Diachronic, areal, and typological Linguistics [Reprint 2019 ed.] 9783111418797, 9783111054414

Table of contents :
EDITOR'S INTRODUCTION
CONTENTS
MASTER LIST OF ABBREVIATIONS
PART ONE: INTRODUCTORY
THE HISTORY OF LANGUAGE CLASSIFICATION
PART TWO: METHODOLOGICAL
INTRODUCTORY REMARKS
THE COMPARATIVE METHOD
INTERNAL RECONSTRUCTION
MATHEMATICAL DEVELOPMENTS IN LEXICOSTATISTIC THEORY
ON COMPARATIVE LINGUISTICS: THE CASE OF GRASSMANN'S LAW
AREAL LINGUISTICS: SOME GENERAL CONSIDERATIONS
THE TYPOLOGICAL METHOD
THE SOCIAL SETTING OF LINGUISTIC CHANGE
WRITTEN RECORDS AND DECIPHERMENT
PART THREE: CASE STUDIES
INTRODUCTORY REMARKS
OTOMANGUEAN ISOGLOSSES
INTERNAL RECONSTRUCTION AND FINNOUGRIC (FINNISH)
THE ROMANCE DATA OF THE PILOT STUDIES FOR GLOTTOCHRONOLOGY
JAPANESE DIALECTS
DEEP-SURFACE CANONICAL DISPARITIES IN RELATION TO ANALYSIS AND CHANGE: AN AUSTRALIAN EXAMPLE
AREAL LINGUISTICS AND MIDDLE AMERICA
DIACHRONIC TYPOLOGY OF PHILIPPINE VOWEL SYSTEMS
SOME ASPECTS OF DECREOLIZATION IN CREOLE FRENCH
LINEAR B
ALTAIC LINGUISTIC RECONSTRUCTION AND CULTURE
BIOGRAPHICAL NOTES
INDEX OF NAMES
INDEX OF LANGUAGES AND WRITING SYSTEMS

Citation preview

CURRENT TRENDS IN LINGUISTICS VOLUME 11

CURRENT TRENDS IN LINGUISTICS Edited by THOMAS Research

A.

SEBEOK

Center for the Language Indiana University

V O L U M E

Sciences

11

Diachronic, Areal, and Typological Linguistics Associate

Editors:

HENRY M . HOENIGSWALD - ROBERT E . LONGACRE

Assistants to the Editor: ALEXANDRA RAMSAY D I LUGLJO - LUCIA H A D D ZOERCHER

1973

MOUTON THE H A G U E • PARIS

© Copyright 1973 in The Netherlands Mouton & Co. N.V., Publishers, The Hague. The research reported herein was performed pursuant to a contract with the U.S. Department of Health, Education, and Welfare, Office of Education, under the authority of Section 602, Title VI, NDEA.

LIBRARY OF CONGRESS CATALOG CARD NUMBER: 64-3663

Printed in The Netherlands by Mouton & Co., Printers, The Hague.

EDITOR'S I N T R O D U C T I O N

With the publication of Vol. 10, Linguistics in North America, earlier this year, our geopolitically oriented survey of linguistics around the globe has ended. All but one of the previous volumes in the series, namely, Vol. 3, Theoretical Foundations, were organized according to some major areal segment of the world. By contrast, this book, and the two to follow, were designed to deal either with selected methodological issues that are, or ought to be, perennially focal to linguistic inquiry, as the problem of language classification; or with subjects currently of renewed internal concern to general linguistics, as its own historiography (Vol. 13); or, again, with burning topics of mutual interest that have emerged at the multiple interfaces of linguistics with its cognate disciplines, a wide array of which is being covered in Vol. 12. The Editorial Board of Vol. 11 ingeniously conceived of Diachronic, Areal, and Typological Linguistics as consisting of a series of chapters devoted to a variety of methodological explorations, each to be followed by a case study exemplifying the particular mode of inquiry described. Accordingly, except for the introductory chapter by Robins, meant to provide a historical setting, this book is divided into two successive sections. These can be read in several ways: for instance, sequentially, through the nine methodological chapters, constituting Part Two, or crosswise, flipping from any methodological chapter directly to one or more of the ten corresponding case studies, which make up Part Three. The fit is something less than perfect. While many pairs of authors wrote their respective contributions to Parts Two and Three in unison, others were unwilling or unable to cooperate, or even to consult with each other. In one instance, because of an editorial decision, there are two case studies for one methodological statement of lexicostatistic theory. In another instance, due to unforeseen difficulties, we have a case study, concerning the Altaic languages, unsupported by a separate methodological chapter, although this discrepancy is adjusted, to a degree, in Hoenigswald's introductory remarks to Part Two. As reported in my Introduction to Vol. 7, the preparation of Vol. 11 was financed by the U.S. Office of Education, in the amount of $33,834, through a contract with the Indiana University Foundation, bearing the identification USOE-OEC-O-9-

VI

EDITOR'S INTRODUCTION

097735-2488. This continuing support for the series is hereby gratefully acknowledged. As announced in my Introduction to Vol. 10, written only two months ago, the colossal bulk of Vol. 12, Linguistics and Adjacent Arts and Sciences, makes it difficult to forecast the precise publication date of the three separate tomes that its seventysome chapters are likely to occupy, but my editorial staff, working in a concerted effort with the publisher's, is still aiming for the end of 1973. There are, however, unprecedented problems to be solved, which may delay the appearance of Vol. 12 until 1974. Not the least of these is our desire to provide that book with a sophisticated Index of Subjects, supplementing the routine Index of Names and Index of Languages. The development of a viable topical index for Vol. 12 will also be good exercise for the immensely more comprehensive and challenging task of preparing an Index of Subjects covering the contents of the entire series, which is intended to constitute a main component of Vol. 14, the Index to Current Trends in Linguistics, Vols. 1-13. The manuscripts for Vol. 13, Current Trends in the Historiography of Linguistics, are now being gathered in, and their editing is progressing apace. This book will certainly appear in 1974. Discussions are under way for the continued implementation of the impulse that has motivated and the ideas that have informed this series since its inception, but, in the light of a decade's experiences, it is sure to alter both its constitution and guise in the future. If the publisher's plans mature by the time expected, an initial announcement about the sequel to Current Trends in Linguistics can be made at the XI International Congress of Linguists, in Bologna, convening August next. In the meantime, preparations go on for a large variety of 'spin-off' books rooted in the series, including Native Languages of the Americas: Linguistic Essays, the only one in which I have personally maintained a direct editorial hand, and which is scheduled for publication simultaneously with Vol. 13. The Editorial Board for Vol. 11 consisted of only two scholars, Henry M. Hoenigswald, who has prime responsibility for the methodological chapters in Part Two, and Robert E. Longacre, who fostered the development of the case studies in Part Three. The technical preparation of this book for press was accomplished jointly by the veteran team of Alexandra Ramsay Di Luglio and Lucia Hadd Zoercher, who also compiled the Master List of Abbreviations and the two Indexes that facilitate access to the contents of this volume. To the two Editorial Board members, both of whom contributed introductory remarks to their respective sections, and one of whom, moreover, wrote an article for the book, to the rest of the eighteen authors, to my editorial staff, and to the invisible stage-managers and hands in The Hague, goes my deep appreciation. Bloomington, April 1, 1972

THOMAS A . SEBEOK

CONTENTS

EDITOR'S INTRODUCTION

v

MASTER LIST OF ABBREVIATIONS

ix

PART ONE: INTRODUCTORY

The History of Language Classification, by R. H. Robins

3

PART TWO: METHODOLOGICAL

Introductory Remarks, by Henry M. Hoenigswald The Comparative Method, by Henry M. Hoenigswald Internal Reconstruction, by Jerzy Kurylowicz Mathematical Developments in Lexicostatistic Theory, by David Sankoff . . On Comparative Linguistics: The Case of Grassmann's Law, by Paul Kiparsky Areal Linguistics: Some General Considerations, by Werner Winter The Typological Method, by Joseph H. Greenberg The Social Setting of Linguistic Change, by William Labov Written Records and Decipherment, by Ignace J. Gelb

45 51 63 93 115 135 149 195 253

PART THREE: CASE STUDIES

Introductory Remarks, by Robert E. Longacre Otomanguean Isoglosses, by Calvin R. Rensch Internal Reconstruction and Finno-Ugric (Finnish), by Raimo Anttila . . . . The Romance Data of the Pilot Studies for Glottochronology, by John A. Rea Japanese Dialects, by Shiro Hattori

287 295 317 355 369

VIII

CONTENTS

Deep-Surface Canonical Disparities in Relation to Analysis and Change: An Australian Example, by Kenneth Hale Areal Linguistics and Middle America, by Terrence Kaufman Diachronic Typology of Philippine Vowel Systems, by Lawrence Andrew Reid Some Aspects of Decreolization in Creole French, by Albert Valdman . . . . Linear B, by John Chadwick Altaic Linguistic Reconstruction and Culture, by John R. Krueger

401 401 459 485 507 537 569

BIOGRAPHICAL NOTES

583

INDEX OF NAMES

589

INDEX OF LANGUAGES AND WRITING SYSTEMS

597

M A S T E R LIST OF A B B R E V I A T I O N S

AAA-M ACIAm AfrS AL AmA AnINA AnL Anthropos ANZAAS AO AOH BAE-B BiOr BSOAS BSE CAJ CAnthr CFS CSSH CTL

