b" > ac" correlation without a'b' > a'c' being demonstrable.
2. Anisomorphic universale But the great majority of Greenberg's universals are not based on any claim of isomorphism, for instance, Universal 19 (Greenberg 1966 a: 87): "When the general rule is that the descriptive adjective follows, there may be a minority of adjectives which usually precede, but when the general rule is that descriptive adjectives precede, there are no exceptions." This universal contains one constant, descriptive adjectives (call this a), and one variable, pre- or postnominal position (b and c respectively). For there to be isomorphism here, the ratio ab : ac must be directly connected with some other twoplace variable ratio on the level of meaning {a'b' : a'c') or on some extralinguistic level (a"b" : a"c"). The case at hand is a simple matter of head-adjunct
30
John E. Joseph
vs. adjunct-head ordering, and since both types are well distributed among natural languages, and there is no obvious rational principle favoring the one over the other, it is difficult to imagine that a significant and general correlation exists on the a or d' level. If no correlation exists, the universal is anisomorphic. The predominance of anisomorphic universale in Greenberg's original list is treated as a clear weakness by Comrie (1981: 88): Perhaps one of the reasons why many linguists following on from Greenberg's results have been less careful in their statements about correlations is that a large number of Greenberg's universals, however valid they may be ... do not seem to lend themselves to any ready explanation or generalization ... Some of Greenberg's empirically ascertained universals do have plausible explanations, and these are the greatest factual merits of his list of universals ....
Comrie goes on to discuss Greenberg's list as essentially a dichotomy of universals more and less intuitively plausible. The years since the list was formulated have seen a steady series of attempts to convert anisomorphic universals into isomorphic ones, by discovering a' and a" contexts for motivated relationships.3 The a contexts have been sought by Haiman, Bybee, and others, with good success considering that such work had been virtually forbidden through most of the history of modern linguistics. The search for β" contexts has been still more successful, resulting in Vennemann's Natural Serialization Principle, Hawkin's CrossCategory Harmony, Givón's several principles, and so on, as well as the various "explanations" for universals pursued by the contributors to Hawkins (ed.) 1988. This search has also produced an uneasy but promising convergence with the goals of Government and Binding theory (cf. Coopmans 1984; Hawkins 1985). I shall discuss the motivation for the rejection of anisomorphic universals further on. But if we approach the issue logically, objectively, nonideologically, what prevents universals from being not just anisomorphic, but actually heteromorphic, with the relationship between signifiers ab and ac not failing to correlate, but correlating inversely, with that between signifieds or extralinguistic elements db' and a'c'i Collapsing for now the d\d' distinction, the six logical possibilities are shown in (4). ab > ac ab > ac ab > ac ab = ac ab ac ab ac -
& & & & & &
a'b' a'b' a'b' a'b' a'b' a'b'
> a'c' a'c' < a'c' > a'c' a'c' < a'c' -
isomorphic anisomorphic heteromorphic anisomorphic anisomorphic anisomorphic
Isomorphic,
anisomorphic,
and heteromorphic
universals
31
Whenever no differential can be found at either the a or a level, the universal will be anisomorphic. Following (4), we would expect anisomorphic universale to outnumber the other types by a factor of two, and this is roughly borne out in Greenberg's original list. We would further expect that heteromorphic and isomorphic universale should occur in roughly the same small proportion, yet such is not the case: isomorphic examples greatly outnumber heteromorphic ones. Pace Comrie (1981), we might turn the tables and ask how it is that so much research has turned up so few logically implausible universals.
3. Heteromorphic universals The most obvious candidate for a heteromorphic universal from Greenberg's list would appear to be Universal 8 (Greenberg 1966 a: 80): "When a yes-no question is differentiated from the corresponding assertion by an intonational pattern, the distinctive intonational features of each of these patterns is reckoned from the end of the sentence rather than from the beginning." The matter of whether a sentence is in the interrogative mode affects its whole processing and interpretation, and thus one might expect the interrogative signal to occur at the start of the proposition or word. In Universal 8, the variables are distributed as in (5). (5)
ab < ac a — interrogative intonation b = initial-reckoning c = final-reckoning
but
a'b' > a'c' a' = question mode b' — pre-signaling c' = post-signaling
It seems intuitively correct that a , which pertains to the whole of the sentence, affects its entire comprehension, and that therefore a'b' > a'c'. This proportion is matched by the structure of at least one language not in Greenberg's sample: the Western Desert language of Australia places a higher pitch on the initial syllable of a question (Douglas 1958; Ultan 1978: 220). Also, Finnish (which was in Greenberg's sample) raises the pitch on all stressed vowels in a question, thus locating the signal at the beginning rather than the end of the intonation contour (Ultan 1978: 220). But even if we weaken 8 to a statistical universal, Greenberg's empirical finding is still that while a'b' > a c', there is a strong tendency for ab < ac. Either this is truly a heteromorphic case, or a b' c' are misidentified as the operative parameters in this case, or a' is more peripheral to the processing
32
John E. Joseph
of the sentence than logic would dictate. If we accept it as heteromorphic, we do not have to stop there, but can devise an a" (etc.) set that will define a deeper isomorphism. In this case, b" and c" are likely to be about general properties of intonation: Hyman (1975: 209), for instance, has claimed that pitch change is more easily perceived at the end than at the beginning of the intonation contour (cf. also Bolinger 1985). It would thus appear to be possible for isomorphism at one level (here a") to coexist with heteromorphism at another (here a').
4. Refining Greenberg's Universal 9 If Universal 8 is truly heteromorphic for the reasons I have stated, then what about Universal 9 (Greenberg 1966 a: 81)? "With well more than chance frequency, when question particles or affixes are specified in position by reference to the sentence as a whole, if initial, such elements are found in prepositional languages and, if final, in postpositional." Question particles and affixes (henceforth QPs) perform the same function as intonation, and apply to the sentence as a whole; why then should they tend to pattern with heads, as Universal 9 indicates? First of all, let us note that Ultan (1978: 227 — 228) conveys some discomfort with Greenberg's formulation, since in his own sample, of the languages "with QP initial or enclitic to the sentenceinitial constituent, 24% are postpositional". He does not, however, attempt to reformulate Universal 9 more precisely; that is what I shall do now, using Ultan's sample, and eliminating from it all questionable cases which I was not able to verify elsewhere. 4 Comparing adposition and QP position we find the distribution in (6), where QP-I = sentence-initial QP, and QP-F = sentence-final QP: (6)
Pr & QP-I: 16 (Agta, Albanian, Chontal, French, Louisiana French, Scottish Gaelic, Gunwinggu, Hebrew, Lithuanian, Malagasy, Rotuman, Russian, Samoan, Squamish, Tagalog, Zapotee) Pr & QP-F: 8 (Syrian Arabic, Mandarin Chinese, Diola, Gbeya, Sango, Tajik, Thai, Vietnamese) Po & QP-I: 4 (Finnish, Guarani, Ojibwa, Piro)
Isomorphic,
anisomorphic,
and heteromorphic
universali
33
Po & QP-F: 17 (Aranda, Asmat, Basque, Buriat, Burmese, Ewe, Fanti, Grebo, Hungarian, Japanese, Kannada, Kurku, Khalkha Mongolian, Telugu, Tunica, Twi, Yakut) The proportions figured in (7) show that Greenberg's Universal 9 stands statistically, though rather more weakly than one would wish: (7)
Of Pr languages, Of Po languages, Of QP-I languages, Of QP-F languages,
66.7% (16/24) are QP-I 33.3% (8/24) are QP-F 19% (4/21) are QP-I 81% (17/21) are QP-F 80% (16/20) are Pr 20% (4/20) are Po 32% (8/25) are Pr 68% (17/25) are Po
If we use verb-object order rather than adposition order (as suggested by Ultan 1978: 228) the results, shown in (8), are stronger. (8)
VO & QP-I: 18 (Agta, Albanian, Chontal, Finnish, French, Louisiana French, Scottish Gaelic, Guarani, Hebrew, Lithuanian, Malagasy, Ojibwa, Rotuman, Russian, Samoan, Squamish, Tagalog, Zapotee) VO & QP-F: 11 (Syrian Arabic, Mandarin Chinese, Diola, Ewe, Fanti, Gbeya, Grebo, Sango, Thai, Twi, Vietnamese) OV & QP-I: 2 (Gunwinggu, Piro) OV & QP-F: 14 (Aranda, Asmat, Basque, Buriat, Burmese, Hungarian, Japanese, Kannada, Kurku, Khalka Mongolian, Tajik, Telugu, Tunica, Yakut)
From the proportions in (9), the major parameter that emerges is the bond between OV and QP-F, which however remains statistical. 5 (9)
Of VO languages, Of OV languages, Of QP-I languages,
62.1% (18/29) are QP-I 37.9% (11/29) are QP-F 12.5% (2/16) are QP-I 87.5% (14/16) are QP-F 90% (18/20) are VO 10% (2/20) are OV
34
John E. Joseph
Of QP-F languages,
44% (11/25) are V O 56% (14/25) are OV
Now let us combine the two sets of findings, as in (10). In the right column, for comparison, are the number of languages with the corresponding adposition and verb-object order from Hawkins's (1983) Expanded Sample. (10)
Of QP-I languages, 75% (15/20) are 10% (2/20) are 15% (3/20) are 0% (0/20) are Of QP-F languages, 28% (7/25) are 4% (1/25) are 16% (4/25) are 52% (13/25) are
Pr Pr Po Po
& & & &
VO OV VO OV
(136/336 = 40.5%) (12/336 = 3.6%) (26/336 = 7.8%) (162/336 = 48.2%)
Pr Pr Po Po
& & & &
VO OV VO OV
(136/336 (12/336 (26/336 (162/336
= 40.5%) = 3.6%) = 7.8%) = 48.2%)
Here, finally, we have come up with a zero, and a significant one at that, given that it occurs at the most common language type, Po & OV. We can now formulate a non-statistical universal, plus two very strong statistical universals, as in (11), where " r > " means "implies the existence o f ' . (11)
(I) QP-I 3 (Po 3 VO) (II) QP-F 3 (OV => Po) [1 exception /14; 92.9% accurate] (III) QP-I 3 V O [2 exceptions/20; 90% accurate]
(I) says that if a language has QP-I, and if it is Po, then it must be VO. (II) says that if a language has QP-F, and if it is OV, then there is a greater than 90% chance that it will be Po. (III) says that if a language has QP-I, then there is a 90% chance that it will be VO. (I) and (III) indicate that QP-I patterns with heads; (III) is written with V O rather than Pr because V O is a slightly better indicator (18 VO, 2 OV vs. 16 Pr, 4 Po). Note that there is no universal corresponding to (III) for QP-final languages, indicating that they are more likely to have QP be disharmonious with verbal and adpositional heads. (I) says that if QP-I does occur in an OV language, it will always be one with ambiguous head-position, hence never OV & Po: one or the other parameter has to be harmonious with QP-I: (II) does not make a parallel claim for QP-F (i. e., that if QP-F occurs in a V O language, it will never be VO & Pr). Rather, (II) indicates that a fully
Isomorphic,
anisomorphic,
and heteromorphic
universah
35
harmonious head-final language is likely to pull QP along with it, but not to the same degree as a head-initial language. Still, the force of (II) is questionable given that the exception to it in the sample gives us a 4% rate of QP-F & Pr & OV, consistent with the 3.6% rate of Pr & OV found overall. If we use NP instead of VP orders, we get essentially the same result, as (12) shows. (12)
Of QP-F languages, 20% (5/25) are 4% (1/25) are 4% (1/25) are 4% (1/25) are 4% (1/25) are 0% (0/25) are 28% (7/25) are 36% (9/25) are
Pr & NG Pr & NG Pr & GN Pr & GN Po & NG Po & NG Po & GN Po & GN
& & & & & & & &
NA AN NA AN NA AN NA AN
(104/336 (30/336 (4/336 (10/336 (11/336 (0/336 (68/336 (109/336
= = = = = = = =
31%) 8.9%) 1.2%) 3%) 3.3%) 0%) 20.2%) 32.4%)
Again, a slight skew toward consistent head-final, but an ability to co-occur with any attested language type. QP is more likely to be harmonious or "stable" when it is initial than when it is final. QP-F languages have a much more nearly normal distribution, slightly skewed by head-final, than do QP-I languages, which are heavily skewed toward head-first. It is much easier for a head-first language to be QP-F than for a head-final language to be QP-I, as (13) verifies. (13)
Of Pr & VO languages, Of Pr & OV languages, Of Po & VO languages, Of Po & OV languages,
68.2% (15/22) are QP-I 31.8% (7/22) are QP-F 66.7% (2/3) are QP-I 33.3% (1/3) are QP-F 42.9% (3/7) are QP-I 57.1% (4/7) are QP-F 0% (0/13) are QP-I 100% (13/13) are QP-F
If QP patterned consistently with heads, we would expect a 100-0 split among Pr & VO languages just as we find in Po & OV. Instead we find that QP-F occurs widely throughout language types, apparently driven by some force that favors final position over initial. This is a rather familiar pattern, observed by Hawkins with relative clauses and affixes, and in the opposite direction by Keenan with subjects. Within the framework of isomorphism provided by the Natural Serialization Principle and Cross-Category Harmony, these patterns are clearly heteromorphic. Note
36
John E. Joseph
that they have not been allowed to stand as such. Keenan's (1979) "Subjects Front" and Hawkins's Heaviness Serialization Principle and Suffixing Preference are all attempts to undo apparent heteromorphism by inventing a still deeper level of isomorphic connection. In this spirit, we can identify here a QP-F PREFERENCE (14) and go about considering why it should exist. (14)
QP-F PREFERENCE: Final question particles and affixes occur in many languages in which consistent head-ordering would predict initial particles, where the converse fails.
Since Greenberg and Ultan do not distinguish question particles from question affixes, the possibility cannot be discounted that the QP-F Preference is itself a manifestation of the Suffixing Preference, Hawkins's designation of the fact that "... suffixes occur productively in many languages in which the H[ead] Ordering] Principle] would predict prefixes, where the converse fails" (Hawkins — Gilligan 1988: 236; see also Joseph 1989). This principle is supported by data such as those in (15), which come from Hall (1988: 321). (15) Head-I languages Head-F languages All languages
Prefixes only 8.5% 0% 4%
Both prefixes and suffixes 72.5% 36.5% 54.5%
Suffixes only 19% 63.5% 41%
Certainly the Suffixing Preference seems to be at work in Greenberg's Universal 10 (Greenberg 1966 a: 82): "Question particles or affixes, when specified in position by reference to a particular word in the sentence, almost always follow that word. Such particles do not occur in languages with dominant order VSO." Lexical QPs are naturally more likely than sentential QPs to be affixed. VSO languages are the most rigidly prepositional, and would be the most likely to place QP before a word and to put up resistance to suffixing. Alternatively, we can recognize Greenberg's Universal 9 and the new universals proposed here as doubly heteromorphic, at both the a' and the a" level, since both their semantic and their perceptual dimensions appear to contravene their formal structure. But that is unlikely to be satisfying. The problem is that universals cannot be both entirely heteromorphic and explicable; explaining universals means finding isomorphism at some deeper level than the obvious anisomorphic or heteromorphic one. We can identify an isomorphism preference in universal-typological linguistics, that is dependent upon the need to explain universals. It could have one of several sources: (1) it may result directly from the inherent isomorphism of language structure;
Isomorphic,
anisomorphic,
and heteromorphic
universals
37
(2) it may be purely ideological on the part of the practitioners; (3) it may be a historical accident; (4) it may be a coincidence. Given the opinions and attitudes consciously expressed in, e. g., Haiman (ed.) 1985 and Hawkins (ed.) 1988, we can eliminate (4) as a serious possibility, but probably not any of (1, 2, 3). We shall see, however, that (3) has played an especially strong role.
5. History of the isomorphism preference Most participants at the 1961 Dobbs Ferry Conference were American structuralists who accepted some version of the doctrine of the arbitrariness of linguistic signs and considered any apparent exceptions to it (cases of onomatopoeia are all that are usually cited) to be trivial. The strongest statement of this doctrine in the Universals of Language volume occurs in Uriel Weinreich's paper: "Thus, we would rule out, as nonlanguage, ... systems whose signs are iconic ..." (Weinreich 1966: 142). 6 Yet even Weinreich was no "hard-line" structuralist, or he probably would not have attended the conference, let alone contribute a paper that — apart from scattered remarks like the preceding — is faithful to its spirit. Also present at the conference was the linguist who for over thirty years had done more than anyone else to undermine the arbitrariness doctrine, Roman Jakobson. A better sense of the meeting's mood emerges from the insightful and well-balanced paper by Hockett (1966), who calls for an "inductive" method in formulating universals, i. e., reasoning from particular instances to a general conclusion. He never says against just what views or methodologies he is arguing; this is 1961, and Chomsky has yet to arrive at the forefront of his concern. In fact, Hockett's calls for a method that does not rely too much on empirical observation would seem to be in the Chomskyan spirit. But a couple of pages later he cites the Port Royal Grammar (Arnauld — Lancelot 1660), and clearly not as a model of the right kind of rationalism. It is instead an embodiment of linguistic Eurocentrism: "Latin, of course, stood at the origin in this particular coordinate system. Any other language could be characterized typologically by listing the ways in which its machinery for satisfying the universal scheme deviated from that of Latin" (5). Later on the same page he notes that the historical diffusion of features can occur in an unnatural way, given "a long period of intimate contact. This fact, if no other, would seem to render suspect any generalÌ2ations based solely on the languages of Western Europe" (5).
38
John E. Joseph
The elimination of linguistic ethnocentrism had been at the root of the American structuralist project, the development of a system of linguistic analysis based on forms and their occurrence patterns, without reference to meaning, which was considered an adjunct behavioral domain. The fear that the particular sound-meaning conventions of European languages could influence our analysis and understanding of the different conventions of less familiar tongues — including those of Native America, in which so many of these linguists had been trained — continued to be dominant in 1961, as Hockett's remarks indicate. The danger was greater because so little was known scientifically about meaning, particularly cross-culturally. 7 Matthews (1986) has correctly warned against overstating the structuralist rejection of meaning as if it were absolute. Nevertheless, it was an overarching concern; and it attained its strongest manifestation in the so-called "Sapir-Whorf' hypothesis, which overtly subjugated meaning to form by claiming that thought (which presumably includes signifieds and semantics) is shaped by the distribution of signifiers in a language. Such a forceful denial of the autonomous existence of meaning would seem to border on obsession. By 1961 the Bloomfieldian solution was collapsing under its own weight. Impractical enough for the analysis of any one language, it made crosslanguage comparison virtually impossible. It is however responsible for the absence of references to meaning in Greenberg 1966 a. But if meaning-free grammar could not give the whole solution, the alternative was logical extrapolation — induction — from form to meaning and vice versa, to hypothesize very general logical patterns of form and meaning that no human language could conceivably do without. At this high level of abstraction, isomorphic hypotheses proved extremely useful as heuristics. The obstacle they encountered was the arbitrariness of linguistic signs, which American linguists had accepted as an uncontroversial and even rather trivial fact since Whitney's time. Given the fact that the Bloomfieldian project of meaning-free grammar was short-lived, while arbitrariness as a design feature of language is still preached by introductory linguistics textbooks, it is easy to think of the former as a weak principle and the latter as a strong one. But I would contend that this "strength" has proven to be a deficiency: linguists took the "self-evident" arbitrariness doctrine so for granted that they seldom bothered to defend it against critiques like that of Jakobson 1966. To this day there has appeared no refutation of the critiques of arbitrariness that addresses the data rather than reiterating dogma. Within a few years of the Dobbs Ferry Conference, Chomsky was toppling Bloomfield, and the handful of scholars who continued to work on language universale found themselves challenged by the proponents of Universal Gram-
Isomorphic,
anisomorphic,
and heteromorpbic
universali
39
mar to attain a level of explanatory adequacy equal to their observational adequacy. There can be little question but that the markedly different direction of Greenberg 1966 b came in response to the pressure for explanation triggered by the work of the transformational-generative school. It was here that Greenberg incorporated the Prague theory of the mark into his work on universale, a move that was made possible by the presence of some overtly isomorphic universale among his original 45. The mark had been largely about isomorphism since Jakobson's original formulation of it in response to an offhand comment in a letter from Trubetzkoy (see Joseph 1987 b), and Greenberg's incorporation of the mark changed isomorphism from a mere discovery procedure to an explanatory principle within universal-typological linguistics. This historical accident — a motivated accident, to be sure — is largely responsible for the isomorphism preference identified here.
6. Conclusion The present paper has attempted to demonstrate that three logically possible types of universal linguistic statements exist with regard to motivation, but that one of these has been heavily favored over the others for reasons that are more historical than scientific. In the case of Greenberg's Universal 8 we saw that a single statement can be simultaneously isomorphic and heteromorphic at different levels of analysis. With Universal 9 we saw first that an implicit assumption that isomorphism equals explanation led Greenberg to overgeneralize the relationship between question particles and heads, when in reality this relationship is subject to a preference for question particles in final position. We then noted the logical heteromorphism of this preference, in view of the fact that a question particle affects the meaning and interpretation of the entire statement in which it occurs; but recognized further that such a claim of heteromorphism is unlikely to be satisfying under the current set of working assumptions in the field, which would require instead that we seek isomorphism at some deeper level. This requirement, I suggest, is an artifact of the history of the discipline rather than a point of objective scientific methodology. The pressure for explanation emanating from the insistence on explanatory adequacy in generative linguistics has been both a constructive and a destructive force within typological studies: it has provoked the observation of important principles, but also their too hasty formulation. Certainly we cannot be satisfied with unprincipled description. But ad hoc, quick-fix, pseudo-
40
John E. Joseph
explanation is hardly preferable. Nor is the tendency toward sweeping "principles" based on general notions of perception and cognition (Lee 1988), which become so powerful as to leave no room for what is really anisomorphic and heteromorphic in language. If the history of linguistics teaches us anything, it is that only disciplined and balanced thought will save typologists from letting facile "explanation" become their obsession, as "meaning" became the obsession of structuralism, and led finally to its undoing.
Notes 1. I wish to thank Bernard Comrie, Marc Dominicy, Hansjakob Seiler, and especially Michel Kefer for comments during and after the Antwerp conference which helped guide the revision of this paper; and the late William Gedney for his first-hand reminiscences of the Dobbs Ferry Conference, related to me in personal interviews in 1978. Thanks also to Ulrich Ammon and Johan van der Auwera for their hospitality. This work was made possible by a Semester Research Award and a Travel Grant from the Graduate Research Board of the University of Maryland. 2. There is some disparity in the use of "isomorphic": for certain linguists it indicates merely a numerical correlation (one to one) between (1) and (2) in a particular language, whereas for others, including myself, it indicates a motivated correlation — what is often called "iconic". I prefer to apply "isomorphic" to a motivated relationship for three reasons: because the biuniqueness of correlations is difficult to prove, especially on the side of (2); because biuniqueness is less significant than motivation; and because the term "iconic", deriving from the work of Peirce, strikes me as being misleadingly and prejudicially unidirectional. An icon is a secondary representation of some more "real" thing: χ is an icon of y. But linguistic signifiers and signifieds having a motivated relationship are reflections or representations of each other, and "isomorphic" captures this reciprocality in a more accurate and open-minded way. 3. At the Antwerp conference Marc Dominicy and Michel Kefer questioned my application of the single set of terms iso-/aniso-/heteromorphism to correspondences of both the types aa and ad'. Both indicated that they would prefer to apply these terms at the purely linguistic aa level only, and to designate extralinguistic aa" correspondences differently. To my mind this is a minor taxonomic point, hardly worth taking a stand on. Nevertheless, I would point out that (1) in conflating the two types, I seem to be following the current practice of linguists throughout Europe and North America, and (2) the existence of a motivated relationship involving a, the signifier, appears at this juncture in the development of linguistic theory to be more significant in and of itself than is the nature of its partner in the relationship, so that the introduction of an additional set of terms might prove to be more confusing than enlightening. 4. I recognize that Ultan's sample is not the most genetically balanced, but would argue that given its direct descendance from Greenberg's own sample it remains useful in refining Greenberg's generalizations, and that comparison with better-balanced samples shows it nevertheless to be quite an accurate predictor (see Dryer 1988). Where appropriate I shall insert data from Hawkins's 336-language sample for comparison.
Isomorphic,
anisomorphic,
and heteromorphic
universals
41
5. Compare Greenberg's Universal 12 (Greenberg 1966a: 83): "If a language has dominant order VSO in declarative sentences, it always puts interrogative words or phrases first in interrogative word questions ...". There are in fact exceptions to this (Ultan 1978: 231, n. 14). 6. Weinreich would also rule out "systems which use other than vocal sign-vehicles", an exclusion which today seems positively antediluvian. 7. This problem remains far from a resolution: recent work like Dryer 1988 and Thompson 1988 are still devoted to sorting through what are identified as "adjectives" in Greenberg's and other's samples — a largely semantic rather than morphological problem.
References Bolinger, Dwight L. 1985 Bybee, Joan L. 1985 Comrie, Bernard 1981 Coopmans, Peter 1984 Douglas, W. H. 1958
"The inherent iconism of intonation", in: Haiman (ed.), 97—108. Morphology.
Amsterdam: Benjamins.
Language universals and linguistic typology. Chicago: University of Chicago Press. "Surface word-order typology and Universal Grammar". language
60: 55 — 69.
An introduction to the Western Desert language of Australia. (Oceania Linguistic Monographs, 4.) University of Sydney. Dryer, Matthew S. 1988 "Object-verb order and adjective-noun order: dispelling a myth". Lingua 74: 185-217. Greenberg, Joseph H. 1966 a "Some universals of grammar with particular reference to the order of meaningful elements", in: Greenberg (ed.), 83 — 113. 1966 b Language universals, with special reference to feature hierarchies. The Hague: Mouton. Greenberg, Joseph H. (ed.) 1966 Universals of language. (2nd ed.) Cambridge, Ma: MIT Press. Haiman, John (ed.) 1985 Iconicity in syntax. Amsterdam: Benjamins. Hall, Christopher J. 1988 "Integrating diachronic and processing principles in explaining the suffixing preference", in: Hawkins (ed.), 321—349. Hawkins, John A. 1983 Word order universals. New York: Academic Press. 1985 "Complementary methods in Universal Grammar: a reply to Coopmans". Language 61: 5 6 9 - 5 8 7 . Hawkins, John Α. —Gary Gilligan 1988 "Prefixing and suffixing universals in relation to basic word order". Lingua 74: 2 1 9 - 2 5 9 . Hawkins, John A. (ed.) 1988 Explaining language universals. Oxford: Blackwell.
42
John E. Joseph
Hockett, Charles F. 1966 "The problem of universale in language", in: Greenberg (ed.), 1—29. Hyman, Larry M. 1975 Phonology: Theory and analysis. New York: Holt, Rinehart & Winston. Jakobson, Roman 1932 "Zur Struktur des russichen Verbums", in: Caristeria Gulielmo Mathesio ... oblata. Prague: Linguistic Circle, 74—84. [Reprinted in Selected writings, II. The Hague: Mouton, 1971, 3 - 1 5 . ] 1966 "Quest for the essence of language". Diogenes 51: 21 — 37. [Reprinted in Selected writings, I. (2nd ed.) The Hague: Mouton, 1971, 3 4 5 - 3 5 9 . ] Joseph, John E. 1985 "Isomorphism and value correlations: evidence from the Latin and French verb phrase", in: Soonja Choi et al. (eds.), ESCOL '85: Papers Jrom the 2nd Eastern States Conference on Linguistics. Columbus: Ohio State University Linguistics Dept., 120 — 131. 1987 a "Subject relevance and deferential address in the Indo-European languages". Lingua 73: 2 5 9 - 2 7 7 . 1987 b Review of La notion de marque che^ Trubet^koy et Jakobson by Michel Viel, Language 63: 665—668. 1989 "Typology, diachrony, and explanatory order". Diachronica 6: 55 — 74. 1990 "The abandonment of nomos in Greek linguistic thought". Historiographia Linguìstica 17: 1 — 13. Keenan, Edward L. 1979 "On surface form and logical form". Studies in the Linguistic Sciences 8.2: 1 —41. [Reprinted in Universal grammar: 15 essays. London: Croom Helm, 1987, 375 — 428.] Lee, Michael 1988 "Language, perception and the world", in: Hawkins (ed.), 211—246. Matthews, P. H. 1986 "Distributional syntax", in: Theodora Bynon and F. R. Palmer (eds.), Studies in the history of Western linguistics, in honour of R. H. Robins. Cambridge: Cambridge University Press, 245—277. Taylor, Talbot J. 1988 The linguistic sign and free will: A forgotten question in the history of linguistics. Paper presented to the Linguistic Association of Great Britain, Exeter, 14 — 16 Sept. Thompson, Sandra A. 1988 "A discourse approach to the cross-linguistic category 'adjective'", in: Hawkins (ed.), 1 6 7 - 1 8 5 . Ultan, Russell 1978 "Some general characteristics of interrogative systems", in: Joseph H. Greenberg (ed.), Universals of human language, Vol. 4: Syntax. Stanford: Stanford University Press, 2 1 1 - 2 4 8 . Weinreich, Uriel 1966 "On the semantic structure of language", in: Greenberg (ed.), 142—216.
Part II Syntactic categories and constructions
Towards a typology of voice1 Ineke Brus
1. Introduction Grammatical voice is a complex phenomenon. Syntactic structures that are considered to be voice options can differ widely in both form and function. Before attempting to set up a typology of any aspect of voice it is therefore vital to define even the most basic notions involved. Furthermore an effort should be made to disentangle, as far as possible, one (or more) of the aspects of the problem of voice from the rest. If this is not at least attempted from the outset, there is a danger that a typology of voice ends up as something as general and unmanageable as "a typology of the transitive clause". By wanting too much too soon, in the end all that is left to conclude will be that "languages are not uniform and do not lend themselves to uniform analyses" (Siewierska 1984: 262). In the last few years studies have appeared on several aspects of grammatical voice, which show that there is much more to voice than our familiar Western European passives would have us suspect. Shibatani (1985) shows the diversity of passive constructions in meaning and structure. He defines a passive prototype in terms of both function and form so that the Western European construction is an example of a prototypical passive, with which constructions found in other languages can have more or fewer properties in common. Foley and Van Valin (1984) propose two functions for the passive, backgrounding and foregrounding, and claim that in some languages a passive construction has one of the two functions, and in others it can have both. Klaiman (1988) introduces the notion of "conceptual status", and shows that some languages have voice options based on the parameters "control" and "affected entity" that are comprised within this notion. Trying to integrate insights from these studies, I will look at a number of constructions that are all connected with the notion of voice in one way or another. Given that the aim is a typology in the literal sense, i.e., a typology that should eventually be able to encompass all languages, it is essential to define basic notions in terms of function exclusively. Therefore I will limit myself as much as possible to an interpretation of the functions of these constructions, i.e., the kinds of real life situations that they express. An in-
46
Ineke Brus
depth discussion of the syntactic structure of these constructions is not my first concern at present. It is therefore a typology of voice functions rather than voice constructions that is presented here.
1.1. Basic notions As a basis for defining different functions a number of assumptions are made: 1. Syntactic constructions are the linguistic expressions or encodings of real-life situations or States of Affairs. 2. For the purpose of this paper States of Affairs are defined on a number of parameters. These parameters will be called semantic, pragmatic, and conceptual status. 3. There is a Standard Transitive State of Affairs, in which these parameters have a default setting, and there are non-standard States of Affairs in which not all parameters have a default setting. 4. All languages can encode a Standard Transitive State of Affairs. In many cases this encoding will be the unmarked option. 5. Non-standard States of Affairs may be expressed in constructions differing from those used to encode the Standard Transitive State of Affairs. Those constructions will be considered to be voice constructions.
1.2. The standard transitive state of affairs A Standard Transitive State of Affairs refers to an action transfer from one referent to another. Action and both referents are characterized by properties of high transitivity (Hopper—Thompson 1980). "High transitivity" implies a (default) setting of a number of parameters (Hopper and Thompson list a number of them). For the purpose of this paper only those parameters that are involved in voice functions are mentioned and defined here. The type of action in a Standard Transitive State of Affairs is highly transitive, that is, it involves true agent and patient referents and no referents with other semantic roles such as source, experiencer etc. (cf. Foley and Van Valin 1984: 59). The agent in a Standard Transitive State of Affairs is highly individuated 2 (e. g., definite, human etc.), and naturally salient because of its semantic role. It will therefore be considered to have the property of focus. The term focus is rather loosely defined as pertaining to that referent that is most salient in the State of Affairs under discussion. Saliency is determined basically by
Towards a typology of voice
47
pragmatics, and hence by context, topic continuity etc., but can also involve semantic notions like animacy. A third parameter is "conceptual status", a term introduced in Klaiman (1988). The term itself is not clearly defined by Klaiman, but its exponents, affected entity and controller, are: "by affected entity is meant the participant perceived as affected or most affected in consequence of the sententially denoted action ... controller status can be ascribed to the argument whose participation is seen as determining the course and/or outcome of a sententially denoted action" (Klaiman 1988: 28). The Conceptual Status parameter thus expresses the way a certain state of affairs is perceived. It is, as Klaiman shows, "neither a grammatical nor a thematic relation" (Klaiman 1988: 72), nor, we might add, a pragmatic function, but a parameter in its own right. In the Standard Transitive State of Affairs the agent has the property control and the patient is affected. We now have the following definition of a Standard Transitive State of Affairs: 3 A standard transitive state of affairs is: — a highly transitive action with two non-identical referents; — an individuated agent that is in focus and in control·, — a patient, which is affected. In many languages the encoding for the Standard Transitive State of Affairs will be the least marked option. 4
1.3. Non-standard states of affairs Having assumed the situation described above as standard, we will now look at a number of situations that deviate from that standard. The two referents in the Standard Transitive State of Affairs each have a number of properties. These same properties can have other distributions than they have in the Standard Transitive State of Affairs. They can, for example, be assigned to the same referents, but in a different combination, or to one or three referents instead of two. A non-standard distribution of properties results, of course, in a nonprototypical state of affairs. Many languages use non-standard encodings to express such non-prototypical States of Affairs. These non-standard encodings might be called the voice options of the language in question. However, languages do not all have the same possibilities in this respect. On the one hand, some languages do not use a non-standard encoding for a non-standard State of Affairs, whereas other languages do. These languages do not express
48
Ineke Brus
the difference between the standard situation and one that is non-standard, at least not by means of syntactic structure. On the other hand, languages that do use non-standard encodings often use one encoding to express different non-prototypical situations. Several combinations of States of Affairs are possible. These combinations are not arbitrary: the States of Affairs that are combined always have elements in common. In the remaining part of this paper I will first describe a number of situations that are non-standard and, as far as possible, give examples of languages that do indicate this in their syntactic structure, and examples of languages that do not. The latter has not always been possible, sometimes because some parameters (as those labeled "conceptual status") have not yet been sufficiently researched, sometimes because of lack of data, or perhaps because some States of Affairs can only be expressed in a non-standard encoding. After defining these non-standard States of Affairs I will give a number of combinations that are sometimes expressed in one encoding, and try to show why it should be possible for language to make these combinations.
2. The encoding of non-standard states of affairs The Standard Transitive State of Affairs has been defined in terms of three parameters. States of Affairs that are non-standard with respect to one of these parameters will now be discussed separately. This does not imply that the parameters are completely unrelated. The parameters termed "semantics" and "conceptual status" are closely connected, and could both be considered semantic notions. For the purpose of this paper, "conceptual status" is nevertheless defined as a separate parámeter. Yet it is beyond doubt that "conceptual status" and "semantics" are more closely related to each other than either of them is to "pragmatics".
2.1. Semantics A State of Affairs can differ from the Standard Transitive State of Affairs in terms of semantics. As we saw, a Standard Transitive State of Affairs includes two non-identical participants, an agent and a patient. A situation can obtain in which a State of Affairs is non-standard in this respect only.
Towards a typology of voice
49
2.1.1. Reflexive A first possibility is that a State of Affairs has only one participant, which is both agent and patient, creating a Reflexive State of Affairs. In most languages the Standard Transitive State of Affairs and the Reflexive State of Affairs do not have the same encoding. Cross-linguistically there are two major encoding options for the Reflexive State of Affairs. Either two referents are expressed, the form of the second referent indicating coreferentiality with the first, or only one referent is expressed, with a morphological indication on the verb that this referent should be interpreted as having the semantic roles of both agent and patient. English is an example of a language using the first strategy (1 b), while Aranda (Strehlow 1944)5 uses the second (2b): (1) a. b. (2) a.
b.
John hits Peter. John hits himselj. a¡a eripa tu-tji-na 1SG-TR 3SG-ACC hit-FUT-SG Ί strike him.' jiï[a tü-la-ma ÎSG-INTR strike-REFL-PERS Ί strike myself.'
(194)
(153)
There are a few languages that encode the Standard Transitive State of Affairs and the Reflexive State of Affairs in exactly the same way. Piraha is an example (Everett 1986): (3)
hi hi xibáobáhá he he hit 'He hits him' / 'He hits himself.'
(217)
In languages like Piraha ambiguity can arise, albeit only in situations involving third person participants. Even so, though Piraha can be said to have no unique reflexive construction, it still has a possibility to express a Reflexive State of Affairs. One additional remark has to be made. The interpretation of the "Reflexive State of Affairs" as given is straightforward enough. In some cases however, there is a complicating factor. A language sometimes uses both types of encoding, the choice between the two depending on the action described. Some actions can be interpreted as more likely done to oneself than to others. Typical examples are the so-called "grooming verbs", like wash, dress and shave. In such languages these actions are not encoded in a Reflexive State of
50
Ineke Brus
Affairs as other, more prototypically transitive actions are. In Ancient Greek (Risselada 1985), for example, a reflexive pronoun is used to express a Reflexive State of Affairs with a prototypically transitive action like "to harm": (4)
blaptö emauton I-harm-ACT myself-ACC Ί am harming myself.'
(125)
while more inherently reflexive actions are expressed in the so-called "middle encoding": (5)
louein louesthai wash-INF-ACT wash-INF-MIDDLE 'to wash (s. o. else)' 'to wash (oneself)'
(125)
One explanation for this difference in encoding might be that the second referent in the middle encodings is interpreted as affected entity in the sense of Klaiman (1988) rather than as a real patient. Klaiman's notion of "affected entity" as part of the "conceptual status" of a State of Affairs will be discussed in more detail in 3.2.1. 2.1.2. Spontaneous A second way in which a State of Affairs can differ from the Standard Transitive State of Affairs in terms of semantics is the situation where one of the two referents in the Standard Transitive State of Affairs is missing. "Missing" should be interpreted literally here. In a State of Affairs where the agent is missing it is not just unfocused, indefinite or non-referential, but there is simply no known or implied referent that causes or performs the action described in the State of Affairs. In the literature, the missing-agent State of Affairs has been labeled in many different ways, such as "middle", "neuter", "anti-causative" and "spontaneous". The label "spontaneous" (Shibatani 1985) describes, in my opinion, the essence of this situation and shall be used here from now on. The Spontaneous State of Affairs is less known than for example the Reflexive State of Affairs. This is, in all likelihood, to a large extent due to the fact that the number of (highly transitive) actions that are conceivable as "happening spontaneously" is limited. One sees the same set of verbs appear time and again when encodings of the Spontaneous State of Affairs are described — break, open, close, and boil are a few examples.
Towards a typolog) of voice 51 It is not unlikely that it is for this reason (limited occurrence of Spontaneous States of Affairs) that there are few, if any, languages that have an encoding that uniquely describes this State of Affairs. In English, verbs of this type can be used intransitively to express a Spontaneous State of Affairs: The window breaks, as opposed to a transitive use of the same verb for a Standard Transitive State of Affairs: I broke the window. In Duala (Epee 1976) a (productive) morpheme -ea exists that is used to encode the Spontaneous State of Affairs, as in (6 b). (6) a.
b.
o si kwasa bekwen ba mbot-ango you not tear-off buttons of clothes-your 'Do not tear the buttons of your clothes.' ekwen-am e ta e kwasea button-my AGR PAST AGR come-off 'My button came off.'
(101)
(103)
In Maasai a suffix -a indicates a Spontaneous, but also a Reflexive State of Affairs. Tucker and Mpaayei (1955: 135) label it "neuter form", and define it as follows: "Its main function is to express the state a person or object gets in without the help of an outside agency.... From this point the meaning may be stretched to cover "reflexive" action — no outside agent implying that the action was done by the subject itself." Examples: (7) a.
b.
c.
« -gil (135) 3sgS-break 'He breaks it.' e-gil-a 3sgS-break-NEUT 'It gets broken by itself / 'It breaks' e-isuj-a (^36) 3sgS-wash-NEUT 'He gets washed without outside help' / 'He washes himself.'
There are other languages in which the Spontaneous State of Affairs is encoded in the same way as the Reflexive State of Affairs. In Fulani the interpretation of the "middle" encoding (Macintosh 1984) depends on the properties of the patient. If the patient is animate, and thus fits for the semantic role of agent, the interpretation is reflexive, as in (8 b). If the patient is inanimate, the interpretation is spontaneous, as in (9 b): (8) a.
mi-suud-ii -ma I- hide- ACT -you Ί hide you.'
(109)
52
Ineke Brus
b.
(9) a.
b.
'a -suud -ake you -hide -MIDDLE 'You hide (yourself).' ' e-min-don-Beyd-a ceede maaje we-increase-ACT money their 'We're increasing their price.' ceede maaje 'e-don-ßejd-oo money their increase-MIDDLE 'Their price is going up.' / 'Their price increases.'
Some Western European languages also encode Spontaneous as Reflexive. French is an example: (10) a.
U tue Jean 3SG.NOM kill-3SG.PRES J. 'He kills John.' b. Jean se tue J. REFL kill-3SG.PRES 'John kills himself.'
(11) a.
b.
U ouvre la porte 3SG.NOM open-3SG.PRES ART door 'He opens the door.' la porte s'ouvre ART door REFL-open-3SG.PRES 'The door opens.'
2.2. Pragmatics A State of Affairs can differ from a Standard Transitive State of Affairs in terms of pragmatics. Logically and semantically speaking, the situation remains as in the Standard Transitive State of Affairs, the same action, the same referents having the same semantic roles. The difference between the Standard Transitive State of Affairs and potential deviants lies in the relative salience of the two referents. In the Standard Transitive State of Affairs the properties of focus and agent are united in one referent, or in other words, the agent in a prototypically transitive action will be the most salient referent in standard situations. However, situations can occur in which the agent is not most salient. This
Towards a typolog) of voice
53
causes a State of Affairs to be non-standard, which in its turn can lead to a non-standard encoding. An agent can be non-salient for two reasons. First, the agent itself can have properties that make it non-salient (or rather lack those that make it salient). Secondly, the other referent can have properties that make it more salient than the agent is. The distinction between these two possibilities has been made by Foley and Van Valin (1985: 149 — 168), among others. They use the terms "backgrounding" and "foregrounding" passive for what I will describe as "Impersonal State of Affairs" and "Focused Patient State of Affairs". In the following sections the Impersonal State of Affairs and the Focused Patient State of Affairs will be discussed separately and examples will be given of encodings that express only the Impersonal State of Affairs or only the Focused Patient State of Affairs. Encodings that can express both types of State of Affairs will also be discussed. 2.2.1. Impersonal A referent that has the property "focus" is salient in a given State of Affairs. A referent having the semantic role of agent is the most likely candidate for the property "focus", given that it has other prototypical properties such as definiteness, -(-human, etc. In the Standard Transitive State of Affairs the referent having the role of agent also has the property of focus. To be focused, i. e., to be regarded as most salient in the situation that is described, a referent must have certain properties that make it important enough to be in this position. Being human (or at least animate), as agents prototypically are, is an important factor, but not the only one. In order to be salient, a (human or animate) agent should also be identifiable. There are, however, situations in which it is not. Sometimes it is clear that an agent exists, but not who it is, as in the general was murdered in his study. Alternatively, it is possible that an agent exists, but that it is not important enough to identify it in the situation, as in this letter was delivered an hour ago. In these situations it is clear that there is an agent, but it is not identified. I will refer to the situations described above as the Impersonal State of Affairs. Languages can encode this State of Affairs in a way differing from that of the Standard Transitive State of Affairs. In English, for example, the passive construction is often used to encode the Impersonal State of Affairs, as can be seen in the examples above. Therefore, labels that are used to describe encodings of the Impersonal State of Affairs often involve the term "passive"
54
Ineke Brus
or "impersonal passive" even though they differ structurally from the passive as we know it in English. A language that uses a unique encoding for the Impersonal State of Affairs is Ika. In Ika the verbal suffix -kuma codes "impersonal agent". Frank (1985: 148) states: "-kuma makes reference to some unidentified non-focal participant as subject." (In this context "agent" would be a more appropriate term than "subject"). Compare: (12)
in'qui kartutsu a'we-kuma-na gui nän-na one shell give-IMPERS-DIST also be-DIST Ά shotgun shell had been given to him.'
(148)
Koasati (Kimball 1985) uses the morpheme ho-¡oh- to encode this State of Affairs. This morpheme has a more general meaning. It is described by Kimball (1985: 125) as basically indicating "subjects or objects that are multiple or scattered over a space". (This is Kimball's terminology, in this context the terms "agent" and "patient" instead of "subject" and "object" would be preferable.) Thus, it is a morpheme used to indicate non-individuated referents in general. A specific use of this morpheme is the encoding of a non-individuated agent and thus of an Impersonal State of Affairs, as can be seen in the following example: (13)
mó':-toho-n halkí oh-í:bi-to-V be-then-SR wife DISTR-kill-3SGPAST-PHRTERM 'After that someone killed his wife.'
(127)
A last example of an encoding for the Impersonal State of Affairs is from Finnish. In many Finnish grammars this form is called "passive". Lehtinen (1963: 237) labels it as "indefinite person form", and explains that it indicates "that an action has a personal performer, but the number and person of the performer is not specified". She adds that "the indefinite person can often be translated into English as a passive". (14)
täälä puhutaan suome-a here speak-IMPERS Finnish-PART 'Finnish is spoken here.'
(238)
Thus, we see that languages like Ika, Koasati and Finnish have a nonstandard encoding for the Impersonal State of Affairs. These encodings are sometimes called "passives". In contrast to those just mentioned, quite a number of languages do not encode this State of Affairs in a manner different from the Standard Transitive State of Affairs. Such languages encode an Impersonal State of Affairs as
Towards a typology of voice
55
they would encode a Standard Transitive State of Affairs. A non-salient agent is encoded as a non-emphatic pronoun, in most cases a third person plural, as this is the least individuated Person/Number combination. In English this encoding for Impersonal State of Affairs is possible next to the passive construction mentioned above: They murdered the general in his study and They delivered this letter an hour ago. In grammars this encoding is sometimes explicitly mentioned as a replacement for a non-existing passive. When discussing the non-existence of the passive in Aranda, for example, Strehlow (1944: 111) remarks: "...ignorance of the active subject... is easily disposed of in Aranda. Where the subject is genuinely unknown Aranda either uses the third personal pronouns or omits them altogether." Examples (15) and (16) are from Aranda and Yoruba (Ashiwayu 1968) respectively: (15)
etna eriya tuma 3PL 3SGO kill-PRES 'They kill him.' / 'He is killed.'
OH)
(16)
nwqn mû çkunrin kan láná (49) 3PL.NOM.NONEMPH arrest man ART yesterday 'They arrested a man yesterday.' / Ά man was arrested yesterday.'
Languages that can have non-overt subjects when they are non-emphatic express the impersonal State of Affairs without an overt subject, as can be expected. Examples (17) and (18) are from Gumbainggir (Eades 1979) and Mandarin (Li and Thompson 1981) respectively: (17)
biyagay jiaiivar) gulama not see-PAST 3SGO 'He has never been seen.' (lit: 'No one has seen him.')
(18)
nèi-bên shü chübän LE that-CL book publish PFV/CRS 'That book, (someone) has published it.'
(312)
(88)
2.2.2. Focused patient In the previous section we have seen that an agent can be non-salient when it is unknown or unimportant. Another reason for the agent to lack the property of focus can be that the other referent, the patient, is more salient in a particular State of Affairs and therefore gets assigned the property of focus. This situation will be called the Focused Patient State of Affairs. This Focused Patient State of Affairs is similar to the Impersonal State of Affairs
56
Ineke Brus
in that in both States of Affairs the agent is not in focus. The difference between the two lies in the motivation for the agent's loss of focus. In the Impersonal State of Affairs the agent is not focused because it is not individuated in its own right, in the Focused Patient State of Affairs it is the patient that causes the agent to lose focus. Another difference is that the Impersonal State of Affairs does not need to have a focused patient, whereas the Focused Patient State of Affairs by definition must. The fact that the two States of Affairs have an important element in common, viz., a non-focused agent, makes it difficult, or perhaps in some cases impossible to distinguish between their encodings. An encoding uniquely indicating a non-referential agent is, as is shown above, in some cases easily interpretable. Encodings clearly indicating that the patient has the property of focus without implications for the agent are less easy to find. In Foley and Van Valin (1985) a number of encodings for these two States of Affairs are discussed. One encoding which Foley "and Van Valin present could be considered as uniquely describing the Patient Focus State of Affairs. In the following construction from Palauan the patient is cross-referenced in the verb by the same suffix as in the Standard Transitive State of Affairs, but, unlike in the Standard Transitive State of Affairs, it has sentence initial (i. e., focus) position. The agent is marked in the verb by a prefix in the Patient Focus State of Affairs (19 b) and can therefore not be non-referential, while it is not crossreferenced at all in the Standard Transitive State of Affairs encoding (19 a): (19) a.
a
?ad
a mos-terir
a
ART man see-3PLHUMAN.UNDERGOER 'The man saw the children.' b-
ngalek
ART child (144)
a ngalek a le-bos-terir a ?ad ART child PASS/3SG-see-3PLHUM.UND ART man 'The children were seen by the man.'
Another type of construction that could be seen as specifically encoding a Patient Focus State of Affairs is the "goal-focus" constructions as found in Philippine languages. In older grammars this construction was sometimes also labeled "passive". The following example is from Tagalog (Schachter— Otanes 1972): (20) a.
bumabasa
ng
diyario
ang
titser
read-ACT.FOCUS ART newspaper T O P teacher 'The teacher is reading a newspaper.'
(69)
Towards a typology of voice
b.
57
binabasa ng tit ser ang diyario read-GOAL.FOC ART teacher TOP newspaper 'The teacher is reading the newspaper'
The fact that in this type of constructions agents are overtly expressed in the majority of cases (Shibatani 1988: 93; see 2.2.3. below) makes it likely that they are first of all an encoding of a patient focus State of Affairs. In spite of this, an Impersonal State of Affairs in languages with this type of construction will be encoded in the same goal-focus construction albeit without an overt agent. This possibility, however, is mentioned as a kind of "afterthought". Schachter—Otanes (1972: 73) observe about Tagalog: "In the case of most goal-focus verbs, the actor may be omitted, if the sentence expresses an action performed by unknown persons, or if the performer of the action is irrelevant." Although goal-focus constructions can encode both the Patient Focus and the Impersonal State of Affairs, their basic/primary function seems to be the encoding of the Patient Focus State of Affairs. 2.2.3. Passive constructions In the previous section we saw that the label "passive" is used for constructions expressing two different States of Affairs: The Finnish "passive" in (14) encodes the Impersonal State of Affairs. Since the patient in this construction does not have the property of focus, the construction does obviously not encode the Patient Focus State of Affairs. In the Palauan "passive", on the other hand, the patient is in focus position. The construction can therefore express the Patient Focus State of Affairs. The agent is cross-referenced in the verb and cannot be dropped — from which one can conclude that this construction cannot encode the Impersonal State of Affairs. There are many languages that encode these two functions in one single construction. Among these are the canonical passives we know from English and other Western European languages, but also constructions like the Tagalog goal-focus construction discussed above. Combining the properties of the Impersonal State of Affairs and the Patient Focus State of Affairs, we can say that a construction which has a focused patient and the possibility to have an implied rather than an expressed agent can potentially express both the Focused Patient and the Impersonal State of Affairs. This situation obtains for the Tagalog sentence (20 b) — where, as Schachter —Otanes explain, an unidentified agent can be dropped, and for English (21 b):
58
Ineke Brus
(21) a. b.
John hit the dog. The dog was hit (by
John).
There is, however, a difference between the Tagalog and the English construction that only shows in text-counts: 80% of Tagalog goal-focus constructions have an overt agent. This indicates that the Tagalog construction is primarily an encoding of the Focused Patient State of Affairs. Since Tagalog can leave out non-focused elements, an additional function of the construction is the encoding of the Impersonal State of Affairs. The English passive does not differ much from the Tagalog construction at first sight. Both have a focused patient, a verb form indicating that the patient is in focus, and an optionally expressed agent. Yet there is a significant difference. Whereas in Tagalog the majority of goal-focus constructions have an overt agent, 80 — 85% of English passives do not (cf. Shibatani 1985: 835). From this one might conclude that the English passive expresses an Impersonal State of Affairs rather than a Patient Focus State of Affairs. Yet the English construction does have a focused patient. A possible explanation would be that English, unlike Finnish, has to have a filled subject-slot. When an unidentified agent does not fill this slot, the next best candidate, the patient, does. In this way the English construction, by default, expresses the Patient Focus State of Affairs as well as the Impersonal State of Affairs. In this section it was shown that constructions labeled "passive" are possible encodings of the Impersonal State of Affairs (as in Finnish), of the Patient Focus State of Affairs (as in Palauan), or of both (as in Tagalog or English). 6
2.3. Conceptual status As defined in the first part of this paper, "conceptual status" expresses the way in which an action is perceived. The notions considered here as part of the "conceptual status", control and ajjected entity, are prototypically connected with the semantic notions of agent and patient respectively. 7 Klaiman (1988) stresses that it is important to consider "conceptual status" as a notion in its own right, differing from both semantics and pragmatics. Unique encodings exist for States of Affairs indicating that the agent is not the controller or the patient is not the affected entity. Even though more research is needed to flesh out the details, it is worthwhile discussing the notion "conceptual status" since differences with the Standard Transitive State of Affairs on this level can be established.
Towards a typology of voice
59
2.3.1. Affected agent First, a State of Affairs can obtain in which not the patient, but the agent is perceived as being most affected by the action, 8 while on the level of semantics, i. e., agent and patient roles, the two referents remain as in the Standard Transitive State of Affairs. The agent is doing something for i t s e l f , not to i t s e l f , as in the Reflexive State of Affairs. Tamil encodes this State of Affairs using the "weak voice", while the "strong voice" encodes the Standard Transitive State of Affairs: "In Tamil, ... the source-controller 9 of action does generally bear the subject relation in a non-derived sentence, 10 while voice indicates whether this argument is also viewed as the affected entity." (Klaiman 1988: 46) The choice of strong or weak voice in the following pair of Tamil sentences is purely determined by the perception of the act as primarily affecting either the agent or the patient. In (22 a) the act of embracing is interpreted as primarily affecting the mother, while (22 b) stresses the fact that the child is affected by the embrace: (22) a.
b.
täj kulantaiyai mother-NOM child-ACC 'The mother embraced the täy ku^antaiyai mother-NOM child-ACC 'The mother embraced the
anain-t-äl embrace-WEAK-PAST.AGRT child.' anai-tt-äl embrace-STRONG-PAST.AGRT child.'
(43)
Klaiman is careful to point out that both passives and reflexives are encoded differently in Tamil: (23)
avan tannai utaittuk kontän (42) he-NOM self-ACC kick-STRONG-PTCPL took(RFL.VERB) 'He kicked himself.'
(24)
cannai ciru paiyanäl utai-kk-ap window-NOM little boy-INSTR break-STRONG-PTCPL pattata fell 'The window was broken by the boy.'
(45)
A second example of a language which has a specific encoding for an Affected Agent State of Affairs is Ancient Greek (Risselada 1987). The encoding is labeled "middle" in traditional grammar; even though the Greek middle can encode other situations, the Affected Agent State of Affairs is
60
Ineke Brus
often regarded as the most characteristic situation the middle encoding expresses. Compare: (25)
ho strato s paraskeua^etai tas naus the army-SUBJ prepares-MIDDLE the ships-GOAL 'The army is preparing the ships for itself.'
(124)
2.3.2. Causative A second exponent of conceptual status is "controller". I have assumed that in the Standard Transitive State of Affairs the referent that is agent also has the property "controller". A State of Affairs is possible in which this is not the case. Assuming that there is a controller, but that it is not the agent, 11 there are two logical possibilities: either there is a third referent having the property "control", or the patient is seen as having control or causing the action described. Encodings for the first option, i. e., a State of Affairs where not the agent, but a third party has control are usually labeled "causative". In these encodings it is always the controller that retains primary syntactic position, being "subject" in most languages. 12 Possible encodings for the Causative State of Affairs are a verb expressing causativity and an embedded sentence as in English (26), or a marking in the verb, as in Turkish (Lewis 1967) (27): (26)
I got the director to sign the letter.
(27)
mektub-u müdür-e in^ala-t-tim letter-DEFACC director-DAT sign-CAUS-lSGPAST Ί got the director to sign the letter.'
(147)
2.3.3. Controlling patient A second option is that the patient is considered as having control over the situation, or in other words, the patient itself is to some extent held responsible for what is happening to it. This Controlling Patient State of Affairs can be encoded in constructions labeled "catalytic passives" (cf. Barber 1975). Such an encoding can be a combination of Causative and Reflexive. The Reflexive State of Affairs indicates a coreferentiality of the patient with the primary referent, which, in a Causative encoding, is the controller. In this way the combination of Causative and Reflexive encodings indicates a controlling patient. The reflexive ¿«/-passive in English encodes the Controlling Patient State of Affairs:
Towards a typology of voice
(28) a. b. c. d.
John John John John
appointed his friend. got his friend appointed. appointed himself. got himself appointed.
61
(Standard State of Affairs) (Causative) (Reflexive) (Catalytic Passive)
An example of a language using a verbally marked combination of causative and reflexive to express a Controlling Patient State of Affairs is Tarma Quechua (Adelaar 1977): (29)
aman warmi-ivan-xa maxa-ci-ku-n-mañ-cu (98) not woman-INSTR-TOP beat-CAUS-REFL-3sg-POT-NEG 'He should not let himself be beaten by a woman.'
Another example is West-Greenlandic (Fortescue 1984). In West-Greenlandic the encoding for the Causative State of Affairs is expressed by means of a verbal suffix. The Reflexive State of Affairs can be expressed by using intransitive inflection on transitive verbs. A combination of the causative encoding with intransitive inflection forms the "pseudo-passive" construction. Fortescue's remark that "in suitable contexts this construction may imply an element of purposefulness [on the part of the patient, I. B.]" (Fortescue 1984: 265) shows that this construction encodes a Controlling Patient State of Affairs, as could be expected. It seems, however, that this encoding is also used to encode States of Affairs mentioned under "pragmatics", i. e., the Impersonal State of Affairs or the Focused Patient State of Affairs. WestGreenlandic does have another passive construction, but: "the pseudo-passive with -tit is often colloquially preferred." (Fortescue 1984: 266) (30) a.
b.
(31) a.
b.
(32)
miiqqat paaraakka (269, Standard Transitive State of Affairs) children-ABS look.after-lsg.3pl.INDIC Ί looked after the children.' miiqqaat uan-nut paari-tip-pai 269, Causative) children-ABS me-ALL look.after-CAUS-3sg.3pl.INDIC 'He had me look after the children.' tuqqur-paa hide-3sg.3sg.INDIC 'He hid it.' tuqur-puq hide-3sg.INDIC 'He hid himself.'
(157, Standard Transitive State of Affairs)
(qimmi-mut) kii-sip-puq (dog-ALL) bite-CAUS-3sg.INDIC 'He got bitten (by a dog).'
(157, Reflexive)
(265, "pseudo passive")
62
(33)
Ineke Brus
nanuq taku-tip-puq polar.bear see-CAUS-3sg.INDIC 'The polar bear let itself be seen.'
