Language typology 1985 : papers from the linguistic typology symposium, Moscow, 9-13 December 1985 9789027235411, 9027235414

Table of contents :
Table of Contents
Preface
Primes
1. Introduction
2. Language as a "social reality" based on relationships
3. Analysis of language by means of categories
4. Treatment of language as a system of communication with three sections: grammatical, semantic, pragmatic
5. Syntax as the central component
5.1. Determination of primes in syntax
5.2. Primary government relationships
5.3. Primary modification relationships
5.4. Characteristic patterns in compound sentences
5.5. Secondary role of morphology and selection
5.6. Possible role of sandhi
5.7. Role of modulation (intonation)
5.8. Fundamental relationship in V and O
5.9. Other bases for syntactic typology
5.10. Phonological typology
6. Textual patterning
7. Semantic and pragmatic typology
8. Typology based on the treatment of the noun as primary; the fallacy of taking traditional logic as basis
9. Proposed tasks for typological study
9.1. Historical typology
9.2. Exploration of 'absolute universais'
9.3. Language typology and language acquisition
Basic Typological
Units
Areal Phonetic
Typology in Time: North and East Asia
1. The Systems
2. Subsystems
3. Areas, Foci
Lexico-semantic Reconstruction and the Linguistic Paleontology of Culture
Universals Specials and Typology
Commensurability of Terms
1. Morphology
2. Alignment
3. Voice and Changes of Grammatical Relation
4. Complementation
5. Conclusion
Metalanguage
1. Introduction
2.
Consistency of Orientation
3. Hierarchization
4. Predicate-Actant Structure
5. Predicate Formation Operators
6. Correspondence of Function-Argument Structure
7. Conclusion
On the Notion of Language Type
On Typological Shift
Discourse Function and Word Order Shift: A Typological Study of the VS/SV Alternation
1. The VSO/SVO/SOV Typology: Some General Remarks
2. Some Functional-Grammatical Preliminaries
2.1. Preferred Argument Structure
2.2. VS/SV Alternating Languages
2.3. Categoriality
2.4. Transitivity
3. Malay
3.1. Clause Types
3.1.1. Preferred Clause Structure
3.1.2. Clauses with Lexical Agent and Patient
3.2. Typological Shift in Malay
3.3. Morphological Correlates of the VS-to-SV Shift?
3.4. Categoriality and Typological Shift
3.5. Reporting and Predication
4. Germanic
5. General Discussion
6. Conclusion
On Form and Content in Typology
1. Introduction
2. Reflexivity
2.1. Split intransitivity
2.2. Syncretism
3. Split possession: The semantics of alienable and inalienable possession.
4. Form determines content
4.1. Alienable and inalienable possession
4.2. Split intransitivity
4.3. Case meanings
5. Lability
6. Conclusions
The Meaning-Form Correspondence in Grammatical Description
A Survey of Major Alaskan Language Types
1. Phonology
2. Morphology
3. Syntax
References
Index

Citation preview

LANGUAGE TYPOLOGY 1985

AMSTERDAM STUDIES IN THE THEORY AND HISTORY OF LINGUISTIC SCIENCE General Editor E.F. KONRAD KOERNER (University of Ottawa)

Series IV - CURRENT ISSUES IN LINGUISTIC THEORY

Advisory Editorial Board Henning Andersen (Copenhagen); Raimo Anttila (Los Angeles) Thomas V.Gamkrehdze (Tbilisi); Hans-Heinrich Lieb (Berlin) J.Peter Maher (Chicago); Ernst Pulgram (Ann Arbor, Mich.) E.Wyn Roberts (Vancouver, B.C.); Danny Steinberg (Tokyo)

Volume 47

Winfred P. Lehmann (ed.) Language Typology 1985

LANGUAGE TYPOLOGY 1985 Papers from the Linguistic Typology Symposium, Moscow, 9-13 December 1985

Edited by

WINFRED P. LEHMANN University of Texas at Austin

JOHN BENJAMINS PUBLISHING COMPANY AMSTERDAM/PHILADELPHIA 1986

Library of Congress Cataloging in Publication Data Language Typology 1985. (Amsterdam studies in the theory and history of linguistic science. Series IV, Current issues in linguistic theory, ISSN 0304-0763; v. 47) Revised papers from the Colloquium in Linguistic Typology, held in Moscow, December 8-13, 1985 and sponsored by the Institute of Linguistics, USSR Academy of Sciences. Bibliography: p. Includes index. 1. Typology (Linguistics) — Congresses. I. Lehmann, Winfred Philipp, 1916. II. Col­ loquium in Linguistic Typology (1985: Moscow, R.S.F.S.R.) HI. Institut iazykoznaniia (Akademiia nauk SSSR) IV. Series. P204.T97 1986 410M2 86-26341 ISBN 90-272-3541-4 (alk. paper) ® Copyright 1986 - John Benjamins B.V. No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publisher.

T a b l e of C o n t e n t s

Preface

vii

Primes

1 W. P.

Lehmann

Basic Typological Units V. TV. Yartseva

19

Areal Phonetic Typology in Time: North and East Asia Robert Austerlitz

27

Lexico-semantic Reconstruction and the Linguistic Paleontology of Culture T. V. Gamkrelidze

43

Universais Specials and Typology V. M. Solntsev

49

Commensurability of Terms Alice C. Harris

55

Metalanguage Alan

77 Timberlake

On the Notion of Language Type G. A. Klimov

105

On Typological Shift M. M. Gukhman

111

vi Discourse Function and Word Order Shift: A Typological Study of the V S / S V Alternation Paul J. Hopper On Form and Content in Typology Johanna Nichols The Meaning-Form Correspondence in Grammatical Description A. E. Kibrik

123

141

163

A Survey of Major Alaskan Language Types Michael E. Krauss

169

References

187

Index

205

Preface

The essays included here are revised forms of papers prepared for the Colloquium in Linguistic Typology which was arranged in Moscow, 8-13 December 1985 by the Institute of Linguistics, USSR Academy of Sciences, under the direction of Dr. Viktoria Nikolayevna Yartseva. The colloquium was held under the auspices of the ACLS-Soviet Academy of Sciences Commission on the Humanities and Social Sciences administered in the U.S. by The International Research and Exchanges Board. It was planned at a meeting in Austin in December 1984; participants representing the Soviet side were Drs. Yartseva, Aleksandr Davidovich Shveitser, Vadim Mikhailovich

Solntsev;

representing

the

American

side

were

Robert

Austerlitz, Wesley Fisher, and the editor. At the Austin meeting eight topics were selected for papers: 1. Primes 2. W h a t is language type? 3. Form versus content 4. Metalanguages 5. Commensurability of terms 6. Different structures and their relevance to typology 7. Areal typology 8. Typological shift These topics were treated by members of each delegation in accordance with their own views, in papers t h a t were circulated, then presented orally in briefer form, and discussed.

I regret t h a t the excellent paper of

V. A. Vinogradov, "Different structures and their typological relevance,"

viii did not reach us for inclusion here.

The remaining papers, revised after

presentation, obviate the necessity of reporting here on the discussions. Colloquia and other I R E X programs are designed to acquaint scholars of each country with one another's contributions. gratitude

to

IREX,

its officers

and

linguistics among its programs.

funding

We

express

agencies, for

our

including

We give special thanks to

Wesley

A. Fisher, the indefatigable secretary to the Commission, for arranging the visit and to IREX for making preparation for publication possible. Print-ready copy of the papers was produced by Earl Roy, who in addition to his skillful formatting of papers submitted in a remarkable array of computer conventions and typescripts, contributed to the editing. Johanna Nichols also deserves credit for turning the odd impenetrable phrase into lucid text, as does Gail Roy for her ever capable management throughout the project. We also express our gratitude to our hosts in the Academy, notably Dr.

Yartseva, who was leader of the Academy delegation, for excellent

arrangements

and

generous hospitality.

Members

of our

delegation

appreciated very much the opportunities to discuss linguistic views with the Academy delegation and also with other members of the vigorous linguistic groups in the USSR.

In addition to views sharpened through

these opportunities, readers may detect in occasional phrases of the essays the rhythms of the Bolshoi Ballet, even of Svatislav Richter, or reflections of the ancient monuments at Zagorsk.

Comments on the essays will be

welcome, especially for planning the second meeting of the two delegations which will be held in this country. Winfred P . Lehmann Austin, Texas 28 July 1986

Primes W. P. Lehmann

1. I n t r o d u c t i o n Linguistic typology has many achievements, as reported in many recent publications, but also many differences in approach, even contradictory approaches. Hjelmslev held out high hopes for it, in listing its goals: "Its ultimate aim must be to show which linguistic structures are possible, in general, and why it is just those structures, and not others, t h a t are possible . . . [Adding], only through linguistic typology can we hope to understand what laws govern linguistic change and what possibilities of linguistic change a given linguistic type implies.

Only through typology

does linguistics rise to quite general points of view and become a science" (1970:96). Vennemann by contrast specifically rejects each of Hjelmslev's proposed aims (1982:41).

His negative attitude contrasts strikingly with

t h a t of the participants in the Stanford Project on Language Universais, who maintain a "generalizing approach designed to uncover and explain the common properties of language"

(Greenberg ed. 1978 I.1), while

contributing in their four volumes the results of a large-scale investigation into many characteristics of language. Students of typology like Hjelmslev see in this study the means for achieving an explanatory goal, which linguists from the time of Schleicher have set as the highest goal of linguistic study, as reiterated subsequently by Noam Chomsky, also earlier by William Dwight Whitney. The bibliography on typology is massive, too great for complete attention here, even when leaving the Soviet contributions to our hosts. To assess the achievements as well as the differences in viewpoint, to

2

W. P. LEHMANN

determine the place of typology in linguistics and thereupon its current tasks, it is highly important t h a t we draw on the contributions of all previous research.

It is also important to come to terms on primary

principles, as the planning group for this conference has proposed.

For it

is essential to proceed from a secure position achieved on the basis of previous

linguistic

scholarship

if one

is concerned

with

more

than

temporary acceptance in the task of understanding language and its role in human society. We are very grateful to our hosts in the Soviet Academy of Sciences, as well as to their counterparts in the International Research and Exchanges Board and in the American Council of Learned Societies for making the colloquia possible. 2. L a n g u a g e as a "social reality" b a s e d on r e l a t i o n s h i p s Adequate typology of language, like adequate

attention to

language,

requires a theory of language, a theory based on the achievements of previous scholars dealing with language and derived from knowledge of various tongues.

[Archaic usages have their problems, but the English

lexicon is so confusing in terms for speech and language t h a t I here use 'tongue' for a specific language.]

In proposing such a theory we need to

state our fundamental positions, or in accordance with the topic and title selected by the planning group, our primes.

Chief among these is the

position we adopt regarding language. Typologists treat language in a phrase of Lazio Antal's as an "objective social reality" (Paikeday 1985:59). Linguistics has as task the analysis and explication of that reality. Because of a kind of attention to language fashionable in this last generation it is necessary to contrast this position with t h a t propagated by Noam Chomsky, to contrast it briefly and without polemics. By his view "a language is a system L-s, it is the steady state attained by the language organ" (Chomsky's statement in Paikeday 1985:58).

I do not deny the

possibility t h a t this position is valid. It is simply beyond our reach today. With current knowledge of the brain we cannot identify a language organ. As research on mental activities progresses, one may be found, and the structures may be mapped which determine the projected steady state. Jerne in his Nobel lecture of 1984 provides some indication of the extent of the task by noting in his reflections on language with reference to the

3

PRIMES

immune system t h a t the immune system includes more than 10,000,000 different proteins (1985). He discusses the immune system with reference to language because both systems are independent of the brain. It has long been acknowledged that language acquisition is not an intellectual activity. Further, both systems are obviously complex. While biologists are now at work identifying the proteins of the immune system, linguists have a far greater problem with the language system. Language cannot be explicated by means of physical entities like molecules, however many, but only by means of relationships. These relationships are ample enough to permit communication not only regarding ten million previously unknown entities, but also regarding countless numbers more in other spheres of knowledge. In short, language is an immensely complex system, capable of comprehending all possible phenomena, yet open to ready scrutiny as a social reality. The key to its capabilities lies not in recognizable physical substances, but rather in its constructions through relationships among categories. 3 . A n a l y s i s of l a n g u a g e b y m e a n s of c a t e g o r i e s Among linguists this position is without dispute.

Linguistic entities,

whether phonological, syntactic or semantic, are not physical objects. Scientific study with increasingly advanced instruments has continued to demonstrate t h a t a linguistic entity like / s / or / i / cannot be established by locating a physical correlate like a molecule, an atom, a quark, whatever.

or

Nor can more complex linguistic entities, such as syntactic

elements like: See! be identified by physical phenomena. / s / may be an element in a phonological system of a given tongue by virtue of its relationships to other elements, as it is in English. sentence See! She!

Similarly / i / .

also is a unit through contrast with others such as Sew!

The or

Because linguistic elements are determined by relationships and not

by physical characteristics, linguists analyze language for categories, not for physical objects of whatever

size or

complexity.

discussion in this paragraph may seem unnecessary.

For

linguists

the

When however a

contrary view is presented by an eminent biologist and circulated among hundreds of thousands of readers, the statement is not without pertinence in noting our primes.

4

W. P. LEHMANN The categories of language may be comprehensive, like the two

categories consonant and vowel. They may be less comprehensive, like any one of the five vowels of Japanese. Even entities of such small categories may include variants, such as the voiceless variant of the Japanese vowel / i / between voiceless consonants.

To treat these, linguists may propose

phonological features, which again are not physical entities, but rather categories.

Accordingly, whether we identify more or less comprehensive

features, we deal with categories. We do so as well in dealing with forms, identifying nouns, verbs, transitives, intransitives, and so on; with sentences, identifying statements, commands, and so on; with semantic matters, identifying sets of mental verbs, sets of speech act verbs, and so on; with pragmatic

concerns,

explicating matters such as the avoidance of center-embedding.

The

categories we adopt may vary. Some scholars operate with highly abstract logical categories. Acceptance by others is achieved in accordance with the general procedures of scientific work.

Whatever the acceptance, adequate

presentation of one's categories is essential. 4. T r e a t m e n t of l a n g u a g e as a s y s t e m of c o m m u n i c a t i o n w i t h three sections: grammatical, semantic, pragmatic It is also essential that the framework for these categories be identified. From the statements above it is clear t h a t the framework is to be viewed as consisting of identifiable portions, such as the phonological portion. Less plain terms for such portions are level, stratum, component. A highly regarded framework owes much to Charles Sanders Peirce, who placed language among communication systems. three major relationships:

To deal with these we posit

(1) those between members of categorial sets;

(2) those between members of categorial sets and the outside world, or, if preferable, between such sets and concepts concerning the outside world; (3) those between categorial sets and the users of the system.

Treatment

of the first now is generally referred to as grammar, rather t h a n Peirce's syntactics; of the second, as semantics; of the third, as pragmatics (Morris 1964:44-48). Comprehensive analysis of a system of communication accordingly requires a grammatical, a semantic, and a pragmatic section.

Scientific

5

PRIMES

attention to the semantic and pragmatic sections is in its infancy. linguists even deny a distinction between them.

Some

These sections of the

system offer many opportunities for future work in typology.

General

agreement has not even been reached on the grammatical section. But we can at least point out components which have been proposed and identify the set of these selected in further attention.

Such components deal with

the text, the sentence, the word-form, and the phonological units. Identification of the named components implies attention to human language, further, to language as an oral means of communication.

For

Peirce, any type of communication might be examined by his scheme. Among examples are communication of humans by means of written symbols, by their selection of clothing among other activities, and of communication systems of non-humans, such as anthropoids, porpoises, birds.

Each of these types of communication may be worth scientific

study, and may contribute to the understanding of other systems of communication such as the oral human communication system.

But this

has specific characteristic features, and I, presumably all of us here, concentrate on the features of oral human language.

I assume the

grammar consists of 4 components: phonology, morphology, syntax, textual analysis. 5. S y n t a x as t h e central c o m p o n e n t Among these I take syntax as the most characteristic and most specifically linguistic component, that is, one with the least effect from non-linguistic forces.

The phonological component is affected by constraints imposed by

the human vocal apparatus.

The morpho-lexical component is affected by

constraints imposed by the outside world; gender systems, for example, reflect some kind of non-linguistic classification, whether or not they are based

on

the

transparent

classification

of sex.

Only the

component is free, or virtually free, of such constraints.

syntactic

Support for this

conclusion may be taken from efforts at iconic effects in use of language. Poets and rhetoricians must take special steps to achieve such effects, especially at the syntactic level.

The textual component, while based on

the syntactic, includes characteristics which have more direct reference to the outside world, and is therefore less purely linguistic. Further evidence

6

W. P. LEHMANN

for the characteristic prominence of the syntactic component in oral human communication

may

be

provided

by

comparison

with

systems

of

communication among other animals, such as birds. These systems consist of sounds with specific meanings; but they lack syntax.

Among other

possibilities, high or low notes cannot be reversed to mark a question as against a statement.

The calls, cries, howls are stereotyped chunks.

For

these reasons I take the syntactic component as central in human language, and therefore central in the typology of human language. 5 . 1 . D e t e r m i n a t i o n of p r i m e s in s y n t a x Determining primes in syntactic typology we may note the basis of its essential relationships, or as Bloomfield put it, the "ways of arranging linguistic

forms"

selection,

modulation,

(1933:163-64). sandhi.

He identified four such ways: order, Of

these

I take

order

as

primary.

Modulation and sandhi rely heavily on phonological possibilities. Selection relies on availability of items in sets often reflecting the outside world. Only order has no such constraints. linguistic forms with

It rests solely on arrangement of

regard to one another.

As a major

step

in

establishing our typology we sort out the forces governing the meaningful relationships in the syntactic patterning known as order. Like Bloomfield I identify these forces as government and agreement (his concord/congruence), leaving aside cross-reference (1933:191-93). these, government has the dominant role in the clause.

If the clause

consists of more than a verb, such as Latin pluit '(it) is raining', '(it) hurts', the basic force is government, as in videt canem [the] dog'.

These two items: videt

and canem,

categorial positions of a simple, transitive clause.

Of

paenitet

'(s/he) sees

occupy the

primary

The arrangement may

vary among languages, as in Japanese inu o miru 'dog see [s/he]'. But the force is the same in the two languages, as may be observed from the Japanese particle  or the Latin -em < -is.

In formation of the simple

clause one item, typically a verb, maintains a relationship to another which we may refer to as government.

The force bringing about the government

relationship—transitivity—is the focus of considerable recent attention, as in textual studies by Hopper and others. As in any scientific research we look forward to advances from scholars dealing appropriately with data; but past work on government provides ample information for our purposes.

7

PRIMES 5.2. P r i m a r y government relationships

The primary relationship in the smallest unit of language, i.e., the clause or simple sentence, is brought about by government, the control of one or more objects by another, such as the verb. Clauses to be sure may consist solely of a verb, as in Sanskrit varsati

'rains', or of an intransitive verb, as

in Sanskrit gacchati 's/he goes'. When they include an object, the order of verb and object is diagnostic, revealing a characteristic feature of the language concerned, as numerous recent studies demonstrate.

For the

order is reflected in other patterns, such as adposition and noun.

The

prominence of such parallel patterns in consistent languages leads to the basic classification: OV and VO languages.

The order is also reflected in

the pattern: adjective and standard in comparison of inequality.

[Such

comparison may be expressed in different ways, as Andersen has indicated at some length (1983); but when expressed with adjective and standard, the order reflects t h a t of verb and object.

Andersen's book is valuable in

reporting on the richness of language, illustrating t h a t it is not the preprogrammed system of communication which sole attention to English may lead some to believe; see also Beloded 1977:20.] patterns of order

are titles, family

and

Among

further

given names, additive

teen

numerals, even patterns recent in human activities, such as addresses (Lehmann 1978:16-17). 5.3. P r i m a r y m o d i f i c a t i o n r e l a t i o n s h i p s Of other major constructions related to order of verb and object, equally striking is the number of those known as modifiers. If one takes relationship as basic in language, one cannot merge government and modifier constructions, using a device of formal logic, as Vennemann does (1974 and elsewhere). For the patterns of relationship in government and modifier constructions differ totally, as the designations suggest. The difference is widely manifested in surface morphological phenomena; modifiers do not control inflection of their heads, unlike the control exerted by governing elements. It is also essential to emphasize that modification extends far more widely than the few noun modifier patterns commonly cited, i.e., relative clauses, genitive constructions and descriptive adjectives. To these we add

8

W. P. LEHMANN

adjectival numerals and limiting adjectives. Also of major significance are patterns of verb modification, such as interrogation, negation, middle, modality, aspect/ tense; see for example Dryer 1985u.

aspect,

Here too

patterns of relative order with reference to verb and clause boundary are found.

The contrasting treatment of tense, a deictic category, and aspect,

an internal linguistic category, provides further insights into the primacy of non-deictic syntactic constructions. Tense indication is far less predictable for

relative order than

is aspect,

an irregularity we may

relate

to

irregularity in such lexical sets as deictic pronouns, personal names, and so on.

These

various

constructions

and

lexical

sets

exhibit

many

irregularities, in contrast with non-deictic constructions and lexical sets. The effects of deixis even within the syntactic system illustrate once again the importance of looking first of all to t h a t component of language which is least affected by non-language constraints, the syntactic. 5.4. C h a r a c t e r i s t i c p a t t e r n s in c o m p o u n d s e n t e n c e s OV and VO languages also differ in compound and complex sentences. Just as verbal objects precede verbs in OV language, so object clauses precede the principal clause. OV languages commonly include quotatives, like Japanese to, to indicate quotations. VO languages, on the other hand, may indicate quotations with special verb forms and the like. Further, hypotactic clauses precede principal clauses. Among such hypotactic clauses are complements, which are often reduced, as in English, e.g., beside:

we find and

We expected t h a t they would arrive late We expected them to arrive late We expected their arriving late

I return below to evidence from historical syntax, but mention here t h a t among the most remarkable and unexplored phenomena in historical syntax is the development of new complement patterns when a language shifts from one fundamental pattern to another. This recognition provides the key to explaining the diverse syntactic phenomena of complex sentences in the early Indo-European tongues, a topic which has scarcely been touched. In sum, any adequate treatment of syntactic typology must deal with

9

PRIMES

the comprehensive set of modifier and government constructions. And as I have indicated, besides being the central component of language, syntax with its essential basis in order reflects the necessary linearity of language. Order is therefore the central device, recognized as such by many authorities on language. 5.5. S e c o n d a r y role of m o r p h o l o g y a n d s e l e c t i o n Inflectional morphology, by contrast, simply amplifies selection devices, as does derivational morphology.

For example, if the verb in Sanskrit, in

contrast with Japanese or Chinese, includes patterns for number, the selection devices are merely expanded.

person

and

Similar expansion is

made available in the Sanskrit noun through inflection for case and gender, neither morphologically expressed in Japanese or Chinese.

The devices

found in a given language for selection are thus secondary.

It was

accordingly a major error for 19th century scholars to take morphological characteristics like synthesis and agglutination in contrast with analysis as primary.

It is a more serious error to maintain their procedure.

Finck's

classification by morphological structure should have provided through its inadequacies a definitive example of the secondary position of selection devices (1909).

Later attempts by Sapir (1921) and Greenberg (1960) to

rescue morphological classification serve to underwrite this verdict.

There

is little point in citing other work in this vein. This is not to say that morphological characteristics should not be taken into consideration in typological studies.

Of major interest are

possible relationships between morphological patterns and syntactic type. These are well-recognized in derivational morphology, as in the VSO pattern of Ibn-Saud versus the OV pattern of Svenson. Moreover, VSO languages rely largely on prefixation, OV languages on suffixation. also

recognized

that

morphological

patterning

is

It is

conservative.

As

languages change their fundamental structure, they may well preserve earlier patterns of morphology.

The relationship between morphological

and syntactic structure is therefore of great interest also in historical study. Inflectional

morphology

as

well

observes

principle of OV or VO structure, though

the

again

general

patterning

ordering may

be

conservative, W h a t is of further interest is the kind of patterning, such as

10

W. P. LEHMANN

the apparent tie betwen OV order and agglutination. F o u n d in languages with no areal contact, such as Quechua and Turkish among many others, the tie needs attention. 5.6. P o s s i b l e role of s a n d h i The tie may be associated with another syntactic device, sandhi.

Joining

of phonological, and hence also morphological, elements tends to be loose in OV languages, with a minimum of sandhi changes. inflectional

elements

agglutinative structure.

thus

retain

their

Derivational and

independence,

resulting

in

As with selection, the device of sandhi is less

central than is arrangement, and consequently more readily affected by areal forces and conservatism. In this way it too will yield further insights in synchronic and diachronic study as we learn more about its relationships to the central pattern of arrangement. 5.7. R o l e of m o d u l a t i o n ( i n t o n a t i o n ) Bloomfield's fourth device of arrangement, modulation, consists of patterns made up of stress, pitch, and length phenomena, kinds of which have been referred to as juncture. My only concern here with this syntactic device is its possible relationship to the device of order. Unfortunately intonation is poorly known. Moreover, its patterns seem to be conservative in language, and also reflect areal influences.

With these cautions, I have suggested

t h a t the use of stress in phonemic suprasegmental patterns may be favored in VO languages, in contrast with the favoring of pitch to the exclusion of phonemic stress in OV languages.

Support of such generalizations must

await assembling of information, and study of this in connection with historical and geographical situations of given tongues.

Until t h a t time

any general conclusions must be highly tentative. 5.8. F u n d a m e n t a l r e l a t i o n s h i p in V a n d  In sum, the primary syntactic constructions depend on the relationship of V and  in a specific tongue. Correlated with it, in my view dependent on it, in consistent languages are many elementary patterns of order, as noted here, which make up much of the basic clause. Further correlations exist with morphological structures, though these reflect historical patterning even more than do less central patterns of order. These interrelationships,

11

PRIMES

which I formulated in a structural principle some time ago (1973) have received further attention, as I note below. In view of the basis of typology in "language as a social reality," further elaboration will be determined by improved and broadened attention to data in accordance with standard scientific procedures. 5.9. O t h e r bases for s y n t a c t i c t y p o l o g y Syntactic typology has also been pursued with other bases, such as the ergative relationship (Klimov 1983). I look forward to discussions on these at the conference. But it may be clear that I have regarded the proposed opposition between so-called nominative and ergative languages as secondary. The contrast is essentially based on morphological characteristics. VO and OV languages may be ergative as well as nominative. The very labels O V / V O , with omission of an. agent/ subject category, imply a non-central status for this contrast in language. At one time, to be sure, it was suggested that ergativity had been found to be syntactic in a specific tongue. But t h a t proposed analysis was not wellfounded, and accordingly I maintain the position t h a t ergativity is a selection device, hence secondary to the patterns of order centering around the O V / V O relationship. Other proposed types, active and class, also seem secondary to me, like ergative as opposed to nominative, characterized by patterning of less central linguistic features. Exploration of their position in the structure of language is however of great interest. 5.10. P h o n o l o g i c a l t y p o l o g y The considerable amount of attention to phonological typology, as by Trubetzkoy (1939), Hockett (1955), Ruhlen (1976), indicates t h a t it must be treated independently of content-oriented typology. no relationship to those of specific syntactic types.

Its elements have

If, however, syllable

structure is included among patterns of phonological typology, there may be correlations, as of open syllables to agglutinative morphology and OV syntax.

But I do not pursue phonological typology here in view of its

independence, in spite of the widely known achievements. It may be useful to recall briefly the fundamental

position of

relationship also in phonological typology. Relationships are noted through

12

W. P. LEHMANN

features, such as High, Front, Strident, which are proposed as universais. These then are represented through configurations, whether triangular as in the Latin vowel system or rectangular as in the Turkish vowel system, and so on (Jakobson/Halle 1956).

Such relationship patterns are well

understood, for example, t h a t a three-vowel system will consist of / i a u / , not /Ü  å/ or the like.

Further, a five-vowel triangular system will

consist of / i e a  u / .

Hierarchies of feature combination, and of

patterning may therefore be demonstrated in phonological as well as content-based typology.

The underlying principles are then comparable,

though the two systems are largely independent. 6. T e x t u a l p a t t e r n i n g The

increasing

number

of

textual

studies

have

led

to

increased

understanding of the entities employed in textual patterning and the processes involved.

For simplicity I note here only the terms theme and

rheme, omitting comparable terms, and other proposed entities such as focus.

Definition of theme and rheme as carriers of 'old' versus 'new'

information

are

widely

accepted,

as

is their

order

relationship;

in

straightforward patterning themes stand earlier in the clause than do rhemes.

But other devices than order are employed in textual structures.

These devices are associated with the basic syntactic patterns, though in ways t h a t need further exploration. Moreover, since they are concerned with marked uses, the effects are open to varied interpretation.

For the

present any statements must be tentative. Among

the processes are passivization,

particles, shifts in intonation.

clefting,

employment

of

Passivization in the strict sense refers to a

process by which an object in a basic clause is promoted to subject in a derived clause, as in: Mary was given the book by John The book was given to Mary by John John gave Mary the book

versus

(It may be noted that the term passive was imported for verb forms indicating other shifts in use, to the possible confusion of non-specialists in the grammar concerned, e.g., Japanese. The so-called Japanese passive had three uses: potential, honorific, expression of misfortune. Through

13

PRIMES external influence it is increasingly used like the passive In English.

This

observation on the term is introduced to illustrate problems in dealing with the large array of materials required for adequate typological conclusions.) In

its strict

sense,

passivization

is especially

characteristic

of

VSO

languages, as in the examples found in H a r t m a n n (1954); for OV languages the pattern is not well-designed.

Other language types employ other

devices, such as the use of particles in OV Japanese and Homeric Greek.

ambivalent

Clefting is one of the favored topicalization devices in

SVO languages. Yet at present generalizations are hazardous because the nuances conveyed through such devices are delicate and open to different interpretations, and because of the scanty information t h a t has been assembled.

I am

primarily

concerned

with

proposing

that

textual

patterning is secondary to that of syntax. 7. S e m a n t i c a n d p r a g m a t i c t y p o l o g y If textual typology is in its infancy, semantic and pragmatic typologies have scarcely been approached.

No volume is dedicated to them in the

Stanford series. Semantic features may well have been noted, as have been pragmatic features.

But there has been little treatment of them in the

formation of patterns.

Somewhat unexpectedly the most useful attention

may be noted in historical linguistics.

Traugott has provided impressive

evidence for distinctions between propositional and interpersonal sets, in accounting for semantic change; similarly, for mental verbs, e.g., believe, know in change to speech act verbs, e.g., promise, as well as English (1982, Pavia paper 1985u). semantic

typology

will

sort

out

additional

report, these in Japanese Further concern with pertinent

configurations, much as phonological typology has done.

features

and

At present we

can best point to the promise of such study, also in pragmatics. 8. T y p o l o g y b a s e d o n t h e t r e a t m e n t of t h e n o u n as p r i m a r y ; t h e fallacy of t a k i n g t r a d i t i o n a l logic as basis The major example of the use of logical formulations is in Vennemann's Principle of Natural Serialization, with its reliance on Operator

and

Operand. In his VX languages these have the order O p e r a n d / Operator, in XV languages the reverse order.

The V in this set is operator; in the

adjective noun set or other modifier noun constructions, the noun is

14

W. P. LEHMANN

operator.

Whatever the history of this conception, it may be compared

with the approach of logicians to language, for example Richard Montague. For him there are two primes, t = clause, e = noun. Further entities, like intransitive verbs are combinations of these; the intransitive verb is represented as t / e . Only in later derivations are transitive verbs generated. The procedure is based on the traditional use in logic of nominal sen­ tences, such as: Socrates verbs, only the copula.

is a man.

In such formulations there are no

T h a t is to say, the noun is the basic entity.

This

analysis of communication is clearly inadequate, as advances in modal logic have made clear in the last century. Maintained in typology, it also leads to inadequate formulations.

In one of its most serious weaknesses, it

neglects verbal modification, as in interrogation, negation, aspect marking, and so on.

For proper treatment of these, language must be approached

directly, rather t h a n as a variant of a logical system. Such an approach requires treating government and agreement as distinct devices. Constructions in which either is involved cannot therefore be

associated

under

the set

Operator/

Operand.

Only by

dealing

separately with constructions resulting from government and agreement will typological patterning be clarified. 9. P r o p o s e d t a s k s for t y p o l o g i c a l s t u d y While the major goal of typology is determination of the possible linguistic structures, it also has other tasks, as Hjelmslev indicated. These can only be mentioned briefly. 9 . 1 . Historical t y p o l o g y Among these tasks is the understanding of linguistic change, noted also by Hjelmslev.

There has been a good amount of attention to this task, and

also considerable misunderstanding, leading even to scepticism of the value of

is

the

identification of the forces at work in the development of languages.

typology

in

historical

study.

Among

major

problems

We

may note briefly two major forces in change: (1) structurally regulated change, often referred to with Sapir's term 'drift'; (2) externally induced change, as in language areas.

Since both forces are often at work at the

same time, exploration of change is highly difficult.

Determining which

force is effective in a given situation must be made through knowledge of

15

PRIMES

social conditions. We may lack adequate information to make a definitive judgement in a given situation, especially in dealing with proto-languages. Proper procedure requires identification of characteristic social situations and observation of the languages used in them. In establishing proper procedures, we must seek social situations in which languages have remained undisturbed by outside influences for lengihy periods. An excellent example is Japanese, used in a self-isolated society for a millennium. It is relatively consistent, with the expression for the middle the only major unexpected construction.

Exploration of other

isolated languages would be of great benefit, especially since few such situations can be identified. Conclusions

can

also be

drawn

from

the

major

languages

of

civilization, which by definition have been open to external influences. All of these are SVO in type, including the major Indo-European languages spoken by more than a hundred million speakers, P u t o n g h u a Chinese, and Modern Standard Arabic. patterning.

Pragmatic reasons have been proposed for the

Speakers using languages non-native for them apparently find

a distinctly different category in the verb useful for distinguishing two nouns in the roles of subject and object. Yet in proposing such conclusions we are dealing with a broad brush. We assume t h a t languages are of one type.

The admirable work done on

Russian should dispel any such notion.

Treated cogently by Kijsper

(1985), these investigations have made it clear t h a t the written language is V O , in contrast with spontaneous speech, which is OV. basic

patterning

is

reflected

also

in

other

The contrast in

characteristics,

such

as

intonation, correlations which require extensive consideration. Attention to such matters will amplify our understanding of language as well as extend benefits from typological study.

Even now the Russian data should make

clear the complexity of a language in its historical development. The results of dedicated study are demonstrated in an excellent paper of Subbarao's (1985u). By in-depth study of reflexives and intensifiers in the languages of the sub-continent with their many shared characteristics across language family boundaries, Subbarao determined t h a t these two patterns permit demarcation of 'Indo-Aryan languages from the other

16 language

W. P. LEHMANN families'.

His impressive

study

gives hope t h a t

informed

attention to findings from typological study will clarify many problems in linguistic research.

One of his conclusions based on many studies parallels

those of Hjelmslev: "Typological studies have important implications for the understanding of languages with 'mixed or contrasting structures' and these studies help us in predicting the direction of change that is likely to take place" (1985u:17). 9.2. E x p l o r a t i o n of ' a b s o l u t e u n i v e r s a i s ' Typological study directed at determination of greater specification of principles underlying language structure is being widely pursued, as by Hawkins (1983).

Scrutinizing typological characteristics in this way,

Hawkins proposes specific principles, making great use of serialization observations. Theory

(see

He further now

associates these under the

Nichols

1986).

While

Head-Dependent

admirable

in

aim,

the

generalizations need further scrutiny. Subbarao on the one hand furnishes evidence

against

the Accessibility Hierarchy

of Keenan

and

Comrie

(1985u:9). And Dryer argues in favor of his Branching Direction Theory, while admitting even its inadequacies (1985u).

Before ascribing greater

certainty to specific generalizations, we must extend our exploration of languages and compare findings from other investigation of language. 9 . 3 . L a n g u a g e t y p o l o g y a n d l a n g u a g e acquisition Support

for

typological

language acquisition.

generalizations

may

come

from

findings

on

Observations on these are increasingly precise, as in

the work of Eve Clark and Dan Slobin, among others (Slobin 1982).

The

observations have led to the recognition of 'organizing points' in the construction of grammars by children.

Strikingly, these cluster around

functions which verbs fulfill in language.

One such point deals with

transitivity, the interrelationships between representations of action and agents as well as patients.

Another focusses on events, as the results of

actions, expressed by means of aspect/ tense indicators, middles, passives. Still others have to do with location, closely related to possession etc., and expressed through case inflections in languages with such forms, through adpositions in others.