ESA FL FLing GL HJAS HMAI IF IJAL IUPAL JanL JAOS JASPsych JPS

American Anthropological Association, Memoirs. Menasha, Wise. xxxv Congress internacional de americanistas, México, 1962. Actas y memorias. México, 1964. African Studies. Johannesburg. Acta Lingüistica Hafniensia. International Journal of Structural Linguistics. Copenhagen. American Anthropologist. Menasha, Wise. Anales del Instituto Nacional de Antropología e Historia. México. Anthropological Linguistics. Bloomington, Ind. Anthropos. Revue internationale d'ethnologie et de linguistique/Internationale Zeitschrift für Völker- und Sprachenkunde. Freiburg, Switzerland. Australian and New Zealand Association for the Advancement of Science. Archiv Orientální. Praha. Acta Orientalia Academiae Scientiarum Hungaricae. Budapest. Bureau of American Ethnology, Bulletin. Smithsonian Institution, Washington, D.C. Bibliotheca Orientalis. Leiden. Bulletin of the School of Oriental and African Studies, University of London. London. Brno Studies in English. Brno. Central Asiatic Journal. The Hague and Wiesbaden. Current Anthropology. A World Journal of the Sciences of Man. Chicago. Cahiers Ferdinand de Saussure. Geneva. Comparative Studies in Society and History. An international quarterly. The Hague. Current Trends in Linguistics, ed. by Thomas A. Sebeok. 14 vols. The Hague. (CTL 3 = Theoretical Foundations, 1966; CTL 4 = Ibero-American and Caribbean Linguistics, 1968; CTL 10 = Linguistics in North America, 1973; CTL 11 = Diachronic, Areal, and Typological Linguistics, 1973.) Etnakeele Seltsi Aastaraamat. Tallinn. Foundations of Language. International Journal of Language and Philosophy. Dordrecht, The Netherlands. Folia Lingüistica. Acta Societatis Linguisticae Europaeae. The Hague. General Linguistics. Lexington, Ky. Harvard Journal of Asiatic Studies. Cambridge, Mass. Handbook of Middle American Indians, ed. by G. R. Wiley. Vol. 5, Linguistics, ed. by Norman A. McQuown, 1967. Austin, Texas University Press. Indogermanische Forschungen. Zeitschrift für Indogermanistik und allgemeine Sprachwissenschaft. Berlin. International Journal of American Linguistics. Baltimore. Indiana University Publications in Anthropology and Linguistics. Bloomington and The Hague. Janua Linguarum. Series maior, minor, practica, and critica. The Hague. Journal of the American Oriental Society. New Haven, Conn. Journal of Abnormal and Social Psychology. Washington, D.C. Journal of the Polynesian Society. Wellington, N.Z.

X JPSocPsych JRSNSW JSocI Kadmos KZ Langages Lg Lingua Linguistics L & S Minos MSFou MSLL MT NTS Oceania OL OLM Orbis PAPS PhJL Phonetica PICL 8

PICL 9

PJS PL PRSNSW PSR RLaR RLR RMEA RomPh SA Science SCL SG SIL SJA SL S1PR SO SS SSlav

MASTER LIST OF ABBREVIATIONS Journal of Personality and Social Psychology. Washington, D.C. Journal and Proceedings of the Royal Society of New South Wales. Sydney. Journal of Social Issues. Society for the Psychological Study of Social Issues. Ann Arbor, Mich. Kadmos. Zeitschrift für vor- und frühgriechische Epigraphik. Berlin. Zeitschrift für vergleichende Sprachforschung auf dem Gebiete der indogermanischen Sprachen, begründet von A. Kuhn. Göttingen. Langages. Paris. Language. Journal of the Linguistic Society of America. Baltimore. Lingua. International Review of General Linguistics / Revue internationale de linguistique générale. Amsterdam. Linguistics. An international review. The Hague. Language and Speech. Teddington, Middlesex. Minos. Revista de filología egae. Salamanca. Mémoires de la Société Finno-ougrienne. Helsinki. Monograph Series on Language and Linguistics, Georgetown University. Washington, D.C. Mechanical Translation. Cambridge, Mass. Norsk Tidsskrift for Sprogvidenskap. Oslo. Oceania. A journal devoted to the study of the native peoples of Australia, New Guinea, and the islands of the Pacific Ocean. Sydney. Oceanic Linguistics. Special Publication. Pacific and Asian Linguistics Institute. University of Hawaii. Honolulu. Oceania Linguistics Monographs. Sydney. Orbis. Bulletin international de documentation linguistique. Louvain. Proceedings of the American Philosophical Society. Philadelphia. Philippine Journal for Language Teaching. Quezon City. Phonetica. Internationale Zeitschrift für Phonetik / International Journal of Phonetics. Basel and New York. Proceedings of the Eighth International Congress of Linguists / Actes du Huitième Congrès International des Linguists, Oslo, 5-9 August, 1958. General editor, Eva Sivertsen. Oslo, Oslo University Press, 1960. Proceedings of the Ninth International Congress of Linguists, Cambridge, Mass., August 27-31, 1962. Ed. by Horace G . Lunt. Janua Linguarum, series maior 12. The Hague, Mouton, 1964. Philippine Journal of Science. Manila. Pacific Linguistics. Series A, B, C. Canberra. = JRSNSW Philippine Sociological Review. Manila. Revue des langues romanes. Montpellier. Revue de linguistique romane. Lyons and Paris. Revista Mexicana de Estudios Antropológicos. Mexico. Romance Philology. Berkeley and Los Angeles. Scientific American. New York. Science. Journal of the American Association for the Advancement of Science. Washington, D.C. Studii si Cercetàri Lingvistice. Bucharest. Studium Generale. Berlin, Göttingen, and Heidelberg. Studies in Linguistics. Buffalo, N.Y. Southwestern Journal of Anthropology. Albuquerque, N . M . Studia Lingüistica. Revue de linguistique générale et comparée. Lund. Slavistic Printings and Reprintings / Slavistische Drukken en Herdrukken. 's-Gravenhage. Studia Orientalia, edidit Societas Orientalis Fennica. Helsinki. Slovo a slovesnost. Praha. Studia Slavica Academiae Scientiarum Hungaricae. Budapest.

MASTER LIST OF ABBREVIATIONS

TAPS TCLP TPhS TRÜT UAJb UAS UCPAAE UCPL UUÂ VFPA Vir VJa Word ZAS ZPhon

XI

Transactions of the American Philosophical Society. Philadelphia. Travaux du Cercle Linguistique de Prague. Prague. Transactions of the Philological Society. Oxford. Tartu Riikliku Ülikooli Toimetised. Tartu. Ural-Altaische Jahrbücher. Wiesbaden. Indiana University Publications, Uralic and Altaic Series. Bloomington, Ind. University of California Publications in American Archaeology and Ethnology. Berkeley and Los Angeles. University of California Publications in Linguistics. Berkeley and Los Angeles. Uppsala Universitets Arsskrift / Recueil de Travaux publié par l'Université d'Uppsala. Uppsala. Viking Fund Publications in Anthropology. Wenner-Gren Foundation for Anthropological Research, Inc., New York. Virittäjä. Kotikielen seuran aikakauslehti. Helsinki. Voprosy jazykoznanija. Moskva. Word. Journal of the Linguistic Circle of New York. New York. Zeitschrift für allgemeine Sprachwissenschaft. Berlin. Zeitschrift für Phonetik, Sprachwissenschaft und Kommunikationsforschung. Berlin.

PART ONE

INTRODUCTORY

THE HISTORY OF L A N G U A G E CLASSIFICATION

R. H. ROBINS

The three methods of language classification comprised in the title of this volume, diachronic, areal, and typological, are recognized by modern scholarship as three legitimate and fruitful approaches to the comparative study of languages. It would be a source of satisfaction to be able to say that linguists have now overcome the confusions that have in the past beset comparative work. Modern linguists are more explicit about the three separate systems they use, but it becomes quickly apparent that the same three have in one way or another controlled and characterized language classification from its earliest European days. Progress in scholarship, in this field as in others, has in great part lain in clarifying concepts and in sharpening the distinctions between theoretically separate but factually related methods, rather than in inventing entirely new modes of approach. Logically the three classificatory methods are distinct, because the criteria employed to assign languages to classes in each are quite separate: diachronic (historical, or genetic, or genealogical) classes, or 'families', rest on the assumption of gradual divergence through time from a once more unitary state (cf. Vendryes 1921:349: 'Dire que le français est sorti du latin, c'est dire que le français est la forme prise par le latin en une certain région au cours des âges'); this was crudely expressed by Ross (1950, 1958:28) as 'Two languages are defined as related if and only if they were once one language', and dismissed by Allen (1953:88): 'Brothers if and only if they were once their father', which is a stronger objection to putting a cash value on the literal interpretation of much metaphorical terminology current in this field than it is to the methodology itself. Areal classes group together languages spoken in geographically contiguous areas, and typological classes rest on shared but not universal features, at any level of analysis, that make certain languages significantly alike in structure in some respects. However, while these three modes of classification are theoretically distinct, in fact their resultant classes overlap in membership, as a result of the natural course of events. The difficulties encountered in disentangling the three systems and their proper criteria comes largely from this empirical overlap of theoretically different methods. Languages change in the course of transmission from one generation to the next. In the prolonged absence of a continuing need and opportunity for mutual under-

4

R. H. ROBINS

standing over an entire area, de Saussure's 'force d'intercourse', nothing prevents these changes from diverging further and further, as in the classic case of dialects of spoken Latin passing, imperceptibly at any one time, into the distinct Romance languages. But, without overseas migration or a successful penetration by speakers of a different language (e.g. the splitting of the Slavic speech area by Magyars c. 900 A.D.), historically related languages will continue to be areally contiguous; hence during the process of separation there is the possibility of changes spreading in 'waves' across a historically related community of diverging dialects and producing effects that blur an otherwise clearcut tree structure of historical relationships. Particular words may be individually replaced more readily than phonological and grammatical structures; and not all parts of phonology and grammar change at an equal rate. The period of the 'great vowel shift' in English saw far more radical phonological changes than syntactic changes. In the absence of interference from extensive bilingual or substrate contacts, historically related languages are for some time likely to exhibit various structural and typological features in common. So far empirical grounds have been given for expecting historical classes, genealogical families, of languages to share some areal contiguity and some structural similarity, at least for a certain time. But additionally areal and structural classes are themselves likely to overlap independently of inheritance. It hardly needs saying that intercourse across language divisions is most likely to occur in contiguous areas; and such intercourse, whose extreme developments are a bilingual community on the one hand or the suppression of one language by another in a few generations on the other, is a well known cause of the spread of structural features, phonological and grammatical, as well as some features of lexical structure, over whole regions (Sprachbiinde) of genealogically distant or unrelated languages (Sandfeld 1930, Jakobson 1936, Emeneau 1956). Were this not the case, an areal classification of languages would be without linguistic significance. Jakobson (1957:524) summarizes the three methods: 'The genetic method operates with kinship, the areal with affinity, and the typological with isomorphism.' It is legitimate for any linguist so to define the field of his operation as to exclude a particular type of classification from what he regards as strictly within the province of linguistics (cf. Allen 1953). It is also legitimate, though it may be dangerous, to apply criteria from more than one system of classification at once and to recognize the mixed nature of the resultant classes (as Guthrie 1948). More radically, the terminology of one system may be reinterpreted within the methods of another, as Trubetzkoy (1939a) suggested in a quasi-typological definition of Indo-European. The danger of such a mixture of methods lies in the likelihood of one's assuming that a dilution of criteria need not be accompanied by any weakening of inference. Trubetzkoy was perhaps himself misled in this respect, in his fallacious assumption that IndoEuropean could be simultaneously defined as a typological class and a genetic class having precisely the same membership. Possibly the best example of the near coincidence of geographical contiguity, typological similarity, and genetic relationship is to