(265, Controlling patient)
As we have seen, the Controlling Patient State of Affairs is a situation in which a controller and a patient are coreferential. This leaves two referents in the State of Affairs, an agent and a controller/patient. This State of Affairs could be considered a combination of the Causative State of Affairs and the Reflexive State of Affairs, and can indeed be coded as such in West-Greenlandic and Tarma Quechua. One result of this way of encoding is that the referent having primary syntactic position, i. e., the controller in Causative encodings, also has the semantic role of patient (but not that of agent, as in "ordinary" reflexives). Since in constructions of this type the primary referent has the role of patient, they can apparently in some cases come to encode a Patient Focus State of Affairs, when the semantic role of this referent is considered its most important feature. This is what seems to have happened in West-Greenlandic, as in example (32) above. 2.3.4. Dative Having interpreted the notion "control" as possibly pertaining to the agent (in a Standard Transitive State of Affairs), the patient (in a Controlling Patient State of Affairs) or a third referent (iin a Causative State of Affairs) we now go back to the notion of Affected Entity discussed earlier. As mentioned, the notion "affected entity", can pertain to the patient (in the Standard Transitive State of Affairs), or the agent (in the Affected Agent State of Affairs), but also, in parallel to the possibilities given for the notion "controller", to a third referent. In this State of Affairs, it is neither the agent nor the patient that is seen as being most affected by the action, but a third referent. The encoding for a referent having the property of affected entity, but not the semantic role of patient or agent, is often labeled "Dative". 13 In English a Dative State of Affairs can be encoded using either a noun phrase or a prepositional phrase to describe the affected referent: (34)
I bought him a present.
(35)
I cooked a meal for my father.
In some languages the Dative State of Affairs can be encoded in the verb, as for example in Swahili (Ashton 1966) and Maasai (Tucker—Mpaayei 1955).
Towards a typology of voice
63
In Swahili, the "Prepositional or Applied form" expresses "to do for, or on behalf of someone, or to the detriment of such a one" (Ashton 1966: 218). In Maasai the "Dative form" describes "action for somebody else" (Tucker — Mpaayei 1955: 130). In both languages it is the affected referent that retains the position that is held by the patient/affected entity in the Standard Transitive State of Affairs, as it does in English (34). Examples are from Swahili (36) and Maasai (37) respectively. (36)
ni -ku -pikie chakulaì lsgS -2sgO -cook-APPL food 'Shall I cook some food for you?'
(218)
(37)
á -bòi -ókì papá olbéné lsgS -open -DAT father basket Ί opened the basket for father.'
(129)
Keeping in mind the Affected Agent State of Affairs it appears that, just as the Controlling Patient State of Affairs can be interpreted as a Causative State of Affairs with a coreferential controller and patient, the Affected Agent State of Affairs can be interpreted as a Dative State of Affairs in which the agent and the affected entity are coreferential. The English translation for the Greek middle encoding in (25) clearly supports this: 'The army is preparing ships for itself — a Dative State of Affairs, with a reflexive pronoun indicating coreferentiality of agent and affected entity.
2.4. Summary All the States of Affairs discussed in the preceding paragraphs can be summarized in the following table showing what properties the different referents have in each case. The table also brings out the fact that Reflexive, Affected Agent and Controlling Patient are coreferential versions of the Standard Transitive State of Affairs, Dative, and Causative respectively. In Table 1, R stands for Referent, STSOA for Standard Transitive State of Affairs, A for Agent, C for Controller, Ρ for Patient, and AE for Affected Entity. If there is a Focused referent, it is R l .
64
Ineke Brus
Table 1. RI STSOA REFLEXIVE DATIVE AFF. AGENT CAUSATIVE CONTR. PATIENT SPONTANEOUS IMPERSONAL FOC. PATIENT
A/C A/C A/C A/C C
c
R2
— coref — — coref — — coref —
P/AE P/AE P/AE
P/AE P/AE AE AE P/AE P/AE — -
R3 -
Ρ Ρ A A —
(A implied) A
3. Combined States of Affairs Having listed a number of non-standard States of Affairs that are associated with the notion of "voice", we will now have a look at their interrelations. As we have seen in the examples presented here, in many cases two or more of these States of Affairs can be expressed by the same encoding. I will now list a number of possible combinations 14 and try to show why a particular combination is likely to occur. It is, however, by no means the case that all languages combine States of Affairs in the same manner. In the previous paragraphs combined encodings within the different levels have been mentioned already, viz., combinations of Reflexive and Spontaneous, and of Impersonal and Focused Patient. Those combinations, and some other combinations pertaining to different levels, will be discussed in the following sections.
3.1. Spontaneous and Reflexive I have already mentioned the fact that Spontaneous is often coded as Reflexive, as in Maasai (7 b, c) and Fulani (8 b), (9 b). This connection between Reflexive and Spontaneous seems to be a logical one. In both cases there is only one referent, which has the semantic role of patient. The difference between the two is that this referent is also the agent in the Reflexive, while in the Spontaneous there is no agent at all. In this light it seems logical that the
Towards a typology of voice
65
animacy of the referent, and thus its potential to fulfill an agent role, influence the interpretation of a particular instance of such a Reflexive/Spontaneous encoding. This is the case for Fulani (cf. 2.1.2). It would be interesting to know whether this would also be the case in other languages having this combination. My data are not clear enough to confirm this with any certainty. Although Spontaneous/Reflexive is a relatively common combination, there are also languages that express them differently. An example is Classical Arabic (Saad 1975). In Classical Arabic, Spontaneous is encoded by means of a verbal prefix 'in, while a reflexive pronoun in a transitive frame expresses the Reflexive: (38)
'inkasara
al-^ujäju
(190)
'in-PAST-break the-glass 'The glass broke.' (39)
qatala
Zaydun
nafsahu
(200)
kill-PAST Z. himself 'Zaydun killed himself.' The same situation obtains in Duala, which expresses the Spontaneous State of Affairs by means of a verbal morpheme (cf. (6 b)), but encodes the Reflexive State of Affairs in a transitive construction with a reflexive pronoun: (40)
moto
a
pane
momene
o
bwele
(138)
man AGR hang himself LOC tree Ά man hanged himself from a tree.'
3.2.
Semantics and Conceptual Status
3.2.1. Affected Agent and Reflexive One might also expect that the Affected Agent State of Affairs and the Reflexive State of Affairs could be encoded in the same way. In the Affected Agent State of Affairs the agent is also the affected entity, while in the Reflexive State of Affairs the agent is not only affected, but also patient (or "goal"). As we have seen in section 2.1.1., the middle encoding in Ancient Greek can also be used to express Reflexive, but only in the sense of "inherent reflexive" as discussed above. Thus louesthai 'to wash (oneself)' and keiresthai 'to shave' have middle forms, while blaptô emauton Ί am harming myself has an active verb form and a reflexive pronoun (Risselada 1987: 125). An explanation for this combination might be that the referent is seen as doing
66
Ineke Brus
something for himself rather than to himself, so that the coreferential relation is with an affected entity rather than with a patient, as it is in the Affected Agent State of Affairs. I have not yet found languages that clearly encode Affected Agent and Reflexive States of Affairs of all transitive actions in the same way. It would be interesting to know whether such languages exist at all. 3.2.2. Affected Agent and Spontaneous In Tamil (Klaiman 1988), the weak diathesis encodes the Affected Agent State of Affairs as sentence (22 a) shows, whereas the Reflexive State of Affairs and the passive (Impersonal and/or Patient Focus State of Affairs) are expressed differently, as shown in examples (23) and (24). From one of Klaiman's examples it appears that the Spontaneous State of Affairs can also use the weak diathesis: (41)
cannai utain-t-atu window-NOM break-WEAK-PAST-AGRT 'The window broke.'
(45)
This shows that in Tamil the Affected Agent State of Affairs and the Spontaneous State of Affairs can be encoded in one construction. 3.2.3. Affected agent, Reflexive and Spontaneous From Klaiman's data on Sanskrit one can infer that this language uses one encoding for Affected Agent (42 b), Reflexive (43), and Spontaneous (44): (42) a. b.
katam karoti 'He makes-ACTIVE a mat.' katam kurute 'He makes-MIDDLE a mat (for himself).'
(34)
(43)
Satyakäma èva jäbälo bravithäh (33) 'You must call-MIDDLE (i. e., call yourself) Satyakama Jabala.'
(44)
namate dandah bends-MIDDLE stick 'The stick bends.'
(36)
Although the reflexive example is not quite clear ("call" is not a prototypical transitive verb), the affected agent and spontaneous are. A similar situation exists for Ancient Greek, where the middle can also encode spontaneous (cf. Risselada 1987: 127).
Towards a typology of voice
67
3.3. Pragmatics and other levels The pragmatic types of non-standard States of Affairs, "Impersonal Agent" and "Focused Patient", can also be encoded in combination with one or more of the above, or in combination with each other. It has been proposed by Foley and Van Valin (1985: 148), among others, that encodings for "passive" in many cases originally expressed other States of Affairs as mentioned above and extended their scope to the pragmatic level. In the following paragraphs I will discuss some of the languages that seem to combine States of Affairs from pragmatic and other levels, and try to indicate how the relations between the different States of Affairs could be structured. 3.3.1. Spontaneous and Impersonal In section 2.1.2., it was shown that the Reflexive State of Affairs and the Spontaneous State of Affairs can be expressed in a single encoding. Another possible combination of encoding is one for the Spontaneous State of Affairs and the Impersonal State of Affairs. This combination seems logical, since it can be expected that the difference between a situation where no agent is implied (the Spontaneous State of Affairs) and one where there is an implied, but unknown and unexpressed agent (the Impersonal State of Affairs) is not necessarily always clear-cut. In Duala (Epee 1976) the encoding for the Spontaneous State of Affairs can in some cases express a situation that can only be interpreted as implying an unknown agent, and thus as encoding an Impersonal State of Affairs: (45)
Cornell ba ta ba dipea sondi ni tombi C. AGR PAST AGR beat-EA Sunday which pass 'Cornell got beaten last Sunday.'
(102)
In addition to the -ea construction, Duala also has a "passive" construction, that is encoded by means of the morpheme -be·. (46) a.
b.
na longi bolongi I build house Ί build a house.' bolongi bo longa-be (na mba) house AGR build-PASS (by me) 'The house was built (by me).'
(115)
Assuming that this latter encoding can express the Impersonal State of Affairs and the Patient Focus State of Affairs, Duala would have two ways
68
Ineke Brus
to express the Impersonal State of Affairs. The use of these constructions in actual speech would have to show what determines the choice between the two possibilities. One would expect that the more "unknown" an agent is in a particular situation, the more likely the choice of -ea would be. 3.3.2. Reflexive and Impersonal In section 3.3.1 we established the existence of Spontaneous/Reflexive and Spontaneous/Impersonal combinations. Now we will turn to another possibility, a combination of the Reflexive State of Affairs and the Impersonal State of Affairs. Some languages do combine the expression of these two States of Affairs in one encoding, which is sometimes labeled "pseudopassive" in the literature. Hixkaryana (Derbyshire 1979) and Turkish (Lewis 1967) provide examples in (47 b) and (48 b) respectively: (47) a
b.
(48) a.
b.
biryekomo yompamnobyako Nonato children-COLL he-taught-them N. karaywa rwon hoko non-indian talk-of concerning 'Nonato was teaching the children Portuguese.' n-os-ompa-hohyatxoko they-DETR-teach-RPASTCOMPLCOLL biryekomo karaywa rwon hoko children-COLL non-indian talk-of concerning 'The children were taught Portuguese.' 'The children taught themselves Portuguese.' çocugu yika-di child-ACC wash-3SGPAST 'She washed the child.' çocuk yika-n-di child-NOM wash-PASS-3SGPAST 'The child was washed.' 'The child washed itself.'
(62)
(150)
The Reflexive State of Affairs and the Impersonal Agent State of Affairs have a common denominator in that there is only one referent expressed. In the Reflexive, this referent has both agent and patient roles, in the Impersonal Agent State of Affairs, it functions as patient only, while an agent is implied but not expressed. In cases like the above, where the two are coded in the same manner, one can expect a third State of Affairs, the Spontaneous, also to be encoded in this way, since it encodes a situation that is in between
Towards a typology of voice
69
Reflexive and Impersonal. A part of Table 1, put in a somewhat different order, might clarify this. Table 2. R1 REFLEXIVE SPONTANEOUS IMPERSONAL FOC. PATIENT
A/C P/AE P/AE P/AE
R2 — coref —
P/AE —
(A implied) A
We have already seen how Reflexive and Spontaneous can combine in one encoding. Going from Spontaneous, having no agent, to Impersonal, having an agent, but one that is only implied and not expressed, is only a small step. None of my data contradict the assumption that, when an encoding expresses both Reflexive and Impersonal, it can also encode Spontaneous. 15 Lack of data however makes it impossible to be certain in every single case. For Hixkaryana and Turkish the assumption is corroborated by the facts. Another translation for the Hixkaryana sentence (47 b) is 'The children learnt Portuguese.' The Turkish sentence (49) encodes a Spontaneous State of Affairs (Van Schayk 1985: 129): (49)
pencere
aç-tl-dt
window open-PASS-3SGPAST 'The window opened.' / 'The window was opened.' I must point out that I have used the term "Impersonal" in this paragraph where most grammars use "passive", since they do not distinguish between the two States of Affairs brought together under this label. In my opinion it is Impersonal rather than Focused Patient that is expressed in the encodings discussed in this paragraph. There are two arguments for this assumption. First, the combination Reflexive/Spontaneous/Impersonal indicates a step-bystep transition, where Focused Patient suggests one further step, i. e., from "implied" to "expressed" agent (see Table 2). Secondly, most languages that have one encoding for this combination of States of Affairs do not allow the expression of an agent at all, or if they do, only marginally. In Turkish, for example, an agent can be expressed with the postposition tarafindan (lit. 'from the side o f ) , but this is judged to be unnatural by speakers of the language. This fact would also point to an Impersonal interpretation. For those languages marginally allowing such an expressed agent it could be assumed that they take the range of States of Affairs that such an encoding can express one step further: from Reflexive to Spontaneous to Impersonal to Focused Patient.
70
Ineke Brus
3.4. Pragmatics Within this level we have distinguished the Impersonal and the Focused Patient State of Affairs. Examples have been given of languages having an encoding for either one of them (as Koasati or Palauan) or for a combination of the two (as the European "canonical passives"). Foley and Van Valin (1984) give examples of both "foregrounding" and "backgrounding" passives that have also come to express the complementary function. A thorough discussion of these examples would lead to a detailed discussion of syntactic phenomena and is therefore beyond the scope of this paper. There is one more point I would like to make, however. The expression of an agent in a "passive" encoding seems to me to indicate that such a particular instance of this encoding is more likely to express Focused Patient than one without an agent. Because of the very fact that the agent is expressed, it can never encode the Impersonal. 16 Taking this possibility of expressing an agent as a criterion for the interpretation of an encoding as "expressing the Focused Patient State of Affairs", some encodings for the Impersonal State of Affairs can be interpreted as having taken the possibility to express Focused Patient as well. In Welsh, for example, an "impersonal form" is given in Morris Jones (1913): "Each tense has in addition [to third persons singular and plural, I. B.] an impersonal form, whose implied indefinite subject means 'someone', 'they' ": (50)
eerir fi love-IMPERS 1SGACC Ί am loved.'
(316)
Nowhere does Morris Jones mention the possibility of expressing an agent with this form, which would, of course, be in contradiction with the interpretation he gives. Awbery (1976), however, gives the same form as an ordinary (i. e., agentive) passive, with an expressed agent: (51)
Rhybuddiwyd fi gan y dyn warned-IMPERS 1SG-INDEP. PRON by the man 'was warned by the man.'
(148)
Awbery also mentions the status of the patient as being unclear, having both subject and object properties, whereas according to Morris Jones "the impersonal form takes after it pronouns in the accusative case." In Morris Jones' interpretation this Welsh form clearly encodes an Impersonal State of Affairs, while Awbery's data suggest that a Focused Patient
Towards a typology of voice
71
interpretation is possible. Taking into account that the Welsh Morris Jones describes is that of an earlier period than Awbery's Welsh, a possible explanation would be that an extension of the encoding possibilities has taken place, so that this encoding can now express both Impersonal and Focused Patient. Encodings for the Impersonal State of Affairs in languages that have no "passive constructions" (see examples (15) and (16)) can in some cases undergo the same extension. Shibatani (1985: 845) gives some examples of languages where this seems to have happened. In these cases the verb is marked for 3PL Agent-Subject (just as in the examples in (15) and (16)) and patientobject, but the patient is in subject-position and the agent appears in an oblique case. Shibatani quotes two examples, Kimbundu (quoted from Givón (1979: 211)): (52)
n^ua a-mu-mono (kwa mane) John they-him-saw (by me) 'John was seen (by me).'
and Trukic (from Jacobs (1976: 121)): (53)
Waan re-liila-β ree-t John they-kill- him by-me 'John was killed by me.'
Examples of this kind indicate that an encoding for the Impersonal State of Affairs in some cases also expresses the Focused Patient State of Affairs, if the fact that an agent can be expressed is taken as a sign that such an encoding does express such a situation. As I said before, only a study of the way in which such constructions are used in actual speech can really solve this problem. With the information that is available at this point we can only speculate. In the preceding paragraphs I have shown different ways in which languages combine States of Affairs. I also tried to show what elements different situations have in common and offered some speculations about combinations like Reflexive/Spontaneous/Impersonal.
4. Conclusion In this paper I have defined voice as a reaction of language structure to a number of situations that deviate from an assumed standard situation. I have called those situations Standard and Non-standard States of Affairs.
72
Ineke Brus
In this way I have tried to distinguish a number of situations that can be encoded in so-called "voice constructions". These same situations can in some cases be expressed in encodings that are not normally considered "voice options". This, and the fact that many languages use a single encoding for several States of Affairs often make it difficult to recognize which encoding expresses which State of Affairs. Yet in my opinion it is useful to make an attempt at separating and defining functions first, and to form an idea about how and why they should be related before looking at the exact forms in which such functions can be expressed. Such morphosyntactic explorations have not been pursued in this paper. This being a first attempt, I am aware of the limitations of the outline presented here, limitations in the scope of this paper, in my own knowledge of several areas I touched upon, but also in the data that are available. With respect to the data, one of the problems has been that of labeling. Middle, neuter, passive, impersonal, semi- and pseudo-passive are some of the labels that are used for encodings expressing the States of Affairs that I have described. This often causes problems in understanding exactly which State of Affairs is expressed in a particular encoding. Another difficulty is the often total lack of information about the actual use of these encodings in spoken or written language in older traditional grammars and grammars written in one of the many formalistic frameworks. This type of information is indispensable for further study, especially, but not only, where pragmatic functions are concerned. (It would, for example, be interesting to know whether all languages make an equally extensive use of their "passives" — which, I suspect, is not the case.) As for the scope of this paper, I have mentioned those aspects that I could relate to each other at this point. There are some other aspects that have not been discussed and that I will just mention briefly: 1. In a number of languages (notably well-known specimens like English) "passives" are — at least in the form of their encoding — related to stative or perfective expressions. 2. Ergative constructions and antipassives have not been mentioned at all. They should be taken into account, especially where discourse functions are concerned. Since this is an outline of a rather extensive area of grammar, it was impossible to avoid generalizing, not only for lack of space, but also because my own knowledge of some of the aspects I touched upon is too limited. Some of the definitions, as those of focus, non-referentiality, causative and "dative" are grossly oversimplified and certainly need a good deal of refinement before anything more definite can be said. All the same, I hope that the perspective developed here can lead to some new empirical questions.
Towards a typology
o j voice
73
Notes 1. I am grateful to Ann Cooreman, Ad Foolen, Leon Stassen and Harrie Wetzer for helpful comments on earlier versions of this paper. In the examples in this paper the following abbreviations are used: ABS — absolutive, ACC — accusative, ACT — active, ACT. FOCUS — actor focus, A G R ( T ) — agreement, A L L — allative, APPL — applicative, A R T — article, CATJS — causative, C O L L — collective, C O M P L - completive, DAT - dative, DEF - definite, D E T R - detransitivizer, DIST — distal, DISTR - distributive, F U T - future, GOAL FOC - goal focus, I M P E R S impersonal, INDEP. PRON — independent pronoun, INDIC — indicative, INF — infinitive, INSTR — instrumental, INTR — intransitive, LOC — locative, NEG — negative, N E U T - neuter, N O M - nominative, N O N - E M P H - non-emphatic, O - object, PART — partitive, PASS — passive, PFV/CRS — perfective/currently relevant state, P H R T E R M — phrasal terminator, PL — plural, POT — potential, P R E S — present, PTCPL — participle, REFL - reflexive, R E C P A S T - recent past, SG - singular, S ( U B J ) - subject, T O P topic, TR — transitive. 2. For the term "highly individuated" see Timberlake (1975). 3. It will be clear that this definition does not imply that no other parameters are involved in a Standard Transitive State of Affairs. I have only mentioned those that are relevant to the discussion, without the pretense that these parameters exhaust the possibilities connected with or related to voice. An obvious example is the parameter of aspect. 4. This is clearly not so for all languages, notably ergati ve ones. Cooreman (1988), however, shows that even in a notoriously ergative language like Dyirbal referents with the semantic role of agent have the property of topic more often than those with the role of patient (Cooreman 1988: 742). Thus, although the encoding mechanisms in Dyirbal may suggest otherwise, text counts show that the Standard Transitive State of Affairs is indeed standard, even in Dyirbal. 5. For examples from exotic languages the source is mentioned in the text. Numbers in brackets after examples refer to page numbers where sample sentences were found in the sources. 6. Foley—Van Valin (1984) give more examples of both backgrounding passives with an additional foregrounding function and foregrounding passives with an additional backgrounding function. 7. This is my interpretation, not Klaiman's. Klaiman does not consider one of the voice options she discusses as "basic voice" to be unmarked in contrast to others. In terms of the Standard Transitive State of Affairs as described above, however, I consider the combinations agent-controller and patient-affected entity to be prototypical, without the implication that the encoding for this combination will be unmarked for all languages. 8. "Affectedness" should be interpreted as a relative notion. It is likely that the patient too is affected in, e. g., (22 a), but the fact that the agent is affected by its own action is conceived as more inportant in this State of Affairs. 9. In this context: the referent having the properties of "agent" and "control". 10. Tamil also has a passive construction, which is "derived" in Klaiman's analysis. 11. The notion "control" is best considered in relative terms. In a Causative State of Affairs ( e . g . , one that can be expressed as "I made him sign the letter"), the agent will in most cases retain some degree of control over his action (short of being physically forced to perform it). What is important here is that there is another referent in the situation that is perceived as having the greater control over or being the cause of what is happening.
74
lneke
Brus
12. Even though arguments exist to show that the term "subject" does not apply to all languages I assume that it is sufficiently clear in the present context. 13. See, for example, Jakobson 1971: 52: "der Dativ gebe nur den Adressaten an, er besage die bloße Gerichtetheit der Handlung ohne Berührung des Gegenstandes." 14. When an encoding is said to express a specific combination of States of Affairs it is not implied that this is an exhaustive listing. In some cases an encoding can express more States of Affairs than is mentioned in the examples. In others a language might have encodings for States of Affairs not under discussion at that point and so they are not mentioned. My objective in this paper is merely to show that certain combinations do exist — not to list them exhaustively. 15. The same claim has been made by Haspelmath (1989) and Geniusienè (1987: 328). 16. Passives with expressed agent could also be considered as having an agent that is new information, and for that reason an unlikely candidate for topic position in the sentence. Also, such an agent could be said to be non-referential up to the time it is first mentioned. The relatively infrequent agentive passives in Dutch and English could be interpreted in this way. This would mean that the (primary) function of the passive encoding still is "impersonal agent", in a modified version.
References Adelaar, Willem F. H. 1977 Ashiwayu, M. 1968 Ashton, E. O. 1966 Awbery, G. W. 1976
Tarma Quechua. Grammar, texts, dictionary. Lisse, The Netherlands: De Ridder Press. Lehrbuch der Yoruba-Sprache. Swahili grammar
Leipzig: VEB Verlag.
(including intonation).
The syntax of Welsh: A transformational University Press.
London: Longmans. study of the passive. Cambridge: Cambridge
Barber, E. T. W. 1975 "Voice — beyond the passive." Berkeley Linguistic Society, Proceedings 1: 16 — 24. Cooreman, Ann 1988 "Ergativity in Dyirbal discourse." Linguistics 26: 717—746. Derbyshire, Desmond C. 1979 Hixkaryana. Amsterdam: North Holland Publishing Company. Eades, Diana 1979 "Gumbainggir," in: R. M. W. D i x o n - B . G. Blake (eds.), Handbook of Australian languages, Vol 1. Amsterdam: John Benjamins. Epee, Roger 1976 Generative syntactic studies in Duala. Ph. D. Dissertation, Cornell University. Ann Arbor, Michigan: University Microfilms International. Everett, Daniel L. 1986 "Piraha." in: Desmond C. Derbyshire—Geoffrey K. Pullum (eds.), Handbook of Amazonian languages, Volume 1. Berlin/New York: Mouton de Gruyter.
Towards a typology of voice
75
Foley, William A . - R o b e r t D. Van Valin 1984 Functional syntax and universal grammar. Cambridge: Cambridge University Press. Fortescue, Michael 1984 West Greenlandic. London: Croom Helm. Frank, Paul Stephen 1985 A grammar of Ika. Ph. D. Dissertation, University of Pennsylvania. Ann Arbor, Michigan: University Microfilms International. Geniusienè, Emma 1987 The typology of reflexives. Berlin/New York: Mouton de Gruyter. Givón, Talmy 1979 On understanding grammar. New York: Academic Press. Haspelmath, Martin 1989 "The grammaticalization of passive morphology." To appear in Studies in Hopper, Paul J. —Sandra A. Thompson 1980 "Transitivity in grammar and discourse." Language 56(1): 251—299. Jacobs, Roderick 1976 "A passive continuum in Austronesian." Papers from the parasession on diachronic syntax, Chicago: Chicago Linguistic Society, 118 — 125. Jakobson, Roman 1971 Selected writings II — Word and language. The Hague. Mouton. Jones, J. Morris 1913 A Welsh grammar, Historical and comparative. Oxford: Oxford University Press. Kimball, Geoffrey David 1985 A Descriptive grammar of Koasati. Ph. D. Dissertation, Tulane University. Ann Arbor, Michigan: University Microfilms International. Klaiman, M. H. 1988 "Affectedness and control: A typology of voice systems," in: M. Shibatani (ed.), Passive and Voice, Amsterdam: John Benjamins. Lehtinen, M. 1963 Basic course in Finnish. Uralic and Altaic Series Vol. 27. The Hague: Mouton & Co. Lewis, G. L. 1967 Turkish grammar. Oxford: Clarendon Press. Li, Charles N. — Sandra A. Thompson 1981 Mandarin Chinese: A functional reference grammar. Berkeley: University of California Press. Macintosh, Mary 1984 Fulfulde syntax and morphology. London: K. Paul International. Risselada, R. 1987 "Voice in Ancient Greek: Reflexives and passives," in: Johan van der Auwera—Louis Goossens, Ins and Outs of the Predication, Dordrecht: Foris. Saad, G.N. 1975 Transitivity, causation and the derivation of passives in Arabic. Ph. D. Dissertation, University of Texas at Austin. Ann Arbor, Michigan: University Microfilms International.
76
Itieke
Brus
Schachter, Paul and Fe T. Otanes 1972 Tagalog reference grammar. Berkeley: University of California Press. Schayk, Gerard van 1985 "Valentiereduktie in het Turks." Interdisciplinair Tijdschrift voor Taal- en Tekstaetenscbap 5: 127 — 139. Shibatani, Masayoshi 1985 "Passives and related constructions: A prototype analysis." language 61(4): 821-848. Siewierska, Anna 1984 The passive. A comparative linguistic analysis. London: Croom Helm. Strehlow, T. G. H. 1944 Aranda phonetics and grammar. Sydney: The Australian National Research Council. Timberlake, Alan 1975 "Hierarchies in the genitive of negation." Slavic and European Journal, 19. Tucker, A. N. and J . T. Mpaayei 1955 A Maasai grammar with vocabulary. London: Longmans, Green and Co.
Non-verbal predicability 1 Kees Hengeveld
1. Introduction The relations between locative, existential, possessive and other constructions based on non-verbal predicates have received considerable attention during the last two decades (e.g., Lyons 1968; Clark 1978; Bickerton 1981; Wilson 1983). The major concern of these studies was to determine to what extent the overlap in lexicalisation of these constructions is systematic, and to find explanations for these overlaps. It was more or less taken for granted that constructions expressing locative, existential, etc. meaning can be viewed as locative, existential, etc., constructions. A few examples will show that this is not always the case. In Yagaria, adjectives cannot be applied predicatively. In order to predicate properties of an object, speakers of Yagaria have to construe a noun phrase within which the adjective is applied attributively to a head noun. This noun phrase is then applied predicatively in a classifying or equative construction. For instance, the adjective okavu 'blue' in (1) is turned into a noun phrase by applying it to the dummy noun na 'one, thing'. This noun phrase serves as the predicate of the construction. Only in this way can the property 'blue' be predicated of an object in Yagaria. Yagaria (Indo-Pacific, Renck 1975) (1)
ma-gaveda okavu'-na DEM-string blue-NOM 'This string is (a) blue (one).'
This is just one of the many examples in which the lack of a certain construction type, in this case a property assigning construction, is remedied by an alternative construction, in this case an equative construction. Another example involves the expression of possessive and existential meaning in Fijian: Fijian (Austronesian, Milner 1956) (2) a.
e dua na nona waqa PRED one ART his canoe 'He has a canoe.' ("His canoe is one.")
78
Kees Hengeveld
b.
(3)
e rua na nona waqa PRED two ART his canoe 'He has two canoes.' ("His canoes are two.") e so na wai PRED little ART water 'There is a little water.' ("The water is a little.")
Possession or existence of an indefinite object is expressed in Fijian by applying a quantifier as a predicate to a term which may contain a restrictor indicating the possessor. These examples show that Fijian has constructions expressing possessive and existential meaning, but that it does not have possessive and existential constructions in the narrower sense of the term. The aim of this paper is to investigate the extent to which languages make use of non-verbal predication types in this narrower sense. In order to do so I first define the non-verbal predication types that will be investigated in terms of the Functional Grammar approach to non-verbal predication, as proposed in Dik (1980). 2 Secondly, I try to determine how these non-verbal predication types can be recognized. Thirdly, I show that the extent to which languages make use of non-verbal predication can be described in terms of a number of hierarchies, and finally I try to find some explanations for the existence of these hierarchies.
2. Non-verbal predicates in Functional Grammar In Functional Grammar underlying predications are built on the basis of predicate frames. Verbs, adjectives, and nouns are all represented in the lexicon according to the same format, of which some examples are given in (4)-(6). (4)
g i v e y (Xi)Ag (X2)go (X3)rcc
(5) (6)
intelligent (x,)„ carpenter N (x,) 0
The verbal predicate give in (4) has three argument positions, provided with the semantic functions Agent, Goal and Recipient. The adjectival predicate intelligent in (5) has one argument position, provided with the semantic function Zero, and so has the nominal predicate carpenter in (6). In order to form a predication, terms are inserted in the argument positions of these
Non-verbal predicability
79
predicate frames. This leads to an underlying structure like (7) in the case of adjectival predicates: (7)
intelligent A (dlx¡: carpenter N (x,)0)0 'The carpenter (is) intelligent.'
In (7) the adjectival predicate intelligent is applied to a term referring to a definite (d) singular (1) entity Xi which has the property carpenter. Note that the inserted term is itself construed on the basis of a predicate frame, and that the underlying predication does not contain a copula. Copulas are considered to be semantically empty supportive devices, which are introduced by a copula-support rule. This allows for a generalization including both those languages which do and those which do not make use of a copula. In what follows I intend the term "non-verbal predication types" to include copula constructions. Three different groups of non-verbal predicates can be distinguished. The first consists of bare non-verbal predicates: adjectives and nouns. I have given an example of a predicatively used adjectival predicate in (7). In some languages it is also possible to apply nominal predicates predicatively in their bare form. Consider the Basque examples (8) —(9). Basque (Isolate, Lafltte 1944) (8)
(9)
Soldado %en soldier COP.PAST.IMPF.3sg 'He was (a) soldier.' gtKpn-a hiζ man-sg COP.PRES.2sg 'You are a man.'
In (8) the nominal predicate soldado 'soldier' is used predicatively in its bare form. This is shown by the absence of the article -a, which basically signals singularity. The presence of this article in (9) signals that in this sentence the predicate is a noun phrase (a term in Functional Grammar) used predicatively. The two sentences can be represented as in (10) —(11). (10)
soldado N (x¡: he (x¡)0)0 'He is (a) soldier.'
(11)
{( ix ¡ : g i z o n N (Xj)„)} (dx k : you (x k ) 0 ) 0 'You are a man.'
Example (10) is a representation of the predicative use of the bare nominal predicate soldado 'soldier'. The representation in (11) shows the predicative use of a term. Whereas bare predicates are used in the construction of property
80
Kees
Hengeveld
assigning expressions, term predicates are used in the construction of classifying or equative constructions. The predicate in (11) is itself represented as an indefinite (i) term (x¡), the predicative function of which is indicated by means of the curly brackets. In this case, too, there is no copula in the underlying representation. The third group of non-verbal predicates contains adpositional and casemarked predicates, as illustrated by the capitalized constituents in the Quechua examples (12) —(13): Imbabura Quechua (Andean, Cole 1982) (12) (13)
chay wasi ÑUKA-PAJ-mi that house 1-POSS-FOC 'That house (is) mine.' (lit. 'of me') Ju^i-ka UTAVALU-PI-mi José-TOP Otavalo-LOC-FOC 'José (is) in Otavalo.'
In Functional Grammar adpositions and case markers the expression of semantic functions carried by terms, ca ti vely used terms. This is accounted for by allowing be assigned to terms used within a term predicate, as of (13) given in (14): (14)
are considered to be in this case by predisemantic functions to in the representation
{(dlx¡: Utavalu N ( Xi )) Loc } (dix- Juzi N ( Xj )) 0
Here the definite (d) singular (1) term (x,) specified as Utavalu carries the semantic function Locative). This construction as a whole is applied as a predicate to a term Xj specified as Ju^i. Again the uderlying representation does not contain a copula. Existential constructions are considered to be a special subtype of locative constructions, in the sense that existence is viewed as being located at an unspecified location. The different types of non-verbal predicate can be used in the construction of a number of non-verbal predication types, each expressing a different semantic relation, as listed in (15). (15)
Non-verbal predication types (cf. Dik 1980) Predication type based on: Bare predicate PredA (x¡)0 PredN (x¡)0
Semantic relation expressed by predication type Property assignment Status assignment
Non-verbal predicabilitj
Term predicate {(dx,)} (χ.). {(ix,)} (x,)o Adpositionalj Case- marked {(xOloc) (X,)0 {(0)loc) (x>)b {(Xi)poss) (Xi)0
81
Identity Class membership predicate Location Existence Possession
The application of bare predicates to terms leads to constructions expressing Property Assignment in the case of adjectival (A) predicates, and Status Assignment in the case of nominal (N) predicates. The term status assignment is chosen since the nominal predicates that can be used predicatively in their bare form in some languages generally designate social categories such as professions, membership of political or religious groups, and kinship relations. The application of term predicates to terms leads to constructions expressing Identity or Class Membership, depending on whether the term from which the predicate is derived is definite (d) or indefinite (i). The application of adpositional or case-marked predicates to terms leads to constructions expressing a variety of semantic relations, depending on the semantic function assigned to the predicatively used term. The examples given here include Location (Loc), Existence, and Possession (Poss), since these seem to be the most central in most language systems, but other examples would be possible, such as, for instance, a predicate derived from a term with Agent function, as in (16). (16)
This book is by Shakespeare
A further subdivision can be made, following traditional terminology, by contrasting term predicates on the one hand with bare predicates and adpositional or case-marked predicates on the other. The usual terminology when using such a subdivision is "equative" vs. "ascriptive". In what follows I will concentrate on ascriptive constructions. Equative constructions seem to be universal or nearly universal, and are therefore less interesting in the present context, although they will show up later in this paper as an often used substitute for ascriptive construction types lacking in particular languages.
82
Kees
Hengeveld
3. Distinctive features of ascriptive non-verbal predication types Given the central question of this paper, "to what extent do languages make use of non-verbal predication types?", it is important to have some criteria to decide whether a certain construction qualifies as an instance of the construction type looked for. Ascriptive non-verbal predication types are constructions based on an ascriptive non-verbal predicate, of which the main characteristics are listed in (17). (17)
A predicate can be considered an ascriptive non-verbal predicate if that same predicate: i. can be used in the construction of a term phrase, either as a head or as an attribute, without further measures being taken, 3 and ii. cannot be used as an independently referring unit without further measures being taken.
In other words, if a predicate can be applied directly within the term, and if that predicate is not itself a term, it is an ascriptive non-verbal predicate. In order to illustrate the first restriction consider the Latin examples (18) — (19). Latin (Italic, Bolkestein 1983) (18) a.
b.
(19) a.
b.
¡¡ber puer-i est book.NOM boy-GEN COP.PRES.3sg 'The book is the boy's.' liber puer-i book.NOM boy-GEN 'The boy's book.' Uber puer-o est book.NOM boy-DAT COP.PRES.3sg 'The boy has a book/There is a book to the boy.' *liber puer-o book.NOM boy-DAT 'The book to the boy.'
The examples in (18) show that, next to the predicative use of the genitive constituent, there is an attributive use of this constituent, whereas the examples in (19) show that the dative constituent can be used at the level of the predication, but not as an attribute within a term. The dative constituent
Non-verbal predicability
83
therefore does not comply with the first restriction on ascriptive non-verbal predicates. In order to illustrate the second restriction consider the English examples (20)-(22) (for (20) cf. de Groot 1983). (20) (21) (22)
John's book my book *mine book
The book is John's *The book is my The book is mine
I prejer John's *I prefer my I prefer mine
The examples in (20) show that the genitive constituent John's can be used attrivutively and predicatively, but also as an independently referring unit provided that the possessed item is understood from the context. In its predicative use the context by definition makes clear what the possessed item is, and there is no reason to assume that in this case the genitive constituent is not an independently referring constituent. This assumption is corroborated by (21) and (22), which show that the attributive my cannot be used as an independently referring unit or be applied predicatively, and the independently referring mine can be used predicatively but not attributively. These facts lead to the conclusion that possessive predicates in English do not comply with the second restriction imposed on ascriptive non-verbal predicates, and that possession of a definite object is expressed in English through an equative construction rather than through an ascriptive construction.
4. Non-verbal predicability With these two criteria in mind, it is possible to determine for any language to what extent it allows the predicative use of ascriptive non-verbal predicates. In other words, it is possible to determine whether an ascriptive non-verbal predication type is predicable in that language, where by (non-verbal) predicability 4 I understand the following: (23)
Predicability:
The possibility of grammatically acceptable application
of a predicate β to an argument OC.
Non verbal predicability: The possibility of grammatically acceptable application of a non-verbal predicate β to an argument a. I have checked 35 languages in order to determine their degree of non-verbal predicability. These languages were selected through a procedure to be fully described in Rijkhoff et al. (forthcoming). The main feature of the procedure is that it aims at maximal diversity within the sample, where diversity is
84
Kees
Hengeveld
calculated on the basis of an objective measure. For reasons of presentation and space I give subsets of my data rather than full tables. In each case I have selected some languages for which my data are complete. A full description of the sample and an exhaustive account of the data will be given in Hengeveld (forthcoming). The data collected can be arranged in such a way that the result is a twodimensional 5 implicational hierarchy. In order to arrive at this result a distinction has to be made between constructions with definite and with indefinite arguments. Table 1 shows a selected subset of the data for the constructions with a definite argument. In this table a " + " indicates that a language makes use of the construction type, a " — " that it does not. Table 1. The predicability of ascriptive non-verbal predication types with definite arguments — some examples Language
LOC(dx)
A(dx)
Tamil Abkhaz English Basque Vietnamese Spanish
+ + + + + +
+ + + + +
N(dx)
POSS(dx)
-
—
+ /+ + +
-
+ /+
Table 1 shows that if a language can use possessive predicates predicatively, it can also use nominal, adjectival and locative predicates predicatively; if it can use nominal predicates predicatively, it can also use adjectival and locative predicates predicatively, etc. This goes for all the languages in my sample for which my data are complete and is not contradicted by the languages for which my data are incomplete. For those languages which do not have a separate class of adjectives the adjective column was considered irrelevant. The data in Table 1 reflect a hierarchy which can be represented as in (24). This hierarchy should be read in the following way: if a construction at a certain point in the hierarchy is predicable in a certain language, then all constructions preceding it in the hierarchy are also predicable in that language. (24)
Predicate hierarchy — definite arguments LOC(dx) > A(dx) > N(dx) > POSS(dx)
Table 1 not only shows that languages can be arranged according to their degree of predicability in the domain of construction types with a definite argument, but also contains some illustrations of another feature of hierar-
Non-verbal
predicability
85
chies: variations around the cut-off point. For a simple illustration of this kind of variation consider the English examples (25) —(26). (25) (26)
John is chairman *John is carpenter
In English a nominal predicate can be used predicatively only if the function it designates is unique, as in (25), but not if this function is non-unique, as in (26). In the case of constructions with an indefinite argument, the results have to be evaluated in a different way. Of the four types of predicate under consideration only possessive and locative predicates are regularly used with indefinite arguments. This is not surprising, since, as Rijkhoff (1988) observes, in order to make an entity identifable for an addressee, either its existence has to be asserted, or it has to be related to another entity which is known to the addressee, such as its location or possessor. The arguments of possessive and locative constructions can therefore be both definite and indefinite, whereas the arguments of adjectival and nominal predicates are almost always definite, or at least specific. Probably it is because of this discrepancy that the studies of relations between non-verbal predication types referred to earlier have concentrated on possessive and locative constructions, and have not paid much attention to adjectival and nominal constructions. Within the domain of constructions with an indefinite argument the predicate hierarchy can thus only be checked for possessive and locative constructions, where I intend the group of locative constructions to include the existential ones. Table 2 shows the results for a selected subset of the data. Table 2. The predicability of ascriptive non-verbal predication types with indefinite arguments — some examples Language
LOC(ix)
POSS(ix)
West Greenlandic Ngiyambaa Krongo Koryak
+/-
-
+ +
-
+
What this table shows is that those few languages which have an indefinite possessive construction also have the indefinite locative construction. Again we can write this down as a hierarchy, as in (27).
86
Kees
(27)
Hengeveld
Predicate hierarchy — indefinite arguments LOC(ix) > POSS(ix)
The terminal points of the two versions of the predicate hierarchy contain the same predicates. The difference is that the second hierarchy does not contain the adjectival and nominal predicates in between the two terminal points. The fact that two versions of the hierarchy have to be postulated on the basis of the (in)definiteness of the argument term suggests that definiteness should be treated as an independent parameter. The definiteness hierarchy is relevant in other domains too, and is given in (28) in its simplest form. 6 (28)
Definiteness hierarchy definite > indefinite
This hierarchy is relevant in both construction types that regularly occur with definite and indefinite arguments, locative constructions and possessive constructions. This is illustrated in tables 3 and 4. Table 3. The predicability of possessive construction types with definite and indefinite arguments — some examples Language
POSS(dx)
!Xü Vietnamese Ket Tagálog Burushaski
+ /+ + +
POSS(ix)
— —
+ /+
Table 4. The predicability of locative construction types with definite and indefinite arguments — some examples Language
LOC(dx)
LOC(ix)
Ngalakan Navaho Nasioi
+ + +
_ +1-
+
It appears from the data in Table 3 that those few languages which use indefinite possessive constructions also use definite possessive constructions. The data in Table 4 show that those languages which use indefinite locative constructions also use definite locative constructions.
Non-verbal
predicability
87
Having separated the definiteness parameter from the predicate parameter, the two can be combined in the two-dimensional hierarchy given in (29). (29)
Predicate and definiteness hierarchy Hardly predicable
INDEF.
LOC(ix)
DEF.
LOC(dx)
LOC(x)
POSS(ix)
A(dx)
>
A(x)
. N(dx)
>
N(x)
POSS(dx)
>
POSS(x)
Easily predicable
The predicate hierarchy is projected horizontally, the definiteness hierarchy vertically. In this way the terminal points of the two versions of the predicate hierarchy are connected, and the irrelevant combinations are simply not realized. The bottom left box in (29) represents an easily predicable construction type, the upper right box a hardly predicable construction type. Thus this picture shows that in a typological perspective the definite locative construction and the indefinite possessive construction are diametrically opposed as to their degree of predicability. So far I have approached the different hierarchies from the point of view of predicable non-verbal predication types. The same hierarchies appear to be relevant if one looks at the alternatives that particular languages use for non-predicable construction types. Let me first present the alternatives for non-predicable constructions with definite first arguments. The alternatives used for non-verbal predication types in this domain are very homogeneous. In nearly all cases the alternative is an equative construction. Consider, for example, the following Tamil alternatives in (30) —(32). Tamil (Dravidian, Asher 1982) (30)
Raaman
nalla-van
Raman good-NMLZR 'Raman is good.' ("Raman is a good one.")
88
Kees Hengeveld
(31)
Avaru (oru) daktar he one doctor 'He is doctor.' ("He is one doctor.") Inta pustakam raaman-atu this book Raman.POSS-NMLZR 'This book is Raman's.' ("This book is the one of Raman.")
(32)
In all three cases the predicate is a term phrase. Adjectival and possessive predicates are nominalized in order to allow them to occur in predicative position; nouns need not be nominalized, but some speakers prefer to provide them with the elsewhere entirely optional oru 'one', as if stressing the termnature of the predicate. 7 In all these cases the predicative application of one of the three predicate types is avoided by reducing it to the status of head or attribute within a predicatively used term. Table 5 gives some examples which show that the predicate hierarchy is relevant in the use of alternative construction types in the sense that if the equative construction is used as an alternative at one point in the hierarchy, it is also used at subsequent points in the hierarchy. Table 5.
The use of the equative construction as an alternative for non-predicable construction types with definite arguments
Language
LOC(dx)
A(dx)
N(dx)
POSS(dx)
Basque
+ + +
+ +
+ EQ
EQ
EQ
EQ
EQ
Abkhaz Chinese, Mandarin
EQ
+ = predicable construction type, E Q = equative construction
The alternatives used for non-verbal predication types with an indefinite argument are very heterogeneous. A well known alternative is the "have"construction, not only used to express possessive meaning but also used to express existential and locative meaning. Another alternative is the predicative use of a quantifier, as in the Fijian examples (2) — (3) given earlier. Yet another frequently encountered alternative is the proprietive/privative construction, as in examples (33 a —b) from Ngalakan. Ngalakan (Australian, Merlan 1983) (33) a.
gu-we?-ji-meniñ CL-water-PRIV-COP 'There was no water.' ("It was waterless.")
Non-verbal predicability
b.
89
y u-may-ci 1 sg-vegetable.food-PRIV Ί have no food.' ("I am foodless.")
A final alternative for possessive constructions is the existential construction, in several variants, paraphrased in (34): (34) a. b. c. d.
There is my book. There is a book to me. A book there.is. to.me. Talking about me, there is a book.
(example: (example: (example: (example:
Turkish) Latin) Nasioi) Mandarin)
In (34 a) the possessive predicate has the status of an attribute of the argument term. In (34 b —c) the possessor is expressed as an entity experiencing the existence of the possessed item, in (34 b) through the expression of this second argument in the dative case, in (34 c) through the expression of this second argument in a benefactive stem of the existential verb. In (34 d) the possessor is expressed as a theme, the constituent with respect to which the existence of the possessed item is relevant. For each of these alternative constructions the observation holds that if it is used as an alternative at one point in the predicate hierarchy, it is also used at subsequent points in the hierarchy. Table 6 a shows that those languages which use the proprietive/privative construction to express locative or existential meaning also use this construction to express possessive meaning. It furthermore shows that languages may use more than one alternative, as in the case of Ngalakan, but that even in that case the alternatives so to speak add up from left to right, the widest variety being shown in the case of possessive constructions. Table 6 b shows similar facts for some languages which use the lexical or "have"-construction as an alternative. Table 6 a. The use of the proprietive construction as an alternative for non-predicable construction types with indefinite arguments Language
LOC(ix)
POSS(ix)
Ket Ngiyambaa Ngalakan
+
PROPR PROPR PROPR, QUANT, HAVE
+/PROPR PROPR, Q U A N T
+ = predicable construction type; PROPR = proprietive/privative construction; Q U A N T = predicative quantifier; H A V E = "have"-construction.
90
Kees
Hengeveld
Table 6 b. The use of a "have"-construction as an alternative for non-predicable construction types with indefinite arguments Language
LOC(ix)
POSS(ix)
Babungo Guarani Spanish
+ +
HAVE, EXIST HAVE HAVE
HAVE
+ = predicable construction type, HAVE = "have"-construction, EXIST = existential construction.
What Tables 5, 6 a and 6 b show is that an alternative used at a certain point in the predicate hierarchy is also used as an alternative for the constructions to its right. Summarizing the observations, one can say that: (35)
The predicate hierarchy gives a systematic account of i. the extent to which languages make use of non-verbal predication types; ii. the extent to which languages make use of regular alternatives for non-predicable non-verbal predication types.
It seems, then, that the predicate hierarchy, with its two realizations in the definite and indefinite domain, is highly relevant in systems of non-verbal predication. Even more so, since so far I have found no clear counterexamples. 8
5. Explanations Although the relevance of the predicate hierarchy seems to be firmly established, so far I have not gone into the factors motivating its existence. Although I certainly have no final answers to this problem, I do have some suggestions. The first concerns the range of arguments that the predicates under consideration can be applied to. Consider the adapted and simplified version of Sommers' predicability tree in (36). (36)
The predicate hierarchy and the predicability tree (Cf. Sommers 1967; Keil 1979; Bickerton 1981)
Non-verbal predicabilitj
91
LOC lst/2nd order
2nd order
POSS non-human The position of a predicate in this tree shows to what kinds of arguments it can be applied. For the interpretation of this tree, the distinction between first and second order entities (cf. Lyons 1977: 442—443) is relevant. First order entities are those that can be located in space, such as persons, animals, cars and houses. Second order entities are those that can be located in space and time, such as meetings, weddings, and destructions. In terms of this distinction, locative predicates have the widest range of applicability. The same locative predicate can be applied to both first and second order arguments. Many adjectival predicates can be applied to all kinds of first order arguments. Nominal predicates and possessive predicates, on the other hand, have a much more limited range of applicability. Nominal predicates can be applied to human first order arguments only, and possessive predicates are generally applied to non-human first order arguments only. A second factor motivating the existence of the predicate hierarchy can be discovered if one looks at the properties expressed by non-verbal predicates. These properties exhibit different shades of abstractness, as indicated in (37). (37)
The predicate hierarchy and degrees of abstractness LOC(x)
> A(x)
> N(x)
> POSS(x)
Spatial properties Concrete Perceptible
> Physical properties > >
> Social properties > >
> Conventional properties > Abstract > Imperceptible
Roughly speaking, locative predicates designate spatial properties, adjectival predicates physical properties, nominal predicates social properties, and pos-
92
Kees
Hengeveld
sessive predicates p r o p e r t i e s that are d e p e n d e n t o n legal o r social c o n v e n t i o n s . In this sense t h e predicate h i e r a r c h y reflects d i f f e r e n t shades o f abstractness, a n d as such is an instantiation o f a m o r e f u n d a m e n t a l parameter. T h e d e g r e e o f abstractness is also r e f l e c t e d in the n u m b e r o f d i f f e r e n t conceptualizations that o n e finds as a l t e r n a t i v e s f o r the n o n - v e r b a l p r e d i c a t i o n types. N o t e that this same p a r a m e t e r can, t o s o m e extent, also be d e f i n e d in t e r m s o f the perceptibility
o f t h e p r o p e r t i e s r a t h e r than in t e r m s o f their d e g r e e
of
abstractness. T h u s t w o m o t i v a t i n g f a c t o r s p r e s e n t t h e m s e l v e s : o n e f r o m the p o i n t o f v i e w o f the r a n g e o f a r g u m e n t s that the a s c r i p t i v e n o n - v e r b a l predicate can be applied to, the o t h e r f r o m the p o i n t o f v i e w o f the p r o p e r t y designated b y t h e a s c r i p t i v e n o n - v e r b a l predicate. E a c h o f these e x p l a n a t i o n s p o i n t s at an e v e n m o r e basic h i e r a r c h y at a h i g h e r l e v e l o f abstraction. It m a y be that it is the i n t e r a c t i o n b e t w e e n the t w o o f t h e m that has such
important
consequences.
Notes 1. I am indebted to Simon Dik and Hotze Mulder for their valuable comments. Abbreviations used throughout are "A" ("adjective"); "ART" ("article"); "CL" ("classifier"); "COP" ("copula"); "DAT" ("dative"); "DEM" ("demonstrative"); "FOC" ("focus"); "GEN" ("genitive"); "LOC" ("locative"); "IMPF" ("imperfective"); "N" ("noun"); "NMLZR" ("nominalizer"); "NOM" ("nominative"); "POSS" ("possessive"); "PRED" ("predicate"); "PRES" ("present"); "PRIV" ("privative"); "sg" ("singular"); and "TOP" ("topic"). 2. See-also Hengeveld (1986, 1987, 1990, forthcoming). 3. This restriction, among other things, excludes adjectival verbs from being classified as nonverbal predicates, since these have to be relativized when used in the construction of terms. See, e.g., Lehmann (1988, ch.2), Wetzer (1991). 4. Predicability can also be understood in an ontological sense. See Sommers (1967) and the discussion in Hengeveld (1990). 5. There is a third dimension that is relevant for determining the degree of non-verbal predicability of a language. Constructions with a first order argument are more easily predicable than those with a second order argument, and these in turn are more easily predicable than those with a third order argument (see Hengeveld 1990). 6. See Comrie (1981: 128) on refinements of this hierarchy. 7. Asher (1982: 108) notes, with respect to the expression of nominal complements: In these there is no distinction between defining, identity and role types, in the sense that the complement will in each case be a nominal form in the nominative case. For some, though by no means all, speakers there is, however, a difference between the defining and the role type, in that a noun in the latter can optionally be preceded by oru 'one' used as a marker of indefiniteness, while a noun in the former cannot. 8. The hierarchy might also be relevant from a psycholinguistic and from a diachronic perspective (see Hengeveld forthcoming).
Non-verbal predicability
93
References Asher, R. E. 1982 Tamil. Amsterdam: North-Holland. Bickerton, Derek 1981 Roots of language. Ann Arbor: Karoma. Bolkestein, A. Machtelt 1983 "Genitive and dative possessors in Latin", in: Simon C. Dik (ed.), Advances in Functional Grammar, Dordrecht: Foris, 55 — 91. Clark, Eve. V. 1978 "Locationals: existential, locative, and possessive constructions", in: Joseph H. Greenberg (ed.), Universals of human language, Vol. 4, Stanford: Stanford University Press, 85 — 126. Cole, Peter 1982 Imbabura Quechua. Amsterdam: North-Holland. Comrie, Bernard 1981 Language universals and linguistic typology. Oxford: Blackwell. Dik, Simon C. 1980 Studies in Functional Grammar. London: Academic Press. Groot, Casper de 1983 "On non-verbal predicates in Functional Grammar: the case of possessives in Hungarian", in: Simon C. Dik (ed.), Advances in Functional Grammar, Dordrecht: Foris, 9 3 - 1 2 2 . Hengeveld, Kees 1986 "Copular verbs in a functional grammar of Spanish", Linguistics 24: 393 — 420. 1987 "A functional analysis of copula constructions in Mandarin Chinese", Working Papers in Functional Grammar 23. 1990 "Semantic relations in non-verbal predication", in: A. Machtelt Bolkestein — Jan Nuyts —Co Vet (eds.), Layers and levels of representation in language theory: a functional view. Amsterdam: Benjamins, 101 — 122. forthcoming "Non-verbal predication: theory, typology, diachrony". Keil, FrankC. 1979 Semantic and conceptual development; an ontological perspective. Cambridge, Mass.: Harvard University Press. Lafitte, Pierre 1944 Grammaire Basque: Navarro-labour din litter aire. Bayonne. Lehmann, Christian 1988 "Studies in general comparative linguistics". Arbeiten des Kölner UniversalienProjekts 71. Lyons, John 1968 "Existence, location, possession and transitivity", in: B. Van Rootselaar—J. F. Staal (eds.), Logic, methodology and philosophy of science III, Amsterdam: NorthHolland, 4 9 5 - 5 0 4 . 1977 Semantics. 2 Vols. Cambridge: Cambridge University Press. Merlan, Francisca 1983 Νgalakan grammar, texts, and vocabulary. Canberra: Australian National University.
94
Kees
Hengeveld
Milner, G. B. 1956 Fijian Grammar. Suva: Government Press. Renck, G. L. 1975 A grammar of Yagaria. Canberra: The Australian National University. Rijkhoff, Jan 1988 "A typology of operators", Working Papers in Functional Grammar 29. Rijkhoff, Jan — Dik Bakker—Kees Hengeveld — Peter Kahrel forthcoming "A method of language sampling". Sommers, Fred 1967 "Types and ontology", in: Peter F. Strawson (ed.), Philosophical Logic, Oxford: Oxford University Press, 138 — 169. Vet, Co 1983 "Possessive constructions in French", in: Simon C. Dik (ed.), Advances in Functional Grammar, Dordrecht: Foris, 123—140. Wetzer, Harrie 1991 "A typology of predicative adjectival constructions", in this volume. Wilson, Bob 1983 "An examination of crosslinguistic constraints on the lexicalization of predications of ownership, possession, location and existence", University of Hawaii Working Papers in Linguistics 15(2): 1 — 18.
Impersonal constructions as a strategy for second-order predication Zbigniew Kañski
1. Introduction This paper is concerned with a special class of sentences commonly and conveniently labelled "impersonal"; more precisely, this term will refer here to a subset thereof with Polish sii, Italian si, and English one,1 whose properties will be specified in section 2. Some apparently precise but inadequate solutions in terms of Transformational Generative Grammar will be pushed to unacceptable conclusions. A competing approach will be suggested in terms of a categorial grammar with predicate abstraction; it will be argued that impersonal sentences are a special class, for in spite of their apparent first-order syntactic structure, their semantics will be shown to be best analyzed in terms of second-order predication. This will be argued to follow from the nonreferentiality of the impersonal subject, and an independently observable constraint whereby impersonal sentences, to be interpretable as propositions, must contain a complement or modifier, even if the latter is not required by verbal subcategorization in other types of structure.
2. The corpus I will be concerned here only with those sentences in Polish, English, and Italian that meet the following structural and distributional criteria: (i) the occurrence of free morphemes (lexical items) sig, si, and one in Polish, Italian, and English sentences, respectively; (ii) the substitutability of nominative NPs for these lexical items without inducing ungrammaticality, pace some adjustments in concord and word order required in particular languages; 2 (iii) the exclusion of cooccurrence of si^si/one with (other) nominative NPs, including coordination with the former; (iv) the lack of influence of lexical transitivity of the finite verb on the occurrence of si$\si\one.
96
Zbigniew Kanski
The following Polish sentences and their English equivalents meet all the above conditions: 3 (1)
(2)
(3)
(4)
(5)
(6)
Ν a r j b j ώοάχί siç tvc^estiie rano. for fish 3sg.neut.-go early morning O n e goes fishing early in the morning.' Do tego domu wchod^i siç od podwór^a. to this house 3sg.neut.-enter from backyard 'One enters this house from the backyard.' W Wigilk sk posci. On Christmas Eve 3sg.neut-fast 'One fasts on Christmas Eve.' Pracuje siç dia pr^yjemnosci. 3sg.neut.-work for pleasure 'One works for pleasure.' Tu sii nie plywa. here not 3sg.neut.-swim 'One doesn't swim here.' Jak siç nie ma co sii h*bi, If not 3sg.neut.-have what 3sg.neut.-like to sii ¡ubi co sii mathen 3sg.neut.-like what 3sg.neut.-have 'If one doesn't have what one likes, one likes what one has.'
Italian examples will be cited and discussed in section 3. I leave the Polish word sii unglossed to evade commitment as to its exact lexical translatability into one. It will be more important in this paper to examine the semantic effects of the word on the whole sentence, and these will be shown to correspond in the relevant respects to the semantic function of the nominative one in English, and impersonal si in Italian. As regards the (surface) syntactic function of sig the four distributional criteria cited above suggest that it is an unstressed nominative NP; negatively, they suggest that sii i n (1) through (6) cannot be taken to function as the homophonous reflexive/reciprocal pronoun in the oblique case because it either cooccurs with such intransitive verbs as chodt(i 'goes' (1); pracuje 'works' (4); posci 'fasts' (3); or else it cooccurs with a transitive verb whose direct object is identifiable as an NP other than siç. Since the impersonal siç always cooccurs with a neuter singular finite verb and there is no morphological distinction between nominative and accusative forms of neuter nouns in Polish, systematic ambiguity results with transitive verbs and neuter NPs:
Impersonal
(7)
constructions
97
Sugeruje sii ro^wiq^anie. 3sg.neut.-suggest sg.neut.-solution
If the neuter singular NP ro^wiq^anie 'solution' is intended or interpreted as nominative, the most obvious English translation will be (7a); if the NP is taken to be accusative, either impersonal (7 b) or passive (7c) is more adequate. (7) a. b. c.