The basic typological structures consist of patterns

for expressing these functions of language. Observations on the acquisition

PRIMES

17

of these patterns in languages of the differing structures promise to be illuminating. In short, typological findings of the last two decades are receiving support from various approaches to the study of language. Some of the findings may still be general. And the languages examined may be only a small number of those in use. But the results indicate t h a t the avenues chosen for investigation are secure, and promise many further insights into language as the primary form of human communication. They also suggest t h a t , as Hjelmslev indicated, typology opens the way for linguistics to "become a science."

Basic Typological Units V. N. Yartseva

Linguists have long been faced with the problem of choosing a principle of describing and classifying the structural properties of languages of the world.

Extreme diversity observed at all levels of the language

structure

been

has

revealed

even

more

clearly

in works

aimed

at

comparative-typological analysis of languages of the world. In this analysis the grammatical structure of the language was always given a prominent place, which can be explained by the necessity of taking into consideration, in comparative-typological analysis, the specific properties of each linguistic level. It stands to reason t h a t each of them is suitable and necessary for typological analysis.

However, structured levels of the language are more

suitable. The variety and considerable number of elements typical of the vocabulary of any language hamper its systematization.

Though widely

used in searching for language universais, the phonological level is less suitable for typological analysis (due to a relatively small number of its components and to the absence of the independent notional ingredient in them) than morphology and its adjacent fields like morphonology and phonomorphology.

It was not without reason that such 19th century

linguists as A. Schleicher and W. von Humboldt, searching for foundations of the typological classification, used mainly morphologic elements of the language.

Though there exist numerous works on general and specific

questions in grammar,

fundamental

issues remain uncoordinated

and

continue to stir disputes between linguists. Comparison of languages belonging to different morphological types

20

V. N. YARTSEVA

should take account of the structure of the grammatical subsystems.

The

content of the grammatical categories and the semantic links between parts of speech as lexical and grammatical series are organized in language primarily through morphology. with

a

fully

developed

However, if we are to study a language

nominal

paradigm,

we

have

to

take

into

consideration the syntactic level in order to understand the possibility of complete expression not only of such meanings which usually belong to the sphere of morphology, but also of those which a number of languages express by means of different syntactical devices.

The relations between

syntax and morphology, between lexis and grammar, are usually effected through the syntagma which are structural units in syntax, but are not units of the language.

Since the sentence patterns are abstract entities, a

study of the individual features of the languages being compared should proceed from morphology through syntagma to syntax. Of course it seems that syntactic properties of the language might be considered to be most appropriate for typological analysis. opinion is t h a t the abstract

My own

character of syntactic patterns and

universality of such notional categories as, for instance, the

the

subject-

predicate relation, limit the possibility of setting out the particular features of different languages when we submit them to typological comparison. In treating the syntactical level, it is important to bear in mind the following parameters: • the logical and grammatical sentence pattern with its more or less abstract properties (in terms of the subject—predicate relation) • the rhythmical structure of the sentence and its communicative functions • the specific instantiations of the sentence pattern morphologically complete items of the morphological level

by

• lexis in the syntactic pattern, particularly when it serves to resolve ambiguity of syntactic homonyms. Although a linguistic investigation cannot do without a discussion of lexical problems essential for understanding how form and content are to be related, one can hardly adopt the tendency especially prominent in some

21

BASIC TYPOLOGICAL UNITS

contrastive studies to use lexical paraphrase in order to establish the same categories and concepts for all languages of the world,

Synonymous

substitutions of one pattern for another as a means of interpreting and comparing their content can only be made if we are contrasting two languages within limits delineated

by the specific

character of their

grammatical structures; the concept of analytical language structure as opposed to inflectional will have a different interpretation in the light of data from agglutinating languages.

Elements of analytical structure are

distributed in different ways in synthetic, analytical, and incorporating languages. This is essential not only to morphological paradigmaticity but also in adjacent spheres, such as derivation. For registering and describing material in various languages, it is necessary to choose a uniform category which could be called a common basis

for

their

comparison.

The

difficulty

lies

in

preserving

the

individuality of a certain language when describing different languages on the basis of comparison drawn from the general model of formal and functional indicators.

Contrasting and singling out the availability of a

constant property or a number of variable properties are used in the process

of

identifying

interconnected.

language

elements.

These

processes

are

To some degree the contrast is always present when

comparing any two elements. When the Nom./Acc. form Old Norse 'hus' is compared with the Dat. 'húse', these forms are simultaneously contrasted with each other.

However, the contrasting phenomenon possesses some

specific features. I have pointed out elsewhere t h a t contrast of elements of the language system is at the same time a way of connecting them (see Yartseva 1968). This is explained by the fact t h a t usually only elements of a single order are compared which are contrasted by several markers, but can be united in a single class by any one marker.

It is meaningless to

compare any phenomena using the formula: "nothing is similar in them." The aim of comparison is to reveal the number of markers among the objects compared.

Therefore revealing the objectively existing contrasts

between the language elements requires a preliminary analysis of properties and markers of these elements.

When analyzing the many constructions

which emerge as a result of the combination of rows of different semantic

22

V. N. YARTSEVA

content, the possibility of distinguishing among them depends on: (a) the coordination of these rows, and (b) the combination of modes.

Adverbial

complements in the predicate can demonstrate the first tenet. The nature of the main member of such syntagma either contributes to or, limits the opportunities of their extension.

An adverbial complement based on a

noun of verbal action (infinitive, gerund, supine, and the like) can be expanded to a considerable extent (cf. infinitive constructions in all the Indo-European

languages,

gerund

constructions

in

Modern

constructions with so-called infinitive in Old Irish, etc.).

English,

By contrast,

adverbial components with the adverb are usually limited to structures with homogeneous parts.

There is another marker of complex structures,

in the form of special auxiliary elements (connective words, so-called conjunctions) in many languages which indicate the forms of connection between different parts of the sentence. Methods of scientific description are determined by two conditions: (1) the aim of description (the task) and (2) the specific features of the material to be described.

The choice of one and the same notion for

comparing individual parts and details of the grammatical structure of a language is a very difficult task because of the diversity of structural units. For

instance,

the

category

of number,

in its content,

applies

very

differently to nouns or pronouns. Countability of the objects of the outer world

is not

confined

to

"one"

or

"plurality."

Lexicogrammatical

groupings of nouns includes such notions as names of substances (gold, iron), qualities and properties (blackness, durability, attentiveness), and so on. Quantitative measurements of linguistic phenomena are generally thought of as determining the frequency of one linguistic form or another. However,

the

notion

of

"quantity"

in language

cannot

be

reduced

exclusively to the question of frequency, despite the importance of this aspect of linguistic phenomena.

The problem cannot be adequately dealt

with on a purely formal plane, but only through mutual comparison of grammatical constructions with reference to their content—and one can also take up the matter as a correlative analysis of the plane of expression and the plane of content.

BASIC TYPOLOGICAL UNITS

23

Grammatical categories which characterize the parts of speech, are expressed in paradigms. In the instances when the root and the stem in different parts of speech coincide in their phonetic form, the paradigm may prove to be the only distinguishing feature of the parts of speech. The world's languages show a very uneven distribution of the paradigms among the parts of speech. For instance, in Modern English, not only grammatical words but many other lexico-grammatical classes as well do not have paradigms. The verb alone possesses a well-developed paradigm, its peculiar trait being a combination of both analytic and synthetic forms. The specific feature of the analytic form is t h a t it constitutes one unit yet consists of forms of words belonging to different classes. Thus, for example, the English verb form 'has been doing' is a unit and belongs to the class of forms recognized by grammarians as Present Perfect Continuous. At the same time, its constituent parts belong to different classes as, for example, 'has' belongs to the class 'has, have, had', 'been' in 'be, was, been', and 'doing' in the set 'do, does, did, done, doing'. Among linguists, works on the analytical structure of language in the late 19th and early 20th centuries agree that, historically, analytical structure succeeded inflectional structure, which consequently reflected progress in language. If we assume t h a t this view reflects a universal law of language development, one is struck first of all by the fact t h a t languages which are in many respects quite different from each other often are all defined as "analytic." English, French, Danish and Welsh are usually considered to be analytic languages, notwithstanding the fact that despite their similarities, these languages reveal such great differences t h a t no one standard of what is an analytical language can emerge from them. In a language of analytical structure where the grammatical (and sometimes lexical) content of the word is realized only in context with the help of adjacent words, and may not be expressed in the form of the word itself, there is a close grammatical connection between connectives and autonomous words as well as between the autonomous words within an analytic structure. Therefore any definition of analytic structure t h a t takes account of contextual links must be sufficiently broad and must be consistent both with the features of positional interchangeability of words and with the links between a

24

V. N. YARTSEVA

connective and an autonomous word, because there are no grounds for considering "grammatical" words non-words. Wide use of syntactic (auxiliary) words seems to be a characteristic feature

of

analytic

paradigmatics

languages.

gives

rise

to

Lack

of

inflectional

lexico-syntactical

devices

indicators for

in

rendering

grammatical notions, but such procedures are not confined to analytic languages.

The use of auxiliary verbs is common in the

typically

agglutinative language Turkish. The analysis of language facts implies t h a t one is comparing them for the purpose of identification or for determining the sum of their distinctive features.

If we assume t h a t general laws

regulating the frame of linguistic systems can operate synchronically as well as diachronically, we must decide what is more important in the historical development of languages: the features by which languages differ one from

another

or

a feature

(property)

shared

by

all

languages.

Lexicalization of syntagmas which can result in idiomatic collocations and sometimes in the emergence of complex words, on the one hand, and of grammaticalization of certain types of word groups, on the other, can be considered a general law or language universal.

Yet, for the purpose of

identifying any one language, this universal is of little value. It is probably necessary to discover not one but a set of interrelated features which may constitute typological grounds for distinguishing one language from others. In his well-known paper on the structure of the English language, Lehmann (1981) maintains the thesis t h a t such traits as the existence of auxiliaries, the use of reflexive pronouns, the presence of subjects and objects in most sentences, the frequency of passives, and so on, depend on the SVO language type.

The question arises whether only SVO languages

have these specific features and, what is perhaps more important, whether all SVO languages are typologically similar and equal in their grammatical properties. Sometimes it is possible to transfer conclusions obtained in the synchronic investigation of languages to the diachronic plane. However, a synchronic

particular

rule holds good for

diachronic

investigation

of

languages if we bear in mind one important restriction: the diachronic explanation of particular language features is more difficult just because one and the same trait can have different values in language structure,

25

BASIC TYPOLOGICAL UNITS owing to different

historical realization of general tendencies and

coexistence with other features characterizing a given language. more important for the comparative-typological

its

W h a t is

analysis is a kind of

connection between a certain notion and other notions relative to it and embodied in the given language, rather than the very fact that,

for

instance, the notion of tense can be expressed by this or t h a t means in the given language. analysis

with

Therefore, it is very important to combine typological the

semantic-functional

approach

to

phenomena

of

a

language. At the level of morphemic analysis, certain elements of the word are singled out in conformity with their functions.

T h a t is why one and

the same sound complex can play the role of root morpheme in the contrasting word series and be the central part of a grammatical category for a paradigm. The method of describing of the grammatical properties of languages suggested here does not disclaim the traditional concept of grammar as a science (categories of syntax and morphology, their division, parts of speech and meanings related to them).

These traditional notions can be

used in the "boundaries" of the description of one language, and works based on them can yield material for comparing languages according to the analytic scheme suggested here. conditioned

by

the

fact

that,

The introduction of the scheme is according

to

their

distribution,

the

grammaticized meanings themselves are differently realized in various parts of the grammatical systems in various languages of the world.

Moreover,

many meanings expressed in one language at the grammatical level are conveyed in other languages by lexical means, and therefore cannot always be taken into consideration.

Their comparison on a purely notional basis

(cf. the problem of universais) is fraught with the danger of depersonalizing languages, since the description of each language should be secured by methods appropriate for setting forth its peculiarities.

It should be noted

t h a t structural and semantic descriptions may be given not only separately, but also stage by stage (for instance, an analysis of word structure can be done with regard to its meaning, the structure of the given type of paradigm, and its semantics).

Yet, as an illustration of the very scheme

for describing multistructural languages of the world, the sequence of analysis suggested here seems to be the most expedient.

Areal Phonetic Typology in Time: North and East Asia Robert Austerlitz

In an earlier paper (1970) I tried to show that, on the plane of grammar we may reckon with an agglutinative belt in northern Eurasia, stretching from the bend of the Volga eastward through Siberia as far as the Lena and then southward in easternmost Asia, as far as Japan.

Grosso

modo, the belt consists of most of the Uralic and "Altaic" languages and of Yukagir, Nivx (Gilyak), Korean and Japanese.

Many of the languages in

this belt are situated near or next to others, which are not in the belt. These are: (1) the Eniseian languages (i.e., Ket, formerly also called EniseiOstyak, and its extinct congeners), (2) the Luorawetlan languages (i. ., Chukchi, Koryak

[Nymylan], and Itelmen

[formerly

Kamchadal])~two

small families in central and northeasternmost Siberia respectively--and (3) Ainu, in the southeast.

For reasons which issued from both areal and

purely typological considerations it was suggested in the 1970 paper that agglutination is an innovation in this part of northern Asia. In the present paper I would like to explore the possibility of recapturing an earlier stage of the phonetic and phonological map of roughly the same area: northern Asia east of the River Ob.

The ultimate

and ideal purpose of this enterprise would be to gain an insight into areal dynamics or, more explicitly, to search for foci from which specific and identifiable sound features have spread out in the past. Even if this ideal cannot be realized it is hoped that the adduced data and interpretation

ROBERT AUSTERLITZ

28

will lead to a fresh way of thinking about the area and about the subject. 1 W h a t can we learn from this? (1) Something about how

Sprachbünde

come into being, how they cohere, and (perhaps) how they disintegrate. (2) Something about how specific sets of features identified with this area, e.g., the distinction between

a velar and

a post-velar series or

the

correlations of palatalization or vowel harmony may be connected with each other historically.

(3) Perhaps something about the movements of

speaking populations who are, after all, the bearers of these features. W h a t are the difficulties involved?

(1) The precise determination of

a point in time for the reconstructed picture. (2) The geographical location of a given focus: how are we to know t h a t a certain feature or cluster of features emanated from a specific point on the map? The ultimate answers to

these

questions—like

the

ultimate

answers

to

other

fundamental

questions about language, such as meaning or linguistic change—lie beyond the pale of linguistics. (3) It is not taxing to the imagination to assume, as I shall, below, t h a t both related and unrelated languages in a large area can share phonological features of subsystems. agree with the

procedure

followed

It is more demanding to

below, namely,

that

phonological

systems from different epochs (some attested and others reconstructed) are comparable.

We will work with information from the earliest available

periods, whether reconstructed (through comparison or philologically) or attested, for better or for worse. Since the question of how linguistic features migrate from one point to another in such a vast area as t h a t considered here is inevitably asked, I will propose

at this point t h a t

it is along trajectories

dictated

by

topological features, principally littoral and riparian (coastal and along rivers). Such migrations are seasonal—they depend on climatic conditions— and

accordion-like:

expansion

is followed

by

resolidification.

The

movements are triggered primarily by economic factors: nomadism (the

Thanks are due to Daniel Abondolo, Roy Andrew Miller, S. Robert Ramsey, Jr., S. A. Starostin, and Alan Timberlake, who, at one or another stage in the development of this paper, have provided data or helped in the interpretation of data. Detailed phonetic information is from personal experience only in the case of Ainu, Japanese, Nivx, Orok, and Yakut.

29

AREAL PHONETIC TYPOLOGY IN TIME

search for food), spouse-seeking, slave-seeking. They may also be triggered by cataclysmic events such as sudden and extreme changes in living conditions (climate, food supply and pestilences, affecting both people and animals). We are dealing with very small migrant societies, perpetually in search of food under inclement conditions, governed by codified societal conventions (kinship, family, inheritance, religion), periodically but not always in contact with other societies (barter, warfare, wife-raiding, slaveraiding).

We

should

also

reckon

with

a

considerable

amount

of

bilingualism and even multilingualism. The design thus proposed is in very bold strokes and, in a manner of speaking, experimental.

Whether, in the last analysis, it buttresses or

undermines hypotheses of the Nostratic type remains to be seen. Special attention will be paid to the following themes: (1) Coronal (Russian ) spirants and affricates and features associated with these, such as (a) the hissing : hushing opposition s : š; (b) palatalization, s : á; () affrication, s : c, š : č; (d) the presence or absence of s a n d / o r the s > h change; (2) the relationship between velar  and post-velar (so-called uvular) q and other obstruents at these two points of articulation; (3) the number of series of stops in a given language; (4) vowel systems, vowel harmony, vowel rotation. The languages and language families examined are: (1) Ainu, (2) Middle Korean, (3) Nivx (Gilyak), (4) Proto-Tungus, (5) Proto- or Old Japanese, (6) Chukchi and Koryak, (7) Itelmen (Kamchadal), (8) Yukagir, (9) Yakut, Samoyed.

(10)

Orxon-Turkic,

(11)

Proto-Eniseian,

and

(12) Proto-

Older stages of specific languages (2, 5, 10) and reconstructed

states of families or sub-families (4, 11, 12) rather than their modern descendants, are considered because working with earlier stages renders the inquiry more challenging. results is therefore

The skewed perspective in time which thus

intentional: it should

encourage

alertness

in

the

procedure and lend a strong historical bias to the inquiry from the very beginning. Two sets of facts can be summarily discussed and dismissed at this point.

All

of

the

languages

discussed

have

a

w

: j

opposition.

Theoretically, these semivowels are ultimately reducible to one feature,

30

ROBERT AUSTERLITZ

semi-vocalicity: [u/ ' : [i/ ], governed by position.

Thus, w and j would

need to be specified only when they are adjacent to each other. question is of no further interest here. tabulated nor discussed.

The

These semi-vowels will be neither

The l : r opposition, its weakness (as one goes

from West to East) and its absence, will be given short shrift here, unless l or r require comment. (In Chukchi and Nivx, r forms part of the obstruent system.) Vowel systems will be referred to as as consisting of (a) five units, (b) six units, or (c) eight units.

See Table 1. The letter y stands for a high,

back, unrounded vowel (as in Turkish dotless i). (a)

(b)

(c)

Five vowels

Six vowels

Eight vowels

Table 1: "Vertical" vowel harmony refers to the Chukchi, Tungus, Nivx type: i ~

e, y ~ a, u ~ o.

"Horizontal" vowel harmony refers to the Turkish

(or Finnish) front- vs.-back type. 1. T h e S y s t e m s 1.

Proto-Ainu, based on data from the Hokkaido and Saxalin dialects.

Initial h seems to be old.

In the Saxalin dialect, medial and final h

corresponds to medial and final stops in the other dialect.

Initial r is an

anomaly in this area of the world.

Five vowels

Table 2: 2.

Middle (or Late Middle) Korean,

ca. 1443-1600, based

on

AREAL PHONETIC TYPOLOGY IN TIME

31

philological evidence. The so-called "strong" series (pp-, etc.) is from initial clusters. The aspirated series is thought to be from another kind of cluster (*pk-?). The exact quality of h is not clear: [x] or perhaps pharyngal [h]? The status of ζ is unresolved. The seven-vowel system is a compromise between imputed phonetic reality and phonological accommodation; note the absence, in the table, of a pure mid-front vowel. There are vestiges of vowel harmony, now obscured by an earlier vowel rotation.

Table 3: 3. Nivx (Gilyak), Saxalin dialect, 20th century. aspirated initially and the b-series unaspirated.

The p-series is

Six vowels

Table 4: The p-series alternates with the /-series and the b-series with the v-series so systematically t h a t the proto-system requires no fricatives (Table 5(a)). There are indications that the dentals (t, d) and palatals (č, ¢) may once have been in complemetary distribution. If t h a t is so, and if the velars were in complementary distribution with the post-velars (q, etc.), then a simple tripartite proto-system issues, as in Table 5(b). Note, however, t h a t two series of stops remain. Vowel harmony: vertical.

32

ROBERT AUSTERLITZ

(a)

(b) Table 5:

4. Baikal.

Proto-Tungus, perhaps 12th century, perhaps north of Lake After Benzing.

alternated with

an

The status of *k is unclear.

*x but the modern

It may have

Tungus languages require

hypothetical *x (or, at any rate, one fricative in addition to *s). opposition in the stops cannot be further

a

The voice

reduced or eliminated.

In

addition to the six proto-vowels there were two more, one high and one low; these participated in vowel harmony, which was vertical.

There was

vowel length, presumably.

Six plus 2 vowels

Table 6:

5. evidence.

Old (or Proto-) Japanese, ca. 750-850, based on philological From Miller and Syromiatnikov, with free interpretation.

The

voiced series needs to be reconstructed for medial position only; it may have consisted of prenasalized obstruents or there may have been nasalized vowels in the language which simply voiced obstruents in their vicinity. Whether *p was already a bilabial fricative in the ninth century (now it is h initially) is unimportant. The eight vowels have been assigned phonetic values (as a departure from tradition) so as to render them a little more realistic.

There were restrictions in vowel occurrence

(called

vowel

harmony by some); they were neither of the vertical nor of the horizontal type (see Mathias 1974). 6.

Chukchi, 20th century.

stops here.

There is no trace of a second series of

Note the absence of s and the presence of r, as if it were a

AREAL PHONETIC TYPOLOGY IN TIME

33

Table 7:

surrogate of a more likely fricative. A glottal stop is reported, but its quality is reported as pharyngal. This has prompted the pharyngal symbol in the table. The vowel t functions as the high partner of a and as the low partner of i in vowel harmony. Functionally, therefore, there are 3 X 2 vowels; in fact there are only five. The Koryak system is similar in substance and identical in system. Six minus one vowels

Table 8:

7. Itelmen, 20th century, uncritically copied from Soviet sources.

In

addition to the glottal stop, there may be a glottalized series of stops. (plus glottalization?) Six vowels

Table 9:

8. Yukagir, T u n d r a dialect, 20th century. According to some sources, q does duty for [q], for [x], and perhaps also for [γ] (post-velar voiced fricative or approximant). This amounts to saying t h a t the status of ϋ is not clear. Vowel length has been reported but the evidence is not convincing. In addition to a five-vowel system, a sixth vowel (a front mid rounded H ö H ) has been reported. 9. Yakut, 19th and 20th centuries. There is a voice correlation in the stops in all positions except the labial, as is to be expected in a Turkic

34

ROBERT AUSTERLITZ

Five vowels (plus one?) Vowel length?

Table 10:

language,  is [] or [χ] by position and [s] and [h] alternate, by position. The status of is uncertain, j is a nasalized [j] and probably identical with n.

Eight vowels; horizontal vowel harmony

Table 11:

10. Orxon-Turkic, ca. 680-740, from inscriptions found near the Orkhon and Selengga Rivers, Mongolia. Among the stops, there is a voice distinction only in the dental area.

Beside d, there is also a

beside s

there is also a z. Furthermore, there is a hissing : hushing opposition, s : š.

This, then, is the first language in the series which has a developed

system of coronal fricatives. 11.

Proto-Eniseian,

after

geographical location indeterminate.

Starostin

(1982),

time

depth

and

(The only living language in this

family, Ket, is spoken in the area of Turukhansk.)

Starostin's very careful

reconstruction sets up 28 proto-consonants and nine proto-vowels.

The

following points are surprising in the reconstruction, bearing in mind the bias espoused in this paper: (1) reconstructed voice; (2) no palatalized s, while palatalization is otherwise strongly represented (in the r-row, the /-row, and the nasals); (3) three hissing coronals (c, dz, s), augmented by

AREAL PHONETIC TYPOLOGY IN TIME

35

Eight vowels, perhaps also e; horizontal vowel harmony

Table 12:

r 2 . This complex reconstructed system may be simply a by-product of the process of reconstruction itself—designed to account for existing reflexes found in daughter-languages--or it may suggest t h a t Proto-Eniseian was not spoken in northern Asia. The two features of the reconstruction which are reconcilable with the area as a whole are the presence of palatalization and the opposition in the  : q sphere.

Table 13:

12.

Proto-Samoyed, more or less identical with proto-Uralic.

Time

depth: ±1000 B.C. A fully developed system of palatalization, presence of a hushing series, but no post-velars.

Beside the reconstructed six-vowel

system, as indicated, one could possibly reckon with a five-vowel system (i e   u), but in either case a high front vowel (ü) strongly suggests itself. The three reconstructed approximants (two deltas and a gamma)—which, by the way, are reminiscent of the voiced fricatives in Orxon-Turkic-- are by-products of the exigencies of reconstruction.

36

ROBERT AUSTERLITZ

Table 14:

2. S u b s y s t e m s 1. The  : q opposition is thoroughly grounded in Chukchi and Koryak (and probably Itelmen) and seemingly in Eniseian, i.e., in northeasternmost and in central Siberia. In modern Ket (Eniseian), as well as in Yukagir, q alternates with a velar or a post-velar fricative (voiced or unvoiced).

In

modern Nenec (Samoyed),  also alternates with a corresponding fricative. (In Nivx each stop alternates with the fricative which corresponds to it.) We thus have one area in which the  : q opposition is firmly established and so to speak inert and another area in which a velar or post-velar stop alternates with corresponding fricatives. In

Nivx,

q and

 are

in

a state

of vestigial

complementary

distribution: q in the environment of a and  and  near the other vowels. This would suggest t h a t  and q split from a proto-*K" and that the split was triggered by the vocalic environment (low-back vs. other). This looks very much like horizontal vowel harmony and Orok (Ulta), a SouthTungus language has this distribution to this day sub-phonemically.

But

both Nivx and Tungus have vertical and not horizontal vowel harmony. We may therefore wish to postulate vowel rotation, which left its traces in the distribution of  and q, In Luorawetlan there are no such traces; these languages also have vertical vowel harmony. Dare we assume t h a t   : q split, connected with horizontal vowel harmony, spread from the northeast toward the south and the center of Siberia, and t h a t vowel rotation followed, erasing the traces which originally triggered the split?

By the

same token, dare we assume a spread of the spirantization of velars (and

37

AREAL PHONETIC TYPOLOGY IN TIME

post-velars) as in Nenec, Ket, and Yukagir? Where would the point of origin of such a spread be located? 2. The system of coronals differs from area to area in terms of the number

and

quality

of

the

units

involved.

Proto-Samoyed

had

palatalization in the sibilants, laterals, nasals, and perhaps approximants. It also had a hissing : hushing opposition.

Furthermore, it had

two

affricates, ć and č, which corresponded to three fricatives, s : ś : š. (What is written ć and ś may represent hushing

palatalized sounds,

and

,)

Furthermore, there seems to have been a statistical tendency for the spirants

to occur

initially

and

for

the

affricates

to occur

medially

(although, intuitively, the opposite distribution would be expected).

Are

we dealing with spirantization again, as in the case of the velars above? 3. Palatalization is patent in Samoyed and very strongly operative in Eniseian.

Tailleur (1960) seems to have evidence that Yukagir s was also

palatalized presumably without an unpalatalized partner. sources write š and not s. This hushing fricative

palatalized θ. A system with only a palatalized hushing Yakut initial s is palatalized to this day. palatalized.

(Older Yukagir

may have been a is improbable.)

Nivx s and ζ are always

Dare we assume that a wave of palatalization spread from

west to east? Was such palatalization connected with horizontal (i.e., front : back) vowel harmony?

Or was it perhaps connected with vertical ( i.e.,

high : low) vowel harmony?

If, for the sake of the paradox, we assume

t h a t vertical vowel harmony originally spread from the west to the east and horizontal vowel harmony spread from the east to the west, then we are postulating a situation which completely contradicts today's

situation,

when the east is the home of vertical harmony and the west the home of horizontal harmony.

Could be this the result of widespread

vowel

rotation? 4. The three obstruent series in Korean can be explained historically (see above). The Japanese voice distinction is, to say the least, weak.

This

leaves Nivx as the only non-Turkic, non-Tungus language in the eastern half of our area which has unequivocally more than one series of stops. Can the two series be reduced to one, on the Korean model? It is tempting to speculate t h a t Nivx aspirated pre-vocalic ρ is from a

38

ROBERT AUSTERLITZ

cluster such as *ps- or *pp- or from *ə- and that unaspirated b ( = [p]) is from plain *p, but there is little lexical support for such speculation. Nevertheless, the Nivx initial clusters px-, kf-, etc., are indeed from

*pk-,

*kp-, etc.; grammatical and lexical evidence for this assumption is ample (see Austerlitz 1982, 1984). 5. It is striking that Chukchi and Koryak have no sibilant at all. If we consider that Nivx s and z are from  and ¢, then Nivx is also an s-Less language. The languages in our area display, then, complete s-Lessness, an alternation between s and h (Yakut), an inert, un-aLternating h (Nivx, Eniseian), and various degrees of sibilant subsystems ranging from Ainu (only s) to Samoyedic and Eniseian. It is difficult to distill a prehistorical map from this picture, but the s : s distinction is incontrovertibly traceable to the West and the South (proto-Samoyed, Orxon-Turkic). 6.

The voice correlation in the sibilants, as it is found in Orxon-

Turkic is also disturbing because the voice correlation in the stops

of

Orxon-Turkic is weak. The situation in middle Korean is similar: s : z but no voice correlation in the stops. discussed.)

(Japanese and Nivx have already been

Dare we assume an archaic situation in which a series of

approximants

which

were

inherently

(but

not

distinctively)

voiced

parallelled a series of stops which was inherently voiceless (more or less *p *t *k : *w *ε * )?

The affricates would then be parallelled by the

corresponding sibilants and shibilants.

(This image is borrowed from the

reconstructed picture of Uralic.) 7. Finally, the coronal nasals deserve a word. The η : ň distinction (where ň is the palatal nasal or a palatalized dental nasal) seems to be firmly rooted everywhere except in proto-Tungus and in the southeast (Korean, Ainu, Japanese), where it is sub-phonemic.

Is this η : ň

distinction perhaps a vestige of palatalization at a very early time (along with s : è), perhaps one of the prime movers of palatalization in the area in general? (In Nivx ň tends to occur in roots and n in suffixes.) 3. Areas, Foci It would be presumptuous to try to assign focal areas, that is, areas from which particular features radiated, to specific points on the map. The best we can do is to identify certain features with general areas.

39

AREAL PHONETIC TYPOLOGY IN TIME

Thus, the Baikal area in the South seems to be the home of voice (Yakut and proto-Tungus) or of incipient voice (Orxon Turkic). voice in Japanese, a thorough voice correlation

Incipient

as reconstructed

for

Eniseian by Starostin, and the fortis : Lenis correlation in Nivx will require different explanations. The Baikal area is also the place where we saw the occurrence

of

š

in

addition

to

s

in older

Turkic

and

where

the

reconstruction of proto-Tungus requires—it seems—a (non-velar?) fricative in addition to s.

This also seems to be the area in which both types of

vowel harmony are represented: vertical in Tungus and horizontal in Turkic.

This situation is hardly an incipient one; rather it must be the

product of something very old.

(See the speculations on vowel harmony,

the  : q opposition, and palatalization above.) Another area, which could be called North Pacific is the home of a patent  : q distinction, of s-lessness (except for Itelmen) and a paucity of non-stop obstruents in general (again except for Itelmen), of vertical vowel harmony

and

a strong

n

:

ñ

opposition.

The

typical

(and

only)

representatives are Chukchi and Koryak. A third area, perhaps historically South of the Sayan mountains, would be represented by Samoyedic and Eniseian in their earlier forms. Typical of this area is palatalization, especially as revealed in the pairs s : ś and l : ľ and the highly developed affricate: fricative opposition, ś-š-s : ć-c.

This

is

also

the

area

with

three

degrees

of

vowel

height

( e æ :   a), as against the other areas in which two vowel heights are found (unless a given language has a five-vowel system). Finally, a fourth area would be the Okhotsk area in the Southeast, represented by Ainu, (Middle) Korean, and Japanese and typified

by

economy in its obstruent system ( p t k in Japanese and p t č k elsewhere) and negatively by the absence of an n : ñ opposition, the absence of coronal fricatives other than s.

This is also the area in which h occurs in

two languages (Korean, Ainu).

The vowel systems in this area, as far as

they can be reconstructed, are different from those of the other areas. Since the use of geographical nomenclature may be too suggestive, the four areas are plotted under cardinal points in Table 15 (V=vowel). The strongest polarity seems to be found between SOUTHWEST

and

40

ROBERT AUSTERLITZ

N O R T H E A S T : note the difference in vowel height and in coronals.

The

SOUTH gives the appearance of a wedge, carrying the voice opposition, a pair of vowels (front-rounded in the case of Turkic) in addition to the six vowels in the SOUTHWEST (Samoyed) and the N O R T H E A S T , and a true hushing fricative š. The SOUTHEAST can best be defined negatively, as above, and by the presence of h.

Table 15:

Table 16 is a replica of Table 15, with the languages identified by name. (Itelmen is a perennial problem and will have to be accommodated more accurately when newer and more precise materials become available.) Table 16 also contains two mystery languages, χ and y.

The former is

Yukagir, which fits none of the four cardinal points because it either lacks a given feature

or contains it only in a rudimentary

information is incomplete).

way

(or

our

Yukagir, as shown, is therefore best assigned

an intermediate position between SOUTHWEST and N O R T H E A S T , with adumbrations toward the SOUTHEAST (economy in the system; vowel system?).

The other mystery language, y, is Nivx, with its opposition in

the obstruent system, its putative s-lessness, its six-vowel system

and

vertical harmony, and its h. It seems to fit nowhere and therefore occupies an indeterminate position in the scheme. Finally, a brief look at adjacent phonetic-typological areas is in order, if only to sharpen the contours of what has been said above.

The

larger areas which skirt our own area are occupied by (1) Slavic and

AREAL PHONETIC TYPOLOGY IN TIME

41

Table 16:

Iranian, (2) Northwest and Northeast Caucasian, (3) Tibeto-Burman, (4) Sinitic, and (5) West-Indonesian (Formosan) languages in the South and (6) Eskaleut and (7) Athapaskan languages in the East. view will be restricted to stop systems.

This cursory over­

With the significant exception of

Eskaleut, all of the languages in the two peripheral areas cited have at least two series of stops.

The correlations within the stop systems are:

voice, glottalization, aspiration, and pre-nasalization.

Eskaleut, which is

literally adjacent to Chukchi, has only p t č k q, like Chukchi. In our own area, it will be remembered, only Tungus and Nivx had two explicit series of stops (voice, aspiration). Starostin reconstructs voice for proto-Eniseian, perhaps with the arriere-pensée South and West.

that it belongs, historically, further to the

The Yukagir system is reported to have a voice

distinction in some stops but internal evidence suggests t h a t the report may be due to an insufficiently detailed analysis. The Japanese opposition p t k s : b d g z

(medially) is secondary and incomplete and probably

connected with or due to nasal-plus-stop sequences or an earlier series of

42

ROBERT AUSTERLITZ

nasalized vowels. The Korean tripartite system (t : th : tt) is reducible to a single series (see the discussion above).

The Old Turkic system and its

descendants, including Yakut, is defective, with t : d as the only fully established pair distinguished by voice. The great majority of the languages in our own area, then, typically has or had one series of stops, while the adjacent Eskaleut, have more than one series.

languages, except

This sort of contrastive procedure

should be applied to the  : q opposition, to the question of vowel systems, to coronal subsystems, to palatalization, to the nasals—in short, to all the aspects of the problem treated above.

If we implement a little historical

imagination, the conclusion suggests itself t h a t our (northern) area (say, the area with typically only one series of stops) was sealed off from the southern belt for a considerable period, t h a t the eastern area (the New World) is either radically innovating or stubbornly archaic, and Eskaleut

occupies

an enigmatic

position

in the typological

that

spectrum

described. W h a t is now needed, aside from more raw data and more work, is a similar approach to questions of grammar, in all the geographical areas involved.

Lexico-semantic Reconstruction and the Linguistic Paleontology of Culture T. V. Gamkrelidze

The picture of areal affinities drawn from the study of grammatical, phonological, and lexical isoglosses among dialects of a proto-language provides a way of determining how dialectal differentiation of the protolanguage proceeded and hence of establishing what contacts took place among speakers of the dialects.

In other words, purely linguistic facts

make it possible to establish extra-linguistic factors such as historical interactions among speakers of the linguistic systems. Of

particular

value

and

significance

for

linguistically

based

reconstruction of non-linguistic historical and social relations is formal semantic analysis of the dialect lexicons, since they can reflect all essential aspects of the historical existence of their speakers.

The

historical

existence of the speakers includes the ecological environment (fauna, flora, geographical surroundings, climate) and human habitation and migration in the environment, as well as culture in the broadest sense (including both material

and

spiritual

culture).