THE HISTORY OF LANGUAGE CLASSIFICATION

5

be found in the Bantu languages, on the narrowest interpretation of the 'Bantu line'. In the earliest European tradition of language classification these three systems are all found in operation to some extent. The ancient Greeks showed an almost total lack of interest in the structures and relationships of the non-Greek languages, despite the amount of bilingualism and professional interpreting that must have gone on in trading centers and at the periphery of the Greek world especially in and around the Greek 'colonies'. But at the same time they left records of a more detailed awareness of the dialectal differences within Greek than is available for any other language in antiquity. This is to be accounted for largely by the fact that among the Greek dialects several were the vehicles of literature recognized over the Greek world as a whole, and more than that number were literate, as is attested by the surviving inscriptions. The linguistic and racial unity of the Greek-speaking world was recognized as a factor overriding its political fragmentation and frequent internecine conflicts. Herodotus (8.144.2) writing not long after the Persian wars, makes this, together with much common religious observance, a main ground for a united struggle against the barbarian Persians. Indeed, throughout Greek antiquity the major classification of the known languages of the world was a binary one of Greek versus barbarian. Within Greek, however, a system of dialect classification appears to have been developed quite early in literary history and to have been maintained unaltered in essentials, save for the addition of the Hellenistic koiné, through the Byzantine period, and in fact up to very recent Greek scholarship, in which detailed epigraphic research has been applied to the traditional system, a system which had been based wholly on literary texts and, like so much Greek linguistics, had been greatly motivated by literary studies (Coleman 1963, Hainsworth 1967). The correctness and the adequacy of the Greek tradition are not here a prime concern; but what we see is a definite framework of four (pre-koiné) dialects: Attic, Ionic, Doric, and Aeolic. These dialects were themselves recognized as abstractions based on shared features, from a more delicate recognition of many individual city-state dialects. The terms diàlektos and glossa were in use when this distinction in delicacy was at issue. A Byzantine grammarian (Uhlig 1883:302-3) wrote: 'istéon dè hóti diaphérei diàlektos glosses, hóti he mèn diàlektos emperiektiké esti glósson: Doris gàr diàlektos mia, hyph'hén eisi glossai pollai, Argeiòn, Lakonòn, Syrakosiòn, Messénòn, Korinthión ; kaì Aiolìs mia, hyph'hén eisi glossai pollai, Boiòtón kaì Lesbión kaì àllón' (Note that dialect differs from subdialect in that a dialect embraces more than one subdialect; Doric is one dialect, under which there are many subdialects, those of the Argives, the Laconians, the Syracusans, the Messenians, and the Corinthians; and Aeolic is one dialect, under which there are many subdialects, those of the Boeotians, the Lesbians, and others). The historical origin of the pre-koiné dialects in the movements of separate divisions (éthne) of the Hellenic people was recognized in the tradition of Aeolos, Doros, and Xouthos, sons of Hellen and the founders of the Aeolic, Doric, and Attic-Ionic com-

6

R. H. ROBINS

munities (Uhlig 1883: 462-3). This tradition went back as far as Hesiod (perhaps c. 800 B.C.). Plutarch (Moralia 9.15.747) quotes a fragment: 'Héllènos d' egénonto philoptolémou basilèos / Dórós te Xoùthós te kaì Aiolos hippiochàrmès' (Of Hellen the warlike king there were born Doros and Xouthos and the horseman Aeolos). The close kinship of Attic and Ionic was acknowledged, with Attic elevated to the status of a principal dialect, one suspects, because of its literary and cultural prestige. The special case of the koiné was admitted, but otherwise the picture of the Greek dialect situation was a static one, of a fixed classificatory tree; regional types of speech that did not fit the system were ascribed to racial mixtures from within the recognized subdivisions of the Hellenic stock (e.g. Thucydides 6.5, on the dialect of Himera in Sicily). Where the Byzantine grammarians referred to the phonological differences between dialects they stated these descriptively by reference to spelling conversion rules in relation to the koiné (e.g. Uhlig 1883:464: 'he Atthìs trépei tò s pei mèn eis t hoion thàlatta pei dè eis x hòion xymphórà'; 466: 'he Doris tói a antì toù e chrètai, hoìon hamérà' [The Attic dialect turns s in some words into t, as in thàlatta ('sea', koiné thàlassa), but in others into x, as in xymphórà ('misfortune', koiné symphórà); the Doric dialect uses à instead of é, as in hamérà ('day', koiné hémérà)]). There is no suggestion of any historical presuppositions or of historical ordering. No such classificatory frame was applied to the barbarian languages, the other term in the over-all Greek system. Strabo, the first century B.C. geographer, stated the Greek dialect system in its accepted form (8.1.2); but in reference to the non-Greek languages of Asia Minor he goes no further than suggesting the appropriateness of neighboring languages being alike in various respects, without further comment (1.2.34). Despite the great interest evinced by Greek and Latin scholars in etymology, this was never seen in a really diachronic perspective, but rather as a means of discovering or inferring the correct meaning of a word from its relations with other words in the language (cf. Uhlig 1883:14: 'etymologià estìn he anàptyxis ton léxeòn di'hes tò aléthès saphènizetai' [Etymology is the unfolding of words by means of which their true meanings are made plain]). Examples of ancient etymologizing, which persisted from Plato until the Middle Ages, are notorious and incidentally cast a wholly unfair light on ancient competence in the west in linguistic studies generally. The Romans might have enjoyed a linguistic advantage over Greek scholars, in that they were more interested in Greek than most Greeks were in Latin; and some comparative studies were made between Greek and Latin structures, such as those of Varrò (first century B.C.) and Quintilian (first century A.D.) on the case systems, and that of Macrobius (c. A.D. 400) on the Greek and Latin verbs. Varrò was probably the Latin scholar best disposed towards the historical treatment of languages, and he treated Latin at length both historically and descriptively in his partially extant De lingua Latina. But he, like all other Roman scholars, misinterpreted the (West) Greek origin of the Roman alphabet and the one-sided dependence of Roman art and literature on Greek sources so as to derive the Latin language from Greek dialects with

THE HISTORY OF LANGUAGE CLASSIFICATION

7

barbarian admixtures (cf. Collart 1954: chapter 3). All Latin words bearing an obvious formal resemblance to a semantically comparable Greek word were treated by Varro as direct inheritances from Greek, the relation of domus to domos 'home' being for him just the same as the relation of feretrum to pheretron 'bier' (5.160, 5.166). Secular antiquity showed no signs of further progress in language classification, either in the Latin-speaking west or in the Greek-speaking east. But the recognition of Christianity as the official religion of the Empire and the study of Biblical literature introduced new influences. Most obviously, Hebrew and afterwards Arabic in the later Middle Ages presented themselves as languages unlike Latin and Greek but not to be dismissed as mere barbarian tongues; but more importantly, western scholars became aware of the Hebrew system of language classification such as is preserved in Genesis, chapters 10 and 11, wherein the nations and speech communities known to the ancient Hebrews at the time were classified under three lists as being descended from the three sons of N o a h : Shem, Ham, and Japheth. The integration of this Biblical system into the existing and developing western scholarship determined the course of language classification for centuries; and it must be regarded as part of the more general process whereby secular knowledge and thought were assimilated within a Biblical framework, as is seen, for example, in the reconciliation of sacred and profane history and chronology by such Christian writers as Augustine of Hippo and Orosius. The most important figure in the formalization of early Christian and subsequent mediaeval doctrine on the classification of languages was Isidore of Seville (seventh century), who in part followed Augustine of Hippo. Hebrew was the first language spoken on earth, and Hebrew together with Latin and Greek constituted the three principal and sacred languages of mankind (Origines 12.1.2). Each was distinct typo logically, as regards its predominant phonetic characteristics, Hebrew being 'guttural', Greek 'palatal', and Latin 'dental' {Ibid. 9.1.8); Isidore's first observation here may be seen as a natural reaction of a Latin writer to the phonemically distinct velar and uvular consonants of Hebrew. Isidore established the Japhetic branch of the seventy-two languages ordained by G o d (Origines 9.2.2, 9.2.37) as the parent of the languages of Europe, building on the Biblical inclusion of Gomer (the Cimmerians) and Javan (the Ionian Greeks) within this group of languages, which in general lay to the north and west of Palestine. The Spaniards, for example, were descendants of Tubal, a son of Japheth (Origines 9.2.29). This system of classification, like that of the Greek dialects, was historical in form, tracing back the separate languages to ancestral figures and seeing the dispersal and spread of populations as the carriers of linguistic divisions. It was, however, largely static history. As with the rest of the early creation story in the Old Testament, the languages of the world, like its plant and animal species, were made by God in their present and permanent form, once for all time. Languages from outside the Biblical