Λ solution suggests itself. One suggests a solution. A solution is suggested.
For the sake of clarity, most of the Polish examples in the remainder of this paper will contain intransitive finite verbs, and the ambiguity of the type shown above will thereby be precluded.
3. Impersonal constructions in Transformational Generative Grammar In the earlier Transformational Generative Grammar (TG) accounts of impersonal constructions in the standard (ST) or extended standard theory (EST) framework (best represented by Napoli (1976) for Italian and Wolmska (1978) for Polish), these constructions are claimed to be transformationally derived from disambiguated (deep) structures with a nominative NP other than si/sig. In Napoli (1976) the deep subject is a nonlexical PRO carrying a specified bundle of syntactic and semantic features. In Wolinska (1978) the impersonals with siç are disambiguated into deep structures with lexical NPs — mostly personal pronouns of both numbers and all three persons, which suggests multiple polysemy and, counterintuitively, complete reduncancy of impersonals in the language. What appears to be common to the ST-oriented descriptions of impersonal sentences is that on some disambiguated level of linguistic analysis (say, deep structure) such sentences have subject NPs invested with referential properties or at least well defined collections of semantic features. Therefore in interpreting an impersonal sentence compositionally, one should eventually proceed as with any other NP-VP structures, i. e., interpret the subject NP, the predicate VP and apply the subject-predicate rule. It will be shown in what follows that this is precisely what does not work with impersonal sentences, which clearly distinguishes them from other types of surface NP-VP struc-
98
Zbigniew Kañski
tures. I shall argue in the next sections that none of the intrinsic semantic or quasireferential features explicitly proposed by Napoli to be anchored to the disambiguated impersonal subject can be consistently maintained to explicate the semantics of impersonal sentences. Some more recent accounts of Italian impersonale in terms of Government and Binding theory (GB), notably those in Borer (1986), seem to be programmatically unconcerned with model-theoretic interpretation of their disambiguated levels of syntactic representation (D-structure, logical form) and it is sometimes difficult to see what formal-semantic implications those accounts support. Manzini (1986: 242), for instance, says that "the interpretation of impersonal si, as of PRO arb , is that of a free variable, as for example in: Si lava volentieri i bambini O n e gladly washes the children'." This claim is left without a comment. But if the term free variable is to be understood as in formal semantics, one can but infer that all sentences with impersonal si are open sentences which do not express any propositions and have no truth value. I doubt whether Manzini would endorse this consequence. Later in this paper I will demonstrate that there is indeed a subset of sentences with impersonal si\si$\one that deserve to be analyzed as open sentences, namely those with what will be called minimal predicates. It will be shown that the complement set of that class in the actual corpus (including Manzini's example quoted above) can be analyzed as denoting truth-valued propositions despite the free variable denotation of impersonal si\si$\one. This will require a shift from first-order to second-order predication supported by a syntactic constraint licencing impersonal sentences with nonminimal predicates only. Another recent TG concept that might be of some help in understanding the semantics of impersonal constructions is that of implicit arguments. But it is again very difficult to translate the term and the theory behind it into a language of model-theoretic semantics. One is not helped even if the exposition is partly formulated in the formal-semantic vocabulary, as in the following: The question for semantics is what does it mean for a theta role in an argument structure to be disjoint in reference from some NP. A n unrealized theta role always receives some interpretation and is assigned some reference. Often, it is existentially quantified, or gets generic reference, which is what happens to the implied agent of agentless passives. (Williams 1987: 153)
Since I will argue later in this paper that impersonal NPs are not referring expressions and they are not synonymous with indefinite generics, existential and universal quantifier phrases, they are probably not unrealized theta roles or implicit arguments. If, on the other hand, unrealized theta roles or implicit
Impersonal
constructions
99
arguments are simply those argument variables for a given predicate which have not been bound or replaced by meaningful expressions of NP (names, quantifier phrases) that do not denote free variables on any semantically interpreted level of analysis, then impersonal NPs might indeed be called implicit arguments or unrealized theta roles. There would be two consequences: first, the term "reference" applied to such arguments would mean no more than an arbitrary referential index (variable name) attributed to a free variable; 4 second, if impersonal NPs, as implicit arguments, are treated as free variables, then the question for the semantics of impersonal sentences is, again, how to account for the fact that (a subset of) syntactically well formed impersonal sentences can denote propositions but have free variables in them. This issue will be addressed in more detail in sections 4—9. First, let by Napoli impersonal perspicuity the paper.
us examine two of the semantically relevant features postulated (1976) to be associated with the disambiguated subject NP of sentences: plurality and indefiniteness. For reasons of formal the feature [+HUMAN] will be discussed in the last section of
3.1. Plurality The problem with Napoli's (1976) argument for intrinsic plurality of PRO underlying si is that it rests significantly on the surface-syntactic plural marking in the nominal predicates of impersonal sentences, such as the following (Napoli 1976: 165): (8) a.
b.
Quando si è medici, bisogna when 3sg.pres-be pl.m.-doctors need fare spesso dei sacrifici. make often sacrifices 'When one is a doctor, it's necessary to make sacrifices often.' Jak siç jest lekar^em, tr^eba if 3sg.pres.-be sg.m.instr.-doctor need csgsto si( poswiçcac. often oneself sacrifice
If the plural form of the predicate noun medici were the conclusive argument for intrinsic plurality of the subject in (8 a), then the singular form a doctor (in Napoli's own translational equivalent) and lekar^em (in what I take to be the translational equivalent in Polish (8 b)) should by analogy determine intrinsic singularity of the underlying subject in both translations. But that
100
Zbigniew
Kanski
would mean that neither the Polish sentence with siç nor the English one with one are translational equivalents of the Italian sentence, at least in respect to what is most important for us, namely the semantics of the subject. In short, either the translations provided are not translations (in the relevant respect) or the grammatical number of the predicate noun does not determine the intrinsic number of the subject NP in si¡sifone impersonal sentences. Since the referential and pragmatic potential of all three sentences seems to be identical, I would rather assume that the first of the above disjuncts is false and the second true. It follows that if anything underlies sijsiçjone on the interpretive level, it need not be intrinsically plural or singular. Therefore those morphemes or whatever underlies them must be intrinsically neutral or undetermined with respect to number as possibly distinct from surface syntactic number (and number agreement). The fact that the predicate noun assumes the plural form in Italian siimpersonals, and singular in English and Polish impersonals only shows that the syntactic number agreement in impersonal sentences is a matter of a language-specific convention. The predicate noun must, as any old noun, be spelled out as either singular or plural, tertium non datur. There is no distinct form in the inflectional paradigms of the languages at issue that would mark lexical nouns as neutral with respect to number. For essentially the same argument, see Manzini (1986).
3.2. Apparent indefiniteness of one\si\siç The disambiguated subject NP in Napoli's (1976) analysis of impersonals is specified as indefinite. Indefiniteness is defined in the same terms as it is usually defined for the contrast between NPs with the definite and indefinite article. This analogy is fortunate because it makes the notion "indefinite Pro" explicit enough to be falsiflable. It is unfortunate because to the extent it is explicit it is false. Napoli's argument for the indefiniteness of Pro is largely based on the conjecture that both the latter Pro and NPs with the indefinite article can have the specific reading and the nonspecific or generic readings. Thus, for instance, si in (9) (Napoli 1976: 171) is argued to be interpreted as a specific pronoun, and in (10) (Napoli 1976: 167) as a nonspecific generic pronoun. (9)
L'unico Van Gogh si è venduto un'ora fa. the-only Van Gogh is sold an-hour ago 'Someone sold the only Van Gogh an hour ago.'
Impersonal
(10)
constructions
101
In Italia, quando si incontra un amico, in Italy when meets a friend si va a prendere un c a f f è . goes to take a coffee 'In Italy, when one meets a friend, one goes for coffee.'
In (9) si is argued to refer to "some particular individual who sold the painting" (Napoli 1976: 171), while in (10) it is argued to refer to the typical human in Italy. The problem is that even if those informal interpretations are intuitively correct, they in no way follow logically from any formal disambiguation of si into a referring indefinite plural third person Pro-NP. Without any explicit treatment of reference it is difficult, if indeed possible, to reconcile the claim that Pro is a referring NP with, for instance, the "normative" interpretation of (10) whereby si "refers" neither to any particular individual nor to each and every individual or any indefinite group of individuals, but rather to any individual that possibly or potentially satisfies the predicate of the quando clause. The treatment presented so far is inadequate precisely because it cannot predict that the relative referential specificity, if any, of the putative Pro is determined by the referential specificity of the predicate whose subject si (or Pro) is, and not by any intrinsic features independently specified for the disambiguated subject of //-sentences. The putative Pro subject of (9) has a relatively more specific reference because the direct object l'unico Van Gogh is obviously specific and perfective aspect plus a definite time reference of un'ora f a are, trivially, more specific than the simple present of (10). This combined with the opaque context provided by the quando-clause in (10) should lead to such a formal explication of impersonal sentences which would on the one hand predict that si, siç and one are not polysemous items, and on the other hand would predict that the denotation of those items is a function of the denotation of their predicates. Hence the specific/generic interpretation of impersonals (if the distinction is worth maintaining in this case at all) will not entail any intrinsic polysemy of si/sig/one itself. Let us now show that any identification of the meaning of impersonal subjects with "generic" indefinite NPs, to the extent it is explicit, leads to false conclusions. As an example of the indefinite NP with generic reference, Napoli cites the following: (11)
Una tigre è feroce. a tiger 3sg.-be sg.-ferocious Ά tiger is ferocious.'
102
Zbtgnisw Kañski
Now if si, si{ and one indeed referred in the same way as a tiger in (11) with the only difference that they should denote the typical human or the class of humans (or whatever indefinite generics are assumed to denote) rather than tigers, the following pairs of syntactically well formed sentences should be semantically equivalent: (12) a. b. c.
? Si è feroci = IJuomo è feroce 3sg.-be pi.-ferocious the-man is sg.-ferocious ? One is ferocious. = Λ man is ferocious. ? Jest sii okrutnym = C^fowiek jest okrutny is instr.-ferocious man is nom.-ferocious
Yet the predicted equivalences simply do not hold, even if the right-hand indefinites are interpreted as nonspecific generics. I claim that the ultimate reason of the intuitive nonequivalence of the above pairs is that (11) and the right-hand expressions in (12) denote independent falsifiable propositions, while the left-hand expressions in (12) are syntactically well formed sentences but denote unfalsifiable propositional functions. Sentence (11) and the right-hand expressions in (12) are falsifiable in that the set-denoting predicative adjective ferocious is therein predicated of the class/set of tigers and humans respectively, where the latter classes/sets are explicitly stated by means of common nouns tiger and man. Now the expressions at issue are independently falsifiable propositions inasmuch as the set of tigers (humans) either is or is not a proper subset of the set of ferocious individuals: (13) a. b.
{x: tiger(x)} c {z: ferocious(z)} {x: human(x)} cz {z: ferocious(z)}
The left-hand expressions with si, one and siç in (12) are not independent falsifiable propositions (and therefore are not equivalent to the right-hand expressions) because if their (nominative) subjects denote anything, this denotation must at best be seen as the improper subset of the set denoted by the predicate ferocious, and therefore identical to the latter. This really amounts to saying that si, one and siç are ultimately translatable into lambda-abstracted individual variables, and the whole expressions are translatable into tensed but unfalsifiable predicate abstracts of the form χ [is ferocious(x)]. Before this proposal is elaborated, let us see some of its immediate consequences and advantages over any equivocation with NPs built of the indefinite article and a noun. Consider the following sentences, which are both syntactically well formed and are translationally equivalent:
Impersonal
(14) a. b. c.
constructions
103
Si è morti. is pl.-dead One is dead. Jest siç martwym. is sg.instr.-dead
Yet in spite of their syntactic-lexical well-formedness, these sentences may not independently and falsifiably assert anything about the world. For they do not mean that some particular (specific) individual is dead, or that a group of individuals is dead; nor do they mean that the class of humans (human kind) is extinct. On no formal or even informal understanding of the notion "generic human" can si/one/siç in (14) be held to be coextensional with that understanding. If one assumes that the generic human is but the set (or intension thereof) denoted by such predicates as human (being) or man, then si\one\siç do not denote the generic human because (14) is not equivalent to the falsiflable proposition whereby the set of human individuals is a proper subset of the set of dead individuals (or an intensional variant therof). Neither is (14) equivalent to a generic proposition where the subject NP is interpreted, like bare plurals in Carlson (1977), as a name of kind (human kind in this case).
4. Impersonal NPs as abstracted variables The assumption that structures such as si è morti denote as tensed predicate abstracts has an important consequence. These structures, in spite of their legitimate syntactic NP-VP constituency, cannot be semantically represented and/or interpreted as empirically or logically falsifiable propositions, and hence it should be impossible to provide any coherent (more or less formal) truth conditions for them. The foregoing discussion showed that any attempt to equivocate between impersonals on the one hand, and existential, universal or generic propositions on the other, is misguided. The latter can each be analyzed as having a bipartite semantic structure corresponding to the NPVP syntactic structure (e. g., in terms of the generalized quantifiers theories of NP). The impersonals in (14 a—c) cannot be thus analyzed because the predicate is dead does not seem to be predicated of any entity or a variable bound by a conceivable quantifier (except the lambda operator of functional abstraction); alternatively, there seems to be no straightforward answer to the question what property-set denoted by the NP siI siç ¡one has the property
104
Zbigniew
Kañski
is dead as its element, if it is neither the set of properties of some (specific or nonspecific) man nor the set of properties that all men have or else the set of properties of the kind homo sapiens. In short, those who would insist that one is dead expresses a proposition should answer what exactly it asserts and how to get to prove whether it is true or false. The assumption that (14 a—c) denote merely tensed predicate abstracts obviously does not pose these problems, for predicate abstracts cannot assert anything by themselves. It is useful at this point to introduce a working definition of a minimal predicate. I shall use this term to refer to syntactically simple (nonderived) one-place predicates. Thus transitive verbs with their objects will not be minimal predicates, nor will intransitive verbs (and predicative adjectives) with their optional modifiers (e. g., adverbial VP-modifiers). In this sense ferocious in (15 a), (is) dead ((14 a —c), and eats in (16 a) below are minimal predicates; in the same sensc, ferocious in a very hot weather in (15 b), eats swordfish in (16 b), and eats smrdfish with two forks in (16 c) are not minimal predicates: (15) a. b. (16) a. b. c.
?? One is ferocious. One is ferocious in a very hot weather. ?? One eats. One eats swordfish. One eats swordfish with two forks.
Now minimal predicates qua legitimate one-place predicates should by definition take one NP subject argument and thus produce a syntactically well formed sentence; if the argument is not a free variable, the sentence should be interpretable as a falsifiable proposition. As we have shown earlier, this works with canonical indefinite NPs (including those interpreted generically) but does not work with impersonal NPs si/sifone. In both cases syntactically well formed NP-VP sentences are produced but only in the former (indefinite NP) case, semantically "complete", are falsifiable propositions produced. The situation changes significantly when the impersonal NP is combined with nonminimal predicates, as in (15 b) and (16 b, c). While (15 a) does not really assert anything (the minimal predicate ferocious is not really predicated of anything), (15 b) can already be interpreted as asserting, informally speaking, some verifiable relationship between ferocity and very hot weather; similarly, while (16 a) virtually lacks any propositional content, (16 b) can already be interpreted as claiming some relationship between eating and swordfish (the Polish equivalent with sig would be a good paraphrase of Swordfish is edible), and (16 c) as asserting a relationship between eating swordfish and the manner of doing it. Generally, it may be observed that the
Impersonal constructions
105
more otherwise optional a complement or adverbial modifier is the better (more "complete") the impersonal proposition is. One sensible way of explaining the behaviour of the impersonal NPs with minimal and nonminimal predicates is that the binary syntactic NP-VP constituency is not matched in the main binary semantic constituency. With minimal predicates, the nonreferential subject NP cannot really be evaluated as having or not having the property denoted by the predicate and, as a result, there is no semantic argument for the predicate to characterize. With nonminimal predicates, the nonreferential subject NP can enter a falsifiable ("full") proposition only if the semantic argument-predicate structure does not mirror the syntactic NP-VP structure, but is shifted to within the syntactic predicate itself. This reanalysis is possible for, by definition, the nonminimal predicate consists of more than one constituent. Thus (15 b) above may be reanalyzed semantically as having the property of being ferocious as the argument of the second-order property expressed by the syntactic adverbial in a very hot weather. We defer a more formal analysis of similar examples to section 7.
5. Impersonal conditionals Conditional sentences such as (17 a—c), from Napoli (1976: 164), are a frequent context in which impersonal constructions with si/sig/one occur. We noted above that the antecedent of (17 a —c), containing a minimal predicate, is virtually meaningless taken as an independent proposition. However, (17 a—c) as a whole is a meaningful, falsifiable proposition. Significantly, this conditional, rather than a sentence with a minimal predicate is given by Napoli to show that si denotes like a generic indefinite NP. (17) a.
Una volta a time b. Once one is c. Jak siç jest when is ιviçcej nie more not
che si è morti, non si vede più la luce. that is dead not see more the light dead, one doesn't see the light any more. martwym, to siç ju^ dead then already ivid^i swiatla. sees light
The problem with analyses of si\si$\one in terms of indefinite generic (human) NPs is that it cannot coherently explain why si è morti by itself is meaningless, yet is miraculously meaningful as an antecedent of a conditional
106
Zbigniew Kanski
whose consequent is usually also an impersonal structure and the whole complex sentence is a meaningful proposition. Again, the antecedent itself, although it is syntactically a well formed NPVP sentence, does not denote an independent proposition. In our proposal it denotes merely a (tensed) property abstract. One important corollary of this is that the syntactic subject NP cannot be taken to denote like an indefinite NP ( = indefinite article + noun) or like a restricted or unrestricted quantifier phrase (existential/universal quantifier + a variable possibly restricted to some domain, e. g., human). If it could, the antecedent si è morti would be an independent proposition, and it is not. One reasonable way out of this dilemma is to treat the whole conditional (17 a—c) as combining two impersonal sentences denoting predicate abstracts directly by the inclusion relation: (18)
{x: is dead(x)} c {y: doesn't see the light, .(y)}
In short, (17 a—c) assert (falsifiably) that the property of being dead "involves" the property of not seeing the light any more, and they do so without independently predicating either property to specific individuals or classes of individuals other than those denoted by the predicates themselves. 5 But this interpretation of (17 a—c) is obviously equivalent to a formula with the universal quantifier in (19): (19)
V x[tífoí¿/(x) —» doesn't see the light, .(x)]
It is equivalent to the extent that any assertion of set-inclusion is expressible in terms of universal quantification. Thus the only difference between conditionals combining two "full" propositions (as in propositional calculus) and those combining two impersonal structures would be that the former state the inclusion relation between two sets of worlds ( = propositions), and the latter state the inclusion between two sets of individuals. In both cases the connective if (or related temporal-conditional connectives such as when) combine syntactically well formed sentences, and in both cases the connective may well be considered to denote the proper inclusion relation itself. In spite of the equivalence of the two representations of (17), there are good reasons not to represent (17) and related sentences directly in terms of universal quantification over individuals, but rather in terms of set-inclusion. One reason may be stated as Partee's (1979) Well-Formedness Constraint or Keenan's (1979) Naturalness Condition whereby, roughly, the disambiguated formal representation of natural language expressions should as closely resemble the "natural" syntactic structure of these expressions as possible. On this account (18) is preferable if only because there is no overt expression in (17)
Impersonal
constructions
107
that could be represented directly as the universal quantifier (and each of the three languages considered obviously has expressions of the everj/aII kind which are comfortably represented as quantifiers). By the same token, (17 a—c) do not show overtly any binding mechanism (say, personal pronouns qua bound variables) whereby on any value assignment to the variable in the antecedent, the consequent variable will have that same assignment (which is directly expressed in (19) by means of identical variable names in the scope of the universal quantifier). If (19) were indeed the representation of (17), one would be forced to disambiguate each consecutive occurrence of onejsij siç (except for the first!) into "an identical bound variable." This obviously would be inconsistent with our findings about impersonals with minimal predicates and, as will be shown in section 7, nonminimal predicates in simple intransitive sentences. This problem does not even arise if (17) is represented as (18), where each instance of one\si\sii is represented independently as an abstracted free variable restricted only to the domain expressed by the appropriate set-denoting tensed predicate. The name of the variable in each occurrence is any name we wish, quite independent of the other, since each is locally abstracted and two predicate abstracts are thereby produced; combined by the conditional (inclusion) connective, they yield a fasifiable proposition (in that the set of dead individuals is or is not a proper subset of the set of individuals who do not see the light). An important feature of this approach is that it links the ultimate interpretation of impersonal conditionals such as (17) in terms of universal quantification with the overt occurrence of a subordinating expression {if, when, una volta che, etc.) interpretable as a conditional or implicational connective. Hence the universal interpretation is not related to the representation and interpretation of the impersonal NP itself. This way I also wish to resist any temptation to treat overt conditionals as the paradigm context for impersonal NPs and force the conditional/universal representation on simple impersonal sentences with nonminimal predicates (see section 7). 6 Note also that the introduction of unselective quantification along the lines of Lewis (1975) to bind all instances of free variables denoted by impersonal NPs would be superfluous, for it was designed to quantify admissible assignments of values to several variables in a formula; if two instances of si in (17) were represented as two instances of a variable denoting indefinite NPs in the scope of unselective universal quantification, (17) would mean that under every admissible assignment of values to χ and y, if χ is dead y does not see the light any more — an interpretation that has little to do with the meaning of (17). In terms of our representation with two predicate abstracts
108
Zbigniew Kañski
combined by set-inclusion, we can give the variables different names (say, χ and y) and still get the intuitively desirable interpretation equivalent to the orthodox selective universal quantification (binding two variables of the same name). This also naturally explains why each consecutive occurrence of sijsi^jone in conditional and other multi-clausal impersonal sentences is somehow interpreted as anaphorically related to the preceding one, and yet the expression si/sig/otie is not by itself an anaphor (both loosely and in the sense of recent versions of TG) or a bound variable, for its first occurrence in a complex sentence or its only occurrence in a simple sentence does not have to be anaphorically related to anything; at the same time we do not seem to be dealing with any polysemy of si¡si(¡one within one complex sentence such as (17)! It might be objected that the interpretation of impersonal NPs as indefinite human generics is an equally plausible explanation of this phenomenon. For once we somehow establish the extension of something like "a/the generic human" or "a/the typical human", each of its consecutive occurrences in a complex sentence will be coextensional with its first or only occurrence, and the problem of apparent anaphoricity would not even arise (on the sound assumption that there are no two different generic or typical entities of some kind, unlike two different specific entities of that kind). However, this would still leave us with the polysemy between specific and generic impersonals, and the latter distinction was earlier in this paper dismissed as untenable, for the relative specificity/generality of the impersonal NP was shown to be a function of the relative specificity of the information in the predicate. Likewise, there are good independent reasons not to treat impersonal NPs as human generics, some of which appeared in the preceding sections and will be discussed in more detail in the following parts of this paper. Since our approach in terms of abstracted variables handles both the dependence of impersonal NP denotations on predicate denotations and the apparent anaphoricity issue, it is preferable.
6. Obligatory complement constraint In her attempt to confront her transformational analysis of Polish impersonal .«'^-constructions with a representative corpus of literary Polish data, Wolinska (1978: 81) notes what I take to be a crucial property of the data:
Impersonal
constructions
109
The use of the analytic form with sig does not alter the right-hand connotation of the verb which preserves its characteristic type of complementation. The transformation does not effect any qualitative changes in the connotation of modifiers and complements; but at the same time the transformation introduces the requirement that at least one possible modifier be present in surface structure. In each of of the 450 analyzed sentences there occurs some modifier narrowing the range of the predicate... In unmarked uses the structuralization of one complement/modifier is obligatory even if the latter constituent is traditionally recognized as marginal and optional... Especially frequent in the surface are manner, place and time adverbials. 7
Let us take Wolinska's observation for granted in terms of some Obligatory Complement Constraint (OCC) on impersonal sentences, where the term "complement" should be taken to comprise, as in Jackendoffs (1977) X' theory, both the right arguments strictly subcategorized by the lexical verb (V' complements) and various otherwise optional modifiers (such as V" adverbial complements). (I assume without any corresponding statistical evidence, that the same regularity holds in Italian and English data.) The problem with the empirically verifiable OCC is that it in no way follows from Wolinska's or Napoli's transformational derivations, and is completely ignored in recent generative accounts of impersonals known to me, e. g., Manzini (1986), Hyams (1986). This is all the more striking in view of the fact that Manzini wants to interpret the impersonal si as a free variable (without discussing the logical consequences of this decision) and both authors explicitly identify the semantics of the impersonal si with that of PRO arb . The trouble with this PRO is that its semantics had never been explicitly defined to the best of my knowledge. The term "reference" suggests that it is a referential expression like other NPs, and the cryptic term "arbitrary" is an ingenious way to evade commitment as to what is referred to by this NP; it invites associations with both universal quantification and indefinite NPs, both inadequate in handling the semantics of impersonals. OCC is inexplicable in any theory that more or less explicitly interprets si/ siç/one as referentially nonempty NPs (however indefinite or arbitrary its reference is taken to be). For why would a syntactically and semantically complete NP-VP structure require at least one otherwise optional constituent in the VP? In our framework whereby impersonal NPs are represented as abstracted free variables, the syntactic constraint OCC is but a corollary of the semantic incompleteness of impersonal sentences with minimal predicates as defined above. The most general conclusion that follows from OCC and the imcompleteness of impersonal sentences with minimal predicates is that the main syntactic-categorial division of the impersonal sentence cannot possibly corre-
110
Zbignieiv Kañski
spond to the main binary semantic division of this sentence. If the main syntactic relation between two binarily distinguished well formed constituents (subject-NP and predicative-VP) and the main falsifiable semantic relation between some model-theoretically defined constituents are subsumed under one general term "predication", then in the case of impersonal sentences the syntactic predication does not correspond to the synthetic semantic predication. Nonminimal impersonal sentences (i. e., those in which OCC is satisfied) can be analyzed as synthetic truth-valued propositions only if the synthetic predication obtains not between the denotation of the nominative subjectNP argument (deep or surface) and the denotation of the syntactic VP, but between other set-theoretic objects. In the preceding sections it was suggested that at least one of these main constituents of predication is the tensed predicate abstract itself (which in overt syntactic structure consists of the impersonal NP si\si^\one on the one hand, and the minimal VP or a part of the nonminimal VP, on the other). Thus we seem to be dealing with the shift from first-order to second-order predication in semantics, while the overt syntax (NP-VP) remains the same as in first-order predication, except that the NP dominates a designated impersonal lexical item — a signal for the shift. In the preceding section on impersonal conditionals, the synthetic second-order relation was shown to obtain between predicate abstracts corresponding to impersonal clauses. But as suggested in section 4, OCC is also satisfied in simple sentences, provided that their predicates are nonminimal; most frequently they contain otherwise optional adverbial modifiers. We return to these in the following section.
7. Adverbial modifiers as second-order properties Consider the following examples: (20) a.
b.
c.
Pracuje siç dia pr^yjemnosci. works for pleasure 'One works for pleasure.' Maria pracuje. Mary works/is working 'Mary works.' ?? Pracuje siç. 3sg. pres.-work 'One works/is working.'
Impersonal
constructions
111
(21) a.
ψ Wigilk sii posci. in Eve fasts 'One fasts on Christmas Eve.' t>. Jan posci. John 3sg.pres.-fast 'John fasts/is fasting.' c. ?? Posci si{ 3sg.pres-fast 'One fasts/is fasting.' (22) a. Tu sii η™ plywa. here not swims 'One doesn't swim here.' t>. Kot nie plywa. cat not swims 'A/the cat doesn't swim.' c. ?? Nie plywa siç. not swims 'One doesn't swim.'
Both the impersonal (a) sentences with adverbial modifiers and the "personal" (b) sentences without adverbials can be judged as asserting something even without any context. In this sense they are semantically complete and denote falsifiable propositions. The completeness of the (b) examples is not impaired by the fact that their predicates are minimal. Their truth values depend ultimately on whether or not the entity denoted by a proper name or an indefinite generic (whatever the latter entity is) is an element of the set denoted by the tensed predicate. In contrast to the (a) and (b) sentence, the (c) sentences are semantically incomplete (open) in that the intransitive minimal predicates in them are not falsifiably satisfied by any set-theoretic object distinct from the denotation of the predicate itself. Since the only observable difference between the complete and verifiable (a) examples and the unverifiable (c) examples is the presence in the former of otherwise optional adverbials, it is sensible to assume that the denotation of the adverbial is one of the two main semantic constituents in the (a) sentences. In short, the presence of otherwise optional adverbials ensures that the (a) sentences can assert something. Again, OCC in this case can be coherently explained only if one admits that the semantic value of impersonal sentences cannot be computed from the denotation of the intransitive predicate on the one hand, and some nonempty denotation of the
112
Zbigniew KaAski
nominative subject NP. Let us now see what it can mean in more formal terms. Traditional VP adverbials (lexical or phrasal) are defined in the syntax of categorial grammars as functions mapping Intransitive Verbs into Intransitive Verbs (where the category Intransitive Verb (IV) is roughly equivalent to the TG category VP and includes combinations of transitive verbs with direct objects). In terms of set-theoretic denotation type, these adverbials will canonically denote functions from sets (or properties) of individuals to sets (properties) of individuals. Since the sets which are values of such functional application rules are narrower than the argument sets to which the adverbial functor applies, the resulting combination can be seen as denoting a subset of the set denoted by the argument IV. Hence such adverbials as for pleasure, on Christmas Eve, and here in the above examples can be conveniently referred to, after Bennett (1975), as subsective modifiers. Canonically then, the nonbasic adverb for pleasure as a functor IV/IV will apply to an IV, eg., works, and thereby yield another IV, i. e., works for pleasure. The latter, by definition, will denote a set which is a subset of the set denoted by the IV works (in that the set of individuals who work for pleasure is a subset of individuals who work). From the syntactic-categorial definition of IV as a function S/NP (mapping NPs into sentences) it follows that our derived IV works for pleasure combines with an NP argument and thereby yields a syntactically well formed sentence. The logical value of the resulting sentence depends on whether or not the denotation of the argument NP satisfies the set-denoting IV on an assignment (pace the generalized quantifiers framework where NPs denote property-sets). The special status of impersonal sentences such as our (20 a) —(22 a) consists precisely in that the derived set-denoting IV works for pleasure legally combines with a nominative NP argument one\si\si$, but at the same time the latter argument is referentially empty (does not denote any set-theoretic object that could be assessed to satisfy the predicate). As noted before, the nonreferentiality of the syntactic subject-argument directly predicts the oddness of minimal impersonal sentences such as (20 c) —(22 c). On the other hand, the independently observable OCC and the semantic completeness of the (a) sentences imply that the otherwise optional subsective modifier must be reanalyzed as one of the two main semantic constituents of the proposition. The latter postulate will be perhaps most straightforwardly realized if the subsective adverbial, which otherwise maps sets into their subsets, is reanalyzed in impersonal sentences as that function which maps sets (or properties) of individuals directly into truth values; as such, the adverbial will change its semantic type from a subsective modifier to a second order property (i. e.,
Impersonal constructions
113
a propertie of properties of individuals). Thus reanalyzed, the denotation of the adverbial for pleasure in the context of the impersonal subject can be represented as the following second-order predicate abstract: (23)
Ρ [Jor pleasure'
(P)]
where Ρ is a variable ranging over first-order properties of individuals. Now a possible value for Ρ is the first-order property represented by our tensed predicate abstract χ [works' (x)] — a representation that was argued for the sequence one works. Thus the functional application gives the following proposition: (24)
Ρ [tfor pleasure' (P)] (x [works' (x)]) = [ t f o r pleasure' (x [works' (x)])]
Represented and interpreted this way, (20 a) will be true if and only if the property (say, "activity") of working is in the set of properties (activities) that have the property of being "done" for pleasure. The closest less clumsy paraphrase I can think of is work ¡working is (done) for pleasure. By analogy, on Christmas Eve in (21 a) is also made to characterize a set of properties of individuals, and the property of fasting is asserted there as being in that set. There are two immediate advantages of this reanalysis theory of impersonal sentences: firstly, it corroborates the syntactic constraint OCC on impersonal sentences (the obligatory presence of otherwise optional complements and modifiers); secondly, it solves the tricky problem of why (21 a) may be true even if not everybody, not even most people actually fast on Christmas Eve. Indeed, (21 a) may be true even in a rather perverse situation whereby nobody actually observes the fasting. For fasting may well be declared or recognized as true for Christmas Eve rather than Christmas Day, without any commitment whatsoever as to how many people do or should actually observe it; for we are really talking about a set of properties characterized by the day of Christmas Eve, rather than mentioning or quantifying individuals characterized by the property of fasting on that day. This way there is no need to postulate any overt or covert first-order quantification for the semantics of impersonal sentences satisfying OCC.
8. Impersonale vs. indefinites revisited In section 3.2. the identification of si/siç/one with indefinite NPs was questioned because it could not explain the different behavior of the two kinds of NP with minimal predicates; likewise, the constraint OCC applies only to imper-
114
Zbigniew Kañski
sonáis and not to indefinite NP subjects. This does not mean that there is no affinity between impersonal sentences and the way indefinite NPs denote. The problem is worth considering especially in the light of some formal analyses of indefinite NPs as variables, e. g., Heim (1982), Krifka (1987). For lack of space, I will not present Heim's theory of indefinites in any detail. Suffice it to say, indefinite NPs such as a man are represented semantically as propositional functions (predicates with individual variables); their apparent existential or universal quantifier reading is not thereby inherent in the semantics of the indefinite NP itself but results from the binding of its variable by a quantifying adverb (cf. Lewis (1975)) or another operator that is represented as having the indefinite in its scope (a default quantifier in the case of indefinite generics in Krifka (1987)). The indefinite thus is represented just as a restrictor for the quantifier or operator, rather than a quantifier plus a restrictor, as in traditional formal analyses. To the extent that indefiniteness is represented in terms of free variables, the impersonal expressions si/si^/one are indefinites in my approach. However, the "surface" syntactic impersonal NPs si\si£\one cannot be represented on a disambiguated level as Heim's indefinite NPs are for one simple reason: there is no property-denoting common noun (explicit or implicit) within the impersonal NP itself to be represented as a predicate (and a restrictor) for the individual variable. This obviously prevents minimal impersonal sentences from being represented in Heim's system as tripartite structures consisting of an operator, restrictor (antecedent), and nucleus (consequent), and could be just another way of explaining the semantic anomaly of impersonal sentences which do not satisfy OCC. On the other hand, the analogy between impersonale in my system and Heim's indefinite NPs can be significantly extended if we take the impersonal NP (si/sq/one) to be analogous just to the indefinite article (represented as a variable), and the tensed verbal predicate (or a part thereof) in the impersonal sentence to be analogous to the common noun in the canonical indefinite NP. If so, the whole sentence One swims would be semantically analogous just to the indefinite NP subject a dog in A dog swims·, if Heim's invisible operator is postulated in the representation of these two sentences, the predicate swims will play two different roles in them: it will be a mere restrictor in the impersonal sentence, and a nuclear (consequent) predicate in the sentence with the canonical indefinite NP (with dog as a restricting predicate). The semantic incompleteness of minimal impersonal sentences (and OCC as its corollary) would be due to the lack of the nuclear (consequent) part in them; by the same token, the conditionals with two impersonal clauses, such as (17) above, would be complete propositions insofar as the first
Impersonal
constructions
115
impersonal clause would act as a restrictor-antecedent and the second one as a nucleus-consequent. In this sense Heim's and related theories of indefinites might as well explain such relevant facts about impersonale as their incompleteness with minimal predicates on the one hand, and their role in conditionals, on the other. However, I am not so sure whether this theory would work as well on those impersonals which satisfy OCC by means of otherwise optional adverbials, such as (20 a) —(22 a) above. There is one important aspect of Heim's (1982) analysis of indefinites that I do not think is applicable to impersonal constructions without some costly revisions (however well it works on canonical indefinite NPs). While indefinite NPs are represented there as variables (with a domain restricted by a common noun denotation) rather than quantifying expressions, these variables are ultimately bound, through the back door, as it were, by some quantifier or modal operator residing outside the indefinite NP itself. The latter quantifier may be present in the immediate context in the form of one of Lewis' (1975) unselective quantifying adverbs, such as always in (25); in the absence of an overt quantifying expression, one is postulated in the semantic representation by convention: it will be an "enriched" necessity operator over worlds and individual-sequences in the case of bare conditionals, such as (26), and indefinite generics, such as (27); it will be an existential quantifier (performing what Heim calls "existential closure") in other well defined cases. (25) (26) (27)
If a man owns a donkey, he always beats it. If a man owns a donkey, he beats it. A cat that has been exposed to 2,4-D goes blind.
This is a gross oversimplification of Heim's project and I do not intend to question its value as a solution to numerous problems concerning standard indefinite NPs (notably the notorious puzzle of "donkey" sentences). However, I would not endorse the "invisible" quantification part of the project as applicable to the semantics of impersonals of the sijsi^jone kind. It may be spurious at best, and counterintuitive at worst. Compare our (21 a) to (28): (21) a. (28)
One fasts on Christmas Eve. One always fasts on Christmas
Eve.
If always is a quantifying adverb in (28) and one is a variable-denoting indefinite (restricted by the predicate fasts, as argued above), then always should act like an (unselective) universal quantifier that binds the individual variable (and perhaps a variable over times), and we would get a first-order formula to the effect that "every individual that fasts does it (fasting, that is)
116
Zbigmew Kanski
on every Christmas Eve." I doubt whether this reading would correctly represent our intuitions about the meaning of (28); since a similar representation would be suggested for (21 a) with no overt quantifying adverb, it would be counterintuitive in both cases. I believe that (21 a) really says something to the effect that Christmas Eve is a day of fasting; in (28), always universally quantifies only individual times called Christmas Eves, and the sentence may be paraphrased accordingly as "Every Christmas Eve is a fasting day." In neither sentence is the individual variable denoted by one unselectively bound by the quantifying adverb (or any other operator). As argued before, these intuitions and the syntactic constraint OCC may be accounted for more straightforwardly if we let the variable stay free (more precisely, abstracted by the lambda-operator) and make the otherwise optional adverbial such as on Christmas Eve denote a second-order property. This way the property of fasting would be "associated" with Christmas Eve in (21 a), and (somewhat redundantly) with every Christmas Eve in (28). This approach frees one from pondering on how many fasters it requires for (21 a) and (28) to be true in the actual or ideal world. It may well be that the pragmatics of impersonal sentences depends precisely on escaping this kind of commitment. For predicating something of a property does not entail predicating it of every or any individual that is in the extension of that property.
9. How human si\siç\one is It was argued in the foregoing sections that the lexical NP sijsi^one cannot be represented as containing any restrictor of the variable-domain that would function like a common noun in a "regular" indefinite NP. The issue of impersonale with minimal predicates and OCC decided against any equivocation between impersonal structures and indefinite human generics. Yet it is unquestionable that impersonal sentences somehow "refer" to human individuals rather than any kind of individuals from the model. This is so even if the tensed predicate in the impersonal structure does not select exclusively human subjects, and in some cases the human reading may override a normally nonhuman selectional requirement of the predicate, as in (29): (29)
One doesn't bark at guests.
This sentence has a human reading despite a strong canine selectional requirement of the predicate bark. Yet (29) is not synonymous with No man
Impersonal
constructions
117
barks at guests or A man doesn't bark at guests (in a generic or existential sense). Identical judgements apply to Polish impersonals with sii an Across VP > Extraction > Raising > P-stranding
Movement which does not cross any maximal projection, e. g., scrambling in Korean, hardly diminishes transparency at all since it affects only the respective order of the various constituents of a clause, each of which remains clearly marked for its role by its accompanying case marker. The operation of scrambling does not allow stranding of case markers, nor is it capable of moving anything out of, or into, any noun phrase or clause. Hence, no reduction of transparency results. Among movements which do cross maximal projections, VP is easiest to cross (e. g., passive in German) since the moved noun phrase remains morphologically identified for its role by a combination of its nominative case marking and the passive verb morphology. Movements across a clause boundary, i. e., extraction and raising, are clearly much more transparency reducing as they remove an element from the predicate that it needs to be interpreted with, and surface-align it with another predicate with which it contracts no semantic relationship. Still, extraction ranks higher in the hierarchy, for at least the extracted element does not become a syntactic argument of the higher predicate, as is the case with
Towards a semantic typology of language
145
raising. The extracted element, hence, never becomes syntactically or semantically associated with the higher predicate, leading to less of a reduction in transparency than raising processes. Where a more transparently oriented language permits extractions at all, as in German, the extracted elements carry their case markers along, thus remaining marked for their role, albeit in the "wrong" clause. Raising, on the other hand, always involves the syntactic integration in the higher clause as either a subject or an object, and thus more grammaticization and less semantic transparency. P-stranding, finally, exemplifies grammaticization in its purest form since stranding an adposition removes the only element which indicates the role of the noun phrase it governs. Consequently, that noun phrase is literally left stranded as a foreign body in the sentence which at least temporally remains without any indication of its syntactic or its semantic function. More transparently oriented languages can avoid this loss of transparency by resorting to pied-piping as a means for retaining the integrity of a moved element. The lower leg of the movement hierarchy is, thus, motivated by the degree to which a given process disintegrates the indication of the syntactic and the semantic function of an element. Independently, the transparency of each movement type can be somewhat increased by leaving a pronoun copy in the extraction site. For these reasons, highly transparent languages restrict movement to the first type, using scrambling processes for coding pragmatic functions. The possibility of encoding pragmatic information makes this kind of movement the only one that can be argued to actually increase the transparency of an utterance. Interestingly, a highly grammaticizing language like English does not just relinquish pronoun copies, at least in its standard variety, but is also forced to do without exactly this transparency-inducing movement type as a result of its strongly grammaticizing grammatical relations. 4 Extremely grammaticizing languages like English are, then, characterized by the presence of the four other movement types. If it is correct that deletion under identity, as, for example, in conjunction reduction or equi structures, reduces transparency, highly transparent languages should tolerate them to a much lesser degree, and, to the extent that they do occur, should prefer to provide pronoun copies at "extraction sites" much more frequently than grammaticizing languages. Finally, semantically transparent languages differ from grammaticizing languages on the lexical level. First, they exhibit greater lexical specificity. Secondly, they need to employ different verb forms or different verbs altogether if a different argument structure is used. In this property, they differ
146
Fran\
Müller-Gotama
from languages like English where zero derivation is frequently used in such cases. The properties discussed in this section combine to yield the following constellations for languages that are consistently located close to the two ideal types defined by the Semantic Typology. Table 1. Strongly transparent
Strongly grammaticizing
Little semantic diversity of grammatical relations
Much semantic diversity of grammatical relations
Overt role markers for grammatical relations
N o overt role markers for grammatical relations
Free scrambling N o raising No out of clause extraction No P-stranding Little deletion of noun phrases
Fixed intraclausal word order Widespread raising Frequent extraction Frequent P-stranding More deletion of noun phrases
High lexical specificity
Low lexical specificity
It is evident that the portrayal of English given in Hawkins (1986) positions that language fairly close to the ideal grammaticizing type. English nouns have no case markers apart from the genitive, which does not serve as a marker of basic grammatical relations. Remnants of a more extensive earlier case marking system are only found in its pronouns. Moreover, English has a rather fixed word order within a clause, semantically extremely diverse grammatical relations, several raising processes, which are triggered by a large number of predicates, frequent preposition stranding, significantly lower lexical specificity, and more deletion of noun phrases than German. By contrast to the extreme position of English on the scale, German diverges significantly from the ideal transparent set of values in several ways. While it indeed has overt case marking, low semantic diversity of the basic grammatical relations, and fewer deletions of noun phrases (Hawkins 1986: 115 — 120), it allows both out of clause extractions and some raising, though less so than English. In addition, its word order, while freer than that of English, does not approach free scrambling as do some other languages. Finally, the level of lexical specificity in German approaches that of English, contrasting sharply with the situation in Russian (Russian data from Comrie, 1986b).
Towards a semantic typology of language
147
Table 2. Lexical specificity in English, German and Russian English
German
Russian
make/sew a dress makejbake bread makejbrew tea make ¡weave a rope makejweave a nest make]lay a road ? make ¡build a house
ein Kleid nähenjmachen Brot backenjmachen Tee kochenjmachen ein Seil webenjmachen ein Nest bauenjmachen eine Straße bauenjimachen ein Haus bauenjimachen
sit' plat'e pec' xleb varit' caj vit' verevku vit' gne^do prolo^it' dorogu Stroit' dom
One goal of the present research has therefore been to identify a language which more faithfully represents the transparent end of the scale. This effort is necessary to verify the existence of this type as a real linguistic constellation. Section 3 of this paper presents the results of this endeavor, a portrait of Korean as it relates to the Semantic Typology. Section 4 then presents a view of a language which appears to violate the predictions made by the system developed here, namely Indonesian. It will be shown that the apparent violations can be explained independently and the typology remains intact.
3. High transparency: Korean Korean is presented here to illustrate a language with a remarkably high degree of semantic transparency. It does not completely reach ideal transparency as it was defined in section 2.3., primarily because the range of its grammatical relations is not identical to the various semantic roles. Nonetheless, Korean turns out to be considerably more transparent than German, and consistently so over a wide range of linguistic properties. In order to show this, we will in turn address its basic morphological and lexical properties, the semantic range of its basic grammatical relations, and the nature of movement processes in Korean.
3.1. Basic morphological and lexical properties Korean has a rich set of inflectional morphemes. Verbs carry suffixes for tense, aspect, honorifics, politeness, modals, and "mood", i. e., the status of a sentence as declarative, interrogative, etc. In addition, complementizers are
148
Franz
Müller-Gotama
suffixed to the verb of an embedded clause. Kim (1987: 891) notes that over 400 different suffixes can be attached to a verb base. Nouns are traditionally said to lack grammatical morphology, but there is evidence that the so-called case particles cliticize to the noun. For example, the forms of some case markers are phonologically conditioned by the stem phoneme that precedes them. The Yale system of transliterating Korean, which is used in this study, recognizes this by representing the case markers as suffixes. Table 3. Principal Korean case markers -ka -lui -eykey
Nominative Accusative Dative
-nun -eyse -ey
Topic Ablative Temporal
-lo -ey -tff
Instrumental Locative Genitive
Some inflectional categories familiar from European languages are quite restricted in Korean. The expression of number is optional, for example, and agreement occurs only with honorifics. Nevertheless, the extent of inflection in Korean verbs and nouns is extensive enough that, if the amount of inflectional morphology indeed correlates with the level of transparency, as was assumed by Sapir as well as by Hawkins, Korean should be a highly transparent language. This impression is reinforced by the level of lexical specificity required in Korean as measured against the possible semantic range of objects of the English all-purpose verb make. It turns out that English and German, and as we will see below also Indonesian, are almost equally liberal here, as opposed to Russian, which needs to employ different verbs depending on the semantic field of the object. Korean patterns more like Russian here. Table 4. English
Korean
makejsew a dress makejbake bread makejbrew tea make\weave a rope makejweave a nest makejlaj a road ?makejbuild a house
os-ul kkomay-ta\mantul-ta ppang-ul ka>up-ta¡mantul-ta cha-lul kkulhi-tajmantul-ta tongacwul-ul cca-ta\mantul-ta twungci-lul tul-tajmantul-ta kil-ul noh-tajmantul-ta cip-ul cis-ta\*mantul-ta
Table 4 shows a misleading liberality for Korean, which hides a number of semantic differences that are not made in the English set. The Korean set
Tomarás α semantic typology of language
149
contains only one ungrammatical item, interestingly the same one that is questionable in English and in German. Even that item can be grammatical, however, only when it is used for a child "making" a toy house. When it comes to sewing, both the verb kkomay 'sew' and the verb mantul 'make' are allowed, but the former can only be used if the dress is merely repaired; production of a new dress necessitates mantul. Similarly, mantul was at first judged ungrammatical by two native speakers as a stand-in for cha-lul 'brew'. At last, one informant accepted it, but only with the meaning that one is "making", i. e., growing, tea plants. If we require, then, that the same meaning must be expressed by the two verbs in a given row, then the general verb mantul can substitute for the more specific verb only in four of the seven items tested for, indicating a higher degree of lexical specificity for Korean than for English or German, albeit a lower degree than that observed in Russian above.
3.2. The semantic range of basic grammatical relations If Korean were an ideal transparent language, we would expect the range of its grammatical relations to coincide with the various semantic roles. The Korean noun suffixes could then be interpreted as semantic case role markers. However, this is not a correct characterization of the Korean facts, although this section will show that the case markers come close to the semantic constellations due to the comparatively narrow semantic range permitted by the grammatical relations they signal. Prime evidence that the grammatical relations of Korean do not completely correspond to individual semantic roles comes from the existence of the passive. (2) a.
b.
Jinhee-ka mwun-ul ye-n-ta. Ν door-ACC open-PRES-DECL 'Jinhee opened the door.' Na-uy guitar-nun Jinhee-ej euhajse cwul hana-ga I-GEN TOP LOC by string one-N pwule-ci-ess-ta. break-PASS-PAST-DECL 'As for my guitar, a string was broken by Jinhee.'
Notice that the patient cwul hana-ka 'one string' in (2 b) carries the same suffix -ka which marks the agent Jinhee in (2 a). The passive, therefore, exists in Korean. With it, the subject relation, which is marked by a nominative suffix,
150
Fran\ Müller-Gotama
absorbs all those patient noun phrases that are promoted in the passivization process. Comrie (1986 b) explains the tight restrictions on the passive in Russian by arguing that grammatical relation-changing processes involve the syntactic reorganization of a basic structure and, hence, constitute a distortion of the semantic constellation. Such processes should, consequently, be lacking or strongly restricted in languages with high semantic transparency. This is exactly what we find in Korean. Although passive constructions undoubtedly occur in this language, Kim (1987: 893) remarks that they are exceedingly rare, occurring much less frequently than in English and in Japanese. Indeed, our informant had to go through some hesitation and false starts when he tried to come up with the correct form of several passive sentences. His performance problem may well be symptomatic for the status of the passive in his native language. 3.2.1. Subjects Korean subjects are marked with the suffix -ka for nouns ending in vowels or -i for nouns ending in consonants. Much more typically than their English counterparts, Korean subjects are agents. Noun phrases with other semantic roles typically do not become subjects, especially not in transitive active sentences. For completeness sake, it must again be pointed out, however, that passivization is a rarely applied but possible strategy for promoting underlying objects, i. e., typically patient noun phrases, to subjecthood. Moreover, the sole argument of an intransitive will regularly be construed as a subject; in such a situation, a somewhat wider semantic range can reasonably be expected than in transitive sentences, in which the more agent-like noun phrase can be construed as subject and the more patient-like noun phrase as object. The non-agentive subject in the intransitive sentence (3) is therefore not surprising. (3)
Kiyong-i chwuk-kess-ta. Ν die-FUT-DECL 'Kiyong will die.'
Sentence (2 a) has already shown an example of an agentive subject. Such examples are the core cases and could be multiplied at length. Following Fillmore's hierarchy of case accessibility (Fillmore 1968, 1971), experiencer noun phrases unproblematically become subjects in the absence of an agent noun phrase. Korean here contrasts with German, which often treats such experiencers as dative subjects.
Towards a semantic typology of language
(4) a.
b.
151
Yongin-i yi chayk-ul coaha-n-ta. Ν this book-ACC like-PRES-DECL 'Yongjin likes this book.' Junmie-ka chwup-ta. Ν cold-DECL 'Junmie is cold.'
Instrumentals are coded as such and do not become subjects. (5) a.
b.
(6)
* Yi yelsoy-ka yi mwun-ul yel-swuiss-ta. this key-N this door-ACC open-can-DECL 'This key can open this door.' Yi yelsoy-lo-nun yi mwun-ul yel-swuiss-ta. this key-INSTR-TOP this door-ACC open-can-DECL 'As for this key, it can open this door.' II nyen ceney-nun il won-ulo twu kay na sey one year before-TOP one won-INSTR two CL or three kay-uy pin-ul sa-l-swuiss-ess-ta. CL-GEN pin-ACC buy-can-PAST-DECL Ά year ago, one won could buy two or three pins.'
Despite the fact that sentence (6) has no overt subject, any attempt to make the instrumental won-ulo 'with one won' into a subject parallel to the structure of the English gloss makes this sentence totally ungrammatical. Interestingly, however, there are examples where this constraint appears to be violated. (7) a.
b.
Yelsoy-ka mwun-ul yel-n-ta. key-N door-ACC open-PRES-DECL 'The key opens the door.' Mangchi-ka mwun-ul kkey-ess-ta. hammer-N door-ACC break-PAST-DECL Ά hammer broke the door.'
Example (7 b) is adapted from Yang (1972). Yang emphasÌ2es that this sentence is only grammatical if mangcbi 'hammer' is interpreted as a personification. One informant probably characterized such sentences best when he stated that they can only describe fairytale worlds where keys and hammers are alive and conscious of their actions. As such, they do not constitute counterexamples to our claim that instrumentais can not become subjects in Korean. Similarly, locatives do not become subjects.
152
Franse Miiller-Gotama
(8)
Na-uy guitar-nun cwul hana-ka pwule-ci-ess-ta. I-GEN TOP string one-N break-PASS-PAST-DECL 'As for my guitar, a string was broken.' = 'My guitar broke a string.'
Replacing the topic marker -nun on guitar with the nominative marker -i (and the nominative marker -ka on hana with the accusative marker -lut) results in the ungrammaticality of the sentence, as we would predict by now. This sentence also shows the inability of patient noun phrases to become subjects in Korean. The English verb break is a so-called "ergative" verb, i. e., it is a transitive verb which can also be used intransitively, with the underlying object becoming the surface subject. In Korean, such a process does not exist. The only way for the patient noun phrase cwul hana 'one string' to become the subject is be applying passivization. Sentence (8) is therefore an example of the usefulness of passivization even in Korean. Some noun phrases which Case Grammar would classify as locatives can appear as subjects in Korean, as (9) shows. (9)
Kay-nun yi hotel-i kumchiha-n-ta. dog-TOP this Ν forbid-PRES-DECL 'As for dogs, this hotel forbids them.'
However, one must question whether such subjects are truly locatives. Rather, it seems that the hotel in (9) is understood as an organization, hence as an agent. This is also possible and occurs commonly in German, as (10) demonstrates, notwithstanding the ungrammaticality of the literal equivalent of (10 a). (10) a. b. c.
* Dieses Hotel verbietet Hunde. this hotel forbids dogs Dieses Geschäft verkauft Zahnpasta. this shop sells toothpaste Die Regierung hat Hunde verboten. the government has dogs forbidden 'The government has forbidden dogs.'
Even for the apparently exceptional sentence (10 a), it can be shown that the ungrammatically is less the result of the semantic role of the subject than of the choice of object. In addition, the sentence improves quite considerably if the perfect is used in place of the present tense, for reasons that still need to be investigated.
Towards a semantic typology of language
(11)
?Dieses this
Hotel hat Hunde hotel has dogs
153
verboten. forbidden
Sentence (11) is anomalous primarily because it implies that the hotel has outlawed dogs in the society at large; the native speaker doubts whether a hotel (as opposed to the government in (10 c), which may well have been concerned about the spread of rabies, etc.) might reasonably be assumed to have such far-reaching powers. In keeping with this line of reasoning, the sentence becomes completely acceptable in German if the right tense is chosen and the choice of the object is properly constrained. (12)
Dieses Hotel hat es verboten, Hunde mitzubringen. this hotel has it forbidden dogs to-bring-along 'This hotel forbids it to bring along dogs.'
Far from being anomalous, example (9) therefore illustrates once more the fact that subjecthood in Korean is constrained to agents and some experiencers. Yang (1972: 8) points out that no other semantic roles can become subjects in Korean. He specifically cites such well known examples as English The food eats (well). (13) presents a further example of an unusual English subject. Our informant made it clear that the phrase han chang-i 'one chapter' can only be the subject of a passive sentence, cf. the verb morphology. Its active equivalent must either have no overt subject, or a suitable agent (the author of the book, its publisher, etc.) must be chosen as subject. No other subject is possible. Notice that the Korean equivalent of the English subject the latest edition only has the role of topic in this sentence. (13)
Y i chajk-uj choykun ho-nun han chang-i this book-GEN latest edition-TOP one chapter-N cwuka-toy-ess-ta. add-PASS-PAST-DECL 'As for the latest edition of this book, a chapter was added.' — 'The latest edition of this book has added a chapter.'
This subsection has shown that only a narrow range of semantic roles have access to subjecthood in Korean. This result is in keeping with the consistently transparent nature of Korean, as it is now taking shape with our observations that Korean has an extensive set of grammatical morphology, relatively high lexical specificity, overt marking of grammatical relations, and restricted use of the grammatical relation-changing process of passivization. It appears that a major reason for the widely expanded set of semantic roles which can become subjects in English is their need to function as topics.
154
Franζ Miiller-Gotama
When English word order became fixed in the SVO frame, English was forced to develop other mechanisms which could take over the pragmatic functions that were previously performed by simple word order rearrangements. Two developments in the English language were particularly important. Both of them had as a major consequence the reduction of semantic transparency in English: first, there was the extended applicability of the passive, which now allowed the promotion of the old class of dative objects; secondly, the extension of the range of noun phrases that are eligible to become subjects in underived environments allowed more and more semantic roles to function as subjects. According to Kirkwood (1969), expanding the set of subjects allowed the English language to retain a good deal of the pragmatic topic-comment ordering despite the now grammatically fixed word order. Faced with the same problem of ensuring the expression of topic continuity through foregrounding, Korean has opted for a strategy that is diametrically opposed to that of English. Instead of fixing word order, Korean allows free scrambling of the major maximal constituents of a sentence except for the verb, which must remain in final position (cf. section 3.3.), making it possible for any constituent to be foregrounded. In addition, the topic marker -nun can attach to any nominal expression. (14) a.
b.
c.
d.
Na-nun yi
chayk-ul
coaha-n-ta.
I - T O P this book-ACC like-PRES-DECL Ί like this book.' II nyen ceney-nun Kiyong-i Yunsoo-lul ttayli-ess-ta. one year before-TOP Ν ACC hit-PAST-DECL 'As for one year ago, Kiyong hit Yunsoo.' Yi yelsoy-lo-nun yi mwun-ul yel-swuiss-ta. this key-LOC-TOP this door-ACC open-can-DECL A s for with this key, it can open this door.' Kay-nun yi hotel-i kumchiha-n-ta. dog-TOP this Ν forbid-PRES-DECL A s for dogs, this hotel forbids them.'
Even multiple topics are allowed. (15)
Mia-nun
Pusan-ey-nun
ka-ess-eyo.
TOP LOC-TOP go-PAST-POL 'Speaking of Mia, Pusan she went to (but not to...).' (Chang 1983: 222)
Towards a semantic typology of language
155
As a final strategy of Korean, non-agentive subjects can be construed as agent-like by making them the subject of a causative construction since causers prototypically are intentionally acting humans, and effects are prototypically the result of an action. (16) Cbongal-i Jongjin-uy taly-lul tacy-key ha-ess-ta. bullet-N GEN leg-ACC wound-COMP do-PAST-DECL 'The bullet wounded Jongjin's leg.' = lit.: 'The bullet caused to wound Jongjin's leg.' The combination of free scrambling, causativization, and overt topic marking ensures that virtually any nominal expression can be foregrounded or topicalized. Therefore, the corresponding strategies which perform these functions in English are much less needed in Korean. Hence, they are severely restricted, like the passive, or virtually non-existent, like the collapsing of a large number of semantic roles onto the grammatical function of subjecthood. 3.2.2. Objects Korean distinguishes between dative and accusative objects. The former are marked by -eykey, the latter by -ul. Only accusative objects can undergo passivization, cf. above for examples. (17 b) is ruled out as it involves the promotion of an indirect object (sentence (17 a) is adapted from Yang 1972). (17) a.
b·
Yunsoo-ka ai-eykey Bible-ul kaluchi-ess-ta. Ν child-DAT ACC teach-PAST-DECL 'Yunsoo taught the Bible to a child.' *ai-ka enehak-ul Jeongdal-ey euhayse child-N linguistics-ACC LOC by kaluchi-eci-ess-ta. teach-PAS S-PAST-DECL Ά child was taught linguistics by Jeongdal.'