Semantic reconstruction

of

relevant

lexemes from the daughter languages gives us a general representation (although sometimes only a fragmentary one) of the speakers' historical existence.

In such

reconstructions,

linguistic facts

are

a source of

information on prehistoric culture and its historical evolution. 1 Reconstructing elements of the extra-linguistic world of daughter-

For details, see Gamkrelidze/ Ivanov 1984.

44

T. V. GAMKRELIDZE

language speakers in t u r n gives a clearer picture of the linguistic affinities among the daughter languages and their development in time, i.e., of purely linguistic factors. This is particularly true of the semantic structure of language, which simply cannot be studied in isolation from the external world that is reflected in the content plane of language. reconstruction the proto-meaning often

In

cannot be established

formal without

going beyond strictly linguistic facts and bringing in typological data on the history of the cultural domains involved. For example, for the ProtoIndo-European base Haios-, whose reflexes in the daughter languages mean variously 'copper', 'bronze' and, 'iron', we reconstruct the

prehistoric

meaning 'copper', not 'bronze' and especially not 'iron', since the period of Indo-European linguistic unity was during the Copper Age. The analysis of language as closely linked to its speakers' culture was developed during the early stages of Indo-European studies.

The classical

figures in Indo-European comparative linguistics, such as Jakob Grimm, pointed out the need for joint analysis of language and culture.

Grimm

indicates this need in the following: Sprachforschung, der ich anhänge und von der ich ausgehe, hat mich nie in der Weise befriedigen können, dass ich nicht immer gern von den Wörtern zu den Sachen gelangt wäre; ich wollte nicht bloss Häuser bauen, sondern auch darin wohnen. [The linguistic research I am engaged in which is my point of departure, could never have satisfied me if I had not been able to constantly turn back to the things they denote; I wanted not only to build houses but also to live in them.] This trend produced studies, dating to the middle and especially late 19th century, which have remained significant to the present day.

Such

investigations were often called examples of linguistic paleontology, after Pictet, who analyzed (1859-63) ancient Indo-European culture on the basis of linguistic data. This

trend

in

linguistics

might

better

be

called

'linguistic

paleontology of culture' since its object of investigation is not the protolanguage but the proto-culture of the speakers; what is reconstructed is not so much the language itself as the extra-linguistic world reflected in the linguistic data.

LEXICO-SEMANTIC

45

RECONSTRUCTION

The Saussurean principle of studying language 'in and for itself' became standard in the subsequent decades, causing a decline in languageand-culture studies and narrowing the focus of linguists' attention to the structure of language.

This restriction to structure alone is beginning to

shift, as of the past two decades, to the study of language, viewed as a product of human culture, in close linkage with culture and with the history of its speakers as a frame of reference. This development entails that the study and reconstruction of lexical semantics and the distribution of lexemes among the daughter languages are of particular importance to the structure and history of the protolanguage.

The lexemes themselves, once we have reconstructed

their

original semantics, give direct evidence for the ecological and cultural environments

of

the

daughter

languages,

which

gives

us

a

first

approximation to aspects of material and spiritual culture, and points to changes in their environment brought on by migrations of

daughter

languages. Comparison of formally cognate words from the daughter languages and their attested meanings yields reconstructions of both proto-forms and proto-semantics.

These lexical archetypes may go back to proto-linguistic

times and be posited for the proto-linguistic system; or they may belong to the various chronological layers reflecting dialect groupings within the proto-language. Comparing the reconstructed forms and meanings makes it possible to discern lexico-semantic fields.

These fields define sets of semantically

grouped lexemes which designate extra-linguistic classes such as animals, plants, construction and dwellings, handcraft tools, and others. Establishing such lexico-semantic fields for the proto-language makes possible a conclusion, extremely important for phonetic correspondences among etymologically related words, about the possibility of departures from regular sound correspondences.

Such departures can be observed in

words from the semantic fields of animal terms and body-part terms.

This

can be explained by the frequent deformation of words in these lexical fields due to euphemism

and tabu.

This fact necessitates a special

provision in the Neogrammarian principle of exceptionless sound laws, as

46

T. V. GAMKRELIDZE

can further be illustrated by many examples of individual words from particular semantic groups. The study of semantic fields reveals striking differences in the lexical stability of individual lexemes, depending on whether their primary usage was nominal or verbal.

For example, the ancient nouns referring

to

various kinds of tools undergo frequent semantic shifts in the individual daughter languages as technology and production evolve, which complicates

reconstruction,

while

the

verbs,

with

their

less

often specific

terminological meaning, display greater stability. Proto-linguistic lexemes naturally reflect the general features of the environment,

material

existence, and

culture of the speakers of

the

daughter languages. These same lexemes provide the oldest lexical stratum of a proto-language and its separate branches. A classificatory lexicon of the proto-linguistic semantemes may be set up arranged in individual semantic fields, bringing in whatever

information

about

the

possible

historical denotata of these lexemes is essential to shed light on the ecological environment and culture type of the speakers of the daughter dialects. Such a proto-linguistic lexico-semantic system can be reconstructed through

comparison

Comparing

this

of

system

cognate with

forms

in

the

extra-linguistic

daughter

reality,

we

languages. obtain

an

approximation to a general picture of the economy, material culture, and social organization of the speakers of the proto-language or its dialect groups. The posited general picture can be given historical reality through typological comparison with actual cultures of the past and present. is a typological verification of a reconstructed culture against cultures. material

In verifying side,

plausibility

of

a reconstructed

archeological the

facts

reconstructed

culture, and in particular

which material

demonstrate culture

the

receive

This

attested its

typological particular

significance. A reconstructed proto-linguistic lexeme must be listed in the semantic dictionary together with the mythological and ritual motifs its referents participate in.

This takes us from the words and their referents to the

comparative analysis of mythic and ritual motifs and to the reconstruction

LEXICO-SEMANTIC

47

RECONSTRUCTION

of semantic archetypes for mythic and ritual elements which go back to proto-linguistic times, as well as to the description of their transformations in separate historical traditions. The reconstructions of common mythic and ritual elements is a purely semantic reconstruction which can be independent of the particular words describing these elements in attested historical traditions.

In

practice, however, mythic and ritual elements established for individual daughter traditions and going back to the proto-linguistic models are attested in texts which frequently contain etymologically related words or even whole expressions.

This further testifies to the common origin of

mythic and ritual elements which illustrate the conceptual and spiritual culture of the speakers of a proto-language.

In addition, it enables us to

reconstruct not only words with their meanings but also fragments of proto-text, i.e., sections of text longer t h a n a word.

These fragments

provide examples of word-level syntagmatics, arrived at not by combining words reconstructed in isolation (an example of which is Schleicher's wellknown fable), but by directly reconstructing integral fragments of text on the

strength

of

etymologically

identical

compounds in the daughter traditions.

phrases,

expressions,

and

They enable us to reconstruct in

broad outline the spiritual culture of the carriers of a proto-language and their conceptions of the world around them, their classification of natural phenomena and their conceptions of humans and human society. The comparative semantic analysis of words from separate daughter dialects of a proto-language and text fragments in separate traditions, together

with

the

method

that

may

be

called

daughter 'linguistic

paleontology of culture' makes it possible to reconstruct a semantic lexicon of the proto-language and a picture of the proto-culture of the speakers of the proto-language or its dialect groups.

This points to a possible original

homeland and suggests routes of migration by which speakers of the protolanguage or its dialects may have reached their historical territories.

Universais Specials and Typology V. M. Solntsev

In a previous paper I have pointed out t h a t it is not an immediate task of typology to determine and to formulate

language

universais.

Rather, universais belong to "pretypology" (Solntsev 1978). Some authors, however, believe that the study of universais is, if not the main, then at least one of the principal tasks of typology (Uspenski 1975:11). Universais are, as their name suggests, universal or general properties of human languages, organized, by Greenberg's definition, according to one pattern. I do not use here the term 'universais' to designate statements of a universal nature, even though that usage of this term is widespread. Universals make it possible to bring to light a field of objects (in the broad sense of the term), a certain set of relatively homogeneous objects, and to distinguish this set from other sets of objects belonging to other fields. As regards typology (including linguistic typology), its main task, as I see it, is, first and foremost, to reveal, on the basis of the structural similarity of objects, various subsets within a certain set. Such subsets of structurally similar objects within a set are usually called types. To define them, knowledge of general or universal properties (universais) is of little value.

Thus, the determination of human races cannot be done on the

basis of universal properties inherent in the representatives of mankind which differentiate them from all other living beings. Subsets (i.e., types) may be established only on the basis of specific or special properties of these subsets.

Accordingly, the determination of

universais is important at the pre-typological stage of study, that is, at the

50

V. M. SOLNTSEV

stage of revealing the field of objects (a set of objects) within framework of which types are distinguished.

the

In linguistics, the knowledge

of universais is essential for the study of the properties of human language in general, as represented by a multitude of natural languages. This is, of course,

important

for

the

subsequent

purely

typological

Therefore, the study of universais may be regarded

analysis.

as pre-typology.

Typology as such, the determination of the types of languages, is based, as has been already mentioned, on special properties inherent in subsets. W h a t are these special properties? W h a t is their connection with the universal properties? Special properties inherent in groups (or subsets) of objects, and the general properties of languages, do not form special groupings. belong to sets as their integral parts.

Subsets

Special properties of subsets are

nothing but a form of manifestation or a mode of existence of the general properties (universais). This, as I see it, may be applied to various fields of objects. Thus, if roots, trunks, branches, and crowns are general properties of a set of trees, then in the subsets (specific kinds of tree) these general properties have a specific (i.e., special) manifestation.

The crown may be

coniferous or leaf-bearing, roots may go deep into the ground or creep on the surface, etc. In linguistics, for instance, the bilateral division of units into meaningful units and auxiliaries is, apparently, a general property of all human languages. It is a universal. But the mode of manifestation of this universal differs in different

languages.

In some languages, the

auxiliary elements exist as a developed system of affixes and auxiliary words; in others, mainly as auxiliary words with a limited number of affixes.

In some languages affixes are agglutinative, in others they are

inflectional, and so on. All languages have sets of certain universal features which make them

human

languages and distinguish them

communication.

from

other systems of

But different manifestations of universal properties in

different languages result in the existence of certain phenomena in some languages and their absence in others. Thus, for instance, in the languages where auxiliary elements exist mainly in the form of auxiliary words, and affixes are not highly developed, as a rule, a morphological category of case

UNIVERSALS SPECIALS AND TYPOLOGY

does not exist.

51

In the languages where the verb has no personal forms,

there is no morphological category of person, and so on. The absence of a category in a language does not mean that some universal property is not manifested in t h a t language.

All universal

properties are manifested in all languages. Otherwise they would have lost their universality. If the existence of auxiliary elements and the expression of relations between words are universal properties, then in language where, for example, there are no cases or personal conjugation, universal properties are manifested otherwise:

these

auxiliary elements exist

mainly as auxiliary words, and the relations between words are expressed by auxiliary words and word order. personal

conjugation,

But in languages with cases and

these serve as a form

of manifestation

of

the

universal properties. We

can

manifested

give

more

examples

in special properties.

properties

are

All languages have consonants

of

how

universal

and

vowels. But in different languages the sets of these sounds form different phonological systems.

In all languages, sounds fulfill two very important

functions, the function of differentiation expression of meaning.

of meaning and t h a t of the

In some languages (for instance, Russian), these

two functions are performed by separate sounds (phonemes).

In other

languages they are divided among various sound units: separate sounds (vowels and consonants) and syllables. For example, in Chinese and many other of the so-called isolating languages, the function of differentiation of meaning belongs to sounds, and the function of the expression of meaning, to syllables. Syllables exist in all languages.

This is a universal property of

human language. But the role of syllables in the organization of language systems is different in different languages. In many languages (e.g., IndoEuropean),

syllabification

meaningful

units (morphemes and simple words).

does

not

coincide

with

the

division

In other

into

languages

syllabification coincides with the division into meaningful units.

This

occurs in many languages of East and Southeast Asia. We say that a law of morphological significance of syllabification applies in these languages. Now we may try to define specific or special language properties.

V. M. SOLNTSEV

52

The special properties of languages are specific features of groups (subsets of languages).

They are a form of manifestation of universal properties.

These properties, by analogy with universais, may be called 'specials'. Specials are properties and phenomena which are repeated in a number of languages.

The morphological category of case, for instance, may be

regarded as belonging to the specials, because it is characteristic of one group of languages, but

absent in others.

Specials should

also be

distinguished from the properties inherent in only one given language. For example, every language has its own phonological system with its own original features.

Of course, in all phonological systems one can find both

general properties and special features, but on the whole they are unique. Individual features may be called 'individuals'.

In their turn, they

are a mode of existence of general, as well as of special, properties. Special properties ('specials') in their totality make up a language type. Each language type is a subset of languages having the same set of specials.

In other words, universais (general language properties) are

manifested in a similar way in the subset of languages belonging to one type.

The search for the language type requires bringing to light the

specials repeated in a group of languages. language type may be defined

In an abstract form, the

quite clearly.

But in the

languages

belonging to that type, some specials characteristic of that type may be absent.

To put it differently, some language of a given type may have an

incomplete set of specials characteristic of the type. Some specials may be found in the languages belonging to different types. All this may hamper the distribution of languages according to type and the definition of the type itself.

Yet, the definition of types is quite possible, because specials

and their sets are real. In a sense, specials play, in respect to their subsets, a role similar to the one universais play in respect to the entire set. They make it possible to distinguish subsets in a set. Universais distinguish sets from still larger sets.

Thus, language sets differentiate the totality of natural languages

from other communication systems.

All communication systems may be

treated as a set in respect to which the natural languages, in their totality, act as a subset.

When the universal properties of communication systems

UNIVERSALS SPECIALS AND

TYPOLOGY

53

are distinguished, language universais become, in respect to them, a kind of special. In conclusion I would like to show how one may distinguish subsets of languages on the basis of specials. The establishment of relations between words is a manifestation of one of the most general, universal properties of languages. Apparently, it is one of the most fundamental properties of language systems.

Generally

speaking, only units and relations between them lie at the very basis of the language.

This

fundamental

languages in different ways.

property

manifests

itself

in

different

In reality there are two basically different

methods of establishing relations between words: (a) by means of words themselves, and (b) by means of word order and auxiliary words. Both (a) and (b) reflect specials, each of them revealing a subset of languages. Using these specials it is possible to distinguish two large subsets in the set of the languages of the world: (a) non-isolating languages and (b) isolating languages. These two subsets in their totality exhaust the languages of the world.

They are the largest groups of languages of the world and may be

called macrotypes.

The existence of such macrotypes is supported by a

whole set of specials characteristic of every type. The universal of expression of relations between units is perhaps the most fundamental property of human languages. And t h a t is why specials based on this universal, firstly, can serve as a basis for the first typological division of all languages into two macrotypes, and, secondly, condition the existence of other properties characteristic of these macrotypes. Languages display similarity to each other within each macrotype. This means t h a t between the units of different levels there exist similar relations in each of the macrotypes.

The relation of sound to syllable, of

syllable to morpheme, of morpheme to word, and of word to sentence takes one specific form

in the non-isolating languages and another in the

isolating languages.

In the isolating languages, in contrast to the non­

isolating, the syllable has, as a rule, a fixed composition, and the total number of syllables is strictly limited.

As has been noted previously, in

isolating

languages,

the

coincides

with

in

contrast

morphological

to

articulation

non-isolation,

syllabification

(the

morphological

law

of

V. M. SOLNTSEV

54

significance of syllabification acts in these languages). Further, the borders of the morpheme and of the simple word almost always coincide. There is similarity in the properties of the morpheme and the simple word; the relations between words in a sentence are expressed only by word order and auxiliary words. A sharply different picture is found in the non-isolating languages. Even in those where isolation and analytical structure are prominent because of the loss of morphology (for instance, in the English language), word order does not always play a grammatical role, and at least some grammatical relations are expressed by words themselves (cf. he says,

says

he). Smaller subsets of languages, which may be called types, can be distinguished within the larger sets: those of non-isolating and isolating languages.

Thus,

agglutinative

and

inflectional

distinguished among the non-isolating languages. languages,

there

are

languages

with

languages

may

be

Among the isolating

agglutinative

morphology

(e.g.,

Chinese) and languages with agglutinative-inflectional morphology (e.g., Tibetan). Such singling out of types within macrotypes makes it possible to speak of a "gradual" typology and to maintain t h a t isolation, on the one hand, and agglutination and inflection, on the other, are not phenomena of one rank.

Isolation is broader than agglutination or inflection and may

include both of them in its framework.

Commensurability of Terms A l i c e . Harris

Since most typological studies depend upon primary analyses of individual

languages,

the

typologist

must

take

into

consideration

differences in the use of terms in these descriptions and the effects these differences

may have on the analysis.

In evaluating and

comparing

typological studies, the linguist must be keenly aware of the ways in which various terms are used. important.

At both stages terminology turns out to be

1

Much of the terminology of typology is the same as that used in the description of individual languages. Since one cannot separate these two, I will discuss the use of terms in both areas, but with emphasis on those that play an important role in comparative studies devoted exclusively to typology.

Several of the issues raised here extend beyond terminology to

differences of approach or to problems in the analysis of languages. I have found it difficult to draw a strict distinction between terminology and the issues related to terminological problems. I

will

concentrate

in

this

paper

chiefly

on

problems

of

the

incommensurability of terms in a few areas, dealing in t u r n with a few issues

concerning

morphology,

alignment,

voice

and

changes

in

grammatical relations, and complementation and compound and complex sentence structures.

I am grateful to Joseph Greenberg and the participants in the Moscow Colloquium for comments on an earlier version of this paper or on the presentation based upon it. This material is based upon work supported in part by the National Science Foundation under Grant No. BNS-8419143.

56

ALICE C. HARRIS

At the very outset, we find differences even in the term .

In both English and Russian, typology/

typology/

 means

'classification' in a general sense, and language typologies classify languages according to a variety of parameters. typology

Examples in English of the use of

to refer to classifications of languages in various ways

are

contained in the four volume work published in 1978 by Joseph Greenberg and his co-editors and in the three volumes just published under the editorship of Timothy Shopen; an example in Russian is Sharadzenidze's (1982) 



   

. But when an English speaker thinks of typology, he thinks first of word order typology, in particular Joseph Greenberg's classic (1966) article and the many studies that have been inspired by it, such as two recent books in this area, Hawkins 1983 and Andersen 1983. Note, for example, t h a t in Lehmann's 1978 book entitled Syntactic

Typology, seven out of the

eight chapters deal mainly or exclusively with word order typology.

In

Russian, on the other hand, the unmarked meaning of  seems to be classification according to ergative, active, nominative-accusative, and related types. simply

For example, Klimov and Alekseev's 1980 book is entitled



Languages),





(Typology

of

Caucasian

but deals primarily with ergative, active, and nominative-

accusative types and their correlates.

The ergative/ active/ nominative-

accusative parameter is discussed further below. There is a further difference in the meanings of typology/ . For many linguists, a typology is a classification according to only one parameter.

For example, word order typology deals with classification

according to word order alone.

Many studies have as their goal the

correlation of word order facts; for example, a number of attempts have been made to correlate the order of various constituents of noun phrases with the order of the main constituents of clauses (subject, S, object, O, and verb, V). For many, the meaning of typology relates only to this kind of classification along one parameter, together with correlations along the same parameter. For others, typology/

 has a more extensive meaning, the

correlation of classifications along more than one parameter. For example,

57

COMMENSURABILITY OF TERMS

the monograph by Klimov and Alekseev cited above correlates ergative case marking a n d / o r verb agreement with lexical divisions of verbs into transitive vs.

intransitive and with SOV word order.

Another recent

example of this approach is a paper by A. Jaxontov presented at the 1981 Leningrad conference on typological methods, 2 proposes

correlations

between

In this paper, the author

morphological

type

(inflectional,

agglutinative, isolating) and types of complementation permitted

(finite

clause, infinitive, masdar, etc.). To accommodate this extended meaning of typology/ C.F. Voegelin has used the term whole-language typology),

distinguishing this from subsystem

,

typology (or

whole-system

typology (e.g., Voegelin, et al.

1960). 3 1. M o r p h o l o g y For Edward Sapir and others preceding him, typology was foremost

involved

agglutinative/

with

morphological

classification.

, inflectional/  and analytic/

and terms

, seem

have the same ranges of meanings in English and Russian. synthetic/

first

The

to

Similarly

 seem to

be

comparable terms. Following Sapir's idea of dividing morphological types into several distinct bases of classification, quantifying these indices.

Greenberg 1960 proposes a method of

One base classifies according to the number of

morphemes per word; the opposing poles on this scale are isolating and polysynthetic.

Greenberg refers to this as the index

of synthesis.

A

second scale relates to the extent to which the morphemes are readily segmentable according to meaning/ function; the extremes on this scale are fusionai (inflecting) and agglutinating. This is Sapir's technique; and in his

The proceedings of this conference is not yet available in the U.S. The conference is known to me only from Enukidze's report, published in Iberijsko-Kavkazskoe Jazykoznanie 23:308-318. 3 In the discussion of this paper in Moscow, Professor Klimov pointed out that his typology does not implicate the "whole language," as it relates only to the syntax, morphology, morphophonology and lexicon, omitting the phonology.

58

ALICE C. HARRIS

quantificational agglutination.

approach,

Greenberg

refers

to

it

as

the

index

of

In a simplified approach based on Greenberg 1960, Comrie

(1981:43) calls the same classificatory base the index

of fusion.

One

advantage of this overall approach to classification is t h a t it permits the linguist

to

distinguish

between

synthesis

that

is

accomplished

by

agglutination, as in Eskimo, from synthesis by fusion (inflection), as in Southern Tiwa.

The degree of synthesis is logically distinct from the

degree of fusion. Languages

may

be

classified

 or suffixing/

as

predominantly

prefixing/

. In general, our uses of

these words coincide. When we look at infix/

, however, we find

some variation in both languages. Some linguists, however, use infix

only

with respect to affixes which, from a synchronic point of view, occur within another

morpheme

morpheme).

(or

between

the

two

parts

of

a

discontinuous

An example of this meaning may be taken from Udi, where

person-number markers are, under certain conditions, infixed, from synchronic point of view. In cam-zu-p-i

a

'I write', -zu- 'I' is infixed to camp

'write', which is followed by the suffix -i (Panchvidze 1974:148). 4

Others,

however, use the word in a weaker sense to mean an interior prefix or suffix, t h a t is a prefix preceded by another prefix or a suffix followed by another suffix.

This weakened usage seems to occur in both Russian and

English. Another characteristic of languages which has morphological aspects is agreement (and/or noun marking) according to grammatical person vs. grammatical class. This, too, may form the basis of a typology, and here again we find different uses of terms.

The variation in meaning/ usage

within each language seems to be at least as great as the

differences

between Russian and English. In their grammar of Avar, Ilia Cercvadze and Arnold Chikobava (1962:84-86) draw attention to the fact that both / gender skesi)

and / class/

k'lasi

(Georgian

have been used in the same meaning in

Diachronically, it appears that cam may have been an incorporated object of pesun; if this is correct, infixes in Udi originated from prefixes (cf. Greenberg 1957:92).

COMMENSURABILITY OF TERMS

59

describing Avar and other languages of the North East Caucasian family. Cercvadze and Chikobava themselves advocate using these words in the following ways: / gender should be reserved for languages like those Indo-European languages in which the 'masculine' category may contain not only words for male humans (e.g.,  'father'), but also words for male (lower) animals (e.g.,  'bull'), and even words for things (e.g.,  'table'). / class should, in their view, be applied to languages such as those of the North East Caucasian family, where only male humans belong to 'masculine'.

According to these authors, it is not gender t h a t

forms the basis for classification in class languages, but social worth, and this fact should be reflected in terminology (1962:84-86). terminology

is used

in more

recent

analyses,

and

In general, this

(the

appropriate)

languages of the North East Caucasian family are considered to have grammatical / class in both Russian and English. Allan

(1977:285),

in

his typology

of

noun

distinguishes gender from class in another way. criteria defining

classifiers,

classifying

systems,

He cites as one of two

"they have meaning, in the sense t h a t

a

classifier denotes some salient perceived or imputed characteristic of the entity to which an associated noun refers (or may refer)."

According to

him, a class system has a semantic basis; while a gender system, such as is found in European languages, is for the most part semantically empty (Allan 1977:290-291).

Since the classificatory systems of Avar and other

North East Caucasian languages are semantically based, Allan's definition leads to the same classification of Andi as Chikobava's (see especially Cercvadze 1965:94ff.). Many Africanists, however, define class language

differently.

For

them, a class language must mark class on the nouns themselves by means of an affix; some would add t h a t there must be agreement as well.

For

these linguists, languages in which class is made explicit only by affixes outside the noun are gender languages instead. By this definition, it would be inappropriate to consider languages of the North East Caucasian family to be class languages, since they have class marking on the nouns only in a few fossilized forms.

For example, in Andi we find vocci 'brother'/ yocci

'sister', but this occurs synchronically in only a few word pairs (Cercvadze

60

ALICE C. HARRIS

1965:102).

Thus, these different definitions of class language

result in

different classifications of the languages of the North East Caucasus. 5 Although

/

class

has

largely

replaced

/ gender

in

descriptions of languages of the North East Caucasian family, the term does not have a uniform meaning.

It is perhaps easiest to see the

differences with reference to the analysis of a specific language, and I will use Andi for the purpose of illustration. In Andi, four markers, v, y, 6, and r are used in the singular and plural according to the system schematized below: SINGULAR υ  b b r

PLURAL v   b r

According to one analysis of these facts, Andi has five grammatical classes, characterized by the singular/ plural pairs Class I v/v, Class II y/y,

Class

III b/y, Class IV b/b, and Class V r/r (Cercvadze 1965:94). Bokarev 1959 makes

an

analogous

analysis of the Dido

languages.

Andghuladze

1968:21-24 objects t h a t this analysis is wrong, stating t h a t classes must be counted by distinctions in the singular only, and thus Andi has four classes. This really is a problem of definitions and terminology. For Cercvadze, Bokarev, and others,  / class/

k'lasi is a group

of nouns distinguished by the marking it conditions; each

difference,

whether in the singular or plural, determines a class. These linguists find it useful to refer to the difference between b/y and b/b as one of class. For Andghuladze,

on

the

other

associated with the singular.

hand,

/ class/

k'lasi

is

primarily

This may be related to the fact t h a t he was

interested in reconstructing the system of the proto-language, and the situation in the singular is clearer than t h a t in the plural for this family. Classificatory marking on the nouns themselves may have once been productive in the languages of the North East Caucasus (Cercvadze 1965:98-99). Therefore it might still be argued that this family once had a class system, even if it is held that its members do not today.

COMMENSURABILITY OF TERMS

61

For some Africanists writing in English, class has yet a different meaning, namely a group of nouns having a common affix with respect to the singular

or with respect to the plural Thus a given noun may belong

to singular class S and to plural class P . A gender G then is a group of nouns characterized by a singular class S and by a plural class P , while a class t h a t constitutes a gender by itself is termed a single class (see, for example, de Wolf 1971:38). From Africanists, who deal with much more complex class systems, we may borrow some terms useful in the description of North East Caucasian languages.

(The definitions given below follow de Wolf 1971.)

The class content

of class  is the set of all items belonging to C. The

class distribution

is the arranging of a given class within a given class

system.

For example, the distribution of class b in Andi is (1) single class

in the gender 6, and (2) singular class in the gender b/y.

The use of these

and other terms might make our analyses more explicit.

Certainly a

greater uniformity of terminology would aid comparison. As a final point on morphology, we may remark on the /

polyp ersonali sm.

This

useful

term,

term

denoting

agreement of a verb with more than one nominal, is completely unknown to many American linguists, and deserves to become more widespread. 2. A l i g n m e n t When we t u r n to the terms / ergative, / active, and related words, we find many discrepancies in use on many levels. We may begin by noting t h a t the whole area of typology dealing with ergative, nominative-accusative, etc. is often referred to as ergativity/ (see works cited below, this section).



However, since this term seems to

give special weight or importance to one particular type, other terms have also been used.

Plank is, as far as I am aware, the first to have used the

neutral term alignment

to refer to this parameter (1979:4), and its use has

been taken up in some other works. There are some differences in names for cases. In both Russian and English, / ergative is the name of a case. 6 The unmarked case

'In Georgian, motxrobiti

'narrative' is used interchangeably with

ergat'iuli.

62

ALICE C. HARRIS

in an ergative system is often referred to as  / nominative

case/

saxelobiti

/ absolutive,

brunva,

though many people prefer to use

reserving nominative

for nominative-accusative

systems. In the Soviet Union,  ergative

construction

of

the

sentence

 / is used to describe a sentence

characterized by a subject marked by the ergative case. In parallel are the terms nominative

construction

and dative construction,

sentences in which the subject respectively.

used to describe

is in the nominative or dative case,

Generally these are then correlated with transitivity, by

means of statements to the effect t h a t the ergative construction,

for

example, is used with transitive verbs. There is no parallel to this usage in English. In

Soviet

usage, the

ergative

construction

and

the

nominative

construction together compose an  / ergative

structure.

These terms are used in this way in Klimov's work on typology (Klimov 1973, Klimov 1977, Klimov and Alekseev 1980), and often in descriptive grammars. For most linguists writing in English, ergativity and related notions are not

defined

in terms of sentence construction,

but in terms of

distribution of the cases. I like to use Sapir's chart to show this, adding to it the modern names for each distribution (Sapir 1917). Direct Object

Subj of Intrans Inactive

Ergative

A

Active

A

Accusative

A

Subj of Trans

Active A

 

A 

 

In the chart, "A" refers to one case, and " B " to some other case. an ergative

system

Thus,

means in English one in which the subject of a

transitive verb (or clause) is marked with one case, while the subject of an intransitive verb (or clause) and the direct object of a transitive are

COMMENSURABILITY OF TERMS

63

marked with a different case. An active system means one in which the subject of a transitive and the subject of an active intransitive are marked with one case, while a different case marks the subject of an inactive intransitive and the direct object of a transitive. An accusative system in English refers to one in which one case marks the subject of transitive and intransitive alike, while a different case marks direct objects. Although I have described this system in terms of case, exactly the same definitions can be extended to agreement and to a variety of syntactic phenomena. Charles Fillmore (1968:54) has represented the same facts in a different way, and many people prefer to use this representation:

accusative

nominative

nominative

ergative

active

inactive

In Fillmore's diagrams, "A" represents the agent and " O " the direct object.

The lines with both  and A represent a transitive sentence.

The

64

ALICE C. HARRIS

lines with A only represent an intransitive sentence with an active subject, and the lines with  only represent an intransitive sentence with an inactive subject. The lines drawn around them represent cases, agreement markers, or other grammatical phenomena that group the arguments of the sentence in the ways shown. Dixon uses "A" and "  " in much the same way as Fillmore, but defines them as, respectively, 'transitive subject' and 'transitive object' (1979:61). He represents 'intransitive subject' as " S " , and displays the difference between nominative/ accusative and absolutive/ ergative systems in the following way (loc. cit.): ERGATIVE NOMINATIVE

ABSOLUTIVE ACCUSATIVE While the distribution is differently d i s p l a y e d by Sapir, Fillmore, and Dixon, in terms of content the analyses are much the same. 7 Similar definitions in terms of distribution are given in other contemporary discussions in English (e.g., Anderson 1976; Comrie 1973, 1978; Plank 1979:4; Silverstein 1976). Although both Sapir and Fillmore, in their separate treatments, described the active system, the active type is unfortunately mostly ignored in recent works in English treating these types broadly (see, for example, those cited above, this section). While a number of American Indian languages have case marking or agreement that is active or similar to active, few recent descriptions make use of this term (see, for example,

In the discussion in Moscow, Professor Kibrik pointed to a problem in defining various alignments according to Sapir's schema, namely that it is based on the notions subject and direct object; he felt that for this reason Fillmore's approach was preferable. I might add here that the problem with Fillmore's schema is that not all subjects of transitives are, in fact, agents.

65

COMMENSURABILITY OF TERMS

Chafe 1970 on Onondaga, Davies 1981 on Choctaw, Van Valin 1977 on Lakhota).

However, I have used active

in the sense defined above in

referring to various case phenomena in Kartvelian languages (especially Harris 1985). It is unfortunate that the active type is sometimes grouped with socalled split

ergativity

(see below); doing so suggests, erroneously in my

view, t h a t such systems are subtypes of ergative ones. misleading

because

it

groups

under

one

heading

This term is also so many

different

phenomena (see below, this section). Dixon 1979 uses the phrase split marking

S-

type (see especially pp. 80-85), where " S " means 'subject of an

intransitive'; I understand this to refer to the same system as the term active

type. Klimov, in the works cited, proposes correlations between various

structures and other grammatical phenomena.

For example, for him a

language with an ergative system is also characterized by SOV word order, divisions of verbs into transitive construction (dative construction). typology

for this approach.

and

intransitive,

and

the

We may use the term

affective

whole-language

In practical terms, these correlates

may

become the basis on which the typology of a particular language is determined (e.g., Klimov 1976 of Laz). Because the same terms in English are defined instead on a strictly distributional basis, as outlined above, we find different labels applied to the same language (compare, for example, Klimov 1973 and Harris 1985 of the Kartvelian languages). In

English

nominative-accusative

we

find

language.

replaced with simply accusative

the

terms

ergative

language

and

More recently the latter has been partly language,

the reasoning here being that

the system should be named with the marked case, just as ergative to the marked member in the pair ergative-absolutive.

refers

Some linguists have

preferred instead to label the systems with the names of both unmarked and

marked

accusative

case;

for

example,

and absolutive/

ergative.

The terms ergative language,

Dixon accusative

(1979:61) language,

uses

nominative/

etc., are likewise

suggestive of whole-language typologies, though many linguists note that

66

ALICE C. HARRIS

this is not what they intend by their usage. 8 It is an often-noted fact t h a t a language

may

have

ergative

case marking

together

with

ergative

agreement (such as Avar), ergative case marking with no agreement (such as Lezgian) or with non-ergative agreement, or ergative agreement with no case marking (such as Abxaz). This variety may be multiplied many times when additional grammatical phenomena are taken into account (see Dixon 1972).

It is my feeling t h a t current usage in English is inconsistent with

the accepted reality.

With the realization of the many varieties available,

it would be far more precise to state the alignment of specific rules for a particular language, rather than calling it either an ergative language or an accusative language. Language L has, for example, ergative case marking and nominative-accusative agreement. As an alternative way of recognizing that in a specific language different rules have different alignments, some descriptions have referred to degrees of ergativity.

This terminology seems to be based on the view that

alignment is essentially a property of a whole language, together with the recognition t h a t the alignment of all rules in a particular language may not be the same.

It contrasts with the view that alignment is essentially a

property of individual rules, which requires t h a t the description of a language must specify

alignment

individually for

Because a language could satisfy

a particular

each relevant

"degree of

rule.

ergativity"

through different rules, this approach is just as imprecise as characterizing alignment as a property of a whole language. Several recent discussions of rule alignment have focused on ergativity,

sometimes called coexistent

systems

split

(e.g., in the first instance,

Delancey 1981, Dixon 1979, Silverstein 1976). Split ergativity

refers to the

fact that in some languages there is an "animacy hierarchy" (itself known under other names), such t h a t nominals toward one end of the hierarchy are marked with an ergative pattern, while those toward the other end are not marked or are marked with an accusative pattern.

This hierarchy

typically includes, in this order 1st and 2nd persons—3rd person, humans— non-human animates—inanimates. A second "split" may exist on the basis

Dixon 1979 does not use the comparable terms to refer to whole languages.

67

COMMENSURABILITY OF TERMS

of aspect (sometimes tense), such that imperfective aspect (or durative aspect, or present

tense) has a case or agreement

encoding rule of

accusative alignment, while the same rule in the perfective aspect (or past tense) has ergative alignment. "Splits" along any one of these parameters still appear to be unrelated to one another, in spite of attempts to treat them in a unified way.

Because of their distinct nature, it seems rather

unfortunate to group them under a single label, until such time as they may be shown to be related. 3 . V o i c e a n d C h a n g e s of G r a m m a t i c a l R e l a t i o n The term voice itself means the same as Russian  but is not used a great deal today by American linguists.

Diathesis

is used even less

frequently, though it has the same meaning in English as in Russian. Since the latter term has played a role in recent Soviet work on typology, we may briefly examine the nearest analog in recent American studies. In relational grammar, the two terms advancement

and

demotion

make reference to a hierarchy of grammatical relations: SUBJECT > DIRECT O B J E C T > INDIRECT O B J E C T > N O N T E R M . Subjects, direct objects, and indirect objects are referred to as

term

grammatical

that

relations,

or simply terms.