8

R. H. ROBINS

text were fitted into it, without the suggestion of a continuous process of linguistic change and development. Somewhat similar and based on the same felt need to integrate sacred and profane knowledge were the various lists of languages and peoples which are known f r o m the third to the seventeenth century. These lists mostly contain seventy-two names, though there is a slight variation on either side, and they are explicitly referred to the postBabel dispersion of the descendants of N o a h . The numbers are, however, the feature of these lists most faithful to the Biblical tradition; languages and peoples unknown to the Bible story are included in them, at the expense of others. The first list includes the British, and a list of 1650 includes Japanese, Chinese, and 'Brazilian'. These lists and the tradition that they represent have been severely treated by later scholarship (Pott 1863:59-62, Borst 1957-63:931-2). But they do expound, for all their excentricities and failings, part of the long enduring conception of the languages of the world within and against which contemporary and later attempts at more genuinely historical classifications had to work. U p to the nineteenth century any study of the classification of languages had to take the Biblical tradition into account. Different people treated it differently; it was attacked, defended, and adapted; what was not possible was to ignore it. As late as 1863 Pott found it worthwhile to publish a book expressly aimed at demolishing an interpretation of it put out only two years earlier (Pott 1863, Kaulen 1861). This was not an isolated occurrence in the history of western culture; it must be seen in the same context as the struggle of the catastrophists and the uniformitarians in relating geological discoveries to the tradition of the Flood and ultimately disassociating them f r o m a strictly literal historicist interpretation of the early chapters of Genesis (cf. Greene 1959). Pott's defense of scientific evidence against a traditional system of explanation must be related to the near contemporary conflict of conscience suffered by nineteenth century thinkers in having to reconsider fundamentally man's place in the living world in the light of Darwin's evolutionary arguments. During the long span of years between the later Middle Ages and the end of the eighteenth century one observes a number of strands of thinking about the languages of the world and their interrelations, and the ways in which by virtue of these they may be classified. Two major factors predominate: the progressive discovery of extra-European languages not hitherto envisaged and the development of a definite historical perspective on linguistic relationship. The Biblical tradition made no allowance for newly discovered languages, and its historical organization, like the rest of the creation story, was on a once and for all basis. It would seem that from the time of Dante the study of the Romance languages in relation to Latin provided the main stimulus to historical classification. Their affiliations were obvious, and unitary Latin survived as a lingua franca of learning and in the usage of the Roman Church, still universal in western Europe. One notes f r o m his De vulgari eloquentia that Dante (1265-1321) is clearly conscious of the major divisions and their subdivisions within the languages derived f r o m Latin. But

THE HISTORY OF LANGUAGE CLASSIFICATION

9

a century before, Roderic, Archbishop of Toledo, drew up the first list of specifically European languages and assigned just one, namely Latin, to the whole western Romance area (Schott 1603:29). This difference of centuries reflects not an actual linguistic fact but the change in linguistic awareness on the part of Dante and the century he lived in. In the Renaissance, with the rise in status and prestige of the vernacular languages of Europe, definite historical links were traced in the differences in word forms and in grammatical constructions between parent Latin and its modern offspring (e.g. the replacement of separate case forms by prepositional phrases). Etymology was now .envisaged in a more diachronic sense, and the new historical perspective afforded greater security in etymological studies. Quite apart from divine intervention, languages could be seen to have changed in the course of time as the result of natural causes, in particular through the movements of peoples; the men of the second millennium were well aware of the part played by invasions and Völkerwanderungen in the formation of modern Europe. The languages of Europe and the rest of the world so far as they were known and taken into account were classified on a historical basis, linked in one way or another with the Biblical divisions of the descendants of Shem, Ham, and Japheth. By the seventeenth century American-Indian languages as well as old world languages were being assigned a specific lineage within this system (Alsted 1650:251-66). This continued into the eighteenth century (several references in Pott 1863:178-80); but Thomas Jefferson (1784:181-2) kept an open mind on the ultimate origin of the Indian tribes of the continent; in an eloquent passage he set out what were later to be taken up as the methods of historical research into the native languages of the Americas and their genetic classifications, and he appealed for support for just the sort of programs now being undertaken by American universities and other agencies: 'Were vocabularies formed of all the languages spoken in North and South America, preserving their appellations of the most common objects in nature, with the inflections of their nouns and verbs, their principles of regimen and concord, and these deposited in all the public libraries, it would furnish opportunities to those skilled in the languages of the old world to compare them with these, now, or at any future time, and hence to construct the best evidence of the derivation of this part of the human race.' Classifications were basically genealogical, diachronic, tracing different languages back to a common source in the Biblical tradition. But there was also a general feeling that such historical groups or families would show some sort of structural or typological kinship as a continuing mark of their common origin; this could be both in the form of obvious similarities in semantically associated words (there was scarcely as yet, prior to the nineteenth century, the more sophisticated concept of systematic correspondences of which mere similarity is only a special and superficially obvious case), and of similarities in grammatical structure and in pronunciation. The thirteenth century Bavarian Chronicle (Pertz 1879:221) distinguished the lan-

10

R. H. ROBINS

guages descended from Shem as guttural, those descended from Ham as palatal, and those descended from Japheth as dental. In this system Greek was classed as Semitic, Latin, German, and the languages of northern Europe as Japhetic, and the Slavic languages as Hamitic ('Horum generationes ex consono idiomatum vel loquelae discernuntur'). The designation of the Slavic languages as predominantly palatal presumably relates to their phonemic exploitation of the 'soft' and 'hard' and nonpalatal) consonant series ('filii Cham [loquuntur] in palato, ut Rutheni et Slavi'). This assumption of congruence between typology and history has already been seen in a somewhat different form much earlier in the formative work of Isidore. Three centuries after the Bavarian Chronicle, J J . Scaliger (1540-1609), in setting up his four major language families of Europe, the modern Greek, Romance, Germanic, and Slavic groups, consistently denied any agreement either in word form or grammatical structure between them (1610:119-22: 'Matricum vero inter se nulla cognatio est, neque in verbis neque in analogia'). Akin to this structural assumption went the quest for a single diagnostic feature, synchronically available as evidence of genetic affiliation. Scaliger's choice is well known; each of his four groups (matrices linguae) show common forms for 'God' in their member languages, which are, however, distinct and unrelated to each other: theos, deus, godt, and boge. Earlier Dante had distinguished the Germanic group by their use of forms related to id for affirmation, and he had further subdivided the neo-Latin group into three subgroups by their use of si, oc, and oil(De vulgari eloquentia 1.8). The relation of languages, especially the modern languages of Europe and of the newly discovered lands beyond the seas, to the Noachian tradition differed somewhat from person to person and from age to age. Isidore's recognition of the languages of Europe being in general Japhetic maintained itself, though we have seen an example of the Slavic group being assigned, perhaps partly from political motives, to the descendants of Ham. As long as Hebrew was regarded as the first language on earth, and as the language of God and Adam, a monogenetic theory of linguistic origins had to be preserved. Such an assumption was common to Dante and to Bibliander (1504— 64), among many others. But monogenesis was compatible with other claimants for primogeniture; Goropius's assertion that Dutch-Flemish ('Cimmerian') represented the surviving original of the Germanic languages and moreover the first language of mankind, untouched by the post-Babel dispersion (Goropius 1569) was only one of a number. Mylius (1563-1637) made similar claims, partly with the patriotic desire to raise at least one European vernacular language to the same level as the three sacred tongues, Hebrew, Greek, and Latin (Metcalf 1953). J. Webb (1611-72) argued (Webb 1669) that Chinese was Adam's tongue, and in a suitable reinterpretation of Biblical history maintained that the Chinese speakers had split off from the rest of mankind before the building and the destruction of the Tower of Babel. Webb argued structurally; Chinese, being monosyllabic and of the simplest possible structure (far simpler than Hebrew), must clearly be the primal language. This is just one facet of

THE HISTORY OF LANGUAGE CLASSIFICATION

11

the great impression made on European linguistic thought in the seventeenth century by the discovery of Chinese and of its allegedly unique and perspicuously simple structure; Wilkins's Essay towards a real character and a philosophical language and other attempts at a universal language were in part inspired by admiration for what people thought they saw in the grammatical organization of classical Chinese. Others, rejecting the claims of Hebrew as known from Biblical texts to be the oldest language, rejected also a monogenetic account of the languages of the world as being unreasonable in the light of the evidence available. J.J. Scaliger refused to go back further than the various families (matrices linguae) that he set up. J.Perizon (16511715) and P. F. Suhm (1728-98) declared that the pre-Babel language was for ever lost and that all the known languages of the world had arisen since then, partly by natural causes (Perizon 1736:175-82, Suhm 1769:83). Leibniz (1646-1716) did not explicitly rule out monogenesis, but he firmly placed Hebrew, as just one language like all the others, into one of the two major classes of language into which he divided the languages of the world, 'Aramaic' or 'Arabic' (Semitic-Hamitic), languages of the south, and 'Kelto-Scythian' (Japhetic), languages of the north (Leibniz 1765:18-20, Arens 1955:79-81). Leibniz's classification was in the main shared by H.S. Reimarus (1694-1768; 1781:93-8) and A.L. von Schlozer (1735-1809). Both were radically critical of the Biblical tradition, though they were monogenesists, Schlozer on the ground that God would do his work of creation in the most economical manner and that the splitting and multiplication of languages was a product of natural causes (Schlozer 1785:35, 149). Reimarus (1751:300-43) held that the story of Babel was allegorical rather than historical, and he rejected the traditional number of the post-Noachian languages. This number had caused trouble all along, but the strength of the Biblical tradition had made it necessary either to try to reconcile it with the facts or to give grounds for its rejection or modification. One of the latest lists of languages to preserve the Biblical number of seventy-two, by J. H. Alsted (1588-1638), comprises a set of languages bearing little resemblance to those set out in Genesis 10-11. Up to this point we have seen the development of a genuinely historical conception of linguistic relations and a system of classification based thereon. But the elaboration of the full theory and method had to wait until the cumulative work of nineteenth century linguistics had borne its fruit. Indeed, one of the features of Renaissance and later pre-nineteenth-century historical linguistics was its rather fragmented character. Scholars observed and considered the vocabularies and the grammars of languages, and they paid particular attention to the relations between classical Latin and the Romance languages, but one does not find the connected chain of scholarly work in this field of linguistics, whereby each writer takes cognizance of the work of his contemporaries and immediate predecessors, such a striking characteristic of nineteenth century historical linguistics from Rask to Brugmann. During the seventeenth and eighteenth centuries a different set of ideas was developing among those thinking about languages, in the direction of classification by