Whereas English has lost the case marking distinction between dative and accusative and allows former dative objects to become promoted to subjecthood by means of passivization, Korean strictly differentiates the two categories. As was the case with subjects, Korean objects preserve the respective underlying semantic distinctions more closely than is the case in a grammaticizing language like English.
156
Fran^
Müller-Gotama
3.3.
Word order effects
3.3.1. Scrambling Korean has a highly flexible word order. As has frequently been noted in studies of Korean (for example Kim 1987), all logically possible permutations of the major constituents result in a grammatical sentence as long as the verb remains in clause-final position. Therefore, sentences (18 a) through (18 f) are all grammatical. (18) a.
b. c. d. e. f.
Kiyong-i Mia-eykey chayk-ul cu-ess-ta. Ν DAT ACC give-past-DECL 'Kiyong gave Mia a book.' Kiyong-i chayk-ul Mia-eykey cu-ess-ta. Mia-eykey Kiyong-i chayk-ul cu-ess-ta. Mia-eykey chayk-ul Kiyong-i cu-ess-ta. Chayk-ul Kiyong-i Mia-eykey cu-ess-ta. Chayk-ul Mia-eykey Kiyong-i cu-ess-ta.
Sentences (18 g) and (18h), however, are not grammatical because material has been scrambled to the right of the verb. (18) g. h.
* Kiyong-i Mia-eykey cu-ess-ta * Mia-eykey chayk-ul cu-ess-ta
chayk-ul. Kiyong-i.
The verb-finality constraint ensures that rightward clause boundaries can be clearly identified. It is especially important when there are embedded clauses. With this constraint, the rightward boundary of all clause level constituents is marked in the surface structure. In nominal expressions, the boundary is demarcated by the case marker or the postposition accompanying a noun, while clauses signal their rightward edge by the position of the verb. The identification of the role of the clause is further enhanced by the fact that embedded verbs typically have a complementizer among their affixes. By comparison, leftward clause boundaries are left curiously unsignalled. Their locations must be inferred by the hearer. (19)
Nay-ka [Mia- ka yi chayk-ul I-N Ν this book-ACC coaha-ki-lul] kitayha-ess-ta. like-COMP-ACC expect-PAST-DECL Ί expected that Mia likes this book.'
The permutations that occur because of scrambling do not lead to a reduction in semantic transparency, unlike other types of movement, because
Towards a semantic typology of language
157
the grammatical roles of the constituents that are moved around remain unequivocally identified by the case marker and/or the postposition accompanying each nominal expression. Constituents can only be scrambled as a unified whole, and nothing can be scrambled into, or out of, another constituent. The violation of this constraint is the cause of the ungrammatically of (20 b), where the matrix subject nu-ka 'who' has been scrambled into the center-embedded clause. (20) a.
Nu-ka [Yunsoo-ka Mia-eykey chayk-ul who-N Ν DAT book-ACC
cu-ess-ta]-ko
malha-ess-ni? give-PAST-DECL-COMP say-PAST-Q 'Who said that Yunsoo gave Mia a book?' t>- *[ Yunsoo-ka nu-ka Mia-eykey chayk-ul Ν who-N DAT book-ACC cu-ess-ta ]-go malha-ess-ni? give-PAST-DECL-COMP say-PAST-Q Of course, the statement that clause-level constituents must only be moved as a whole and without violating the integrity of other constituents entails the claim that neither Adposition-stranding nor extractions exist in Korean. The stranding of a postposition implies that it has somehow been removed from the remainder of the postpositional phrase in violation of the requirement that phrases can only be moved as a whole. Extractions, on the other hand, move an element out of the integrity of the receiving clause. Indeed, none of these offending movement processes exist in Korean, showing once more how the task of maintaining transparency lends a common directionality to a large number of seemingly unrelated structural facts about Korean. If case markers or postpositions are deleted, as can optionally be done in Korean, the language resorts to its basic SOV word order in order to identify the grammatical function of the various constituents: The first noun phrase in such a sentence will be interpreted as the subject and the remaining noun phrases as objects. Interestingly, the respective order of dative and accusative objects is not fixed and must be recovered from the pragmatic context (A. Kim 1985). English makes a clear distinction between these on the basis of word order. If a sentence like John gave Mary Bill can plausibly be uttered in some context, it is clear that the recipient is Mary and the thing received is Bill.
158
Fran% Müller-Gotama
3.3.2. Raising There is no evidence of subject to subject (S — S) raising in Korean. The equivalent construction to English raising works exactly as we would expect for all embeddings: Kat 'seem' appears as the matrix verb in sentence-final position; there is no overt expletive; and the embedded clause has the same shape as its simple counterpart, with the exception of the morphology of its verb, which is nominalized and must select from a different set of tenseaspect markers than a main verb. This is shown in the contrast between sentences (21 a) and (21 b). (21) a.
b.
Jeongdal-i
yi chayk-ul Ν this book-ACC 'Jeongdal likes this book.' [Jeongdal-i yi chayk-ul Ν this book-ACC 'Jeongdal seems to like this
coaha-n-ta. like-PRES-DECL coaha-nun kos] kat-ta. like-COMP NOM seem-DECL book.'
Sentence (21 c) illustrates that scrambling applies freely within the embedded clause, including scrambling of the subject. c.
[Yi chayk-ul Jeongdal-i coaha-nun kesJ kat-ta. this book-ACC Ν like-COMP NOM seem-DECL
This last fact alone indicates that the subject has not been raised, given the constraint formulated above that scrambling into any other constituent is not permitted. Add to this the fact demonstrated by (21 d) that this subject cannot be scrambled within the matrix clause, and the argument against a raising analysis is complete: d.
*[Yi chayk-ul coaha-nun kes] Jeongdal-i kat-ta. this book-ACC like-COMP NOM Ν seem-DECL
The productive use of constructions like those in (22) suggests that objectto-subject (O —S) raising may exist in Korean. Some evidence for a raising analysis comes from the fact that the underlying object can be scrambled inside the matrix clause, as shown in (22 b). (22) a.
b.
Enehak-i kongpwuha-ki-ka swuip-ta. linguistics-N study-COMP-N easy-DECL 'Linguistics is easy to study.' Kongpwuha-ki-ka enehak-i swuip-ta.
With regard to its effect on the position of Korean in the Semantic Typology, the apparent availibility of O —S raising constitutes a problem. In
Towards a semantic typology of language
159
the properties discussed thus far, Korean has exhibited an exceptional overall consistency. Raising, however, is a process which reduces semantic transparency considerably, if the reasoning behind the movement hierarchy presented in section 2.3. is correct. It must, therefore, be asked, whether an O —S raising analysis is actually the correct derivation of sentences like (22), or whether the possibility of scrambling within the matrix is not sufficient to prove that raising has occurred. In fact, there is evidence that O —S raising is not the correct analysis of (22). Comrie and Matthews (1990) have recently pointed out that similar constructions in Serbo-Croatian are not to be interpreted as O — S raising since the relevant examples in that language are only grammatical if the corresponding string Ν Ρ (Nominative) be Adj is also grammatical. Korean presents a parallel problem for the O —S raising analysis: that analysis is forced to claim that only underlying objects can be "raised" into the subject position of the matrix. However, sentences like (23) demonstrate that this is not true in Korean. Significantly, the locative phrase jipen kumyoil 'this Friday', which logically belongs with the embedded predicate, can appear as the matrix subject here, despite the fact that this phrase is not a direct object of the lower predicate. (23)
Yipen kumyoil-i kongpwuha-ki swui-wul-kesi-ta. this Friday-N study-COMP easy-FUT-DECL 'This Friday is easy for studying.'
While (23) is not the preferred structure in such a situation, it is perfectly grammatical when studying is usually difficult on Fridays, say, because of the noise level in the students' dormitory; (23) could be uttered to refer to a Friday when the noise problem won't occur. Summing up it can be said that Korean proved to be a highly transparent language throughout a wide range of its grammatical properties. The most interesting part about this consistency is the discovery of how various seemingly unrelated properties of the language interact to preserve the advantages of a consistent location on the continuum close to its transparent end.
4. A problematic case: Indonesian Indonesian is presented as a detailed study in this paper for three reasons: First of all, some superficially observable properties of this language, such as its inflectional morphology, word order properties, etc., make it a possible
160
Fran\ Müller-Gotama
contender as a member of the grammaticizing type, like English. While several languages with close affinities to the transparent end of the Semantic Typology have been identified so far, the grammaticizing type is represented only by English as of now. Secondly, the treatment of Indonesian from the typological perspective developed here yields a unifying view of a number of grammatical properties of this language, some of which have been recognized individually in the literature and some of which have not previously been noted. The Semantic Typology provides a rationale for the cooccurrence of these properties and, thus, can be said to constitute an explanation for their coexistence in Indonesian. Thirdly, the detailed study of Indonesian reveals a clear split in the alignment of its properties on the typological continuum, with some features patterning with English, as we had expected, but others positioned close to the transparent end of the scale. This split could represent a crucial violation of the framework, but the systematicity of the apparent violations leads to a solution which yields further insight into the nature of the Semantic Typology.
4.1. Basic morphological and lexical properties Even a look at some quite superficial lexical and morphological properties of Indonesian reveals the typical grammaticizing patterning. For one thing, it lacks grammatical morphology almost completely. The grammatical affixes that do occur are limited to the set of verb affixes listed in figure 4. ber-
+
BASE
+
BASE
+ -kan
Figure 4. Indonesian verb affixes
These affixes function grammatically as indicators of the argument structure of a sentence, i. e., they serve strictly syntactic purposes. Thus, ber- indicates intransitivity of the verb, meng- its transitivity, and di- the passive. The suffixes -i and -kan serve to create locative and benefactive object positions, respectively. Because of its affixes, an Indonesian verb is regularly marked for its status with respect to transitivity. This fact, besides being transparency increasing in itself, has further transparency increasing effects for the range of subjects in Indonesian, as will be shown below.
Towards a semantic typology of language
161
Even with respect to the marking of the transitivity of a verb and the respective roles of its arguments, the morphological system of Indonesian is deficient, however, since it is characterized by many lexical exceptions. For instance, while meng- verbs are overwhelmingly transitive, a minority of them are not, e. g., me-laut 'sail' and meny-alak 'bark'. There are even a few transitive verbs with the prefix ber-·, ber-buah-kan 'result in' is a transitive verb, notwithstanding the prefix ber-·, the regular transitive form should be mem-buahkan, of course. The latter form exists but has a different meaning, namely 'produce, yield'. Such exceptions are widespread enough to make a clear distinction into inflectional and derivational affixes doubtful. Most affixes serve both functions, and only the prefix di- functions consistently as an inflectional affix, marking passivization. Even regular applications of an affix may still not identify the roles of all noun phrases because the suffixes -i and -kan are multifunctional: -kan regularly indicates the causative in addition to marking the benefactive (mati 'die' —>· me-mati-kan 'kill'; hidup 'live' —> menghidup-kan 'bring to life, create', etc.), while -i marks iterative verbs (me-mukul 'hit' —> me-mukul-i 'hit repeatedly'; meng-gigit 'bite' —• meng-gigit-i 'bite repeatedly', etc.). Other, more semantically based morphological distinctions, such as tense, number, or gender, do not exist in Indonesian. In as far as the pure "richness" of the grammatical morphology of a language determines its location in the typology, Indonesian should, therefore, belong to the English type. Furthermore, Indonesian patterns unequivocally with English with regard to the degree of required lexical specificity. Table 5. English
Russian
Indonesian
makejsew a dress makejbake bread makejbreni tea makefweave a rope makejiveave a nest makejlaj a road ?makejbuild a house
sit' plat'e peC xleb varit' caj vit' verevku vit' gne\do prolo\it' dorogu
buat\jahit baju buatjpanggang roti buatjmasak teb buatjtenm tali buatjbangun sarang buatjbangun jalan ??buat¡bangun rurnah
Stroit'
dom
When compared with the same expressions in Korean and German, a clear hierarchy of required lexical specificity emerges from the data of these five languages. At one and of this hierarchy, Russian requires a different verb for almost all of these items, while Korean requires four distinctions. German
162
Fran£
Müller-Gotama
allows the general verb machen 'make' in the upper five rows and marginally in row 6, whereas English permits its general make with all of the listed nouns, albeit only marginally in the lowest row of Table 5. The Indonesian data closely mirror those of English with respect to lexical specificity although the final entry more strongly approaches unacceptability in this language than in English.
4.2. The semantic diversity of basic grammatical relations Like English, Indonesian is characterized by considerable semantic diversity of both objects and subjects; however, underived subjects display relatively more semantic specificity. The class of direct objects in Indonesian actually covers an even wider range than in English. Like that of English, it includes not only strongly patient-like objects, but also other roles which are likely to show up as nonaccusative objects in more transparent languages, as in (25) and (26). This fact is evident from the possibility of passivization. The German equivalents are given to illustrate this point. (24) a.
b.
(25) a.
b.
(26) a.
Alt
me-mukul orang itu. TA-hit man that 'Ali hit that man.' 'Ali schlug diesen (ACC) Mann.' Orang itu di-pukul oleh Ali. man that PASS-hit by 'That man was hit by Ali.' 'Dieser (Ν) Mann wurde von Ali geschlagen.' Ali me-nolong orang itu. TA-help man that 'Ali helped that man.' 'Ali half diesem (DAT) Mann.' Orang itu di-tolong oleh Ali. man that PASS-help by 'That man was helped by Ali.' *'Dieser (N) Mann wurde von Ali geholfen.' Dia he 'He 'Er
me-merlu-kan bantu-an-mu. TA-need-BEN help-NOM-2sg needs your help.' bedarf deiner (GEN) Hilfe.'
Towards a semantic typology of language
b.
163
Bantu-an-mu di-perlu-kan-nya. help-NOM-2sg PASS-need-BEN-3sg 'Your help was needed by him.' *'Deine Hilfe (Ν) wurde von ihm bedurft.'
The Indonesian direct objects in (24), (25), and (26), like those of English, correspond to accusative, dative, and genitive objects in German, respectively. An interesting example is (26), since the verb perlu 'need' requires the benefactive suffix -kan. In one of its productive functions, besides forming causatives, this suffix alternates with the preposition untuk 'for', creating a direct object from a prepositional phrase (Chung 1976). Similarly, the suffix -i alternates with the preposition ke 'to', leading to the creation of locative objects. Objects created by either of these processes undergo passivization freely, as shown in (27) and (28), although their equivalents in English must frequently appear in a prepositional phrase and resist the application of the passive. (27) a.
b.
c. d.
(28) a.
b.
c.
Hendra
mem-beli mobil itu untuk Yoyo. TA-buy car that for 'Hendra bought that car for Yoyo.' Hendra mem-beli-kan Yoyo mobil itu. TA-buy-BEN car that 'Hendra bought Yoyo that car.' * Hendra mem-beli Yoyo mobil itu. Yoyo di-beli-kan mobil oleh Hendra. PASS-buy-BEN car by 'Yoyo was bought a car by Hendra.' Hendra datang ke rumah saya. come to house I 'Hendra came to my house.' Hendra men-datang-i rumah saya. TA-come-LOC house I 'Hendra came to my house.' Rumah saya di-datang-i oleh Hendra. house I PASS-come-LOC by *'My house was come to by Hendra.'
The class of direct objects in Indonesian is, therefore, expanded relative to the transparent state of German as well as relative to English. Interestingly, we are again dealing with a relationship of proper subsets between these
164
Fran^ Miilkr-Gotama
languages, where the class of direct objects in German is properly included in that of English, which is in turn included in that of Indonesian. The semantic generality found with objects in Indonesian largely carries over to subjects as well. However, the range of subjects turns out to be somewhat more restrictive in this language than in English because of syntactic reasons (cf. below). (29) a.
b.
(30) a.
b.
(31) a.
b.
(32) a.
b.
(33)
(34)
(35)
Pak Topo men-jual mobil-nya. [AGENT] Mr. TA-sell car-3sg 'Mr. Topo sold his car.' Jahja mem-buka pintu itu. John TA-open door that 'John opens that door.' Kunci ini mem-buka pintu itu. [INSTRUMENTAL] key this TA-open door that 'This key opens that door.' Waktu itu se-puluh rupiah bisa mem-beli se-buah nangka. time that one-ten rupiah can TA-buy one-fruit nangka 'At that time, ten rupiah could buy a Nangka fruit.' Kamar ini bau. [LOCATIVE] room this smell 'This room smells.' Hotel ini me-larang tamu-tamu untuk mem-bawa anjing. hotel this TA-forbid guest-guest for TA-carry dog 'This hotel forbids its guests to bring dogs.' Saya men-jadi dingin. [EXPERIENCER] I TA-become cold 'I'm getting cold.' Saya senang buku itu. I like book that Ί like that book.' Besok akan jadi dingin sekali. [TEMPORAL] tomorrow will become cold very 'Tomorrow will be very cold.' Edisi terachir dari buku ini tambah dua pasah. edition last from book this increase two chapter 'The last edition of this book has increased by two chapters.' Mobil ini me-muat lima orang. car this TA-load five people 'This car seats five people.'
Towards a semantic typology of language
165
Indonesian naturally allows all of the semantic roles illustrated in examples (29) through (34), and even in (35), to access subjecthood in these sentences, including those where a language with greater surface semantic transparency like German would require or prefer to express the semantic relations in an adpositional phrase. So far, Indonesian behaves like English once again. However, it is also evident from examples like (36) that not all English subjects match an Indonesian subject. (36)
*Kemah ini tidur empat tent this sleep four 'This tent sleeps four.'
orang. people
Fortunately for the present argument, the exceptionality of (36) finds an explanation on independent syntactic grounds: Verbs in Indonesian, as has been pointed out above, are regularly marked for their status with respect to their syntactic argument structure, and tidur 'sleep' is unmistakably an intransitive verb. Whereas it is possible for a language like English, which lacks such obligatory marking, to easily change the transitivity of a verb, the same cannot be done in Indonesian. In fact, the collocation 'sleep somebody', Indonesian men-idur-kan, can only be interpreted as a causative in this language. (37)
Kemah ini men-idur-kan lima orang. tent this TA-sleep-CAUSE five people 'This tent put four people to sleep.'
Hence, (36) is ruled out because of the intransitivity of the verb form used. For the same reason, the so-called "ergative" verbs of English, i.e., verbs like open, which occur both as transitives with agentive subjects and as intransitives with objective subjects, must be expressed by different forms in Indonesian. (38) a.
b.
c.
Hasan
mem-buha pintu itu. TA-open door that 'Hassan opened that door.' Pintu itu ter-buka. door that open 'The door opened.' *Pintu itu mem-buka. door that TA-open
The verb mem-buka is unequivocally transitive, hence the ungrammaticality of (38 c). Notice that non-agentive subjects do occur with the prefix meng- as
166
Fran\
Miiller-Gotama
(30 a), (30 b), (31 b), (32 a), and (35) have shown. So the ungrammaticality of (38 c) really must be attributed to the intransivity of the sentence. The assignment of semantic roles to specific grammatical relations, unlike some other grammatical properties, necessarily involves some distortion of semantic subtleties since, first of all, any given semantic role incorporates an infinite set of different real-world relationships and since, secondly, it is not always clear to which semantic role a given real-world relationship belongs. For example, the role "agent" comprises not only the subjects in (39), but probably those in (40) as well; even those in (41) can be interpreted as metaphorical extensions of agency since, in a very real sense, the flood in (41) is the entity that performed the action of lifting the tree. Certainly floods are far removed from prototypical agency: they are neither human nor animate and lack any concept of intentionality. However, nobody could seriously claim much animateness or intentional control for the behavior of microbes and the like, and yet we have no difficulty conceiving of statements where such barely animate microbes function as agents of a verb like attack. Even inanimate entities can fulfill this role with equal ease given the right circumstances, as when a meat-eating plant is said to devour its "prey". Conversely, a human acting while under hypnosis does not thereby become "instrumental" of the psychologist who controls him. In other words, the flood can be held to at least combine elements of "cause" and "agent". That it is subsumed under the agentive pattern, allowing it to become a subject even in a much more restrictive language like German (cf., Die Flut hat das Land verwüstet 'The flood devastated the land'), is therefore not surprising. (39) a. b. (40) a. b. (41) a. b.
John hit Bill. Mike kissed Mary. Sam tried to sleep. The robot opened the door. The flood water lifted the tree. The key opened the door.
The clear cases of (39) involve intentionally acting humans; as such, they illustrate prototypical agency. Certainly, the relationship expressed in (40 a) is quite different from those in (39) since trying is obviously not as typically a kind of action as hitting or kissing are. In (40 b), the deficiency lies with the presumed agent, the robot, at least if intentionality is an important aspect of the prototype; nonetheless, the robot clearly must be an agent here. The subject in (41 b) would normally be considered an instrumental, with the key serving as a tool of a human agent for opening the door; but in a very real sense, the key can also be thought of as the one thing/agent that performs
Towards a semantic typology of language
167
the action of opening the door. At the present state of technological development, one only needs to imagine a pre-programmed motor-driven key to drive home the affinity of the key to both agency and instrument. Two conclusions follow from this discussion: First, all languages will have considerable semantic distortion here since the semantic roles, of which there are, presumably, only a limited number, are already abstracted away from the relationships that obtain in the "real world" (or any possible world). For instance, agency with respect to hitting is quite different, as a real-world relationship, from agency with respect to kissing, although linguistically, the same semantic role "agent" occurs in the subject of either verb. In languages where the range of the grammatical relations is not in a one-to-one relationship with the semantic roles, those already abstracted semantic roles must in turn be mapped onto the further set of grammatical relations. Significantly, the latter situation seems to be prevalent in the world's languages. Secondly, the affinity between different semantic roles that has been demonstrated for agency and instrumental may be a clue as to how languages can go about extending or restricting the set of permitted roles for any given grammatical relationship. They can increase grammaticization by reducing the number of grammatical relations, as in the collapsing of dative and accusative objects in the history of English, or by extending the range of admissible semantic roles from core relations to relations that are intermediate between two different prototypes until all relations covered by the latter prototype become admissible. Semantic transparency can be increased by providing a larger number of oblique cases and/or prepositions, or by reducing the range of semantic roles admissible to a grammatical relation.
4.3. Movement and deletion processes The assumptions of this study about the semantic component of language comprise three types of requirements, as was stated in section 1: semantic structures must be unambiguous, arguments must be ordered with their predicates, and all surface-removed material must be present. Movement out of a containing clause and deletion rules have in common that they can lead to distortions in all three of these respects. As a result, Indonesian should exhibit considerable freedom in all of these aspects if it is to be unequivocally analyzed as a highly grammaticizing language. Moreover, the permitted movement and deletion rules in this language should be systematically related to those of English if a strict version of the scalar model of the typology is
168
Franz Miiller-Gotama
correct. This section examines in how far English and Indonesian are indeed related in this way. We will in turn examine deletions, raising structures, and extractions out of an S-bar. 4.3.1. Deletions Though not involving movement, the deletion of a noun phrase under identity with a second noun phrase shares with raising and extractions the property that it creates a gap. Such deletions become problematic if the two noun phrases have different case marking; English, therefore, allows them more freely than German, having collapsed the accusative, dative, and genitive objects of older stages of the language into a single direct object category. Now, we have already pointed out that the class of Indonesian objects is at least defined as widely as that in English and that additional objects can be created by suffixation on the verb. We would, therefore, expect that deletion applies freely among objects. This is indeed born out by the data: (42)
Orang itu mem-bangun dan men-datang-i rumah saya. man that TA-build and TA-come-LOC house I 'That man built and came to my house.' Perusahan ini mem-bangun dan perusahan itu men-jual rumah kami. company this TA-build and company that TA-sell house we 'This company built and that company sold our house.'
(43)
4.3.2. Raising Given the overall thrust of the argument in the previous sections, it may come as somewhat of a surprise that raising is a rather restricted phenomenon in Indonesian. For one thing, an investigation of the verbs typically associated with subject-to-subject (S — S) raising appears to indicate that this language does not make use of this process at all (but compare the discussion of Indonesian "tough" constructions below). To express the standard examples with seem, the language resorts to sentential adverbs like sepertinya 'like, as, apparently', rupanya 'apparently', or kelihatannya 'it looks like, apparently'. (44) a.
b. c.
Sepertinya apparently 'Apparently Rupanya dia Kelihatannya
dia senang gadis itu. he like girl that he likes that girl.' = 'He seems to like that girl.' senang gadis itu. dia senang gadis itu.
Towards a semantic typology of language
169
Other verbs which serve as triggers for S —S raising in English require nominal objects (cf. (45 b)) instead of a verb complement as in the ungrammatical sentence (45 a). (45) a.
b.
* Dia me-nerus-kan meng-kuliah-kan tentang Eropa. he TA-continue-CAUSE TA-lecture-CAUSE about Europe 'He continued to lecture about Europe.' Dia me-nerus-kan kuliah-nya dalam ilmu pasti. he TA-continue-CAUSE lecture-3sg inside science precise 'He continued his lecture about mathematics.'
Subject-to-object (S —O) raising is quite productive in Indonesian with a wide range of "control" verbs, but is decidedly bookish. The corresponding constructions with the subordinating complementizers supaya 'so that' or bahwa 'that' are much more natural and preferred in anything but the most formal contexts. (46) a.
b.
Saya meng-hendak-i Budiman untuk mem-baca buku ini. I TA-want-LOC for TA-read book this Ί want Budiman to read this book.' Saya meng-hendak-i supaya Budiman mem-baca buku ini. I TA-want-LOC so that TA-read book this *'I want that Budiman reads this book.'
The contrast with the construction in (47) shows that (46 a) really involves the raising of the embedded subject out of its clause by clarifying that the complementizer untuk 'for' appears in a clause-initial complementizer position, ruling out an analysis which would claim that untuk 'for' generally occupies a clause-second position. (47)
Penting sekali untuk Sandra pergi sekarang. important very for go now 'It is very important for Sandra to leave now.'
There are some constructions in Indonesian which look as if they are instances of O—S raising. The relevant examples are presented in (48). (48) a.
b.
Biologi susah untuk di-pelajar-i.5 biology hard for PASS-learn-LOC 'Biology is hard to study.' Danny mudah di-senang-kan. easy PASS-like-BEN 'Danny is easy to please.'
170
Fran%
c.
Miiller-Gotama
Roti ini baik untuk di-potong dengan pisau. bread this good for PASS-cut with knife 'This bread is good/easy to cut with a knife.'
However, these sentences require obligatory passivization in the dependent clause and, hence, should be considered examples of S — S raising. Even then their occurrence is restricted to a small set of trigger predicates in addition to the fact, observed above, that no other instances of S —S raising exist in Indonesian. The extent of the applicability of these constructions, therefore, approaches the tight restrictions of German, rather than the more liberal range found in English. In particular, their grammaticality begins to deteriorate rapidly as one tries to introduce other triggering predicates in (49). (49) a. Buku ini
b.
susah mudah ìmem-bosan-kan *men-arik
untuk di-baca.
book this for PASS-read 'This book is hard/easy/boring/interesting to read.' *Computer ini mabal untuk di-beli. computer this expensive for PASS-buy 'This computer is expensive to buy.'
All other things being equal, the most straightforward analysis is, therefore, that both S —O raising and S —S raising exist in Indonesian. Since the raised noun phrases cannot be interpreted as arguments of the surface predicate that they stand with, they are revealed as true instances of Argument Trespassing. Clearly, though, there are several factors which severely delimit the impact of raising as a creator of such trespassing. These are the total absence of O —S raising; the restriction of S —O raising to formal registers and its preferred substitution by dependent clauses without raising: the general avoidance of S — S raising; and the limited number of triggers for both S — O and S — S raising. We can conclude, therefore, that raising processes do exist in Indonesian and account for some loss of semantic transparency, but that, as in German, their effect in this regard is rather limited. 4.3.3. Extractions Indonesian sentences are characterized by stringent word order requirements. Nevertheless, variation from the basic S AUX V O order can be achieved by means of topicalization, passivization, and extraction, which is also used in the formation of relative clauses. Argument scrambling is not permitted.
Towards a semantic typology of language
171
Examples (50 a) and (51a) illustrate the basic word order of Indonesian for the simple active and for the Object Preposing passive (Chung 1976), respectively, a construction which figures prominently in the discussion of long-distance extraction below: (50) a.
(51) a.
Teman-nya Subianto mau mem-bikin kue. friend-3sg want TA-make cake 'Subianto's friend wants to bake a cake.' Anjing itu harus ku-panggil. dog that must I-call 'That dog must be called by me.'
Example (50 b) demonstrates that topicalization can take place from subjects. (50) b.
Subianto,
teman-nya mau mem-bikin kue. friend-3sg want TA-make cake 'As for Subianto, his friend wants to bake a cake.'
Topicalization of objects is not possible, though, unless the sentence has first undergone passivization. (50) c.
* Kue itu, teman-nya Subianto mau mem-bikin. cake that friend-3sg want TA-make 'That cake Subianto's friend wants to bake.'
Finally, (50 d) exemplifies the process of extraction in Indonesian. In a striking parallel to the topicalization case, it again turns out that, whereas extraction of subjects is possible, the extraction of an object renders (50 d) ungrammatical unless the sentence has first been passivized. (50) d.
e.
Siapa yang mau mem-bikin kue? who that want TA-make cake 'Who wants to bake a cake?' * Apa yang teman-nya Subianto mau mem-bikini what that friend-3sg want TA-make 'What does Subianto's friend want to make?'
Indonesian allows long-distance extraction with a few (possibly only two) bridge verbs such as pikir 'think' and kira 'think, guess'. (50) f.
Siapa-kah yang kau-pikir yang [ x j mau mem-bikin kue? who-Q that you-think that want TA-make cake 'Who is thought by you that wants to bake a cake?' = 'Who do you think wants to bake a cake?'
172
Fratta Miiller-Gotama
(51) b.
Apa jang kau-kira jang [x] mau Hendra beli [t]? what that you-guess that want buy 'What is thought by you that is wanted to be bought by Hendra?' = 'What do you think that Hendra wants to buy?' 6
These examples indicate that such extractions are possible out of complement clauses. Notice, however, that passive must obligatorily apply in these sentences to satisfy the constraint that only noun phrases which are subjects at some level are extractable. The underlying object in (51 b) is, therefore, first promoted to subjecthood before it is moved out of the clause. As a result, no difference of the kind observed by Hawkins (1986) for German ensues between extraction out of two and three place predicates. No extraction out of other complement types is allowed, whether the embedded clause is passivized or not. It is even prohibited from complement clauses which are marked with any complementizer other than jang. (52) a.
b.
Kau-pikir bahwa Sandra mau mem-bikin kue. you-think that want TA-make cake 'You think that Sandra wants to bake a cake.' *Siapa jang kau-pikir bahwa [x] mau mem-bikin kue? who that you-think that want TA-make cake 'Who do you think that wants to bake a cake?'
Extraction from purpose clauses is also not allowed. (53) a.
b.
Ibu-nja datang supaja bisa mem-bikin kue. mother-3sg come so that can TA-make cake 'His mother came in order to bake a cake.' *Apa jang ibu-nja datang supaja bisa di-bikin? what that mother-3sg come so that can PASS-make *'What did his mother come in order to make?'
Neither is it possible from adjunct clauses. (54) a.
b.
Dia pulang karena ibu-nja mau mem-bikin kue. he go home because mother-3sg want TA-make cake 'He went home because his mother wants to bake a cake.' *Apa-kah jang dia pulang karena mau what-Q that he go home because want di-bikin oleh ibu-nja? PASS-make by mother-3sg 'What did he go home because his mother wants to make?'
Towards a semantic typology of language
173
The prohibition against the extraction of objects has pervasive consequences for the grammar since it explains at the same time the ungrammaticality of object topicalization and of object-wh-movement, including the exceptional status of the language in the Keenan — Comrie Accessibility Hierarchy to relativization (Müller-Gotama 1987). Moreover, this constraint also explains the curious status of the Indonesian "tough" construction (cf. section 4.3.2. above). First, the Object Extraction Prohibition is a constraint on movement, so it only applies when movement takes place. The fact that it is operative in the Indonesian "tough" construction therefore proves that such sentences indeed involve movement. Secondly, the prohibition explains why the Indonesian "tough" construction requires passivization of the embedded clause with subsequent S —S raising since it prohibits objects from being raised, i. e., moved, directly. We conclude, therefore, that there is a pervasive constraint in Indonesian, the Object Extraction Prohibition, which limits all movement to subjects. This constraint is operative throughout the grammar.
4.4. Indonesian in the Semantic Typology We have noted in the preceding sections that raising and extractions in Indonesian are subject to severe constraints, particularly the pervasive Object Extraction Prohibition. Given the tight constraints on extractions and raising, these grammatical processes are even more limited than in German. The location of Indonesian in the overall typology, thus, is split between a close affinity with the transparent type in all grammatical properties which involve movement and with the grammaticizing type in all other cases, including the low degree of lexical specificity, the accessibility of diverse semantic roles to subjecthood, and the coordinate deletion facts. This peculiar split of properties can be viewed as a direct result of the limitation of extraction to subjects. Since the extractability of a noun phrase is essential for a great many grammatical processes, including relativization, the "tough" construction, the formation of wh-questions, and topicalization, it is extremely important for a noun phrase to gain access to extraction. For agentive noun phrases, this access is unproblematic since they become prototypical subjects. For subcategorized objects of verbs, accessibility is guaranteed by means of passivization; the unified treatment of all objects, contrary to the differentiation made in German, exploits this strategy to the largest possible extent. But what of oblique noun phrases? Indonesian offers them two solutions: First, some obliques have the opportunity to become objects
174
Fran^
Mülkr-Gotama
through the suffixation of the verb with -kan or -i, which makes them eligible for promotion to subjecthood through passivization. Secondly, Indonesian applies a more direct solution for oblique noun phrases by loosening semantic restrictions on subjects. Far from being a possible but accidental collocation of properties, the regularities we observed in Indonesian, thus, follow logically from one another. In particular, it seems to be the case that the restriction of extraction to only one grammatical relation, the subject, is a major cause for the liberal, English-like properties in other areas of the grammar. Pending the crosslinguistic confirmation of this collocation through a large scale typological survey and a comparative study of closely related languages, like the strongly subject-prominent Malagasy as well as the Philippine languages (Keenan 1976), it is therefore possible to state the following implication. (55)
If a language restricts extraction to only one grammatical relation, it will compensate for this constraint by creating strategies which allow other noun phrases to access this same grammatical relation.
5. Conclusion Our analysis of Indonesian has shown that there is a split in the properties of that language, with some aspects behaving in a strongly grammaticizing way and others in a very transparent manner. I have argued that this split is triggered by an overriding force in Indonesian, namely the centrality of subjects and the consequent importance for noun phrases to acquire subjecthood. Only subjects are accessible to processes like wh-movement, topicalization, the "tough" construction, etc. Therefore, the language has developed several strategies for permitting more noun phrases to become subjects, a move which has resulted directly in a loss of semantic transparency in many areas of the grammar. In a modular theory of language organization, we should expect that the interaction between competing principles creates similar overriding pressures in other languages and with respect to other typologies. Split properties like those found in Indonesian do not, therefore, constitute counterarguments against the validity of the scalar model of the Semantic Typology. The deviations of Indonesian were neither unsystematic nor was the language characterized by intermediate scores between the two extreme. Indeed, the very preciseness of the split between properties relating to Argument Tres-
Towards a semantic typology of language
175
passing on the one hand, and the degree of syntactic conflation of syntactically or semantically diverse categories that a language allows on the other, reinforces the assumptions of the Semantic Typology. Based on our analysis of the Indonesian data, the notion of semantic transparency can be interpreted as an aggregate of two partially independent subdimensions. The first of these is defined by the Argument Trespassing Principle (Hawkins 1986: 97), whereas the second is an indicator of the degree of syntactic conflation of syntactically or semantically diverse categories that a language allows. Among the languages which have been examined from this perspective, German, Russian, and especially Korean score low on both subdimensions, tolerating only little syntactic conflation and argument trespassing. Conversely, English scores high on both counts, evidencing the strongly grammaticizing tendencies of the language. Our study of Indonesian has introduced a further language into the discussion and has demonstrated that it does not pattern with either of the extreme types across the board. Rather, one part of the grammar consistently patterned with the grammaticizing type, while another patterned much closer to the transparent type. The question that immediately arises is, of course, whether there can be any languages of the fourth, as yet unattested type, or if the required combination of parameters, a high score on argument trespassing and a low score on syntactic conflation, can be ruled out by an implicational universal. A priori, all that can be said is that the presumed properties of this fourth type are not self-contradictory nor necessarily disadvantageous: a type 4 language severely restricts the kinds of possible semantic roles for individual grammatical relations while at the same time permitting long-distance movement. It is easy to see how such a semantic role restriction should facilitate the recognition of extractions on-line, particularly if a high score on syntactic conflation does in fact correlate with extensive grammatical morphology: as soon as the hearer encounters two noun phrases marked for the same syntactic/ semantic role, he is thereby alerted to the likelihood that extraction has taken place. At this point, however, we must leave the determination of the status of the fourth type to future research.
Notes 1. I would like to thank my principal informants, Jeongdal K i m for Korean, and Jacintha and Caecilia Gotama for Indonesian. My thanks also to the numerous Indonesians and Koreans w h o have provided occasional data and grammatically judgements. Yale romanization is used for Korean.
176
Fran%
Miiller-Gotama
For their comments on a previous version of this paper, I am endebted to Bernard Comrie, John Hawkins, Joseph Aoun, and Jacqueline Schachter. It goes without saying that any remaining inadequacies are mine alone. 2. Of course, we would expect a strong correlation among the positions of the various properties of a language L, all things being equal. In addition, we would expect it to be the case that significant deviations arise as a result of clearly identifiable reasons, cf. the discussion in the text. 3. The conception of the extremes defined by the typology differs from that drawn in Hawkins 1986, where complete transparency means that every possible meaning distinction has its own distinct surface form, yielding an infinite number of surface forms, while complete grammaticalization is understood as the collapsing of all possible semantic differences onto a single grammatical form (Hawkins 1986: 126). Needless to say that no usable natural languages could be located anywhere near the extreme values of a such defined typology. In contrast to Hawkins, the present paper takes a narrower view of the range that needs to be defined for the Semantic Typology. In our view, transparency becomes a measure of the degree to which the syntax preserves the distinctions made in the semantics. An example from the realization of semantic roles should illustrate this difference: we take it for granted that the semantic component of a language distinguishes a certain, finite number of semantic roles. In the most extreme transparent scenario, the syntax would preserve all of these semantic distinctions, but the syntax cannot increase transparency by going beyond the number of semantically distinguished roles since any deviation from the semantics constitutes a loss of transparency. In other words, the syntax cannot reintroduce linguistically any real-world distinctions that are not made in the semantics. Introducing an infinite number of surface syntactic distinctions, therefore, would not increase semantic transparency as it is understood in this paper. 4. Standard English is the only variety of English investigated for its degree of transparency and grammaticization so far (in Hawkins 1986). There appear to be some interesting differences among the varieties of English since some dialects readily admit pronoun copies. Such differences are relevant for the Semantic Typology. Their extent and their implications, however, must be relegated to future research. For the purposes of this paper, any mention of "English" should be understood as refering to its standard variety. 5. The complementizer untuk is optional if the embedded clause has no overt material other than the verb. 6. Both verbs in (51) are in the so-called Object Preposed (OP) construction (Chung 1976), which is a passive strategy in Indonesian. Verbs with object preposing carry no overt passive affix, but, unlike active verbs, are characterized by the cliticization of the underlying subject to the verb (hence kau-kira in (51)). Modals, which usually stand between the subject and the main verb, obligatorily precede the logical subject in the OP-passive and, hence, form a test for cliticization. Notice the sentence Buku itu mau Hendra beli (Book that want Hendra buy), which shows this sequence of preposed object — modal — logical subject — verb and which is, therefore, an instance of the OP-passive, as are the verbs in (51).
Tomarás a semantic typology of language
177
References Abasolo, Rafael 1974 Basic semantic structures of Korean. [Dissertation, Georgetown University.] Bowers, John S. 1986 Grammatical relations. New York: Garland. Chang, Sok-Chin 1983 "Reference in Korean discourse", in: Korean National Commission for UNESCO (eds.), The Korean language, Pace International Research, 219—263. Chomsky, Noam 1977 "On WH-movement", in: Peter Culicover—Thomas Wasow—Adrian Akmajian (eds.), Formal syntax, New York: Academic Press. Chung, Sandra 1976 "An object-creating rule in Bahasa Indonesia", Linguistic Inquiry 7: 41—87. Comrie, Bernard 1981 Language universals and linguistic typology. Chicago: University of Chicago. 1986 "Review of Kvam, Sigmund and Sven-Gunnar Andersson, Satzverschränkung im heutigen Deutsch." Studien \ur deutschen Grammatik, 24. Tübingen: Narr (1984) and S. Kvam, Linksverschachtelung in der deutschen und norwegischen Gegenwartssprache. Linguistische Arbeiten, 130. Tübingen: Niemeyer (1984)", Linguistics 24: 1133-1134. 1986 b "Contrastive linguistics and language typology", in: Dieter Kastovsky —Aleksander Szwedek (eds.), Linguistics across historical and geographical boundaries. Vol. 2: Descriptive, contrastive and applied linguistics, Berlin: Mouton de Gruyter, 1155-1163. 1987 The unity of syntactic contrasts in English and Russian (MS). Paper presented at the University of Southern California. Comrie, Bernard — Stephen Matthews 1990 "Prolegomena to a typology of tough movement", in: William Croft—Keith Denning — Suzanne Kemmer (eds.), Studies in typology and diachrony: Papers presented to Joseph H. Greenberg on his 75th birthday ( = Typological Studies in Language 20.), Amsterdam/Philadelphia: John Benjamins, 43 — 58. Fillmore, Charles J. 1968 "The case for case", in: Emmon Bach—Robert T. Harms (eds.), Universals of linguistic theory, New York: Holt, Rinehart and Winston. 1971 "The case for case reopened", in: Peter Cole—Jerrold M. Sadock (eds.), Grammatical relations ( = Syntax and semantics 8), New York: Academic Press. Hawkins, John A. 1986 A comparative typology of English and German. London: Croom Helm. Keenan, Edward L. 1976 "Towards a universal definition of subject", in: Charles N. Li (ed.), Subject and topic, New York: Academic Press. Kim, Nam-Kil 1987 "Korean", in: Bernard Comrie (ed.), The world's major languages, London: Croom-Helm, 8 8 1 - 8 9 8 .
178
Fratta Miiller-Gotama
Kirkwood, H. W. 1969 "Aspects of word order and its communicative function in English and German", Journal of Linguistics 5: 85 — 107. Müller-Gotama, Franz 1987 Indonesian and the accessibility hierarchy: filling the gap, Paper presented at the 17th Annual Western Conference on Linguistics, University of Washington, Seattle. Plank, Frans 1980 Verbs and objects in semantic agreement: minor differences between languages that might suggest a major one (MS). Sapir, Edward 1921 Language: An introduction to the study of speech. New York: Harcourt Brace. Seiler, Hansjakob and Christian Lehmann (eds.) 1982 Apprehension: Das sprachliche Erfassen von Gegenständen (2 vols.). Tübingen: Narr. Yang, In-Seok 1972 Korean syntax: case markers, delimiters, complementation, and relativization. [Dissertation, University of Hawaii.]
A hierarchy of main predicate encoding Leon Stassen
1. General background This paper aims to investigate the cross-linguistic variation in the encoding of main predicates, and to arrive at implicational statements which define the limits of this variation. It thus represents a report on a typological linguistic research project, in which the typology is based on the various ways in which the parameter "main predicate" is manifested in the structure of the languages under discussion. The research was based on data from a sample of 200 languages, which are listed in the appendix. With regard to the parameter under investigation, a few preliminary definitions and restrictions will have to be made explicit. First of all, no attempt will be made here to define the notion of "predicate". This notion will be taken to be axiomatic, and will be used here in much the same way as it is done in standard logic: anyone who understands what the 'Ρ' stands for in logical formulas like the ones in (1) will understand what I mean by the term "predicate": (1) a. b.
Ρ (a) V χ ((P) (x))
As for the encoding of these logical predicates in natural languages, I have confined myself to the way in which main predicates are morphosyntactically manifested. Thus, while in a sentence like (2)
The old man
slept.
all three items old, man and slept can be said to be linguistic manifestations of logical predicates, only the way in which items like slept in (2) are encoded will be the subject of our investigation. As a further point, I must stipulate that only main predicates which occur in main clauses will be included in the data base of our typology. That is, while in a sentence like (3)
John said that Harry
cheated.
both the items said and cheated represent main predicates in their respective clauses, only items which function like said in (3) will be taken into consid-
180
Leon S fassen
eration. (The main reason for including this restriction is that, in quite a few languages, main predicates of subordinate clauses often have non-finite or finite forms which are specifically marked for subordination, so that by limiting ourselves to main predicates of main clauses we eliminate a possible interfering factor in the morphosyntactic representation of main predicates). Lastly, at this stage of our investigation we will discuss only cases of oneplace main predicates, typically encoded as intransitive predicate items. Thus, the sentences in (4) are prototypical of the data which are relevant to us: (4) a. b. c. d.
John John John John
walks. is tall. is a carpenter. is in the kitchen.
Given these restrictions, our first, descriptive, aim can be formulated in the following question: what does a main predicate of an intransitive main clause look like in the languages of our sample?
2. Differentiation of encoding At the very start of our investigation, it is possible to formulate a striking difference between the ways in which intransitive main predicates are encoded in natural language, and the way in which such predicates are represented in formulas of standard logic. In the large majority of natural languages, we can observe that there is no (or no complete) uniformity in the morphosyntactic encoding of intransitive main predicates. Unlike logic, which treats all intransitive main predicates in the same syntactic manner, most natural languages show a diversification in their encoding of intransitive main predicates, depending on the kind (i. e., the class) of the predicate at issue. Thus, whereas in logic all intransitive main predicates have the same syntax, regardless of whether they are to be interpreted as "walk", "tall" or "carpenter", natural languages tend to reflect the semantic (or cognitive, or conceptual) differentiation of intransitive main predicates wholly or partially, in that they tend to employ different formal, morphosyntactic encoding strategies for different semantic classes of intransitive main predicates. This formal differentiation is, of course, one of the major criteria by which various parts of speech are usually distinguished for a given language: the morpho-syntactic behaviour of lexical items when employed as main predicates is generally seen as one of the most important distributional properties of those items.
A hierarchy of main predicate encoding
181
Within the domain of semantic functions that can be expressed by intransitive main predicates, four subtypes are commonly distinguished. In traditional Western grammar, these four subtypes are generally associated with a particular part of speech, which is seen as the prototypical morphosyntactic encoding of the predicational function in question. Table 1 shows these four semantic subtypes, and their typical morphosyntactic counterparts. Table 1. S E M A N T I C FUNCTION
TYPICAL ENCODING
Activity/State
(Intransitive) Verbs
Property
Adjectives
Class Membership/Identification
Nouns or Noun Phrases
Location/Existence
Adverbs or Adverbial Phrases
With respect to this table, it should be added that the classification of intransitive predicates into four subclasses may very well be in need of further refinement. In particular, the subtype designated here an Class Membership/ Identification may be divided further into types such as Class Membership, Role, Specification and Identification (see Graham, 1965). For the purpose of our typology, however, the rough stratification of the semantic domain as given in Table 1 will do.
3. The encoding of the semantic functions Given that the semantic domain covered-by intransitive main predicates can be divided into four subtypes, our first task will be to sample the ways in which these four semantic functions are encoded in the languages under consideration. Following that, we should be able to group the languages of our sample into types, on the basis of whether or not they differ in the strategies which they employ in the encoding of the various functions of intransitive main predicates. Now, once the relevant data have been collected, there are a few general observations which can be stated with some degree of confidence. First, it turns out that, in a limited group of languages, all semantic functions that intransitive main predicates may have are encoded by one single morphosyntactic strategy. That is, in a way which is parallel to formal logic, in these
182
Leon Stassen
languages an item functioning as the main predicate of an intransitive sentence will always be encoded in the same morphosyntactic manner, regardless of its specific semantic content. It will be clear that in languages of this kind an important criterion for the distinction of different parts of speech is not available. Seen from a particular perspective, one might call these languages "syntax-oriented", since it is the syntactic function of being an intransitive main predicate which is taken as the decisive factor in the morphosyntactic encoding. As such, these languages are opposed to what might be called "lexicon-oriented" languages, in which the lexical class of the predicative item (and, by prototypical connection, the semantic function expressed by the item) is, at least to some degree, the determinant factor in the choice of the morphosyntactic encoding strategy. Examples of languages in which this uniform strategy for main predicate encoding can be observed are Piro, Mordvin and Turkish. In Piro, an Arawakan language of Bolivia, intransitive main predicates are marked by two successive sets of suffixes, the first of which refers to tense and aspect, and the second of which cross-refers to the subject. As the examples in (5) show, this marking of predicates takes place regardless of the semantic function expressed by the predicative item. (5) a.
b.
c.
d.
e.
f·
Piro (Matteson, 1965) Ya — tka — na go - PAST - 3PL 'They went.' (139) Hitsko — 0 — na strong - PRES - 3PL 'They are strong.' (143) Kapakle — 0 — ru hunter - PRES - 3SG 'He is a hunter.' (45) No — prapa — tka yi my - pet - PAST 2SG 'You were my pet.' (39) Hew i — 0 —no here - PRES - 1SG Ί am here.' (84) Ko ko — yma — 0 na 3PL uncle — with — PRES 'They are with Uncle.' (93)
A hierarchy of main predicate encoding
183
In Mordvin, a Finno-Ugric language, all intransitive main predicates are marked by means of a set of inflexional suffixes, which constitute a combined reference to tense and to the subject of the sentence. (6) a.
b.
c.
d.
e.
Mordvin (Wiedemann, 1865) Mon mol — an 1SG go - 1SG.PRES Ί go.' (69) Mon par — an 1SG good - 1SG.PRES Ί am good.' (61) Mon tsor — at —an 1SG son - your - 1SG.PRES Ί am your son.' (61) Min te oss —to — tanok ÎPL this town - from - ÎPL.PRES 'We are from this town.' (61) Oss — o — lin town - in - 1SG.PAST Ί was in town.' (61)
A similar situation can be attested for Turkish, where, as sentences (7 a —d) illustrate, subject-marking by means of inflexional suffixes is employed for all intransitive main predicates, regardless of whether they must be rated as verbs, adjectives, nomináis or adverbials. (7) a.
b.
c.
d.
Turkish (Lewis, 1967) Gel — iyor — sun come - PRES.PROG - 2SG 'You are coming.' (109) Gii^el — sin beautiful - 2SG 'You are beautiful.' (177) Türk — sün Turk - 2SG 'You are a Turk.' (98) Ev — de — sin house — in — 2SG 'You are at home.' (98)
The situation in Turkish, however, is a bit more complex than in the other two languages discussed here. While in Piro and Mordvin the uniformity of
184
Leon S fassen
main predicate encoding is fairly absolute, in Turkish this unity of encoding strategies is restricted to sentences that have present tense. If the sentence has any other tense, the uniformity of encoding breaks down, and gives way to a diversification of encoding, based upon the recognition of various classes of predicative items. Thus, while verbs in Turkish keep a full inflexional system throughout all tenses and aspects, other classes of predicative items are morphologically inert if the sentence has non-present tense: adjectives, nomináis and adverbial predicates in non-present tense sentences remain uninflected, and require the presence of some suppletive verb with a meaning that is equivalent to English to be. (8) a.
b.
c.
Turkish Sen %engin i — di —η 2SG rich be - PAST - 2SG 'You were rich.' (Lees, 1972: 65) Türk i — di —η Turk be - PAST - 2SG 'You were a Turk.' (Lewis, 1967: 99) Ev — de i — di — η house — in be — PAST — 2SG 'You were at home.' (Lewis, 1967: 99)
We may conclude, then, that the set of languages which show a uniformity in their handling of intransitive main predicates is rather small, and that at least some of the members of this set need additional specification as to the conditions under which this uniformity may be exhibited. Opposed to languages in which every intransitive main predicate is encoded by the same morphosyntactic strategy, we find a large majority of languages which show formal differentiation to at least some degree. Thus, in most languages the various semantic functions are grouped formally into various types for morphosyntactic encoding. In Table 2, I present the eight logically possible options in intransitive main predicate encoding, given that there may be no formal diversification, or diversification of encoding to a greater or smaller degree. In this table, the following notational conventions should be observed. A " - f " in a column means that the semantic function at issue is encoded by the morphosyntactic strategy that is employed in the language to encode Activity/State, regardless of the exact formal nature of this encoding. Thus, a " + " signals the verbal strategy of the language, and its occurrence in the matrix indicates that the semantic function in question is formally encoded as an intransitive verb. In contrast, a " — " means that the relevant semantic function is encoded by a morphosyntactic strategy that is not in use
A hierarchy
of main predicate
encoding
185
for the encoding of Activity/State in the language; again, this qualification holds regardless of the exact formal make-up of this non-verbal strategy. Table 2.
ACTIVITY (V) PROPERTY (ADJ) CLASS MEMBERSHIP (N) LOCATION (ADVERB)
I
II
Ill
¡V
Y
VI
VII
Vili
+ + + +
+ + +
+ +
+
+
+
-
+ +
+
-
-
+ +
—
+
• •
-
On the basis of this set of options, which exhausts the logically possible combinations of verbal and non-verbal encodings for intransitive main predicates, we should be able to categorize the languages in the sample into at most eight types. In this connection, it should be remarked that not every language in the sample can be unequivocally categorized in this way; there are quite a few languages which allow the choice between two, or even more, options. This fact, however, will not damage the validity -of the conclusions that can be drawn from the typology. As far as I am aware, in languages with more than one option all choices conform to the tendencies that can be formulated on the basis of those languages which have only one option in intransitive main predicate encoding.
4. The Verbalization Hierarchy If we examine the question of whether all of the eight logically possible encoding types are actually attested in the languages of the sample, we find that the languages in the sample are distributed over these eight types in a manner that is almost certainly non-random. To be specific, it turns out that examples of the types I, II, III and IV are fairly easy to come by, whereas the types V, VI, VII and VIII hardly, if ever, occur. That there are languages with an encoding strategy of Type I will, of course, come as no great surprise to the reader. It is this strategy which embodies the uniformity of intransitive main predicate encoding; examples for Piro, Mordvin and Turkish have been given in section 3. As an example of a language which has an encoding strategy of Type II we can cite Biloxi, a Siouan language. In this language, intransitive activity/state predicates are marked for person, number and tense
186
Leon Stassen
by a set of inflexional prefixes. As can be seen from sentences (9 b —c), this same strategy is in use for cases where the predicate designates a property or a class. If, however, the predicate designates a location, the verbal strategy can no longer be maintained; predicative locationals cannot be inflected, and require that the inflexional morphology be carried by a suppletive verb, which is chosen from the set of so-called "posture verbs" that Biloxi possesses. (9) a.
b.
c.
d.
e.
Biloxi (Dorsey and Swanton, 1912) Nk — anki 1SG.PRES - sit Ί am sitting.' (290) Nk — anyasahi 1SG.PRES - black Ί am black.' (179) Nk — sankt 1SG.PRES — woman Ί am a woman.' (179) On ti ja ti tei η — anki bear the house in 3SG.PRES — sit 'The bear is in the house.' (53) Ewa η — y like — di there ÌPL - stand - DUR.PAST 'We were there.' (135)
A second example of Type II encoding, which parallels the case of Biloxi in all relevant respects, is represented by Pipil, a Uto-Aztecan language. (10) a.
b.
c.
d.
Pipil (Campbell, 1985) Ni — yaw 1SG — go Ί go.' (142) Ni — tumawak 1SG - fat Ί am fat.' (55) Ni — ta:kat 1 SG — man Ί am a man.' (55) Ni — ne m i tik techan 1 SG — be in town Ί am in town.' (112)
A hierarchy of main predicate encoding
187
The option of the type designated as Type III in Table 2 represents a very common split in the encoding strategies for intransitive main predicates. This type contains those languages in which adjectival predicates can be said to be "verby" (cf. Wetzer, this volume); property predicates are treated as (some subclass of) intransitive verbs, whereas predicative nomináis and adverbials are not treated in this way. A clear example of a case in point is the encoding strategy of Guajajara. In this Brazilian Tupi language, both "verbs" and "adjectives" show person/number-agreement with their subjects, by means of inflexional prefixes. In contrast, predicatively used nomináis and adverbials do not show this inflexion: they remain morphologically inert, and are linked with their subjects by means of a "zero-verb". (11)
Guajajara a.
U
—
ku%a?i
3PL - fall 'They fell.' (Bendor-Samuel, 1972: 131) b.
H
— urywete
aie
3SG — happy 3SG 'He is happy.' (Harrison, 1986: 411) c.
Inamusin
h
—
eimaw
bird his — pet 'His pet is a bird.' (Harrison, 1986: 411) d.
Zawar
i
— pjr
dog him — with Ά dog was with him.' (Bendor-Samuel, 1972: 161) In the case of Guajajara, the split of encoding strategies which is embodied in Type III is rather straightforward: predicative verbs and adjectives follow an inflexional verbal strategy, whereas predicative nomináis and adverbials are both encoded by means of a zero-strategy. It should not be thought, however, that such a situation holds for all of the languages which can be categorized as members of Type III. A somewhat more complex case of this type is represented by Goajiro, an Arawakan language of Northern Colombia. Here we find that, as in Guajajara, verbs and adjectives share the same verbal strategy; in the case of Goajiro, this strategy consists in obligatory number/ gender agreement with the subject (which, Goajiro being a VSO-language, follows the predicate). Both predicative nomináis and predicative adverbials lack the possibility of this type of cross-reference to their subjects. However, while predicative nomináis in Goajiro employ a zero-strategy, predicative adverbials require the presence of an overt locational verb.
188
Leon Stassen
(12)
Goajiro (Holmer, 1949) a.
Λ
— wate
MOMENTAN. 'He flew up.' (476) b.
Ana
fly
— s
— c
-
—pa
MASC.SG. -
wayu
—
PAST
kaih
good — MASC.SG Indian — the 'The Indian is good.' (476) c.
Wayu
way a
Indian 1PL 'We are Indians.' (492) d.
Ya
— s
taya
sotpa
b e a t - MASC.SG 1SG beach Ί am on the beach.' (468) Yet another variant of the encoding strategies subsumed under Type III can be encountered in Lushai, a Chin language of Northern Burma. Again we find that predicative verbs and adjectives are treated by the same verbal strategy, which in this case takes the form of an obligatory subject pronoun in front of the predicate. Both predicative nomináis and adverbials differ from this verbal strategy, in that they require the presence of an overt suppletive verb. The point is, however, that the suppletive verb employed with predicative nomináis (i. e., the Lushai copula) is not the same lexical item as the suppletive verb used in sentences with an adverbial predicate. (13)
Lushai (Lorrain and Savidge, 1898) a.
Mi
a
hon
e
man he come INDICATIVE 'The man came.' (29) b.
In
a
lain
e
house it big INDICATIVE 'The house is big.' (6) c.
Thualthat
a
ni
murderer he COPULA 'He is a murderer.' (26) d.