The hierarchy states

subjects rank higher than direct objects, that direct objects

outrank

indirect objects, and that all these rank higher than other nominals. Advancement

refers to a syntactic rule which, as it were, moves a nominal

up the hierarchy; for example, passivization makes a logical direct object into a subject.

Demotion

refers to the opposite process, movement of a

nominal down the hierarchy; for example, inversion makes a logical subject into a grammatical indirect object, as in the Svan sentence, (1)

bepšv-s kartvil-d loxgərgla child-DAT Georgian-ADV h e / s p e a k / I I I 'The child spoke in Georgian' The approach of relational grammar to the question of diathesis is

rather different from that of traditional grammatical analysis. diathesis

refers to forms

While

of the verb and takes these as a starting point,

ALICE C. HARRIS

68

relational grammar begins with the varieties of (logical and grammatical) relations. Morphological marking on the verb is treated as signifying that a syntactic rule has applied.

Although some morphological indication is

usually given on the verb, it is not diagnostic; some advancements and demotions go unsignalled.

Such an instance is the rule in English which

makes an indirect object a direct object; it relates sentences like (2) to sentences like (3). 9 (2) Mary gave the apron to John. (3) Mary gave John the apron. Note t h a t these sentences, while examples of an advancement fall within the bounds of what is generally regarded as

rule, do not diathesis—either

from a syntactic or a morphological point of view. The terms are thus not truly compatible, and they reflect a difference in approach to the analysis of syntactic constructions. The approach of relational grammar is quite similar to t h a t of the Leningrad School; both begin with what we may loosely call underlying structure, or logical form. For both approaches, morphology is a signalling device, not necessarily present. 10 This approach is in sharp contrast to the tradition of analyzing only those constructions and categories that are morphologically approaches, different.

signalled.

in other

In

respects,

In particular, from

spite

of

including

above; passive

this

similarity

terminology,

published definitions

Xolodovich 1974), it appears t h a t passive include many of the advancements

the

they

of

these

are

very

(Xrakovski 1973,

for the Leningrad group would

and demotions,

in the sense defined

in relational grammar refers only to the

construction

resulting from the promotion of a direct object to subject, other names

This rule is often called Dative Movement; in relational grammar it is referred to as 3-to-2 Advancement, where " 3 " denotes indirect object, and "2" direct object. The same is true in areas other than diathesis; for example, in analyzing causatives, both approaches place greater emphasis on the syntax and semantics of the arguments/actants than on the difference between morphological causatives and those employing a causative word.

69

COMMENSURABILITY OF TERMS

being given to the other constructions. 11 The term passive, as observed by Xrakovski (1973:69-70) is used in a variety of ways by traditional grammarians. One use that falls outside the usual range is the application of this term to a non-agentive intransitive in the Kartvelian languages. An example from Georgian is (4)

u^rebi

i

ba

da

ik'et'eba

drawers i t / o p e n / I and it/close/I 'The drawers open and close' (Gogeb., KEGL, v. 4. 887) Although these are called passives (vnebiti gvari) traditionally, it has often been noted that they are not t r u e passives. In relational grammar the difference between this construction and the passive is recognized terminologically; (4) is labelled an unaccusative, an advancement of a direct object to subject in t h e a b s e n c e of a logical s u b j e c t , while the label passive is reserved for (5), where the same advancement occurs but a logical s u b j e c t m a y be present. (5)

k'arebi-k'i isev ise dak'et'ili iq'o... doors-tho again thus closed be/11 'The doors, though, were closed that way again' (Ak'ak'i, KEGL, v. 3. 454)

According to this analysis, the prefix i- is the productive marker of the unaccusative. 1 2

According to the analysis of the Leningrad group, this

prefix, specifically in the verbs of (4), is the marker of the (Gecadze, Nedjalkov, and Xolodovich 1969:134ff.). Antipassive

is

a term

used

in English

anticausative

13

in the

description

of

constructions like (6) from Greenlandic Eskimo, where the object is in an oblique case (examples modified from Woodbury 1977:323). Constructions encompassed by advancement or demotion, but not by the Leningrad group's use of passive, include some instances of inversion, where the logical subject is not an agent, and unaccusative (see below). In a small, closed set of verbs ablaut grade distinguishes the unaccusative; the prefix i- has many additional uses. A construction in Lak and one in Tsova-Tush, which appear to be essentially similar to Georgian (4), are included as passives by Xrakovski 1973:68, 70.

70

ALICE C. HARRIS

(6)

piniartuq-Ø qinmi-mik hunter-ABS dog-INST 'The hunter beat a dog'

unata-i -šuq beat-APAS-IP:3:sg

(7)

piniartu-p qinmiq-Ø) unatar-aa hunter-ERG dog-ABS beat-TP:3sg,3sg 'The hunter beat the dog' While some linguists include only sentences like (6) as antipassives,

e.g., Comrie 1978, others include also a construction in which the logical direct object of a transitive is unexpressed, e.g., Dixon 1979:119.

This

second type is illustrated in (8) from Adighe (Rogava 1975:277). (8)

ar ma-z°,ǎ 3rd-NOM plow ' 'He plows'

(9)

aš ar yǎ-z o , ǎ 3rd-ERG 3rd-NOM plow' 'He plows it' (e.g., 'the earth') Georgian linguists use the Russian and Georgian term /

labiluri

to

refer

to

the

alternation

in

(8)-(9)

(Chikobava

1942).

/ labiluri refers to an alternation in which a single form of the verb may be found in both transitive and intransitive clauses (or, a single form of the verb may be either transitive or intransitive). The similarity between antipassive

and  has been noted

by Winfried Boeder (1979:462-463). Rogava 1975 has remarked also on the relationship between the construction with an unexpressed object, (8), and t h a t with the object marked by an oblique case. 4. C o m p l e m e n t a t i o n Perhaps the discrepancies in terminology are greatest when we go beyond the bounds of the simple clause. Among linguists writing in English, there is little agreement on terminology, and the same seems to be true among those writing in Russian.

I can only scratch the surface in a discussion of

variations in meaning/ usage; in this section I will concentrate on recent work (1980's).

71

COMMENSURABILITY OF TERMS

The typology group in Leningrad has turned its attention to this problem in recent years.

As reported by Enukidze (1985, see note 2), a

conference on this theme was held in Leningrad in 1981; the comments below are based on his report. Among other things, the conferees proposed various revisions in terminology,

and

some of these

are

specifically

discussed by Enukidze. In order to understand differences in terminology in this realm, it is helpful to understand fundamental

differences between points of view

concerning the actual structures involved.

The first difference I discuss

relates to views on various nonfinite verb forms: infinitives, absolutes, masdars, etc.

gerunds,

Many American linguists see these as reduced

forms of clauses; many Russians view them as exactly what they appear to be: substantives or substantive-like constituents of the main clause. A second difference

derives, I believe, from the first.

Although

American linguists are at present very much divided in their views of clause structure, many consider the entire dependent clause to be an argument of the main verb. Russian linguists seem to view the dependent verb, rather than the clause of which it is a part, as an argument of the main verb. Each position seems to result naturally from the corresponding view of nonfinite verb forms. Both differences are reflected in the terminology of typology. The Leningrad group refers to these constructions as constructions a predicate argument

actant.

containing

The use of actant here is similar to t h a t of the word

in English. 14

Thus, the constructions included are those in

which some argument of the verb is a predicate.

Clearly, this reflects the

view t h a t it is the dependent verb, rather than the clause of which it is a part, t h a t is an argument of the main verb. In English the whole range of such constructions is referred to most often as complementation.

The range of this term reflects the prevailing

In the usage of the Leningrad group, actant includes only nominals that are obligatory with a particular verb, others being circumstants. In the usage of most linguists writing in English, arguments includes nominals that are optional with a particular verb, though it is not so broad as to include all circumstants. Because the exact scope of argument varies a great deal in English, I will not try to define it further.

72

ALICE C. HARRIS

view among Americans that nonfmite verb forms constitute reduced clauses and t h a t it is these clauses t h a t are dependents of the main verb. usage of most linguists, complementation

In the

includes constructions in which a

dependent clause functions either as subject or as object of a main verb; it does not include either adverbial clauses or relative clauses. The range of the Russian phrase cited above seems to be the same as t h a t of the English term

complementation. Xrakovski has proposed to distinguish between verb and predicate

the following way. parts of speech, complex functors.

in

(a) Predicates may be verbs, but may also be other (b) Verbs may be predicates; however, main verbs in

constructions

(e.g., modals)

are

not

predicates,

but

rather

Statement (a) corresponds to current usage in both the US and

the USSR. Statement (b), on the other hand, is an innovation in both countries, and is not accepted by all (Enukidze 1985:17). Xrakovski

provide

Qasi-predicates

a tentative semantic would

include

at

classification

least

phase

Kasevich and

of main

verbs:

(e.g.,

'begin',

verbs

'continue', 'cease'), aspectual, temporal, modal, and orientational verbs. Supra-predicates

would refer to verbs of saying, thinking, knowing, feeling,

and emotion, among others. The second group is also referred to as modus predicates

(Kozinceva, at the 1981 Leningrad conference).

same conference, suggests the term operator

Birulin, at the

to refer to main verbs in

constructions with phase verbs, modals, or causatives; these are expressed in some languages by morphology, and it is thus useful to distinguish operator from verb proper. Complementation conjunctions.

may

be

overtly

marked

by

subordinating

Nowadays in English this phrase is usually replaced by

complementizer,

which includes subordinating conjunctions, but has a

slightly wider meaning. Complementizers include not only separate words, but

also

clitics,

subordination.

affixes,

or

any

combination

of

these

that

signal

Some linguists writing in English include markers of

nonfinite verb forms (such as English -ing of the gerund and participle) as complementizers,

while

others

do

not.

Some

would

include

as

a

complementizer the subjunctive mood or any other grammatical category which alone or in conjunction with other markers signals subordination; others would not include verbal categories of this kind,

73

COMMENSURABILITY OF TERMS

Parataxis

and hypotaxis

each in English.

are two terms with at least two meanings

According to one traditional definition, (a),

parataxis

refers to a structure in which two complete clauses are juxtaposed with no conjunction other than intonation or punctuation; hypotaxis

then refers to

a structure in which clauses are combined with an overt conjunction or conjoining paratactic

word. 15 According

to

a second

traditional

definition,

constructions are coordinate ones, while hypotactic

have subordination. 1 6

(b),

constructions

In this meaning, either paratactic or hypotactic

constructions may occur without a conjunction; in this case, refers to the absence of a conjunction.

For some, this

asyndetic distinction,

according to the second meaning, is not a formal distinction, but one of relationship of the two clauses. As far as I am aware, paratactic only in meaning

(b), t h a t

and hypotactic

are used in Russian

is, a contrast between subordination

and

coordination. Noonan 1985 uses paratactic above.

He speaks of paratactic

in a different way, related to the first

complementation

as having the following

properties: 1. A subject NP followed by a series of verb phrases. 2. Each verb phrase contains a fully inflected verb. 3. No marker of coordination or subordination links the two verb phrases. 4. No special verb forms are used. (Noonan 1985:55) Noonan's use of paratactic

differs from the first of the traditional ones

given above in that, in the examples he provides, the meaning of the whole cannot be inferred from the meanings of the parts.

For example, in the

Lango sentence,

This corresponds to the definition given by Crystal 1980:177, 257; it is the sense in which the words are used in several contemporary studies, including Lehmann 1980, esp. p. 113. 16

This corresponds to the definition given by Hamp 1966:33, 44; it is the sense in which the words are used in several contemporary studies, including Romaine 1982, esp. p. 53.

74

ALICE C. HARRIS

(10)

ìcó man

òdìá

àcégò

d gólá

pressed (3SG SUBJ 1SG OBJ) closed(lSG SUBJ)

door

'The man pressed me, I closed the door' (The man forced me to close the door)

(Noonan 1985:77)

the last line reveals that the sentence does not mean merely 'The man pressed me and I closed the door'. This may be contrasted with the more traditional use of the term parataxis, which would not include this additional meaning dimension. For example, the following Georgian sentence would be an example of parataxis for many grammarians, but its meaning is straightforwardly the sum of the meanings of its parts. (11) daik 'ripa

venaxebic,

daic'ura

q'ur3eni,

dadurda

grapevines

pressed

grapes

boiled

picked

mac'rebi wine

'The grapevines were picked, the grapes pressed, [and] the wine boiled' (Cited by Basilaia 1974:21.) An

additional

difference

between paratactic

complementation

in

Lango and traditional parataxis as in (11) is that the former does not permit a nominal subject after the first, though the notional subject may change, as in (10) (Noonan 1985:77). Other constructions similar in some respects to Noonan's "paratactic complementation" have been discussed in recent years, largely on the basis of African languages, though similar constructions have been described in Mandarin, Hittite, and other languages.

Each kind shows a range of

syntactic properties. Serial verbs is the name given to constructions where two or more verbs follow a single subject and refer to a single event (Lord 1973:269). Consecutives

likewise refer to a sequence of two or more verbs,

but the verbs in this instance relate to separate events that take place in sequential

order

(Hyman

1971).

In general, serials have no special

marking, and each verb is fully inflected. Consecutives, on the other hand, typically use a special verb form after the first.

Noonan's

paratactic

75

COMMENSURABILITY OF TERMS

complements differ from both, but have some characteristics of each. Like serial constructions, both verbs in the paratactic complement are fully inflected. Like consecutive constructions, the paratactic complements in Lango refer to distinct events. Constructions of these types are testing the limits and definitions of syntactic analysis and syntactic typology. We are grappling with the same problems (compare Nadjalkov's 1981 paper, reported by Enukidze 1985, with the those of the Africanists cited above), but with rather different data, and equipped with different sets of terms and the distinctions implicit in them. 5. C o n c l u s i o n It seems fair to say that the greatest problem in comparing the meanings of terms between Russian and English is the difference in usage of a single label within one of these languages (or both). Certainly in English there is great diversity between traditional uses, more modern uses outside any specific framework,

and the various uses within various

grammatical

frameworks. Having worked back and forth between multiple systems for years, I do not believe there is any problem in the comparability of terms in the analysis and description of individual languages.

As long as each author

defines his terms clearly and uses them consistently, the reader can follow the intention of terminology that is different from his own. However, problems may arise in studies which compare

primary

analyses by label, since the labels may mean very different things to different primary authors, as exemplified above.

Classifications based on

comparison of language analyses are accurate only when the differences in the meanings of terms are taken into account.

Metalanguage Alan Timberlake

1. I n t r o d u c t i o n By its nature typological research presupposes t h a t the facts of different languages can be compared. languages,

however,

metalanguage.

facts

In order to compare facts from must

be

translated

into

an

different

appropriate

In its most general form a metalanguage for typological

research can be taken to be simply a model of language expressed with whatever degree of notational explicitness.

structure, While the

notion of metalanguage in this sense is virtually commensurate with the notion of linguistic theory, we can nevertheless focus our attention on some issues t h a t arise in formulating a metalanguage for typological research in syntax.

These all involve the degree of correspondence among different

subsystems of language. It has become common practice to recognize t h a t syntax (broadly construed) involves a number of partially distinct domains.

For the

purposes of the discussion here we may distinguish the following: (1) basic syntactic relations (including notably the relations between predicates and actants);

(2)

predicate

formation

(including

such

operations

as

passivization, causativization, and the like); (3) morphology (including case assignment

and

agreement);

(4)

referencing

(including

not

only

reflexivization, pronominalization, and relative clause formation, but also the

coreferencing

that

is implicit

in the formation

of infinitival

or

participial verbal forms); (5) temporalization (including tense, aspect, and mood); and (6) linearization.

It is natural to assume, as do both many

traditional

some

approaches

and

contemporary

theories,

that

each

78

ALAN TIMBERLAKE

Operation

on

form

interpretation.

is

correlated

with

a

corresponding

semantic

Further, it is natural to assume t h a t any syntagmatic

operation (either in the specification of form or the interpretation of a form process) can be stated so that one element of the syntagm is a function applying to the other element; t h a t is, all syntagmatic operations involve functors and arguments. Against the background of these assumptions, many contemporary approaches to syntax make three additional assumptions; these all involve an assumption of correspondence in one form or another. First, it might be supposed that (for any syntagmatic process) the element t h a t is the functor and the element that is the operand would be the same both in form and in interpretation.

Using the notation of

categorial grammar, the combination of some functor category x/y with y (to give something of category x) is interpreted in a way such t h a t the meaning associated with x/y is a function from y-type meanings to x-type meanings. If we recall t h a t 'interpretation' here is equivalent to 'function' in its traditional sense of 'usage, value', then this assumption is simply the assumption

of

correspondence

between

form

and

function

(the

correspondence discussed by Nichols, this volume). Second, one might be tempted to assume that elements are either functors or arguments across all the six domains listed above.

Thus if, as

seems entirely natural (see Lehmann, this volume), noun phrases are taken to be arguments of predicates in the domain of basic relations, then according to this assumption one would expect in the morphology t h a t predicates would agree with their arguments (as predicted by Keenan 1974); further, one might expect the rules of linearization to be stated so that predicates determine the position of their arguments rather than vice versa. The third assumption, or set of assumptions, concerns predicateactant relations in particular.

An observation t h a t has been central to

much typological work in syntax is the observation t h a t many processes define an alignment, or orientation,

of actants across classes of predicates.

Thus for example, a process is said to be 'ergative' if it identifies the sole argument of a one-place predicate with the specifically Patient argument of

79

METALANGUAGE a two-place predicate. been

utilized

more

This kind of comparison across predicate types has consistently

for

morphological

processes,

and

in

particular for case marking, than for other processes, but it is in principle possible for any process.

Accordingly, much of the typologically oriented

work in syntax over the last twenty years has been directed to establishing the orientation of actants with respect to various syntactic processes. Describing the orientation of actants defined by any syntagmatic process is then one of the central tasks of any typologically oriented metalanguage (as observed by Harris, this volume). With respect to the orientation of actants, most approaches make a set of related assumptions. It is generally assumed t h a t there should be a single representation of predicate-actant relations that will serve for all domains.

This assumption is implicit in any approach that, for example,

uses reflexivization (specifically constraints on what actants may serve as antecedents for reflexives) to determine the 'subject' of a predicate; an argument of this sort presupposes t h a t the actant t h a t is the 'subject' in one domain must also be the 'subject' in another.

This in effect is the

assumption t h a t there is a primary orientation of actants for a given language.

Next, an assumption of uniqueness is commonly adopted--the

assumption that only one actant can hold any given predicate-actant relation within a given syntactic domain.

And finally, many approaches

make the additional assumption t h a t the syntactic relations of actants to predicates have a particular geometry, specifically, t h a t they can be stated in a hierarchy.

This last assumption is most clearly articulated in the

framework of relational grammar, which assumes a universal hierarchy of subject/ object/ second object, etc., but it is implicit in most instantiations of

categorial

transformational

grammar syntax;

and these

also

in

the

assumptions

various appear

redactions to

be

of

shared

additionally by role and reference grammar (Foley and Van Valin 1984), to the extent t h a t the distinction of 'pivot' vs. 'non-pivot' is a (minimal) set of syntactic relations. To recapitulate briefly, we find in most contemporary approaches to syntax a set of assumptions that in one way or another involve a high degree of correspondence between different

domains of the syntax: an

80

ALAN TIMBERLAKE

assumption of correspondence between form and function; an assumption of correspondence

among domains in the

definition

of functors

and

arguments; and a set of related assumptions about the orientation of actants with respect to classes of predicates (notably the assumption of a primary, hierarchized orientation). W h a t I wish to do in the following is to reexamine the second and third assumptions. First I will review the evidence—which is in fact wellknown--that argues against the assumption that predicate-actant relations can be characterized in terms of a unique, primary structure in which these relations are hierarchized.

After that I will suggest an alternative view of

the metalanguage that would be appropriate for typological research in syntax. In outline, the syntax of a language can be viewed as a set of form processes, each of which has a corresponding semantic effect.

These form

processes, such as those listed above as the six domains, are in principle independent and parallel components of grammar.

Each process can in

principle define its own view of which element is a functor and which an argument. Further, each process (to the extent t h a t it involves predicateactant relations at all) can in principle define its own orientation of actants. 2, C o n s i s t e n c y of O r i e n t a t i o n Let me now cite some examples--many of them familiar from

recent

literature—that

about

are

orientiation t h a t consistency

problematic

of orientation

Case Marking

for

were listed above.

the

set

of

assumptions

I will start

with the issue of

across domains, organized by pairs of processes.

and Agreement.

It is not at all uncommon for a

language to be partially or rigidly ergative in its case marking,

but

accusative in its predicate-actant agreement. This is the case, for example, in almost all P a p u a n languages, including Enga (Li and Lang 1979). Conversely, Finnish and (probably by areal diffusion) northern dialects of Russian

and

Lithuanian

allow

the

Patient

to

be

expressed

in

the

nominative under certain conditions, roughly when the predicate could not allow the Author--using

this term

here in preference

to the

better

established but more loaded term Agent—to be expressed as an overt constituent in the nominative; this 'nominative object', however, cannot

81

METALANGUAGE

control predicate-actant agreement in the predicate, and does not have any of other properties of 'subjects' (Timberlake 1974). Morphology

and Reflexivization.

Chechen is strictly ergative in its

case marking system, but antecedence of the reflexive pronoun stays with the Author of two-place predicates (Nichols 1980). Abxaz is ergative in its predicate-actant agreement pattern, but again antecedence of the reflexive pronoun stays with the Author of two-place predicates (Anderson 1976). The Tindi language of Daghestan (Kibrik 1985) allows reflexivization in two-place predicates either from Author to Patient or—remarkably from a cross-linguistic perspective—from Patient to Author, apparently as long as the source precedes the target reflexive in linear order; thus Tindi does not hierarchize the two actants of two-place predicates in the process of reflexivization, although it is morphologically ergative in both case marking and agreement.

In Russian the privilege of anteceding reflexive pronouns

normally resides with the actant that is nominative and controls agreement in the predicate (the subject, in the traditional use of the term), but with certain 'inverse' predicates (those with existential, quantifying, modal, or experiential meaning) the privilege of anteceding reflexive pronouns goes over to an actant t h a t has oblique case marking (Timberlake 1980). Morphology

andRelativisation.on.

A mismatch between case marking

and the strategies used for relativization is found in those Polynesian languages that are ergative in their case marking (Chung and Seiter 1980). Rennellese uses the same strategy (deletion) for Patients of two-place predicates as for

the sole actant

of one-place

predicates; it uses a

pronominalization strategy for the Author of two-place predicates.

This,

then, is a case where the classes of actants defined by case marking and by relativization do agree.

But the ergative language Niuean uses the same

strategy (deletion) for the sole actant of one-place predicates and for both actants of two-place predicates.

Thus, in contrast to other

ergative

Polynesian languages, Niuean defines a different orientation of actants in case marking than in relativization.

The contrast between these two

orientations in ergative Polynesian languages is striking in part because it undermines the possible conjecture t h a t accessibility to relativization might be determined not by any language internal orientation of actants, but

82

ALAN TIMBERLAKE

instead but a universal, hierarchized orientation (as in Keenan and Comrie 1977). Rennellese, where the orientation of relativization is identical to the orientation defined by case marking, violates the universal accessibility hierarchy. Relativization

and Reflexivization.

Schachter

(1976, 1977) has

pointed out t h a t the antecedent of reflexives in Tagalog is invariably the Author, regardless of the 'focus' form that the predicate occurs in.

Since

the focus form of the predicate determines which actant receives neutral (ang)

case and simultaneously which actant is uniquely accessible to

relativization, the actant t h a t is entitled to antecede reflexive pronouns need not correspond to the actant t h a t is privileged with respect to relativization.

The situation appears to be similar in Eskimo, although

some caution is necessary because the equivalent available from the same dialects.

information

is not

It appears to be true that the reflexive

(in Eskimology, the 'fourth person') can be anteceded by Authors of twoplace predicates, which creates an accusative orientation for antecedence of reflexives. The actant most accessible to relativization is the sole actant of one-place predicates, followed by the Patient of two-place predicates; Authors of two-place predicates are apparently not directly accessible to relativization.

This implies an ergative orientation for relativization (see

Smith 1984, although with a different interpretation). Morphology

and Linearization.

It is well-known that most languages

with ergative case marking have the primary word order SOV.

The word

order of the barest form on a one-place predicate (namely, SV) could be said to be ambiguous. If one argued that SV is determined by a rule that places the pivotal actant immediately to the left of the predicate, then one would have created an ergative alignment of SV and (S)OV. Alternatively, if one argued that the pivotal actant is placed in clause-initial position, one would have created an accusative alignment of SV and S(0)V. It seems to be generally agreed that the second strategy more realistically describes the distribution of actants with respect to predicates, for the reason that oblique actants of basically one-place predicates go between the absolutive actant and the predicate. In short, it appears that languages with ergative case marking systematically have different alignments in case marking and

METALANGUAGE

83

linearization of actants. A less extreme example of conflict between morphology and linearization is provided by 'inverse' predicates in SVO languages like Russian; such predicates place the morphologically oblique actant in the initial position preferentially reserved for nominative subjects. The examples cited above, which could be multiplied, are familiar. Curiously, the result they have led to is the attempt to typologize languages as 'deeply' vs. 'superficially' ergative (respectively, accusative); this typologization still maintains the assumption t h a t there is a single, primary orientation of actants. It seems to me that such examples should be taken as indicating rather t h a t it is uninteresting to ascribe a primary orientation to any given language. Instead, orientations are defined by processes. 3. Hierarchization Next

I would

like to

consider

three

cases of

orientation

that

are

problematic for the assumption t h a t predicate-actant relations can be stated as a monotonic hierarchy. The first comes from the process of relativization and formation in Chamorro.

question

Chamorro is a (predominantly) VSO language of

the Western Austronesian family spoken on the Mariana Islands (see Chung 1982). Agreement is complex, but simplifying somewhat, agreement for person/ number on the predicate (by means of pronominal enclitics or proclitics) has a different

orientation depending on the mood of the

predicate. In the irrealis mood, agreement is accusative in orientation, and in fact a two-place predicate shows agreement only with the Author.

In

the realis mood, agreement is ergative in orientation, and then a two-place predicate agrees with both the Author and the Patient.

When major

actants ( = A u t h o r and Patient) are expressed as independent constituents, they are both marked with a neutral, or absolutive, case in realis and irrealis moods, in contrast to a limited set of oblique cases. Relativization and question formation are accomplished by modifying the form of the predicate in accordance with the relation of the relativized or questioned

actant.

There

are four

distinct

modifications

predicate t h a t are utilized, which I will discuss in turn.

of

the

84

ALAN TIMBERLAKE UM: One strategy is to mark the predicate with the infix -um-,

with

the rest of the clause intact (this marking eliminates the possibility of any other agreement). This strategy is available exactly and necessarily for the Author actant of two-place predicates in the realis mood, the same actant t h a t triggers ergative agreement in ordinary clauses. 0: A second strategy involves no change in the case marking or agreement patterns of the clause except that the relativized or questioned actant is missing.

This strategy would appear to be the central or pivotal

strategy in t h a t it is the only strategy available for the sole argument of one-place predicates.

It is also the strategy used for the Author actant of

two-place predicates in the irrealis mood, an actant t h a t otherwise would trigger accusatively oriented agreement. This strategy is also the preferred, but not the unique, strategy for Patient actants of two-place predicates. IN N M Z : A third strategy involves creating a nominalization of the predicate by infixing the morphome -in-.

In this nominalization, the

expected subject of the predicate is expressed as a possessor. This strategy is only used for Patient actants of two-place predicates, where it then competes with the 0 strategy mentioned just above. Ø NMZ : The fourth

strategy is nominalization

without

any

overt

morphology on the predicate, when the expected subject of the predicate is expressed as the possessor (and a Patient would be expressed as an oblique).

This strategy is used for questioning various kinds of peripheral

actants (instruments, comitatives, and oblique case marked actants of certain lexical stative predicates like 'fear'). We can summarize the distribution of these strategies as in the following chart. Here I assume t h a t the sole actant of one-place predicates ( = P 1 ) has some thematic relation t h a t falls in the range between t h a t of the Author ( = A U ) and the Patient ( = P A ) of two-place predicates ( = P 2 ) ; this range of thematic relations is noted as 'AU . . . TH': (1)

AU, P 2 r e a l i s :

UM

AU, P 2 i r r e a l i s :

ø

AU...TH, P 1 :

ø

METALANGUAGE

PA, Ρ 2 :

ø,

Oblique:

85

INNMZ øNMZ

W h a t this gives is an orientation that is not strictly linear, but instead one that has a center and two peripheries.

The 0 strategy is

presumably central, since it is the strategy that is used for the sole actant of one-place predicates; additionally, it is also used for some other major actants, the Author of irrealis two-place predicates and (optionally as one of the two strategies) the Patient of two-place predicates. In contrast, the Author

actant of a realis two-place predicate uniquely uses the UM

strategy; on the other side, Patients of two-place predicates may also uniquely

use

the

IN N M Z

strategy.

This

geometry

of strategies

for

questioning and relativization, with a center and two peripheries, is quite obviously not monotonically hierarchical. A second example of a non-monotonic orientation is provided by the case marking system of Nez Perce (Aoki 1970, Rude 1982). Nez Perce has three cases used for major actants: ergative, absolutive (with no overt realization), and accusative. Absolutive case is used for the sole actant of one-place predicates, and is therefore presumably pivotal; absolutive case is also used for 1st and 2d person Authors of two-place predicates.

Ergative

case is used for 3d person Author actants of two-place predicates, as well as for possessors.

Accusative is used for the Patient of a two-place

predicate. This by itself is a non-monotonic orientation. Nez Perce provides more of interest.

W i t h three-place predicates

such as 'give', the Locus actant (the typically animate recipient of the Theme, the transferred object) is evidently treated as a major actant, in the sense t h a t it is expressed in the accusative. The Theme actant of such predicates is expressed in the absolutive case. Nez Perce also has a variety of predicate formation operators t h a t make potential oblique actants into the Patient actant (expressed by the accusative); if the original predicate was a two-place predicate, the original argument again goes into the absolutive case. And finally, Nez Perce has an antipassive operator (used productively especially when the possessor of the Theme is coreferential with the Author), and again the relevant actant~the actant that would

86

ALAN TMBERLAKE

have been expressed in the accusative if it were not for the application of anti-passive-- into the absolutive case; the Author is now likewise expressed in the absolutive. This is curious, since it is an instance where a 'demoted' actant is not assigned an oblique position, but instead is identified as central with respect to its case morphology. The orientation, then, of Nez Perce can be described as in (2): (2)

AU 3d , P 2 :

ergative

AU

absolutive

lst,2d :

AU...TH, P 1 :

absolutive

PA(TH), P 3 :

absolutive

AU, ANTIPASS(P n ):

absolutive

[n = 2, 3]

PA, ANTIPASS(P n ):

absolutive

[n = 2, 3]

PA, P 2 :

accusative

PA(LC), P 3 :

accusative

The absolutive case is pivotal, since it is the case used for the sole actant of a one-place predicate and also for a variety of other actants that are evidently 'central' or 'internal'. Authors and Patients of two-place predicates are then marked by special (in fact, morphologically overt) cases. Both of these examples, then, are cases where a process (relativization/ question formation in Chamorro, case marking in Nez Perce) imposes an orientation on the actants of predicates t h a t is not monotonic. Instead, one set of actants (including by definition the sole actant of one-place predicates) is pivotal, and the Author and Patient of two-place predicates are potentially (to some extent, under some circumstances) treated as peripheral. Another kind of problem for the assumption t h a t predicate-actant relations can be strictly hierarchized arises when the hierarchization required for one domain is distinct from the hierarchization required for

87

METALANGUAGE

another domain.

As an extended illustration of this, let us consider the

behavior of passive in Icelandic (Zaenen, Maling, and Thrainsson 1985). Passives of ordinary two-place predicates (for which the expected object would be expressed in the accusative in the active form of the predicate) are

unproblematic.

More interesting

is the

behavior

of

three-place

predicates. It is useful to recognize the classes and subclasses of three-place predicates given in (3); here I give the actants of the active forms in the linear

order

they

would

normally

appear

in,

with

the

relevant

morphological case along with translated illustrations: (3)

1. DAT L C , ACC T H : 'tell/give someone D A T something A C C ' 2a. ACC L C , DAT T H : 'conceal (from) someone A C C something D A T ' 2b. ACC L C , GEN T H : 'ask (of) someone A C C something G E N ' 3a. DAT L C , GEN T H : 'promise (to) someone D A T something D A T ' 3b. DAT L C , GEN T H : 'wish (to) someone D A T something G E N ' It is perhaps of some interest t h a t these valence patterns can be

described in terms of two thematic relations, Locus and Theme

(as

discussed below); in all instances of three-place predicates the Locus precedes the Theme. When passive applies to a three-place predicate, it usually (with one complication mentioned below) makes the left object argument the subject, at least in two respects.

This actant can be manipulated by further

predicate formation rules as if it were the highest ranking actant (for example, it undergoes raising). Also, it behaves as the subject with respect to a variety of linearization properties:

normally it occurs in preverbal

position (#SVX); it occurs in immediately postverbal position in questions (#VSX?); and it allows exactly one constituent to be topicalized in front of it (#XSVY). In these respects the relevant actant behaves as subject. If it is also an actant t h a t would otherwise be accusative (as in Patterns 2a, 2b), it also occurs in the nominative and controls agreement; but actants t h a t would otherwise be dative (Patterns 1, 3a, 3b) remain in the dative and do not control agreement.

Curiously, in the passive of P a t t e r n 1 the right

88 object

ALAN TIMBERLAKE does become

nominative

and

controls

agreement,

although

it

remains postverbal. Minimally, then, the passive of three-place predicates in Icelandic reveals discrepancies between the morphological definition of 'subject' and other (linearization, predicate formation) definitions.

In this

way Icelandic provides an example of non-consistency among domains. Let

us

consider,

though,

hierarchization of actants.

what

these

facts

say

about

the

Here there are two possible approaches, and

either leads to a problem.

Suppose we presume t h a t the hierarchical

structure is iconic to the linear structure.

Under this assumption the left

object actant would rank higher than the right actant, and quite naturally would be the

actant

that

is accessible to promotion

under

passive.

However, if this assumption is adopted, one cannot use an ordinary branching structure to produce the correct linearization: in the ordinary conventions of hierarchically branching tree structures, the actant t h a t is added at the deepest level should be closest to the predicate, yet under this assumption this would not be true. Under the opposite assumption the converse problem arises.

If we

adopt the hierarchical structure that works smoothly for linearization, then the right object actant would be higher. In this instance passive would not promote the higher ranking actant, but instead would have to select the lower ranking actant. Under either approach, then, we get the result that the

hierarchization

that

is necessary for

determining

accessibility

to

predicate formation rules is not the hierarchization t h a t is appropriate for determining linearization. Thus,

two

types

of

problems

arise

with

the

assumption

that

predicate-actant relations can be monotonically hierarchized: there exist orientations with a pivotal set of actants and two peripheries; and the hierarchizations defined by different domains need not be identical. 4. P r e d i c a t e - A c t a n t S t r u c t u r e If we do not adopt the assumption t h a t predicate-actant relations should be represented in a hierarchical fashion, it is natural to ask what form of metalinguistic representation might be appropriate. A natural possibility would be a modified form of categorial grammar in which a predicate is represented as sentence from which some number of actants is missing. In

89

METALANGUAGE

a hierarchical representation one would write, for a two-place predicate, < [ S / A ] / A > (using S for sentence and A for actant here, in place of the more traditional noun phrase).

Removing the hierarchization of actants,

we could write simply < S / A n , . . . , A m > , where η and m are indexes. Such a representation, of course, will not adequately

distinguish

actants unless the indexing system has some structure, which is exactly what we wish to avoid. As a metalinguistic indexing system it would seem most natural to attempt to use a substantive theory of thematic relations. Although most theories impose a number of restrictions on thematic relations (e.g., a given actant bears one and only one thematic relation), it seems more promising to think of thematic relations as functions that assign a scalar value to each actant. As a consequence, any actant can have a given thematic role to a greater or lesser extent, and (in principle) an actant can have positive values for more t h a n one thematic relation.

In

practice, of course, in most instances it will still be possible to characterize a given actant by a single label, which means t h a t it has a positive value for one thematic function but neutral or negative values for others. As a tentative proposal, I suggest the following. We can understand a predicate as stating a succession of states over time (and over worlds). The

most

basic thematic

function

is t h a t

of theme

(=TH),

which

characterizes the extent to which the actant is t h a t actant whose situation is reported by the predicate.