12

R. H. ROBINS

reference to structural type, on a basis different from that of etymological derivation, whatever inferences may have been drawn (then or since) from the one system to the other. This can be ascribed to a number of factors. In the first place the sheer volume and structural diversity of languages of which Europeans became aware after the end of the Middle Ages made a profound impression on a tradition of scholarship hitherto based almost exclusively on Latin and Greek, and later on Hebrew and Arabic, although the different structure of these two languages as compared with that of Latin had already been noticed (Reuchlin 1506). In particular the grammar of classical Chinese, though its proper analysis was misunderstood, and the structures of some of the languages of North and South America as reported to Europeans revealed new ranges of grammatical organization, the one supposedly of an extreme simplicity, the others of a frightening complexity. Secondly the realization of the composition and history of the Romance language family showed people how in the course of centuries the more obvious aspects of grammatical structure could change and produce radically different types of language, as case inflection gave way to prepositional constructions with invariant noun forms and the variety of Latin tense and person inflections were in large part replaced by auxiliary verbs, periphrases, and the unmarked presence of subject pronouns preceding verbs. This consciousness of the possible extent of structural change was made all the greater by the fact that the Latin that was studied in historical relation with the Romance languages was classical literary Latin rather than the more directly ancestral colloquial ('Vulgar') Latin, which shared several of the features typical of the later Romance languages, for example a greater grammatical fixity of word order and the more extensive syntactic use of quod as a conjunction instead of the accusative and infinitive type of construction so typical of classical Latin and so untypical of the descendants of colloquial Latin. From a somewhat different point of view, the liberation of modern English grammar from the Priscianic framework, the work of a succession of empiricist-inspired English grammarians, of whom Wallis (1616-1703) was the most radical and is perhaps the best known, contributed to the same effect. Here again since Latin and the categories of Latin grammar were always present in the minds of such grammarians, the similar contrast between the Latin structural type and the structural type of English when subjected to an unbiased formal analysis brought to the forefront of men's attention the possible differences of typology between one language and another. This increasing awareness of the typological or structural diversity of languages coincided with the rather general eighteenth century interest in the classification of nature and of the natural products of the earth. This was part of the rise of natural science that took place in the seventeenth and eighteenth centuries and as much as any other movement of thought helped to bring into being the intellectual stance of our own contemporary world. Of course systems of classification of one sort or another are as old as any form of science, and a comprehensive classification of living creatures was made by Aristotle (W. D. Ross 1923:114—7) and passed on through

THE HISTORY OF LANGUAGE CLASSIFICATION

13

the Middle Ages as the Scala naturae. But eighteenth century naturalists felt the need and the possibility of a general systematization of all nature, whose diversity had been so much increased as the result of geographical expansion and the discovery of fossil remains. The century saw several classifications of nature, of which probably Linnaeus's (1707-78) Systema naturae and Philosophia botanica are the best known and have left their author's name recorded in the Linnaean system of plant nomenclature and in the Linnaean Society in London. Linnaeus's classifications were based on the empirical observations of attributes considered to be of particular significance or importance in the economy of nature. He declared (1775:249, 331-2, 438; 1787:lxviii) that in his work he acknowledged 'no authority but that of inspection' (in the case of plants giving prime attention to the reproductive organs); such an empiricism was not considered as necessarily contrary to piety, for he went on to say 'they [the parts of plants] are written by the hand of God; it should be our study to read them'. Buffon (1707-88) also stressed the necessity of observation and comparison between species and species in natural history (1785a:216, 1785b:62-3). Classification was carried through by reference to the structure, external form, and arrangements of the parts of whatever was under consideration. Such work and the theory lying behind it occupied some of the greatest minds of the period. It was held to be compatible with the view of nature as the creation of fixed species (Cuvier (1769-1832) and Linnaeus, with reservations), and with an evolutionary view envisaging species developing and changing through the course of time (Lamarck [1744-1829] and Charles Darwin [1809-82]). Buffon's concern (Greene 1959:142-60) over the rival claims of a fixed creation and the apparent evidence for changes in species over the years, in part deliberately induced by man's actions, is typical of the intellectual agony that resulted from the impact of scientific observation on received authority during this period. Today, with these passions largely spent, it is hard to reconstruct such a climate of tension, but it dominated scientific and religious thinking, particularly as regards man's place in nature, until well into the nineteenth century. The classification of nature could not be wholly divorced from the history of nature, though historical considerations could be stressed in varying degrees. Kant (1785), appropriately as one of the leading intellects of the eighteenth century, explicitly distinguished natural description (Naturbeschreibung) from natural history (Naturgeschichte), each leading to its own form of classification. In the study of language we find attitudes corresponding to those of eighteenth century natural science. In addition to the historical perspective already considered, systems of classification by structural type, comparable to the classification of natural products by the structure and arrangement of their parts, made their appearance. Earlier generations had tended to assume that historically related languages would exhibit structural similarities, but by the eighteenth century the evidence of Romance history had shown that this was not a necessary inference; despite the ease with which

14

R. H. ROBINS

individual lexical items could be replaced, over the course of centuries structural type could change radically while leaving a substantial core of inherited roots and inflectional morphemes. Though typological and genetic classifications continued to interact on one another, some people were prepared to disassociate them. In the entry 'langue' in the French Encyclopédie (Diderot and d'Alembert 1772) the writer distinguished two fundamental types of language, 'langues analogues' and 'langues transpositives', the distinction being based on the general character of their grammatical structure. Analogical languages were those with relatively little syntactically relevant morphology, and having a word order analogous to the order in the thought expressed; transpositive languages were those in which the grammatical construction of sentences involved a good many differences in the inflected forms of words. Hebrew, French, Spanish and Italian were said to be analogical languages, Latin and ancient Greek being transpositive languages. German was also transpositive, but under the influence of neighboring European analogical languages it had become less markedly transpositive than formerly. For prudential or other reasons, the writer accepted the Biblical tradition of the Tower of Babel, the dispersion of languages, and the primogeniture of Hebrew, which as the oldest language appropriately exhibited a primitive, analogical character, a character to which the Romance languages had now returned. From a modern point of view this binary distinction in grammatical typology was in some respects unclear and crudely drawn; the important point, however, that the article set itself to make was that the 'filiation' of the Romance languages with Hebrew, on the basis of their structural similarity, was of far more significance and certainty than their etymological connection with Latin, because syntax was 'le génie principal de langue', and on this count Latin was declared to be 'étrangère' to the modern Romance languages. During the earlier years of the nineteenth century, when genealogical classification was itself making rapid strides towards its present day position, these typological classifications were being actively debated, not at first with direct reference to historical relationships. F. von Schlegel maintained a binary distinction similar to the one set forth in the Encyclopédie but based on a wider range of structurally diverse languages, between those in which the finer semantic specifications associated with plurality, time reference, and other grammatical categories were indicated by an alteration in the form of the word root, and those in which such specifications were indicated by the addition of specific words to it (1808:45: 'Entweder werden die Nebenbestimmungen der Bedeutung durch innere Veränderung des Wurzellautes angezeigt, durch Flexion; oder aber jedesmal durch ein eignes hinzufügtes Wort, was schon an und f ü r sich Mehrheit, Vergangenheit, ein zukünftiges Sollen oder andre Verhältnissbegriffe der Art bedeutet.'). These two types, he said, were absolutely basic; all other typological species of language were only modifications of one or the other. He allowed for the division'of non-inflecting languages into those making use of affixes and those lacking all grammatical

THE HISTORY OF LANGUAGE CLASSIFICATION

15

form, thus finding a place for the class of agglutinating languages. Languages of this sort, that rely on affixation, lay between the purely unstructured languages like Chinese and the fully inflected ones like Sanskrit, Latin, and Greek, but the distinction between independent words and independent affixes as the means of grammatical designation was of secondary importance (46-9). Schlegel envisaged the possibility of a historical progression in structural types, recognizing the stage reached by languages like Arabic (48), where the affixes had begun to lose their individual identity and so to pass towards the status of inflections. He also recognized the opposite process in the loss of inflections by the modern Romance languages and Germanic languages in relation to their earlier structural type (56). But he made it clear that while as structural types the inflecting languages were the best developed, individual languages of other types can and do produce their own excellences (e.g. Chinese [49] and Hebrew [55]). Moreover, again as exemplified by Chinese, there was no universal movement in typology; and, of great importance in the subsequent history of language classification, he clearly separated typological classes from genetic classes established by formal correspondences in the composition of 'roots'. Thus he recognized (46) many totally unrelated languages on the American continent, but like most later writers in the first half of the nineteenth century he grouped them all together into one typological class of a complex but still unsystematic affix-using type. The now familiar threefold typological classification into the co-ordinate classes of isolating, agglutinating, and inflecting languages was set out soon afterwards by F. von Schlegel's brother, A.W. von Schlegel, again as a non-historical system of classes (1818:14): 'Les langues qui sont parlées aujourd'hui et qui ont été parlées jadis chez les différents peuples de notre globe, se divisent en trois classes: les langues sans aucune structure grammaticale, les langues qui emploient des affixes et les langues à inflexions'. A.W. von Schlegel, like his brother, made the point that the difference between affixing and inflecting languages was not simply a matter of whether serially placed affixes were used or not in the surface morphology, since a good many affixes were found in the inflecting languages. What was relevant to the distinction was whether the affixes themselves had a definable and isolatable meaning or whether as affixes they were devoid of identifiable content in themselves and only served to form a complete inflected word unity with its totality of specified meaning (15). This latter distinction turns largely on the number of formally distinct affixes and the uses to which they are put. In typically agglutinative languages each affix has a specific single category and range of associated meanings assigned to it; in inflectional languages a single affix form may have a variety of entirely separate categories and meanings according to the class of word stem with which it is being used (e.g. Latin -es may be nominative singular, nominative or accusative plural, second personal singular indicative, or second person singular subjunctive). This last class of languages he called 'organic' (1818:15), and in a manner increasingly found in succeeding generations he regarded such languages as the most superior in internal structure. Thereafter A.W. von Schlegel (1818:16-17) subdivided his organic (inflectional)