Leilawn
a
om
bridge it be over there 'The bridge is over there.' (38) On the basis of these data from Guajajara, Goajiro, and Lushai, we can conclude that these three languages can all be categorized as members of Type III in our typology of intransitive main predicate encoding. These
A hierarchy of main predicate
encoding
189
languages all conform to the defining feature of Type III: they show a split between a verbal strategy for predicative verbs and adjectives, on the one hand, and some non-verbal strategy for predicative nomináis and adverbials, on the other hand. This said, however, we must also note that the languages which can be categorÍ2ed as members of Type III allow for a certain measure of diversification in their strategies, with the result that one might consider a subcategorization of this type. Thus, it might be possible to construct a set of subtypes for Type III, on the basis of whether or not the non-verbal strategies for predicative nomináis and predicative adverbials are identical, and on the basis of the exact formal nature of the non-verbal strategy or strategies employed. In particular, the question of whether a zero-strategy or an overt suppletive strategy is employed for one or both of the non-verbally encoded functions might be seen as a criterion for a further subdivision of Type III. I am of the opinion that such a subcategorization may indeed be rewarding, and that a preliminary examination of the relevant data seems to suggest that the combinations of options that make up this subcategorization are not totally unrestricted. However, a full exploration of these possibilities lies beyond the scope of the present paper. Turning now to Type IV in our typology, we notice that, once more, this type constitutes a very frequent option in intransitive main predicate encoding. The languages in which only activity predicates are encoded verbally while all other semantic functions of intransitive predicates are encoded in a non-verbal way form a fairly extensive set, which comprises a number of language groups that are very familiar to Western readers: all Germanic, Romance and Slavic languages belong to this type. English is a fairly typical example. As the sentences in (4) illustrate, English is a language in which there is at least rudimentary inflexion for activity predicates, but in which all other semantic classes of intransitive predicates are encoded by way of a nonverbal construction, which requires the presence of the suppletive verb to be. Other languages which are completely 2 parallel to English in this respect are Latin (see the examples in (14)) and Russian (compare the examples in (15)). But the "English-style" option is by no means confined to Indo-European languages: the examples in (16) show that a completely unrelated language like Imbabura Quechua also adheres to this typological option. (14)
Latin (Troll, 1975) a-
Femina
cur rit
woman walk-3SG.PRES 'The woman walks.' (46)
Leon Stassen
b.
c.
d.
a.
b.
c.
ά.
a.
b.
c.
d.
Rosa pulchra est rose-NOM.SG.FEM beautiful-NOM.SG.FEM be-3SG.PRES 'A/the rose is beautiful.' (91) Rosa fio s est rose-NOM.SG.FEM flower-NOM.SG.MASC be-3SG.PRES Ά rose is a flower.' (91) Navis est in portu ship-NOM.SG be-3SG.PRES in harbour-ABL.SG 'The ship is in the harbour.' (91) Russian On govorit 3SG.MASC speak-3SG.PRES 'He speaks.' (Fennell, 1961: 18) Okno byl vysokoe window-NOM.SG.MASC be-3SG.PAST high-NOM.SG.MASC 'The window was high.' (Raptschinsky, 1948: 15) Ona budet ucitel'nic — ej 3SG.FEM be-3SG.FUT teacher - ABL.SG 'She will be a teacher.' (Raptschinsky, 1948: 27) On u menja veer a byI 3SG.MASC at 1SG.GEN yesterday be-3SG.PAST 'Yesterday he was with me.' (Raptschinsky, 1948: 170) Imbabura Quechua (Cole, 1982) Nuka wawki ri — nga my brother go — 3SG.FUT 'My brother will go.' (78) Nuka wast — ka yuraj — mi ka — rka my house - TOP white - VALIDATOR be - 3SG.PAST 'My house was white.' (67) Juan — ka mayistru — mi ka — rka J. - TOP teacher - VALIDATOR be - 3SG.PAST 'Juan was a teacher.' (67) J u y j — ka Utavalu — pi — mi ka — ju — J. - TOP U. - in - VALIDATOR be - PROG η 3SG 'José is in Otavalo.' (67)
LS was the case with Type III, the languages which make up the set under e IV might be categorized further on the basis of the kind of non-verbal
A hierarchy of main predicate encoding
191
strategy or strategies which they employ. English, Latin, Russian, and Imbabura Quechua are all cases of Type IV languages in which there is a uniform encoding strategy for all non-verbal functions and where, moreover, this non-verbal strategy is (optionally or obligatorily) characterized by the presence of an overt suppletive verb. Tiwi, a Northern Australian language, presents a somewhat different variant of Type IV. In this language, too, there is one single non-verbal strategy, but this non-verbal strategy is characterized by obligatory zero-marking of the predicate in question. (17)
Tiwi (Osborne, 1974) a.
Ji
—
3SG.MASC.PAST 'He fell.' (70) b.
Tungkwaltiringa
pauligi
fall
pumpuka
stringy bark good Ά stringy bark is good.' (60) c.
Purukuparli
m am tina
P. boss 'Purukuparli is boss.' (60) d.
Ningani
kanitawa
today bread 'Today there is bread.' (66) Furthermore, Type IV comprises a number of languages in which there is no uniformity of non-verbal strategies. For instance, Zapotee, a language of Mexico, offers the situation in which predicative adjectives and nomináis are linked with their subjects by means of a zero item, whereas the predication of location requires an overt suppletive locational verb. (18) a.
b.
Zapotee — nya — ba lo j u PAST - fall - 3SG.ANIMAL to ground 'It (i.e., a dog) fell to the ground.' (De Angulo —Freeland, 1935: 123) Ri
Nadjici
ve
angry he 'He is angry.' (Radin, 1930: 83) c.
Karpinteru
la?a
be
carpenter person 3SG 'He is a carpenter.' (Pickett, 1960: 55) d.
Jju
bet si
nye
—
ba
be-there lice foot — his 'There were lice on his feet.' (De Angulo —Freeland, 1935: 123)
192
Leon
Stassen
However, I must point out once again that, for the purpose of this paper, I have chosen to neglect the differences between the languages under discussion here, and hence I have categorized all these languages as members of the broadly defined Type IV. As I said above, the types I, II, III and IV in our typology cover practically all instances of intransitive main predicate encoding that have been attested in the sample. Languages which choose as their option Type V, VI, VII or VIII simply do not occur in my data, and although my sample, like any other sample used today, is far from perfect, I feel that other samples would not present a picture that differs radically from the one presented by the sample used here. Thus, we may formulate the following universal statements with a certain degree of confidence: (19) a.
b.
c.
d.
There are no languages in which predication of properties is encoded non-verbally, while at the same time all other semantic functions of intransitive main predicates are encoded by the verbal strategy. There are no languages in which predication of class membership is encoded non-verbally, while at the same time all other semantic functions of intransitive main predicates are encoded by the verbal strategy. There are no languages in which predication of both properties and class membership is encoded non-verbally, while at the same time predication of location is encoded by the verbal strategy. There are no languages in which predication of both properties and location is encoded non-verbally, while at the same time predication of class membership is encoded by the verbal strategy.
The statements in (19) can be rated as negative absolute universals of intransitive main predicate encoding: they are meant to specify that a number of logically possible options in the typology at hand are excluded on empirical grounds. On the positive side, we can also take a closer look at the options which do have their basis in the data. These options are the types I, II, III and IV in Table 2, which, for the sake of convenience, I reproduce as Table 3. Table 3.
ACTIVITY (V) PROPERTY (ADJ) CLASS MEMBERSHIP (N) LOCATION (ADV)
I
II
Ill
IV
+ + + +
+ + +
+ +
+
-
—
-
-
-
—
A hierarchy of main predicate
encoding
193
If we look at Table 3, keeping in mind that the four options contained in it represent the only four empirically attested possibilities, we cannot fail to observe a number of highly specific constraints on the possibilities of intransitive main predicate encoding. In particular, Table 3 allows us to formulate the following two implicational universale: (20) a. b.
If, in a language, Location is encoded by the verbal strategy, then all intransitive main predicates will be encoded by that strategy, If, in a language, Class Membership is encoded by the verbal strategy, then Property will be encoded by that strategy.
An alternative way to formulate the conclusions that can be drawn from Table 3 is to state that the options contained in it represent a hierarchy of semantic functions of intransitive main predicates, with respect to their "accessibility" for encoding by means of the verbal strategy. This hierarchy, which may be called the Verbalization Hierarchy of Intransitive Main Predicates, can be represented by the following formula: (21)
The Verbalization Hierarchy of Intransitive Main Predicates Semantic: (activity) > property > class > location Morphosyntactic: (verb) > adjective > noun > adverb
This formula should be read as shorthand for the following claim about intransitive main predicate encoding: if a language can use the verbal strategy for a semantic function (c. q., the corresponding lexical class) on a given position on the hierarchy, then it can use this verbal strategy for all positions higher (i. e., to the left) on the hierarchy. Furthermore, the hierarchy is meant to formulate the claim that for each position on the hierarchy, there is some natural language which can "verbalize" (i. e., employ the verbal strategy) on that position and all higher positions, but not on lower positions. Finally it should be noted that the highest position on the hierarchy, i. e., activity or verb, is put in brackets. This notation is meant to stipulate that this highest position is verbalized by definition, and is therefore not a part of the hierarchy in the same way as the other positions are.
5. Explanations for the Verbalization Hierarchy To the extent that the data upon which my research was based are reliable and representative, the Verbalization Hierarchy of Intransitive Main Predicates can be said to constitute a hard empirical fact. Given this, a further step in
194
Leon Stassen
the research project should consist in the quest for an explanation of this fact: we should ask ourselves why the typology, and the hierarchy into which the typology can be restated, are the way they are. As for solutions to this question, one possibility which comes readily to mind is the so-called Time Stability Hypothesis, which was first developed by medieval philosophers of language (see Bursill-Hall, 1971), and whose most recent formulation has been given by Talmy Givón. In its essence, the Time Stability Hypothesis claims that (classes of) semantic predicates, as well as their prototypically corresponding lexical classes, can be ordered as to whether they denote states that are, to a greater or lesser degree, "permanent" or "stable over time". According to Givón (1984: 55), the major predicate classes of verbs, adjectives and nouns form a time-stability scale, in that verbs tend to denote states that can be subject to (often rapid) change, nouns tend to denote the most permanent or time-stable states, while adjectives denote states that are somewhere intermediate between these extremes. In (22), this Time Stability Hypothesis is illustrated; left-to-right ordering corresponds with an increase in time stability. (22)
Time Stability Scale VERBS ADJECTIVES
NOUNS
Now, if we examine the Time Stability Hypothesis as a possible candidate for the explanation of the Verbalization Hierachy, we can note that this hypothesis works well for at least part of the hierarchy. In particular, the three highest positions on the hierarchy correspond exactly in their ordering to the Time Stability Scale presented in (22): the left-to-right increase in time stability on the scale has a perfect match in the left-to-right decrease of verbalization on the hierarchy. Given this situation, one might venture the hypothesis that Time Stability is the underlying, explanatory factor in the Verbalization Hierarchy, and that therefore the following statement is valid: (23)
The less time-stable an intransitive main predicate is, the greater its chances are of being encoded by the verbal strategy.
As noted, hypothesis (23) provides an explanation for the ordering of verbs, adjectives and nouns in the Verbalization Hierarchy, and can therefore be said to be at least partially successful. The problem with this explanation, however, lies in the position of adverbs on both the scale and the hierarchy. In Givón's formulation of the time stability scale, adverbs are not included. We may, however, assume that, if adverbs are going to be included in the scale, they will not be situated at the right-hand extreme. It is certainly counter-intuitive to rate the states which are denoted by adverbs as being
Λ hierarchy of main predicate
encoding
195
more time-stable than the states that are typically denoted by nouns. Predicatively used locationals (i. e., the class of adverbials to which we have restricted ourselves in this study) typically indicate that the subject is located in a given space at a given point in time. This state of location must be rated as relatively unstable over time, certainly more unstable than class membership, and presumably also more unstable than properties. Thus, if the Time Stability Hypothesis were to function as a perfect match for the Verbalization Hierarchy, we would expect the hierarchy to be ordered along the lines of the revised time-stability scale as given in (24), instead of the actual ordering in (21), where adverbs occupy the lowest position: (24)
Revised Time Stability Scale VERBS ADVERBS
ADJECTIVES
NOUNS
In short, it seems that we are caught in a dilemma here. It cannot be denied that Time Stability, as an explanation, is fruitful for a major part of the Verbalization Hierarchy, but it does not predict the correct position on the hierarchy as far as adverbs are concerned. Obviously, one way to excape from this dilemma is to abandon Time Stability as an explanation for the facts embodied in the Verbalization Hierarchy. Another approach would be to maintain Time Stability as an over-all explanatory principle, and to attribute the deviant behaviour of adverbs to the influence of a second, intervening, factor. It is this latter approach which I want to pursue here. In my opinion, the fact that adverbs form a category which does not behave in the way predicted by the Time Stability Scale can be attributed to a second principle, which, in the case of adverbs only, overrides the requirements posed by Time Stability. This principle, which I have called the One Word Principle of Verbalization, may be formulated as follows: (25)
The One Word Principle of Verbalization In order for a semantic function to be encoded by the verbal strategy, a necessary condition is that the class of lexical elements which are used to encode that function should (at least in a large majority) consist of single phonological words. 3
This principle thus embodies a lexical-phonological constraint on the possibilities of verbalization: even if a certain semantic function in a given language would be eligible for verbalization on the grounds of time stability, the use of the verbal strategy for that function will be blocked if the class of associated lexical elements does not, in general, consist of single words. In my opinion, adoption of this second principle will enable us to account for the fact that it is the class of adverbials which constitutes a deviation of the
196
Leon S tasseti
otherwise generally applicable principle of time stability, as formulated in (23). In presumably all natural languages, verbs, adjectives and nouns form classes which, in an overwhelming majority, consist of single phonological words: the states of affairs denoted by the members of these classes typically lend themselves to a "labeling" by means of single-word items. With adverbs, however, the situation is often quite different. In most languages, adverbials form a class which is very hard to generalize in terms of lexical structure, even if we restrict ourselves to locative adverbials. The typical situation is that the class of locative adverbials in a language comprises only a very limited number of single-word items; most often, these singleword locations are confined to a small, closed set of demonstrative or pronominal locative adverbs such as here, there, where and somewhere. In addition to these items, the largest part of the class of locative adverbials is typically made up of phrases which consist of more than one phonological word; adpositional phrases like in the house are a very common manifestation of this class. In other words: opposed to the other major predicate classes, adverbials usually cannot be counted upon to consist of a single word, at least not in a majority of the languages of the world. Hence, adverbials often fail, as a class, to meet the requirement posed by the One Word Principle of Verbalization, and they are therefore seldom eligible for encoding by the verbal strategy. It should be added here that the introduction of the One Word Principle is not just a theoretical move, motivated by the desire to eliminate a recalcitrant exception to an otherwise successful generalization. In at least some of the languages in the sample we can find direct evidence for the claim that the One Word Principle is a realistic factor in the choice of encoding strategies for locational main predicates. Turkish is a case in point. As we saw in sentence (7d), which I repeat here as sentence (26 a), at least some Turkish locative adverbials can, when employed predicatively, be treated by the same strategy as intransitive verbs. It should not be thought, however, that just any locational predicate can be handled in this way in Turkish. Sentence (26b) shows a case where the locative adverbial predicate, i.e., the phrase okulun dipnda 'outside the school' cannot be conjugated as an intransitive verb; instead, a suppletive verb, chosen from a set of locative verbs, is required here. (26) a.
Turkish (Lewis, 1967) Ev — de — sin house — in — 2SG.PRES 'You are at home.' (98)
A hierarchy of main predicate
b.
encoding
197
Taksi okul — un dipnda bulun — mak — ta — taxi school — GEN outside be — INF — i n dir 3 SG.PRES 'The taxi is outside the school.' (91)
Now, the difference between the locational predicates in the sentences (26 a) and (26 b) is the following. In (26 a) the locational predicate is a phrase which consists of the noun ev 'house' and the nominal case-suffix -de 'in'; declined nominal forms such as these are considered single phonological words in Turkish. In contrast, the locative predicate in (26 b) is considerably more complex. It is a phrase which contains the noun okul 'school', which is marked for genitive case; this declined nominal form is followed by the postposition dipnda 'outside', which, at least in its etymology, must itself be considered to be a declined case form of the noun dipn 'the outside'. In Turkish, adverbial phrases in which the "postposition" is in fact a declined nominal form are not regarded as consisting of one single phonological word. It will be clear that the facts in (26 a — b) can be explained immediately once we assume that the One Word Principle is a decisive factor in the encoding of main predicates in Turkish. As a second, somewhat similar, case we can point to the facts in Chemehuevi. In this Uto-Aztecan language, the verbal strategy consists of a marking of predicates for aspect and tense, by means of a string of suffixes; there is no subject agreement in the language. As examples (27 b) and (27 c) will show, this verbal strategy is also available for locational predicates, provided that they consist of case-marked forms of pronominal items: such forms constitute single-word items in Chemehuevi. If, however, the locative predicate consists of a nominal form marked by a case suffix or postposition, the predicate no longer counts as a single phonological word. As a result, the verbal strategy can no longer be maintained in such a case (cf. the unacceptability of (27 e)). Predicatively used adpositional phrases that have a nominal complement cannot be marked for aspect or tense, but require the presence of a suppletive locative verb, which takes over the tense-aspect marking. (27) a.
b.
Chemehuevi (Press, 1975) John minimi — j J. dance - PRES 'John dances.' (169) Nini tua — η iva — ni?i — j my son - my here - CONTIN - PRES 'My son is here.' (113)
198
Leon Stassen c.
Τ imp
ζ
— υαα
— ni?i
—
rock this - on - CONTIN 'The rock is on this.' (140) d.
Huu
wibi
— vajiwi
uni
j
PRES
— kai
—
vi
arrow knife — beside be — RESULT — PAST 'The arrow was beside the knife.' (143) e.
*Huu
wihi
— vajiwi
arrow knife -
—
beside -
vi
PAST (143)
6. Conclusion In this paper, I have proposed a hierarchical ordering between various classes of intransitive main predicates, with respect to their eligibility for encoding by verbal strategies. Furthermore, I have argued that this Verbalization Hierarchy can be explained on the basis of the (semantically motivated) scale of Time Stability. Especially in the case of locative main predicates, however, this semantic principle is blocked by the requirements posed by the One Word Principle, which has a lexical-phonological motivation. Thus, if my analysis is correct, we must conclude that intransitive main predicate encoding is a grammatical process which is determined by the interaction of two parameters that must be situated in different components or modules of the grammar. As a final remark, I should point out once more that the analysis presented here constitutes only a very global approach to the intricate problem of intransitive main predicate encoding in natural language. Fascinating questions that might be asked in this connection, such as the motivation for zerostrategies, or the combinatory possibilities of various verbal and non-verbal strategies within a language, have hardly been touched upon. Like practically every other matter of scientific interest, these questions will have to await further research.
Appendix The research reported in this paper has been based on data from the following languages: Abkhaz, Acehnese, Aghu, Agul, Akan-Fante, Alabama, Alamblak, Albanian, Amele, Andoke, Apalai, Arabic (Cairene), Aranda, Archi, Arme-
A hierarchy of main predicate encoding
199
nian, Asmat, Avar, Babungo, Banda, Banggai, Barea, Bari, Basque, Bedauye, Bilin, Biloxi, Birom, Bororo, Breton, Bribri, Buli, Burmese, Burushaski, Cambodian, Canela-Kraho, Car, Cebuano, Chacobo, Chamorro, Chatino, Chechen, Chemehuevi, Choctaw, Chontal (Highland), Chuckchi, Cornish, Dafla, Daga, Dinka, Diyari, Dumaki, Dutch, Dyirbal, English, Esseeja, Ewe, Falor, Fijian, Finnish, Gaelic, Georgian (Modern), Gidabal, Goajiro, Gothic, Grebo, Greek (Modern), Guajajara, Guanano, Gumbainggir, Gurenne, Haida, Haitian Creole, Hebrew (Modern), Hindi, Hiroi-Lamgang, Hixkaryana, Hungarian, Igbo, Indonesian, Irish (Modern), Irish (Old), Jabem, Japanese, Kabyl, Kalispel, Kapauku, Kâte, Keiese, Khalka, Khasi, Kiowa, Kobon, Korean, Kpelle, Kunama, Kwaio, Kwakiutl, Ladakhi, Lak, Lamutic, Latin, Latvian, Lepcha, Lezghian, Lillooet, Lithuanian, Loniu, Luiseno, Lushai, Maasai, Malagasy, Mamvu, Manam, Manchu, Mandarin, Mangarayi, Manipuri, Manobo, Marind, Mende, Miskito, Mixtee, Moore, Mordvin, Mutsun, Nama, Nepali, Ngalakan, Nkore-Kiga, Nubian, Nuer, Ona-Selknam, Ormuri, Ostyak, Otomi, Paez, Palauan, Papiamento, Persian (Modern), Persian (Avestan), Pipil, Piraha, Piro, Pitjantjatjara, Quechua (Imbabura), Quileute, Romany, Rumanian, Russian, Samoan, Sango, Sedang, Seneca, Senufo, Serbo-Croat, Serrano, Shona, Shuswap, Sinhalese, Songhai, Spanish, Squamish, Swahili, Swedish, Tagalog, Tajik, Tamil, Temne, Tera, Thado, Thai, Tibetan (Classical), Tigak, Tiwi, Tocharic (West), Tolai, Tondano, Toradja, Tubu, Turkish, Tuscarora, Tzutujil, Urubu-Kaapor, Usan, Vai, Vietnamese, Wappo, Waropen, Waskia, Welsh, Wichita, Wolof, Yaqui, Yavapai, Yoruba, Yukaghir, Yurak, Yurok, Zapotee, Zoque. For reasons of space, a full exposition of the relevant data in these languages is impossible. Readers who are interested in the data of those languages that are not explicitly discussed in the paper are invited to contact the author.
Notes 1. Throughout this paper, the numbers enclosed in brackets after the last line of an example refer to the page number in the work where the example can be found. 2. Latin and Russian also have a zero-strategy for their non-verbal predicative constructions. In both languages, this zero-strategy is restricted to indicative sentences with present tense. In Latin, zero is particularly common with third person subjects, whereas Russian does not normally use a copula in any person in the present indicative. 3. For a discussion of the notion "phonological word" see, for instance, Lyons (1968: 195 — 206).
200
Leon Stassen
References Bendor-Samuel, David 1972 Hierarchical structures in Guajajara. Norman, Oklahoma: Summer Institute of Linguistics. Bursill-Hall, G. L. 1971 Speculative grammars of the Middle Ages: the doctrine of the Partes Orationis of the Modistae. The Hague: Mouton. Campbell, Lyle 1985 The Pipil language of El Salvador. Berlin/New York/Amsterdam: Mouton. Cole, Peter 1982 Imbabura Quechua. Amsterdam: North Holland. De Angulo, J. —L. S. Freeland 1935 "The Zapotecan linguistic group", International Journal of American Linguistics 8: 1 - 3 8 . Derbyshire, Desmond C. —Geoffrey K. Pullum (eds.) 1986 Handbook of Amazonian languages. Vol. 1. Berlin: Mouton de Gruyter. Dorsey, James Owen—John R. Swanton 1912 A dictionary of the Biloxi and Ofo languages. Washington, D. C.: Government Printing Office. Fennell, John L. I. 1961 The Penguin Russian Course. Harmondsworth, Middlesex: Penguin. Graham, A. C. 1965 "'Being' in linguistics and philosophy: a preliminary inquiry", Foundations of Language 1: 223—232. Givón, Talmy 1984 Syntax. A functional-typological introduction. Volume 1. Amsterdam/Philadelphia: John Benjamins. Harrison, C. H. 1986 "Verb prominence, verb initialness, ergativity and typological disharmony in Guajajara", in: Desmond C. Derbyshire — Geoffrey K. Pullum (eds.), 407 — 439. Holmer, Nils 1949 "Goajiro (Arawak) III: verbs and associated morphemes", International Journal of American Linguistics 15: 145 — 157. Lees, Robert B. 1972 "The Turkish copula", in: John M. W. Verhaar (ed.), 6 4 - 7 3 . Lewis, G. L. 1967 Turkish Grammar. Oxford: Clarendon Press. Lorrain, J. H. —F. W. Savidge 1898 A grammar and dictionary of the Lushai language (Dulien dialect). Aizawl: Tribal Research Institute. Lyons, John 1968 Introduction to theoretical linguistics. Cambridge: Cambridge University Press. Matteson, Esther 1965 The Piro (Arawakan) language. Berkeley/Los Angeles: University of California Press.
A hierarchy of main predicate
encoding
201
Osborne, C. R. 1974 The Tiwi language. Canberra: The Australian National University. Pickett, Velma B. 1960 The grammatical hierarchy of Isthmus Zapotee. Ph.D. Dissertation, University of Michigan. Press, Margaret L. 1974 A grammar of Chemehuevi. Ph.D. Dissertation, UCLA. Radin, Paul 1930 "A preliminary sketch of the Zapotee language", Language 6: 64—85. Raptschinsky, B. 1948 Russische Spraakkunst. II. Groningen: Noordhoff. Troll, W. 1975 Lateinische Grammatik. Frankfurt am Main: Moritz Diesterweg. Verhaar, John M. W. (ed.) 1972 The verb 'be' and its synonyms. Part 5. Dordrecht: Reidel. Wetzer, Harrie M. " 'Nouny' and 'verby' adjectivals: a typology of predicative adjectival constructions", (this volume). Wiedemann, F. J. 1865 Grammatik der ersa-mordwinischen Sprache. Berlin/Riga: Eggers.
On explaining double object constructions Lieve Van de Walle
1. Introduction1 The bulk of literature on double object constructions and related areas of research (such as dative shifting) amply demonstrates the continued attraction this construction type exerts on linguists. However, in spite of the crusade of many a language theorist and/or typologist for a unified, eclectic approach aiming at the integration of both formal and functional aspects in one theory, in most articles the issue is approached from only one angle. There is nothing wrong per se with opting for either a formal or a functional orientation; both indubitably yield valuable insights. However, language cannot be conceived of as a unidirectional process — the pragmatic component feeding into the semantic component and the semantic component into the syntactic component, etc. — rather it results from a constant interplay between structure and function. So, one may lose touch with the overall picture if one opts for one particular orientation at the exclusion of the other one. Basically voicing the same idea, Hyman (1984: 68) states that "far from being contradictory of one another, [internal and external explanations] together provide a fruitful avenue for the pursuit of explanations of language universale." He finally concludes that "the totality of languages will be accounted for only by a combination of 'explanations'" (1984: 82) (cf. also Comrie 1984 b: 355; Östman 1987: 155 in this respect). In the remainder of this introductory section, Hyman's distinction between internal and external explanation, which will be used as the main structuring device in this paper, will be briefly expanded upon, followed by an attempt to define "double" object constructions. In the body of the paper (section 2 and 3), the pros and cons of both an analysis in purely syntactic terms and an analysis that goes beyond the realm of the syntactic component will be weighed against each other. As a conclusion, a "symbiotic" approach will be proposed and tested with a sample of Sanskrit data. Of course, this "program", if it were to be scrupulously executed, would be too ambitious for a paper of this format. However, despite its inevitable tentative character, it is hoped that the article will at least provide a better insight into the problem of the double object construction and will demonstrate the necessity of an integrated approach of linguistic phenomena.
204
Lieve Van de Walle
Coming back to Hyman's internal/external dichotomy, it should be noted that it was introduced to avoid the more familiar but overused formal/ functional dichotomy, with which it roughly corresponds. As suggested above, an internal explanation aspires to account for the facts from within the component under discussion at that time, e. g., a syntactic problem will be settled "in terms of the nature of syntax itself." On the contrary, an external explanation will go beyond the borders of the one component: e. g., a syntactic problem will be settled by relating it to other phenomena "outside the realm of syntax" (Hyman 1984: 67). Trying to define "double" 2 object constructions in a more or less satisfactory and unambiguous way seems as hopeless an endeavour as trying to explain them. Since a generally accepted, unbiased definition of even a single object is non-existent — and non-impending for that matter — I will for the present time confine myself to presenting a working definition solely aiming at preventing major misunderstandings. In the course of the discussion different views will be dwelt upon. Supposing that a syntactic function can best be defined in terms of semantic roles, the following views are tenable: a) the prototypical object is the semantic role patient of a two-place predicate (cf. Borg and Comrie 1984: 110) b) a double object construction occurs when two different semantic roles (usually a patient and a recipient) are encoded in an identical way, i. e., not having "syntagmatische Objekt-Differenzierung" (cf. Plank 1987: 37). There may be some "case recoverability" problems (cf. Givón 1984: 154).
2. An internal explanation of "double" object constructions 2.1. Traditional Relational Grammar Before exploring the adequacy of Relational Grammar (henceforth "RG") in dealing with "double" object constructions (henceforth "dO" constructions), my choice of precisely this framework should be briefly motivated. In spite of its inherent limitations and continuous decrease in popularity, basically due to the numerous falsifications of their proposed laws (cf. Dryer 1986: 809; Mallinson-Blake 1981; Newmeyer 1980), RG still has some credit. Thanks to Dryer's reformulating the RG laws into markedness principles
On double object constructions
205
(1986) and his adding a new dimension to the syntactic relations (cf. below), RG seems to provide a good instrument to analyse the complex dO construction in purely syntactic terms. To guide the reader towards a correct understanding of the issues discussed below, let us first give a few instances of dO (the examples under (a)) and alternative constructions (the examples under (b)). (1) a.
b.
(2) a.
b.
German (cf. Plank 1987: 41, 46) Frl. Schmidt hörtjfragt die Mädchen ΝΟΜ.-Ms. Schmidt interrogates ACC.-the girls den Katechismus ab. ACC.-the catechism 'Ms. Schmidt interrogates the girls about the catechism.' Frl. Schmidt hört ¡fragt den Mädchen ΝΟΜ.-Ms. Schmidt interrogates DAT.-the girls den Katechismus ab. ACC.-the catechism 'Ms. Schmidt interrogates the girls about the catechism.' Sanskrit (AV 12.4.48; Dkc 116.9) 3 vasäm ced enam jäcejuh\ ACC.-cow if ACC.-him OPT.-they should request 'If they should request the cow from him.' Dkc 116.9 .. .pauramukhjebhyah smasänaraksäm ABL.-citizens ACC.-watch over the cemetery ajäcata— alabhata ca¡ IMPERF.-he asked IMPERF.-he obtained and 'He asked from the citizens the watch over the cemetery and he got it.'
Case languages are very clear in respect to dO constructions, both the patient and the "partner" role 4 are encoded in the accusative, whereas in the alternative construction one of the roles is marked differently. One of the basic tenets of RG is the SUL (Stratal Uniqueness Law), which says in essence that there is only one direct object (one subject or one indirect object for all that matter) at a given stratum. At first sight the dO constructions seem to violate and, hence, falsify this law (cf., for instance, Hock 1985: 67). However before jumping to conclusions, let me try to roughly reconstruct the line of thought the adepts of the original RG model, as proposed by Postal and Perlmutter (1983), most probably follow in respect to these constructions. First and foremost, the alternative constructions (the b-sen-
206
Lieve Van de Walle
tences), in which the patient is unquestionably the direct object, are conceived of as basic and the dO constructions as derived from them. In this derivation process then, the original indirect object ("IO") or oblique object ("OO") is promoted to direct object ("DO") and the original direct object, losing its status, is turned into a chômeur, which implies its non-qualification for advancement to subject. To summarize, based on the assumption that there is an alternative construction, the dO construction is seen as an optional derivation. Although at first sight, this all sounds fairly reasonable, there are quite a few problems and pitfalls in the argumentation. First, accepting the fact that the DO, by analogy with the subject (henceforth "SU") (cf. Keenan: 1976), can be characterized as a set of properties (cf. Borg—Comrie 1984: 123), one can normally expect that the potential candidates for the DO slot are characterized by at least some of these properties. Let me refer to German once more (cf. Plank 1987: 44—45). (3) a.
b.
c. d. e. f.
Die von Frl. Schmidt (Ìden Mädchen) abgefragten The by Ms. Schmidt DAT.-the girls interrogated Vokabeln werden von Frl. Müller nicht mehr abgefragt. words are by Ms. Müller not anymore interrogated 'Ms. Müller does not interrogate (the girls) anymore about the words about which they were interrogated by Ms. Schmidt.' * Die von Frl. Schmidt (die Vokabeln) abgefragten Mädchen the by Ms. Schmidt the words interrogated girls 'The girls interrogated by Ms. Schmidt.' die Lehre des Tocharischen (?? für die ... Studenten) 'the study of Tocharian (?? for the ... students)' *die Lehre der fortgeschrittenen Studenten 'the study of the advanced students' die Abfragung der Vokabeln 'the questioning of the words' die Abfragung der Mädchen 'the interrogation of the girls'
In cases a) through d) the original IO and OO's do not pass the tests which are typically used to identify the DO: the perfective participle of the ditransitive abfragen (cf. (3a) and (3b)) can only be attributed to its corresponding DO, Vokabeln, and not to the original IO, den Mädchen-, the nominalisation of the ditransitive lehren only allows the original DO, das Tocharische, in the genitive and not the original IO, den fortgeschrittenen Studenten. However, as is illustrated in (e) and (f), the latter restriction does not apply to ditransitives (such as abfragen) which can take both a patient and a recipient DO when
On double object constructions
207
used monotransitively (both Frl. Schmidt fragt die Vokabeln ab and Frl. Schmidt fragt die Mädchen ab are possible — cf. Plank (1987: 45); and Hock (1985: 61 — 62) for similar examples in Sanskrit). All things considered, I agree with Dryer (1986: 834) that the "chômeur-status" attributed to the original DOs, suggesting that they are "non-terms", is rather unfortunate, since "they occupy a higher position on the relational hierarchy than non-terms". Second, one of the corollaries of the RG analysis, as presented above, is that an IO/OO, eligible for advancement to DO, is consecutively eligible for advancement to SU. Although this is indubitably a valid guideline in a lot of languages, counter-examples are easy to find: (4)
German (Plank 1987: 42) * Die Mädchen werden von Frl. Schmidt NOM.-the girls are by Ms. Schmidt den Katechismus abgehört. ACC.-the catechism interrogated. 'The girls are questioned about the catechism by Ms. Schmidt.'
Third, formulating a hypothesis — albeit a tentative one — to delimit the domain within which dO constructions typically occur, seems to be outside the scope of the theory. As a consequence, one has to settle for ad hoc explanations. Given below is a selection of facts to be accounted for in a hypothesis. a. Only a limited set of ditransitives, typically conceptualizing a transfer of either ideas (e.g., verbs of asking; ordering; teaching) or material goods (e. g., verbs of giving; taking; stealing), actually occurs with dO constructions. b. dO constructions are optional (cf. the examples given in (1) and (2)) in some languages but obligatory in others. It should be noted here that an obligatory advancement rule, which is necessary to cope with the latter group, is not commensurable with the "Transparency Law", which basically says that "Grammatical relations referred to by specific rules in a language should be transparent from surface evidence in the language" (Dryer 1986: 809). c. Within that limited set of ditransitives some verbs favor the dO construction (cf. ji 'to win something from someone'; mus 'to steal something from someone' in Sanskrit), whereas others have a decided predilection for the alternative construction (cf. the Sanskrit verb vad 'to speak', occasionally used in the sense of 'teach'). d. In the alternative constructions (cf. (5b) and (5d)), depending on the verb involved, the encoding of either the patient (5b) or the "partner" role (5d) varies, as illustrated in the following examples from Ancient Greek.
208
Lieve Van de Walle
(5) a.
b.
c.
d.
Od.IX.364 Kùklòps, eirôtàis m' önoma klutón. VOC. PRES.-you ask ACC.-me ACC.-name famous 'Cyclops, you ask me a famous name.' Hdt.1.32 epeirötäis me antbrôpëiôn prëxmâtôn péri. PRES.-you ask ACC.-me GEN.-human affairs about 'You ask me about human affairs.' I1.XXII.295 eiteé min dóru makrón. asked ACC.-him ACC.-big spear 'He asked him for a big spear.' X.An.1.3.16 egemóna aiteîn parà toutou. ACC.-leader INF-to ask from Gen.-him 'To ask a leader from him.'
2.2. Relational Grammar as adapted by Dryer (1986) As I suggested in the introduction, Dryer's approach is a promising improvement on basically two fronts. First, by substituting the term markedness principle for the objectionable notion law, Dryer emphasizes that a linguistic theory should not be primarily concerned with the "outer limits of the set of possible human languages" but should rather concentrate on the typical, "normal" properties. Consequently and importantly, a small number of counter-examples cannot endanger any longer the whole theoretical construct (Dryer 1986: 808 — 813; cf. especially his example of Ojibwa). A second major point of progress lies in the field of grammatical relations itself. By analogy with the generally accepted distinction between Absolutive — Ergative and Subject— Object oriented languages, Dryer proposes a distinction between what he calls Primary Object—Secondary Object (henceforth abbreviated as "PO" and "SO") and Direct Object—Indirect Object languages. Four sets of hierarchically ordered pairs are to be discriminated (Dryer 1986: 818), of which the first two account for the assignment of the first syntactic relation and the last two for the assignment of the second syntactic relation: (a) SU > OBJ (b) ABS > ERG
(c) DO > IO (d) PO > SO
On double object constructions
209
The pairs (a) and (c) are the ones with which we are most familiar. Subjects in both transitive and intransitive sentences and direct objects in both monoand di transi ti ve sentences are encoded in an identical way, whereas the objects in transitive sentences and the indirect objects in ditransitive sentences are encoded differently. However, the pairs (b) and (d) are characterized by another distribution. Just as in an Absolutive oriented language, there is a close bond between the subject of an intransitive and the object of a transitive clause, in a Primary Object oriented language, there is a close bond between the DO in a monotransitive and the IO in a ditransitive clause. 5 Of course, the very fact that it makes a nice analogy does not prove that the PO/SO dichotomy is a useful and, above all, a necessary distinction. The major argument Dryer cited — and profusively illustrated — is that a lot of languages "treat IO's like DO's, and assign them grammatical properties that place them above DO's of ditransitive clauses" (Dryer 1986: 819). Furthermore, Dryer sensibly argues that even the English ditransitive construction is more efficiently dealt with when adhering to the PO analysis. To uncover the line of thought underlying this point of view, the classical dative analysis (cf. (6)) will be confronted with the alternative one, i. e., the antidative analysis (cf. (7)) (cf. Dryer 1986: 821 for a more detailed discussion). Note that only the final stratum is given. (6) a. b. (7) a. b.
Dave gave a cookie to me. SU DO IO Dave gave me a cookie. SU DO chômeur Dave gave me a cookie. SU PO s o Dave gave a cookie to me. SU PO chômeur
These are the main points of divergence between the two approaches. First, both analyses treat a different sentence as basic ((6a) and (7a), respectively). Second, in (7b) the original SO (a cookie) is advanced to PO, and the original PO ends up as a chômeur instead of as a final IO, as would be the case under the traditional dative analysis. Thus, according to Dryer, the IO eventually has a status similar to the by-phrase in a passive sentence. The biggest asset of the alternative approach is undoubtedly that, using the presence (nonterm) or absence (term) of prepositions as basic criterion, the discrimination between terms and non-terms is considerably facilitated. Under the dative analysis, the findings are much more disparate: SU, DO and DO-chômeurs appear without preposition, whereas non-primaries other than DO-chômeurs
210
Lieve Van de Walle
occur with preposition. It should also be mentioned that the antidative analysis is an excellent tool to check the valency of the verb. Compare in this respect the respective configurations under (6b) and (7a); the latter immediately reveals that the verb is used as a ditransitive, as opposed to the former. Having roughly sketched the major advantages of Dryer's adaptations to RG, the implications for the analysis of the dO construction should now be approached. In this respect, basically two questions are to be addressed. First, given the typical/atypical dichotomy, as proposed in the first paragraph in this section, are the dO constructions to be situated in the former or the latter zone? Second, does the PO/SO analysis offer a viable way out of the problems discussed in section 2.1.? The first question calls for a terse but clear answer. Taking a crosslinguistic perspective, I argue that the dO construction should not be conceived of as a syntactic anomalous case, in spite of its rather borderline status in most Indo-European languages. This view is motivated not only by the fact that we have at our disposal crosslinguistic evidence of languages in which identically marked objects are the rule rather than the exception (cf. Plank 1987: 39), but also by the fact that the dO constructions in Indo-European languages (and perhaps others), no matter how limited in number, are confined to a small set of verbs (cf. above). If we were to treat the dO constructions as purely anomalous, one would have to consider the crosslinguistic convergences as sheer coincidental. Before responding to the second question, Dryer's notion of Split Objectivity should be introduced. Obviously, this term is reminiscent of Split Ergativity, hence, it can be analogously defined: in languages characterized by Split Objectivity some rules refer to the grammatical relations PO and SO, while others refer to the grammatical relations DO and IO. As far as I can see, the concept of Split Objectivity may be very fruitful indeed in explaining the syntactic aspect of the dO constructions. Assuming that the dO constructions, as opposed to the other construction types, operate on a PO/SO basis, at least the problem of "double" objects is eliminated. The recipient (or whatever semantic role functions as the "partner") and the patient do not have to compete any more for the same syntactic status, because one is now the PO and the other the SO. Moreover, according to this Split Objectivity analysis, the typical object properties retained by the patient (DO-chômeur in the traditional analysis) are adequately handled by giving both the PO and the SO fcrw-status. In section 3, some Sanskrit examples will illustrate the feasibility of this approach. Despite its strong advantages, a purely syntactic analysis always has a limited scope of use because it does not explain the motivation of the
On double object constructions
211
construction type studied. However, it is hoped that the foregoing discussion, no matter how sketchy and incomplete, has demonstrated that the Split Objectivity analysis at least contributes to a thorough insight in the problems at hand.
3. An external explanation of "double" object constructions 3.1. A Functional Grammar proposal In Functional Grammar (FG) a distinction is made between three components hosting pragmatic, semantic and syntactic functions, respectively. In FG language is basically treated as a unidirectional process: first, the pragmatic functions are assigned, followed successively by the semantic and the syntactic ones. To deal with subject and object assignment, which are situated in the semantic component, FG appeals to the Semantic Function Hierarchy (SFH) (AG > GO > REC > BEN > INSTR > LOC > TEMP; cf. Dik 1978: 70) in combination with the Continuity Hypothesis (Dik 1978: 76), grosso modo a translation into semantic terms of Johnson's Advancement to Subject Chaining Constraint (ASCC) (Johnson 1977: 158): "For any language, if Subj or Obj function can be assigned to some semantic function S¡, then Subj or Obj can be assigned to any semantic function S¡, such that S¡ precedes S¡ in SFH." According to FG, syntactic function assignment 6 is a matter of choosing a particular perspective: subject assignment is associated with the primary and object assignment with the secondary 7 perspective. Since, as indicated by the SFH, the agent usually is more central to the predication than the goal, and the goal is more central than the recipient, and so forth, the language-user will preferably describe the state of affairs from the viewpoint of the agent ( > GO...) and, hence, the agent is a better candidate for subjecthood. The same line of thinking is pursued to account for object assignment, except, of course, that the agent is not an eligible candidate for the object slot. It is also worth noting that languages vary as to the exact cut-off point for syntactic function assignment (for further detail cf. Dik 1978; 1989). After this "theoretical" intermezzo, we should focus again on the main topic under discussion, the dO construction.
212
Lieve Van de Walle
The Workgroup on Functional Grammar (1981: 207) unfolded a proposal to deal with teaching verbs and distinguished between two underlying predicate-frames (cf. also Van de Walle 1989 in this respect): - the GIVING model: teach John (AG) maths (GO) student (REC) - the OPERATING model: teach John (AG) student (GO) maths (complement). According to the GIVING model, the "GO" function is bestowed upon the knowledge transferred (maths), while, according to the OPERATING model, the GO function is attributed to the transferee (student). Taking this distinction as a point of departure, one can argue that languages in which dO constructions and/or advancement of the IO to DO do not occur operate according to the GIVING model. In this set of languages the information or goods transferred is/are more nuclear than the transferee and is/are attributed subject or object status. The SFH is not violated since the FG analysis refers to information or goods transferred as the semantic functional goal and goal precedes recipient on the hierarchy. In contrast, languages in which dO constructions and/or obligatory advancement of the IO are the rule rather than the exception (for instance, Georgian, cf. Plank 1987: 39), the OPERATING model should be considered basic. Here, the transferee occupies a more central position and consequently has prior access to a syntactic function. Again, the SFH and the related hypothesis about syntactic function assignment is correct, since the FG analysis analyses the transferee as a goal and the goal precedes the complement on the hierarchy. A third group of languages, which allows both dO and alternative constructions, can be argued to boast two possible underlying predicate-frames. Indubitably, this approach reminds us of the Objectivity Split, proposed in the foregoing section. However, in sharp contrast with my readiness to adopt this purely syntaxoriented analysis, the FG hypothesis provokes some criticism, most of it finding its origin in the FG decision to situate the problem in the semantic component only. For example, think of a language in which the verb ask usually occurs with a dO construction and, thus, is — at least following the FG analysis — organized according to the OPERATING model. One of the corollaries of the latter claim is that me in He asks me a question is thought of as a GO, in other words me has the same semantic case role as d book in He gave a book to me, which seems quite counter-intuitive. Whereas I can easily go along with the statement that me and a book share syntactic features, I have difficulties with the idea that both are semantically related. Another counter-argument corroborated by the present discussion is that the semantic functions lose their inherent value when one subscribes to an analysis like this. Difficult as it already is to define the semantic functions in
On double object constructions
213
an unambiguous way, why complicate things further by turning REC into GO, etc.?
3.2. Givón: Direct Object and Dative Shifting Unlike Dik, who confines syntactic function assignment to the semantic component, Givón interprets it as a conflation of both semantic and pragmatic features. Like Dik, Givón is convinced of the predictive power of a hierarchy. However, their eligibility for topic status being the basic criterion for ranging the semantic roles, their order on the hierarchy is significantly different: AG > DAT/BEN > ACC > LOC > INSTR > others (cf. Givón 1984: 151). To summarize, SU and DO are conceived of as pragmatic case roles, grammaticalizing the primary and secondary clausal topic, respectively (for a detailed discussion of topic cf. Givón 1983). Also note that the DAT/BEN precedes the ACC in the hierarchy, i. e., from a cross-linguistic perspective, the DAT/ BEN of a ditransitive construction in a language that syntacticizes DO's has better access to the DO slot since it is a better candidate for topic-status. Of course, this DO concept is not completely at odds with the traditional, almost dogmatic view (cf. the introduction), which defines the patient as prototypical object and treats dative shifting as an "optional", "stylistic" device. Givón does agree that the ACC, due to its frequency of occurrence, has a higher potential in comparison with the DAT/BEN to be interpreted as the "cardinal" prototype of DO (1984: 176). He even provides a couple of arguments underscoring this fact: the ACC is typically morphologically less marked (a typical feature for "pragmatic" cases) and it is seldom ellipted in the sentence as opposed to the DAT/BEN. Referring to statistically processed, concrete language data, however, Givón (1984: 155 — 156) takes the edge off the "generally accepted surmise" that DAT/BEN is only rarely eligible for the DO slot: "A careful study of frequency distributions in live text reveals that in such unsolicited discourse the overwhelming bulk of DAT/ BEN objects appear as DO rather than IO", because "whenever the ACC and DAT/BEN objects are pitted against each other allowing both, theoretically, to become DO, the DAT/BEN object wins in 84% of the cases, the ACC object only in 16%". Interestingly, Givón's topic hierarchy, as will have appeared from the foregoing discussion, is largely compatible with the PO/ SO dichotomy. His hierarchy recognizes that a lot of languages treat DAT/ BEN (PO) as higher than ACC (SO), a fact which is reflected in their syntax. Let us concentrate now on the implications of this approach for the dO constructions. Most likely Givón would describe them as constructions in
214
Lieve Van de Walle
which the DAT/BEN occurs in the DO slot and assumes the usually unmarked morphemic status of the ACC at the expense of its own characteristic case marking. Thus, there is no morphological marking whatsoever to disambiguate 8 between the semantic case roles. As is amply demonstrated by the multilingual set of examples in his article and the basic orientation of his hypothesis, Givón recognizes the "topic-potential" of the DAT/BEN and its importance for the organization of the information structure in the discourse. Definitely he does not consider dO constructions as either "accidental" or "anomalous". Without wanting to obfuscate the assets of this approach, let me quickly point out what I perceive as a drawback. The problem basically concerns passivization of ditransitives. On the basis of Givón's philosophy, the most topical element in the sentence and in the larger discourse is most eligible for the subject function. Thus, one would expect that at the elimination of the AG for whatever reason, the role immediately following the AG on the topic hierarchy takes over control. However, even though he cites "Maintopic assignment" (i. e., "an argument other than the subject/agent of the active becomes the main topic of the passive clause, by whatever means" (Givón 1984: 164)) as one of the "three distinct though related functional domains" of passivization (the other ones being detransitivization and impersonalization), Givón himself signals that in the majority of the cases, the passive is restricted primarily to the ACC, which ranges only third on the hierarchy. Briefly, in an active clause the DAT/BEN wins the competition for the DO slot, whereas in the passive turn the ACC wins, implying that in the former case the DAT/BEN is more topical, whereas in the latter case this feature characterizes the ACC. This leaves a question unanswered: What motivates this divergent behaviour of active (bitransitive) and passive (monotransitive) clauses as to topicality?
4. Towards a symbiotic approach: evidence from Sanskrit I will now try to combine the findings of the various sections in one global synthesis, using examples from Sanskrit as a "control apparatus". First, to facilitate the interpretation, a brief outline will be given of the Sanskrit dO construction.
On double object constructions
215
4.1. The Sanskrit dO construction in a nutshell On the basis of the semantic roles they generally trigger, three clusters of ditransitives emerge: first, the verbs that denote a transfer of ideas and are characterized by the semantic roles patient (henceforth "PAT") and + recipient 9 (henceforth " + REC"), second, the verbs that denote a transfer of goods and are characterized by the semantic roles PAT and — REC, and, third, the causative constructions 10 . The first cluster comprises the categories of asking; speaking; and teaching verbs. The second cluster combines a set of heterogeneous verbal roots, which can be conceived of as anomalous cases since they only occasionally occur. Often cited are: ji 'to win something from someone either by gambling or by fighting'; mus 'to steal something from someone'; and finally duh 'to milk' (i. e., to extract something from someone). For convenience's sake, dadnay 'to punish someone with something' or 'to inflict something on somebody' has been hosted in this subgroup, although there is not a —REC but an INSTR involved in the verbal process. When categorizing one generally concludes that the borderlines are somewhat blurred. Thus, two clarifying points need to be made. First, the conceptual field of asking is very extensive, ranging from transfer of ideas ( + REC) to transfer of goods (—REC). Consider the examples under 8. (8) a.
b.
Cu V 3.5 panca mä räjanyabandhuh prasnän apräksit, ... five ACC.-me ΝΟΜ.-military-man ACC.-question AOR.-ask Ά man from the military has asked me 5 questions.' M 2.50 mätaram... bhikseta bhiksäm / ACC.-mother OPT.-should beg ACC.-alms 'He should beg alms from his mother.' (Cf. also (2a) and (2b))
Presenting the field as a continuum with the transfer of ideas ( + REC) at the left pole and the transfer of goods ( — REC) at the right pole 11 , prach (cf. (8a)) lexicalizes the conceptual areas around the left pole and yäc and bhiks (cf. (8b); (2a) and (2b)) lexicalize the conceptual areas around the right pole. Since it is impossible to delimit exactly the meaning of a lexical item, ambivalence inevitably occurs in the middle section (cf. Verschueren 1985, chapter 1). As a result, one may find instances of prach used in a more concrete sense and instances of yäc and bhiks used in a more abstract sense. The syntactic difference between both groups is closely related to their
216
Lieve Van de Walle
different semantic content. Not only the lexicon but also the syntactic structure gives us a glimpse of the organization of the conceptual field. Second, it is noteworthy that the verbs of teaching are somehow semantically related to the verbs of speaking (compare (9a) and (9b)) and that the borderline between the verbs of teaching and the verbs of ordering (compare (9c) and (9d)) is fairly blurred. (9) a.
b.
c.
d.
Cu IV 2.1 anu mä etäm bhagavah devatäm DAT./ACC.-me ACC.-this VOC.-lord ACC.-God sädhi, yäm devatäm upässa itiI IMP.-teach REL.-which God PERF. M.-you believe 'Teach me, Lord, the God you believe in.' Ku I 15 lokädimagnim tarn uväca tasmai, ... ACC.-fire sacrifice the PERF.-he told DAT.-him 'He told/taught him the fire-sacrifice, ...'. Dkc 121.9 ädiksam ca tarn sakhe .../ AOR.-I ordered and ACC.-him VOC.-friend 'And I ordered/instructed him: "Friend, ...".' Dkc 120.17 ädistas ca tena= ayam .../ gaccha .../ VA-NOM.-was ordered and INSTR.-him NOM.-he IMP.-go 'And he was ordered by him: "Go ...".'
As a result, the verbs of teaching seem to be squeezed between two big conceptual areas, they share semantic features with both the verbs of speaking and the verbs of ordering, which is neither surprising nor exceptional from a cross-linguistic point of view. For example, the verb "to instruct" in English can be used in three different senses: teaching (The teacher instructed us in chemistry), informing (The jury instructed the lawyer), and ordering (The president instructed his s t a f f to boycott the commies). The context always determines the interpretation of the verb. If one were to study in detail the lexicalization of this domain, a hierarchy could possibly be established, ranging from verbs that can be used only in commands to verbs that can be used only for transfer of pure knowledge. Obviously a wide spectrum of meaning lies between the two extremes, a semantic fuzziness reflected in the syntactic behavior of the teaching verbs.
On double object constructions
217
4.2. A combination of explanations as a possible solution In order to gain a clear insight into the workings of the dO construction I started with Dryer's approach. Though not claiming to explain how the construction came about or what its underlying motivations are, his hypothesis does claim to be able both to cope with the potential structural differences among languages and to present an adequate structural analysis. According to Hock (1985), in Sanskrit the dO constructions, only scarcely appearing in the early language, gradually superseded the alternative constructions in importance. Thus, provided Hock's claim is correct, Sanskrit can be said to have developed from a rather rigid DO/IO system (no dO constructions at all) to a really mild form of Split Objectivity (dO constructions with a limited set of verbs). I talk of a "mild" form of Split Objectivity because the DO/IO distinction still largely outbalances the PO/SO distinction. In other words, only with a limited set of verbs are the DO's of the monotransitive variants syntactically cognate to the IO's of the ditransitive variants and is the use of the terms PO/SO motivated (cf. section 2.2.). Of course, one could argue that with languages like Sanskrit even referring to a split is nonsensical. However, one should remember the endless discussions in the literature concerning the "true" direct object, and the controversy about the term or non-term status of the initial DO (the PAT). When adhering to the Dryer approach, this problem is eliminated because the initial DO is conceived of as a secondary object and, hence, bears term status. Furthermore, as argued above, if one wants to interpret language data from a crosslinguistic perspective, a broad framework is indispensable. Since the exploration of the syntax of a certain construction type is a "means" rather than an "end", there is one logical next move: to unravel the network of indubitably closely intertwined functions underlying the syntactic configuration, in order to ultimately establish the relation between function and form. Thus, one needs a semantic/pragmatic superstructure, complementary to and fully compatible with the Dryer approach. Let me first present a hypothetical reconstruction of the shift from rigid DO/IO to mild PO/SO orientation. In the initial stage, all case roles were perfectly recoverable, the grammar being semantically quite transparent (cf. Plank 1987). However, in the course of time, a number of typical pragmatic features were grammaticalized, such as high involvement or prominence of the REC. In other words, grammar started encroaching into a domain until then reserved for pragmatics only, by encoding explicitly (or semi-explicitly in the case of the dO construction in Sanskrit) what was left implicit so far.
218
Lieve Van de Walle
As it may have appeared from the discussion in section 3, neither Dik's, nor Givón's hierarchy really satisfies the demands; both inadequately deal with the complex network of factors involved in this process of syntactic function assignment. Whereas the "flexibility" of the semantic role "goal" was considered a drawback for the FG proposal, the problem pointed out in respect to Givón's analysis was the seemingly divergent approach to the active and passive turns of the ditransitives. Instead of rejecting these hierarchies tout court, however, I opted for conflating them. In Van de Walle (1990), a paper which stresses that subject assignment in some types of constructions is a syntactic emanation of the prominence or rather comparatively high degree of agentivity of the REC, a prototype of a "merged" hierarchy was presented. The general idea was that not one but several levels had to be posited. I argued that for the bulk of predicates the deep level and the surface level overlap, the surface syntax disclosing the underlying hierarchy (AG > PAT > REC), but that for a minority of predicates this very process did not apply. Thus, to be able to account for these happy few, which give the recipients rather than the patients prior access to syntactic functions, an intermediate level was posited to host the shift from AG > PAT > REC to AG > REC > PAT. However, given the cross-linguistic situation, dative shifting does not deserve the epithet "exceptional structure", nor does the multilevel approach comply with the neutrality requirement for a theory because it considers the order AG > PAT > REC as basic. Therefore, yet another suggestion is made below. The difference between the previously presented hierarchies and the hierarchy which will be presented below is that the former ones principally focus on the differences among the semantic roles, while the latter also values the convergences among them. In some cases the recipient is very closely involved in the verbal action and displays typical agentive characteristics, whereas in other cases it is a more random role, not intriguing in the sentence nucleus. For instance, in Sanskrit the role of the hearer in a command (and in a question for all that matter) is of primary importance and is, thus, appropriately assigned syntactic functions. In contrast, with the verbs of speaking the REC is less involved in the verbal action (the addressee merely has to listen) and, hence, appears less often in syntactic slots. The apparent schizophrenic syntax of the verbs of teaching may be ascribed to its relation to the ordering verbs, in which the REC is closely involved in the verbal action, on the one hand, and to the speaking verbs, in which the REC has a slightly less important status, on the other hand. These facts motivate the following order of ranking: AG > REC > PAT > REC > .... Thus, I propose a split of the REC using its kinship to the AG as basic criterion. PO/SO languages
On double object constructions
219
tend to activate through default the REC role for object assignment and demonstrate their preference to make use of a syntactic function to encode primarily semantic and/or pragmatic information, such as topicality, humanness, givenness etc. In other words, they emphasize the convergence between agent and recipient. In DO/IO languages the exact opposite arises: there is a decided predilection for the patient, which means that semantic/pragmatic information of the type mentioned above is left implicit. To sum up, while in the former case the semantic/pragmatic convergence between the agent and the recipient is at the bottom of the assignment of syntactic functions, in the latter case the divergence between the agent and the patient is basic. Split Objectivity languages seem to need an extra stimulus to give priority to the recipient. The factors topicality, givenness, and humanness alone do not trigger syntactization. Only when the recipient displays an additional typical agentive feature, i. e., high involvement in the verbal action, is it eligible for the object slot. So, to come back to the hierarchy in case of rigid PO/SO languages the agentive-like recipient, which has a prominent status in the information structure, is activated and the other one is neglected. In case of the rigid DO/IO languages not the agentive-like recipient but the patient is conceived of as more prominent. A merged process prevails in Split Objectivity languages. It is hoped that in spite of its fairly premature form, the solution offered is helpful when analyzing a dO construction in no matter what language and gives at least some useful suggestions about the factors that trigger it.
Notes 1. My thanks are due to George De Schutter, Johan van der Auwera and Michel Kefer for numerous valuable comments and suggestions. 2. Actually, the term "double object" — at least as far as I can see — is a contradiction in terminis, hence my putting "double" between quotes. The whole idea behind a syntactic function assignment to a particular constituent is to separate it from the rest for whatever reason. Assigning the same syntactic function twice within the same sentence unit basically means that this function is nullified. I fully agree with Plank (1987: 40) who warns "daß die beiden Objekte in ihrer Kasus-Markierung identisch sind, muß jedoch nicht notwendigerweise heißen, daß sie als syntaktische Relationen gänzlich ununterschieden wären" (cf. also Blansitt 1984: 127 in this respect). 3. Abbreviations of textual references: AV = Atharva-Veda; Dkc - Dasakumäracaritam; Hdt = Herodotos; IL = Ilias; X.An = Anabasis; Cu = Chändogya-Upanisad; M = Mânavadharmasästra; Ku = Katha-Upanisad; OD = Odyssey. Abbreviations of grammatical terms: ABL = ablative; ACC = accusative; AOR = aorist; DAT = dative; GEN = genitive;
220
4.
5.
6. 7. 8.
9. 10.
11.