Next, an actant is a locus ( = L C ) to the

extent t h a t it is the actant in terms of which the situation of the Theme is reported; Locus includes such alternative labels as goal (recipient), source, path, and the like.

Not all predicates report the location or change of

location of the Theme; an actant is instead predicative

( = P R ) to the

extent t h a t the predicate reports a property of the Theme. These thematic relations can be viewed as inner

relations.

In addition, actants are also

characterized by the outer functions of author ( = A U ) , which characterizes the extent to which a given actant is responsible to the succession of states reported by the predicate, and of patient

( = P A ) , which characterizes the

extent to which a given actant changes as a function of time.

Presumably

the three functions TH, LC, and P R are mutually exclusive, and the two functions AU and P A are likewise mutually exclusive. Evidently, though,

90

ALAN TIMBERLAKE

an actant can be both AU and TH or both P A and TH (for example, with one-place and two-place predicates reporting motion of a Theme). This is written AU(TH) and PA(TH), respectively. Let us consider some typical examples. predicates shows the most variation.

The class of one-place

Thus the sole actant of such a

predicate will most likely be a TH, and in addition will have some value (or range of values) for the functions

of AU and PA. One way of

interpreting the survey of Merlan (1985) is that active-stative languages divide the class of one-place predicates either by attending to one or the other extreme of the AU . . . P A scale. For example, Lakhota assigns most one-place predicates 'objective' predicate agreement, which is to say t h a t 'subjective' marking is (for the most part) reserved for actants of one-place predicates t h a t

are definitively

authorial; in contrast, Seneca

assigns

'subjective' marking to most one-place predicates, and reserves 'objective' marking for a limited class. As Merlan's discussion shows, the assignment of any given predicate is to some extent conventional.

But in those

instances where both markings are available for a single lexical predicate (for example, extensively in Batsbi), it is always true t h a t the 'subjective' marking is used for the relatively authorial sense of the predicate and the 'objective' is used for the relatively patientive sense. Incidentally, there is no requirement under the system proposed here t h a t any given predicate have a single set of thematic values; it can rather display a range.

More

generally, predicates need not have a single valence pattern, but can instead display a range of patterns. Two-place predicates typically have one actant t h a t is definitively authorial, and another t h a t is patientive.

Quite commonly it does not

matter whether the P A actant is TH or LC, but in some instances this secondary

distinction

correlates

with

case

marking.

Thus

in

the

Tabassaran language of Daghestan (Kibrik 1985) predicates like 'throw', whose

non-authorial

actant

is a

TH,

will

typically

have

transitive

morphology (AU will be ergative, the TH absolutive), but predicates like 'hit (at)' or 'look at', whose non-authorial actant is rather a LC, will have somewhat different morphology (a noun as AU will be ergative, and a 1st or 2d person pronoun still triggers 'ergative' agreement, but the LC will be

91

METALANGUAGE

expressed in an oblique case rather than the absolutive). In Maori, a nonAU actant will typically receive the 'accusative' preposition i, but a nonAU actant t h a t is LC will receive the 'dative' preposition i. Thus, just as the sole actant of a one-place predicate shows a range of values for AU . . . PA, the sole non-AU of a two-place predicate shows a range of values for TH . . . LC. W i t h three-place predicates there is typically a clear

distinction

among AU, TH, and LC; observe t h a t the Icelandic predicates listed in (3) above—and these seem by and large representative of the cross-linguistic patterns—have two non-AU actants, one of which is clearly TH and the other as clearly LC. Just as there is interesting cross-linguistic variation in the way t h a t the sole actant of one-place predicates is oriented with either the AU or the P A of two-place predicates, there is also interesting variation in the way t h a t the sole non-AU actant of two-place predicates is oriented with either the TH or the LC of three-place predicates. Perhaps the more familiar pattern is the accusative/dative pattern of Russian or Turkish, whereby the typically non-AU actant of two-place predicates is oriented specifically with the TH of three-place predicates (and marked with the accusative case), while the LC of three-place predicates is marked with a distinct, dative, case.

But it is also possible to orient the sole non-AU

actant of two-place predicates with the LC of three-place predicates, either in case marking or predicate agreement; this is true in most Amerindian languages in North America (for example, Algonquian languages), and it is also true in some Western Austronesian languages.

Thus, Chamorro uses

the neutral (absolutive) case for the LC of predicates like 'buy' and 'sell', while the TH is expressed in the all-purpose oblique case. In Tagalog the so-called 'goal focus' evidently unites the single non-AU of two-place predicates and the LC of three-place predicates.

(The variability in case

marking of three-place predicates has been discussed by Dryer 1985, from a somewhat different theoretical stance.) As has emerged in the framework of relational grammar (Perlmutter 1980), many languages commonly distinguish a special class of two-place predicates in which one actant receives some sort of oblique case marking and the other direct case marking; commonly the privilege of controlling

92

ALAN TIMBERLAKE

reference pperations (reflexivization, adverbial participles, switch reference, and the like) is assigned to the oblique actant rather than the direct one. This class of predicates, which might be termed inverse,

shows at least

some

selected

cross-linguistic

similarities,

and

typically

is

from

existential/possessive, experiential, quantifying, a n d / o r modal predicates. For these predicates it seems generally that the actant marked with a direct case is a TH (it is this actant t h a t is quantified, asserted to exist, and

so on).

The oblique, or

'inverse', actant

is more

difficult

to

characterize, but it might be construed as a LC, in the sense t h a t it states the locus on which existence or quantification or modality is evaluated. Quite likely, both actants are neutral with respect to the outer thematic relations of P A and AU. Identifying the two actants of inverse predicates as LC and TH has the advantage that it does not require the introduction of yet other thematic relations such as 'percept' and 'experiencer' or the like. In any event, this class of predicates is distinguished from the usual 'transitive'

two-place

predicates

by

lacking

distribution of values for AU and P A .

the

canonically

clear

Consistent with this, it often

happens t h a t the 'inverse' actant, even when it is entitled to some of the same privileges as the AU of non-inverse two-place predicates, does not make use of these privileges to the same extent as the AU of a two-place predicate. For example, the dative 'inverse' actant in Russian is allowed to antecede reflexive pronouns, but only in the other major actant, not in circumstantial

targets.

In Chechen

'inverse'

actants

are allowed

antecede various cross-clause coreference markings, but the

to

asymmetry

between the two actants is less for inverse predicates than for non-inverse predicates (Nichols 1980).

In Choctaw (see Davies 1984), clauses are

marked for same vs. different subject. With actants t h a t are AU subjects, there is no ambiguity in usage; but with inverse actants either same or different subject marking is possible.

Although the dual behavior of

inverse actants could be taken as an indication t h a t the inverse actant has changed syntactic

relations (as it is in the framework

of

relational

grammar), it can just as easily be understood as an indication t h a t the inverse actant is intermediate in behavior between the AU and P A actants of transitive two-place predicates.

93

METALANGUAGE

These basic patterns are stated in summary form in (4). Here commas delimit distinct arguments, while dots indicate t h a t a given actant may have a range of values:

(4)

P1:

{AU(TH) . . . PA(TH)}

P2transitive: transitive*

{AU, PA(TH . . . LC)}

P2

{LC, TH}

: inverse

P3:

{AU, PA(TH), PA(LC)}

The system of substantive thematic relations proposed here is of course by no means novel, and derives from work such as Gruber 1965, Fillmore 1968, and others; the distinction of TH and LC is essentially that of 'trajector' and 'path' in Talmy 1985. innovations.

There are, however, two small

By allowing thematic relations to have scalar values, we do

not impose any implicit orientation on thematic relations of actants.

In

particular, the sole actant of a one-place predicate does not have to be identified primarily either with the AU or the P A of two-place predicates; this presumably is the purpose of Dixon's (1979) three-way distinction of Agent ( = A U ) , Subject ( = A U

. . . PA), and Object ( = P A ) .

And by

allowing the distinction of two layers of relations, the sole actant of a oneplace predicate can be both TH and (to some degree or another) either AU or PA; this allows us to state the respects in which the sole actant of a oneplace predicate can exhibit behavioral properties either like the AU or the P A of a two-place predicate. I should repeat, though, t h a t the purpose of introducing these thematic relations is not to claim t h a t by themselves they constitute a theory of predicate-actant relations, but rather to use them as an indexing system for distinguishing actants from each other.

Any

orientation, or alignment, of actants must be defined by a given process, and may to some extent be arbitrary or conventional. 5. P r e d i c a t e F o r m a t i o n O p e r a t o r s One potential argument for the assumption that there is a uniquely primary, hierarchized definition of predicate-actant relations could come from predicate formation operations--operations t h a t create new predicates

94

ALAN TIMBERLAKE

from other predicates; I have in mind here such familiar operations as passive, raising, causative, and the like. Now if the predicate t h a t resulted from the application of one of these operations behaved identically to an ordinary lexical predicate, then it would at least be convenient to describe basic predicates and derived predicates in terms of a single predicateactant structure.

For example, given that passive typically creates a one-

place predicate from a two-place predicate, then one could expect this derived one-place predicate to be identical in all respects to lexical oneplace predicates; in languages where the notion of 'subject' is defined by the constellation of a number of behavioral properties (Keenan 1976), then the sole argument

of the

derived one-place predicate should

behave

specifically as a subject. There are various indications that this expectation is not completely fulfilled, although of course it is fulfilled for the most part. For example, the derived subject of a passive in Russian is to some extent allowed to antecede reflexive pronouns (as one would expect of any subject), but the range of actant positions is limited; for a target that is a temporal or locative circumstantial, it is more natural to use the non-reflexive pronoun than the reflexive pronoun under coreference to the derived subject.

For

these same circumstantial actants a true subject (of an active lexical predicate) would have to use a reflexive pronoun. In this respect a derived one-place passive predicate is not equivalent to an active lexical one-place predicate. Further, predicate formation rules cannot apply with total freedom to predicates that are themselves derived by the application of some other predicate formation rule. Thus it is universally almost impossible to derive a passive from a predicate that is already passive, even in those languages t h a t otherwise form passives of one-place predicates.

It is also fairly

common t h a t it is difficult or impossible to form a causative from a passive; this is especially striking, since cross-linguistically causatives apply most freely to one-place predicates, and in the typical case a passive is exactly a derived one-place predicate. It should be pointed out t h a t the various approaches to grammar which operate with a primary predicate-actant structure can be modified

95

METALANGUAGE

in one way or another to account for the observation t h a t

derived

predicates are not equivalent to lexical predicates. The point is rather that once it is established that derived predicates are not necessarily equivalent to lexical predicates (of the same valence), then the positive motivation for positing a single primary structure that states predicate-actant relations disappears. If predicate formation operations do not oblige us to posit a single, hierarchical

structure

for

predicate-actant

relations, we still

need

to

consider how such operations might be described. In fact, a body of recent work has suggested how such operations as passive, causative, benefactive, raising, equi, and the like can be defined directly as operations t h a t create predicates from predicates (Keenan 1982, Dowty 1982a, 1982b, Keenan and Timberlake

1985).

Although

in this literature

the

predicate-actant

relations are treated hierarchically, this is not, I think, strictly necessary. W h a t one has to do is to specify the properties of the actants of the derived predicate in relation to those of the source predicate.

The

morphological properties of derived predicates can be stated simply as part of the predicate formation operation; the real complication concerns how to deal with the indexing system—on the approach advocated here, the thematic relations--of derived predicates.

While a general answer to this

question is not clear, we can illustrate with the example of causatives what a solution might look like. By definition, a causative can be taken to be an operator introduces an AU actant over and above the prior actants.

that

Suppose we

index the new actant as AU and the original AU of the source predicate as AU'.

This introduces an additional parameter into the model that is not

wholly defined (what does it mean to mark a thematic relation as prime?), but we could at least specify t h a t the degree of authorship of A U ' is less than that of AU; more radically, one could suggest t h a t A U ' is a patientive author--that is, it has the thematic relation A U ' = P A ( A U ) . this A U '

In any event,

often behaves like one of the primitive thematic

available for lexical predicates.

relations

The general pattern is t h a t the A U ' of

causativized one-place predicates will behave like a PA, although there are exceptions to this generalization.

For example, Gilyak ( = N i v x )

(as

96

ALAN TIMBERLAKE

discussed by Comrie 1976) has a special case used for the A U '

in

causatives of both one- and two-place predicates, and Japanese optionally allows the use of the oblique case for this actant, depending on the degree of its authorship and animacy. ausativization

of

two-place

predicates

shows

interesting

cross-

linguistic variation, which leads to an interesting correlation with the case marking of lexical three-place predicates (see Dryer 1985).

One type

(sometimes asserted to be universal, as in Comrie 1976 and Foley and Van Valin 1984) puts the P A of the source predicate in the ordinary case for PA--accusative or absolutive--but puts the A U ' in an oblique or dative case, the case normally used for LC.

Familiar examples here are the

accusative languages Turkish and French; the morphologically ergative language Tindi of Daghestan (Kibrik 1985) also does this. But there is a distinct type puts the A U ' in the object case and the P A of the original predicate in an oblique case.

Curiously, the languages t h a t employ this

strategy are those t h a t independently use primary object case specifically for the LC of three-place predicates, but use a secondary object case (or oblique) for the TH of lexical three-place predicates; an example here is Chamorro (Gibson

1980).

The regularity, then, is that the

derived

predicate thematic relations {AU, A U ' , P A } are treated as parallel to the lexical thematic relations {AU, PA(LC), PA(TH)}, respectively.

In both

types A U ' is treated as analogous to PA(LC). Incidentally, this parallelism is natural if one thinks of A U ' as PA(AU), given that in other respects LC can be analogous to AU; for example, the control of reference operations by LC

(in inverse predicates)

is analogous to control

by AU.

The

significance of this pattern is not entirely clear, but I think it suggests t h a t the case marking patterns of derived causatives can be stated in terms of a slightly modified set of thematic relations, close to those t h a t are necessary for lexical predicates. Thus, in principle it should be possible to use thematic relations even in derived predicates as an indexing system for distinguishing actants from each other. More generally, it appears to be possible to describe predicate formation

processes

without

utilizing

an

independent,

structure for representing predicate-actant relations.

hierarchized

METALANGUAGE

97

6. C o r r e s p o n d e n c e of F u n c t i o n - A r g u m e n t S t r u c t u r e To this point I have suggested that a metalanguage for typological research need not include a single, hierarchized (oriented) structure of predicateactant relations. Instead, each domain of syntax defines its own orientation of actants with respect to predicates; actants can be distinguished from each other by referring to thematic relations. Thematic relations can then be used both as substantive description and as an indexing system. The discussion above was intended as one possible answer to the question of how much correspondence there is among different domains with respect to predicate-actant relations. There is a further question to be considered here concerning the degree of correspondence between distinct domains of syntax. It might be t h a t (as argued here) the predicate-actant relations would not be uniform across domains, but that still the characterization of what is a functor category and what an argument category would remain constant across domains. In particular, if actants are treated as argument categories and predicates as functor categories in the domain of basic syntactic relations, then one can ask whether actants are always arguments and predicates always functors in all domains of syntax. I am not confident that there is a clear answer to this question, since much depends on how each domain is structured. But we can consider briefly what would be required in order to maintain the possible hypothesis that domains correspond with respect to functor/ argument structure, starting with morphology. Let us suppose that we write morphological features as a separate entry in the categorial statement, separated from the statement of the syntactic category by ';'. Thus if an actant shows the arbitrary morphological feature x, we can write it as having the category < A ; x > . For a predicate category t h a t shows morphological features, we can write minimally < S / A n , . . . , A m ; y > , where again y represents whatever morphological features are expressed by the lexical predicate. Morphological processes typologize into (at least) two kinds, agreement and case marking. To represent t h a t a given predicate assigns a certain case to one of its actants, we could write the predicate as

98

ALAN TIMBERLAKE

demanding the case cn for the actant position An.

In general terms, a

predicate would have the structure in (5): (5)

This is to say t h a t a predicate with its own morphological features y demands actants t h a t occur in cases cn, . . . , c m . If one adopts this view, then, t h a t case marking is (by definition) a certain morphological expression on actants, case can be viewed as an argument category in the domain of morphology. To describe agreement of predicates with actants, we might write a predicate as expressing a certain complex of features, say gnp for gender, number, and person (to take the features most commonly expressed under agreement) on the left of the functor symbol ' / ' and again on the right in the actant position t h a t controls agreement. For example, a one-place predicate would have the structure in (6): (6)

This is to say t h a t the predicate expresses the same gnp features in itself t h a t it expects to find in a certain actant position, here the position subscripted by n.

In instances where a predicate agrees with more than

one actant, the left position of the predicate has to carry more than one set of features.

For example, agreement in a two-place predicate with both

arguments would give the predicate

the

category

in (7), where

'+'

concatenates two sets of features expressed simultaneously. (7)

A

m;

gnpn + gnp m /gnp n , g n p m >

On this approach, then, both agreement and case marking can be expressed as properties that the predicate dictates for its actants.

In this

way the predicate is the functor in the morphological domain and the actants are arguments, just as in the basic statement of the syntactic categories. With respect to morphology, then, it would appear possible in principle to maintain the assumption t h a t there is a correspondence in the definition of functor and argument categories across domains. Things get progressively more difficult, however, when we look at the reference domain and at linearization.

Under reference operations we can

99

METALANGUAGE

include those t h a t express coreference between a target and a source and distinguish between direct and indirect

coreference.

In direct coreference,

target and source are actants of the same predicate, and the coreference device

indicates

pronouns. different

coreference

In indirect predicates

alone;

the obvious

coreference, target

and, more

example

is

reflexive

and source are actants of

importantly,

coreference

is

indirectly

stipulated as part of the more general process of subordinating a predicate to some other category.

Under indirect coreference three subtypes can be

distinguished, depending on the syntactic relation of the subordinated predicate to the host category: the predicate can be subordinated as an argument category (e.g., infinitives), as a nominal modifier (e.g., relative clauses), or as an adverbial modifier

(adverbial participles and other

constructions marking 'switch reference'). Coreference could potentially be viewed as a relation directly between the target and the source. coreference--even

There is, however, considerable evidence that

in the most transparent instances involving

reflexive

pronouns—crucially involves the predicate of which the target and source are actants.

Thus, for example, some predicates in Russian specifically

require a reflexive pronoun in a given actant position (Padučeva 1985:181). In addition, in Russian the source for reflexive pronouns is normally the 'subject', defined by the coincidence of nominative case marking and control of agreement in a finite predicate.

An accusative object can

nevertheless antecede reflexive pronouns with certain predicates, but not with all.

Thus coreference is possible between the accusative TH actant

and the oblique LC actant of, for example, zaščiščat'

kogo-to ot sebja 'to

defend someone from self' but not between the accusative LC actant and the oblique TH(?) actant of *rassprosit' someone about self'.

kogo-to

o sebe 'to question

Thus the ability of an accusative actant to serve as

source for an oblique reflexive pronoun depends on the specific predicate and the thematic relations it assigns to its non-AU actants.

In short,

coreference--even when it is expressed directly by means of a reflexive pronoun--is in part a property of predicates, and not just of the two actant positions. To formalize coreference in a more or less categorial form (that is, in

100

ALAN TIMBERLAKE

a form t h a t might allow us to maintain the assumption that a category is either a functor or an argument category across all domains), we might assign referential indices to actants and write these indices as a separate entry in the categorial statement statement and morphology).

(in addition to the basic

category

Coreference itself could be understood as an

asymmetric and transitive relation, which can be denoted (arbitrarily) as 'Δ'.

Then iAj will mean t h a t j is (asymmetrically) coreferential with i.

A

predicate with coreferential subject and object might be written as (8)

< S / A A U , A PA ; x / n o m , acc(rfl); i Δ j / i , j >

to be understood as a two-place predicate takes an accusative reflexive pronoun as P A actant with reference j and a nominative AU actant with reference i and assigns coreference ( = iAj) of the P A to the AU. A threeplace predicate zaščiščat with reflexive pronoun in the oblique actant position would have two entries: (9)

a.

< S / A A U , A PA(TH) , A PA(LC) ; x / n o m , ace, obl(rfl); iΔk/i, j , k >

b.

< S / A A U , A P A ( T H ) , A PA(LC) ; x/nom, ace, obl(rfl); jΔk/i, j , k >

depending on whether there is coreference between the oblique actant and the nominative Author ((9a)) or the oblique and the accusative Patient ((9b)). Similar formalisms can perhaps be developed for instances of indirect coreference. To illustrate the situation with an infinitive that occurs as the P A actant of a two-place predicate--when coreference will be required between the AU actant position of the infinitive and the AU of the host predicate--we might write the formalism in (10).

Here [S/A AU ] is to be

understood as the infinitive of a one-place predicate. (10)

< S / A A U , [S/A AU ]; x/nom, inf; i Δ j / i , [iΔj/j]>

To illustrate relative clauses, we might think of a relative clause as essentially an adjective, so t h a t it would have the category < A / A > , and of a relative pronoun as a functor that converts a predicate t h a t is missing one actant to an adjective. In the simplest situation—a relative clause on a

101

METALANGUAGE

subject position--a relative pronoun would have the category < [ A / A ] / [ S / A ] > , inasmuch as it converts a one-place predicate missing its subject actant to an adjective. Coreference holds between the host noun and the relativized argument position--subject in the instance just discussed. Thus, ignoring morphology, a relative pronoun on the subject would have the category in (11): (11)

< [ A / A ] / [ S / A ] ; [i/i]/[iΔj/j]> Details aside, the point here is to suggest t h a t it should be possible to

develop a formalism for expressing coreference between actant positions in which coreference is stated as a property of predicates.

In this way

predicates remain functor categories in the domain of reference, just as they are presumably functor categories in the basic categorial statement, and the correspondence of functor and argument categories across domains is maintained. Finally, let us ask whether linearization can be represented in terms such t h a t the correspondence of functor and argument categories can be preserved.

Here the results of attempting to maintain the correspondence

are somewhat dubious.

In order to represent linearization, let us suppose

that

statement

each

categorial

includes

an

explicit

statement

of

linearization that the functor category imposes on its arguments. Thus we might revise the statement of categories to distinguish between ' r / ' , which will be concatenation of the argument on the right, and ' 1 /', which will be concatenation of the argument on the left.

For example, a two-place

predicate in an SVO language could have the shape in (12), in the typical situation: (12)

< S 1 /A A U ' r / A P A > If a formalism of this type is attempted, however, some problems

arise, two of which I will mention here. First, it is difficult on this view to give the correct linear order to two or more arguments when they occur on the same side of the predicate. This problem arises systematically for two- or three-place predicates in verbmarginal languages (SOV, VSO, VOS), and it also arises for three-place predicates in verb-medial

languages (that is, for SVO 1 O 2 predicates).

102

ALAN TIMBERLAKE

Without some further mechanism for ranking the order of arguments, even the directionalized notation just introduced would not be able to linearize correctly two actants t h a t occur on the same side of the predicate.

The

response would be to hierarchize linearization. For example, to describe a VSO language, we might write parentheses around the instruction for concatenation of the innermost actant; thus an actant concatenated by r

will be closer to the predicate than the actant concatenated by / .

(r)

/

Then a

two-place predicate in a VSO language will have the category in (13): (13)

And a three-place predicate in an SVO language (like those of Icelandic listed schematically in (3) above) might be written as: (14)

< S /A A U ,

(r)

/APA(LC)> r /A P A ( T H )>

But if the linear order is hierarchized, it may well be true t h a t the hierarchization for linear order will not match the hierarchy for some other domain.

This kind of mismatch was noted above in connection with the

Icelandic passive. Second, it may be difficult on this approach to express cleanly some of the observed word order regularities across categories. One firm and by now generally known result of the research on word order is the correlation between modifier/head order, adposition/argument, and predicate/actant order.

To take the most consistent correlation, (S)OV languages would

presumably combine actants with predicates consistently on the left, so t h a t the predicates in such languages would consistently use

1

/. The same

would be true of adpositional elements, which are presumably

functor

categories; in the (S)OV language type adpositions are postpositional, so they would also concatenate using

1

/.

In (S)OV languages, however,

modifiers commonly precede the modified element.

If, as is traditional,

modifiers are treated as functor categories, modifiers in (S)OV languages would consistently use the linearizing operation r / , given t h a t the head of the modifier occurs to its right.

Thus we arrive at something of an

impasse. If, in order to have a generalized account of the linearization of functor and argument categories, we attempt to use a single directionality of concatenation in a given language type, then we will be forced to the

METALANGUAGE

unlikely

conclusion

that

modifiers

are

103

not

functor

categories.

Alternatively, if we stick with the traditional wisdom t h a t modifiers are indeed functors, then even the highly consistent word order type (S)OV will use 1/ for some functor categories and r / for others.

In this instance

there is no consistency in the linearization assigned by functor categories. In turn, this means that it is not possible to give a generalized account of the cross-categorial regularities in word order in terms of functor vs. argument categories. For at least these two reasons, then, the assumption t h a t there is a correspondence between functor and argument categories across domains is somewhat problematic with respect to linearization in particular. 7. C o n c l u s i o n In this paper I have considered two related problems t h a t arise in the attempt to construct a metalanguage for typological research in syntax, both involving the degree of correspondence among different domains of syntax.

First, I have suggested t h a t there need be no correspondence in

the orientations of predicate-actant relations t h a t are defined by different syntactic processes from domain).

different

domains (or even within the

same

As a consequence, the metalanguage should not contain any

implicit orientation of predicate-actant relations. Given t h a t of course any formal statement of predicates must in some way or another distinguish among actants, I have proposed a substantive set of thematic relations that might serve this purpose. If such thematic relations can be assumed to be universal, they might provide the basis for comparison of orientations across languages.

Second, I have suggested that it may be possible to

maintain the assumption t h a t functor categories and argument categories remain constant across domains, although the assumption is problematic with respect to linearization. The discussion here was carried out in terms of a faintly categorial metalanguage, but it is worth pointing that similar problems would

arise

in

non-categorial

approaches,

for

example,

in

dependency grammar. More generally, the discussion here is intended to expose what Bazeli (1952/66) terms the "correspondence fallacy," namely "the assumption t h a t two distinct (sets of) criteria will necessarily lead to isomorphous

104

ALAN TIMBERLAKE

analyses"; the fallacy is characterized equivalently by McCawley (1980:169) as "the idea t h a t there is a single linguistic level to which all combinatoric restrictions basically relate." This fallacy arises, I suspect, when we make the assumption t h a t structure determines form rather than vice versa.

On the Notion of Language Type G. A . K l i m o v

Contrary to the viewpoint which assumes language type to be an entirely optional

notion

in typology,

I proceed

from

the

statement

(consecrated by a long-standing tradition) which holds t h a t this notion is the basic one for typology. To my mind, linguistic research proves that, in the opposite case, it would be impossible not only to assign any language to a definite typological class of languages, but even to identify

any

structural phenomenon as a fact of typological relevance. Hence, favorable conditions arise for confusing typological inquiry with any other structural treatment of language, or (at least) for depriving typology of the enviable purposefulness which is so characteristic of the two other

fundamental

branches of the comparative study of languages--genetic (or historicalcomparative) and areal lingustics. I suppose t h a t in any event a proper typological inquiry (i.e., not a contrastive,

characterological

or

universalistic

one)

becomes

virtually

impossible without working out the notion of language type as the abstract model for the comparison of languages which are structurally different. Precisely, these types prove useful as means for overcoming arbitrariness in criteria choice.

This is still rather popular in some modern typological

schools, because it is only by having recourse to strict criteria t h a t we may expect to demonstrate t h a t far-reaching typological similarities between certain languages are not accidental.

This is due, as seemed already clear

in Sapir's time (Sapir 1921:121-122), to parallel development of languages. By analogy to abstract models of language family (i.e., the parent language) and of areal grouping (i.e., to the unity of areally correlated

106

G. A. KLIMOV

features), the language type first of all must be explicitly divorced from the class of actually existing languages, which are in fact described in the terms of language type. Thus, taking into account t h a t a typological class of languages is a phenomenon of the same order as a language family or an areal group {Sprachbund),

it will be necessary to recognize that a language

type should represent a construct comprising a set of structural features, serving

as diagnostic

criteria for

attributing

languages to a

definite

typological class. In other words, unlike the corresponding typological class of languages represented on the linguistic map of the world, the language type is a concept inferred by a linguist. with

the

widespread

formulation

It is therefore difficult to agree

according

to

which

a

typological

classification groups languages together into language types. One of the most important consequences of such delimitation may be the possibility of constructing language types which, due to historical accident, happen not to be among existing languages. Meanwhile, only by obeying this delimitation will typological inquiry be able to give an answer to the question posed by Hjelmslev: what language structures are possible and why is this so, whereas others are impossible (1970:96)? Proceeding from the fact t h a t genetic and areal lingustics deal in their

procedures with

criteria which

are

not

arbitrary,

but--on

the

contrary--strictly determinate criteria, it would be natural to expect also t h a t typological linguistics should elicit determinate rather than arbitrary fundamentals.

Such an approach is well known to have been formulated

repeatedly already in the history of linguistics.

Thus, according

to

Jakobson, "linguistic typology based on arbitrarily selected traits cannot yield satisfactory results, any more than would a classification of the animal kingdom which instead of the productive division into vertebrates and non-vertebrates, mammals and birds, etc., used the criterion of skin color and on this basis grouped together, for example, white people and light pigs" (1971:525). persists

that

criteria

Nevertheless, in modern typology the idea still for

typological

analysis

should

be

arbitrary.

Therefore, the conventionality of using them in typological classifications is a commonly held opinion, though their explanatory power is absolutely incommensurable with the analogous potentials of genetic (genealogical)

ON THE NOTION OF LANGUAGE TYPE

classification.

107

This view seems to be clearly reflected in the realm of

formally oriented typology, where the number of classificatory schemes recognized as typological, is increasing.

Incidentally, an entirely different

situation arises in content-oriented typology, where conceptually prevails, confirming essentially one and the same set of language types: nominative (accusative), ergative, active, and possibly some other systems. If it is granted that the proper procedure for constructing language types is non-arbitrary, we have to assume that a type cannot rest on a single structural feature.

It is quite evident that, if we attempt

to

construct a unidimensional type, we have no guarantee of the typological relevance of the chosen feature.

I hold t h a t the total complex of features

making up a language type, e.g., a set of its differential features, should comprise

only

phenomena.

logically

interdependent--and

not

freely

combinable--

One may assume also that such phenomena will form a

certain hierarchy corresponding to the ranked organization of

different

levels of language. The principle of interdependence of structural elements within a language type implies in its turn a principle of "purity" of type. T h a t is, it

presupposes

a

logical

consistency

of

the

type's

foundations

and

conversely, dubiousness of the notion of "mixed" (transitional) type, which still survives in some linguistic works (it is clear that sometimes the term "mixed t y p e " must be qualified as a mere lapsus dicendi).

Although this

principle does not seem to have gained a clear formulation in typology so far, some of its prerequisites may be distinguished in many publications. For example, Skalička defines a type as an extreme language stage, with the

mutually

favorable

structural

phenomena

maximally

developed

(1979:123). Also, Bechert indicates explicitly the logical incompatibility of accusative and ergative systems in the framework of content oriented typology

(5).

It

seems to

(Mischentypus, type mixte)

me

that

the

very term

"mixed

presents a very clear example of

type"

contradictio

in adjecto, for the intention of constructing a type presupposes abstraction of all elements of the non-typical. the inner raison

d'être of the type.

As research demonstrates, this reveals The history of linguistics illustrates

clearly an intuitive, yet nevertheless always perceptible, aspiration

to

108

G. A. KLIMOV

postulate

such

contradictions.

a notion of type, which would

be free

of

intrinsic

Thus, in the history of formal (so-called morphological)

typology of languages we find no example of the development of a mixed (e.g., agglutinative-inflectional)

system, and in the history of

content

typology we find no examples of the rise of, say, an accusative-ergative system. In view of what has been said, we may infer t h a t a combination of typologically

heterogeneous

structural

features

is

in

principle

a

phenomenon represented only in actually existing languages. Naturally, for adequate typological description of these languages we need a formulation of rules describing their deviation from the nearest type. conceptual

analogies

existing

between

the

theoretical

In view of

foundation

of

typology and the other main branches of comparative linguistics, we may conclude t h a t the notion of typological class of languages belongs to the same order of notions as the linguistic family (genetic class) and the Sprachbund

(areal class).

Assumption of the non-arbitrariness of the structural features of language type involves the question of the genuine systemic nature of the type.

Holistic ideas, as is well known, have developed in linguistics in a

way t h a t is far from straightforward, though the need for working out typology based on interlevel correlations was already expressed by von Humboldt.

But by the failure of some attempts to construct a holistic

typology, which resulted in the extraordinary stability of the traditional morphological classification of languages (which has solved this task only partly), led to long lasting popularity of the idea t h a t only what has been called subsystem typology might be built.

The conviction that creating

holistic typology is impossible received support from doubts concerning the legitimacy of lexical typologization, which are formulated now and then in linguistic studies. It seems, however, that refusal to deal with whole-system typology, embracing different

levels of linguistic structure, is at odds with the

commonly recognized principle of systemic organization of language.

Such

a refusal is not consistent with the ultimate aim of linguistics, to describe languages as systems consisting of subsystems. The fact t h a t the two other

ON THE NOTION OF LANGUAGE TYPE

109

main branches of comparative linguistics, genetic and areal--unlike modern typology-- deal with the whole language system, also deserves special attention. Therefore I agree with Skalička's opinion, t h a t

" a typology of

languages should observe and describe languages in toto, i.e., both in terms of content

and expression, form of expression and form

of

content,

morphology and syntax, vocabulary and phonology" (1983:281). I hold t h a t such an approach is also supported by the development of typology, established at the present on selected language structures, clearly demonstrating not only definite principles of lexical organization, but also their strict

agreement with the relevant syntactic

mechanisms of language.

and

morphological

Moreover, based on the theoretical assumption

(consecrated by long tradition in linguistics) of the primary role of lexical function and of the secondary role of the grammatical component (an assumption based on the greater depth of grammatical abstraction as compared to lexical), it is possible to conclude t h a t precisely in the realm of lexicon some fundamental traits of language structure reside, and that they cause the functioning of most of the phenomena observed at other levels.

This assumption receives empirical support from content-oriented

typological

inquiry,

demonstrated

that

in the some

course of which

syntactic

and

it

has

been

morphological

repeatedly

phenomena

of

language are strictly related to the organizational principles realized in the verbal and nominal lexicon. I do not wish to be misunderstood.

I willingly agree with the

possibility and usefulness of any structural comparison between languages within the limits of a single level. proceeding typological.

from

the

notion

But I fail to see why comparison not

of language

type

should

be

labeled

as

In my opinion, a comparison will remain typological only if

the features of the given level are really pertinent to the language type. For example, when studying assimilative processes affecting the consonants which represent

the personal prefixes in Northwest

Caucasian

verbal

structure, and their interaction with ergativity in these languages, we

G. A. KLIMOV

110

remain in the realm of typology. 1

But if we start to study the same

assimilative processes apart from the language type, such comparison will t u r n out to be contrastive rather than typological. In conclusion, something must be said about correlation between the language types, constructed in the framework of formally oriented and content-oriented typological studies.

Formal types are constructed on the

basis of some formal relationships existing between structural facts.

By

contrast, content-oriented types are constructed with obligatory orientation to the mental content expressed.

One of the most promising trends of

modern typology seems to be the further working out of content-oriented language types, according to how they render subject-object Research has demonstrated t h a t it is the expression of

relations.

subject-object

relations with its structural projections at different levels of the language system (lexicon, syntax, morphology, morphophonology) that is essential in determining the general shape of the system.

Meanwhile, the various

formally oriented types usually involve relatively restricted sets of related surface features, which often do not extend beyond the limits of a single level.

To my mind, content-oriented types possess the advantage in this

respect, for they reduce significantly the arbitrariness in the choice of typological

criteria,

and

(as one

might

expect)

are

able to

typological inquiry with a more definite historical perspective.

endow

This gives

grounds for hope that, as far as explanatory power is concerned, this trend of modern typology may achieve a status on a par with genetic and areal linguistics.

Cf. the assimilation of 1st person singular prefix s- in the Abkhaz transitive verb wordform i-s-gojt > i-z-gojt ['I carry it'], beside the absence of such assimilation in the corresponding intransitive.

On Typological Shift M. M. Gukhman

1. Identification

of typologically-marked

changes ("typological

shifts")

in

language evolution is linked to the problems of historical typology which have

recently

become

(Gukhman 1981).

an

independent

domain

of

typology

studies

However strict the selection criteria for objects of

historical typology, they include investigation of categories identical neither in scope nor in significance. In addition, much uncertainty still remains as to what processes are typołogically valent, what are the specific features of their content, and what is the extent of linguistic change caused by these processes.