16

R. H. ROBINS

languages into synthetic and analytic, according to their recourse to usages such as prenominal articles, preverbal pronouns, and auxiliary verbs, in preference to extensive tense inflection and person inflection, and prepositional constructions in place of case inflection. The passage from synthetic organic to analytic organic might be a matter of history within a single language family as was instanced by the comparison of Latin with its Romance descendants. A.W. von Schlegel's system of classification was a more complex one than that of his brother, and in particular quite a wide range of languages could be included in the membership of his organic languages. Certainly at this time a language structure that could be so designated was to be favored, when a powerful movement of European thought was seeing the highest ideal of science in the conception of an organized whole that was so much more than the sum of its parts pieced together ('das zusammenbrennende, zusammentreffende Ganze'). The typological classification of languages, especially the triad isolating, agglutinating, and inflectional (sometimes called 'fusional' later) is traditionally and widely associated with the work of Wilhelm von Humboldt (1767-1835). He did indeed make frequent use of these terms, more especially in the Verschiedenheit des menschlichen Sprachbaues (1836), but his position is less simple and clear-cut than it is often represented as being, for example by Schmidt (1926:24-5). Humboldt spoke of the contrast between isolation, agglutination, and inflection as the cardinal point (Angelpunkt) in the consideration of language structure (1836:114); but elsewhere, rather like F. von Schlegel, he recognized two typological poles only, highly developed inflection as best exemplified in Sanskrit and extreme isolation as exemplified in Chinese. The enthusiasm aroused by Sanskrit and the study of Sanskrit during the early nineteenth century had an effect on linguistic opinion comparable to that of the European discovery of Chinese and European misinterpretation of Chinese grammar in the seventeenth and eighteenth. This misunderstanding persisted in Humboldt; Chinese had, he said (1836:124, cf. 1827:66) no indications of different word classes. Between these two poles agglutination, a kind of hybrid species ('Zwitterwesen'), was not a clearly separated type but just an intermediate stage (1836:124-5, 294). He expressed doubts (298) whether a really comprehensive classification of languages was feasible on this basis, and inclined more to something like a continuous scale of structural types from the most highly organized (Sanskrit) to the least organized (Chinese), with the agglutinating languages (125: 'diese sogenannten agglutinierenden Sprachen') ranged within it (169-75). It is noticeable that Humboldt regarded Hebrew as lying near to Sanskrit and not among the isolating ('analogical') languages as had the writer of the Encyclopédie article. Beside this system based primarily on word form in relation to the expression of grammatical relations and categories, Humboldt also set up a fourfold system of typological classification partly overlapping with the former, having reference to sentence structure : the Chinese type, the agglutinative type, the inflecting type, and the incorporating type, again with languages having the possibility of exhibiting the characteristics of more than one type (1836:272). Languages like Sanskrit showed

THE HISTORY OF LANGUAGE CLASSIFICATION

17

(151) by their word forms the grammatical relations of words within sentences; languages like Chinese had to rely on word order or the presence of other words ; but a separate structural possibility was exemplified in particular among American-Indian ('Mexican') languages (152-3), wherein the grammatical core of the sentence was contained in a single word, more especially by the incorporation of the object noun or pronoun (151-4). Humboldt's description of this type is strikingly echoed by Allen's characterization of a Caucasian language, Abaza (1956:139): 'The typical Abaza sentence simply lists the nominals in the function of "topics", and the semantic relations between them are functions of the verbal complex.' Humboldt's incorporating type is reflected in Duponceau's (1819) category of polysynthesis, whereby 'the greatest number of ideas are comprised in the least number of words', a language type which he and a number of early nineteenth century American linguists very generally took as characteristic of the native languages of the American continent as a whole. That this system is partly independent of the isolating — (eagglutinating) — inflectional scale is shown in another passage where Humboldt wrote (1836:294) that since incorporating languages cannot include all their syntactic relations within single word units they must themselves lie somewhere along this scale. One would agree with Schmidt (1926:25) that the incorporating type does not fit well into the otherwise tripartite system of classification, but it is scarcely fair to ascribe its introduction, as he does, to Pott (cf. Pott 1863:113). So far we have been considering the typological classifications that were developed at the end of the eighteenth century from a purely synchronic point of view just as systems of descriptive classification. Thus, it seems, they were considered by the Schlegels; while they recognized the possibility and the fact of typological change in a language over the course of years, they clearly distinguished this form of language classification from classification into genealogical families on the basis of etymologies. Humboldt too laid his main emphasis on descriptive typology, but he also saw the different types as stages in an evolving sequence. This is certainly how he treated them in his 1822 Entstehen der grammatischen Formen, in which he regarded the order isolating — agglutinating—inflecting as marking a course of development and progress towards the full realization of the potentialities of human language. In 1836:295 he designated the inflecting languages as more advanced ('vollkommnere') ; and he assigned the isolating type to a lower grade (1827:65-6, 69), but at the same time he insisted on the special excellence of the Chinese language, at least within its own type, praising the close correspondence of its sentences with the bare sequence of thought, in terms very reminiscent of the seventeenth century pioneers in search of a 'real character' and universal language (1827:64: 'fondant la construction presqu'exclusivement sur la suite des idées'; cf. Robins 1967:113-15). On the other hand he expressly rejected (1836:295) the equation of more advanced with later in time, less advanced with earlier in time, as if 'Chinese were the earliest and Sanskrit the most recent language', repeating that the scale was one of typology not history ; and in fact he anticipated the position of von der Gabelentz (1901:255-58) and Schmidt (1926:26)

18

R. H. ROBINS

that languages are subject to a constant movement from isolation to inflection and then back, not to the original type of isolation, but to a comparable structural type (1836:169, cf. 258, where he contrasted an imagined form of Sanskrit, that like English had lost almost all its inflecting character, with the persistent and conservatively isolating Chinese). During the late eighteenth century and early in the nineteenth grammatical typology was considered and studied from these two points of view, first as a means of classifying languages by reference to their observed structures and arrangements, in line with other contemporary classifications of natural phenomena in science, and secondly as an accompaniment and consequence of the historical process as it affected language. This latter became the dominant interpretation of structural typology in the eyes of most nineteenth century European linguists. The supremacy of historical linguistics in that century is well known, so well known indeed that historical linguistics is often referred to as traditional linguistics by those who choose to ignore the long tradition of European linguistic thought prior to it, cast largely in the frames of descriptive linguistics and of general synchronic theories of language. The course of historical classification from 1800, largely the work of German or German-trained scholars, is familiar and well recorded (Pedersen 1931: chapter 7, Leroy 1963:15-60, Ivic 1965: chapters 1-12, Robins 1967: chapter 7). Genealogical, historical classification was now the ultimate goal of linguistic science, and etymological relationships of lexically essential root forms and affix forms, systematically established, were the evidence upon which these classifications were founded. Tracing the progress of this aspect of linguistics through Rask, Grimm, Bopp, Pott, Schleicher, and the Neogrammarians one sees the continuous development of method, theory, and resultant classification, for the most part worked out in Indo-European and in the genealogical families bearing some cultural kinship with I-E (i.e. written records of earlier stages of the family, possession of at least one language with an enduring literature, etc.). F. von Schlegel (1808:28) held comparative grammar to be the key to genealogical relationships, and in Bopp's Vergleichende Grammatik (1833), as the title indicates, the elements brought into comparison were primarily the inflectional morphemes of the languages in association with their roots. At this time 'comparative grammar', still in use for example in Brugmann and Delbriick's (1886-1900) magnum opus and in Buck's (1933) classic textbook, was literally the proper designation of the comparative method. Later scholars, notably Pott (1833-36), broadened the method systematically to include the etymological study of whole lexica as far as the evidence permitted. As etymological studies progressed, mere resemblance of forms gave way to the more subtle but much more powerful conception of systematic sound correspondences, and thereafter a stage of scientific maturity was achieved in the insistence of the neogrammarians (as part of method if not as literal matter of fact) that sound changes and therefore sound correspondences must admit of no haphazard, inexplic-

THE HISTORY OF LANGUAGE CLASSIFICATION

19

able, exceptions. In theory, the century saw the concept of a surviving 'oldest' language (still lingering on in the supposed 'antiquity' of Sanskrit as nearest to the Ursprache, held by F. von Schlegel among others) yielding place to the universal recognition of a non-extant 'reconstructed' language, with Schleicher in mid-century laying more stress on actual reconstruction than did the later neogrammarians. In the actual system of classification, the main outline of the Indo-European family, which became the model for other families and the principal field for linguistic research, emerged fairly early in its main membership, Indo-European dating from 1814 and Indogermanisch f r o m 1823. More detailed research placed the subfamilies of I-E in their respective places, notably resulting in the downgrading of Sanskrit from its supposed position near the head of the family to membership of just one branch which itself had undergone at least two major shifts, in the reduction of the earlier vowel system to (at first) three vowels and in the consonantal changes that characterized the divergence of what were later to be called the satem languages (Whitney (1873:204) marks a stage when Sanskrit was not 'mother' but 'eldest sister in the family'). The subsequent discovery of Tocharian and the decipherment of Hittite in this century brought about certain changes in the internal organization of the I-E family. The geographical distribution of the centum and satem branches was no longer as simple as had first been supposed; and after some years of controversy over the relation of Hittite to the rest of the family, leading for a time to the assumption of 'Indo-Hittite', the language has now been generally accepted as a member of the Anatolian branch of I-E (Puhvel 1966). In all this work typology played no major part methodologically. F. von Schlegel had already dissociated it from any necessary connection with genealogical classification; the neogrammarians were not particularly interested in it; and in articles first published in 1914 and 1924 Meillet summarily dismissed typological classification as no more than an 'amusette' (1948:76-7), and declared (1938:53): 'La seule classification linguistique qui ait une valeur et une utilité est la classification généalogique.' 'Nineteenth century historical linguistics' is generally said to have begun with Sir William Jones's 1786 paper, and Hockett (1965:185) recently expressed the view that this was the first of the only four really important breakthroughs in linguistics as a whole. A single passage from just one paper of Jones's numerous writings is reprinted in modern publications and accorded pride of place as having instituted scientific historical linguistics and the historical classification of languages. In a number of ways, however, Jones's position rests rather less on his actual scholarship, great though it was — and not just in historical linguistics as a glance at his collected works reveals — than on the fact that he first brought Sanskrit, both descriptively and historically, to the full attention of European linguists, and did so at the time most favourable for Sanskritic studies and historical linguistics to be taken up with enthusiasm and successfully developed (cf. Hoenigswald 1963). Jones based the historical relations between Sanskrit and the classical languages of Europe on 'a stronger affinity, both in the roots of verbs and in the forms of grammar,

20

R. H. ROBINS

than could possibly have been produced by accident'. But his eulogy of Sanskrit as 'more perfect than the Greek, more copious than the Latin, and more exquisitely refined than either', while perhaps owing most to the work of the Sanskrit grammarians, whom Jones was one of the first to study, scarcely represents a scientifically objective description. Against his anticipation of later methods of historical classification must be set his denial, in the paragraph preceding his most famous quotation, of a descent of Hindi from Sanskrit, despite 'five words in six' being derived from it, because of a typological diversity between the two languages 'in the inflections and regimen of verbs'. His conception of a common source, no longer extant, for the languages he brought into historical relationship, though not yet universally accepted, had. in fact been anticipated in the seventeenth century by the Swedish scholar Jäger (1686).