Lieve Van de Walle IMP = imperative; I M P E R F = imperfect; INF = infinitive; INSTR = instrumental; LOC = locative; O P T = optative; PERF = perfect; P R E S = present; T E M P = temporal; VA = verbal adjective (past perfective participle). Further abbreviations are: Advancement to Subject Chaining Constraint (ASCC); AG = agent; BEN = beneficiary; dO constructions = double object constructions; DO = direct object; FG = Functional Grammar; GO = goal; IO = indirect object; OO = oblique object; PAT = patient; PO = primary object; REC = recipient; R G = Relational Grammar; SO = secondary object; SFH = Semantic Function Hierarchy; SU = subject. For convenience's sake, sandhi has occasionally been undone in the Sanskrit examples. This is signalled by the use of " = "; for example tena= ayam replaces tenäjam. Of course, recipients are not the only semantic case roles functioning as "partner" in a dO construction. For instance, in Sanskrit and Greek one has to discriminate between at least recipient and something like source (cf. section 3). Dryer (1986: 814—815) mentions that in the literature a lot of similar suggestions have been made. Compare in this respect Blansitt 1984; Bresnan 1982; Comrie 1982; Kisseberth and Abasheikh 1977. FG discriminates only between subject and object, arguing that there is no independent motivation for an indirect object. Due to the fact that Dik never really explains what is meant by "secondary" perspective, it is a rather vague notion and, hence, a rather vague explanation. Note in this respect that the so-called "case-recoverability problem" is hardly ever a real problem. In order to identify the patient and the recipient (or some other semantic role for that matter), one can rely on the context. Even when both constituents are characterized by the feature -(-human, it is usually easy to figure out how the roles are distributed (cf. also Plank 1987: 38 on this matter). With + R E C I mean a person to whom either information or goods are given, with —REC I mean a person of whom either information or goods are required. One might object to my including the causative constructions in this analysis arguing that they simply cannot be put on a par with the other dO constructions because of the radically different characteristics of the semantic roles causee and recipient. However, it will soon become clear that a parallel treatment is not a random decision but is seriously motivated by the fact that both construction types obey the same basic principles. I am aware that there are many verbs that lexicalize one or the other conceptual area of the conceptual field of asking. I do not pretend to be exhaustive but merely intend to give a rough sketch.
References Blansitt, Edward L. 1984 "Dechticaetiative and dative", in: Frans Plank (ed.), Objects. Towards a theory of grammatical relations, London: Academic Press, 127 — 150. Borg, Albert J . and Bernard Comrie 1984 "Object diffuseness in Maltese", in: Frans Plank (ed.), Objects. Towards a theory of grammatical relations, London: Academic Press, 109 — 126. Bresnan, Joan 1982 "The passive in lexical theory", in: Joan Bresnan (ed.), The mental representation of grammatical relations, Cambridge, M A : M I T Press, 3 — 86.
On double object constructions
221
Comrie, Bernard 1984 a "Grammatical relations, semantic roles and topic-comment structure in a New Guinea Highland language: Harway", in: R. Steele—Terry Threadgold (eds.), Language topics. Essays in honour of Michael Halliday, Amsterdam: Benjamins, 355-367. 1984 b "Form and function in explaining language uni versais", in: Brian Butterworth—Bernard Comrie —Osten Dahl (eds.), Explanations for language universals, Berlin: Mouton Publishers, 87 — 105. Dik, Simon C. 1978 Functional Grammar. Amsterdam: North Holland Publishing Company. 1989 The theory of Functional Grammar, vol. 1: clause structure. Dordrecht: Foris Publications. Dryer, Matthew S. 1986 "Primary objects, secondary objects, and antidative", Language 62, #4, 808— 845. Estival, Dominique 1985 "Syntactic priming of the passive in English", Text 5 (1—2), 7 — 21. Givón, Talmy 1983 Topic continuity in discourse: a quantitative crosslanguage study. Amsterdam: J. Benjamins. 1984 "Direct object and dative shifting: semantic and pragmatic case", in: Frans Plank (ed.), Objects. Towards a theory of grammatical relations, London: Academic Press, 1 5 1 - 1 8 2 . Hock, Hans H. 1981 "Sanskrit causative syntax: a diachronic study", Studies in the Linguistic Sciences, Vol. 11, #2, 9 - 3 3 . 1985 "Sanskrit double objects: will the real object please stand up?", Indian Journal of Linguistics 12, 51—70. Hyman, Larry M. 1984 "Form and substance in language universale", in: Brian Butterworth—Bernard Comrie —Osten Dahl (eds.), Explanations for language universaIs, Berlin: Mouton Publishers, 6 7 - 8 6 . Johnson, David E. 1977 "On relational constraints on grammars", in: Peter S. Cole and Jerrold M. Saddock (eds.), Syntax and semantics 8: Grammatical relations, New York: Academic Press, 151 — 177. Keenan, Edward 1976 "Towards a universal definition of subject", in: Charles Li (ed.), Subject and topic, New York: Academic Press. Kisseberth, Charles W. — Mohammad Iman Abasheikh 1977 "The object relation in Chi-Mwi:ni, a Bantu language", in: Peter S. Cole— Jerrold M. Sadock (eds.), Syntax and semantics 8: Grammatical relations, New York: Academic Press, 1 7 9 - 2 1 8 . Kühner, R. 1898 Ausführliche Grammatik der griechischen Sprache. Zweiter Teil: Satzlehre. Hannover und Leipzig: Hahnsche Buchhandlung. Luraghi, Silvia 1986 "Der semantische und funktionelle Bau des althethitischen Kasussystems", KZ 99, Heft 1, 2 3 - 4 2 .
222 1987
Lieve Van de Walle
"Reconstructing Proto-Indo-European as an ergative language: a test", J I ES Vol. 15, # 3 and 4, 3 5 9 - 3 7 9 . Mallinson, Graham and Barry J. Blake 1981 Language typology. Amsterdam: North Holland. Newmeyer, Frederich 1980 Linguistic theory in America. New York: Academic Press. Ostman, Jan-Ola 1987 "Implicit involvement in interactive writing", in: Jef Verschueren — Marcella Bertucelli-Papi (eds.), The pragmatic perspective, Amsterdam: Benjamins, 155 — 179. 1988 "Adaptation, variability, and effect: comments on IPrA Working Documents 1 & 2", IPrA Working Document 3, 7 - 4 0 . Plank, Frans 1987 "Direkte indirekte Objekte, oder: Was uns lehren lehrt", Leuvense Bijdragen 76, 37-61. Postal, Paul and David M. Perlmutter 1983 "Some proposed laws of basic clause structure", in: David M. Perlmutter (ed.), Studies in Relational Grammar 1, Chicago: University of Chicago Press. Speyer, J. S. 1886 Sanskrit syntax. Delhi: Motilal Banarsidass. Van de Walle, Lieve 1990 "Semantic functions and subject assignment revisited. Evidence from Classical Sanskrit", in: Mike Hannay — Elseline Vester (eds.), Working with Functional Grammar: descriptive and computational applications, Dordrecht: Foris Publications, 8 7 - 1 0 1 . Verschueren, Jef 1985 What people say they do with words. Norwood, New Jersey: Ablex Publishing Company. 1987 Pragmatics as a theory of linguistic adaptation. IPrA Working Document 1. Workgroup on Functional Grammar 1981 "On the Functional Grammar of teaching verbs", in: Teun Hoekstra—Harry van der Hulst, Michael Moortgat (eds.), Perspectives on Functional Grammar, Dordrecht: Foris Publications, 203 — 234.
"Nouny" and "verby" adjectivale: a typology of predicative adjectival constructions 1 Harrie Wetzer
1. Introduction This paper reports on work in progress. It contains the first results of a typological-universal research project addressing the question of the formal encoding of adjectival concepts in language. The terms "adjectival concept" and "adjectival meaning" are used here rather loosely, i. e., as those concepts referring to qualities or properties, which in Indo-European languages generally belong to the linguistic category "Adjective". Further, with the notion "adjectival (word/item)" I will refer to words expressing adjectival concepts, irrespective of their word class status. Every languâge seems to distinguish the major word classes Noun and Verb. Although the question of the universality of the noun-verb distinction has long been subject of debate, most linguists now adhere to the view that every language has the two basic parts-of-speech Noun and Verb 2 (see, for instance, Robins 1952; Jacobsen 1979; Hopper —Thompson 1984; Schachter 1985). Adjectives, however, do not constitute a universal linguistic category: many languages have no separate adjective class at all, whereas other languages only have a rather small, closed set of adjectives at their disposal. In the past decades, linguists have paid considerable attention to the question how languages without a distinct open adjective class encode concepts that are expressed through adjectives in languages which do have this major class. As to the question of how adjectival meanings are encoded in language, Schachter (1985) distinguishes three groups of languages. 1. Languages with a distinct open adjective class. 2. Languages with a rather small, closed set of adjectives. Adjectival concepts not included in the adjective class are typically expressed as verbs or nouns. 3. Languages which have no distinct class of adjectives at all, either open (as in 1.) or closed (as in 2.). Two subgroups of languages can be distinguished here:
224 a. as b. as
Harrie Wetter adjectival-noun languages: adjectival concepts are primarily expressed (subclasses of) nouns. adjectival-verb languages: adjectival concepts are primarily expressed (subclasses of) verbs.
Schachter's description is representative for the way the variation in the expression of adjectival meanings is generally discussed in the literature (for a similar view see, for instance, Locker 1951; Dixon 1977; Givón 1979, 1984). Basically three types of lexical categorization are distinguished: adjectival concepts are encoded cross-linguistically as Adjectives, as (adjectival) Verbs or as (adjectival) Nouns. This more or less standardly accepted description, primarily based on the part-of-speech status of adjectivals 3 , can be thought of as a typology of the ways adjectival meanings are expressed in language. This typology generates the following two questions for further explanation: The first question concerns the occurrence of the attested types of lexical categorization: why should it be the case that adjectival concepts are found to be distributed across the three lexical categories Adjective, Noun and Verb in the world's languages? The second question concerns the distribution of languages over these three types of lexical categorization: why does a language choose a particular strategy in the expression of adjectival meanings? Why, for instance, do IndoEuropean languages generally have a distinct open class of adjectives? Why are adjectival concepts in languages like Quechua expressed as nouns, whereas adjectivals in Mandarin Chinese, for instance, cluster with the verbs? These questions are based on the assumption that the cross-linguistic variation in the expression of adjectival meanings is adequately captured by a description in terms of the three parts-of-speech Adjective, Verb and Noun. In this context, it is worth noting that the typology at issue here is not the direct result of a comprehensive cross-linguistic study of the grammatical behavior of adjectival words; at least as far as I know, a systematic investigation of this kind has never been carried out. In fact, the typology is constructed on the basis of secondary sources, i. e., the various linguistic analyses of adjectivals presented in the descriptive grammars of individual languages. As such, the set-up of the proposed typology contains a potential flaw: although the word class distinctions made within each individual language may be perfectly acceptable for the purpose of describing the grammatical structure of the language in question, there is no a priori reason to assume that they are equally valid for cross-linguistic comparison as well. Thus, if we intend to use this typology as a basis for further research, along the lines indicated by the questions formulated above, we should at
A typology of predicative
adjectival constructions
225
least make sure that the three types of lexical categorization that are distinguished are worthy of explanation at all, i. e., that Adjectives, (adjectival) Verbs and (adjectival) Nouns actually represent clearly identifiable and distinct cross-linguistic categories. In this paper I will argue that the proposed typology — reflecting the standard view on the variation in the expression of adjectival concepts — does not comply with this requirement and is therefore unsatisfactory as a basis for further explanation. Next, I will suggest a different point of view for looking at the cross-linguistic variation in the encoding of adjectival concepts. Starting with the alternative perspective, I will present a typology of predicative adjectival constructions.
2. The linguistic categorization of adjectival concepts: adjectives, verbs or nouns? 2.1. Some observations If we take a closer look at the grammatical properties of adjectivale — crosslinguistically defined as adjectives, verbs or nouns — we arrive at these two conclusions. A. If in a language a separate class of adjectives is distinguished, members of this class tend to show morphological and/or syntactic similarities with verbs or nouns as well. In different languages, members of the category "adjective" display varying degrees of affinity for one of the major word classes verb or noun. B. If in a language no separate class of adjectives is distinguished, adjectival words, defined as (a subclass of) verbs or nouns, typically have at least some distinctive grammatical properties not shared by "core" nouns or verbs. Before discussing the implications of these findings in section 2.2., let us briefly consider some of the sorts of evidence that lead to these conclusions. The first conclusion (in A. above) concerns the affinity of "adjectives" — defined as a separate word class — for the nouns or the verbs. Nkore-Kiga, a Bantu language spoken in South-Western Uganda, has a restricted set of about twenty "true" adjectives (Taylor 1985). Adjectives may function as modifiers in a noun phrase and are not subcategorized for inherent gender; unlike nouns, adjectives can take any noun class prefix in concord with the noun they qualify, compare:
226
Harrie
(1) a.
b.
Wetter
omu CL1 'The eki CL7 'The
-ntu omu-rungi -person CLl-good kind person.' -shushani eki -rungi -picture CL7 -good beautiful picture.'
(49)4
(49)
As Taylor notices, however, "the true adjective has a form and function similar to that of a noun" (Taylor 1985: 85). Adjectives, for instance, take the same set of class prefixes as nouns do (compare example (1)). Further, adjectives occur in the same predicative constructions as nouns do. Consider the following examples of predicate adjectives and nouns with the copula -ba 'be'. (2) a.
b.
m 1SG Ί am m 1SG Ί am
-ba omu -COP CL tall.' -ba omu -COP CL a doctor.'
-raingwa -tall
(176)
-fumu -doctor
(38)
Finally, adjectives can be used as nouns. Omu-raingwa (CLl-tall) in example (2a), for instance, may function as a head noun, meaning 'the tall one' (Taylor 1985: 90). Whereas in Nkore-Kiga the adjectives share many grammatical properties with the nouns, in Tigak, an Austronesian language spoken in New Ireland, the opposite affiliation has taken place (Beaumont 1980). Adjectives, described as constituting a distinct basic word class, have obvious verbal characteristics: In defining the basic word classes, reference frequently has to be made to characteristics of other classes. For example, intransitive verbs are separated from adjectives because they cannot be used as modifiers in a Noun Phrase as adjectives can. Use in Verb Phrases does not indicate the difference (Beaumont
1980: 85).
When used predicatively, adjectives are treated on a par with intransitive verbs. Like verbs they are obligatorily marked for subject by means of preposed subject pronouns, two sets of which occur, viz., present tense forms and past tense forms. In addition to these obligatory subject pronouns, a subject expressed by a noun or an independent pronoun may optionally be added (see examples (3b) and (4b). Consider the following examples of adjectival predicates (3) and verbal predicates (4) with the third person singular subject pronouns gi (present tense) and ga (past tense):
A typology of predicative
adjectival constructions
227
Adjectival predicates: (3) a.
b.
gi lavu 3SG.PRES big 'It is big.' tang iai ga lavu A R T tree 3SG.PAST big 'The tree is/was big.' 5
(72)
(40)
Verbal predicates: (4) a.
b.
gi ima 3SG.PRES come 'He is coming.' na Gamsa ga ima A R T Gamsa 3SG.PAST come 'Gamsa came.'
(74)
(58)
Thus, adjectives and verbs are indistinguishable when used as predicates. The defining characteristic of adjectives in Tigak is their function as a modifier in a noun phrase. When used as modifiers, adjectives directly follow the noun they qualify, whereas verbs are obligatorily accompanied by the preposed subject pronouns, just as in main predicates. Compare the attributive use of the adjective lavu 'big' and the verb tara 'see' in the following examples. (5) a.
b.
tang lui lavu A R T house big 'The big house.' (naga po etok suna) 1SG.PAST PERF talk to tang lakeak ga tara-i tang muata A R T child 3SG.PAST see-it A R T snake '(I spoke to) the boy who saw the snake.'
(41)
(51)
Nkore-Kiga and Tigak are instances of languages in which adjectives, defined as a separate word class, have grammatical properties in common with nouns or verbs respectively. Analogous arguments for the affinity of "adjectives" with either nouns or verbs can be made for virtually every language which has such a class, albeit that the degree to which adjectives resemble nouns or verbs may vary from one language to the other. The second conclusion regarding the grammatical characteristics of adjectival words (stated in B. above) concerns the distinctive properties of (adjectival) verbs and (adjectival) nouns, compared with "core" verbs or nouns.
228
Harrii Wetter
An example of what Schachter (cf. section 1) calls an "adjectival-noun" language is Imbabura Quechua (Province of Imbabura, northern Ecuador). In Imbabura Quechua "there does not appear to be a category 'adjective' which is formally distinct from the category 'noun'" (Cole 1982: 186). Cole's definition of nouns as "elements which can be the object of a postposition" (Cole 1982: 99) also applies to adjectivals. Adjectivals, for example, may be marked by the accusative postposition -ta, just like (other) nouns: (6) a.
b·
Ju^i jatun
-ta
-mi
chari-n
(97)
José big -ACC -VAL 6 have-PRES3 'José has a big one.' pay -paj tayta -ka chay wambra -ta -mi wajta he -of father -TOP that child -ACC -VAL hit 'His father hit that child.'
-rka (69) -PAST3
Further, adjectivals appear in the same predicative constructions as nouns do. Predicate nouns and adjectivals occur as the complement of the copula verb ka- 'be', which is obligatory, except when the verb is third person in the present tense (in that case the copula is normally omitted). Compare: (7) a.
t>-
ñuka wasi
-ka
yuraj
-mi
ka
-rka
my house -TOP white -VAL COP -PAST3 'My house was white.' Juan-ka mayistru -mi ka -rka Juan-TOP teacher -VAL COP -PAST3 'Juan was a teacher.'
(67)
(67)
Finally, both adjectivals and (other) nouns can serve as noun modifiers: (8) a.
t>-
jatun
runa
(73)
big man Ά big man.' rumi wasi stone house Ά stone house.'
(120)
These examples show that adjectivals in Imbabura Quechua pattern very much like nouns. However, adjectivals do not resemble nouns in all respects. First of all, unlike (other) nouns, adjectivals can be modified by adverbs such as yapa 'too', maymi 'very', asha(lla) 'slightly', etc. The sentence in example (9b), for instance, is ungrammatical:
A typology of predicative
(9) a.
b.
adjectival constructions
229
(99)
chay warmi maymi sumaj -mi that woman very pretty -VAL 'That woman is very pretty.' *chay warmi maymi duktur -mi that woman very doctor -VAL (That woman is very a doctor)
(100)
Second, the derivational suffix -sha which "suggests that the basic meaning of the word is pleasant" (Cole 1982: 186) is restricted to adjectival nouns: (10)
kushi -sha -mi ka-ni happy -'nice' -VAL be-PRESl Ί am nice and happy.'
(186)
Third, the suffix -ta is used to derive manner adverbs from nouns. The distribution of this suffix, at least in its function as adverbializer 7 , is limited to nouns expressing adjectival meanings. Cf.: (11)
toy ta -ka sumaj -ta trabaja father -TOP beautiful -ADVLZR work 'Father worked well.'
-rka -PAST3
(186)
Finally, inchoative verbs are derived from nouns by means of the suffix -ja. Again, however, -ya is largely restricted to adjectival nouns (cf. the unacceptability of (12b)): (12) a.
jatun -ya -rka big -'become' -PAST3 'He became big.' b. *libru -ya -rka book -'become' -PAST3 (It became a book)
(179)
(179)
Even though adjectivals in Imbabura Quechua do not behave like (other) nouns in all respects, no separate adjective class is distinguished. Cole rightly notices, however, that the restricted applicability of the suffixes -sha, -ta, and -ya constitutes a problem for the claim that Imbabura Quechua has no category "adjective" which is formally distinct from the category "noun". He then proceeds: The most likely explanation for the existence of a suffix with the distributional limitations of -sha in the absence of a category "adjective" is that the meaning of -sha limits its use to certain classes of meanings (e.g., qualities rather than objects), and that the appropriate meaning classes correspond roughly to the category "adjective" in those languages having such a category. (The same
230
Harrie Wetter
approach would be taken with -ya ... and the adverbializer -ta ...) (Cole 1982: 186).
Thus, we see that the obvious similarities between adjectivals and nouns in Imbabura Quechua are considered by the author to be more salient than the observed grammatical differences which seem to be motivated on semantic grounds. The Yuman language Mojave is an example of an "adjectival-verb" language: adjectival meanings are formally expressed as stative verbs (cf. M u n r o 1976; Schachter 1985). When used as predicates, adjectivals and (other) intransitive stative verbs are indistinguishable; they take the same (obligatory) subject person prefixes (3rd person 0 ) and tense-aspect suffixes and they optionally co-occur with the same auxiliary. Compare the following examples taken from Schachter (1985): (13) a.
b.
?i:pa man 'The ?i:pa man 'The
-c homi: -k - S U B J tall -TNS man is tall.' -c su:paw -k - S U B J know - T N S man knows.'
(idu: AUX (idu: AUX
-mJ -TNS -m) -TNS
(19)
(19)
However, when they function as modifiers, adjectivals are distinguished from (other) verbs. Verbs obligatorily appear in a relativized form which involves the prefix k"-\ with adjectivals the use of this prefix is optional, cf.: (14) a.
b.
?i:pa man 'The ii:pa man 'The
(kw-) homi: -η* -c iva:k R E L tall - D E M - S U B J is here tall man is here.' kw -su:paw -tf -c iva:k R E L -know - D E M - S U B J is here man w h o knows is here.'
(19)
(19)
Schachter (1985) discusses Mojave as an example of a language whose classification as "adjectival-verb" language — unlike many other languages such as Mandarin Chinese — is somewhat problematic because adjectival words also have at least one distinctive property not shared by (other) verbs: "In the case of such a language (as Mojave, H. W.), one would probably wish to analyze words with adjectival meanings as a distinguishable subclass of verbs, rather than as a distinct part of speech, but this is perhaps an arbitrary choice." (Schachter 1985: 1 9 - 2 0 ) Mojave, however, is definitely not unique in this respect. Many so-called "clear cases" of "adjectival-verb" languages, such as Burmese, Chemehuevi,
A typology of predicative
adjectival constructions
231
Guarani, Mandarin Chinese, etc., are, in fact, quite similar to Mojave: it is the rule rather than the exception that "adjectival-verbs" show at least some distinctive grammatical properties not shared by (other) verbs. Compared with "core" verbs, adjectival verbs typically have a defective or partially different inflectional paradigm or they behave differently when used in attributive position. In Mandarin Chinese, for instance, a notorious "adjectival-verb" language, there is no formal distinction between adjectivals and verbs used predicatively (Li —Thompson 1981; Hopper—Thompson 1984). Compare: (15) a.
b.
tä 3SG 'S/he tä 3SG 'S/he
häo good is good.' päo run runs.'
(Hopper-Thompson 1984: 729)
(Hopper-Thompson 1984: 729)
Adjectivals, however, may be used attributively either with or without the nominalizer -de (Hopper—Thompson 1984: 729): 8 (16) a.
b.
häo rén good person 'a good person' häo -de rén good -NMLZR person 'a good person'
All other verbs may only be used in attributive position if accompanied by -de. Of the following examples (Hopper—Thompson 1984: 729), (17a) is ungrammatical: (17) a. b.
*pâo rén run person pâo -de rén run -NMLZR person 'a running person'
So far, we have discussed two general conclusions concerning the grammatical properties of adjectives, defined as a distinct word class, and adjectivalverbs and adjectival-nouns. These findings may now be summarized thus: whether or not a language has a distinct class of adjectives, words expressing adjectival meanings show the following characteristics to varying degrees across languages:
232
Harrte
Wetter
1. adjectivals tend to display distinctive grammatical properties not shared by "core" verbs or nouns. 2. adjectivals tend to fall into categories which either share many grammatical properties with verbs or with nouns. Cross-linguistically, then, the grammatical behaviour of adjectival words can be characterized by two opposing tendencies: irrespective of their word class status adjectivals tend to show an affinity for one of the major word classes, noun or verb. At the same time, however, they typically have distinctive properties not shared by nouns or verbs.
2.2. The problematic word class status of adjectivals In writing the grammar of any language, a linguist will identify a number of word classes or parts-of-speech. Word classes are established on the basis of language-specific criteria, and are then labelled in accordance with semantic considerations (cf. Lyons 1968, 1977; Schachter 1985). If, for instance, unrelated languages are described as having a category "adjective", the label "adjective" is chosen because of the roughly comparable semantic content of these classes: in each language, the adjective class contains (at least a focal subclass of) lexical items referring to "properties" or "qualities". Since the recognition of these word classes depends on language-specific grammatical criteria, however, the formal characteristics of "adjectives" may vary from language to language and are not necessarily comparable cross-linguistically. The cross-linguistic variation in the expression of adjectival meanings is more or less standardly described in terms of the word class status of adjectivals. Adjectival concepts are formally encoded as "adjectives", as "(adjectival) verbs" or as "(adjectival) nouns" (cf. section 1). In fact, this cross-linguistic description is a mere inventory of how adjectival words are analyzed in the grammatical descriptions of individual languages. In addition, however, the impression is created that the three-way distinction between adjectives, (adjectival) verbs and (adjectival) nouns is linguistically significant from a universal viewpoint, the implicit claim being that these three word classes represent distinct homogeneous cross-linguistic categories as well. The observations discussed in section 2.1. suggest that this (implicit) claim is not justified. As to the cross-linguistic category "adjective" we may note that it actually represents a rather heterogeneous cluster of word classes, containing both verb-like and noun-like adjectives.
A typology of predicative
adjectival constructions
233
Furthermore, the boundaries between adjectives on the one hand and (adjectival) verbs and nouns on the other appear to be rather fuzzy. Adjectives with verb-like characteristics, for instance, are by definition formally distinguishable from verbs. In this respect, however, they are not fundamentally different from (adjectival) verbs, which also typically show distinctive properties not shared by (other) verbs. The same arguments can be made for the absence of a clear-cut dividing line between noun-like adjectives and (adjectival) nouns. These observations give rise to the following question: what does it mean to say that a language has a class of adjectives rather than a subclass of (adjectival) verbs or nouns? Dixon (1977: 62 — 63) suggests that adjectives are "a set of lexical items, distinguished on morphological and syntactic grounds from the universal classes Noun and Verb". Obviously this suggestion is of little help, since not only "true" adjectives, but adjectival words in general tend to show at least some distinctive grammatical properties compared to verbs and nouns. One might further specify Dixon's statement and argue that some grammatical differences do, and others do not justify the assumption of a separate adjective class. This would imply that some distinctive properties must be considered differences of kind, resulting in the recognition of an adjective class, whereas other differences are merely differences of degree and give rise to the definition of adjectivals as (subclasses of) verbs or nouns. This qualified statement, however, is also not without difficulties. If we compare the grammatical characteristics of adjectives, (adjectival) verbs and (adjectival) nouns in the various descriptive grammars of individual languages, there appears to be no consensus among grammarians about the status of distinctive properties as either differences of kind or differences of degree. As a matter of fact, it is impossible to isolate any single distinctive grammatical feature that is consistently treated as either a necessary or a sufficient condition for the assignment of adjectivals to a separate adjective class. Consider, for example, the noun-modifier function of adjectives. In many languages, including Nkore-Kiga and Tigak (see examples (1) and (5)), the syntactic behaviour of adjectivals in attributive position is an important criterion for the recognition of a distinct adjective class. Cross-linguistically, however, adjective classes are not necessarily characterized by distinct behaviour in attributive constructions. Japanese, for instance, has a class of (verb-like) adjectivals, which is commonly viewed as an independent adjective class (cf. Backhouse 1984). These adjectives directly occur as predicates and as adnominal modifiers just as verbs do. When used attributively, both adjectives and verbs occur in a relative clause which directly
234
Harrte
Wetter
precedes the head noun, without the use of a relative pronoun. Consider the following examples of predicative (a) and adnominal (b) constructions with adjectives (example (18)) and with verbs (example (19)) (Backhouse 1984: 170). Adjectives: (18) a.
b.
kono rombun -wa naga-i this article -TOP long-PRES 'This article is long.' naga -i rombun long -PRES article Ά long article.'
Verbs: (19) a.
b.
Suzuki -kun-wa tabete -iru9 Suzuki -Mr.-TOP eatGERUND -AUX 'Mr. Suzuki is eating.' tabete -iru hito eatGERUND -AUX person Ά person who is eating.'
In Japanese, then, adjectives are defined as a separate word class, but syntactically they are indistinguishable from verbs when used attributively. In other languages, however, adjectivals do behave differently in attributive position, but their distinctive behavior is simply not accepted as a criterion for a separate adjective class. In Mojave and Mandarin Chinese, for instance, adjectivals are treated as (subclasses of) verbs, despite their grammatical differences from (other) verbs (cf. examples (14) and (16 — 17)). As it turns out, the distinctive behavior of adjectivals used as noun modifier is neither always a necessary (cf. Japanese) nor a sufficient criterion (cf. Mojave and Mandarin Chinese) for "adjective-hood". Similar observations can be made with respect to other "typical" distinctive properties, such as specification for degree (including comparative and superlative forms, as well as adverbial modification), agreement of adjectives with the noun they qualify (as a modifier in the noun phrase or as a predicate), and whatever other grammatical properties one might think of when talking about adjectives. In fact, it appears to be the case that no formal characterization in terms of "distinctive grammatical properties" can be given that applies to all adjective classes and not to (adjectival) verbs or nouns. Whereas adjectival words tend to display both grammatical similarities with and differences from the major word classes noun and verb, no objective
A typology of predicative adjectival constructions
235
criteria are available to distinguish adjectives from (adjectival) verbs or nouns. As a result, adjectivals are classified more or less arbitrarily as either adjectives or subclasses of verbs or nouns cross-linguistically. Given this situation, it will come as no surprise that grammatical descriptions of the same language may differ in the word class status assigned to adjectivals. As an example, according to Hardjadibrata (1985), Sundanese (Western Java) has a separate class of adjectives. One of his main arguments is that adjectives normally have three morphologically marked levels of comparative degree (cf.: jankung 'tall', jangkungan 'taller', pangjangkungna 'tallest') whereas verbs do not have this property (cf.: nangtung 'stand'; *nangtungan and *pangnangtungna do not exist) (Hardjadibrata 1985: 8). Another argument is that the use of the relative marker anujnu is optional with adjectives when they are used as noun modifier 10 . Furthermore, adjectives, unlike verbs, are not affixed for voice and adjectives have the distinctive property of occurring in so-called "Interjective Predicates" (Hardjadibrata 1985: 62 — 63), i.e., exclamatory sentences of the type "How beautiful this house is!". Notwithstanding the obvious differences mentioned by Hardjadibrata, Sundanese adjectivals also have many grammatical properties in common with verbs. Just like verbs (and contrary to nouns) adjectivals occur with auxiliary verbs and adverbials, expressing aspectual and modal notions, and are generally negated by teu\henteu. Consider the following examples, taken from Hardjadibrata (1985): (20) a.
t>·
teu
rêk
hésé
pisan
NEG will difficult very '(It) will not be very difficult.' kuring teu rék Mit I NEG will go Ί will not go.'
(58)
(49)
In the presence of both similarities with and differences from verbs, Hardjadibrata emphasizes the distinctive properties of adjectivals and postulates the existence of a separate adjective class. In his treatment of Sundanese, Robins (1965, 1968), however, takes the opposite position and states that "adjectives may be regarded as a subclass of (intransitive) verbs" (cf. Robins 1968: 352). The difficulties in determining the word class status of adjectivals clearly illustrate a major general problem in the definition of word classes, i. e., the choice of suitable criteria for the classification of words into classes.
236
Harrie
Wetter
Within the structuralist tradition this problem is, at least in theory, nonexistent: for the classification of words into classes all distributional properties of lexical items are of equal relevance and "as many classes are set up as words of different formal behaviour are found" (Robins 1964: 229). In fact, this point of view implies that the definition of word classes is a purely nonarbitrary, empirical issue: each grammatical difference — without further qualification — enforces the recognition of a distinct word class. Apart from the question whether it is feasible at all to account for all grammatical differences between lexical items, this theoretical approach is obviously far too absolute. It would lead to a gross over-classification, yielding a very large number of very small classes at the expense of the generalizations that can be made on the basis of grammatical similarities. In the practice of grammar-writing the definition of word classes has always involved, either implicitly or explicitly, the ranking of criteria. Words with "essentially identical roles in the structure of the language" (Gleason 1961: 93) are classed together, whereas classes are treated as separate when they show "enough difference" (Hall 1964: 163) from other classes. As to the choice of criteria for the recognition of word classes, the problem arises as to how notions like "essentially identical" and "enough difference" should be defined. In fact, the major word classes of a language are distinguished on the basis of clusters of properties which typically are non-discrete, showing a certain amount of overlap of grammatical properties as well as some differentiation (compare the observations discussed in section 2.1.). Often, in the presence of both similarities and differences, the question arises whether such clusters should be considered as "essentially identical" or as showing "enough difference". The observed similarities might lead to the recognition of subclasses of one single word class (consider, for instance, adjectivals forming a subclass of the verbs); the grammatical differences, however, might equally justify the assumption of distinct word classes (cf. verb-like adjectives, distinguished as independent word classes from verbs). This problem cannot be solved on empirical grounds; it is essentially a matter of interpretation (if not taste). In this context, consider for instance Matisoffs statement (1973: 42 — 43) in his grammar of Lahu: " . . . ultimately the analyst's decisions (whether he admits it or not) will be based on esthetic considerations... The analyst arrives at his major word classes through trial and error, guided by the principles of 'generality', 'economy', and 'simplicity', and his sense of what is 'linguistically significant'." In an elaborate discussion on the definition of word classes Crystal (1967: 47) eventually reaches the conclusion that in the choice of definitional criteria for word classes arbitrariness cannot be eliminated: 11 "... one is forced to
A typology of predicative
adjectival constructions
237
conclude that word classes may be as broad or as narrow as there is need of in a particular situation, and that no one classification is absolutely better than any other ... different linguists for different purposes will make more or less detailed classifications." Summarizing the major points of this section, we have to conclude that the standard description of the cross-linguistic variation in the expression of adjectival concepts, stated in terms of the word class status of adjectivals, is unsatisfactory as a starting point for further investigation. The three types of lexical categorization adjectives, (adjectival) verbs and (adjectival) nouns do not represent clearly identifiable and homogeneous cross-linguistic categories. First of all, the cross-linguistic category "adjective" constitutes a rather heterogeneous cluster containing both verb-like and noun-like adjective classes. Furthermore, the boundaries between adjectives on the one hand and (adjectival) verbs and nouns on the other are extremely fuzzy. Whereas adjectival words generally display an affinity for the verbs or the nouns, there appear to be no objective criteria to decide whether the observed distinctive grammatical properties of adjectivals justify the recognition of a separate adjective class or a (distinguishable) subclass of verbs or nouns. As a result, adjectivals are more or less arbitrarily classified cross-linguistically as adjectives or as (adjectival) verbs or nouns. Even though a chosen classification of adjectivals may be useful and acceptable for the purpose of writing a descriptive grammar for an individual language, 12 the word class status of adjectivals is not a sound basis for cross-linguistic comparison, as there appears to be no consistent pattern according to which adjectives, adjectival verbs and adjectival nouns can be distinguished formally as homogeneous cross-linguistic categories.
2.3. "Nouny" and "verby" adjectivals In the previous section, I have argued that the cross-linguistic variation in the formal encoding of adjectival concepts cannot adequately be described in terms of the traditional parts-of-speech labeled adjective, verb, and noun. In this section, I will propose a different point of view for looking at the ways in which adjectival meanings are expressed across languages. First, however, I would like to discuss a semantic factor which is claimed to underlie the classification of the lexicon into verbs, nouns, and adjectives, and which is also relevant in the context of the alternative perspective to be introduced here. This semantic factor, referred to as time-stability, has been recently worked out by Givón (1979, 1984) in the Time Stability Hypothesis.
238
Harrte
Wetter
Givón (1984: 51—52) argues that the lexical categories verb, noun, and adjective can be ordered as to whether they denote states that are, to a greater or lesser degree, "stable over time". "Experiences ... which stay relatively stable over time ... tend to be lexicalized in human language as nouns... At the other extreme of the lexical-phenomenological scale, one finds experiential clusters denoting rapid changes in the state of the universe. These are prototypically events or actions, and languages tend to lexicalize them as verbs." Prototypical adjectival concepts (i. e., properties) are taken to be "intermediate states". In languages which have a class of adjectives, this class "occupies the middle of the time-stability scale" (1984: 52). Thus, in a language with a distinct adjective class, the time-stability scale is divided into three sections which receive different formal encodings in the language. Whereas the exact cut-off point on the semantic scale may vary from one language to the other, these three sections — prototypically covering "actions/events", "properties", and "time-stable concepts" (see (21a)) — are formally encoded as verbs, adjectives and nouns respectively. The left-to-right ordering of these lexical categories (see (21b)) corresponds with an increase in time-stability. (21) a. b.
Time Stability Scale events/actions — properties — time-stable concepts VERBS ADJECTIVES NOUNS
Within the context of the Time Stability Hypothesis, the cross-linguistic variation in the formal encoding of adjectival concepts can be conceived of as different ways in which languages give formal expression to semantic differences in the degree of time-stability of the concepts encoded. Thus, the standard view according to which adjectival concepts are cross-linguistically expressed as either adjectives, or (adjectival) verbs, or (adjectival) nouns, could be represented schematically by the following three basic patterns (dashes indicate the boundaries between lexical categories): (22) a. b. c.
VERBS VERBS VERBS
/ ADJECTIVES ADJ. VERBS H ADJ. NOUNS
/ NOUNS / NOUNS NOUNS
Languages with a distinct adjective class are represented by pattern (22a), as described above in (21). Patterns (22b —c) represent languages without a separate adjective class. As opposed to type (a) languages, the time-stability scale could be thought of as divided into two rather than three semantic domains. Adjectival-verb languages (cf. (22b)) do not formally distinguish between "events/actions" and "properties" which are both lexicalized as verbs. The cut-off point on the time-stability scale is located between "properties"
Λ typology of predicative
adjectival constructions
239
and "time-stable concepts", the latter being encoded as nouns. Adjectivalnoun languages (cf. (22c)) display the opposite affiliation: "properties" are treated on a par with "time-stable concepts" and are formally expressed as nouns, whereas "events/actions" are categorized as verbs. However, as we already concluded in sections 2.1. —2.2, the standard description based on the distinction between adjectives, (adjectival) verbs and (adjectival) nouns yields a rather distorted picture of the actual grammatical properties of adjectival words. In fact, languages with "verb-like" adjectives and adjectival-verb languages pattern very similarly in the way they formally express adjectival concepts. Whereas a clear distinction is made in these languages between nouns and adjectivals, the formal differentiation between adjectivals and verbs is never that obvious, though always observable. Whether the grammatical differences between adjectivals and verbs should be interpreted as differences of kind or differences of degree is, as we have seen, not an empirical issue, but rather a matter of definition (see 2.2.). On analogous grounds, languages with "noun-like" adjectives and adjectivalnoun languages can be said to display very much the same pattern in the encoding of adjectival meanings: whereas verbs are clearly distinguishable from adjectivals and nouns, the grammatical properties of adjectival words and nouns are very similar, yet never identical. In the light of these observations, the three patterns distinguished in (22a —c) can be replaced by the following two more generalized patterns (23a —b). In (23) " / " indicates a "clear" categorial boundary, whereas "??" refers to the questionable dividing line which is sometimes interpreted as a class boundary, and sometimes as a boundary between subclasses·. (23) a. b.
VERBS VERBS
?? H
ADJECTIVALS / ADJECTIVALS ??
NOUNS NOUNS
A major split can be observed in the way adjectival concepts are encoded across languages. Whether or not there is a distinct adjective class, words expressing adjectival meanings tend to fall into two categories: either they share many grammatical properties with verbs, or they share many properties with nouns. 13 This split will, from now on, be referred to as the distinction between "nouny" and "verby" adjectivals. Although adjectivals will typically not be identical to verbs or nouns in all respects, "verby" adjectivals will show a clear "orientation" towards verbs, whereas "nouny" adjectivals will display the opposite affiliation and have many properties in common with nouns. The "nouny"/"verby" split in the expression of adjectival concepts offers a promising alternative for the generally accepted, though unsatisfactory,
240
Harrie
Wetter
distinction between adjectives, (adjectival) verbs and (adjectival) nouns, and is taken as the point of departure for further cross-linguistic research. Our first, descriptive, task is to investigate, on the basis of a representative sample of languages, whether adjectivale can be considered "nouny" or "verby" in each single language of the sample. In the remainder of this paper I will present the results of this investigation in the form of a typology of predicative adjectival constructions.
3. A typology of predicative adjectival constructions In this section I will introduce a typology of predicative adjectival constructions, based on a sample of 100 languages. Predicative adjectival constructions are classified into a number of different types, depending on whether, and how, adjectivals in the attested constructions display an orientation towards verbs or nouns. Before presenting the typology in section 3.2., I would like to make some remarks concerning the construction of the typology.
3.1. Preliminary remarks The purpose of the present study is to find out whether adjectivals can be considered "nouny" or "verby" in the languages under investigation. The two most important syntactic functions generally associated with adjectival words are their predicative use (as in "the man is tall") and their use as nounmodifier (as in "the tall man"). If we compare the grammatical properties of attributive and predicative adjectivals, the general tendency is for the "verby" or "nouny" character of adjectivals to manifest itself most clearly in predicative constructions. In attributive constructions adjectivals typically display distinctive grammatical properties not shared by verbs or nouns (see section 2.1.). 14 Since we are primarily interested in the question of whether adjectivals are either verb- or noun-oriented, the focus of our investigation will be on linguistic expressions which typically involve the use of predicative adjectivals, i. e., constructions which have the semantic function of assigning a property to a person or an object and which represent the functional equivalent of English sentences such as "the man is tall." Because expressions which satisfy these conditions are generally constructed by means of predicative adjectivals, they will be referred to here as predicative adjectival constructions. It should be emphasized, however, that "predicative adjectival constructions", as they
A typology of predicative adjectival constructions
241
are described above, do not necessarily include predicative adjectivals. Some languages have (sub-)classes of adjectivals which do not occur as predicates at all (see, for instance, Schachter 1985: 16). Instead, these languages use different kinds of periphrastic constructions, in which the adjectivals in question function, for example, as noun modifier in a noun phrase, or as the complement of a preposition (see types 3 and 5 in section 3.2). Since periphrastic adjectival constructions are, in these languages, the direct functional equivalent of expressions like "the man is tall" in English, they are considered instances of "predicative adjectival constructions" and are included in the data base for the typology to be constructed. For the purpose of this investigation I have further confined myself to prototypical adjectivals. This term refers to classes of adjectival items which minimally express property concepts included in Dixon's (1977) "semantic types" of AGE, DIMENSION, and VALUE. In order to clarify my reasons for limiting the scope of the investigation to these "prototypical" adjectivals, let me first discuss the major findings of Dixon's 1977 paper, in which he explores the question of how adjectival meanings are expressed in languages which lack an open adjective class. Dixon (1977: 31) classifies the "basic members" of the English adjective class into seven universal "semantic types". These seven types which make up the word class Adjective are the following: 1. 2. 3. 4.
DIMENSION - big, little, long, wide, ... PHYSICAL PROPERTY - hard, heavy, smooth, ... COLOR HUMAN PROPENSITY — jealous, happy, kind, clever, generous, cruel, proud, ... 5. AGE — new, young, old, ... 6. VALUE — good, bad, delicious, ... 7. SPEED — fast, quick, slow, ... Dixon then goes on to investigate the word class affiliation of these semantic types in "adjective-deficient" languages. The major results of his investigation can be summarized as follows. 1. If a language has a category of adjectives, this class is likely to include at least members of the semantic types AGE, DIMENSION, VALUE and COLOR, however small it is. In languages without an adjective class these four types belong to a single part-of-speech, i. e., either verbs or nouns. In that case, the actual word class membership of these types cannot be predicted. 2. The other three semantic types in Dixon's list above, i.e., PHYSICAL PROPERTY, HUMAN PROPENSITY and SPEED, may be included in the
242
Harrie Wetter
same class which covers the four types mentioned in 1. This is the normal situation in languages with an open adjective class, and in adjectival-verb and adjectival-noun languages. However, these three remaining types are not always included in the class which expresses age, dimension, value and color. In that case, the lexical categorization for these semantic types can be predicted. Physical properties, for instance, are then preferably encoded as verbs, whereas human propensity concepts are typically associated with the category noun. Although Dixon's generalizations appear to be valid on the whole, some additional remarks should be made. First of all, my own observations confirm the claim (cf. 2 above) that adjectival concepts included in the semantic types physical property, human propensity and speed are typically associated with either the verbs or the nouns (that is, when they do not cluster with age, dimension, value, and colour concepts). With respect to the typical word class affiliation of these concepts, however, Dixon's predictions are not always correct. The data suggest that further subclassification of these semantic types is needed and that predictions become far more reliable when the semantic factor time stability is introduced. Within the human propensity type, for instance, a distinction can be made between more permanent, stable properties (such as "wise", "brave", "clever", "wicked", etc.) and temporary, non-stable properties or states (such as "angry", "happy", "sick", "sad", etc.), the former being typically encoded as nouns, the latter as verbs. A second remark concerns the status of Dixon's semantic type color. Contrary to Dixon's findings, my observations suggest that the color type is not as prototypical as age, dimension and value. In fact, colour terms do not always occur in the adjectival class which prototypically covers the latter three semantic types. In that case color is typically found to be encoded by nominal items. 15 Within the context of the "nouny"/"verby" split in the formal expression of adjectival concepts, Dixon's results, modified by my own observations, can be interpreted as follows. In many languages all adjectival semantic types in Dixon's list are encoded in one and the same way. In these languages, the "nouny" or "verby" orientation of adjectivals cannot be predicted from their meaning. There is no semantic reason — as far as we know — why adjectivals are "verby" in one language, and "nouny" in another. In other languages, adjectival concepts are found to be distributed across two or more different lexical classes. For classes containing the "prototypical" adjectival types age, dimension, and value no prediction whatsoever can be made as to their "verby" or "nouny" character. However, for concepts included in the semantic types color, physical property, human propensity, and speed, things are
A typology of predicative
adjectival constructions
243
different. If these concepts do not cluster with the three "prototypical" types of property concepts, there are good reasons to assume that their lexical categorization as "nouny" or "verby" adjectivals is motivated on semantic grounds, even though the semantic principles underlying their typical word class affiliation are not (yet) fully understood. For the purpose of the present study I have decided to exclude those adjectivals whose "nouny" or "verby" character is quite likely motivated on semantic grounds. For this reason the scope of the investigation is restricted to prototypical adjectivals, i.e., those classes of adjectival items which minimally express properties belonging to the semantic types age, dimension, and value. Thus, the examples used in the following section will typically involve adjectivals referring to prototypical properties, such as "old", "young", "tall", "small", "good", and "bad". Some grammars, however, do not give sample sentences with adjectivals expressing prototypical properties and so for practical reasons some of the examples below contain non-prototypical adjectival expressions. Even so, these examples are representative for the form prototypical adjectivals take in that particular language. So far, we have indicated which adjectival constructions are considered relevant for the data base of the typology. The scope of our investigation is restricted to prototypical adjectivals appearing in predicative adjectival constructions. Once the data base has been established, the next stage in the investigation involves the construction of the typology. The attested adjectival constructions are classified into different types, according to whether the adjectivals in question show a "verby" or "nouny" orientation. Generally, the orientation of adjectivals can be determined without serious difficulties. As an example, consider the following predicative adjectival constructions in Cairene Egyptian Arabic (Gary —Gamal-Eldin 1982): (24) a.
b.
h i j j a hilwa she prettyFEM.SG 'She is pretty.' h i j j a kaan -it hilwa she COPwas -3FEM.SG prettyFEM.SG 'She was pretty.'
(61)
(61)
Predicative adjectivals are obligatorily accompanied by the overt copula kaan 'to be' in past and future tense constructions (cf. (24b)). In the present no overt copula is used, and predicative adjectivals are linked to their subject by juxtaposition (cf. (24a)). Nominal predicates are constructed in the same way:
244
Harrie
(25) a.
b.
Wetter
hijja she 'She hijja she 'She
mudarrisa teacherFEM.SG is a teacher.' mudarrisa kaan -it COPwas -3FEM.SG teacherFEM.SG was a teacher.'
(23)
(23)
As opposed to adjectivals and nouns, verbs are inflected for tense-aspect and gender, number, and person: (26) a.
b.
ti -ktib 3FEM -writelMPERF 'She writes.' katab -it writePERF -3FEM.SG 'She wrote.'
(100)
(100)
Thus, we see that adjectivals in Cairene Egyptian Arabic are nouny. They occur in the same predicative constructions as nouns do, opposed to verbs. In most languages of the sample a clear formal distinction can be made between verbal and nominal predicates. Verbs are typically subcategorized for tense, aspect, mood, etc., and may be marked to indicate agreement for person, number and gender categories. Predicate nouns, on the other hand, do not follow the verbal paradigm, and may be accompanied by a (verbal, pronominal or particle) copula. These grammatical differences between verbal and nominal predicates make it relatively easy to determine the "verby" or "nouny" character of predicative adjectivals. For some languages, however, the orientation of predicative adjectivals towards verbs or nouns cannot be so easily determined. These languages use one single morpho-syntactic strategy in the encoding of verbal, adjectival and nominal predicates (cf. Stassen, this volume). As a result, predicative adjectivals share grammatical properties with both verbs and nouns, and cannot unequivocally be characterized as either "verby" or "nouny". This situation is observed, for instance, in Yurak, a member of the Samoyed language family (Castrén 1854 [1966]; Hajdú 1963, 1970, 1975). Just like intransitive verbs, predicative nouns and adjectivals attach the so-called "predicative suffixes" which specify person and number of the subject. Consider the following examples with the first person singular suffix -m¡-dm: (27) a.
man jilê -m I live -1SG Ί live.'
(Hajdú 1963: 68)
Λ typology of predicative
b.
c.
man sawo -dm I good -1SG Ί am good.' man xasawa -dm I man -1SG Ί am a man.'
adjectival constructions
245
(Castrén 1966: 226)
(Hajdú 1975: 13)
Further, predicative nouns and adjectivals can be marked to indicate past tense by adding the past tense suffix -s. According to Hajdú (1963: 68) the same suffix is used to form the past tense of durative verbs like "stand" and "live", compare: (28) a.
jïlê -0 -s live -3SG -PAST 'He lived.' b. sawo - 0 -J good -3SG -PAST 'He was good.' c. pida xasawa - 0 -s he man -3SG -PAST 'He was a man.'
(Hajdú 1963: 68)
(Hajdú 1963: 68)
(Hajdú 1975: 13)
Thus, we see that in Yurak the noun-verb distinction is partially neutralized. With respect to subject agreement and tense marking, predicative nouns are treated on a par with intransitive, durative verbs. Accordingly, these grammatical properties give us no indication whatsoever of the "verby" or "nouny" orientation of adjectivals. However, although verbs, adjectivals, and nouns are treated in a fairly uniform manner when used as predicates, this uniformity is not absolute. On closer examination we find that verbal and nominal predicates can be formally distinguished, and that, on the basis of this distinction, predicative adjectivals can be considered "nouny". First of all, compared with verbs, nouns and adjectivals display a highly defective conjugational pattern. They only occur in the indicative non-past and past tense forms mentioned above, and, unlike verbs, they do not attach mood markers. The grammatical distinction is demonstrated even more clearly in negative constructions. With verbal predicates, negation is expressed by means of a negative verb, usually the verb nis 'not to be'. This negative verb is inflected, whereas the main verb appears in the so-called "negative stem form" (Hajdú 1975: 18). Compare: (29)
ni -dm harwcP not be -1SG want Ί don't want.'
(Hajdú 1963: 69)
246
Harrte
Wetter
In nominal and adjectival predicates we find the same negative verb his which, however, is used in a different syntactic construction. As opposed to verbs, nouns and adjectivals do not have a "negative stem form". Instead, his is followed by an — otherwise "virtual" — copula (cf. Hajdú 1975: 18) which surfaces here in its negative stem form ngcP. Negative constructions take the following form. The predicative noun/adjectival appears in its inflected (affirmative) form, followed by the — also inflected — negative verb his which, in turn, is followed by the negative stem form of a copula, ngcP'. This ngeP is optional and is usually omitted: 16 (30) a.
b.
pidar wesako -η nt -n you old -2SG not be -2SG 'You are not old' (lit.: you are old, man xasawa -dm ni -dm I man -1SG not be -1SG Ί am not a man' (lit.: I am a man,
(ngcP) COP you are not) (ngcP) COP I am not)
(Hajdú 1970: 98)
(Hajdú 1975: 18)
To summarize these observations, we find a fairly uniform pattern in the encoding of verbal, adjectival, and nominal predicates in Yurak. Yet, predicative nouns and verbs are formally distinguishable, albeit that the difference is not as clear as in many other languages. As far as verbs and nouns display distinctive grammatical characteristics, adjectivals appear to cluster with the nouns and not with the verbs. Therefore, adjectivals in Yurak can be considered "nouny". A few other languages in my sample are comparable with Yurak in the sense that they adopt — to a greater or lesser degree — one single morphosyntactic strategy in the encoding of verbal, adjectival, and nominal predicates. In each of these languages the uniformity of encoding is not absolute, and breaks down, so that verbal and nominal predicates can be formally distinguished. This differentiation between verbs and nouns enables us to determine the orientation of predicative adjectivals towards verbs (as in Kalispel and Kusaiean) or nouns (as in Basque, Chamorro and Turkish). A final remark concerns the type-membership of the sampled languages in the typology to be presented. Most languages in the sample can be characterized by one type of adjectival construction. Prototypical property concepts all belong to the same class of adjectivals, for which one single type of predicative adjectival construction is identified. In a considerable number of languages, however, prototypical adjectivals occur in two or more different types of predicative constructions. This phenomenon of "multiple typemembership" manifests itself in two different ways. First, within a given language a split may be found in the formal encoding of prototypical adjectival
A typology of predicative
adjectival constructions
247
concepts, so that they are distributed across different lexical categories. Each class of adjectivals is characterized by one single type of predicative construction. This situation is found, for instance, in Nkore-Kiga. In this language "adjectives are few in number, the function of the adjective being carried out in general by descriptive verbs." (Taylor 1985: 85). Members of the "adjective" class are "nouny", and occur in the same predicative constructions as nouns do. Consider the following examples, in which "adjectives" (a) and nouns (b) are accompanied by the overt copula -ba 'to be'. (31) a.
b.
m 1SG Ί am m 1SG Ί am
-ba omu -COP CL tall.' -ba omu -COP CL a doctor.'
-raingwa -tall
(176)
-fumu -doctor
(38)
Prototypical adjectival concepts not included in the "adjective" class are encoded as descriptive verbs which, when used as predicates, "occur in the appropriate tense, and behave like any other verb" (Taylor 1985: 175). (32) a.
b.
ekitabo ni -ki -tukura book PRES.CONTIN -CL.SUBJ -red 'The book is red.' ni -η Mbarara PRES.CONTIN -1SG.SUBJ -go Mbarara Ί am going to Mbarara.'
(175)
(10)
In the typology, languages like Nkore-Kiga are indicated by an "*" (e.g.: Nkore-Kiga*). A second form of multiple type-membership concerns the use of the same adjectival elements in different types of constructions. It should be noted that it is not unusual for a language to have different types of predicative constructions with the same adjectival item. Generally, however, the use of an alternative construction type involves the derivation of a secondary adjectival form with a different categorial status. As an example, in West Greenlandic (Fortescue 1984) most prototypical adjectival concepts are encoded as stative verbs. These stative verbs are used predicatively just like any other verb. Compare: (33) a.
quianar -puq amusing -3SG.INDIC 'He was amusing.'
(302)
248
Harrie
b.
Wetter
isir -puq come in -3SG-INDIC 'She came in.'
(120)
In addition, the participial form of the stative verb in (33a), quianar-tu (amusing-PARTIC), may also be used predicatively and occurs in the same predicative construction as nouns do (see example (34)). In Greenlandic, nominal predicates are formed by attaching a "verbalizing suffix" -u 'to be' to a nominal base, the result being a derived verbal form. Compare: (34) a.
b.
quianar -tu -u -vuq amusing -PARTIC -be -3SG.INDIC 'He is amusing.' Maalia kalaali -u -vuq Maalia Greenlander -be -3SG.INDIC 'Maalia is a Greenlander.'
(302)
(211)
Whereas the expression with the non-derived verbal form (33a) is a simple affirmation about the narrative present ('he was amusing [on a particular occasion]'), the construction with the derived participial form (34a) indicates a characteristic or permanent quality ('he is amusing [in general]'). Other languages in the sample display a similar pattern of variation in the encoding of adjectival predicates. "Primary", non-derived, adjectivals are used in one type of predicative construction, whereas "secondary", overtly derived, forms may also be used predicatively and appear in an alternative construction. In the typology, this kind of variation will not be accounted for. In such cases, only predicative constructions with non-derived adjectivals will be taken into consideration. In the case of the adjectival constructions in West Greenlandic, for instance, (33a) is included in the typology, whereas (34a) is not. In some languages, however, the same adjectival items may occur in different predicative constructions, without any derivational process involved. In these languages no distinction can be made between "primary" and "secondary" adjectival forms, and both construction types are included in the typology. Mundari is a case in point (Langendoen 1967 a, b). When used predicatively, adjectivals may occur as the complement of the copula menaq 'to be'. The same copular construction is used to express nominal predicates: (35) a.
fado marang menaq -ko -akan -a man -PL tall COP -3PL -PERF -PREDICATOR 'The men have been tall.'
(b 85)
A typology of predicative
b.
adjectival constructions
249
en hodo -ko munda -ko menaq -ko -akan -a (b 83) that man -PL headman -PL COP -3PL -PERF -PREDIC 'Those men have been headmen.'
However, adjectivals appearing in constructions like (35a) may also be used as verbs, without further measures being taken. In example (36a), marang 'tall' is treated as an intransitive verb. According to Langendoen (1967 b: 85), examples (35a) and (36a) represent 'mere stylistic variants': (36) a. b.
hodo man 'The hodo man 'The
-ko -PL men -ko -PL men
marang -akan -a -ko tall -PERF -PREDICATOR -3PL have been tall.' dub -akan -a -ko sit down -PERF -PREDICATOR -3PL have sat down.'
(b 85) (a 44)
Thus, we see that in Mundari adjectivals are ambivalent with respect to their categorial status. Without any derivational process involved, adjectivals are "verby" in one predicative construction, and "nouny" in another. In the listings in the typology, languages like Mundari are indicated by a slash following the name of the language (e. g.: Mundari/).
3.2. The typology of predicative adjectival constructions The typology to be introduced in this section is based on a sample of one hundred languages. Predicative adjectival constructions are classified into five types, depending on whether, and in what way, adjectivals in the attested constructions display an affinity for the verbs, for the nouns, or for neither of these two categories. These five types will be presented in three subsections. In 3.2.1. and 3.2.2. I will deal with predicative constructions involving "verby" and "nouny" adjectivals respectively. In section 3.2.3. I will discuss some types of constructions with "distinct" adjectivals, i. e., adjectivals which, in the relevant predicative constructions, are neither "verby", nor "nouny" in character. 3.2.1. "Verby" adjectivals Type 1: Verbal adjectival predicates. The first type in our typology represents a widespread pattern in the formal encoding of predicative adjectival constructions. This type contains those languages in which prototypical adjectivals are "verby"; when used as pred-
250
Harrie Wetter
icates, adjectivals are treated as (some subclass of) intransitive verbs, whereas predicative nouns are not treated this way. 17 Big Nambas (Fox 1980) is a case in point. In predicative constructions, adjectivals are treated on a par with intransitive verbs. Verbs are obligatorily marked for subject by prefixed actormode portmanteau morphemes which occur in four sets, i. e., realis mode (past/present), irrealis mode (future), irreal condition, and imperative mode. Predicative nomináis do not attach these prefixes, and cooccur with the copula verb v'i 'to be'. Consider the following examples of (a) intransitive verbal predicates, (b) adjectival predicates, and (c) nominal predicates in Big Nambas. (37) a.
b.
c·
/
-vdrvdr
(48)
3SG.REAL -run 'He runs/ran.' i -Iii 3SG.REAL -big 'He is/was big.' a uni -ar i -v'i prapar REF.PART mother -their 3SG.REAL -COP sow 'Their mother is/was a sow.'
(48)
(117)
Other languages with verbal adjectival predicates include Babungo and Wappo. 18 Babungo (Schaub (1985): (38) a.
Lambí jwt
vis-t
(240)
Lambi comePERF in-front 'Lambi came first.' t>- ngwá' ngwäa bày box my redPERF 'My box is red.' c· ngwá lùu wánddng he COP man 'He is a man.'
(234)
(142)
Wappo (Li —Thompson 1977; Thompson, personal communication): (39) a.
c'ic'
-i
Fep'is
nahwelis
-khP
bird -NOM worm hold in mouth -STAT 'The bird is holding the worm in its mouth.'
(p· c.)
Λ typology of predicative b.
he
poi'e?
-i
k'ena
DEM boy -NOM tall 'This boy is tall.' c.
te
ce^(e^)
adjectival constructions
251
(p.c.)