In particular, it needs to be determined whether there are

changes which affect

language type (i.e., typological shifts),

entail

a

complex of implications, and determine language type. The diachronic aspect of typology is carried out by means of data obtained

in synchronic typology.

Identification

of initial

typological

categories and structures undergoing transformation is accomplished within synchronic typology. comparison

between

Only data obtained in synchronic typology allow qualitatively-marked

typological forms with

their

subsequent distribution along the time axis. Definition of language type by a complex of interdependent features or implications gives the shifts being modeled a complex polyaspectual nature. 2. In this paper, language type is to be understood as a multidimensional system of interdependent differential features which reveals the inner structure of language. Of the existing typological models, I will consider formal-structural and (to some extent) content-oriented models.

112

M. M. GUKHMAN Content-oriented typology is understood to be oriented toward the

content of language--in the words of Professor Klimov, "toward ways of representing [semantic] subject-object relations" (1983:38).

This is in fact

a concern of syntactic typology. Formal-structural typology is based on morphological word structure, and

in

its

morphological

external

parameters

classification

of

comes

languages

inflectional, and isolating language types.

close that

to

the

assumes

conventional agglutinative,

However, using word structure

to provide a typological basis for classification does not yield a simple indicator of language type, but

rather

a complex of

interdependent

differential features of varied relevance. These

typological

characteristics--inflection,

agglutination,

and

isolation--constitute only one, external (and therefore more noticeable) aspect of word structure. Meanwhile, these external features are related to indicators determining morphophonological word structure and the nature of morpheme boundaries and word divisibility principles; the correlation of elements like root, stem, and affix; inflectional paradigmatic types and the semiotic structure of grammar indicators (correlation of the content and expression

elements); and,

finally,

constructional

principles of

speech

segments, both simple and complex. Word structure might conventionally be assumed to contain, in a reduced form, implications of the whole formal-structural organization of a language, including the absence or presence of redundancy in marking inflectional categories, the possibility or impossibility of several grammemes being combined in one affix, accordingly, structure.

the

presence or absence of isomorphism

and,

in its semiotic

Eventually, morphological (formal-structural) types are related

to differences in language mechanisms which allow multiple representation of the same meaning.

However, opposing content-oriented typology to

formal typology is in itself rather arbitrary, since each has a complex of content and formal parameters. Yet they differ in their essence as well as in role relations within each typological model. In this connection, it should be emphasized t h a t multidimensional typological

models—

both

formal-structural

and

content-oriented--are

equivalent, as they model different aspects of the complex whole t h a t is

ON TYPOLOGICAL SHIFT language, complementing one another reflecting

the

various

ontological

objectively inherent in language.

113

and not excluding characteristics

and

each

other,

dependencies

This makes any discussion of the

advantages of one typology over others hardly worthwhile. 3. Application of multidimensional typological characteristics reveals that no "pure types" actually exist.

The prominent Czech typologist Skalička

viewed any language type as a construct designed by a researcher from observations of numerous concrete languages (1966). This may be roughly applied to any typological model: most often features of one type do not fill the whole linguistic area, which also contains categories of other types of the same class.

However, in the way this thesis is used by Skalička,

with no regard for the typological dominant invariably present in any language, it involves an incorrect absolutization of a correct observation. The typological dominant--generating the nature of language system--is its principal feature. Typological shift occurs mainly in categories constituting the typological dominant of any particular language. The identification of the

dominant

typological

underlies

leading

and

and

permits

characteristics,

central

(for

a given

differential

heterogeneous features combined in a language.

treatment

language) of

the

In certain instances,

however, the dominant itself can be difficult to identify. The typological homogeneity or heterogeneity and correlation of different

typological strata vary with languages, as differences in the

distribution of feature segments.

Thus,

characteristics are traced to separate

some

languages

display

construction of nominal and verbal paradigms.

typological

language

deviations

in

For example, Bulgarian,

unlike any other Slavic language, has essentially no case paradigm, with only a few relics found beyond the literary language—and those in petrified forms only.

However, in the verbal paradigm the old markers

preserved with some consistency.

are

This includes person, number and tense

grammemes (cf. the present-tense conjugation of 'write 5 : , , ,

,

,

).

paradigms have different indicators.

Besides, present

and

imperfect

A complex of simple verb forms

retains features of the inflectional type. A different typological stratum is

114

M. M. GUKHMAN

represented by a diverse system of analytic forms of preterite and future, consistent with the developed preposition and particle system in the noun. The typological dominant is hardly definable in this "entanglement" of typological features.

An essential objective in studying Bulgarian is to

determine a perspective for its development. Specific parallels are found in Indo-Aryan

languages.

Zograph

(1976)

points

to

differences

in

combinations of inflectional and analytic forms on the one hand, and agglutinative and analytic forms on the other. High homogeneity is found in Turkic languages, which consistently and unambiguously reflect a set of implications of the agglutinative type (see below), which forms a clear typological dominant.

These languages

approach an ideal model for the agglutinative type. Different correlations arise from studies of Finno-Ugric languages, particularly the Baltic-Finnic group, which has numerous traits of the inflectional type. 1 Here, too, the typological dominant cannot be defined without a thorough preliminary analysis of the correlation between different typological strata. 4. In most instances typological heterogeneity of a language, in a synchronic view of an old or a modern language, reflects the paths of its development, typological processes bearing trace of its earlier history. These data provide prerequisites for the reconstruction of typological shift. The potentialities of content-oriented and formal-structural typology are far from being equal. The possibility, and in fact the reality, of formalstructural changes is confirmed by historic and comparative-historic studies of many languages: Indo-Aryan, Iranian, Germanic, Semitic and still others, whose development is traced in written documents over many centuries. The task, therefore, lies not in assessing the possibility of typological shifts, but in studying the nature and direction of these changes as well as the degree of their commonness and predictability. In content-oriented typology a different situation is observed. Here typological shifts seem to progress more slowly. They are rooted in the

Characteristically, Professor Tauli (1966) does not assign Finno-Ugric languages to a "pure" agglutinative type.

115

ON TYPOLOGICAL SHIFT remote past and evade direct observation in written documents.

In

addition, more often than not the initial data are based on languages with no written history, and this calls for verification of the very thesis about the transformations of content-oriented types. The problem of typological changes in syntactic structures first arose with inquiry into whether Indo-European had developed from a nonnominative accusative linguistic type, although the issue was phrased in terms

differing

from

those

used in modern

typology.

Though

the

hypothesis concerning the past of Indo-European languages proved to be wrong, and Uhlenbeck's methods of investigation were clearly outdated, we should not forget t h a t it laid the groundwork for a trend in typology later developed in a series of studies by Meshchaninov, and retains a certain value to this day.

For many years data were selected, problems were

discussed,

generations

hypothesis

concerning

of

scientists

the

succeeded

non-nominative

one

period

another, of

and

the

Indo-European

languages (which would then have experienced a crucial typological shift centuries ago) found ever-expanding application. Of particular significance now is the typological reconstruction of the Pre-Indo-European language as active-stative,

presented

in

the

work

by

G.

V.

Gamkrelidze

and

V. V. Ivanov, which summarizes their long-term investigations (1984; cf. Gukhman 1981, chap. 4).

However, in spite of extensive research into

changes in content oriented types in a number of less explored languages, the most fundamental problems remain obscure.

Though applied to the

earliest history of Indo-European languages, after the publication of the study by Gamkrelidze and Ivanov, the thesis of the active type shift toward the nominative type is gaining wider recognition, and the problem of active-ergative and ergative-nominative relationships is still open to discussion, largely due to the ambiguity of the set of features identifying the ergative type.

These intricate questions naturally exceed the scope of

this paper and demand a series of investigations involving extensive data. 5. Other possibilities are revealed on the level of formal-structural typology. Here we can clearly see that the intensity of typological shift depends on language type. Comparison of developmental trends in inflectional

116

M. M. GUKHMAN

languages with a century-old writing tradition have shown the existence of similar tendencies not only between geographically distant Indo-European languages, but also between Indo-European and Semitic languages. Common

tendencies

are

caused

by

instability

of

important

implications of the inflectional type and a corresponding gradual decline of the old system of marking grammatical categories and processes.

The

results of these changes, though, are far from identical even in kindred languages.

However, only in some of the languages of this type does

restructuring equally involve all the segments of a language system and lead to a typological transformation.

This accounts for the coexistence of

different typological strata in Bulgarian and the Indo-Aryan languages mentioned above. In Germanic languages a major typological shift has occurred only in Afrikaans.

The collapse of the nominal paradigm resulted in the absence

of gender and case categories in the noun, with only the opposition of number retaining stability.

The adjective even lost agreement in number.

Afrikaans, unlike other Germanic languages, has not retained strong verb reflexes except for modal and auxiliary verbs. The verb paradigm is based on simple forms of the present and analytic forms of past and future devoid of person and number categories: the present ek val 'I fall', ons val 'we fall', hy val 'he falls', hul(le) val 'they fall'; the perfect ek het geval 'I have fallen', hy het geval 'he has fallen', etc.

The only grammatical

categories are number in the noun and tense in the verb (i.e., the so called non-syntactic categories, typical of isolating languages).

In word groups

juxtaposition is predominant, whereas agreement is no longer used.

The

typological dominant is isolation. In striking contrast to Afrikaans is modern Icelandic, which exhibits maximal retention of the differential features of the inflectional type.

The

noun system preserves variability in declension classes and instability of the root morpheme in the nominal declension due to numerous umlauts (cf. vollr 'field', vallr 'of the field' [genitive singular], vellr 'fields' [nominative plural]); it also retains the case paradigm, the three-gender system, a variety of strong verb classes, and synthetic forms of the preterite. word combinations agreement is predominant.

In

Typologically, Afrikaans

bears the greatest similarity to modern English, and Icelandic to German.

117

ON TYPOLOGICAL SHIFT

In Iranian languages, processes reflecting the decline of the old morphological type have been even less uniform, partially due to the formation of new synthetic structures of the agglutinative type based on integration of components in phrases by means of postposition, function verbs, and enclitics ("secondary synthetization," (Rastorgueva, 1979:157). While Germanic languages had the same direction of restructuring (though differing in the intensity of their synthesis-to-analysis transition), in Iranian languages the formation of a new stratum of synthetic forms alongside analytical constructions did not promote a complete and uniform transformation of the inflectional type. 6. Typological instability of some Indo-European languages has been ascribed to the influence of a complex of factors.

The first of these are the basic

features of the ancient grammatical structure of Indo-European, namely (1) numerous

subcategories,

often

rather

small,

in

nominal

and

verbal

paradigmatics, and (2) the ability of lexical stem to participate in both nominal and verbal paradigms due to vowel and consonant alternation types.

The latter accounts for the productivity of composite markers (cf.

in Gothic the composite integral marker in u-stem declension is nom. p1.

'sons', and sun-u-m

sun-jus

'to the sons', where bimorphemic case

indicators are formed by a stem vowel alternation + a case indicator). Separable markers of the verbal present/ preterite opposition in the same language—nim-i-s 'you take', nam-t

'you took'—include (a) root vowel

alternation -i/a-, (b) alternation in the stem-vowel -i-/ß, () the opposition of

personal

endings

-s/

-t

(which

led

to

redundant

marking

and

consequently weakened the semiotic function of the separate morph), and (d) the absence of isomorphism between formal structure and content units (i.e., morphemic and grammemic composition) in endings. This situation is typical of inflectional languages (cf. Gothic nominative singular feminine baurg-s 'town' where the simple marker -s is represented by one morph, but its structure combines number, gender, case grammemes determining the position of the form in the case paradigm). These

features

led

to

instability

in

the

paradigms

of

basic

grammatical categories, and were actually the "weak" link in the chain

118

M. M. GUKHMAN

which gave impetus to system transformations.

These processes were

additionally intensified or impeded due to effects of other factors, both external and internal (for details see below). The main changes which independently affected both Germanic and Iranian languages included the formation of regular paradigmatic series, a trend

realized

differently

in different

languages, where

it

sometimes

brought forth new typological dominants, t h a t is, triggered typological shifts (Afrikaans, and to some extent English). 7. Agglutinative languages like Turkic have been described in the literature as rather stable typologically.

This is manifested not only in the retention of

earlier modeling principles in inherited word forms, but also in the fate of the innovations in case plus postposition sequences in the noun system, and groups of nominal verb form or adverbial participle plus function verb in the verb system. The use of postposition in the formation of case affixes is a trend widely represented in various languages and is realized in different periods of their history.

The model for these word forms is characterized

by postposition of a function word or enclitic and by the stress type, which brings

about

innovations

intensifying

structures in agglutinative languages.

the

productivity

of

synthetic

Similar processes occur in the verb

system, comprising nominal verb forms and adverbial participles plus a function verb.

Blending results in various forms of integration of stem-

final phonemes in conjugated verbs and initial phonemes of the second component (cf. in modern Uigur alar edim simplification

of

complex

forms

>

manifested

allatim). in

shifting

This entails morphemic

boundaries between the components, and it leads to their integration, due to assimilation processes changing the original configuration of initial patterns.

This development of analytic constructions in languages is

dominated by agglutinative principles of inflectional composition in forms. In addition, it is symptomatic that the differential feature above to characterize the formal-structural

cited

type of the ancient Indo-

European languages is absent or peripheral in this case: Turkic languages have no variations in inflectional paradigmatics. "the

root

morpheme

is universal

As Shcherbak puts it,

(one declension,

one

conjugation)"

ON TYPOLOGICAL SHIFT

119

(1977:17). In contrast to ancient Indo-European languages, there is no redundance in marking grammatical categories. This is linked to pervasive isomorphism between the formal and content structures of inflectional markers, between morphs and grammemes--in Shcherbak's definition, "one morpheme--one meaning" (ibid.). Thus the Turkic inflectional paradigm structure lacked prerequisites, such as those which caused the instability of the Germanic or Iranian type. However, Turkic languages do not rule out isolated instances of inflections with different features, e.g., in Azerbaijani verb forms kelir-ə 'I come', kəlir-sən 'you come'; and the inflectional markers -em, -sən include not only the person grammeme, but also the number grammeme (cf. indicators of the first and second person plural: -ik, -siniz). Consequently, lack of integration of grammemes in one morpheme is not the only norm in Azerbaijani and other Turkic languages, but it is a typological dominant. The distinction between inflectional and agglutinative languages lies in the part each pattern plays in their structure. Besides, different periods in the development of Turkic languages have been marked by the rise of separable complex constructions of the analytic type, eventually to become synthetic wordforms (see above); this proves the predominance in Turkic languages of a synthetic type and a synthetic typological dominant with the basic nucleus forming agglutinative patterns. No typological shift occurs in this instance. 8. Further description of formal-structural changes in Indo-European languages requires consideration of why the rates of development and depth of transformation have not been uniform in the various languages of this family. The intensity and depth of typological changes in Germanic and Iranian languages seem to be caused also by changes in the accent type which weakened final syllables, that is, primarily inflectional markers. The relation of native to borrowed lexicon in language development is also of considerable importance. In this respect the unique isolation of the lexical system in Icelandic is of some interest: no words are imported even from international vocabulary. Here probably lies the reason for archaic paradigmatics in Icelandic. Presumably, this assumption receives support

120

M. M. GUKHMAN

from comparison with the history of English, whose lexical system has twice

undergone

crucial

changes.

Under

certain

conditions

massive

borrowings cause penetration into the language system of forms which might be more or less at variance with the word structure of a given language. 9. Investigation of the lexical factor brings into focus the problem of the effect of external stimuli, in addition to significant intralinguistic factors (see above), on typological changes of languages.

A possible development

for Bulgarian due to this kind of stimulus may be its involvement in the Balkan

language

Albanian).

community

(it shares features with Romanian

and

In the development of English, the influence of Scandinavian

and Anglo-French bilingualism cannot be ignored.

Finally, in Afrikaans

the process was obviously stimulated by a complex of external factors earlier analyzed in detail by S. A. Mironov (1969). Thus, typological shifts are a result of the interaction of internal and external stimuli.

The problem of the resistance and stability of the

inflectional type calls for special investigation. There is a possibility t h a t shifts of formal-structural types may be cyclical:

in

Iranian

the

disintegration

of

inflectional

patterns

and

intensification of analytic trends are succeeded by secondary synthetism with elements typical of agglutinative languages.

A presumably Indo-

European inflection has been formed in a language where agglutinative elements were combined with isolating elements.

If the assumption is

correct, this is an example of formal-structural type cyclicity. example is found in the history of Chinese. As

the

set

of

formal-structural

Another

2

types

transformational possibilities of those types.

is

limited,

so

are

the

Therefore, it is only natural

t h a t similar changes occur in the history of different languages.

Since

absolutely " p u r e " types do not occur in reality, and in every language elements of another formal-structural type coexist to some extent with the typological dominant, possibilities of shift are inherent in their very status.

'See the works listed in Gukhman 1981:149, and Solntsev, this volume.

ON TYPOLOGICAL SHIFT

121

These transformations, however, form no consistent unidirectional constant. Nor are they related to language development, inasmuch as the effect of external stimuli is an additional reason preventing the construction of universal patterns.

Discourse Function and Word Order Shift: A Typological Study of the VS/SV Alternation P a u l J. H o p p e r

1. T h e V S O / S V O / S O V T y p o l o g y : S o m e G e n e r a l R e m a r k s The present paper grows out of a concern to integrate more closely those generalizations which are made about the typology of word order with some recent work on the relationship between word order and discourse pragmatics. 1 The former, t h a t is, the study of word order universais and word order types, derives mainly from the work of Joseph Greenberg, especially his article of 1966, and subsequent concerns to refine this, including Lehmann 1972, 1978, and Hawkins 1983. The bibliography of work on the pragmatics of word order is by now quite considerable, and includes the papers in Li (ed.), 1975, and numerous studies of individual languages.

A

number of works on word order pragmatics have moreover attempted to combine pragmatics with the word order typology, e.g., Hopper 1971; Givón 1975. The necessity to combine the study of word order types

with

discourse pragmatics emerges from two considerations. One is the problem of the notion "basic word order."

Despite Hawkins' optimism (1983:12 et

pass.), typological generalizations depend on a uniform notion of "basic sentence type" which is far from easy to establish either cross-linguistically

The research and writing of this paper were completed during a Guggenheim Fellowship held in 1985-86. I am grateful to the John Simon Guggenheim Foundation and the State University of New York's Research Foundation for support during this period.

PAUL J. HOPPER

124

or within a single language.

In extreme cases an extensive study of texts

reveals no consistent pattern of word order which could be called "basic" in a syntactic sense, and it becomes crucial to study the discourse uses of various word order types.

Brody (1984) has suggested t h a t

although

individual languages may possess a "basic word order," the discourse function of this basic word order is not consistent from language to language, and may not correspond to any simple criterion such as textual frequency, ease of processing, or lack of ambiguity.

This conclusion, too,

would point to the possibility of broad textual conditioning for word order, and

would

vitiate

any

hope

of

postulating

a

"basic"

word

order

independent of context. In Hopper

1971 a semantic criterion for basic word order was

proposed, which was based on Prague

School syntactic

theory

and

influenced by the work of Wallace Chafe on new and old information (1970). This criterion can be worded as follows: The Basic Sentence Type in a language is t h a t in which the Subject is old or topical, and the focus of new information falls on the Predicate. The typological class membership of a "Free Word Order" (FWO) language would then depend on the word order in the basic sentence type thus defined, and it would no longer be necessary to consider adding a special type of F W O languages. Nonetheless the semantic criterion has the defect t h a t it ignores the fact that the semantic distribution of information in a clause is itself derivative of discourse, and that it is only in a specific discourse context t h a t it makes sense to speak of "new"

and

"old"

material, or of "theme" and "rheme." A second motivation for considering pragmatics in the discussion of word order typology is the need to account for change.

Admittedly this

necessity derives in the first place from a pragmatic view of word order. If word order is viewed as a syntactic structural phenomenon, then change might conceivably be viewed as a set of ordered adjustments, each of which occurs as the language strives to re-establish typological equilibrium. My own view is t h a t "grammar" is not as systematic and homogeneous as the syntactic-structuralist view would hold.

In fact I would see what is

DISCOURSE FUNCTION & WORD ORDER SHIFT

125

called "grammar" as a rather random and disconnected set of regularities which have emerged from discourse over time and are continually in a process of sedimentation (grammaticization), fossilization,

and

perhaps

disappearance; some further discussion of this view is presented in my paper Hopper (1986 [to appear]). This view is at odds with any attempt to study a fixed "synchronic" cross-section of

a language.

Yet

the study

of syntactic

universais

customarily presupposes sets of fixed, discrete organic structures which may be compared, in other words, sets of static synchronic grammars. The dilemma can best be circumvented by stressing the continually changing nature of "structure" and seeking to classify languages not so much as co-existing synchronic entities ("systèmes as

individual

collections

of

sub-systems

où tout se tient")

at

different

but

stages

of

grammaticalization and located at points along the sorts of historical trajectories which Traugott has proposed (e.g., 1982).

Such a project

involves considerable labor, with attention being paid to the typology of change (i.e., the nature of the trajectories themselves) as well as to the processes by which change is implemented in individual languages. It may then

turn

out

restructuring epiphenomenon

of

that

"typological

linguistic which

shift,"

systems

synchronically

i.e.,

along can

the

supposed

universal at

best

lines,

only

be

global is

an

vaguely

characterized, and then only as a set of possible outcomes of general diachronic tendencies. It follows t h a t there is no alternative in typological studies to a careful language-by-language study of textual occurrences of word order (this same point is made by Brody 1984).

Isolated and decontextualized

sentences, even when supplied with a semantic template of the Prague school variety, as suggested in my Semantic Principle, have only a limited validity in typological studies.

It is only from discourse t h a t we can tell

what is "new," what is "old," what is "contrastive," and so on; without textual analysis, the data base for a typology is suspect. In this paper some typological aspects of verb position are discussed primarily with textual data from Malay.

A comparison is made with

earlier Germanic, but the study is limited in both the range of language

126

PAUL J. HOPPER

types discussed and the size of the sample. The basicness of verb position is insisted on despite Hawkins 1983, where the claim is made t h a t syntactic typology should be grounded in adnominal groups (noun and adposition, noun and genitive). Verb position and verb morphology appear universally to form a crucial link between discourse function and clause structure, and are therefore a more realistic basis for a typology t h a n the relatively trivial syntax of nominal groups. 2. S o m e F u n c t i o n a l - G r a m m a t i c a l P r e l i m i n a r i e s Before proceeding with the diachronic-typological examination of Malay which forms the core of the paper, I will briefly discuss some of the grammatical concepts which underlie it. 2.1. Preferred Argument Structure Syntactic investigations have usually proceeded from the study of random isolated sentences. Frequently, such sentences represent only a minority of the clauses and sentences which actually occur in discourse.

The recent

work of J. Du Bois (1985) shows that actual live discourse is made up of a small number of clause types, constructed with rather limited means.

The

most common of these, the "Preferred Argument Structure," consists of a verb with either zero or one lexical noun argument. positions not

represented

by

a lexical noun

may,

The

argument

according

to

the

individual language, be filled by atonic pronouns, clitics, or pronominal affixes, or by zero.

The important point is that clauses with more than

one lexical noun are considerably less common than clauses with only one lexical noun or no lexical nouns at all.

In his 'investigations of spoken

French, Lambrecht (1985, to appear) has also drawn attention to what he refers to as the "Preferred Clause Unit," consisting of a transitive verb with clitic pronouns and one or no lexical arguments. DuBois notes that the one overt (i.e., lexical) argument in this preferred clause type is usually the verb's object; lexical transitive agents are

relatively

rare.

Another

favored

role for

lexical

nouns is the

"presented" (i.e., new) subject of an intransitive verb. It is, of course, no coincidence t h a t these two discourse roles—transitive object and intransitive subject--repeatedly constitute a grammatical class, that of the Absolutive. Most significantly, too, in a majority of ergative languages the ergative/

DISCOURSE FUNCTION & WORD ORDER SHIFT

127

absolutive case system is restricted to lexical nouns, while pronouns tend to follow nominative/ accusative case marking. 2.2. V S / S V A l t e r n a t i n g L a n g u a g e s This term was coined by Myhill (1985) to refer to the widespread type of language whose basic word order cannot be said to be fixed as SVO or VSO, but alternates between the two under discourse conditions. This alternation had already been studied for Spoken Hebrew by Givón (1975), and in Greenberg 1966 the existence of alternative orders in both

Type I (VSO) and Type II (SVO) languages had been

noted.

Lehmann (1972, 1978) had noted the consequence that the position of the Subject was irrelevant to the typology, and indeed pointed out that the "Subject" was in general a dispensible element, leaving only the two basic types

VO

(combining

Greenberg's

(Greenberg's Type III).

Type

I

and

Type

II)

and

OV

Myhill's work points to a syntactic functional

continuum within Lehmann's VO Type, the "VS/SV Alternation." In view of the work of DuBois discussed above, it is evident t h a t the alternation concerned is in fact one of preverbal versus postverbal lexical nouns, i.e., V N / N V , the N normally standing for the Absolutive N P .

In

order to maintain continuity with previous work on word order typology, however,

the

established

terminology

involving

S,V,

and

employed rather than the more accurate terminology, N,V. noted,

however,

that

(1)

the

natures

of

Subject

and

 is here It should be Object

are

problematical where ergative language are involved, and (2) S and  generally refer to lexical nouns and autonomous pronouns. (The notations VS and OV, however, conventionally refer to Verb-initial and Verb-final clauses respectively.) 2.3. Categoriality The notion of CATEGORIALITY was developed in Hopper and Thompson 1984. are

According to this theory, the lexical categories N O U N and VERB not

a priori

grammatically

fixed,

instantiations of proto-typical functions.

but

are

discourse-constituted

For verbs, this function is to

report a new event of the discourse. To the degree t h a t a linguistic form is carrying out t h a t function, it will reflect the grammatical characteristics (morphological and syntactic) of a verb in t h a t language. For nouns, the

128

PAUL J. HOPPER

prototypical discourse function is to identify a new participant of the discourse.

Forms which carry out this function

accompanied

by

whatever

morphosyntax

are predicted to be

grammatically

characterizes

nouns. 2.4. T r a n s i t i v i t y TRANSITIVITY is understood here as a composite characteristic of a clause, only one component

of which

is that

which

figures

in

the

traditionally accepted definition, namely the presence of an Object (the "valency" of the clause). The theory of Transitivity and its associated discourse dimensions was presented

and

discussed in Hopper

and

Thompson

1980.

The

following ten parameters which contribute cumulatively to the Transitivity of a clause were proposed: 1. Valency, presence of an agent and an object; 2. Kinesis, denoting an action or event; 3. Telic aspect, i.e., sense of a completed goal; 4. Punctuality, denoting a sudden action; 5. Volitionality, denoting a deliberate action; 6. Affirmation—an affirmative clause; 7. Modality—the clause is in the realis mode; 8. Agency—the agent is human or otherwise autonomous; 9. Object affectedness--the object is changed in some way; 10. Object individuation, i.e., definiteness-referentiality. Transitivity

interacts

with

Categoriality,

the

Preferred

Argument

Structure, and the V S / S V Alternation in similar ways across a number of languages.

The one which will be considered in the greatest detail here is

Malay; reference will also be made to Indonesian (Bahasa Indonesia), a modern development of Malay. 3. Malay In early 19th century Malay texts assumed to be fairly close to vernacular

129

DISCOURSE FUNCTION & WORD ORDER SHIFT

Malay (specifically the Hikayat Abdullah [1848]), we find the following clause types: 3.1. Clause Types The most important clause types in Malay are here given. -lah

is a discourse

particle

denoting,

roughly,

The particle

"event".

The

gloss

PASSIVE, or PASS., is suggested without prejudice as to whether the form is actually Passive or in fact Ergative in a particular instance. 3.1.1. Preferred Clause Structure (i) Clitic ^V Clusters: di-ambil-nya-lah "he took (it)"

-tr.verb-3p.AGENT-LAH

ku-ambil-lah "I took (it)"

1p.AGENT-tr.verb-LAH

(ii) V^ Noun: datang-lah sa-orang nakhoda "a sea-captain came"

intr.verb-LAH^AGENT

ku-unjokkan satu rupiah lp.AGT.-tr.verb " P A T I E N T "I handed over one rupee" 3.1.2. Clauses w i t h Lexical A g e n t and Patient ERGATIVE: di-ambil/oleh pawang itu/sadikit batang pisang PASS.-tr.verb^AGENT^PATIENT "the elephant-magician(2)/took(l)/a bit of banana stalk(3)" ACTIVE: monyet/men-dapat/bunga AGENT^ME-tr.verb^PATIENT "a monkey/gets hold of/a flower"

130

PAUL J. HOPPER

PASSIVE: aku/di-jualkan/oleh ibu-ku "I/was sold/by my mother" Almost

invariably,

PATIENT^PASSIVE-tr.verb^AGENT

in Malay these basic clauses are preceded

by

a

conjunction and, usually, an adverbial of some kind, such as an expression of time or place. By far the commonest of these is maka,

translatable as

"and" or "then." The clause types which contain a lexical patient are classified along the following lines (Hopper 1983): 1. ERGATIVE: The Verb has passive morphology, patient follows the verb., i.e., the clause is VN.

and

the

2. PASSIVE: The Verb has passive morphology, and the patient precedes the verb, i.e., the clause is NV. 3. ACTIVE: The Verb has an active prefix such as meng- or ber-, the patient follows the verb; this type of clause usually has a lexical or autonomous-pronoun agent, which precedes the verb, and the word order is therefore NV(N). The passive morphology is illustrated in the examples above.

It

consists in the prefixation of di- (glossed as "PASS") to the verb stem if the agent is understood to be 3rd person, and the prefixation of the clitic form of the pronoun if the agent is 1st or 2nd person.

The third person

agent has the preposition oleh if a lexical noun, and is enclitic -nya on the verb if non-lexical. Ergative and Passive thus share a similar morphology, and Active and Passive share a similar word order.

(Even so, Passive and

Ergative differ in more t h a n simply word order; for example, the Passive, but not the Ergative, may lack an indication of the agent.) The discourse functions of these various transitive constructions have been discussed in Hopper 1979a, 1983: the Ergative serves to "foreground" events, especially perfective, sequential ones, while the Passive and the Active are "backgrounding"

constructions, functioning to hold up the

action in order to describe things, states, or repeated occurrences.

and

on-going

The Active in particular has functional affinities with the

construction known as Anti-Passive in a number of ergative languages, as

DISCOURSE FUNCTION & WORD ORDER SHIFT

131

has recently been shown by Polinskaya (1984). In Hopper 1978 [1977], and 1979a a close functional

correlation was described between the focus

structure of clauses and event structure in the discourse, such that the pragmatic sequencing of events universally implicated an identity of agent across the sequenced clauses and new information in the predicate, while in backgrounding newly introduced information was more likely to appear in the topic/subject.

This distribution is of course compatible with the

division of labor between the Ergative on the one hand and the Passive and Active on the other.

It has important consequences for the link

between

Categoriality,

Ergativity,

Verb

Transitivity,

and

Preferred

Argument Structure to be discussed below. 3 . 2 . T y p o l o g i c a l Shift in M a l a y In the early 19th century language, the situation in Malay is generally as outlined here. The Ergative narrates sequential events with a high degree of topic continuity (Hopper 1978 [1977]), while Actives and Passives supply explanatory material: m a k a sa-bentar sa-bentar and from time to time surat itu, letter the

di-ambil-nya PASS take A G T .

di-renong-nya, PASS. stare:at-AGT.

di-letakkan-nya, demikian-lah PASS.put:down A G T . such LAH Maka sa-hari-hari and daily

adat-nya ia habit his

kemudian then laku-nya. behavior his

berkereta pada petang-petang; he go:driving on afternoon

maka pada hari itu sampai malam kereta menanti di-pintu, tiada i and on day that until evening carriage ME:remain at door not he mau t u r u n dari rumah-nya... want go:down from house his and this was his behavior: every now and then he took the letter, stared at it, and then put it down again. It was his custom to go for a drive every day in the afternoon; but on

132

PAUL J. HOPPER t h a t day his carriage remained at the gate, and he did not leave his house (A 86) In this passage, which is quite typical of narration at this period, pre-

verbal lexical nouns (e.g., kereta 'carriage') and autonomous pronouns (i 'he') are found in clauses which do not actually continue the action, but which explain, describe, contrast, justify, etc. the actual reported events. Later in the 19th century, the discourse-functional distinctions among the three constructions can be seen to be breaking down. Rafferty (1985) shows t h a t statistically during this period, and into the 20th century, transitive clauses regardless of their discourse status tend increasingly to be formulated in the Active, i.e., with SVO word order and the Active prefix meng- if this is possible.

This encroachment is largely at the expense of

the Ergative, which at least in the written language becomes an archaism. Since the Ergative was the more highly Transitive construction, in the composite sense discussed above, the trend may be characterized as one in which the Active increases in Transitivity, and the language as a whole becomes increasingly SVO and decreasingly VSO, though there is of course no point at which the transition may be said to have been completed. One

manifestation

of

this

trend

toward

development of copular and possessive verbs.

SVO

syntax

In the spoken

is

the

Malay

vernacular, copular clauses are generally expressed without an explicit verb equivalent to 'to be', and possession is indicated either by the existential ada 'there is/are' or by the form punya. empu-nya

This punya

derives from

'its owner', and the original syntax was (and often still is) of the

form rumah

ini saya punya

'this house I [am] its-owner', requiring the

possessed object to be a definite topic.

Nowadays, punya has long been

reanalyzed as a verb: saya punya rumah

'I have a house'.

Purists

Malay"

(i.e.,

Europeanized), and likewise the use of punya as a genitive case saya

punya

rumah

usually

dismiss

this

usage

as

"Bazaar

'my house' (with its Genitive-Noun word order!).

In the written

language, however, punya

is now decked out with full verbal trappings:

mem-punya-i

is

mempunyai

(where rumah

-i

a

low-grade

'I have a house'.

transitivizing

suffix):

Saya

Also characteristic of the written

language is the development of what was originally a resumptive pronoun

DISCOURSE FUNCTION & WORD ORDER SHIFT

ia-lah as a copula and the use of the verb menjadi equivalent

of Western-style

'to

be'; cf.

133

'become' as the exact

Hopper

1972:129-140.

The

development of such copular and possessive verbs has the effect

of

increasing the text frequency of SVO sentences. 3 . 3 . M o r p h o l o g i c a l C o r r e l a t e s of t h e V S - t o - S V Shift? Ergativity is correlated with basic word order in t h a t apparently only VS and OV languages (Greenberg's Type I and Type III) may be ergative. We might therefore expect t h a t when word order shifts from VS to SV, a concomitant shift from ergativity to accusativity will also begin to manifest itself. It is unfortunate that Ergativity has until quite recently rarely been studied from a discourse perspective.

There are many facts in individual

ergative languages which suggest t h a t discourse-based investigation would be rewarding.

In Abkhaz, for example, certain of the subject prefixes on

the verb are dropped if the subject immediately precedes the verb (Deeters 1963:58-9; Boeder

1979:474).

This

evidently

involves

loss of

overt

indicators of ergativity when the verb is moved out of its clause-final position into the post-subject position--a curious parallel to the Malay situation presented here. Unfortunately, neither of the sources for Abkhaz just mentioned supplies the all-important discourse data which might explain it.