In wider ranges of linguistic history Jones accepted the Babel tradition, both in the paper just cited and more fully in his "Origin and families of nations" (1799:129-42), where he assigned the nations of the world to the three groups descended from the sons of Noah. Noah's language he considered lost, and he denied any historical connection between the original languages of these three groups. From Japhet, he wrote, sprang the north European and north Asian peoples ('Tartarians'), from Shem the Arabs and Jews, and from Ham the Indians, some of the inhabitants of southern Europe, and probably the rest of the human race. A wider study of Jones's scholarship would show him to have been more a man of his time than the instant discoverer of modern historical linguistic theory and method. His claim to historical renown rests on a wider base than on the repeated but isolated quotation of a single paragraph. At the beginning of the nineteenth century F. von Schlegel had regarded his typological distinction between inflecting and non-inflecting languages as marking 'die beiden Hauptgattungen aller Sprache' (1808:45). As we saw, this mode of classification continued and was further refined by von Humboldt; but with the steady and successful advance of historical linguistics and the great interest that this aroused, typological studies tended in mid-century to become subordinated to the search for the supposed stages in the historical evolution of genetic language families, and ultimately they fell out of general favour, as was evidenced in the opinions quoted from Meillet (1938,1948), until the revival of synchronic studies in the present century. Evolutionary ideas were gaining ground before the publication of Darwin's Origin of species in 1859. Steinthal (1850) set out a system of eight structural types of language, in which the more developed languages of the world represented the progressive development of the potentialities of language ('die Entwickelung der Sprachidee'). His views on structural types, when considered in successional relation to each other, were more akin to those of the earlier German Naturphilosophie, as expounded, for example, by Goethe as a man of science, than with Darwininan evolution; but the movement of nineteenth century thought in this respect, like that of contemporary scientific thought in general, is unmistakable (on Goethe's attitude to biological

THE HISTORY OF LANGUAGE CLASSIFICATION

21

classification and its relation to contemporary linguistics, see W. Haas 1956-57). The classic exponent of the thought of this period was August Schleicher (1821-68), who, whatever the actual sources of his thinking (Hoenigswald 1963:6-8, Maher 1966), towards the end of his life saw the history of languages as closely corresponding to the history of living species set forth by his contemporary, Darwin (Schleicher 1863). He integrated typology and history within one view of linguistic evolution. Just as living species arose from simple living cells and passed through successive stages of organization to produce the variety of species actually found in the world, so all languages must be thought to have arisen from the allegedly simple state still exemplified by Chinese (1863:24, notice the centuries long persistence of this myth of the simple structure of Chinese!), and to have passed, through the agglutination of auxiliary words, ultimately into the highest form of inflectional structure represented by Indo-European in its classical period, before these languages began their period of decline. He likened the supposedly oldest stage of the I-E roots, and the roots of comparable language families, while as yet grammatically undifferentiated, to living cells with the potentialities of even more complex development and differentiation (25-6). Typology was now wholly subordinate to history, even though he recognized (28) that genetically unrelated languages in similar stages of typological evolution often occupy contiguous areas of territory. His attitude is best summed up in his own words (1850:15): 'Nur dadurch wird die Geschichte, das Werden dem Systeme adäquat, dass jede Periode einen Repräsentanten zurücklässt, wodurch eben das Nacheinander der Geschichte in das Nebeneinander des Systems umschlägt.' Schleicher's historical conception of evolutionary typology was true to the spirit of his time, for all its naïveté. Such ideas continued to be expressed during the century, but in 1909 Finck (1909 :6) explicitly denied any historical or evaluative grading in his exposition of linguistic types, and in 1921 Sapir broke decisively with all evaluative typology. The revival of such doctrines and their reduction to ideologically inspired absurdity can be seen in the theories of Marr, once highly spoken of in the Soviet Union until Stalin's ipse dixit put a brusque stop to it all (Simmons 1951, Thomas 1957). With Darwin evolution had finally won the day in natural history over static creation. Schleicher saw the genetic families of languages and their subfamilies as the linguistic equivalents of natural species in the animal and vegetable kingdoms. Synchronic typological classifications had already been recognized, for example by von Humboldt, as very broad ones, difficult if not impossible to elaborate in detailed subdivisions before the development of strict descriptive quantification techniques. But close study of etymological relations and of shared innovations had enabled linguists by Schleicher's time to arrange the the I-E languages into groups and subgroups, ending with living and historically attested dialects as infimae species (cf. Schleicher 1863 : table), and the articulation of the I-E and other genetic families went on throughout the century. Schleicher, himself responsible for the Stammbaum model to display such family relationships (with its well known strong and weak points),

22

R. H. ROBINS

likened his picture of language classes to Darwin's picture of the evolutionary classes of living things (1863:13-16, Greenberg 1957:56-65). The triumph of historical classification was the triumph of etymology. Etymological connections cumulatively worked into sound laws provided the justification for historical relationships and the means of refining and extending them. This reached its climax with the neogrammarians, but their field of investigation was I-E and other families of languages spoken in areas to some extent culturally comparable. In the field of the American-Indian languages very different conditions prevailed: almost complete illiteracy of the languages and consequently no records prior to European immigration (except for the 'classical' languages of Mayan and Aztec civilization), and a vast number of languages spoken by relatively small groups, almost all awaiting descriptive analysis for the first time. Here people argued throughout the nineteenth century over the relevance of typology of over-all structures as against the etymologies of individual words and morphemes for the establishment of historical classifications ('grammar or lexicon' [M. Haas 1969, where, as the whole course of the debate shows, grammar means grammatical structure, not the actual inflectional and derivational morphemes, such as Bopp and others in Europe were concentrating on, which would here fall under lexicon]). In colonial days and at the end of the eighteenth century attempts had been made to classify some of the then known Indian languages of North America by reference to lexical cognations, and, as was noticed above, Jefferson had urged the collection of vocabularies for this purpose from the languages of North and South America, just as Leibniz had earlier encouraged the collection of vocabularies from the diverse languages of the Russian dominions (Arens 1955:85-6). But from the beginning of the nineteenth century, as knowledge of the native languages of North America increased with the western advance of the frontier, American scholars in company with several Europeans came to regard the alleged typological unity of the languages of the whole continent as the most significant basis of classification. Duponceau (1819) used the term polysynthetic to designate the supposed type of American linguistic structure, and this term persisted, later reinforced with the Humboldtian incorporating. But as the century wore on the available facts began to make it clear that American-Indian languages were both less homogeneous in structural type and less peculiar in relation to the rest of the world than had been thought. From 1836 Gallatin carried out and published comparative work on the basis of lexical, etymological studies, declaring that this was the only proper foundation for the establishment of language families within the continent, though he still tried to work within the time span of a literally interpreted Biblical tradition. The rival claims of structural type as against lexical correspondences was argued up to the end of the century, when Powell's comprehensive classification (1890), which has in outline survived very well the revisions of later scholarship, confirmed that 'the evidence of cognation is derived exclusively from the vocabulary' (1891:11). The controversy was not just about the claims of typology versus history in pro-

THE HISTORY OF LANGUAGE CLASSIFICATION

23

ducing differently based even though factually related classifications, as had largely been the case in early nineteenth century Europe, but it concerned the wider question, that was to be debated in other areas later on, of the extent to which structural type could itself be used as a criterion of historical relationship. Gallatin (1836:5-6) had himself agreed with the belief in an over-all unity of structure among the AmericanIndian languages proving an ultimate common origin of all the American languages historically anterior to the appearance of the lexically recognizable separate families. This view was shared by Whitney (1870:346-53), who, however, insisted that the immediate need was to apply to the known American languages the methods already being successfully used in I-E and other established families (for this period of American-Indian scholarship, see M. Haas 1969; for a recent view on the typology of these languages, Hoijer 1954). The turn of the century witnessed both the recognition of maturity in the theory and methods of historical classification, at least as far as I-E and similar families of languages were concerned, and the rise, or rather the renewal, of synchronic studies stimulated and strengthened by the various interpretations and applications of de Saussure's structural conception of language. In historical linguistics, with the neogrammarian ascendancy still at the forefront of people's thinking, Pedersen felt able to write in 1924: 'Comparative linguistics has now attained full maturity and a clear consciousness of its methods and undertakings' (1931:245). Structural attitudes were to make their impact on historical studies, as is seen in such works as Jakobson 1931 and Hoenigswald 1950 and 1960, but these changes did not directly affect the basis of historical classification. A new interest was, however, given to structural, typological classification; greater insights had been made into the nature of linguistic structure at all levels, and linguists now felt free to range beyond the classifications of the previous century, which had dealt with only one or two rather obvious facets of grammatical organization, important and revealing as they had been. It is perhaps an exaggeration to say, with Stankiewicz (1966:495) that 'the development of modern linguistics is largely a history of oscillation between two complementary approaches to the comparison of languages, known respectively as the "genetic", or "reconstructive", and the "typological", or "general linguistic" approach'; but the revival of independent interest in the structural comparison and classification of languages and the developments in the classification procedures employed have been a notable characteristic of twentieth century linguistic work. Twentieth century typology was not only wider-ranging than hitherto; it came to be deliberately independent of historical and geographical implications (although, of course, the contingent coincidences of history, geography, and structure were recognized). Schmidt (1926) sharply distinguished between Sprachfamilien, resulting from genetic classification, and Sprachenkreise, typological groupings based on some set or sets of shared features, perhaps correlating with wider Kulturkreise. The material evidence was different as between the two: for the former 'eine Untersuchung des Sprachmaterials im einzelnen, seiner einzelnen Wörter', for the latter 'eine Reihe