-khP
-STAT (1977: 433)
kanituCma
he COP chief 'He is a/the chief.' Verbal adjectival predicates are found in the following languages: Abkhaz Bororo Dakota Korean Malagasy Mokilese Nuer Thai Wappo
Acehnese Burmese Guarani Kusaiean Mandarin Nakanai Samoan Tigak Wolof
Alabama Cambodian Kalispel Lahu Menomini Navaho Sundanese Turkana Yoruba
Big Nambas Chemehuevi Kiowa Lenakel Mojave Niuean Tamazight Vietnamese Yurok
Amharic* Ewe* Japanese* Nkore-Kiga* Vai*
Babungo* Fordat/ Kassena* Oromo/
Banda* Gola* Motu I Pala I
Chitimacha/ W. Greenl.* Mundari/ Shona*
3.2.2. "Nouny" adjectivale Type 2: Nominal adjectival predicates In many languages, adjectivals display an orientation towards the verbs (cf. type 1). The opposite affiliation, however, is also frequently observed. In a great number of languages adjectivals are "nouny"; prototypical adjectivals occur in the same predicative constructions as nouns do, as opposed to verbs. This situation is found, for instance, in Finnish (Fromm —Sadeniemi 1956). Predicative nouns and adjectivals do not participate in the verbal inflectional system and are obligatorily accompanied by a verbal copula, usually the copula olla 'to be'. Compare the following examples of verbal, adjectival, and nominal predicates: ( 4 0 ) a.
han
saapuu
he arrivePRES3SG 'He arrives.'
(App. III)
252
Harrie Wetter b.
c.
tjttö on pieni girl COP.PRES3SG small 'The girl is small.' ystävä -ni on pappi friend -my COP.PRES3SG vicar 'My friend is a vicar.'
(116)
(115)
Languages with nominal adjectival predicates, then, use a uniform encoding strategy for adjectival and nominal predicates, distinguishable from the way in which intransitive verbal predicates are expressed. 19 The formal expression of nominal adjectival predicates, however, may vary from one language to the other. In Finnish, for example (see (40)), an overt, verbal, copula is used to encode adjectival and nominal predicates. In Hausa (Abraham 1941), predicative nouns and adjectivals are generally accompanied by a non-verbal copula, viz., one of the copula particles ne (used with masculine singular and plural subjects) or ce (in case the subject is feminine singular). (41) a.
ya kan sp 3SG.M HAB come 'He regularly comes.' b- jakin nan k'ank'ane ne ass this small COP 'This ass is small.' c· Kano gari babba ne Kano town large COP 'Kano is a large town.'
(13)
(47)
(18)
As a final example, in the Northern Australian language Tiwi (Osborne 1974) nominal and adjectival predication is effected by mere juxtaposition, without the use of an overt copula. (42) a.
b·
c·
¿¡
-pangulimai
3SG.M.NONPAST -walk 'He's walking/ he'll walk.' nginaki palangdmwani pumpuni this dog good 'This dog is good.' anginaki pilimunga this road 'This is a road.'
(40)
(56)
(56)
A typology of predicative
adjectival constructions
253
Nominal adjectival predicates are found to occur in the following languages of the sample .20 Albanian Chamorro Finnish Hungarian Kilivila Mangarayi Mongolian Quechua Swahili Tiwi Yurak
Arabic (Cairene) Diyari Georgian Jabem Lamutic Maranungku Nama Russian Tagalog Tonkawa
Aranda Dutch Gumbainggir Jacaltec Lithuanian Margi Nez Perce Sentani Tajik Tubu
Burushaski Ekagi Hindi Kanu ri Luiseño Miskito Pipil Siroi Takelma Turkish
Amharic* Fordat/ Ika* Mundari/ Shona*
Babungo* Gaelic/ W. Greenlandic* Nkore-Kiga* Spanish/
Basque/ Gola* Japanese* Oromo / Tamil*
Chitimacha/ Hausa* Kassena* Pala / Vai*
Type 3: Periphrastic constructions with abstract nomináis In most languages with "nouny" adjectivals, "predicative adjectival constructions" are formally encoded in the same way as nominal predicates are (cf. type 2). In some languages, however, a different situation is found. Prototypical adjectivals, lexicalized as abstract nomináis, are not used syntactically as the predicate of the clause, but appear in some kind of periphrastic construction which represents the functional equivalent of sentences like "the man is tall" in English. Typically, these periphrastic adjectival expressions parallel constructions which indicate relations of possession. An example of a language which uses this strategy to encode "predicative adjectival constructions" is Hausa (Abraham 1941). This language has a restricted set of a dozen or so "adjectives" or "adjectival nomináis" which occur in the same predicative constructions as nouns do (see example (41)). In addition to this small class of adjectivals, other prototypical property concepts are encoded as abstract nomináis, such as fad'i 'width'; kjau 'goodness'; gir m a 'largeness'; etc. For the expression of "predicative adjectival constructions", these abstract nomináis may occur in a number of periphrastic expressions which — when used with non-abstract nouns — indicate relations of possession. One of these constructions is shown in (43). The abstract nominal denoting the intended
254
Harrte
Wetter
property forms part of a prepositional phrase in combination with the instrumental/comitative marker da 'with' (cf. (43a)). With non-abstract nouns, this construction is used to express possession, the noun in the prepositional phrase referring to the object possessed (see (43b)). (43) a.
kogin nan y a river this 3SG.M 'This river is wide.' b. ja -na 3SG.M -PROGbe 'He has ten horses.'
-na da fad'i -PROGbe with width (it-is with width) da doki goma with horse ten (he-is with ten horses)
(51)
(68)
Given that we have restricted ourselves in this typology to prototypical adjectivale, the number of languages having periphrastic constructions with abstract nomináis is very small. In my sample, this type of construction is found in only two languages: Hausa*
Motu /
3.2.3. "Distinct" adjectivals From the listings in sections 3.2.1. and 3.2.2. we can conclude that in most languages of the sample, adjectivals are either "verby" or "nouny" with respect to their grammatical behavior in predicative adjectival constructions. However, some languages in the sample have adjectival constructions which do not conform to this general pattern; adjectivals are clearly non-verbal, but they do not pattern like nouns either. Adjectivals appearing in these constructions are referred to as "distinct" adjectivals. In this section I will discuss two types of predicative adjectival constructions with "distinct" adjectivals. Type 4: Distinct adjectival predicates As a general cross-linguistic tendency, languages with non-verbal adjectivals appear to use one single strategy for the encoding of both adjectival and nominal predicates (cf. type 2). In some languages, however, different nonverbal strategies are used with predicative adjectivals and nouns respectively. Typically, the difference between adjectival and nominal predicates is indicated by the use of different copula morphemes. In some languages, however, the same copula is used for both types of predicates. In that case, adjectivals and nouns are treated in different ways in order to function as the complement of the copula in question. Adjectivals, for instance, have to be nominalized, as opposed to nouns.
A typology of predicative adjectival constructions
255
An example of a language with "distinct" adjectival predicates is represented by Vai (Weimers 1976). This language has a restricted set of nonverbal adjectivals which, when used as predicates, occur as the complement of the (locative) verb bè 'to be'. For the expression of nominal predicates a different copula is used, i. e., the "identificative morpheme" mù 'it is'. Consider the following examples of (a) verbal, (b) adjectival, and (c) nominal predicates in Vai. (44) a.
b.
c.
à nâ'à he comeSITUATIONAL 'He came.' kàiê bè kündú manDEF COP short 'The man is short.' tig nggóà ko'à Ιέέ-mdò my older brother weaver 'My older brother is a weaver.'
(129)
(77)
mù COP
(131)
Whereas "distinct" adjectivals in Vai are exclusively used with the copula bè (and not with the nominal copula mù), a different situation is observed in (Scottish) Gaelic (Anderson 1909, 1910). Predicative adjectivals occur in "nominal adjectival predicates" (cf. type 2). Both adjectivals and nouns may be accompanied by the copula is 'to be'. (45) a.
b.
is làidir e COP.PRES strong he 'He is strong.' is bean-tighe i sin COP.PRES housewife she that 'She is a housewife.'
(1910: 236)
(1909: 441)
In addition, however, adjectivals also appear in "distinct adjectival predicates", as the complement of the copula tha 'to be'. This copula does not allow a noun as its complement. Of the following examples, (46b) is ungrammatical. (46) a.
b.
tha e làidir COP.PRES he strong 'He is strong.' *tha e duine laidir COP.PRES he man strong *'He is a strong man.'
(1910:236)
(1910: 236)
256
Harris Wetter
Thus, we see that adjectivals in Gaelic are characterized by two different types of predicative constructions. Nominal adjectival predicates (cf. (45a)) indicate more permanent, time-stable properties, whereas distinct adjectival predicates (cf. (46a)) are used to express temporary states. The two languages discussed above, i. e., Vai and Gaelic, are representative of languages with distinct adjectival predicates. Generally, this type of construction is not the only option a language has to express predicative adjectival constructions. In some languages this construction-type is characteristic for only a restricted class of adjectivals (as in Vai). In other languages the same adjectivale occur in other predicative constructions as well (as in Gaelic). In fact, Hixkaryana (Derbyshire 1979) is the only language in the sample which is uniquely characterized by this type of adjectival predication. According to Derbyshire (1979: 42), "adjectives and adjective phrases do not occur. Adjective-like words are treated as adverbs". At first sight, adjectival and nominal predicates in this language are pretty much the same. Both types of predicates may be encoded either in the form of an "equative sentence" with a zerocopula, or by means of an overt copular construction with the verbal copula -exe-. However, adjectival predicates in Hixkaryana are classified as distinct adjectival predicates, because adjectivals and nouns have to meet radically different requirements in order to be used in these constructions. The equative sentence-type requires a nominal predicate. As opposed to nouns, adjectivals have to be nominalized in order to appear in this construction. (47) a.
b.
tewatxarkaxe -mi mok.ro playful -NMLZR that one 'That fellow is playful.' toto noro man he 'He is a man.'
(100)
(89)
The verbal copula -exe- requires an adverbial complement. Adjectival items, being in fact adverbial forms, can be used as complement of the copula without further measures being taken. Nouns, however, must be incorporated in an adverbial phrase which is formed by adding the postposition me 'denominalizer'. (48) a.
b.
tewatxarkax n- a -ha mokro playful 3- COP -NONPAST that one 'That fellow is playful.' toto me n- a -ha man DENOMLZR 3- COP -NONPAST 'He is a man.'
(100)
(89)
A typology of predicative
adjectival constructions
257
Copular and equative sentences formed with nouns can be substituted for each other without any significant change of meaning. The adjectival constructions (47a) and (48a), however, differ in meaning. The construction with the nominalized adjectival item, i.e., (47a), is used for the assignment of more general or inherent properties. Distinct adjectival predicates occur in the following languages: Hixkaryana Basque/ Mundari/
Ewe* Spanish/
Gaelic/ Tamil*
Ika* Vai*
Type 5: Periphrastic constructions with attributive adjectivals In a few languages of the sample, (subclasses of) adjectivals can only function as noun modifiers and do not occur as predicates at all. The functional equivalent of, say, 'the man is tall', is formally expressed by means of a predicate noun phrase, the head of which is modified by the adjectival in question, i. e., 'he/this man is a tall man' for 'he/this man is tall'. The head of the predicate noun phrase is generally some kind of non-informative "dummy" noun, such as "man", "child", or "thing". This type of construction is found, for instance, in Babungo (Schaub 1985: 256) with a small subclass of adjectives: "The adjectives m'ä' 'right'; ngkwîi 'left'; kwî 'important'; day 'different'; and ngò' 'old' can only be used attributively. That is, it is not possible to say 'this thing is important'; one has to say 'this thing is an important thing.' " (49)
nú kê lùu nú km thing this COP thing important 'This thing is (an) important (thing).'
(256)
Periphrastic constructions with attributive adjectivals are found in the following languages: Babungo*
Banda*
Kassena*
4. Conclusion In this paper I have argued that the cross-linguistic variation in the formal encoding of adjectival concepts cannot adequately be described in terms of the traditional parts-of-speech "adjective", "noun", and "verb".
258
Harrte Wetter
Starting from the observation that adjectivals tend to be either "nouny" or "verby", I have introduced a typology of predicative adjectival constructions, based on a sample of one hundred languages. In this typology, predicative adjectival constructions are classified into five types. The distribution of the sampled languages over these five types indicates a clear predominance of two major types, viz. verbal adjectival predicates (type 1), and nominal adjectival predicates (type 2). With the exception of Hixkaryana, all languages in the sample have the possibility to encode predicative adjectival constructions either like (intransitive) verbal predicates, or like nominal predicates. The remaining three types (3 — 5) can be considered minor types. They are attested in relatively few languages which — except for Hixkaryana — have adjectival constructions of one of the major types as well. These findings justify the formulation of the following cross-linguistic tendency: crosslinguistically, a major split can be observed in the formal encoding of predicative adjectival constructions. Either adjectivals are "verby", and are treated on a par with (intransitive) verbs, or they are "nouny", and occur in the same predicative constructions as nouns do. This paper is a report on work in progress. The typology presented here provides the descriptive basis for explanatory research concerning first the occurrence of the attested types, and second the distribution of languages over these types. The focus of further investigation will be on the fundamental split between "verby" and "nouny" predicative adjectivals.
Notes 1. I wish to thank Ineke Brus, A n n Cooreman, A d Foolen, and Leon Stassen for many helpful comments on earlier versions of this paper. I am also grateful to Mark Durie, Kees Hengeveld, Rüssel Schuh, and Sandy Thompson for providing the data for some languages in my sample. For the value of the symbols * and / after language names, refer to the discussion of the phenomenon of "multiple type-membership" on pages 246—249. 2. Albeit that the distinction between verbs and nouns is not equally clear in all languages; see section 3.1. 3. In addition to the word class status of adjectivals, a second parameter concerns the open vs. closed character of the adjective class (cf. Schachter's group 2). This parameter is relevant in the sense that languages with a small closed set of "real" adjectives are characterized by minimally t w o "types" of lexical categorization: adjectival concepts which are not included in the small adjective class are generally encoded as verbs and/or nouns. In sections 1 and 2 of this paper the open vs. closed character of adjective classes is neglected, as it is irrelevant to the discussion to be presented. 4. Throughout this paper the figures in parentheses following each example refer to the number of the page in the grammar where the example can be found.
A typology of predicative
adjectival constructions
259
5. Adjectival past tense forms may refer to both present and past states. In fact, adjectival predicates referring to present states are preferably expressed by means of past tense forms (cf. Beaumont 1980: 40). 6. I^/lLidators are independent suffixes ("independent" in this sense means that they can be used with all parts of speech) which indicate authority for assertion and degree of certainty. The validator -mi used in the examples cited refers to "first-hand information" (Cole 1982: 163-164). 7. It must be noted that the suffix -ta is also the accusative case marker (cf. example (6)) and one of the primary locative morphemes (added to nomináis), meaning "motion through or past", as in: (i) pungu -ta ri -rka -ni (119) door -LOC go -PAST -1 Ί went through the door.' 8. For a detailed discussion of the differences between the constructions in (16a) and (16b), see L i - T h o m p s o n 1981: 1 1 9 - 1 2 3 . 9. One of the differences between Japanese adjectives and verbs is illustrated in example (19). Syntactically, adjectives do not combine with auxiliary verbs such as ir«, shimau, and kureru to form expressions conveying distinctions of aspect, benefaction, etc.: compare tabete-iru 'be eating' (example (19) in the text), tabete-shimau 'eat completely', tabete-kureru 'eat for me', etc. Other grammatical differences between these verb-like adjectives and verbs include the form and number of inflectional endings (verbs having a larger range of forms), the absence of regular morphological passive, causative and potential forms with adjectives, etc. (cf. Backhouse 1984: 170-171). 10. A similar situation is found in Mojave and Mandarin Chinese (see section 2.1.). 11. For a similar view, see Plank (1984). 12. It should be emphasized that, for the purpose of a language-specific grammatical description, the uncertainty concerning the word class status of adjectivals is not necessarily problematic. Whether adjectivals should be analyzed as a separate category or as a subclass of verbs or nouns is not the most important issue in a descriptive grammar. What really counts is that the analyst accurately describes all relevant grammatical properties of adjectivals and that he explicitly states the criteria for his classification of adjectival words. In this way the reader of the grammar is able to make his own judgement on the proposed classification. In this respect, consider for example the way adjectives are discussed in the grammars of Nkore-Kiga (Taylor 1985) and Tigak (Beaumont 1980): in both cases the observed grammatical similarities between adjectives (defined as a distinct class) and nouns/verbs are also given considerable attention (see section 2.1.). 13. For a similar view on the cross-linguistic variation in the formal encoding of adjectival concepts, see Thompson 1988. 14. I suggest the following generalization here. If predicative adjectivals show verb-like or noun-like properties, this is not necessarily the case for attributive adjectivals as well. If, on the other hand, adjectivals used as noun-modifier behave like verbs or nouns, they will also display verby or nouny characteristics when used predicatively. 15. It has struck me that in many cases these noun-like color terms seem to be the result of semantic bleaching of words which originally referred to objects characterized by a specific color, such as orange (fruit) > orange; ashes > grey; coal > black; unripe melon > green; gold > yellow, etc.
260
Harrte
Wetter
16. Note the parallelism between ni-dm harwcP Ί don't want' in (29), and nt-n (ngcP) 'you are not' and ni-dm (ngiP) Ί am not' in (30a —b). According to Hajdú (1970: 97 — 98) the negative construction with nouns and adjectivals developed from a fusion of two sentences, i. e., the affirmative sentence with the predicate noun/adjectival, e. g., 'you are old', and the negative sentence with a verb 'to be' as its main verb, e. g., 'you are not'. 17. In a few languages of the sample (Kalispel, Kusaiean) the distinction between verbal/ adjectival predicates and nominal predicates is admittedly rather vague. As far as verbal and nominal predicates can be formally distinguished, adjectivals cluster with the verbs, and not with the nouns (see section 3.1.). 18. Other examples of verbal adjectival predicates are found throughout this paper: Tigak (3), Mojave (13), Mandarin Chinese (15), Japanese (18), Sundanese (20), Nkore-Kiga (32), West Greenlandic (33), and Mundari (36). 19. In some languages with nominal adjectival predicates, the distinction between verbal and adjectival/nominal predicates is relatively vague. However, insofar as verbal and nominal predicates can be formally distinguished, adjectivals pattern like nouns, and not like verbs. This situation is observed in Basque, Chamorro, Turkish, and Yurak (see section 3.1.). 20. In previous sections of this paper, other examples of nominal adjectival predicates can be found: Nkore-Kiga (2), Imbabura Quechua (7), Cairene Egyptian Arabic (24), Yurak ( 2 7 - 3 0 ) , Mundari (35).
References Abraham, R. C. 1941
A modern grammar of spoken H ausa. London: Published on behalf of the Government of Nigeria by the Crown Agents for the colonies.
Anderson, A. O. 1909 "The syntax of copula IS in modern Scottish Gaelic", Zeitschrift für keltische Philologie 7: 4 3 9 - 4 4 9 . 1910 "The syntax of the substantive verb THA in modern Scottish Gaelic", Zeitschrift für keltische Philologie 8: 236—241. Backhouse, Α. E. 1984 "Have all the adjectives gone?", Lingua 62: 169 — 186. Beaumont, Clive H. 1980 The Tigak language of New Ireland (= Pacific Linguistics Series Β — no. 58). Canberra: The Australian National University. Castrén, M. Alexander 1966 Grammatik der samojedischen Sprachen ( = Uralic and Altaic Series S3). Bloomington: Indiana University/The Hague: Mouton. [Reprint of St. Petersburg 1854: Kaiserliche Akademie der Wissenschaften.] Cole, Peter 1982 Imbabura Quechua (= Lingua Descriptive Studies no. 5). Amsterdam: North Holland. Crystal, David 1967 "English", Lingua 17: 2 4 - 5 6 . Derbyshire, Desmond C. 1979 Hixkaryana ( = Lingua Descriptive Studies no. 1 ). Amsterdam: North Holland.
A typology Dixon, R. M. W. 1977
of predicative
adjectival
"Where have all the adjectives gone?", Studies in Language,
constructions
261
1.1: 19 — 80.
Fortescue, Michael 1984 West Greenlandic (Croom Helm Descriptive Grammars). London: Croom Helm. Fox, G. J . 1980 Big Ν ambas grammar ( = Pacific Linguistics Series Β — no. 60). Canberra: The Australian National University. Fromm, Hans —Matti Sadeniemi 1956 Finnisches Elementarbuch. I: Grammatik. Heidelberg: Winter. Gary, Judith Olmsted — Saad Gamal-Eldin 1982 Cairene Egyptian colloquial Arabic ( = Lingua Descriptive Studies no. 6). Amsterdam: North Holland. Givón, Talmy 1979 1984
On understanding grammar. New York: Academic Press. Syntax. A functional-typological introduction vol. 1. Amsterdam/Philadelphia: John Benjamins.
Gleason, H. A. Jr. 1961 An introduction to descriptive linguistics (revised edition). London and New York: Holt, Rinehart and Winston. Hajdû, Péter 1963 The Samoyedpeoples and languages. ( = Uralic and Altaic Series 14). Bloomington: Indiana University/The Hague: Mouton & Co. 1970 "Zur Syntax der negativen Verbalformen im Samojedischen", in: Wolfgang Schlachter (ed.): 9 0 - 1 0 6 . 1975 "Predikative Nominalflexion in den samojedischen Sprachen", Acta Linguistica Academiae Scientiarum Hungaricae 25: 1 — 30. Hall, Robert A. Jr. 1964 Introductory linguistics. New York: Chilton Books. Hardjadibrata, R. R. 1985 Sundanese: a syntactical analysis. ( = Pacific Linguistics Series D — no. 65). Canberra: The Australian National University. Hawkins, John A. (ed.) 1988 Explaining language universals. Oxford: Basil Blackwell. Hopper, Paul J . —Sandra A. Thompson 1984 "The discourse basis for lexical categories in universal grammar", Language 60: 7 0 3 - 7 5 2 . Jacobsen, William H., Jr. 1979 Noun and verb in Noothan (— Heritage Record 4). Victoria: British Columbia Provincial Museum. Langendoen, D. Terence 1967a "Mundari verb conjugation", Linguistics 32: 39 — 57. 1967b "The copula in Mundari", in: John W. M. Verhaar (ed.), 7 5 - 1 0 0 . Li, Charles N. (ed.) 1977 Mechanisms of syntactic change. Austin and London: University of Texas Press. Li, Charles N. — Sandra A. Thompson 1977 "A mechanism for the development of copula morphemes", in: Charles N. Li (ed.), 4 1 9 - 4 4 4 .
262
Harris
1981 Locker, Ernst 1951
Wetter Mandarin Chinese: A functional reference University of California Press.
grammar.
Berkeley and Los Angeles:
Nominales und verbales Adjektivum ( = Beihefte %ur Zeitschrift für "Die Sprache", Heft 1). Wien —Meisenheim: Verlag A. Sexl.
Sprachwissenschaft
Lyons, John 1968 Introduction to theoretical linguistics. Cambridge: Cambridge University Press. 1977 Semantics. Vol. I and II. Cambridge: Cambridge University Press. Matisoff, James A. 1973 The grammar of Lahu. Berkeley/Los Angeles/London: University of California Press. Munro, Pamela 1976 Mojave syntax. New York and London: Garland Publishing, Inc. Osborne, C. R. 1974 The Tiwi language. Canberra: Australian Institute of Aboriginal Studies. Plank, Frans 1984 "24 grundsätzliche Bemerkungen zur Wortarten-Frage", Leuvense Bijdragen 73: 489-520. Robins, Robert H. 1952 "Noun and verb in universal grammar", Language 28: 289 — 298. 1964 General linguistics: an introductory survey. London: Longmans. 1965 "Some typological observations on Sundanese morphology", Lingua 15: 435 — 450. 1968 "Basic sentence structure in Sundanese", Lingua 21: 351—358. Schachter, Paul 1985 "Parts-of-speech systems", in: Timothy Shopen (ed.), 3 — 61. Schaub, Willy 1985 Bahungo (Croom Helm Descriptive Grammars). London: Croom Helm. Schlachter, Wolfgang (ed.) 1970 Symposion über die Syntax der uralischen Sprachen ( 15. — 18. futi 1969 in Hemhausen bei Göttingen). Göttingen: Vandenhoeck — Ruprecht. Shopen, Timothy (ed.) 1985 Language typology and syntactic description, Vol. 1. Cambridge: Cambridge University Press. Stassen, Leon in press "A hierarchy of main predicate encoding". This volume. Taylor, Charles 1985 Nkore-Kiga (Croom Helm Descriptive Grammars). London: Croom Helm. Thompson, Sandra A. 1988 "A discourse approach to the cross-linguistic category 'adjective'", in: John A. Hawkins (ed.), 1 6 7 - 1 8 5 . Verhaar, John W. M. (ed.) 1967 The verb 'be' and its synonyms. Part 1. Dordrecht: Reidel. Weimers, William E. 1976 A grammar of Vai. Berkeley: University of California Press.
Part III Morphemes and lexical items
Demonstratives and the localist hypothesis Walter De Mulder
1. Introduction Can the localist hypothesis of demonstratives be maintained? In other words, can their local-pointing sense be considered as their basic sense? Recently, this "theory of the pointing finger" has been severely criticised by Kleiber. Yet our thesis is that the localist hypothesis is still attractive from a cognitive point of view, if two provisions are made. First, it must be clearly specified what is meant by "basic sense", and second, demonstratives cannot be expected to express exactly the same relations in the concrete domain of space and in the more abstract discourse domain. We will argue this thesis by discussing a classical formulation of the localist hypothesis by Bühler (1934) and the criticisms of Kleiber (1983). Finally, we will try to describe and explain the regularities in the textual use of the demonstrative pronouns, adverbs or clitics in widely divergent languages.
2. The localist hypothesis Our starting point will be the following classical formulation of the localist hypothesis by Bühler (1934: 121-122): From a psychological point of view, the anaphoric use of the pointing words presupposes that sender and receiver have the discourse as a whole in front of their eyes, so that they can refer forward and backward to its parts. The whole discourse must then be accessible to sender and receiver to the extent that a kind of scanning is possible, comparable to the scanning of a present object with the eyes. [Translation Bosch 1983: 12]
One could provide more sophisticated, up to date formulations of the localist hypothesis, quoting from recent work by Johnson, Lakoff and others, but Bühler's formulation has the advantage of stating very clearly the two parts of the hypothesis: 1. Demonstratives are pointing words: they function as "Wegweiser des Blickes" (Bühler 1934: 122), they point to a visually present object, which is there before the eyes, hence the term "Deixis ad oculos".
266
Walter De Mulder
2. The use of demonstratives as anaphors is an extension of this fundamental use: the discourse is conceived as a visual space before the eyes, in which one can point to words in the same way as one points to things in space. It has been widely recognised that the structuring of space is of primordial importance for human cognition: things in space are concrete and clearly delineated, and the structures which a human being develops in his interaction with the world around him can be used to structure more abstract and less neatly structured domains such as, e. g., time and language. 1 We will be especially concerned with the transfer of the spatial pointing sense of the demonstratives to the discourse domain, since it is precisely the possibility of using this transfer to explain the anaphoric use of demonstratives that has been criticised by Kleiber.
3. The criticism of the localist hypothesis by Kleiber Kleiber (1983) criticises the two parts of the localist hypothesis: he denies that the basic sense of the French demonstratives is the associated ostension and that the spatial use of demonstratives can explain the anaphoric use. In his article, he criticises the thesis of the ostensive sense of the demonstratives, which is exactly the one Bühler has stated, according to which demonstratives, as their name implies, serve to indicate, to point to something. As such, they would replace a pointing gesture and the opposition between French -ci and -là would be an opposition in localisation between "near the speaker" and "far from the speaker", respectively. Kleiber claims that this "strong version" of the ostensive theory can easily be dismissed because it is impossible to omit the pointing gesture while using a demonstrative: when one says this man and omits the pointing gesture, noone will ever know who is being talked about. So the orientation associated with the use of a demonstrative does not seem to be part of its sense: it is given by the gesture the speaker makes. To save the ostensive theory, some propose that the pointing gesture itself is in some way part of the sense of the demonstrative. But this implies that each demonstrative has an infinite number of senses, an idea which may well be acceptable to a logician, 2 but less so to a linguist, who seeks a unifying sense for all uses of demonstratives. Why not propose, then, that this unifying sense of the demonstrative be a kind of pointing sense which would not contain any indication as to the precise direction in which the object should be located? This sense could
Demonstratives and the locatisi hypothesis
267
then be paraphrased as qui est ici — when one argues that demonstratives serve basically to localise a referent, to say where the demonstratum is — or as que je montre — when one argues that their basic use is to identify a referent, to say what is being demonstrated. Thus, in this "weak version" of the localist hypothesis, the demonstrative functions as a kind of signal: it says that something is being indicated, without specifying the direction of the associated demonstration. In the course of his discussions of the "weak version", Kleiber explicitly rejects the localist hypothesis for two reasons: (a) the derived anaphoric senses are more frequent than the spatial ones; (b) the pointing gesture is omitted in the transfer to the discourse domain; therefore, the basic sense only contains the information that the demonstratum has to be searched for by means of the present discourse situation. This means however that the difference between the demonstratives and the other indexical symbols such as je, tu, ici and maintenant, whose referent must also be identified by making use of the current discourse situation, is obscured. As to objection (a), we would like to point out that the basic sense does not necessarily have to be the most frequent. The spatial sense can be called "basic" because it is the first sense to develop in the ontogenetic development: it arises from the interaction that we as bodily beings have with the environment (Johnson 1987; Lakoff 1986). So we first develop the spatial meaning and then the anaphoric sense as we learn to engage in human conversation, as is suggested by Lyons (1975) and confirmed by Atkinson (1979: 234). This would imply that the anaphoric sense becomes more frequent than the spatial sense as we learn to use demonstratives and other devices to give coherence to discourse. In fact, this result could also be interpreted as saying that the child and the adult have a different system, the spatial sense being fundamental for the child and the anaphoric fundamental for the adult. But then it still remains a mystery how the two senses are connected. Therefore it seems more appealing to integrate the two senses in one model and to explain the connections between them by starting from the ontogenetically prior sense, which is the local sense, and explaining how the second develops from it as a result of motivated transfer mechanisms (Fillmore 1982). Objection (b) states that the anaphoric use of demonstratives cannot even be explained by the "weak" pointing sense that accounted for their use in the concrete spatial domain. Therefore, since the pointing sense cannot explain all uses of the demonstratives, it is not basic. But in our view, saying that a sense is basic does not imply that this sense is present each time a demonstrative is used. Should one nevertheless hold on to this conception, it seems
268
Walter De Mulder
clear that the sense of a demonstrative must be reduced to the rule that the referent is to be searched for by means of the present context of utterance. We then have a sense that is available in each context of use. However, one may wonder whether this kind of sense is adequate when thinking about natural language. Should the sense of a word not be conceived of as a "stereotype", a rule of thumb usually followed unless there is good reason to depart from it? (Bosch 1983: 55). Bosch (1985) shows that each "sense" is necessarily a sense-in-context, always open to revision when the word is used in contexts in which it had not been used before. This conception of "sense" allows us to hold on to the localist hypothesis, since we do not have to accept that the basic locational sense is present every time the demonstrative is used. According to this view, the local sense is the one the child learns first as a "rule of thumb" — for instance in presentational sentences — and afterwards, the rule thus apprehended eventually gets changed when the word is used in other contexts. In this way, the basic sense of the demonstrative is locational and is no longer confused with that of the other deictics.
4. The sense of demonstratives in discourse This section is devoted to showing how demonstratives lose their local pointing sense when they are used in discourse and how strategies of discourse comprehension suggest explanations as to why these transfers from the basic local to the anaphoric sense occur. So, if we accept for the time being the frame for a description of local deixis that was developed by Fillmore (1982), stating that the spatial deictic systems of all languages are integrated in a system that distinguishes three distances: D(+Proximal), D( + Medial) and D( + Distal), the question is: what happens when this system is transferred to the discourse space? According to Bühler, demonstratives point forward or backward in the discourse space when they are used anaphorically. This idea seems to work well with textual deixis as exemplified in (1) (cf. Lyons 1980: 289). (1)
X declares: Y replies:
Ça c'est un rhinocéros. Un quoiÌ Tu peux me l'épeler?
Here the stretch of discourse formed by the two sentences can indeed be seen as being "before the eyes" and /' points back to the word, the element of the discourse, mentioned before.
Demonstratives and the localist hypothesis
269
This description however does not apply to anaphors. Consider for instance (2).
(2)
Un homme entra, un chapeau de femme sur la tête. Cet homme était fou.
As Bühler no doubt already realised, in an example like this, cet homme does not really point to the element of the linguistic string that is the word token un homme. Kleiber (1983, 1984, 1986 a/b, 1987 a/b) and Corblin (1986) have tried to show that ce is an instruction to the hearer to identify the referent by making use of the context and the cotext of the utterance, the sole function of the noun being to reclassify the referent, which explains why ce lit can refer back to le plancher du wagon in (3). (3)
Ils considérèrent le plancher du wagon et décidèrent que pour une nuit, ce lit ferait bien l'affaire.
However, if the demonstrative NP finds its referent by making use of the relevant elements of the context, shouldn't we accept that what enables ce lit to refer back to le plancher du wagon is the simple fact that the only element which can be reclassified as a bed in this context is the floor of the train wagon? If so, then the noun does a lot more than simply reclassify a referent already identified by the demonstrative, since in this example, it also plays a role in the identification of the referent. 3 Moreover, this example shows that a lot of inferences are made in the recovery of the referent, whereby the hearer relies on background and contextual knowledge. These inferential processes also explain why anaphors need not be coreferential with their antecedent as can be seen in (4). (4)
Le sapin que Jean a planté dans son jardin a crevé. Ce sapin ne résiste jamais à la chaleur estivale de la plaine.
In this sentence, ce sapin refers to a type of pine tree, and has generic reference, whereas le sapin que Jean a planté refers to a particular pine tree (a token). Nunberg (1978: 29) has established that this is possible because in our understanding of the world, there is a "referential function" that links demonstrata to referents and that goes from type to token and vice versa. If we now reconsider (1), our example of textual deixis, we see that it makes use of the same strategy: what has to be spelled out is not the token of the word rhinocéros but the type (Conte 1981: 123 — 124). Thus even in textual deixis the process is not as simple as the Bühler model claims. What seems to go wrong is that Bühler has a semiotic model of the linguistic sign, in which a form is linked with a meaning "like two sides of a piece of paper". In fact, linguistic signs must be interpreted, and in doing
270
Walter De Mulder
this, the hearer must use linguistic as well as extra-linguistic k n o w l e d g e (the b a c k g r o u n d ) . In other terms, linguistic signs are always interpreted in a certain co(n)text, which can be broad or n a r r o w , implicit or explicit. W h e n the appropriate co(n)text is missing, the anaphoric demonstratives are difficult to interpret and therefore less acceptable, as has been signalled by Corblin (1986) w h e n he c o n f r o n t e d (3) with (5). (5)
?Les ouvriers réparaient le plancher du wagon et enfonçaient des clous dans ce lit.
It appears that it is the first part of the sentence in (3) that motivates and thereby clarifies the use of the w o r d lit in the second part of the sentence. Bosch (1983) has stressed that we thus interpret every sign, including the demonstrative, in the light of a "context m o d e l " . So demonstratives d o not directly catch up with reality: they find their referent in the context model and it is this context model which can b u t need n o t be matched w i t h reality. In this conception, reference is negotiated: the speaker must make use of what he thinks the hearer k n o w s , of w h a t he thinks is already in the hearer's context model. M o r e specifically, in the course of the conversation, it is of the u t m o s t importance that speaker and hearer speak a b o u t the same thing, that they focus, direct their attention o n the same referent. T h e r e f o r e , the speaker must indicate w h e t h e r he continues to focus on the same referent. In other w o r d s , w h e n the speaker uses demonstratives, t w o procedures are possible: the anaphoric p r o c e d u r e , which serves to make the hearer continue to focus on the same referent; the deictic p r o c e d u r e , which serves to reorientate the attention of the hearer to a new focus (Ehlich 1979). O u r thesis is that it is these procedures that are universally m a r k e d by demonstratives. W i t h this hypothesis in mind, we will consider some data f r o m such widely divergent languages as Japanese, Eipo, D u t c h , English, and French. In Japanese the distinction between the ^»-series (kore: 'this one'), the joseries (sore: 'that one') and the Λ-series (are: 'that one there') of demonstratives has been described by K u n o (1973: 290) as follows: (i) The ö-series is used for referring to something (at a distance either in time or space) that the speaker knows both he and the hearer know personally or have shared experience in. (ii) The .«-series is used for referring to something that is not known personally to either the speaker or the hearer or has not been a shared experience between them. 4 (iii) The /éo-series is used semianaphorically as if the object being talked about were visible and were at the speaker's side. The hearer cannot refer to the same object by using the ^«-series in spite of the fact that in the ordinary demon-
Demonstratives
and the localist hypothesis
271
strative use of the /èo-series the same object can occasionally be referred to with ko by both the speaker and the hearer. Once the understanding is established between the t w o that both k n o w the object well, the a-series takes precedence, and the use of the /êo-series results in unacceptability.
The difference between a- and so- can be observed in the following example. 5 (6) a.
Speaker A: Kinoo Yamada toyuu hito ni aimasita. yesterday as named person met. Sono (*ano) hito, miti ni mayotte komatte-ita person way in lose was-in-trouble node, takusete age m asita. because helping gave (in favor of). 'Yesterday, I met a man by the name of Yamada. Since he had lost his way and was having difficulties, I helped him.'
Speaker A knows that speaker Β does not know the referent, so he uses so-. Β now has to answer with so-, because the use of a- would lead to a contradiction, as can be observed in the translations. (6) b.
Speaker B: Sono hito, hige o hayasita tyuunen no hito desyoo? person beard grew middle-aged person is 'That person that you have mentioned whom you have assumed that I do not know — isn't he a middle-aged man with a beard?'
If sono were replaced by ano, the translation would be: (6) c.
'That man that both you and I know very well — isn't he a middle aged man with a beard?'
In the Papuan Eipo language, the demonstrative prefixes a- (here) and da(there) serve to mark topic-comment structures in combination with the suffix -ra. Ara is then used to order the information in successive utterances in which the grammatical subject remains the same. Dara on the other hand is used if there is a change of subject or a change of speaker within a preestablished field of things meant or things spoken of (Heeschen 1982: 87 — 88). (7)
Dingerkon nang ara dabosib balamak gum. Dingerkon people ara lower-regions they-are-usually-going not 'The people of Dingerkon usually don't go to the lower regions of the north.'
272
Walter De Mulder
(8)
Arebkin dara arebnilyam I-have-given-to-you dara you-give-to-me Ί have given to you,· now give to me!'
In Dutch die 'that' serves to mark low deixis, 6 "lesser urging that the hearer find the referent" (Garcia 1975: 65), whereas de^e 'this' marks high deixis, "greater urging that the hearer find the referent" (Garcia 1975: 65). One of the parameters that thus gives rise to a difference in force of deixis is givenness: the speaker will direct the hearer's attention strongest to entities that are not given, in the hearer's consciousness (Kirsner 1979: 360). When, for instance, a noun is used for the first time, its referent is not as given as it will be upon repetition. Therefore, nouns with totally new referents will attract de%e more often than die. Moreover, when a noun is used again, the givenness of its referent may have "decayed" because the referent was introduced a long time before and has not been repeated since. In this case also, higher deixis is required, and de^e should be more frequent than die. Kirsner (1979: 364 — 365) shows that these expectations can be confirmed by statistical means. In languages in which the demonstratives have evolved into definite articles, these serve to mark the anaphoric procedure. 7 To quote Traugott (1982: 250) for English: The definite article the derives in E[arly] Mfiddle] Efnglish] from the nonproximal demonstrative that of 0[ld] Efnglish]. The function of demonstratives is primarily to identify or index things in the world and relate them to the speaker's point of view; the function of the is to signal that whatever is being talked about has already been referred to, or to treat it as if it were already salient in the hearer's consciousness.
Kleiber (1986 b, 1987 a/b) has demonstrated how a parallel difference separates demonstrative from definite phrases in French: demonstrative phrases serve to attract attention to a new referent or a referent that is presented in a new way, whereas definite phrases continue to present the referent as it had been introduced before (the French definite article is derived from Latin Ule). From this perspective, Kleiber compares (9) and (10). (9) (10)
Un avion s'est écrasé hier. ?Cet\l'avion venait de Miami.8 Un avion s'est écrasé hier. Cet/?/'avion relie habituellement York.
Miami à New
In (9) the plane in question clearly is the one that crashed the day before: it is the plane as it was introduced in the first sentence of (9). The second sentence, one could say, is a comment on the first one. In (10) on the contrary,
Demonstratives
and tbe locatisi hypothesis
273
the word habituellement means that the second sentence should not be understood as a comment on the first: the plane the speaker wants to talk about is not the one talked about in the first proposition. Ce indicates that the hearer should not look for the referent in the first sentence, but that he should use the first sentence as a context which enables him to find the referent. In showing that languages use demonstrative forms to express the deictic and anaphoric procedures in text comprehension, the foregoing descriptions also show that the demonstratives do not project their spatial sense as such upon the discourse. In fact, the identification of the referent proceeds by other means in the concrete spatial domain than in the discourse domain: whereas in the concrete spatial domain, distances seem to play an important role, what seems to be important in the discourse domain is whether the referent has already been talked about and whether the speaker may assume that the hearer still has the referent in his context model. In other words, as demonstratives get used in the discourse domain, the local distinctions such as proximal/distal get superseded by discourse theoretic distinctions such as known/not known, in focus/not in focus and theme/rheme. Since in most languages demonstratives reflect a more defined system than the mere proximal/distal distinction, the passage from the spatial to the discourse domain should imply a reduction of the system of demonstratives. This expectation is correct, as illustrated by the description by Heeschen (1982) of the historical evolution of the demonstrative system in the Papuan language Yale. At one point, Yale had a pure deictic system, with ane 'here'; ani 'up there'; anu 'down there'; anet 'across there'; and ano 'across there (on the same level, but not separated by a valley)'. These distinctions could be substituted or accompanied by a pointing gesture, and this was almost obligatory in actual discourse. The system also involved pre- and suffixes that could be attached to the demonstratives to add a further nuance. Thus, the prefix d- involved a reference to a whole area, hence to a collectivity, a set of entities and, finally, a plurality. The prefix ab- meant either "only this, not too many" or it expressed duality. Furthermore, a suffix -ko introduced features mentioned in the discourse or known from the situational context. Heeschen (1982: 92 — 93) now notes two tendencies. First, "The more d-, ab- and -ko formations assume discourse functions — i. e. the more they refer not to points in concrete space but to items previously mentioned in the linguistic context — the more they lose the potential for pointing to those things which are truly 'up there' or 'down there'." Second, "As this development proceeds, forms based on ane [...] supersede those based on ani, anu, anet and ano. These tend to be lost since their use is less economic."
274
Walter De Mulder
Heeschen considers the deictic use of the demonstratives ani, anu, anet and ano less economic because ane, combined with the suffix -ko, which introduces features already mentioned or already known, can be used as a neutral discourse-based demonstrative to refer to the location of some entity, if this location has already been introduced in the preceding discourse by one of the other demonstratives, which have kept their full spatial sense. Since the use of these other demonstratives implies that the entity talked about is not at its permanent ane — 'here', in the area of the village 9 — ane becomes available as a neutral discourse-based demonstrative. Heeschen (1982: 93) thus offers a diachronic explanation of the selection of ane to express an anaphoric procedure. However, the selection of ane is surprising, since in most languages, it is not the proximal demonstrative that is selected to express the anaphoric procedure, but the distal one. This is easily explained: objects in the immediate environment of the speaker and the hearer can be pointed at without any difficulty. In consequence, proximal demonstratives identify their referents more easily than the distal ones. Therefore, "a distinction between deictic and anaphoric forms is far more useful in the non-proximal, non-immediate region than in the proximal or immediate regions . . . " (Heath 1980: 61). So why would Yale select the proximal demonstrative for the anaphoric procedure? In fact, Yale is not the only language where the proximal demonstrative acquires the capacity to refer to just about any space: Greenberg (1985: 273 — 280) gives other examples of such languages. As he sees it, the evolution of their demonstrative system leads to a situation where it becomes synchronically irrelevant to define the demonstrative primarily in spatial terms. The proximal demonstrative should rather be conceived of synchronically as the unmarked element of the demonstrative system. As such, it may be the source of the definite article and the third person pronoun, the usual markers of the anaphoric procedure (Greenberg 1985: 278 — 279). Heeschen does not tell us whether Yale develops articles and third person pronouns on the basis of ane, but his description of the evolution of the meaning of ane fits in nicely with Greenberg's hypothesis: ane also acquires the capacity to refer to "just about any space" and therefore also seems to become, from a synchronic point of view, the unmarked element of the demonstrative system. Yale thus seems to be a nice illustration of Greenberg's thesis.
Demonstratives
and the Idealist hypothesis
275
5. Conclusion: the localist hypothesis revisited In this article, w e h a v e d e m o n s t r a t e d that the v e r s i o n o f the localist h y p o t h e s i s p r o p o s e d b y B ü h l e r ( 1 9 3 4 ) is m u c h t o o simple. T h e r e f o r e , the criticism o f K l e i b e r o n the localist h y p o t h e s i s is justified. T h i s does n o t m e a n , h o w e v e r , that w e h a v e t o reject the localist h y p o t h e s i s a l t o g e t h e r , f o r w e think there is still r o o m f o r a localist h y p o t h e s i s w h i c h states that d e m o n s t r a t i v e s h a v e a local p o i n t i n g sense that gets t r a n s f o r m e d in m o t i v a t e d w a y s , as the d o m a i n o f application o f the d e m o n s t r a t i v e s changes. M o r e specifically, in the c h a n g e f r o m the c o n c r e t e spatial d o m a i n t o the d i s c o u r s e d o m a i n , d u e t o the n a t u r e o f the d i s c o u r s e d o m a i n , d e m o n s t r a t i v e s o r the f o r m s d e r i v e d f r o m t h e m are u n i v e r s a l l y used t o express t w o strategies o f d i s c o u r s e c o m p r e h e n s i o n : o n e w h e r e the h e a r e r m u s t f o c u s attention o n the same r e f e r e n t as b e f o r e and o n e w h e r e he m u s t r e o r i e n t a t e his attention.
Notes 1. See Miller and Johnson-Laird (1976: 375); Lakoff and Johnson (1980: 14, 25); Johnson (1987: 25, 7 6 - 8 1 ) ; and Lakoff (1986: 2 8 2 - 2 8 3 , 4 8 9 - 5 0 6 ) ; for the application of spatial structures to time, Traugott (1975, 1978); Dervillez-Bastuji (1982); Wunderlich (1985); for the application to language, Dervillez-Bastuji (1982); and Vandeloise (1986). 2. See Kaplan (1977) and what he calls the "Fregean theory of demonstratives". 3. Kleiber (1988) comes close to adopting this point of view. 4. Note that so- may be used when the speaker knows the referent and the hearer does not. 5. We left out examples of the ko- series because this would lead us too far astray. Moreover, the example with so- shows that it can be used anaphorically, in spite of Kuno's description, which could suggest the contrary. 6. "Deixis" is defined by Kirsner (1979: 358) as the semantic substance subclassified by the grammatical system of demonstratives. It may be defined as "the force with which the hearer is instructed to find the referent" (Garcia 1975: 65). With such a definition, the dividing line between anaphor and deixis seems to become fuzzy, but this should not cause too much trouble if one accepts that demonstratives first have a local use and then develop their anaphoric use out of it. 7. Hartmann (1982) gives some nice examples of the way this evolution proceeds in dialectal varieties of German. 8. The judgments are Kleiber's; in fact, to a lot of French native speakers, the use of cet in example (9) seems better than the use of Γ in example (10). These divergent judgments could depend on a different conception of the function of the first sentence: does it constitute a proposition in which some predicate is ascribed to the subject, such that it is informative enough to constitute a circumstance of evaluation in which one can look for a referent, or does it merely present a referent — and nothing more? If its function is conceived of in the first way, le is appropriate; if it is conceived of in the second way, ce seems more appropriate.
276
Walter De Mulder
9. The translation of ane as 'in the village', which of course is not a deictic expression at all, is explained by Heeschen (1982: 96): in the restricted living space of the Inlom people — the speakers of Yale — "deictics are mere stand-ins for place names, and their use presupposes a common knowledge of what events are likely to happen at what places." (Heeschen 1982: 96).
References Atkinson, Martin 1979 "Prerequisites for reference", in: Eleonor Ochs —Bambi Schieffelin (eds.), Developmental pragmatics, New York: Academic Press, 229—249. Bosch, Peter 1983 Agreement and anaphora. London: Academic Press. 1985 "Context dependence and metaphor", in: Wolfgang Paprotté — René Dirven (eds.), The ubiquity of metaphor, Amsterdam: Benjamins, 141 — 173. 1988 "Representing and accessing focussed referents", Language and cognitive processes 3: 2 0 7 - 2 3 1 . Bühler, Karl 1934 Sprachtheorie. Frankfurt: Ullstein, 1978. Conte, Maria-Elisabeth 1981 "Textdeixis und Anapher", Kodikas/Code 3: 121 — 132. Corblin, Francis 1986 Indéfini, défini et démonstratif. Genève: Droz. Dervillez-Bastuji, Jacqueline 1982 Relations spatiales dans quelques langues naturelles. Genève: Droz. Ehlich, Konrad 1979 Verwendungen der Deixis beim sprachlichen Handeln. Linguistisch-philologische Untersuchungen \um hebräischen deiktischen System. Frankfurt am Main: Lang. Fillmore, Charles 1982 "Towards a descriptive framework for spatial deixis", in: Robert Jarvella— Wolfgang Klein (eds.), Speech, place and action, Chichester: Wiley, 31—59. Garcia, Erica 1975 The role of theory in linguistic analysis: the Spanish pronoun system. Amsterdam: North Holland. Greenberg, Joseph H. 1985 "Some iconic relationships among place, time, and discourse deixis", in: John Haiman (ed.), Iconicity in syntax, Amsterdam: John Benjamins, 271 — 287. Hartmann, Dietrich 1982 "Deixis and anaphora in German dialects: The semantics and pragmatics of two definite articles in dialectal varieties", in: Jürgen Weissenborn—Wolfgang Klein (eds.), Here and there. Crosslinguistic studies on deixis and demonstration, Amsterdam: Benjamins, 187—208. Heath, Jeffrey 1980 "Nunggubuyu deixis, anaphora, and culture", Papers from the parasession on pronouns and anaphora, Chicago Linguistic Society, 151 — 166.
Demonstratives
and the localist hypothesis
277
Heeschen, Volker 1982 "Some systems of spatial deixis in Papuan Languages", in: Jürgen Weissenborn — Wolfgang Klein (eds.), Here and there. Crosslinguistic studies on deixis and demonstration, Amsterdam: Benjamins, 81 — 109. Johnson, Mark 1987 Kaplan, David 1977 Kirsner, Robert 1979
The body in the mind. Chicago: University Press of Chicago. Demonstratives. An essay on the semantics, logics, metaphysics and epistemology of demonstratives and other indexicals. [Unpublished Mimeo.] "Deixis in discourse: an exploratory quantitative study of the modern Dutch demonstrative articles", in: Talmy Givón (ed.), Discourse and syntax, New York: Academic Press, 355 — 375.
Kleiber, Georges 1983 "Les démonstratifs (dé)montrent-ils?" Le français moderne 51: 99 — 117. 1984 "Sur la sémantique des descriptions démonstratives", Linguisticae Investigationes 8: 6 3 - 8 5 . 1986 a 1986 b 1987 a
1987b 1988
"Déictiques, embrayeurs, 'token-reflexives', symboles indexicaux etc.: comment les définir", Li information grammaticale 30: 3 — 21. "Pour une explication du paradoxe de la reprise immédiate", Langue Française 72: 5 4 - 7 9 . "Article démonstratif et article défini en anaphore fidèle", in: Jean David — Georges Kleiber (eds.), Déterminants: syntaxe et sémantique, Paris: Klincksieck, 169-195. "L'énigme du Vintimille ou les déterminants 'à q u a i ' " , Langue Française 75: 107-122. "Sur l'anaphore démonstrative", in: G. Maurand (ed.), Nouvelles recherches en grammaire, Université Toulouse-le-Mirail, 51—74.
Kuno, Susumo 1973 The structure of the Japanese language. Cambridge: M I T Press. Lakoff, George 1986 Women, fire and dangerous things. Chicago: University Press of Chicago. Lakoff, George and Mark Johnson 1980 Metaphors we live by. Chicago: University Press of Chicago. Lyons, John 1975 "Deixis as the source of reference", in: Edward Keenan (ed.), Formal semantics of natural language, Cambridge: Cambridge University Press, 61—83. 1980 Sémantique linguistique. Paris: Larousse. Miller, George and Philip Johnson-Laird 1976 Language and perception. Cambridge: Cambridge University Press. Nunberg, Geoffrey 1978 The pragmatics Schweizer, Harro
of reference.
Bloomington: Indiana Linguistics Club.
1985 Sprache und Raum. Stuttgart: Metzler. Traugott, Elisabeth 1975
"Spatial expressions of tense and temporal sequencing", Semiotica 230.
15: 207 —
278 1978
Walter De Mulder
"On the expression of spatio-temporal relations in language", in: Joseph H. Greenberg (ed.), Language universals. Volume 3, Stanford: Stanford University Press, 3 6 9 - 4 0 0 . 1982 "From prepositional to textual and expressive meanings: some semanticpragmatic aspects of grammaticalization", in: Winfred Lehmann —Yakov Malkiel (eds.), Perspectives on historical linguistics, Amsterdam: Benjamins, 245 — 271. Vandeloise, Claude 1986 L'espace en français. Paris: Seuil. Wunderlich, Dieter 1985 "Raum, Zeit und das Lexikon", in: Harro Schweizer (ed.), Sprache und Raum, Stuttgart: Metzler, 66 — 89.
Phraseological universale: theoretical and applied aspects Dmitri) Dobrovol'skij
1. Introduction It is only in recent years that the problem of universals in the sphere of phraseology has become an object of linguistic study (cf. Dobrovol'skij 1988 and, to a lesser extent, Makkai 1978). That is why for many linguists the combination phraseological universals itself may still sound odd. There is no doubt, however, that it refers to a promising trend in modern linguistics. On the one hand, phraseology 1 has since Bally (1909) stored a great amount of empirical data that until recently have hardly received any account within the framework of general linguistics. On the other hand, neither universalist linguistics nor linguistic typology has sufficiently dealt with phraseological facts. Yet it should be clear that without linking up phraseology with other issues of general linguistics no holistic model of natural language can be worked out. If we want to do away with this discrepancy it becomes necessary, on the one hand, to make explorers of universals aware of the phraseological facts and of the necessity to integrate them within a linguistic model and, on the other hand, to make phraseologists aware of the opportunities offered by the application of methods originally adopted in the study of universals and linguistic typology. This constitutes the main purpose of this paper. The achievement of this purpose promises advantages from both theoretical and practical points of view. Traditionally phraseology is considered to be an extremely idiosyncratic subsystem of language. Phraseology is beyond doubt an important indicator of the culture-dependent peculiarities of language at a given stage. However, the phraseological system (as well as any other subsystem of language) reveals certain universal features, as, according to Jakobson (1966: 264), "the languages of the world can actually be approached as manifold variations of one world-wide theme — human language." The universal substratum of the phraseology of each language is often disregarded, and I believe this should be viewed as the most serious obstacle in the way of creating a general theory of phraseology.
280
Dmitrij
Dobrovol'skij
2. Phraseological universals The term phraseological universals can be used to refer to fairly diverse phenomena. This is brought about by the heterogeneity of universal and potentially universal phenomena in the phraseological domain, and some kind of taxonomic arrangement of these phenomena is necessary. Classical typology of universals differentiates between phonological, grammatical, semantic and symbolic universals (Greenberg —Osgood —Jenkins 1966: xxii). On the other hand, today more and more writers find it necessary to distinguish between universals that are extralinguistically determined (conceptual or cognitive universals) and those that are specifically linguistic. Die Annahme, daß es Eigenschaften gibt, die allen natürlichen Sprachen gemeinsam sind, ist plausibel (jede natürliche Sprache stellt ein Zeichensystem dar) und kann durch empirische Beobachtungen (das Studium möglichst vieler Sprachen) und theoretische deduktive Methoden gestützt werden. Anderseits gibt es eine Menge von Abbildelementen (Noemen), die der gesamten Menschheit (dem menschlichen Bewußtsein überhaupt) zukommen. Erstgenannte sind linguistische, letztere kognitive Universalien. Die Gleichsetzung dieser beiden Arten von Universalien ist eigentlich eine Identifizierung der Elemente des Systems, mit dessen Hilfe Objekte (Dinge, Prozesse usw.) benannt werden (Laut-Bedeutungs-Zuordnung), mit Elementen des menschlichen Bewußtseins (des menschlichen Denkens), also den Noemen. (Staffeidt 1983: 93)
So Staffeidt, as well as many other scholars, speaks about two types of universals: (a) linguistic and (b) cognitive universals. Thus the study of phraseological universals must begin with classifying them into (a) conceptual or cognitive (extralinguistically determined) phraseological universals, and (b) linguistic or systemic phraseological universals. The latter can be further classified into two groups (cf. for details Dobrovol'skij 1988: 22—61): (bl) lexical-phraseological universals (those that are characteristic of both lexical and phraseological systems), and (b2) phraseological universals proper (those that characterize only the phraseological system).
3. Conceptual universals Conceptual (cognitive) phraseological universals are understood here as cognitive structures that belong to the human intellect and get ontologized with the help of the phraseological means of language. This is the place to mention the so-called "double anthropocentrism" of phraseology (i.e., the semantic
Phraseological
universali
281
direction of both the actual meaning and the figurative ground of most of the idioms toward man, cf. Rajchstejn 1980: 92), the prevalence of a negative estimation, and above all, the concepts that typically acquire phraseological means of expression (consider, for instance, such phraseologically relevant spheres of signification as stupidity, fear, death, fury, blame, and fraud). In the same way Makkai (1978: 427) points out the productivity of such phraseological fields as death or thievery/theft. Phraseological nomination thus exhibits a strong tendency towards naming the subjectively relevant phenomena of the objective world (cf. Rajchstejn 1980: 60). The prevalence of a negative estimation is an important non-trivial regularity of the phraseological system. It is unclear whether it also characterizes the lexical system. On the one hand, namings with negative estimation are said to be more widely represented in language. This view is supported by work in cognitive psychology, purporting to show that knowledge that is associated with negative emotions is far more differentiated (cf. Vol'f 1986: 102). On the other hand, in connection with the Ρoily anna hypothesis of Boucher and Osgood (1969), one has noted the universal tendency towards quantitative predominance (as well as more frequent occurrence in texts) of namings with a positive estimation. The Pollyanna hypothesis finds support in psycholinguistic experiments and is explained by the fact that words with positive estimation (like good) do not only express appreciation, but also the idea of the norm (for details see Vol'f 1986: 103; Arutjunova 1987: 1 2 - 1 3 ) . About the phraseological system, however, we may with a far greater degree of certainty maintain that negative estimation plays the leading role, both from the point of view of the degree of differentiation of namings with a negative estimation and from the point of view of the proportion of the idioms with a negative versus positive estimation in the lexicon. So the Pollyanna principle does not work in the sphere of phraseology. It probably does not work in the sphere of lexemes of secondary figurative nomination either, i. e., in those spheres where estimation is amalgamated with emotivity. Emotive namings never serve the expression of norm and so they elude the Pollyanna principle. Another issue in the sphere of cognitive phraseological universale concerns the universality of the concepts that constitute the basis of phraseological inference. These concepts denote elements of the "visible", physical world. In the course of the nominational activity of different (if not of all) language communities they are used for the linguistic ontologization of certain goal concepts, i.e., as a rule to give names to the elements of "invisible", nonphysical worlds. Thus, in many languages the goal concept "defame" is ontologized with the help of the initial concept "make dirty": 2
282
Dmitrij
(1) a.
b.
c. d.
e.
Dobrovoískij
English: drag sb. through the mud\muck cast dirt at sb. German: jmdn. in den Dreck Riehen 'drag sb. into dirt' jmdn. mit Dreck bewerfen 'pelt sb. with dirt' jmdn. mit Kot besudeln 'make sb. dirty' Dutch: iemands naam door het slijk halen 'drag sb.'s name through the mud' Russian: smesivat' s grja^'ju kogo-l. 'merge sb. with mud' vylivat' pomoi na kogo-l. 'pour out slops on sb.' brosat' grja^ju ν kogo-l. 'cast dirt at sb.' Lithuanian: pama^gomis apipilti 'pour slops' su purvais sumaisyti 'merge with mud'
As a rule the initial concepts cluster around names of parts of the body, articles of clothing, modes of human behavior, various situations of everyday life, familiar animals, etc. For example: (2) a.
b.
c.