There is a need for greater attention to trans-sentential

approaches to the phenomenon of ergativity; cf. Hopper 1979b. It seems to me, too, that by listing morphological paradigms of the "ergative inflection" in such-and-such language, we predispose ourselves to thinking of ergativity as a grammatical given, an "a priori"

fact about a

certain language, rather than as a possibility for arranging called for by discourse.

arguments

In this regard, the current and forthcoming work

of Du Bois is to be viewed

as a very significant

advance in

understanding both of ergativity and of the nature of "grammar"

our in

general. 3 . 4 . C a t e g o r i a l i t y a n d T y p o l o g i c a l Shift It was noted above in the section on Categoriality t h a t linguistic forms which fall away from

their prototypical functions

will tend to

morphosyntactic markings characteristic of the two prime categories.

lack

134

PAUL J. HOPPER

One of the commonest marks of the Active verb is the prefix meng(which has several allomorphs; in the citations below it is glossed as ME). There are some indications that in the earlier texts, the verb with this prefix, which almost invariably is preceded by its subject, is in some sense "less of" a verb than the Ergative di-form (whose order is, of course, VSO). Verbs which carry the prefix meng- often have the same distribution as nouns, i.e., are nominalized verbs; they may be objects, subject/topics, etc., and may have the demonstrative itu: Sa-bermula now aku I

makaada-lah and it: w as

kira-kira about

delapan sembilan bulan eight nine months

bertekun mengaji dan menulis itu.... occupied ME:study and ME:write the

tengah mid

hari day

lepas-lepas after

Makaada-lah and it:was

waktu time

mengaji itu... ME:rstudy the

Now it happened that for about eight or nine months I was occupied with studying and writing. . . And at about midday, after studying was over. . . (20) In a recent article (Myhill 1985), John Myhill has shown t h a t among languages which display the alternation of VS/SV, it is possible to show that languages which are "truly" VS, i.e., Greenberg's Type I language, require t h a t if a noun is placed in front of the verb (in other words, if the clause has the structure SV), the verb will appear in a nominalized form. Languages which shift their typological structure from VS to SV undergo a progressive

"verbalization"

of the verb

in SV clauses, t h a t

is,

the

morphology of the verb changes from a nominal to a verbal one. Myhill's work can be interpreted

in terms of the

Categoriality

Hypothesis (Hopper and Thompson 1984) to point to the conclusion t h a t as a language moves from VS and increasingly accepts SV as an unmarked word order, we can expect to find: (1) t h a t SV clauses INCREASE IN TRANSITIVITY. This increase may involve a loss of constraints against transitive morphology in SV clauses, or a loss of transitive morphology peculiar to VS clauses. Either of these developments would, of course, have the effect of increasing the average transitivity of SV clauses.

DISCOURSE FUNCTION & WORD ORDER SHIFT (2) that the V in SV CATEGORIALITY.

clauses

INCREASES

IN

135 VERB

The increase in verb categoriality in SV clauses may similarly result from a loss of constraints against typically verbal morphology in SV clauses, or from the decay of verbal morphology peculiar to VS clauses. Myhill's work suggests t h a t there are indeed generalizations which can be made about SV as opposed to VS languages, although they are of a more subtle nature t h a n gross observations about word order alone, since they involve also questions of pragmatics and morphology.

It is perhaps

premature to dismiss the verb-based typology in general, and the V S / S V (Greenberg's Type I / Type II) distinction in particular. 3.5. Reporting and Predication One way in which the difference between VS and SV clauses in Malay might be characterized is to say that SV clauses make

predications,

whereas VS clauses report or narrate. Viewed in this way, the Active, in which a lexical agent precedes the verb, and the Passive, in which a lexical patient precedes the verb, would be noun-oriented, or topic-oriented, while the Ergative, with its Preferred Argument Structure and its affinity for foregrounding, would be eventoriented; see Hopper 1983.

This discourse functional distinction is quite

well preserved in the early 19th century texts, but breaks down in the later period. The distinctions of clause types, and ultimately the morphological distinctions also among Active, Passive, and Ergative, depend on what must have been a subtle distinction between "reporting," or "narrating," an event, and "describing" an action. It is essentially the same distinction analyzed by Weinrich in his book Tempus:

besprochene und erzaehlte

Welt

(Weinrich 1964). Distinctions of this kind, which depend on a richness of shared context which is difficult to achieve across ethnic and cultural boundaries, are extremely vulnerable to subversion through the sort of external domination which increased in Southeast Asia during the 19th century (and which Abdullah himself documents).

136

PAUL J. HOPPER

4. G e r m a n i c Germanic provides an instructive parallel to Malay both in regard to the discourse functions of the V S / S V alternation and the change from a period of alternation

to one of highly grammaticized

SVO

syntax.

(This

discussion is based on Hopper 1975 [1967], in which earlier research on this topic is also described, and on Hopper 1979a.) In Old Norse the alternation of SV and VS displays close functional parallels

to Malay:

Lively

narration,

foregrounding,

implicates VS,

backgrounding--both local and global—is accomplished strictly with SV; these are Heusler's bewegte Stellung 1921:173-182).

and Ruhestellung

respectively (Heusler

Essentially, as Heusler shows, these are for Old Norse the

only possible word orders. In Old English a wider variety of word orders is available, and the general picture is less similar to Malay, largely because the earliest Old English prose documents antedate the earliest Norse monuments by several centuries; we find English therefore at an earlier stage t h a n Norse. In the oldest narrative prose, for example the early parts of the Anglo-Saxon Chronicle, verb-final alternation

between

clauses are common,

and

a

discourse-functional

all three clause types can be discerned

(Hopper

1979a:220-226): 1. SV syntax is typical of backgrounding. 2. VS (in prose the verb is invariably preceded by a particle or adverb) is typical of the initial event in an episodic series, usually sharing the same topic. VS correlates very strongly with a change of topic, and therefore can be said to initiate a topic-chain. 3. OV is "interior," being found with both episode-internal events and subordinate clauses. The OV type of clause in Old English displays a number of parallels to the Malay Ergative (Preferred) clause type, including the phenomenon of clitic-clustering at the beginning of the clause, and an affinity for transitivity. The later history of English syntax is dominated by the spread of the verb-medial (SVO) type at the expense of the other two types, and its eventual grammaticization

in the modern language, which is already

evident in the Middle English period (cf. Fries 1940).

DISCOURSE FUNCTION & WORD ORDER SHIFT

137

The same trend can be seen in Old English as in Malay: SV clauses occur with increasing indifference to the original distinction in grounding and transitivity.

T h a t SV clauses were originally lower in Transitivity is

suggested by the following statistic (Hopper 1979a:225). Saxon

Chronicle,

a

selection

of

typically

durative

In the Angloverbs

(such

as

"remained") were found to occur 8 5 % of the time in SV clauses, while a selection of typically punctual verbs (such as "slew") occurred only 41.4% of the time in SV clauses, being found otherwise in VS or OV clauses. This distribution is symptomatic of lower Transitivity in the composite sense for the SV clause type. English appears to be the only Germanic language to have undergone grammaticization of SV word order in all clause types.

The

other

languages have either retained the V S / S V alternation (Icelandic), or have modified it with a verb-second constraint (German, Danish).

Standard

German, of course, has moreover grammaticized verb-final

syntax in

subordinate clauses, thus codifying an old tendency in Germanic as a whole. 5. G e n e r a l D i s c u s s i o n I will conclude by offering some speculative conclusions based on what is admittedly at this stage rather limited evidence. The change in the direction of grammaticized SVO word order can occur through a replacement of a highly Transitive VS clause type by an originally less Transitive SV clause type if the two clause types are distinct morphologically as well as syntactically. This has occurred in Indonesian. In languages in which the different clause types are distinct solely through

word

order,

and

not

through

a concomitant

morphological

difference, it may also occur through a progressive fronting of various types of N P without morphological replacement. Both of these "strategies" for fixing SVO word order have the effect of producing a gradual textual preponderance of SVO word orders, which may be reinforced through the introduction of copular and possessive-verb constructions. English,

or

commands.

The older word orders are either eliminated, as was OV in reserved

for

specific

speech

acts such

as questions

and

138

PAUL J. HOPPER It is important in the evaluation of typological shift in verb-position

not to assume

any inherent

"teleological"

direction, no matter

complex and subtle the conditioning factors might be.

how

No principle

dictates t h a t word order will change, e.g., from SV/VS to SVO.

To the

contrary: the V S / S V continuum appears to be capable of existing over long periods of time in perpetual flux, with only slight but not necessarily consistent movement toward stabilization in either direction.

Where SV

stabilization occurs, it can very often be seen to be the result of language contact and creolization.

It is likely t h a t this is precisely what

has

happened in Malay/Indonesian and English. One can even envisage the sorts of social factors which might be conducive to such a development.

The nature of contact between (for

example) the Norman French ruling class and the Anglo-Saxon serfs, or Dutch colonial overseers and subject societies in Indonesia, involved a certain social discontinuity, even alienation, and an unpredictability with respect to the content of communication. break-down traditionally

of the

intersubjectivity

stratified

society.

In

There is, in other words, a

characteristic an

extreme

of a

homogeneous,

it

probably

is

no

exaggeration to say that a radically different conception of "language" is involved from the situation in which speakers who mutually inhabit the same language are vocally socializing. Certainly a sharply different kind of hierarchy is present. The change involved may be seen as one between an expressive and a referential use of language; but of course, as Bakhtin showed, the "referential" use of language is itself a hegemonizing one (see White 1984). This greater unpredictability would surely have manifested linguistically in a higher proportion of new-information (explicit

meaning-bearing

morphemes),

shorter

and

more

lexical

itself nouns

peremptory

utterances, and a considerably greater explicitness and decreased reliance on context in general.

In my discussion of Malay above, I have tried to

show how SV word order types (Active and Passive) are more nominal (predicative, identificational), while the VS type (the ergative or Preferred Clause Unit) is more verbal (reportorial)—reference to participants being carried on with clitics and even zero anaphora, but not with full lexical

DISCOURSE FUNCTION & WORD ORDER SHIFT nouns.

139

In the situation of social discontinuity which I have described

above, the language is broken down, and its (holo-)phrasal nature (the Preferred Clause Unit, with its intrinsic affinity for ergativity but also other ready-made components of discourse, such as formulas and proverbs) replaced by a more analytic structure in which Verbs and Verb Phrases become predicative of Noun topic/subjects.

The language in a very real

way comes to symbolize the social disruption caused by the external forces at work on it. The influence of written languages should also be considered an important

factor.

Written

language

permits more flexibility

"tracking" of participants (hence a slackening of the rigid

in

the

anaphoric

conventions necessary in the spoken language for marking continuity or discontinuity of topic), and also a greater degree of abstraction, both of which are conducive to changes in topic and an increased use of lexical nouns.

In this sense also the shift toward SVO structure may reflect an

increasingly logocentric ontology. Even with contact-induced change, however, we should not assume that the syntactic changes will be abrupt.

On the contrary, there will

usually be no apparent textual discontinuity, but rather we should expect to find a gradual, and perhaps grudging, extension of SVO syntax from pre-verbal lexical noun to tonic pronoun and eventually perhaps to a re­ ordering

such

that

enclitics

become

proclitics,

(probably

through

cliticization of the older autonomous pronouns), precisely to the degree t h a t the new social order is accepted by the expanded—though precariously balanced--language community.

To a large extent this shift will reflect an

already-present possibility in the language. Thus the "creolization" theory of the development of English is often (wrongly, I believe) rejected on the grounds t h a t the same changes noted in the post-Conquest era have predecessors in the pre-Conquest period. 6. C o n c l u s i o n In this paper, I have suggested that a word order class should be investigated in which the clause-initial verb is in a discourse functional alternation with a noun-verb type of clause, that is, a V S / S V alternation. Languages of this type lend themselves to a number of important grammatical and diachronic generalizations:

140

PAUL J. HOPPER 1. The VS clause is more dynamic and foregrounded, while the SV clause is backgrounded. 2. The VS clause is shown to be higher in Transitivity in the compound sense defined by Hopper and Thompson (1980), and the SV clause is correspondingly lower in Transitivity. 3. The VS clause, moreover, reflects the Preferred Argument Structure in the sense of Du Bois, and is structurally holophrastic, while the SV clause is articulated, being bi- or multi-partite in structure (e.g., subject-predicate). 4. The VS clause is continuous with respect to the discourse topic. The SV clause, on the other hand, is discontinuous with respect to the discourse topic. 5. There may be morphological concomitants of the two clause types. If so, and the two clauses are differentiated by ergative vs. accusative grammatical structure, the VS type will always be ergative and the SV type accusative, never the other way round. 6. If the verb morphology reflects more verb-like as opposed to more noun-like features (Myhill 1985), these features will be distributed in such a way that the verb in the VS type will show the verb-like features and the verb in the SV type the noun-like features. The discourse-functional correlates of this distribution will be as predicted in the Categoriality Principle of Hopper and Thompson (1984). 7. Speculatively: Although not intrinsically unstable diachronically, this type of language may show a "drift" toward stable SVO word order. Such a drift is (always? often?) a result of creolization.

On Form and Content in Typology Johanna Nichols

1. I n t r o d u c t i o n The invitation to examine the relation of form and content provides an opportunity to take a critical look at some of the fundamental goals and findings

of

typology.

The

following

discussion

is

intended

as

a

contribution to the question of the basic building blocks of a good typology.

As I assume

the

participants

in this symposium

would

unanimously agree, typology is not just the cross-linguistic study of the current hot issues in theoretical descriptive grammar.

On the contrary,

some of the clearest examples of good typological practice—Greenberg's work on word order, or the work on ergativity and other aspects of clause organization that began to achieve prominence a decade or so ago—have contributed to descriptive theory rather than drawn from it.

The present

paper does not propose new typological building blocks, but rather it asks whether the fundamental units in some established typological patterns are units of form or units of content. The

discussion

below

presents

some

examples

of

typologically

interesting grammatical patterning in which the form-content relation is unorthodox. The received view of the form-content relation would seem to be t h a t form codes content, or--more elaborately—that, as we progress upwards from semantics to deep syntax to surface syntax to morphology, each level codes the preceding one and is coded by the following one. I will call this the Saussurean

dogma, not because Saussure claimed it explicitly

but because it follows rather clearly from his view of the sign function. With regard to just form and content, the Saussurean dogma entails that

142

JOHANNA NICHOLS

content determines, or limits, or motivates, form, and not versa.

In

Saussure's terms, content is the signifié

A

and form the signifiant.

central goal of this paper is to subvert the Saussurean dogma in two ways: first,

by

adducing

straightforwardly

examples

where

content-to-form

but

the

coding

is reflexive,

relation with

form

is

not

coding

structural properties of itself or content referring to properties of itself; and second, by adducing examples where, in cross-linguistic implicational and predictive statements as well as in individual languages, form limits or determines content rather t h a n vice versa. The conclusion will be t h a t the role of form in typology has been misunderstood and underestimated; t h a t just as form does more than merely code content, so it does more than merely reflect linguistic type. Closely allied to the Saussurean dogma is the well-known principle of duality of patterning.

Although this paper will not discuss duality of

patterning explicitly, it is worth noting t h a t any limitation on content imposed by the structural patterns of form poses a potential challenge to the principle.

An explicit challenge to duality of patterning (which may

also involve another kind of reflexivity) is presented in Woodbury 1985, where it is argued t h a t content resides in form in a way t h a t is unorthodox to the assumptions of structural

and formal

grammar.

Specifically,

phonological rules whose output has a changed meaning are interordered among other phonological rules. The organization of this paper will be as follows. Section 2 describes two instances of what have generally been viewed as semantic values coded in form: split intransitivity (or stative-active inflection) in verbs, and formal syncretisms sharings in case paradigms.

That section argues that,

contrary to the usual assumption, these are examples of content coding content or form coding form. vs.

Section 3 gives a brief typology of alienable

inalienable possession and its coding, and shows where form limits

content

and

where

content

limits

form.

Section

4 discusses

split

intransitivity and case meanings, arguing that the semantics of each is available only to certain formal types, i.e., that form determines content in that it limits its distributional possibilities.

Section 5 surveys lability,

commonly believed to have its basis in clause semantics, and argues that it

143

ON FORM AND CONTENT IN TYPOLOGY is actually based on and predictable from

certain abstract

structural

properties of languages. Some of the discussion to follow will be based on the distinction of head-marking and dependent-marking grammar (Nichols 1986a).

The

distinction has to do with whether syntactic relations are coded by an affix (or other marker) attached to the head or to the dependent member of the relation or constituent.

The dependent-marking type is the familiar Indo-

European pattern which places a formal marker on the dependent element of the relation or constituent: cases on subject, object, etc.; the genitive case on the possessor; cases on objects of prepositions or postpositions. The head-marking type is its mirror image: it marks clause relations by complex verb agreement (or cross-reference,

in Bloomfield's terms), puts a

possessive affix on the possessed noun instead of using the genitive case on the possessor, and uses the same possessive morphology on prepositions or postpositions.

The extreme dependent-marking type is represented by

various Northeast Caucasian languages, with Lezghian a particularly clear example; the Indo-European languages are almost entirely marking, with the single exception of subject-verb extreme

head-marking

type

is rare

in the

dependent-

agreement.

Old World,

The

where it is

represented by Abkhaz and its kin, but common in the New World. Although most languages show a consistent preference for one or the other type throughout their grammars, not all languages belong to the polar types: some use double marking (e.g., both genitive case and possessive suffix), and some use split marking (e.g., dependent-marked clauses but head-marked phrases, as in Hungarian). 2. R e f l e x i v i t y This section surveys two widely attested grammatical patterns for which the form-content relation can best be described as reflexive. In the first, the category meaning somehow makes reference to the set of category members, rather than membership depending straightforwardly on meaning. In the second, the form of the category is designed to facilitate recovery of form rather than straightforwardly signaling content.

144

JOHANNA NICHOLS

2 . 1 . Split i n t r a n s i t i v i t y This term, and the analysis reported here, are taken from Merlan 1985. The term refers to what is commonly known as stative-active, or simply active, inflection, the type (found e.g., in Siouan and Iroquoian languages) where some intransitive verbs take subjects inflected like the agents of transitive constructions (the 'active' class), while others take

subjects

inflected like the patients of transitive constructions (the 'stative' class). Merlan argues t h a t the received view of this opposition as encoding the relatively agent-like vs. incorrect.

Following

patient-like nature or role of the subject is Merlan

(but

rewording

her

description

and

elaborating her examples of the second type), split intransitivity can be broken down into two formal subtypes: head-marked and marked.

dependent-

The difference has to do with whether the opposition is carried

by verbs or by their subjects. The head-marked type opposes two different classes of pronominal markers (typically subjective vs. objective, but both agree with subjects) on intransitive verbs; since the languages using this type generally lack cases, the subject nouns themselves do not differ.

The

dependent-marked type involves a choice between two different

cases

(typically ergative vs. nominative) on the subject; since the languages using this type have little or no verb agreement, the 'stative' and 'active' verbs themselves differ minimally if at all. The head-marked pattern (the formal opposition in verb inflection) is exhibited by Dakota, Seneca, A n k a r a , and Tunica in North America, and by Mangarayi in Australia (all consistently head-marking languages); it is also detectable in Georgian in the South Caucasus, where the clause is actually double-marked rather than strictly head-marked. languages, verbs are strictly classified as statively vs.

In all these

actively inflecting.

Merlan shows t h a t in some languages it is the stative verbs t h a t comprise the

minority,

marked

class, while in others

it

is the

active

verbs.

Nonetheless, there is a constant semantic generalization to be made: in almost all languages the marked, minority subclass contains verbs denoting bodily functions and processes, regardless of whether t h a t class is stative or active. Significantly, particular glosses ('sneeze', 'hiccup', various kinds of sounds and cries) recur in more than one language, appearing sometimes in

145

ON FORM AND CONTENT IN TYPOLOGY

the stative set and sometimes in the active set. Hence any cross-linguistic generalization made about the content of the stative-active opposition must refer not to the relation between the formal class and the semantic role of the subject, but to the relation between the formal class and the predicates t h a t enter into it.

The opposition is thus semantically reflexive.

Viewed

cross-linguistically, split intransitivity is not simply the grammaticalization of agent and patient relations, i.e., is not simply a matter of a formal opposition responding to a semantic opposition.

Not surprisingly, then, in

none of these languages does the speaker have a choice of inflection type which carries semantic consequences; rather, the inflection type is dictated by the lexicon. The

above

generalizations

apply

only

to

head-marked

intransitivity, and not to the dependent-marked type.

split

Dependent-marked

split intransitivity (which uses an opposition of subject cases) is represented in Merlan's corpus by Eastern P o m o and Batsbi (for the latter see also Holisky 1986); it is also found in Northern Pomo (O'Connor 1985).

Here

the form-content relation is not reflexive but straightforward: for many or (for Batsbi) perhaps all intransitive verbs, the speaker

does have

a

systematic choice of inflectional patterns, a choice which carries semantic consequences.

The semantic opposition often has to do with volitionality

urgency: one case is used for subjects which act non-volitionally, the other for subjects which act volitionally. In Northern Pomo, as O'Connor shows, some verbs require subjects in the agent case, some require them in the patient case, and some permit both cases, with a clear semantic difference. The

semantic

difference

is

not

an

invariable

contrast

in

agency,

volitionality, or the like; rather, O'Connor describes it as what can be broadly termed an evidential parameter. In summary, the semantics of head-marked split intransitivity is reflexive, while the semantics of dependent-marked split intransitivity is straightforward.

The significance of this fact will be discussed again in

section 4.2. 2.2. Syncretism Jakobson 1971a,b,c suggests t h a t syncretism functions as a signal of semantic parameters. For instance, in certain masculine noun paradigms in

146

JOHANNA NICHOLS

Russian, the dative and second genitive or second locative endings are formally syncretic, all of them -u.

In terms of Jakobson's componential

analysis of the semantics of Russian cases, this syncretism follows from the fact t h a t all these cases share the feature of directionality.

Jakobson also

argues t h a t the syncretism of genitive singular and nominative plural in various paradigms has to do with the fact that both genitive and plural have to

do with

quantification.

This view

is consistent

with

the

Saussurean dogma. In contrast, I maintain t h a t syncretism is not a signal of abstract semantic categories.

Rather, it is a phenomenon t h a t survives due to its

functional value in language processing and perhaps also language learning. Syncretism is one aspect of the more general propensity of languages to reduce the inventory of elements (segments, sequences, categories) available in certain contexts.

Language after language displays a restricted set of

phonemes used in inflectional affixes (while all phonemes can occur in roots),

syncretism

in

inflectional

paradigms,

and

reduced

complexity in certain parts of speech or inflectional classes.

categorial I maintain

that all these phenomena have the same functional motivation: if there is a reduced inventory of segments or categories available in a certain context, then the hearer's chances of guessing correctly what segment or category has just been uttered are considerably increased.

The greater the number

of / u / or / o j / endings in a given paradigm, the greater the hearer's chances of guessing correctly t h a t he or she has just heard / u / or / o j / . On this analysis, syncretism is reflexive in that formal patterns function to facilitate the recovery of form, rather than (as the received view would have it) functioning purely to facilitate the recovery of content. This analysis does not preclude the possibility t h a t some instances of syncretism may follow semantic lines; it simply means t h a t syncretism is not to be taken a priori

and exclusively as evidence of semantic or

functional overlap, and that reduction of the formal inventory is relevant even if semantic conditioning is also involved.

Unfortunately, the chances

of testing the relative extent to which semantic overlap and

formal

reduction determine syncretism are not good: true syncretism is typical only of inflecting languages with different declension classes of nouns, and

ON FORM AND CONTENT IN TYPOLOGY

147

there are few such languages outside of Indo-European. (Syncretism must be distinguished from isofunctionality, which means marking of two or more semantic roles by a single case, e.g., the use of the Russian instrumental case to mark both instruments and agents. Isofunctionality is of course semantically motivated in a straightforward way.) Neither semantic overlap nor formal reduction could easily be falsified within Jakobson's three-feature framework: since in his system 23 out of 27, or 8 5 % , of the possible pairings of the eight Russian cases share at least one feature value, of course the majority of syncretisms will appear to be semantically motivated on his analysis. Syncretism also functions reflexively on a more abstract level. Virtually all structuralist inquiry (Hjelmslev 1935-37, Jakobson 1971abc, Greenberg 1966) observes that at a more abstract level syncretism is due to, and an index of, markedness. We could therefore say t h a t on this level syncretism codes the markedness of the category. Such coding is reflexive because the formal category codes properties of the category qua category, rather than coding its actual content. 3 . Split possession: T h e s e m a n t i c s of alienable a n d i n a l i e n a b l e possession. Formal distinctions in the marking of possession, associated one way or another with the semantics of the possessed nouns or the relation of possession itself, are common. inalienable possession.

They are often referred to as alienable vs.

This section argues that there is no invariant

semantic content to such oppositions, t h a t whatever semantic values are involved are contingent on the form that signals them rather than vice versa, and that the treatment of 'alienable' and 'inalienable' possession in the literature errs just as Merlan has shown t h a t the treatment of 'stative' and 'active' inflection errs:

consistent with the Saussurean dogma, it

attributes a semantic constant to what actually turns out to be a formal split in the marking of adnominal constructions. inalienable

The terms alienable

and

are still useful, and will be used below for convenience, but the

JOHANNA NICHOLS

148

distinction should not be taken as a purely semantic one. 1 The essential question to ask in typologi zing possessive constructions turns out to be whether possession is head-marked or dependent-marked. In a possessive phrase, the possessed noun is head and the possessor is its dependent. Head-marked possession involves a possessive affix on the head (possessed) noun of the phrase, as in Nanai (Avrorin 1959:157-8): naj person

dili- ni head 3sg

'(the/a) person's head' (inalienable)

while dependent-marked possession is the familiar Indo-European

type

where the head noun is not case-marked for possession and the dependent takes a genitive case, e.g., Russian: golova head

celovek- a person GEN

'(the/a) person's head'

The semantics, and indeed the very existence, of alienability contrasts turns out to depend crucially on whether possession is head-marked or dependent-marked. I begin with a typology of alienable-inalienable oppositions, classified by form: first by the head or dependent marking of possession (Pattern numbers 1, 2, 3, etc.), and then by whether the two possession types are in complementary distribution (subtypes A) or contrast

(B).

Then

the

number of languages in my corpus displaying the pattern is given, followed by a list of the languages. The pattern numbers are ordered by decreasing propensity to use head-marked

possession:

Patterns

1

and

2

use

only

head-marked

possession; P a t t e r n s 3-6 involve various splits, with alienably possessed nouns marked one way and inalienably possessed nouns marked another way (and of those splits, Patterns 3-5 oppose head marking to some other type, while 6 does not involve head marking); P a t t e r n 7 uses only dependent-marked possession.

A glance at the numbers of languages for

The original findings of this section have been expanded and reported separately in Nichols 1986b. The present section draws on the larger survey of that paper and can now be regarded as a summary of it. For sources on languages mentioned, see that paper.

149

ON FORM AND CONTENT IN TYPOLOGY

each

pattern

will

show

that

oppositions

of

alienable

to

inalienable

possession are frequent where possession is head-marked, rare elsewhere. P a t t e r n 1: The language uses only head-marked possession. Inalienably possessed nouns cannot occur without possessive markers; alienably possessed nouns can occur alone. 19 languages: Acoma, Chipewyan, Chiricahua Apache, Diegueño, Greenland Eskimo, Ineseño Chumash, Hopi, Jacaltec, Karok, Kootenay, Menomini, Nanai, Navajo, Pawnee, Seneca, Tarascan, Tzutujil, Washo, Yurok P a t t e r n 2: The language uses only head-marked possession. There are two formal types of possessive affixes: one used for inalienables, the other for alienables. 2A: the type of possessive affix is lexically conditioned. 14 languages: Acoma, Diegueño, Hualapai, Lakhota, Nguna, Nootka, Nunggubuyu, Pima-Papago, Takelma, Tuscarora, Tzutujil, Upper Chehalis, Wiyot, Yavapai 2B: The possessive affixes are in contrast. 9 languages: Chiricahua Apache, Hualapai, Kootenay, Menomini, Nanai, Navajo, Pima-Papago, Seneca, Tzutujil Total languages for P a t t e r n 2: 23. An example of P a t t e r n 2B comes from Nanai, where body-part nouns take regular possessive affixes when inalienably possessed and take an expanded form when alienably possessed (Avrorin, loc. cit.). naj dilli-ni person head 3sg

'(the/a) person's head'

(inalienable)

naj dili-ηo-ni person head 3sg '(detached) head (e.g., of an animal) owned by a person' (alienable) P a t t e r n 3: Possessive marking is split. Inalienably possessed nouns take head-marked possession; for alienables there is no marking. 1 language: Matngala P a t t e r n 4: Split marking. Inalienables take head-marked possession, alienables dependent-marked. 4A: Lexically conditioned. 11 languages and one family: Burushaski, Eastern Pomo, Hittite, Maidu, Maung, Nez Perce, Northern Pomo, Northwest Sahaptin, Patwin, Polynesian (family), Wintu, Yurak Samoyed

150

JOHANNA NICHOLS 4B:

Contrast. 0 languages. Total for P a t t e r n 4: 11 plus one family. An example of P a t t e r n 4A comes from Eastern Pomo, where a set of kin terms takes head-marked possession while other nouns take a dependent-marked construction with a genitive (McLendon 1975:92, 108): wí1sg

bàyle husband

wax šá-ri my-GEN basket

'my husband'

(inalienable)

'my basket'

(alienable)

P a t t e r n 5: Split marking. Inalienables are head-marked, alienables double-marked. 5A: Lexically conditioned. 0 languages. 5B: Contrast. 3 languages: Ge'ez, Turkish, Tuva Total languages for P a t t e r n 5: 3. P a t t e r n 6: Split marking. Inalienables take no marking; alienables take dependent marking. 6A: Lexically conditioned. 4 languages: Dyirbal, Garawa, Uradhi, Nunggubuyu 6B: Contrast. 0 languages. Total languages for P a t t e r n 6: 4. P a t t e r n 7: The language uses only dependent-marked possession. Two different cases, or other possessor forms, are used, one marking inalienables and the other marking alienables. 7A: Lexically conditioned. 1 language family: Polynesian (many members) 7B: Contrast. 1 language: Northern Pomo Total for P a t t e r n 7: one language and one family. This survey of the data leads to the following conclusions. First, as mentioned above, alienability contrasts are most frequent in languages which use head-marked possession, either exclusively (Patterns 1 and 2) or as one of the types in a split system (Patterns 3-5).

They are much less

151

ON FORM AND CONTENT IN TYPOLOGY

frequent in languages t h a t use only dependent-marked possession (Pattern 7) or a split system involving dependent-marked possession (Pattern 6). Second, if a language uses head-marked possession it is highly likely to have an alienability contrast: if it uses dependent-marked possession it is highly unlikely to have an alienability contrast.

I found

languages which use head-marked possession but have no

very

few

alienability

opposition (a total of five, as against 30 languages using head-marked possession and showing alienability oppositions), but many languages which use dependent-marked possession and have no alienability opposition (they include many Indo-European and Northeast Caucasian languages).

The

two instances of P a t t e r n 7 in my corpus both involve languages (Northern Pomo and the Polynesian family) which have other, head-marked patterns of inalienable possession (Type 4A) in addition to these. Third, in split systems inalienable possession is invariably associated with more nearly head-marking, or less dependent-marking, tendencies in the encoding of possessive relations.

Either the inalienables are head-

marked and the alienables are not, or the inalienables are not dependentmarked while the alienables are. These three findings clearly show that head-marked possession is a nearly necessary and nearly sufficient condition for an alienability contrast. Two conclusions may be drawn about the lexicosemantic conditioning of alienability oppositions.

(These facts are not shown above.)

First, in

virtually every language having an alienability opposition the inalienably possessed nouns form

a closed, often small, set, while the

possessed nouns are an open set.

alienably

There are no reversals of this principle.

There is one non-conformity: Polynesian, in which both sets are apparently open. Second, there are strong cross-linguistic consistencies in the lexical membership of the inalienable category, consistencies which can best be stated in terms of an implicational hierarchy: 1. Kin terms

and/or

body parts

2. Ρart-whole and spatial relations 3. Culturally basic possessed items ('house', 'arrows', etc.)

152

JOHANNA NICHOLS There are many languages for which the inalienable nouns include

only kin terms; and several for which they include only body parts.

There

are many for which they consist of both kin terms and body parts, and many which add relational nouns and various artifacts.

Although the

composition and ordering of steps 2 and 3 of the hierarchy are not firm (since not all grammars exhaustively list inalienables), it is quite clear t h a t the hierarchy is co-headed by kin terms and body parts in inclusive disjunction. There is no cross-linguistic semantic invariant to the notion of inalienable possession, then.

The only generalizations t h a t can be made

concern the closed nature of the set of inalienables and the implicational hierarchy governing the lexical membership of that closed set. Even within individual languages, any semantic generalizations have to be phrased in terms of lexical membership rather than the raw semantics of some kind of possession.

This is because membership in lexical classes is not exact for

most languages:

if kin terms are included it is some or most but not all

kin terms t h a t are involved, and likewise for the other classes of nouns, which

means t h a t

we cannot

accurately

predict

membership

in

the

inalienable class from a description, however refined, of the semantics of kinhood or whatever.

Even the instances of contrast--the  patterns

above--apply only to the closed set of alienably possessed nouns (and sometimes only to a subset of them); so here, too, any semantic statement must take the form of a generalization over the category membership.

The

semantics of the alienability opposition, then, is another instance of reflexive semantics: a statement about the content of the category must refer not to the relation between the category and a chunk of pure semantics (organic possession, ownership, etc.) which it supposedly signals, but rather to the relation between the category and the lexical items t h a t enter into it.

This is entirely analogous to the recursive semantics of split

intransitivity in head-marking languages, as described above. Again analogously to the situation with split intransitivity, we find once more t h a t form determines, or limits, the possibility of expression of content: it is head-marked possession t h a t crucially favors oppositions, while its absence precludes them.

alienability

153

ON FORM AND CONTENT IN TYPOLOGY

Historically, it is easy to see how this situation could have come about.

The nouns categorized as inalienables are those which are most

often used in possessive constructions even in languages like English. Inalienable possession in its various forms is simply the reflex of tighter bonding and even fusion between possessor and possessed: it involves obligatory

possession

(Pattern

1) or fewer

morphemes

separating possessor and possessed (Patterns 2-6).

or

phonemes

(The Nanai examples

cited above are typical of P a t t e r n 2 in t h a t the possessive morphology for alienable possession is longer and morphemically more complex t h a t t h a t for inalienable possession.) Apart

from

the form-content

broader implications for typology.

relation, these findings The first

concerns the

have

two

discourse

motivation of grammatical categories. Hopper and Thompson (1984:724-5) find cross-linguistic discourse-based categoriality in the fact t h a t possessive constructions with body-part nouns disfavor dependent marking a n d / o r favor head marking (this is not their terminology).

They argue t h a t the

relevant patterns constitute reduced nominal categoriality for the bodypart noun, and t h a t this reduced categoriality is due to the fact t h a t bodypart nouns are low in discourse salience.

T h a t analysis accounts for the

frequent treatment of body parts as inalienably possessed, but it fails to account for the substantially more frequent treatment of kin terms as inalienably possessed; and a kin term, since it denotes a h u m a n being, is high in discourse salience.

Their analysis and the one given here are at

cross-purposes rather than being contradictory, however. Thompson

proceed from function

Hopper

and

to form, and find examples where

discourse function is mirrored in aspects of form; their object of description is not pre-existent formal categories like inalienable possession as a whole, but those individual formal patterns that happen to show t h a t form can mirror function.

Their function-to-form approach does not obligate them

to consider the whole cross-linguistic range of alienability oppositions or even the whole membership of the inalienable set for any given language. The approach taken here, in contrast, takes the formal categories as given and asks about their relation to content.

It is obligated to consider the

whole range of alienability oppositions and their membership (and it has

154

JOHANNA NICHOLS

attempted to do just that); what it is not obligated to do is consider the entire range of a functional parameter such as discourse salience (and it has not done t h a t ) .

The unsurprising conclusion is t h a t a functionally-

based typology may account for some aspects of form, but the categories t h a t a form-oriented grammarian takes as basic will not lie within its purview; and a form-based typology may account for some aspects of function or meaning, but the categories t h a t a functionalist takes as basic do not come under its purview. The second typological

implication

concerns clause

organization.

Klimov 1977:148-56 argues that an opposition of alienable vs.

inalienable

possession is an implication of the stative-active morphosyntactic type. The present survey has argued that it is contingent instead on the presence of head-marked possession in the language.

It is worth inquiring into

whether head-marked possession (or head-marked morphology in general) predicts the stative-active clause type or vice versa.

The number of

languages known to be stative-active has increased substantially since Merlan's paper, on which section 2.1 above is based, was written; to the best of my knowledge all of the recently discovered ones have head-marked clauses.

Batsbi and the Pomoan languages apparently remain the only

dependent-marking stative-active languages.

Now, there are also many

head-marking languages which are not stative-active. stative-active

languages

are

mostly

head-marking

If it is true t h a t but

head-marking

languages are not mostly stative-active, then morphological marking type limits clause organization type; and if that is so, then it is indeed morphological marking type rather than clause organization type t h a t is the

ultimate

investigation.

predictor

of

alienability

oppositions.

The facts

merit

2

4. F o r m d e t e r m i n e s c o n t e n t The three patterns to be discussed in this section exhibit a straightforward dependency of content on form: in all of them, the availability of a content distinction is dependent on certain formal

or structural

propensities.

Specifically, certain semantic distinctions seem to be available only to head-marked constituents or only to dependent-marked constituents. In fact they are already under investigation, in a separate project.

155

ON FORM AND CONTENT IN TYPOLOGY 4 . 1 . A l i e n a b l e a n d inalienable possession

As argued in section 3 above, in the area of alienable and inalienable possession there are several respects in which the possibility of signaling some content depends on the formal expression available. A true semantic opposition of (roughly) ownership vs. other possession, found in Chiricahua Apache, Menomini, Nanai, and Tzutujil, is apparently available only when possession is regularly head-marked. Oppositions involving referentiality of the possessor arise when there is a possibility of contrasting dependentmarked with non-dependent-marked possessors.

And so on.

In other

words, the particular semantics of 'alienability' or 'inalienability' depends on the type of formal distinction. of

course

be

gotten

across

The semantic distinctions involved can in

a

language

of

any

type,

but

grammaticalization of the distinction appears to take place only when head marking is available somewhere in the system. In addition to such semantic distinctions, there is an opposition which does not signal a semantic contrast, although its distribution is limited by semantic classes of possessed nouns.

This is the pattern where

certain nouns, primarily kin terms a n d / o r body parts, take a form of possessive marking which points to closer formal binding between possessor and possessed.

This opposition is again inherently connected to the form

of the marking, in t h a t the closer binding is associated with increased tendencies

toward

head

marking

and

decreased

tendencies

toward

dependent marking. 4 . 2 . Split i n t r a n s i t i v i t y As shown in section 2, the nature of the form-content relation in split intransitivity depends on its formal marking.

Where the distinction is

head-marked, the form-content relation is reflexive; where it is dependentmarked, the form-content relation is straightforward.

Thus the kind of

semantic generalization we think of as typical--the straightforward, nonrecursive type where there is an opposition of pure semantics, as in Batsbi-is available only for the dependent-marked relation. 4.3. Case meanings The conclusion of section 4.2 can be generalized. Dependent-marked split intransitivity--the choice of subject cases for intransitive verbs--is only one

JOHANNA NICHOLS

156

of many

manifestations

of semanticity

in case choice.

Well-known

examples of semanticity in case choice in Russian include the choice of accusative vs. genitive in direct objects of negated verbs (for the semantic analysis

see

Timberlake

1975)

and

the

choice

of

instrumental

vs.

agreement in predicate adjectives (for the semantic analysis see Nichols 1981, Timberlake 1986). In Chechen-Ingush and in Lezghian, the choice of dative

vs.

all ative

case

for

indirect

objects

signals

permanent

vs.

temporary transfer of possession. In a number of languages in the Turkic and Tungusic families, there is a choice in direct object

cases: the

nominative is used for indefinite or non-referential objects, the accusative for referential or definite objects.

Finnish makes a choice of nominative

vs. partitive in predicate adjectives.

All of these examples involve a free

choice between two cases, with semantic consequences. entirely typical of languages using cases.

These patterns are

Such choices—and, importantly,

the kinds of semantics they entail—appear to be simply unavailable to languages using head-marked clauses. 3 This is another instance where straightforward, non-recursive formal signaling of semantic oppositions is available only through dependent-marking. 5. Lability Lability (or 'diffuseness') is a valence pattern in which the verb can be either

transitive

or

intransitive

without

the

application

of

formal

transitivizing or detransitivizing derivations (English open or break labile

by

this

definition).

The

term

lability

is

used,

are

somewhat

indiscriminately, in three distinct senses in Caucasian studies (as clarified by Hewitt 1982). Here it will refer only to what Hewitt describes as agent deletion.

The labile verb may lack an agent and have only a patient, with

the patient identifiable as subject and the verb hence intransitive; or it may have both

an ergative

agent

and

a nominative

patient,

hence

patterning as a transitive verb. I symbolize the overall valence pattern as (A)0, capturing the optionality of the agent and the obligatory presence of the patient. Unless the choice of agreement vs. non-agreement in gender and number in Russian verbs is an example. The choice is available when the subject is a quantified noun phrase. It is determined by a number of factors, including some semantic ones (such as topicality, animacy, and referentiality of the subject): see Corbett 1983:, Nichols et al. 1980:375-6.

157

ON FORM AND CONTENT IN TYPOLOGY

Lability is generally viewed as a typical feature of ergative languages (see e.g., Klimov 1973:118ff.). But in fact its distribution is more narrow. It is found in two types of languages. First, trivially and uninterestingly, it is found in languages like English, which have very little morphology and hence make extensive use of zero derivation.

Such languages will not be

discussed here. Second, and interestingly because the languages in question are rich in morphology, it is found in a subset of ergative languages which can be defined on formal grounds: languages whose verbal derivations are exclusively transitivizing, or exclude intransitivizing processes.

These in

t u r n will be languages which lack relation-changing rules; the absence of detransitivizing derivations is simply the lexical reflex of the absence of syntactic operations of promotion and demotion. Lability has certain associations with content.

If a language has

labile verbs, they will center on those which, like 'break', 'open', or fill', refer to changes of state which can either happen spontaneously or be brought about by human agency.

These include the type of verb which

Nedjalkov (1969) shows to be cross-linguistically most prone to formally derive the intransitive from the transitive (as with Russian lomat' (trans.) lomat'sja

'break'

'break' (intr.), where the intransitive is derived from the

transitive by reflexivization).

However, while it does seem possible to

predict, for a language with labile verbs, which verbs are most likely to be labile, this semantic factor does not allow us to predict whether a given language will have any lability.

Lability can be predicted only from

abstract structural, and specifically derivational, properties of a language. Some

examples

showing

labile

verbs

in

Northeast

Caucasian

languages: Lak

č'äwussa aramtal Ii-w-t'unni many people-NOM G died 'many people died' (Zirkov 1955:105)

( G = g e n d e r marker)

qqačaytural č'äwussa aramtal li-w-t'unni bandits-ERG many people-NOM G died 'the bandits killed many people' (ibid.) Ingush

áz

xiy

píla

ħal-dizar

158

JOHANNA NICHOLS I-ERG water-LOCglass-NOM 'I filled the glass with water'

up

filled (transitive use)

xiy píla ħal-dizar water-LOCglass-NOM u p filled 'The glass filled with water'

(intransitive use)

Table 1 shows the results of a survey of several languages testing the cooccurrence of lability, ergativity, and other factors. As shown in Table 1, many, and probably most, languages of the Northeast Caucasus share the features of lability, transitivization in the form of a causative, and lack of detransitivization. Klimov & Alekseev 1980:260-62, l'ču 1982.

See e.g., Kibrik 1979,

The latter source argues

t h a t all verbs are basically intransitive in Lezghi; this is not true of most Northeast Caucasian languages, however. Languages of the Northwest Caucasus have relatively few (A)O labile verbs, and some of those verbs exhibit a formal distinction t h a t permits them to be regarded as involving detransitivization rather than lability. These languages detransitivize both by reflexivization and by A(O) object deletion (Hewitt 1982; for detransitivization by reflexivization, see e.g., Šagirov 1967:175, Kumaxov 1967:154). Lability and lack of detransitivizing derivations are usually found together.

The only language found to have only one of these two is Avar,

which has a number of labile verbs and yet also has a detransitivizing derivation.

The

detransitivizing

derivation

appears

to

be

a

recent

innovation; it may represent the first step toward a different syntactic type which will eventually lead to loss of lability.

Lability and lack of

detransitivization imply lack of relation-changing rules, and not vice versa, as is shown by the Tungusic languages, many of which (e.g., Nanai, Oroch) have no relation-changing rules yet have detransitivization and lack labile verbs. Lability is apparently limited to ergative languages, as has long been pointed out.

However, the association with ergativity is not a direct one.

Most ergative languages seem to have antipassives and other relationchanging rules, and hence most of them have detransitivizing derivations; such languages lack lability. Lability and lack of detransitivization are not

ON FORM AND CONTENT IN TYPOLOGY

Language

Area

Source

lability

TransitivDetransiRelation tivizing changing izing morphology morphology rules

(A)O

159

Ergative Ergative morphology syntax

ChechenIngush

field notes

NEC

yes

yes

no

no

yes

no

Lezghi

Mel'čuk 1982

NEC

yes

yes

no

no

yes

yes

Archi

Kibrik 1979 Kibrik et al. 1977

NEC

yes

yes

[no]

[no]

yes

yes

Lak

Žirkov 1955:81, 112,115 Klimov 1973:119

NEC

yes

yes

no

yes

Avar

Klimov & AJekseev 1980:260-61 Klimov 1973:70

NEC

yes (70+) yes

yes

yes

Circassian

Hewitt 1982:164-5

NWCyes*

yes

yes

Dyirbal

Dixon 1972

Austr

no

yes

Yidin

Dixon 1977

Austr

no

yes

Basque

Lafitte 1944 Jacobsen 1985

Eur

no

Tiutujil

Dayley 1985a,b

SAm

Lakhota

Boas &. Deloria

Nanai

Avrorin 1959, 1964

Russian

yes

yes

[no]

yes

yes

yes

yes

yes

yes

yes

yes

yes

yes

yes

no

no

yes

yes

yes

yes

partially

NAm

no

yes

yes

no

no(active)

no

Sib

no

yes

yes

no

no

no

Eur

no

no

yes

yes

no

no

KEY to abbreviations in 'Area' column: NEC = Northeast Caucasus, NWC = Northwest Caucasus, Austr = Australia, Eur = Europe, SAm = South or Central America, Sib = Siberia. "Marginal: there are some such verbs, but several of them show formal change.

T a b l e 1:

Lability and other typological factors

160

JOHANNA NICHOLS

automatic consequences even of canonical or syntactic ergativity, as is shown by Dyirbal, which despite its syntactic ergativity has no lability and has both transitivizing and detransitivizing derivations (Dixon 1972:64). And in the Northeast Caucasus, lability and lack of detransitivization are found both in syntactically ergative languages like Archi and Lezghi, and in syntactically accusative languages like Chechen and Ingush. While there exist accusative languages which do not use relationchanging rules (Nanai, Oroch, and other languages of the A m u r group of Tungusic are examples), there appear to be no accusative languages which lack detransitivizing derivations and exhibit lability. In summary, there is one type of ergative language which makes no use of relation-changing rules and no use of detransitivization. In this type of language, and in no other, we find (A)O lability. This kind of language is not common, even among ergative languages; in fact, the only examples of it t h a t I have found come from the North Caucasus. There is reason to think t h a t the cooccurrence of lability, lack of relation-changing rules, and lack of detransitivization is not an accidental areal or genetic feature but the systematic manifestation of an abstract morphosyntactic and lexical type.

T h a t same abstract type is probably responsible for two other

pervasive features of verbal lexicalization found in the Northeast Caucasus. The first is the tendency to encode what Talmy 1985 calls figure nominative, without

offering

other

patterns

(and specifically

offering the type of option Talmy calls reversed precedence, e.g., load the truck with

in the without

as shown in

hay [reversed] beside load hay onto the

The second is the consistent

lack of distinction

between static

truck). and

inceptive verbs of stance: the semantic opposition expressed formally by Russian sest'

vs. sidet'

is expressed, if at all, in Northeast

languages by tense-aspect inflection and preverbs.

Caucasian

Both of these patterns

imply, at the lexical level, a meaning structure which gives centrality to the moving or located figure and accords a more peripheral status to the human agent of motion, position, or change. lexicalization

pattern

that

gives

these

It may be this preferred

languages

their

distinctive

grammatical cast. In conclusion, then, lability is a structural property and the reflex of

161

ON FORM AND CONTENT IN TYPOLOGY an overall structural type. ultimately

centered

in

the

T h a t overall structural type appears to be lexicon

rather

than

in

the

syntax:

its

manifestations are in the area of verbal valence (ergativity, lability) and verbal word-formation rules (no detransitivization), and it impinges on the syntax only in the form of a statement about the immutability of lexical patterns (absence of relation-changing rules).

It appears to be associated

with other lexicalization phenomena (no reversed precedence, no static/ inceptive distinction in verbs of stance), phenomena which, it should be noted, are not limited in distribution to ergative languages.

However,

although this is a lexical matter, it is one of structure and not content.

It

is important to emphasize that the lexical and even syntactic differences between language groups as diverse as Northeast Caucasian, Tungusic, and Australian are not differences in content:

verbal meanings in all of them

involve agents, patients, etc., and roots whose lexical core translates easily into any Western language. W h a t differs is the structural organization of t h a t shared content.

(A)O lability seems not to be fruitfully incorporable

into a typology based on a semantic analysis of clause types (ergative, stative-active, etc.), because it does not distinguish either these types or their fundamental content elements. But it is of obvious diagnostic value for a form-based typology. 6. C o n c l u s i o n s Three of the grammatical

patterns surveyed here—split

intransitivity,

lability, and (at least in Klimov 1973, 1977) alienable vs. inalienable possession—have

figured

in discussions of clause-level typology, notably

ergativity and the stative-active type.

The discussion above has not

weakened the claims for such associations. It has, however, suggested ways in which they might be better explained. If the patterns surveyed a b o v e split

intransitivity,

syncretism,

split

possession,

case

meanings,

and

lability—are to figure in content-oriented typologies, then it is imperative t h a t we have a full understanding of their content and its relation to their form. We must also have a full understanding of the extent to which form limits content.

Finally, we must have a full understanding of the ways in

which form limits form. For all of the phenomena surveyed here, I have argued t h a t the form-

JOHANNA NICHOLS

162

content relation is not what would be predicted by the Saussurean dogma. Rather, form limits content, form recursively signals form, and

even

content

than

itself

turns

out

to

be

a

matter

of

recursive

rather

straightforward sign relations. On the assumption that linguistic typology should focus on the predicting, limiting, or causal factors rather than on the resultant a n d / o r epiphenomenal factors, I suggest t h a t form has a more central role to play in typology than would appear to be the received view. 4

I am grateful to Mary Catherine O'Connor and Sandra Thompson for comments on an earlier version of this paper. My thanks should not be taken to indicate that they agree with my analysis or my interpretation of their work.

The Meaning-Form Correspondence in Grammatical Description A. E. Kibrik

The ultimate task of a grammatical description is to reveal the correspondence between linguistic forms and meanings, t h a t is, to elaborate rules of the following structure: "such-and-such a form has such-and-such a meaning," "such-and-such a meaning is expressed by such-and-such a form" (plus listing all the contextual restrictions on the domains of the rules).

In principle, such a correspondence has no intrinsic orientation.

However, the usual way of speaking about t h a t correspondence forces us to view one of its members as the source or point of departure, and the other as goal. Therefore the question arises whether the choice of direction is of any importance to linguistic description. proceed (1) from

form

to meaning

Does it matter whether we

or (2) from

meaning

to formt

My

purpose is to suggest t h a t there are reasons for preferring one decision to the

other;

furthermore,

in

contrast

to

tradition,

'meaning-to-form'

description has important advantages over form-to-meaning description. In descriptions of individual languages the first alternative (form-tomeaning)

usually

dominates.

Formal

linguistic

entities

(morphemes,

grammatical categories, phrase markers, etc.) are taken as primary units of analysis and are assigned certain meanings or functions.

This way of

description is motivated by the fact that it reflects the heuristic process of linguistic investigation: surface elements are the starting point of analysis. However, the logic of linguistic description does not coincide with the order of heuristic procedure.

necessarily

Moreover, because of the

exceedingly complex correspondence between forms and meanings, which is

164

Α. Ε. KONRIK

not one-to-one (one form may have many meanings and one meaning may be expressed by many forms), this way of description appears to be unjustifiably

complicated and lacking explanatory force.

The

coding

techniques of language, featuring its surface levels, are not autonomous but are strongly predetermined—first, by underlying semantic structure which these techniques are designed to represent; and, second, by the processes of transition

from

embodiment).

this

structure

to

its

terminal

realization

(material

The approach featuring the semantic level as a starting

point and relating certain coding devices to given semantic units has great advantages. This approach reveals the motivated nature of many syntactic and

morphological

description.

features

Motivatedness

of

language

deals with

and

thereby

establishing

simplifies

one-to-one

the

corre­

spondence between meaning and form or between meaning and some intermediate stage of transition (from meaning to form). To support this thesis I present some specific examples. 1. Having acknowledged the priority of meaning over form, we may dispense

with

useless

discussion

about

the

level

of

a

number

of

classificatory grammatical categories: parts of speech, noun classes, etc. Thus noun classes in class languages may be viewed as either semantic or syntactic or morphological entities.

Furthermore, semantic

treatment of a noun class system seems to be very difficult and even impossible.

T h a t is why many scholars are not inclined to look on a

semantic basis as a necessary condition for positing the existence of noun class systems.

The fact that noun classes are exceedingly widespread in

human languages has not found a good explanation.

It is however likely

that the existence of noun classes is determined not by morphological or syntactic factors, but by the universal semantic nature of the nominal lexicon; in certain conditions it may be overtly categorized, but otherwise this

categorization

is covert.

The

inherent

nature

of

classificatory

categories is always semantic, but in different languages those categories may be provided with coding techniques and assigned derivative functions to a different extent.

Therefore, attempts to define such categories in

syntactic or morphological terms, though more or less justified by language peculiarities, do not represent their universal nature.

THE MEANING-FORM

165

CORRESPONDENCE

2. Meaning-to-form orientation of linguistic analysis is of importance not only to morphology but, to no less an extent, to syntax. I touch briefly on

one

syntactic

problem

which,

despite

a tremendous

amount

of

publication, has not yet received a satisfactory solution: t h a t is, a problem of sentence typology.

This problem is primarily related to the distinction

between Accusative (=Nominative) and Ergative sentence types. There exist various opinions regarding this phenomenon: (a) first, the Accusative: Ergative opposition is a morphological phenomenon, resulting from the caprice of a given language; (b) second, this opposition is syntactic and is related to the notion of transitivity-intransitivity,

to

subject-object categories; (c) third, the phenomenon is a logical one; it is related to the way of thinking in a given language.

Though I would not

presume to make judgments concerning different possible ways of thinking, I am convinced that an explanatory theory of ergativity is possible only on a semantic basis and within the scope of universal sentence typology. It is necessary to develop a calculus of possible sentence types in universal semantic terms, for example, by means of a calculus of possible ways of distinguishing Agent and Patient of one-place- and two-placepredicates, and ways of neutralizing this opposition; or by means of choosing specific language tendencies, actualized in one or another.

Such

an approach to ergative languages shows t h a t they exhibit greater semantic motivatedness

in

expressing

accusative languages.

predicate-argument

relations

that

do

In particular, the lack of voice oppositions and the

presence of labile and affective verbs in ergative languages are due to the semantic motivatedness of the ergative way of argument coding.

T h a t is

to say, and Absolutive- (in alternative terms: Nominative-) coded noun corresponds not to two different entities (subject or object) but to a single entity: the nearest and most immediate participant of the situation.

That

noun is opposed to the argument shaping the given situation, t h a t is, the Agent of a two-place verb. A typical ergative language is a language with a fixed, iconic way of expressing semantic roles; to such a language the syntactic terms of subject and object are inapplicable.

As regards accusative languages, subject and

object positions are held by nouns with various semantic roles.

These

166

Α. Ε. KIBRIK

positions

have

referential

another,

properties

though

of

the

also

noun

semantic, phrases

task

and

of

the

manifesting

communicative

perspective of an utterance. 3. I mention one more syntactic problem, t h a t of syntactic synonymy. The very fact t h a t the attention of linguistics has been drawn to this phenomenon is favorable. devices

which

But it is likely t h a t the role of synonymous

actually

exist

in

human

language

is

considerably

exaggerated; to characterize two constructions as synonymous generally means inability to reveal and describe relevant semantic

distinctions.

Interpreting the same phenomena from the standpoint of detailed semantic analysis provides many interesting results. Thus, one of the arguments in favor of this hypothesis is the established correlation between "free" word order and topic-comment structure of the utterance.

In this connection I

refer to a very fruitful

attempt of my colleague Johanna Nichols to

describe

communicative

semantic

and

factors

forcing

the

choice

of

nominative vs. instrumental case in Russian. 4. The "meaning-to-form" method of description has its basis in typological research. linguistic

entities

For a typological point of view concerning various always

demonstrates

that

the

highest

degree

of

generalization is available on the semantic level, if the functional nature of an entity is clear. Indeed,

the postulate of translatability

(in principle) from

one

language to another implies the identity or essential similarity of their semantic bases. And this, in turn, means t h a t languages are comparable, that they have uniform structure, chiefly on the semantic level. But each language is characterized by a peculiar distribution between lexical and grammatical devices for the expression of meaning.

T h a t is why the

typological study of languages must take place primarily on the basis of semantic and not formal representation.

T h a t is, we must compare the

devices which languages use to express certain meanings and not the meanings t h a t similar forms have in various langauges. It is obvious t h a t for this sort of typology a description featuring meanings as starting points is preferable. It appears, therefore, t h a t descriptions focussing primarily on linguistic form are almost useless for this kind of typological work.

THE MEANING-FORM

167

CORRESPONDENCE

5. From the point of view of historical linguistics the meaning-toform description turns out to be preferable to an even greater extent. Historical studies aim to reveal the causes of linguistic change and to reconstruct prior states of language.

It is natural to believe t h a t the

primary correspondence between a meaning and a form is more direct and closer t h a n the derivative one, and t h a t there always exists a state where the relation between them is motivated.

In other words, from a historical

standpoint the existence of non-motivated relations between form meaning is doubtful.

and

The seeming lack of motivation finds its explanation

in the fact that this relation becomes obscure and demotivated; we must therefore look for a primary state.

Such an approach requires building a

substantial, functional, and not a formal, historical linguistics. Beyond the scope of a narrow grammatical description in the realm of verbal behavior, it turns out that meaning also fails to represent a starting point for linguistic description. communicative

stimulus

and

creative

Behind meaning is hidden the cognitive

behavior.

Semantic

representation is simply a way of coding and shaping thought; t h a t is the reason for the urgent need of models describing the thought-to-MESSAGE correspondence. But that is a topic for another report.

A Survey of Major Alaskan Language Types Michael E. Krauss

The two major Alaskan language types correspond directly to the two major "Alaskan" language families, Eskimo-Aleut and AthabaskanEyak-Tlingit.

(The other two language families marginally present in

southeasternmost

Alaska, Tsimshian

and Haida, are, in our

opinion,

genetically isolated, and also typologically distinct from Athabaskan-EyakTlingit, in the case of Haida contrary to Sapir's Na-Dene hypothesis, which claims genetic relation between Haida and Athabaskan-Tlingit).

Sapir

strongly felt, moreover, that Na-Dene, which fascinated and absorbed him more t h a n any other group of languages, was a late intrusion into the New World, the most "un-American," in fact relatable to Sino-Tibetan, a view very interesting to compare with Joseph Greenberg's very recent and still only informally published view t h a t Eskimo-Aleut was the last documented family to enter the Americas (4000 B.P.), Na-Dene distinctly the secondlast (6000 B.P.), while all the other language families of all the Americas together form a single vast genetic phylum, which crossed much earlier (15,000 B.P.) t h a n the two "Alaskan" families. Accordingly, typologists at this Soviet-American colloquium may find these "Alaskan" families of special interest, as they may belong prehistorically as a link somewhere between the Old World and the New. Here I shall present brief informal sketches of both families, including phonology and morphology as well as the syntax (the part these days that seems to be the main concern of typology), noting points of contact between the two groups, and noting points t h a t seem to me of special typological interest in their structure and their historical development.

170

MICHAEL E. KRAUSS

Branches of both families, Inupiaq or Inuit Eskimo and Athabaskan, still very cohesive, have in the last millenium or so spread far beyond Alaska, all the way across Arctic Canada to Greenland in the case of the former and nearly to Mexico (as Navajo and Apache) in the case of the latter. Though by far the greatest diversity remains in Alaska, I shall in some senses include the entire range of these recently extending families in this account of "Alaskan" types. 1. P h o n o l o g y I hope I shall be pardoned for beginning with and perhaps even dwelling a bit on the phonology of these Alaskan languages, which I am sure is less well known to typologists than some of the morphosyntax, but which I feel is no less interesting. I shall begin with tabular presentation of generalized taxonomic phoneme inventories for each group and a few remarks. For Eskimo-Aleut I present a basic consonant matrix, filled in with units in parentheses that are widespread but secondary a n d / o r allophonic.

where ρ t c k q are plain (voiceless unaspirated) stops,  is mostly [c] (widely p (>h)

> >

s).

Labials

0, and v

concerning this below. Inuit, especially Alaskan.

>

in w,

Aleut >

are

severely

χ in Pacific

The retroflexes generally >

weakened,

e.g.,

Gulf Yupik;

more

y, s, except in some

The velar-uvular distinction is fundamental and

very stable almost everywhere (except syllable-finally in modern Labrador). The vowels are a simple triangle, with a fourth central or neutral vowel:

171

MAJOR ALASKAN LANGUAGE TYPES

These have generally remained very stable, except for Ə, which

has

remained as such throughout Yupik, but in Aleut and (nearly all) Inuit there is only a i u; in Aleut Θ >

a, i, 0 fairly early, which in Inuit

(except at the Yupik border, especially Diomede) Ə merges with i on the surface, but original i differs widely still in morphophonemics and in leaving various degrees of palatalization of following apicais throughout the dialects. The Athabaskan-Eyak-Tlingit taxonomic phoneme inventory is far more extensive and typologically more distinctive.

For one thing, in

contrast with Eskimo-Aleut which has only consonants and vowels, the Athabaskan-Eyak-Tlingit has what I consider four classes or subsystems: obstruent, glides, sonorants, and vowels. As most convenient, I shall present the obstruents as reconstructed for Proto-Athabaskan-Eyak.

I have included here the glottals ? and h. Certainly distinct segments, e.g., as syllable onset, syllable final, this pair constitutes its own subsystem ("glides"), at the same time constituting distinctive features of the obstruents, as can clearly be seen from the table, and of sonorants, and becoming distinctive features of the vowels as well. Note t h a t none of the obstruents are voiced, except much later and only in the case of Athabaskan intervocalic fricatives. The stops (with which the affricates clearly belong) are either aspirated or glottalized (ejective) or neither (plain; but always voiceless, voice being merely a redundant feature of sonorants and vowels). These rows remain very stable throughout the

172

MICHAEL E. KRAUSS

family (though Athabaskan, as mentioned, has developed voiced fricatives, and in some areas has seen some aspirates turn to voiceless fricative). Tlingit, in addition though, has a row of glottalized fricatives: ł s ' , (lacking only the * š ' , which must at one time also have been present, since the only stems in which ts- and tš-series obstruents co-occur involve s ' , e.g., tša-s' 'pink salmon', implying earlier *tšá-š').

This also makes

h

Tlingit unusual in having five laterals, tł, tł , t ł ' , ł, and Γ , but no 1 (!, except very marginally as an unrelated sonorant, to be noted below). The vertical columns of series have remained intact as such only in Tlingit.

Athabaskan-Eyak has lost the distinctive labialization of the

velars and uvulars as such (partly preserved only as ν in reduced vowels), and in the proto-Athabaskan shift of the * -series to *tš w . In Athabaskan the vertical series of affricates (almost always acting together as units, e.g., if *ts >

tθ, then also *ts h

>tθh, *ts'

> t θ ' , *s >

Θ, *z >

) have

tended to develop in two different directions: either the distinctions are preserved by phonetic shifts, especially *ts (or most often, also > fronting also of *

>

>

tθ, *tš

ts, but in extinct Tsetsaut > tš and *q >

>

ts, *tš w

>

tr

pf !), along with

k; or there is little movement in the

position of the series, but then a loss in the number of contrasts, resulting in the set t tł ts (tš) k q, much more similar to the Eskimo set of positions. It is perhaps no coincidence t h a t the languages which have kept the k q and simplified the affricates are neighbors to Eskimo (especially Koyukon, Tanaina, Ahtna), but this is probably an extremely superficial or late

situation.

*k *q >

There

are

in

fact

clear

indications

that

the

shift

tš  may have spread very recently and very rapidly across

most of the vast Athabaskan language-complex territory, since k q is still present in one dialect of Babine (Hagwilgate, at the southwest corner of the northern area), in some Californian Athabaskan (Hupa, Wailaki) which is perhaps an extension from the preceding area, and in some of the oldest documents

(18th

century)

from

Churchill

area

Chipeqyan,

at

the

easternmost corner of the triangle. This type of distribution, at all three points of the great triangle, must indicate a very rapid shift across a still functional language- or dialect-continuum, unless someone can show me cases of a phonological shift tš

>

 in any language, as opposed to what

173

MAJOR ALASKAN LANGUAGE TYPES

is perhaps the most commonplace consonant shift in the world's languages,  >

tš.

Is this

unidirectionality

(k

>

tš very widely,

tš >



nowhere?) a universal? Note the very striking total lack of labial obstruents, along with the partial loss of labials in Aleut especially.

Jakobson and others have

suggested t h a t this might be connected with the wearing of labrets. However, the Eskimos in Alaska wore labrets as well as the Aleuts, but most lost no labials; the Tlingits were coastal and wore labrets, but Athabaskan clearly and Proto-Athabaskan-Eyak presumably (and perhaps also Proto-Athabaskan-Eyak-Tlingit) were interior peoples and are not known ever to have worn labrets, but are also fully lacking in labials. Perhaps more could be understood by correlating with the linguistics what is known about the history and distribution of labret use; a correlation between loss of labials and labret use might indeed be found interesting, something like t h a t between smoking and lung cancer. The

Proto-Athabaskan-Eyak

sonorant

system

has recently

been

recognized as such and reconstructed as follows:

Syllable-finally the nasals were also probably non-occlusive [w (V) y w ' ( ' ỹ'] (while syllable-initial w and y probably also had allophones w and y

in

/_VN).

Tlingit

and

Athabaskan

(mostly)

lost

the

initial

preglottalized contrast, left only in Eyak, though such contrasts are present in Haida and Tsimshian, and even more importantly further south (e.g., in Salish).

In Eyak the nasality, on the other hand, is mostly lost, and in

Tlingit it is completely lost.

So in addition to other peculiarities, Tlingit

has nasality only as a redundant feature of sonorants, which formerly on the coast and still in the interior may be more or less nasalized syllablefinally in the case of y and w, while η generally is nasal but formerly and in the case of a few aged speakers still may be [1] instead.

(This sonorant,

ironically, is the only voiced lateral, hardly related to the obstruents tł t ł ' t łh ł ł'). The *η in Tlingit was denasalized to an unusual sonorant y, now

174

MICHAEL E. KRAUSS

merged with y in most dialects.

In Athabaskan *m and η mostly >

the system thus reduced to w n y.

These have tended to become

progressively obstruentized, most widely the labial (w the n, in a subset of the languages t h a t (η >

n,

>

v, m, p); next

have obstruentized the w

nt, t); and last, in a subset of those, y

>

ž.

Since I do not see

any particular conditions in Athabaskan which would determine

this

priority, first w, then n, then y, perhaps some kind of general principle is operative here. The Athabaskan-Eyak-Tlingit vowel nuclei are basically

each both tense and lax (or full and reduced).

The system remains in

Tlingit, but in Proto-Athabaskan-Eyak the four lax were reduced to three, Θ α υ (with the merger of reduced i and e to Θ), and were generally less stable than the full vowels, more subject to the obstruent environment. Further, the Proto-Athabaskan system

often develops into

or the like, where Θ >

ε, ν

>

ο, α

>

a, with various secondary

developments creating such classical patterns.

However, in Tanaina and

Ingalik, which otherwise as Athabaskan neighbors to each other have little in common other than proximity to Eskimo, a chain of unusual shifts of full vowels has taken place, whereby *u

>

i (=

i


u,

MAJOR ALASKAN LANGUAGE TYPES e > a, and the reduced Θ α υ all > with the Yupik

175

Θ, resulting in a triangle identical

Tanaina and Koyukon have obstruent series most like the Eskimo in point of articulation (not the Ingalik), but Tanaina and Ingalik (not Koyukon, which keeps the Proto-Athabaskan vowels virtually intact) have vowels most like the Yupik. Athabaskan

As will become clear below, though Eskimo and

are both notoriously polysynthetic, there the

resemblance

ends; we see no trace or even perhaps possibility of grammatical influence or diffusion or interpenetration from one to the other, and even lexically we find only a handful of loans in either direction, at the border where contact has been the most intense.

It becomes a question of unusual

interest then, how at the same time there appear to be such profound influences on the basic taxonomic phoneme patterns and inventory of Athabaskan from Eskimo. Are there situations in which a language can be so much more susceptible at t h a t level than any other to

profound

reshaping influence by diffusion from an otherwise unrelated neighbor? Here we see hardly any Yupik influence on Athabaskan other than this phonological one, which we would indeed consider quite profound. Athabaskan languages often develop nasalized vowels from syllablefinal nasals, usually fewer than oral. Eyak also has all four full vowels, each in the form Vh, V·, V?, V·?, and all except e nasalized as well, thus 28 full-vowel nuclei, including the glottal modifications, and no tone.

Proto-Athabaskan-Eyak had at least

full *V· and *V?, which in Proto-Athabaskan became *V· versus *V· (i.e., [±constricted]) This mark of constriction is what widely gave rise to Athabaskan tone; in some languages *V· > marked,",

in others *V· >

V, which we call "high-

V, which we call

"low-marked."

The

geographical distribution of these two outcomes is in partly checkerboard pattern, which cannot be explained by simple diffusions.

In several places

there are even mutually intelligible dialects which differ mainly in having reverse tones.

Are the differences perhaps partly relatable by diachronic

reversal, "flip-flop" as well as opposite development?

176

MICHAEL E. KRAUSS In any case, Athabaskan tonogenesis is clearly a development from

postvocalic glottal segments (? [and h?], and ·, as found in Eyak still) becoming suprasegmentalized as glottal modification of vowel voice-quality, and t h a t in t u r n becoming pitch. Tlingit tonogenesis is now equally clear. Modern Tlingit generally has V· vs. V· in full-vowel stems (reduced always V), but a southern dialect area has V· and V· (both merging as V· further north), and the sole surviving speaker now of the very southernmost sub dialect, Tongass, has V· and V? corresponding to V· (and southern V·, respectively), with

V'

(fading energy) corresponding to V·.

This one

speaker thus still preserves a form of pre-tonal Tlingit, geographically adjacent, incidentally, to Tsimshian, which has a similar system. One of the things t h a t excited Edward Sapir about the possibilities he saw of genetically linking his Na-Dene with Sino-Tibetan was the tonality present in both, and which he was the first to establish or document

in Athabaskan,

Tlingit,

and

Haida

(where

tone,

though

phonetically prominent, proves even more superficial). We now know t h a t on both sides earlier stages can be reconstructed t h a t are pre-tonal, and t h a t the development of tones must be explained as of rather

different

origins, so t h a t even typologically the tonogenesis in Na-Dene and SinoTibetan is quite different. Finally in phonology, I shall make some observation on phonotactics, syllable structure, and prosody. # In Eskimo-Aleut the syllable is generally of the shape CV(V) (C), so that maximal internal consonant clusters are CC, and there are no initial or final consonant clusters.

(Aleut, however, copiously allows a special

type of internal cluster of three:

i.e., apical flanked by χ a n d / o r : , as in Alaxsxa :, the original form of the name Alaska). cluster, thus

Eskimo generally allows any two of the vowels i a u to

177

MAJOR ALASKAN LANGUAGE TYPES

but Aleut has only ii aa uu. Central Siberian Yupik (Chaplinski and St. Lawrence Island), unlike any other Eskimo, also has only ii aa uu, but this is from a synchronically active and transparent principle of assimilation by dominance of i over a over u, so that ai ia ui iu au and ua and here

→

uvular →

→

ii;

aa; and uu only ← uu. Moreover, whole labialized velar

series arise from

au

and

ua

→

aa, thus

auk

'blood'

aak w , no longer interpretable as auk, since e.g., with truncating

suffix -lek 'having', 'having blood' (other Eskimo aulək) here is aalək, with no

labial

segment

remaining.

Elsewhere

in Eskimo

there

are

also

tendencies, though lesser, for assimilation of vowel clusters as active synchronic rules, as in Upper Kobuk Inuit, where ai and ia ua → õ,

or

in

Naukanski

where

ai

and ia → ë,

→

ë, au and

apparently,

but

au → aa, as in Chaplinski, while ua → ō; in Greenlandic aa au ai (except final) all → [æ·], except before uvular, which has pharyngeal effect, thus instead [a·], coincident with Danish (and similar to flatness in Arabic and Chilcotin Athabaskan). The extreme degree of vowel cluster assimilation in Central Siberian Yupik may well be due to the influence of Chukchi (which only very marginally allows unassimilated clusters). Sirenikski Yupik, which is more profoundly different from all other Yupik than most specialists realize, points, on the other hand, to a very different # proto-Eskimo pattern, which allowed no vowel clusters, but had CV syllables

only,

evidently

preserving

old

intervocalic

everywhere else, as e.g., in Sirenikski ə 'blood'