English: keep/have an/one's ear to the ground "notice signs of coming events" an ugly duckling "sb. born less attractive than his/her brothers and sisters who later surpasses them" the early bird "the early riser" take a leaf out of someone's book "copy sb. else" break one's neck "try one's utmost" catch sb. with his pants down "take sb. by surprise" German: das Maul vollnehmen 'have one's mouth full' "twaddle" den Bock %um Gärtner machen 'make the goat the gardener' "make sb. do sth. he/she is not able to do" das schwarte Schaf 'the black sheep' "entirely different from others" jmdm. eins auf den Hut geben 'hit sb. on the hat' "scold" Dutch: iets onder de knie hebben 'have sth. under the knee' "know sth. very well" een vrije vogel 'a free bird' "one's own master" het paard/de paarden achter de wagen spannen 'put the horse/horses after the cart' "do sth. wrong"
Phraseological
d.
e.
universali
283
iem. de¡een pen op de neus fetten 'set the pen upon sb.'s nose' "force sb. to a declaration or action" iem. knopen draaien 'make buttons for sb.' "cheat" Russian: namylit' seju komu-l. 'soap sb.'s neck' "tell sb. o f f ' nosit'sja kak kurica s jajcom 'rush about like a hen with an egg' "make too much of sb./sth." mo^olit' gla^a komu-l. 'make sb.'s eyes callous' "be an eyesore to sb." moloko na gubach ne obsochlo u kogo-l. 'milk hasn't dried up on sb.'s lips' "sb. is too young" cesat' ja^yki 'scratch one's tongues' "gossip" otkinut' kopyta 'throw away one's hooves' "die" Lithuanian: nuslustyti nos[ 'wipe sb.'s nose' "excel sb. in sth." liefyivi praryti 'swallow one's tongue' "refuse to utter a word" ant danties pakliüti 'get upon sb.'s tooth' "be mocked at by sb." kaip slapia vista 'like a wet hen' "soaked" nueiti suniui ant uodegos 'go under the dog's tail' "be wasted" paskutines kelnias atiduoti 'give away the only pants' "be a person of ready sympathy"
Studying the most essential universal tendencies connecting the spheres of the initial concepts and the spheres that get ontologized in their actual meaning with the help of the idioms, we touch upon universal laws of imaginative and associative thinking. Thus we come to problems of language and world picture. The notion of world picture comes from philosophy (e. g., Pavilionis 1983). Applying it to phraseology, we can refer to the two different conceptual spheres as to "phraseological world pictures" ("PWP"): a PWP 1 that is constituted by the initial concepts, and a PWP 2 that is based upon the goal concepts. PWP 1 is created with the use of the meanings of the imagery substrata, i. e., with the use of the idioms taken literally. It is a surreal world whose inhabitants put fleas in each other's ears, bite each other's noses off, talk through their hats, etc. A world like this (PWP 1) is shown in the wellknown picture by Pieter Bruegel the Elder, "Flemish Proverbs". Together with the surreal elements in the structure of PWP 1 there are also "traces" of earlier cultures: habits and customs, historical events and everyday situations. PWP 2 is part of our semantic world picture. Note that PWP 2 is highly subjective and anthropocentric. If people expressed their thoughts exclusively
284
Dmìtrìj Dobrovol'skìj
by means of idioms, it would result in an extremely emotive, estimating, subjectively modal communication. It is obvious that PWP 1 and PWP 2 belong to different realities. However, they are by no means fully autonomous and independent from each other. In the mind of the native speakers they are interconnected through the "inner form" (in the sense of Humboldt). The inner form is understood here as a kind of initial conceptual pattern functioning as a prototypical imagery ground. If it did not exist, decoding the phraseological meaning would be impossible. As for metaphorical and hyperbolical idioms, the connection between the concepts of PWP 1 and PWP 2 is carried out within the fictivity modus as if {cf. Telija 1988: 186 — 187). In this way a possible world is projected upon the real world. Apart from its purely theoretical benefit, the above discussion has a practical relevance. Cognitive phraseological universals build the theoretical foundation for modern phraseographical reference books that follow the thesaurus principle and serve as an adequate model of real linguistic competence. Compiling ideographic phraseological dictionaries (phraseological thesauri) is an urgent task of modern theoretical phraseology and practical lexicography. Obviously the task can be successfully accomplished only within a computer-assisted phraseographical system (cf. the current project Leksikograficeskaja ra^rabotka 1988). The work on a phraseological thesaurus presupposes several stages. At a first stage a certain descriptor is attached to each of the idioms under consideration. The correctness of this conceptual marking of the phraseological stock is a major requisite for the application of the ideographic principle of regrouping the whole phraseological "vocabulary" at a second stage. The computer-assisted phraseographical system must also be supplied with information about certain cognitive structures (i. e., about the phraseological component of the conceptual world picture). With the help of this conceptual system we can determine both coordinative and subordinative connections between separate descriptors which reflect the corresponding basic concepts. For instance, the concept "a stupid man", which in all the investigated languages is often expressed by idioms, comes under the concept "man's intellectual properties", and the latter in turn comes under the concept "man's properties". Without a conceptual system of this kind an ideographic phraseological dictionary cannot be considered a real thesaurus because the linguistic information would only be systematized at the level of the basic concepts. But the thesaurus principle brings forward the systematization of the relationships among the basic concepts — the systematization that reflects the structure of the mental lexicon.
Phraseological
universals
285
Consider the following idioms: (3) a.
b.
c.
d.
e.
English: come down on sb. like a ton of bricks give¡read sb. a lesson tear sb. to bits have sb. on the carpet German: jmdm. übers Maul fahren 'ride over sb.'s mouth' jmdm. eine Gardinenpredigt halten 'make a curtain sermon' jmdn. %ur Sau machen 'make sb. a swine' jmdn. in die Pfanne hauen 'chop sb. into the frying pan' jmdm. geigen, was eine Harke ist 'show sb. what a rake is' Dutch: iem. over de hekel halen 'drag sb. through the flax comb' iem. iets onder ^jjn neus wrijven 'rub sth. under sb.'s nose' iem. iets op %ijn broodgeven 'give sb. sth. on his bread' iem. de les le^en 'teach sb. a lesson' iem. op ^ijn nummer %etten 'put sb. on his number' Russian: dat' ^i^ni komu-l. 'give sb. life' rçadaf %aru komu-l. 'give sb. heat' ustroit' banju komu-l. 'make a bath for sb.' dat' duchu komu-l. 'give sb. air' vyjat' ν oborot kogo-l. 'take sb. into a turnover' namjlit' golovu komu-l. 'soap sb.'s head' ra^delat' pod orech kogo-l. 'make sb. look like a nut' Lithuanian: pipirtf duoti 'give pepper' garo duoti 'give steam' pirti u^kurti 'prepare the bath' vejo duoti 'give wind' velntift duoti 'give the devils' pries plaukq sukuoti 'comb against the hair' kaili nudirti 'strip skin'
In the phraseographical dictionary these idioms acquire the descriptor "blame". Together with the idioms that have such descriptors as "deceive", "humiliate", "intimidate", "frighten", etc., they group under a concept of a higher rank, i. e., the concept "have an active non-physical verbal negative influence upon sb.", and so on, until they come under an abstract notion
286
Dmitrij
Dobrovoískij
such as "act". The corresponding fragment of the conceptual system can be presented in the following way. act 1. perform a directed act 1.1. have an active influence upon the object 1.1.1. have an active non-physical influence upon the object 1.1.1.1. have an active non-physical verbal influence upon the object 1.1.1.1.1. have an active non-physical verbal negative influence upon the object 1.1.1.1.1.1. blame 1.1.1.1.1.2. deceive 1.1.1.1.1.3. intimidate 1.1.1.2. have an active non-physical non-verbal influence upon the object 1.1.1.2.1. have an active non-physical non-verbal negative influence upon the object 1.1.1.2.1.1. ruin sb. financially 2. perform a non-directed act 2.1. perform a wilful act 2.1.1. perform a wilful mental act 2.1.1.1. think 2.1.2. perform a wilful physical act 2.1.2.1. eat (too much, too little, etc.) 2.1.2.2. drink (too much) The fragment "act" of the thesaurus by no means claims to be exhaustive for it has a purely illustrative character. The terminal nodes of this hierarchical tree represent the basic concepts that are the direct abstraction from linguistic empirical data. In our case they play the role of hyperonyms for the corresponding phraseosemantic fields, e. g.: 1.1.1.1.1.2. deceive (4) a.
English: blow dust in sb.'s eyes pull the wool over sb.'s ejes drag a red herring across sb.'s path tell sb. a story turn geese into swans
Phraseological
b.
c.
d.
e.
universals
287
German: jmdm. Sand in die Augen streuen 'throw sand into sb.'s eyes' jmdn. auf die falsche Fährte setzen 'put sb. on the false track' jmdm. ein Märchen auftischen 'tell sb. a fairy-tale' Dutch: iem. voor de gek houden 'hold sb. for a fool' iem. een blinddoek voor de ogen binden 'bind a kerchief over sb.'s eyes' iem. op een dwaalspoor brengen 'put sb. on the false track' iem. een rad voor de ogen draaien 'turn a wheel before sb.'s eyes' iem. bij de neus nemen 'take sb. by the nose' Russian: obvodit' vokrug pal'ca kogo-l. 'lead sb. around one's finger' vodit' nos kogo-l. 'lead sb. by the nose' tçagovarivat' %uby komu-l. 'put on a spell against a toothache' ^ama^at' gla^a komu-l. 'oil sb.'s eyes' Lithuanian: u^kalbèti dantis 'put on a spell against a toothache' vedutoti nosies 'lead by the nose' uodegq kraipyti 'turn one's tail' muilinti akis 'soap sb.'s eyes' miglq püsti i akis 'let the haze into sb.'s eyes'
1.1.1.2.1.1. ruin sb. financially (5) a.
b.
c. d.
English: cut sb.'s throat put a nail in sb.'s c o f f i n take bread out of sb.'s mouth pick sb.'s pocket German: jmdm. den Hals abschneiden 'cut the throat of sb.' jmdn. übers Ohr hauen 'hit sb. on the ear' jmdm. den Gashahn abdrehen 'turn off the gas-valve for sb.' jmdm. das Wasser abgraben 'dig out the water for sb.' Dutch: iem. de neus en oren afeten 'eat up sb.'s nose and ears' Russian: pustit' kogo-l. po miru (s sumoj) 'make sb. go begging (with a pouch)' pustit' ¡vypustit' kogo-l. ν trubu 'let sb. out through the chimney' snjat' poslednjuju rubasku s kogo-l. 'take the last shirt off sb.' obodrat'¡obcistit'joblupit' kogo-l. kak lipku 'bark/clean out/peel sb. like a lime-tree'
288
Dmìtrij
e.
Dobrovol'skij
Lithuanian: paskutinius marskinius nuvilkti 'take off the last shirt' paskutinl kaili nulupti 'take off the last shin' nukirpti vilnas 'cut the hair'
A thesaurus of the initial concepts can be based upon the same principle. Juxtaposing the thesauri of initial and of goal concepts, we get the opportunity to investigate (on the basis of phraseological data) the universal mechanisms of inference in natural languages. In this way a probabilistic model of metaphorical phraseologizing can be worked out. Although a model like this has no operational value, it still has a certain explicative force contributing to the explanatory potentials of universalist linguistics (cf. Gak 1988: 15).
4. Lexical-phraseological universale The study of lexical-phraseological universale proceeds from the assumption that word-stock and phraseology are not isolated domains but constitute a holistic system. They are closely interconnected, revealing common regularities, tendencies of development and interdependent processes. In the sphere of lexical-phraseological universals there are categories like polysemy, homonymy, synonymy, and antonymy. They are linguistic phenomena in the general sense, i. e., not purely phraseological phenomena. Their universality in the lexical sphere was pointed out by Ullmann (1966). The investigation of these phenomena on the basis of phraseological data is important not only for confirmation of their universality but also for determining relevant quantitative peculiarities of phraseology as distinct from word-stock. Thus many scholars (e.g., Cernyseva 1986: 196 — 197) have claimed that polysemy in the lexical system is more highly developed than in the phraseological one. In the sphere of phraseology, polysemy and homonymy are relatively rare. Even in those cases where the dictionaries qualify something as phraseological polysemy it is often the wide meaning of the idiom that is meant, i.e., a diffuse vaguely outlined meaning covering large classes of designates (cf. Fleischer 1982: 170—171). For example, the idiom: (6) a.
jmdm. die Zunge lösen 'set sb.'s tongue free'
with the wide meaning "make sb. speak with the help of sth. (with the help of alcohol, enforcement, hypocritical sympathy, etc.)" is in the better part of
Phraseological
universals
289
modern reference books labeled as polysemantic (cf. MDI, Duden-GW). Or consider the English idiom: b.
not have a clue (informal) "know nothing" (Longman DCE)
which is presented in Longman DEI as polysemantic: 1) be completely uncertain of what to do, what is correct, etc.; 2) lack ability in a general way. Synonymy, on the contrary, is much more frequent in the phraseological sphere than in the lexical sphere, frequent enough to build large series of almost absolute phraseological synonyms. Consider: (7)
alles über einen Leisten schlagen 'put everything on the same last' = alles über einen Kamm scheren 'cut everything to the same comb' = alles in einen Topf werfen 'put everything in the same saucepan' "treat all alike paying no attention to fundamental differences"
The tendency of phraseology to form synonyms is also demonstrated by the examples of phraseosemantic fields (3), (4), (5) given above. Among lexemes, on the other hand, absolute synonyms are rather uncommon (cf. Cernyseva 1980: 7 4 - 7 9 ) . In the sphere of antonymy the peculiarities of phraseology as compared to word-stock are interesting too. Whereas antonymous words are mainly adjectives, phraseological antonyms are mainly verbal (cf. the detailed description in Dobrovol'skij 1988: 45 — 52). As to hyponymy, idioms normally perform the function of the co-hyponyms, and the hyperonym is represented by a lexeme. If, for instance, the word "dismiss" functions as a hyperonym (and at the same time as the descriptor of the respective phraseosemantic field), then the idioms play the role of the corresponding hyponyms. "Dismiss" can be viewed as the hyperonym for the idioms in (8), for each of them has in the structure of its meaning the semantic component "dismiss" as the archseme and, in addition, a set of differential concretizing semes. (8) a.
b.
English: dismiss·. turn sb. out of house and home give sb. the boot give sb. the sack show sb. the door German: hinausweisen·. jmdn. vor die Tür setzen 'put sb. in front of the door' jmdm. den Laufpaß geben 'give sb. walking-papers' jmdm. die Tür weisen 'show sb. the door'
290
Dmitrij
c.
d.
e.
Dobrovol'skij
Dutch: wegjagen·. iem. vierkant de deur uitgooien 'chuck sb. out of the door quadrangularly' ìem. de boender geven 'give sb. the push' iem. %ijn paspoort geven 'give sb. his passport' iem. de %ak geven 'give sb. the sack' Russian: vygnat'·. vytolkat' ν sejujv tri sei kogo-l. 'push sb. out in the neck/in three necks' dat' pinka komu-l. 'give sb. a kick' vysvyrnut' dver' kogo-l. 'chuck sb. out of the door' poslat' kogo-l. k certovoj babuske 'send sb. to the devil's granny' spustit' kogo-l. s lestnicy 'send sb. down the staircase' Lithuanian: isvarytr. sluota paimti 'take the broom' pakarpos ismesti 'throw away by the scruff of the neck' u£ sprando ismesti 'push out in the neck' velniop stesti 'send to the devil' parodyti duris 'show the door'
An important issue in the sphere of lexico-phraseological universale is the possibility of a universal typology of idioms. In addition to the universals suggested above, one can formulate universal types of idioms in the same way as definite communicative-functional classes are postulated for lexemes (cf. Arutjunova 1976; Stepanov 1981). The typology of idioms that is suggested below is understood as universal and many-dimensional in the sense that it is based not on the properties determined by the language structure, but on the main communicative functions, which are universal by nature. The following phraseological types are distinguished: a. Characterizing (qualifying) idioms: (9)
smell a rat 'suspect that something is wrong, that someone is trying to deceive one, etc." (to swallow) a bitter pill "an unpleasant or disappointing fact that must be accepted" hit the bull's-eye "do or say exactly the right thing; succeed completely in what one is doing" the black sheep (of the family, etc.) "that member (or a family of other group) who is thought to be a disgrace to other members of it (because he is a criminal, a wastrel or because he does not measure up to their imposed standards)"
This type of idiom constitutes the center of the phraseological system of each language. Characterizing idioms as units of predicative semantics perform the
Phraseological
universalj
291
function of ontologizing actions, states, processes, intensifications, relations, etc. Thus (10) names an action, and (11) qualifies a person. (10) (11)
pull someone's leg "make fun of a person in a friendly way, e. g., by trying to make him believe something that is not true" a new broom "sb. recently appointed to office or a responsible post"
b. Identifying idioms: (12)
a bot dog "a special sort of long red sausage in a bread roll" the Black Death "the illness that killed large numbers of people in Europe and Asia in the 14th century"
This class consists of fairly diverse units that share the communicative function of identifying objects, which is different from characterizing them. Traditionally, such units are ranked as border-line cases, because they are intuitively felt to be non-standard. The following subclasses can be distinguished: b.l. Proper name idioms: (13)
the the the of
Little Corporal "Napoleon" Iron Lady "M. Thatcher" White House "the official home (in Washington) of the President the United States"
b.2. "Folk terminology" idioms: (14)
Adam's apple "that part at the front of the throat that is seen to move when a person, esp. a man, talks or swallows"
b.3. Idioms that name concrete objects: (15)
bloody mary "a drink made by mixing vodka and tomato juice" the Black Maria "a vehicle used by the police to carry prisoners"
c. Quantifying idioms (i. e., functioning as natural-language quantifiers): (16)
every man jack/Jack (of sb.) "every one of a large group or of a considerable number of people, with no exceptions" nine times out of ten "almost always"
d. Idioms that are not major clause constituents and that function as modal operators: d.i. Idioms that express the speaker's judgment: (17)
God/the Lordjgoodnessjheaven one can only guess"
knows "I don't know; nobody knows;
292
Dmìtrij DobrovoÎskij
d.2. Idioms that express emotion: (18)
(dear) God in heaven ( ! ) — exclamation of surprise or shock; sometimes worked into the structure of an exclamatory appeal or statement
e. Idioms functioning as utterances: (19)
the die is cas ¡¡thrown "a decision or decisive action has been taken and must be kept to"
In the traditional theory of phraseology the latter idioms are called "phraseological expressions" and grouped together with paroemii (see, however, Kunin 1970: 334 — 335, where a class of so-called "nominative-communicative phraseological units" is postulated). Since our typology of idioms is based on a communicative-functional principle, the two are distinguished, for paroemii are known to perform an entirely different semiotic function and cannot be thought of as units of lexicon (for details see Dobrovol'skij 1989). The practical advantages of this typology of idioms, as compared to the conventional one, consist in its universality, due to which phraseological systems of different languages can be typologized and described in accordance with the same principles. This has positive consequences both for the teaching of foreign languages and for lexicographical work. It would also make it possible to compile phraseological dictionaries of two and more languages without any additional unification of the descriptive apparatus.
5. Phraseological universale proper To study phraseological universale proper means to determine the regularities that are unique to the phraseological system and reflect the essential features of its inner organization. One works towards an explicit answer to the question of what makes the phraseological system differ from the lexical one, and thus the creation of a "linguistics of collocations" to complement the "linguistics of words". In the sphere of the phraseological universale proper ("PUP") it is advisable to distinguish between implicational and non-implicational universals. Whereas non-implicational PUPs are regularities of the inner organization of the phraseological system, implicational PUPs are correlations between the structure of the phraseological system and the peculiarities of other subsystems of language.
Phraseological
universals
293
The following inherent characteristics of the phraseological system can serve as examples of non-implicational PUPs: (i) Phraseological constituents of any language represent a definite set of lexemes whose volume varies from one language to another but never equals the volume of the entire word-stock of the language in question. (ii) In all languages there is probably a definite set of formally bound phraseological constituents, i. e., words occurring only in idioms, whose volume varies from one language to another. The following units can be considered as examples of idioms with formally bound constituents. 3 (20) a
b.
c.
d.
e.
English: gone for a burton "killed" play hook(e)j "be absent from school without permission" kith and kin "members of one's family" get one's/someone's dander up "become/make angry" German: bei jmdm. ins Fettnäpfchen treten "cause unintentional offence" Lug und Trug 'lie and fraud" das Hasenpanier ergreifen "run away" Dutch: de loftrompet over iem.\iets steken "praise sb. excessively" in rep on roer "in the state of confusion, embarrassment, anxiety, agitation" Russian: semeja podkolodnaja "scoundrel" so skre^etom £ubovnym "very reluctantly" chodit' chodunom "shake, tremble" bit' baklusi "waste one's time" nicto^e sumnjasesja "without doubting for a moment" Lithuanian: dèkui us[ aba^éli "thank" luisis pastatyti "lend one's ears, listen with attention" lukutus sumesti i kupetq "do sth. together"
(iii) The phraseological activity of the constituents varies between absolute uniqueness and superfrequency. Not surprisingly, the study of the phraseological activity of the words of a language may be hampered by the fact that different dictionaries treat these words differently. So according to the ODCIE-2, the frequency scale of the substantive constituents of English idioms is as follows (only the most active constituents are listed).
294
Dmitrij
(21) a.
Dobrovol'skij
ODCIE-2·. time(s), man\men, day(s), thing(s), way(s), l i f e , hand(s), word(s), mind(s), head(s), eye(s), world(s), heart(s), end(s), face(s), Godjgods
Compare this with the Longman b.
DEI:
Longman DEI·. hand(s), head(s), eye(s), way(s), time(s), day(s), heart, word(s), men, foot/feet, mind(s), face(s), end(s), l i f e , ear(s), world(s)
manj
Correspondingly for verbs: (22) a.
b.
ODCIE-2: be, say, go, know, do, see, come, make, want, stand, think, tell, take, have, need, get Longman DEI\ take, go, put, make, keep, give, come, set, do, know, cut, play, hold, throw, turn, get
For adjectives/adverbs: (23) a.
b.
ODCIE-2·. all, good, right, old, well, best, bad(ly), better, dead(ly), just, long, never, hard( l y ) , far, little, wrong Longman DEI·. good, all, right, dead, down, bad(ly), big, high, old, short, hard, cold, f u l l , black, hot, last, long
The comparison of the corresponding facts in German is made on the basis of the MDI and RKIDS dictionaries. For substantives: (24) a.
b.
MDI·. Handf Hände 'hand(s)', Kopf Köpfe 'head(s)', Herten) 'heart(s)', Auge (η) 'eye(s)', Wort (e) 'word(s)', Fuß! Füße 'foot/feet', Ohr (en) 'ear(s)', Weg(e) 'way(s)', Bein(e) 'leg(s)', Mund 'mouth', Hals 'throat', Teufel 'devil', Nase 'nose', Tag(e) 'day(s)', Boden 'ground', Wasser 'water' RKIDS·. Her£ 'heart', Haar(e) 'hair', HandjHände 'hand(s)', Finger 'finger', Kopf 'head', Ohr (en) 'ear(s)', Wasser 'water', Nase 'nose', Bein(e) 'leg(s)', Feuer 'fire', Zahn\Zähne 'tooth/teeth', Arsch 'ass', Auge(n) 'eye(s)', Luft 'air', Mund 'mouth', Hund(e) 'dog(s)'
Phraseological
universali
295
For verbs: (25) a.
b.
MDI: haben 'have', sein 'be', lassen 'let', machen 'make', kommen 'come', gehen 'go', bringen 'bring', geben 'give', können 'can', tun 'do', stehen 'stand', sehen 'see', bekommen 'get', nehmen 'take', helfen 'help', sagen 'say' RKIDS-. haben 'have', sein 'be', machen 'make', gehen ' g o ' ^ f a l l e n 'fall', lassen 'let', nehmen 'take', bringen 'bring', kommen 'come', Riehen 'pull', halten 'hold', setzen 'set', sitzen 'sit', können 'can', hauen 'hit', liegen 'lie', gucken ' l o o k '
F o r adjectives/adverbs:
(26) a.
MDI·. gut
'good', schivar^
'black', grün
'green', blau 'blue', halb(e)
'half,
recht 'right', kur^ 'short', dumm 'stupid', weiß 'white', bunt 'colored', klein 'small', rot 'red', schlecht 'bad', schön 'beautiful', groß 'big', lang 'long'
b.
RKIDS-. all 'all', gut
'good', kalt
'cold',. voli 'full', groß
'big', alt 'old', blau
dick 'thick', gleich 'same', grün 'green', klein 'small', let^t 'last', rein 'clean', sauer 'sour', falsch 'false', schivar£ 'black', fertig 'ready', 'blue',
heiß 'hot', klar 'clear', kur% 'short', link(s) 'more', trocken 'dry', warm 'warm'
'left', lang 'long',
F o r D u t c h the c o r r e s p o n d i n g facts are o b t a i n e d f r o m the NSG
and
mehr NSSZ
dictionaries. F o r substantives: (27) a.
NSG: hand 'hand', voet 'foot', a p p e l ( t j e ) 'apple', water
'water', wind 'wind',
neus 'nose', oogjogen 'eye/eyes', paard 'horse', hond(je) ' d o g \ ] a n , slag 'stroke', Adam, dood 'death', hart 'heart', vinger 'finger', aarde 'earth', alarm 'alarm', gal 'bile, gall', gek ' f o o l ' , oor 'ear', schoen(en) 'shoe(s)', stuk 'piece'
b.
NSSZ·. hand 'hand', hond 'dog', paard
'horse',
kat 'cat', duivel 'devil', kind
'child', geld 'money', God 'God', oog 'eye', voet 'foot', haar 'hair', mond 'mouth',
water 'water', man 'man', varken 'pig', vinger ' f i n g e r '
For verbs: (28) a.
NSG: %ijn 'be', hebben 'have', komen 'come', maken 'make', gaan 'go', ^itten 'sit', geven 'give', krijgen 'get', ^etten 'put', houden 'hold', nemen 'take', staan 'stand', weten ' k n o w ' , liggen 'lie, be placed', spelen 'play', lopen 'run', vallen 'fall'
296
Dmitrij
b.
Dobrovol'skij
NSSZ-. geven 'give', sparen 'save, put away', spreken 'talk", doen 'do', trouwen 'trust', %wijgen 'keep silence", gaan 'go', horen 'hear', leven 'live', eten 'eat', liegen 'lie, tell a lie', spelen 'play', stelen 'steal', wachten 'wait', Jorgen 'taken care', fluiten 'pipe, whistle', hangen 'hang', vragen 'ask', werken 'work', Jeggen 'say'
For adjectives/adverbs: (29) a.
NSG: goed 'good', lang 'long', oud 'old', eigen 'proper', groot 'big, large, great', kort 'short', koud 'cold', kwaad 'bad', %wart 'black', blauw 'blue', hoog 'high, tall', klein 'little', laat 'late', breed 'broad, wide', dik 'thick', half 'half, recht 'right', sterk 'strong', vol 'full', %uur 'sour' b. NSSZ-. beter 'better', gek 'crazy, mad', oud 'old', %pt 'silly', vet 'fat', wijs 'wise', jong 'young', kort 'short', kwaad 'bad', laat 'late', blind 'blind', goed 'good', vroeg 'early', doof deaf, stil 'quiet, silent, calm', thuis 'at home'
In spite of the differences between the frequency scales from one dictionary to another and from one language to another, various regularities can be observed. For instance, among substantives, somatisms (English: hand, head, eye, heart, foot, face, ear, German: Hand, Kopf, HerAuge, Fuß, Ohr, Bein, Mund, Hals, Nase, Haar, Finger, Zahn, Arsch·, Dutsch: hand, oog, voet, haar, mond, vinger, neus, hart) play a great idiom building role. The voluminous Soviet literature on the subject of phraseological systems of different languages makes it reasonable for us to suppose that this regularity is a good candidate for a non-implicational PUP. Among verbal constituents the most active idiom builders are verbs of superwide semantics or semantic primitives in the sense of Wierzbicka (1972) (English: be, have, take, go, put, make, do, give, come, set, hold, get, turn, need·, German: haben, sein, lassen, machen, kommen, gehen, bringen, geben, stehen, bekommen, nehmen, setzen, sitzen, trieben, halten-, Dutch: %ijn, hebben, komen, maken, gaan, fitteti, fetten, geven, krijgen, nemen, houden, staan, doen, liggen).4 In the spere of adjectives and adverbs we primarily deal with axiological elements (English: good, right, well, bad(ly), wrong, best·, German: gut, schlecht, falsch, schön-, Dutch: goed, kwaad, recht, beter)·, parametrical ones (English: down, big, high, short, f u l l , long, little, far, German: kur^, weit, klein, groß, voll, dich, Dutch: groot, kort, hoog, klein, breed, dik)·, and names of colors with symbolic values (English: black', German: schwarζ, grün, blau, weiß, rot-, Dutch: %]vart, blauw). Implicational phraseological universale proper (PUPs) are particularly relevant for both theoretical and applied approaches. In general, universal
Phraseological
universals
297
implications (as compared to non-implicational universals) convey more information because they register the existence of regular correlations. This is far more important than to discover isolated features or elements inherent in all languages. Implicational PUPs help indicate the regular dependence of phraseology upon other subsystems (modules of language), with the aim of connecting relevant features of different subsystems by one or more intermodular parameters. This should contribute to the effort of creating a holistic typological model (see also Comrie 1987; Klimov 1987; Seiler 1987). A hypothesis based on facts of English, German and Dutch that can serve as an example of an implicational PUP runs as follows: other things being equal, the more analytic the language is, the more regular is its phraseological system. The first point to elucidate is the "other things being equal" condition. It refers to the fact that the phraseological organization of a language can be influenced by a variety of factors. To investigate the role of typological factors it seems reasonable to focus on languages that are of the same area and are closely related genetically, culturally and historically, but that are at the same time typologically heterogeneous. Second, I claim that the degree of analyticity is an important parameter of the desired holistic typological model. Analytic vs. synthetic features can be observed in all subsytems of language. Highly developed analyticity is characterized by the following features: the loss of inflection, the increasing role of auxiliary grammatical elements, the fixed word order, the affinity for semantic word-formation accompanied by the diminishing role of agglutination, the marked growth of lexical polysemy and homonymy, the decisive role of context in the actualization of meanings of homonymous, polyfunctional and polysemantic forms, the increasing role of verbs of superwide semantics, the prevalence of those means of nomination (e. g., conversion) that do not change the sound-form, etc. So one can think of a typological model for which the degree of analyticity is the basic parameter binding features of different modules: morphology, syntax, word-formation, and word-stock. If it turned out that phraseology, which is traditionally thought of as a typological "outsider", is to be a fifth module, it would be a good argument in favor of the possibility of holistic typology. Let us now dwell upon the notion of regularity (and the degree of regularity) of a phraseological system, since the study of phraseology in the typologically relevant direction outlined here comes to the ascertainment of correlations between the degree of analyticity and the regularity of the phraseological system. An indicator of high regularity of a phraseological system is its affinity for discreteness and orderliness, for building sets, series and paradigms, and for repeated use of the same lexical elements. The degree
298
Dmitrij
Dobrovol'skij
of regularity of a phraseological system is thus conditioned by the activity of a structural-combinatorial principle. In other words, the more regular the phraseological system of a language is, the more uniform is its internal structure and the more often phraseologization picks up structural patterns already known and previously used and, in doing so, produces combinations of a limited number of "building blocks". Consider, in this respect, the formal boundness of phraseological constituents, illustrated in (20). It is evident that the small number of the idioms with formally bound constituents in the whole phraseological system is indicative of the increased regularity of this system, and vice versa. In other words, the smaller the number of unique constituents, the more regular the phraseological system. A detailed, quantitative account of the phraseological systems of English, Dutch and German with respect to degree of regularity is given in Dobrovol'skij (1988). This study strongly supports the existence of regular connections between the inner organization of the phraseological system and the typological characteristics of the language. These regularities have the status of hypothetical phraseological universal implications proper (implicational PUP). If this conclusion is correct, then phraseology should be included into a holistic typological model. Thus the investigation of implicational PUPs is important not only for the theory of phraseology but also for the treatment of general theoretical issues in linguistic typology and linguistics of universals.
Notes 1. Phraseology is understood as a linguistic discipline whose object is set expressions of all types, from units of lexicon with strong connotation (e. g., kick the bucket) to units of lexicon with purely nominative function (e.g., pay attention to sb.jsth.) (cf. Fleischer 1982: 128—142). Together with Cernyseva I understand idioms (as distinct from other kinds of set expressions) as fixed word combinations of different structural syntactic types "whose meaning results from full or partial semantic transformation of the constituent structure" (Cernyseva 1986: 178-179). 2. In (1) and later in (3), (4), (5), (6), (8) the gloss only supplies the literal translation of the non-English idioms, their idiomatic meaning being explained in the text. Thus, all the idioms in (1) have the idiomatic meaning "defame". In general, we use single quotation marks for literal meanings, and double quotation marks for idiomatic meanings. 3. In this case we don't give any literal translation because it is practically impossible to do. Formally bound phraseological constituents, i. e., words that as a rule have no meaning whatever outside the corresponding idiom (which also means that they are not self-sufficient units of the lexicon), are for contemporary native speakers untranslatable. 4. Of course, these facts require further examination on the basis of other languages, because in different languages verbs of superwide semantics develop in different ways.
Phraseological
universals
299
References Arutjunova, Nina D. 1976 Predlo^enie i ego smysl: Logiko-semanticeskie problemy [The sentence and its sense: logical and semantic problems]. Moskva: Nauka. 1987 "Anomalii i jazyk: Κ probleme jazykovoj 'kartiny mira' " [Anomalies and language: in connection with the problem of the linguistic 'world picture'], Voprosy ja^yko^nantja 3: 3 — 19. Bally, Charles 1909 Traité de stylistique française. Heidelberg: Winter. Boucher, Jerry —Charles E. Osgood 1969 "The Pollyanna Hypothesis", Journal of verbal learning and verbal behaviour 8: No. 1. Cernyseva, Irina I. 1980 Feste Wortkomplexe des Deutschen in Sprache und Rede. Moskva: Vyssaja skola. 1986 "Phraseologie", in: Maria D. Stepanova —Irina I. Cernyseva (eds.), Lexikologie der deutschen Gegenwartssprache, Moskva: Vyssaja skola, 175 — 230. Comrie, Bernard 1987 "Holistic versus partial typologies", XIV. Internationaler Linguistenkongreß. Vorabdruck der Plenarvorträge. Berlin: Redaktionsgruppe, 213 — 232. Dobrovol'skij, Dmitrij 1988 Phraseologie als Objekt der Universalienlinguistik. Leipzig: Enzyklopädie. 1989 "Linguistische Grundlagen für die computergestützte Phraseographie", Zeitschrift für Germanistik 5: 528—536. Fleischer, Wolfgang 1982 Phraseologie der deutschen Gegenwartssprache. Leipzig: Bibliographisches Institut. Gak, Vladimir G. 1988 "Metafora: universal'noe i specificeskoe" [Metaphor: universal and specific aspects], in: Veronika N. Telija (ed.), Metafora ν ja^yhe i tekste [Metaphor in language and text], Moskva: Nauka, 11—26. Greenberg, Joseph H. —Charles E. Osgood —J. J. Jenkins 1966 "Memorandum concerning language universals", in: Joseph H. Greenberg (ed.), Universals of language, Cambridge (Mass.): Μ. I. T. Press. Jakobson, Roman 1966 "Implications of language universals for linguistics", in: Joseph H. Greenberg (ed.), Universals of language, Cambridge (Mass.): Μ. I. T. Press, 263 — 278. Karaulov, Jurij N. —Viktor I. Molcanov —Vladimir A. Afanas'ev —Nikolai V. Michalev 1982 Anali\ metaja^yka slovarja s pomosfju EVM [Analysis of the dictionary metalanguage with the help of the computer], Moskva: Nauka. Klimov, Georgij A. 1987 "Integral'naja tipologija vs. parcial'naja tipologija" [Integral typology vs. partial typology], XIV. Internationaler Linguistenkongreß. Vorabdruck der Plenarvorträge. Berlin: Redaktionsgruppe, 233 — 249. Kunin, Aleksandr V. 1970 Anglijskaja fraseologia [English phraseology]. Moskva: Vyssaja skola. Leksikograficeskaja ra^rabotka fra^eologi^mov dlja slovarej ra^licnych tipov i dlja Masinnogo fonda russkogo jaiçyka 1988 [Lexicographical representation of idioms in dictionaries of different types and in the linguistic data base of Russian]. Moskva: AN SSSR.
300
Dmitrij
Dobrovol'skij
Makkai, Adam 1978
"Idiomaticity as a language universal", in: Joseph H. Greenberg (ed.), Universals of human language, Stanford: University Press, 3: 401—448. Pavilionis, Rolandas 1983 Problema smysla: Sovremennyj logikofilosofskij anali^ ja^yka [Problem of sense: modern logical and philosophical analysis of language]. Moskva: MysP. Rajchstejn, Aleksandr D. 1980 Sopostavitel'nyj anali\ nemeckoj i russkoj fra^eologii [Contrastive analysis of German and Russian phraseology]. Moskva: Vyssaja skola. Seiler, Hansjakob 1987 "Language typology in the U N I T Y P model", XIV. Internationaler Linguistenkongreß. Vorabdruck der Plenarvorträge. Berlin: Redaktionsgruppe, 250 — 271. Staffeidt, Fred 1983
"Zu einigen philosophischen und linguistischen Aspekten der Universalienproblematik", Linguistische Arbeitsberichte. Leipzig: Karl-Marx-Universität, 39: 82-96.
Stepanov, Jurij S. 1981 Imena, predikaty,
predlofynija
[Names, predicates, sentences]. Moskva: Nauka.
Telija, Veronika Ν. 1988 "Metaforizacija i ee rol' ν sozdanii jazykovoj kartiny mira" and its role in creating the linguistic world picture], in: Boris (ed.), Rol' celoveceskogo faktora ν ja^yke: ]az(yk i Martina mira human factor in language: language and world picture], 173-204.
[Metaphorization A. Serebrennikov [The role of the Moskva: Nauka,
Ullmann, Steven 1966 "Semantic universale", in: Joseph H. Greenberg (ed.), Universals
of
language,
Cambridge (Mass.): M. I. T. Press. Vol'f, Elena M. 1986
"Ocenocnoe znacenie i sootnosenie priznakov'choroso/plocho'" [Estimating meaning and the correlation of the features 'good/bad'], Voprosy ja^yko^nanija 5: 9 8 - 1 0 6 . Wiegand, Herbert E. 1988 "Vorüberlegungen zur Wörterbuchtypologie: Teil I", in: Karl HyldgaardJensen —Arne Zettersten (eds.), Symposium on lexicography III, Tübingen: Niemeyer, 3—105. Wierzbicka, Anna 1972 Semantic
primitives.
( = Linguistische Forschungen, 22). Frankfurt am Main:
Athenäum.
Dictionaries Duden-GW 1976 — 1981
Duden. Das große Wörterbuch der deutschen Sprache, 6 Bde, herausgegeben und bearbeitet vom Wissenschaftlichen Rat u. d. Mitarb. der Dudenredaktion unter Leitung von G. Drosdowski. Mannheim: Bibliographisches Institut.
Phraseological umversah
301
Longman DC E 1982 Longman 1980 MDI 1967
Longman Dictionary of Contemporary English. Harlow: Longman. DEI Longman Dictionary of English Idioms. Harlow: Longman. Friederich, Wolf. Moderne deutsche Idiomatik.
Systematisches Wörterbuch mit De-
finitionen und Beispielen. London: Longman—Hueber. NSG 1974 NSSZ 1967 ODCIE-2 1983 RKIDS 1980
Stoett, F. A. Nederlandse spreekwoorden en ge^egden. Zutphen: Thieme. Ter Laan, K. Nederlandse spreekwoorden, Spreuken en ?egswij\en. Den Haag: Van Goor. Cowie, A. P. —R. Mackin —I. R. McCaig. Oxford Dictionary of Current Idiomatic English. Vol. 2: Phrase, Clause and Sentence Idioms. Oxford: University Press. Görner, Herbert. Redensarten. Kleine Idiomatik Bibliographisches Institut.
der deutschen Sprache. Leipzig:
Scopai quantifiers: some universale of lexical effability David Gil
1. "In place of five hundred, he would say ninenX In a short story titled Funes
the
Memorious,
J o r g e Luis Borges creates a
character endowed with a perfect memory. Funes' phenomenal m e m o r y has an interesting effect on his language: 2 He told me that in 1886 he had invented an original system of numbering and that in a very few days he had gone beyond the twenty-four-thousand mark. He had not written it down, since anything he thought of would never be lost to him. His first stimulus was, I think, his discomfort at the fact that the famous thirty-three gauchos of Uruguayan history should require two signs and two words, in place of a single word and a single sign. He then applied this absurd principle to the other numbers. In place of seven thousand thirteen, he would say (for example) Máximo Pe're%; in place of seven thousand fourteen, The Railroad·, other numbers were Luis Meltán Lafinur, Olimar, sulphur, the reins, the whale, the gas, the caldron, Napoleon, Agustín de Inedia. In place of five hundred, he would say nine. Each word had a particular sign, a kind of mark; the last in the series were very complicated ... I tried to explain to him that this rhapsody of incoherent terms was precisely the opposite of a system of numbers. I told him that saying 365 meant saying three hundreds, six tens, five ones, an analysis which is not found in the "numbers" The Negro Timoteo or meat blanket. Funes did not understand me or refused to understand me. Locke, in the seventeenth century, postulated (and rejected) an impossible language in which each individual thing, each stone, each bird and each branch, would have its own name; Funes once projected an analogous language, but discarded it because it seemed too general to him, too ambiguous. In fact, Funes remembered not only every leaf of every tree of every wood, but also every one of the times he had perceived or imagined it. He decided to reduce each of his past days to some seventy thousand memories, which would then be defined by means of ciphers. Like many other of Borges' stories, Funes the Memorious is fantasy, or science fiction. However, in creating Funes, the author does not suspend any laws of physics, n o r does he conjure into existence any fabulous technology. If Funes the Memorious is science fiction, the science in question is that of the mind: cognitive science, specifically linguistics, even more specifically lexical semantics. Clearly, n o language can have a w o r d to refer to a particular
304
David Gil
person's particular perception of a particular leaf of a particular tree of a particular wood. What Funes has done, then, is to create an impossible language — a language with an impossible lexicon. Funes the memorious raises a problem which may be called the Lexical Effability Problem. Its formulation is very simple: (1)
The Lexical Effability Problem: What are the possible denotations of words in natural languages?
With few exceptions, the Lexical Effability Problem has not been of major concern within current linguistic research. 3 Perhaps this is because many linguists feel that compared to fields such as phonology and syntax, the lexicon is too irregular and idiosyncratic for interesting generalizations to be possible. Perhaps, also, many linguists believe that it is impossible to determine the possible denotations of words in a language without taking into consideration a myriad of cultural and other extra-linguistic factors. However, our strong and unwavering intuitions regarding the impossibility of Funes' language underscore the meaningfulness of the Lexical Effability Problem, and its relevance to any universally-oriented research paradigm whose goal is to define the class of possible natural languages. In order to address the Lexical Effability Problem, we shall find it useful to introduce a number of important distinctions. To begin, one may differentiate between words in general and words of monomorphemic structure. Consider, for example, the denotation "boy". In English, this denotation is borne by a monomorphemic word boy; in Portuguese by a bimorphemic word garoto, 'child-M'; and in Tagalog by a two-word phrase batang lalaki 'childLIG male'. 4 Thus, "boy" is a word denotation in English and Portuguese but not Tagalog; however, it is a monomorphemic word denotation in English alone, not Portuguese or Tagalog. In languages with complex internal word structure, morphological processes thus make additional word denotations available. Nevertheless, the relationship between morphological complexity and word denotation inventory is far from straightforward; for example, although Tagalog has much richer word structure than English, English but not Tagalog has a word for "boy". In examining the distribution of word denotations across languages, one may distinguish between denotations present in all languages, denotations occurring in some but not all languages, and denotations unattested in any of the world's languages. For example, the sun has a word for it in all languages; a few stars, such as Sirius, have names in some languages (e. g., English) but not others (e. g., Biblical Hebrew); while most other celestial bodies do not have names in any language whatsoever. The cross-linguistic
Scopai quantifiers
305
distribution of word denotations leads to a further and more explanatory distinction, within a Universal Grammar perspective, between denotations that are necessary in Universal Grammar, denotations that are possible but not necessary, and denotations that are impossible. Clearly, if a denotation is necessary in Universal Grammar, it will occur in all languages; however, the converse may not hold: a possible word denotation may "accidentally" occur in all languages without being necessary. Similarly, if a denotation is impossible in Universal Grammar, it will not be attested in any languages; however, the opposite may not be the case: a possible word denotation may be absent "accidentally" from all languages without being impossible. In order to infer the status of a word denotation as necessary, possible or impossible within Universal Grammar from its distribution across the world's languages, it is necessary to draw a further distinction, between effability phenomena of grammatical nature, and other effability phenomena whose provenance may be more appropriately characterized as extra-grammatical. As a heuristic, a linguistic principle will be considered grammatical if and only if it has no parallels in other, non-linguistic domains. Conversely, a linguistic principle possessing clear analogues in other, non-linguistic domains will be characterized as extra-grammatical, and its explanation accordingly sought in some other field of inquiry. 5 Application of this rule of thumb to lexical effability phenomena may be illustrated in the domain of color words. Consider, first, the often-discussed cases of cross-linguistic variation in colorword inventory, with different languages carving up the color space in different fashions. A typical instance of this is English blue, corresponding to Hebrew kahol 'dark blue' and txelet 'light blue'. As has been argued by a number of researchers, such variation is not correlated in any fashion with differences in color perception, or with any other known cognitive differences, between speakers of different languages. 6 Accordingly, cross-linguistic variation in color-word inventory may be characterized as grammatical, the color words occurring in each language being specified in the language's lexicon. On the other hand, consider the large number of potential color words that are unattested in any of the world's languages. For example, no language would seem to have a word for the 500 — 610 ηιμ spectrum range, corresponding roughly to the union of green, yellow and orange, but excluding violet and blue (on the lower end) and red (on the upper end of the spectrum). Apparently, it is a fact about human vision that we do not perceive the 500 — 610 ηιμ range as constituting a unitary color. Since this fact is clearly extralinguistic, pertaining to the faculty of vision, there is no reason for it to be restated, redundantly, in Universal Grammar, in order to account for the
306
David Gil
non-existence of the appropriate color word. In this case, then, lexical effability lies outside the scope of grammatical theory. Among lexical effability phenomena of extra-grammatical nature, a further distinction may be drawn between phenomena of a cognitive character, and other phenomena, whose explanation lies ultimately in other areas, outside of cognition. The previous example, the absence of a color word for the 500 — 610 ηιμ range, is cognitive, in that it is associated with a fact about the human visual faculty. Consider, however, the superficially similar fact that no language has a color word for the 300 — 400 ιτιμ range. The reason for this is simple: the 300 — 400 πιμ range is invisible to the human eye. The nonexistence of an appropriate color word is thus explainable not in terms of cognition, but rather with reference to human physiology. Many lexical effability phenomena are clearly attributable to facts about the world. Perhaps the most celebrated of all these, 7 the Eskimos' rich inventory of snow words, is a straightforward consequence of the Eskimos' icy habitat. From a linguist's point of view, such "real world" lexical effability phenomena are perhaps less interesting than those of a cognitive nature; moreover, among the latter phenomena, those of an extra-grammatical character are possibly of less interest than those that have no extra-linguistic correlates and therefore fall within the domain of grammatical theory. The examples discussed above may give the impression that the notion of lexical effability is of more relevance to those aspects of lexical semantics characterizable as idiosyncratic, rather than "logical". However, this is not the case; interesting lexical effability phenomena are also evident in more obviously logical domains. Consider, for example, the ways in which two or more semantic roles may be united into a single word. The following examples show how five different languages synthesize the actor (A), predicate (P) and goal (G) in their renditions of the sentence "I love you": 8 (2) (3)
(4)
(5)
English I love you. Georgian Miq' varxar. l:SG:SJ-love-PRES-2:SG:OJ Bontoc Laylaycok sika. PT-PRES-love-NTOP: 1 :SG TOP:2:SG Indonesian Aku mencintaimu. 1:SG ACT-love-2:OJ
(A, P, G synthesized)
(A, Ρ synthesized)
(P, G synthesized)
Scopai quantifiers
(6)
Tagalog Mahal kit a. love NTOP:l:SG-TOP:2:SG
307
(A, G synthesized)
While in English (2), actor, predicate and goal are associated with three different words, in the other four languages, processes of affixation and cliticization enable two or more of these semantic roles to be identifed with a single word. In Georgian (3), all three roles are telescoped into a single word, a verb with subject and object agreement markers. In Bontoc (4), the actor pronoun cliticizes to the verb, while in Indonesian (5) the goal pronoun cliticizes to the verb. The synthesis of a predicate with one or more of its arguments is a widespread phenomenon in the languages of the world. In contrast, the synthesis of two or more of a predicate's arguments is extremely rare. One instance of this is provided in (6) by the Tagalog pronominal portmanteau form kita, collapsing first person actor with second person goal. However, even this form is clearly suppletive: for any other combinations of actors and goals, two independent forms will resurface, for example, Mahal mo ako 'You love me'. The exceptional nature of multiple-argument synthesis, as in (6), in contrast to the widespread occurrence of argument-predicate synthesis, as in (3) —(5), accordingly exemplifies a lexical effability phenomenon of a clearly "logical" flavor. In the remainder of this paper, we shall present an exploration of lexical effability phenomena within a particular semantic domain of eminently "logical" character, that of quantification. Indeed, it was this specific field which triggered Funes' lexicological enterprise: unhappy with the fact that the number 33 requires two words, he coined a series of number words extending much further than in any extant natural language. In place of five hundred, he would say nine, thereby replacing two words with one. Funes' interest in number words has been shared by a variety of grammarians and lexicographers. 9 In recent years, it has been mirrored also by the work of formal semanticists within the now burgeoning field of Generalized Quantification Theory. 1 0 Some of the more interesting results that have emerged from this research paradigm include a variety of universale of lexical effability pertaining to the possible denotations of words belonging to the syntactic category of Determiner. For example, in an important universal, Keenan and Stavi (1986) claim that any function from the set Ρ of properties into its power set P* satisfying a particular condition, that of conservativity, is expressible with an English word or phrase. In another universal, they maintain that any monomorphemic English determiner must be either increasing or logical. Keenan and Stavi's results effect powerful constraints on the class of possible quan-
308
David Gil
tifier words in English, and by extrapolation, in Universal Grammar. (However, they still fall short of entailing what is probably the case, that no natural language has a Funes-style monomorphemic number word for "five hundred.") In this paper, however, we shall concern ourselves specifically with lexical effability phenomena pertaining to a class of quantifier words that have remained relatively ignored within Generalized Quantification Theory, namely, portmanteau quantifiers. We shall consider a word to be a portmanteau quantifier if it combines a quantificational denotation with one or more denotations of a non-quantificational nature. Following are some examples of portmanteau-quantifier words in various languages, combining the numeral "three" with one or more non-quantificational denotations: (7) (8)
(9)
(10)
(11)
English Three-colored. Hebrew Slostenu. three-l:PL 'The three of us.' Japanese Sanko. three-CLR 'Three round objects.' Tagalog Pakipapagtatluhin. POL-INDACT-three-PT:IMP 'Please let (them) be three.' Maricopa Mxmokntuum. 2-three-SG-also-FUT 'You will also be three.'
English (7) is an adjectival compound based on the numeral three. Hebrew (8) is a noun formed from the numeral slosa by suffixation of a person marker. Japanese (9) is a construction consisting of numeral san plus a numeral classifier. Tagalog (10) is a verb formed from the numeral tatlo by prefixation of politeness and indirect-action markers and suffixation of a patient-topic formative, which by its position marks the verb as imperative. And Maricopa (11) is a verb formed from a singular verbal numeral stem xmok by prefixation of a person marker, and suffixation of a formative meaning "also" and a tense marker. In each of the above examples, then, a numeral "three" combines
Scopai quantifiers
309
with morphemes of non-quantificational denotations, thereby creating a portmanteau-quantifier word. As suggested by the above examples, portmanteauquantifier words form a heterogeneous class, reflecting the variegated morphological and syntactic patterns of the languages in which they occur. A complete and exhaustive investigation of portmanteau-quantifier words would assume monumental proportions. In the present study, we shall be concerned with a restricted subclass of portmanteau-quantifier words, namely those whose non-quantificational component involves a denotation pertaining to the relations of quantifier scope and distributivity. We shall refer to such words as scopai quantifiers.11 In this paper, we shall present some results of a cross-linguistic survey of scopai quantifier words. The survey will be limited to adnominal quantifier words, namely words quantifying nouns, and occurring in construction with the nouns they quantify. Some examples of adnominal quantifier words are the prenominai English quantifiers every, all, three and so forth. Excluded from consideration will be words that quantify verbs, for example twice (as in, say, George has been to Uruguay twice'), and words that quantify nouns but do not occur in immediate construction with them, for example the adverbial always (as in one possible interpretation of Birds always have wings). The results of this survey reveal that the class of possible scopal-quantifier words is surprisingly small. These findings will suggest a number of universals of lexical effability for scopal-quantifier words, of far-reaching empirical and theoretical import.
2. Distributive-key quantifier words In English, there are but two clear examples of scopal-quantifier words: the universal quantifier every, and its near synonym each.12 Consider the following paradigms: (12) a. b. c. d. e. f. g. h.
* Every man gathered at the stroke of midnight. * Each man gathered at the stroke of midnight. All men gathered at the stroke of midnight. The men gathered at the stroke of midnight. Two men gathered at the stroke of midnight. Some men gathered at the stroke of midnight. Several men gathered at the stroke of midnight. Many men gathered at the stroke of midnight.
310
David Gil
i. j· a. b. c. d. e. f. g· h. i. i-
Most men gathered at the stroke of midnight. Few men gathered at the stroke of midnight. Every man carried three suitcases. Each man carried three suitcases. All men carried three suitcases. The men carried three suitcases. Two men carried three suitcases. Some men carried three suitcases. Several men carried three suitcases. Many men carried three suitcases. Most men carried three suitcases. Few men carried three suitcases.
Each of the ten quantifiers exemplified in (12) and (13) above is of plural cardinality; however, the NPs every man and each man exhibit quite different semantic properties from the remaining eight subject NPs. In (12), the predicate gathered at the stroke of midnight imposes a collective interpretation on its subject NP. While most of the semantically plural NPs in (12) may indeed be interpreted collectively, the NPs every man and each man are incompatible with such a collective interpretation; hence, sentences (12a) and (12b) are ungrammatical. In (13), the predicate carried three suitcases allows its subject NP to be interpreted either collectively or distributively; hence all of the sentences in (13) are grammatical. However, the NPs every man and each man permit a more restricted set of scope relations than the remaining subject NPs. In sentences (13c) —(13j), at least four scope relations are potentially available: a strong symmetric relation, in which each of a unique set of men carried each of a unique set of three suitcases; a weak symmetric relation, in which a unique set of men carried a unique set of three suitcases between them; an asymmetric relation with wide scope for the subject NP, in which each of a unique set of men carried a set of three suitcases, and an asymmetric relation with wide scope for the direct-object NP, in which each of a unique set of three suitcases was carried by a set of men. Of these four relations, the symmetric ones, involving unique sets of men and suitcases, are generally more readily available than the asymmetric ones, involving a scope dependency between one NP and another; however, arguably, all four scope relations are available for each of (13c) — (13j). 13 In contrast, sentences (13a) and (13b) permit only two of the above four scope relations: a strong symmetric scope relation, in which each of a unique set of men carried each of a unique set of three suitcases; and an asymmetric scope relation with wide scope for the subject NP, in which each of a unique set of men carried a set of three suitcases. The remaining two scope relations are totally unavailable.
Scopai quantifiers
311
As indicated by the above paradigm, every and each exhibit exceptional semantic behavior among the semantically plural quantifiers of English. We may account for the distinctive behavior of these two words by characterizing them as scopai quantifiers. As many scholars have observed, every and each share a core quantificational denotation, that of universal quantifier. However, as evidenced by the constrasts in (12) and (13), every and each permit a more restricted class of interpretations than the other universal quantifier, all, and most other quantifiers, such as two, some, several, and so forth. This property of every and each may be represented in terms of the binary asymmetric relation of distributivity, in which a constituent Y, the distributive share, distributes over a constituent X, the distributive key.u Specifically, while NPs of the form Q Ν are characteristically free either to enter into a relation of distributivity with another constituent or not to do so, NPs of the form every Ν and each Ν must always enter into such a relation; specifically, as distributive key. Thus, we may characterize every and each as scopal-quantifier words, their denotations consisting of two distinct components: a quantificational component, of universal quantifier, and a scopai component, of distributive key. Referring more specifically to the latter, scopai component, we may characterize every and each as distributive-key quantifiers. The characterization of every and each as portmanteau quantifiers implies that they are more highly marked than ordinary quantifiers, by dint of their additional denotational component, as distributive keys. The characterization of every and each as marked is supported by a number of arguments. To begin with, as noted above, every and each are exceptional, as the only English quantifiers forcing distributive interpretations. Furthermore, as also noted above, asymmetric scope relations, in which one quantified NP distributes over another, are more marked than symmetric scope relations, in which quantified NPs are independently referring, and no relation of distributivity obtains. Thus, for example, in sentences such as (13e) Two men carried three suitcases, interpretations involving exactly two men and exactly three suitcases are much more readily available than interpretations involving two men and three to six suitcases, or two to six men and three suitcases. 15 Facts such as these suggest that distributivity is a semantically marked relation. It is therefore reasonable that quantifiers imposing distributivity also be marked relative to ordinary quantifiers, by dint of their portmanteau structure. Further support for the characterization of every and each as marked is provided by their exceptional morphosyntactic behavior. Thus, every and each are the only semantically plural quantifiers to take a singular head noun; similarly, NPs of the form every Ν and each Ν are the only semantically plural quantified NPs to trigger singular verbal agreement. Moreover, NPs of the
312
David Gil
form every Ν and each Ν are the only semantically plural quantified NPs that may license singular anaphors. In the following examples, antecedents of the form every N, in (14a) and (15a), are contrasted with antecedents of the form all Ν, in (14b) and (15b): (14) a. b. (15) a. b.
Every student here thinks he is going to be a linguist. They have no idea how bad the job situation is. All students here think they are going to be linguists. They have no idea how bad the job situation is. Every graduate will receive his own special diploma. He will also be awarded a free trip to Cambridge, Massachusetts. All graduates will receive their own special diploma. They will also be awarded a free trip to Cambridge, Massachusetts.
While NPs of the form every Ν may license either singular or plural anaphors, NPs of the form all Ν license only plural anaphors. Thus, in (14a), every student licenses a singular bound anaphor he and a plural discourse anaphor they, while in (15a) every graduate licenses a singular bound anaphor his and a singular discourse anaphor he.16 Analogous observations hold with respect to similar NPs containing each. Thus, NPs of the form every Ν and each Ν exhibit distinctive morphosyntactic behavior, as the only semantically plural NPs containing singular head nouns, triggering singular verbal agreement, and licensing singular anaphors. The exceptional morphosyntactic properties of every and each accordingly provide further support for their characterization as marked portmanteau-quantifier words. However, the characterization of every and each as more highly marked runs contra to what seems to be a widespread tradition in linguistic semantics. Ask any linguist to name an English quantifier, and the chances are he or she will say every, most linguists seem to view every as the prototypical member of whatever category it is supposed to belong to. Let us therefore briefly entertain some of the consequences of taking every to be an unmarked quantifier, with a primitive, unitary meaning. First, we would have to characterize the universal quantifier all as ambiguous; specifically, we would be forced to posit "two alls", an allx, synonymous with every, with distributivekey denotation, and an all2 for collective interpretations. However, our task would not end here. As suggested in (13) the range of interpretations permitted by all is similar to that allowed by most other quantifiers. In analogy to