24

R. H. ROBINS

besonders charakteristischer Sprachelemente' (1926:271). But he was still concerned to set up continuous areas of languages showing the extension of particular features, and generally speaking he set out the membership of his Sprachenkreise in terms of the genealogical Sprachfamilien, thus making his two systems of classification less radically distinct than the basis of his working would have permitted. Much more independent of all areal and historical considerations was the rather earlier typology of Sapir (1921). Sapir determined to investigate what he called (127) the 'structural genius' of different languages; he rejected (128), as had von Humboldt, any absolute determinacy of typological classification, such as is necessarily and inherently a feature of genetic classification (cf. p. 29), and reacting against the domination of structural questions by the evolutionary and evaluative conceptions of the previous century he took his ideas on typology far beyond the earlier tripartite and quadripartite schemes, which he felt were too superficial to bring in all that was relevant to the comparative study of linguistic structure. He drew up (chapter 6) three separate considerations which he claimed must all be involved in any satisfactory system of classification, leading to a more complex but more penetrating typology: 1. Types of concepts formally expressed in a language: 1. Roots; 2. Derivatives; 3. Concrete relations; 4. Pure relations (1 and 4 being universal). 2. Languages that primarily develop their morphological processes in relation to derivation, as against those that develop them in relation to inflection. 3. The 'techniques' involved (here belong the traditional distinctions between isolation, agglutination, and fusion (inflection)). Sapir's system is highly interesting, though not easy to follow in detail, in part as the result of the idiosyncracy of his metalanguage, which he developed in advance of the later standardized terminology of American interwar descriptive linguistics. Martinet (1962:93-100) has suggested a more formal revision of Sapir's system, within his own theory of the 'moneme'. Sapir recognized, as one must, that recent genetic divergence is likely to preserve much of the original typology, and that prolonged contacts between contiguous languages will tend to produce areal congruences of type (Sprachbiinde); but the strength of his system lay in its total theoretical independence from areal, temporal, and genetic considerations. Thus Latin and Greek were shown to fall very close typologically to Takelma (S.W. Oregon) within one major type, and Turkish and Yana came together within another (150-2). Sapir's was pure typology; he explicitly rejected any connection between structural type and level of civilization or linguistic progress: 'When it comes to linguistic form, Plato walks with the Macedonian swineherd, Confucius with the head-hunting savage of Assam' (234). In this Sapir in 1921 was more in sympathy with later linguistics than with the rather conservative Schmidt. Hjelmslev (1966, written in 1943), Voegelin (1954), and Saporta (1957), for example, all insisted on the distinction between typology and history as the sources of quite separate classifications, and the latter two have made use of the term cross-genetic in connection with typology. All agree that much detailed

THE HISTORY OF LANGUAGE CLASSIFICATION

25

observation and analysis must be brought in if the results are to be significant. Saussurean structuralism had opened the way for detailed typological classifications at particular levels of analysis, whereby languages would be grouped together simply according to shared features of phonology, grammar, or semantics. Trubetzkoy 1939b: chapter 4 contains a now classical set of phonological system types, and this is followed on similar lines in Hockett's (1955: chapter 2) "Typology of phonologic systems". In both of these, languages are classed together by their possession of similarly structured sets of contrasts between phonemes and the distinctive features characterizing the phonemes. Structural phonology has suggested the possibility of the regrouping of dialects, not in direct line with attested or inferred history, but on the basis of similarities in sets of phonological contrasts. The general theory of this approach was put forward by Weinreich (1954), and with particular reference to the Slavic language area by Stankiewicz (1958); some detailed applications are made to the Scottish dialects by Catford (1957). In morphology, Greenberg (1960a), in an issue of IJAL largely devoted to typology, developed Sapir's system of classification in the direction of greater formalism, setting up ten 'typological indices' (synthesis, agglutination, compounding, derivation, gross inflection, prefixing, suffixing, isolation, pure inflection, and concord), based on a statistical counting of specific features in sample texts of different languages; for example the index of synthesis was calculated by reference to the average number of morphemes to each single word, out of eight languages Eskimo coming at one extreme with 3.72, Sanskrit next with 2.59, and Vietnamese last with 1.06. At the syntactic level Bazell (1949) proposed a typology of languages drawn from the essentially syntactic relations of (overt) succession or word order and (covert or functional) determination and subordination. At the level of semantics Ullmann (1953) has suggested a classification of languages by reference to characteristics of their lexica in relation to such features as their relative proportions of 'motivated' (selfexplanatory) words and 'unmotivated' words (e.g. German Handschuh and Schlittschuh as contrasted with French gant and patin, and with English glove and skate), and their preferential use ofgeneric or of specific words (cf. Martinet 1962:87-9 and Stan 1963). If Whorf's ideas could be worked out satisfactorily with application to a number of languages, his conception of different grammatical-semantic structures fostering different world views in their users would be a basis for a semantic typology of a farreaching kind (cf. Carroll 1956). The special problems facing comparative linguistics in largely or wholly illiterate language areas beset American linguists working on the native languages of their continent in the nineteenth century, and the controversy between typological and lexical comparison as the key to classification has already been noticed. In fact work on the native languages of America and on those of central and southern Africa provides good illustrations of the attitudes taken up during the last two centuries by different scholars on the question of language*classification in largely illiterate areas; and for this reason work in these two fields is worth some particular notice in a historical survey of general principles.

26

R. H. ROBINS

During the nineteenth century a historical perspective was generally maintained by both sides, and the question turned on the relative claims of structure and lexicon as criteria for membership of language families. Some linguists of the present century, however, with the revival of non-historically orientated typology, have suggested that with languages in this condition genealogical classification is not really feasible. Such a view has been expressed by Lehmann (1962:49): 'for languages attested only today we may be limited to classification based on typology'. Much earlier, Boas (1929) had argued that in a field like that of native North America, in the absence of written records it is impossible to distinguish between the results of divergence from a common source and the results of subsequent interinfluences of languages in prolonged contact. This would be comparable to saying that under such conditions the heavy French influence on English since the eleventh century makes it impossible to assign English definitively either to the Germanic or to the Romance family (for a similar view in the Southeast Asian area, cf. Honey and Simmonds 1963). Indeed, within I-E, Schuchardt, who never accepted the full neogrammarian position, raised just such questions in regard to dialect classification ('geschichtlich verwandt oder elementar verwandt'). They are most acute in the classification of dialects because interdialectal intercourse favours the areal spread of features across dialect boundaries (Schuchardt 1928:166-88, 189-204, 248-53). In a significant part of the North American languages the work of twentieth century scholars such as Sapir, Bloomfield, and M. Haas has more than validated the application of historical methods leading to historical conclusions and resulting in historical classifications (cf. Mandelbaum 1949:167-250; Hockett 1948; M . H a a s 1966). In 1945 Voegelin complained of the potentially misleading irrelevance of purely areal groupings, not reinforced by strictly linguistic analyses of the languages involved, in some earlier subdivisions of the larger American-Indian families. In a review of work on South American languages, Rowe (1954) likewise confirmed scholars' confidence that historical classifications could there too be established by the comparative study of correspondences and similarities (cf. the later classification of these languages in Greenberg 1960b). In particular the comparative work of Bloomfield in the Algonquian family (Hockett 1948) demonstrated that differences in the state of the languages or in the culture of their speakers in no way required or permitted any lowered standard of scholarship in the application of comparative methods; in this Bloomfield bore out Meillet's claim (1938:69) that 'la grammaire comparative des langues indo-européennes fournit à l'ensemble de la linguistique historique un modèle à imiter' (cf. M. Haas 1966:120-1). But in this connection one must again stress the lesson learned over three generations of Indo-European historical linguistics, the necessity of working through systematic phonological correspondences between word and morpheme forms in the languages to establish genealogical relations, rather than relying on mere similarities (cf. Lehmann 1958). The languages of central and southern Africa have given rise to a similar debate on the relevant criteria to apply in arriving at adequate classifications. As with expres-

THE HISTORY OF LANGUAGE CLASSIFICATION

27

sions like American-Indian languages the starting point is simply areal. In the first instance a set of languages is looked at for comparative purposes just because they are spoken in a geographically well defined stretch of contiguous territory. Earlier sporadic treatments of certain African languages had been carried out from the seventeenth century (Schmidt 1926:85-6), but the first major classification work in the field was that of Bleek (1862-69), in which the now general term Bantu first appeared. Bleek moved at once from the areal basis of his title to a typological one, citing as specific to the Bantu group of languages certain structural features including the now well known nominal prefix class concord systems (2-3, 93-108). The Bantu family started life as a typological group, however loosely defined, and so it continued at the hands of several scholars. Drexel ((1917-18) wrote of the Bantu type, and Johnston (1919-22) referred to 'Bantu and Semi-Bantu languages'. Meinhof (1899, 1906) was the first to set out a systematic historical treatment of these languages, classifying them by reference to their divergence from a postulated Ur-Bantu, which he established from sound correspondences between the different Bantu languages and the sound laws governing their history. But he classified the Bantu languages themselves within the larger class of Klassensprachen by the typological criterion of prefix class concord (1906:22-3), and elsewhere in Africa he was prepared to resort to typological criteria (e.g. the absence of nominal prefix classes) as evidence for historical classification (1916:110). For a general appreciation of Meinhof's African language work, one may consult Doke 1961 and Doke and Lestrade 1946. These different approaches to what were broadly the same set of languages were reviewed by Guthrie in 1948. Guthrie distinguished three methods of classification (1948: chapter 3): the 'historical', the 'empirical', and the 'practical'. Historical classification he associated with Meinhof, and he dismissed as impossible 'true historical study ... in a field with practically no historical records' (1948:20). Guthrie's 'empirical' is in fact partly typological, based on the drawing of isoglosses to show the distribution of significant grammatical and phonological features, and partly comparative-lexical, involving the occurrence of words in different languages referable by systematic comparison to hypothetical formulae ('starred forms'). By 'practical' (27-8) he meant that where the application of 'empirical' criteria does not by itself lead to a satisfactory classification, decisions based on areal contiguity were to be called in. particularly in relation to the largest units of classification, the sixteen 'zones' (28: 'Since [the present work] is avowedly practical in its intention, similarities between widely separated languages are of little importance'). Following these principles Guthrie's Bantu family depended on a mixture of typological and what amounted to historical criteria. That there is a major congruence between the two may well be a special attribute of these languages; that there will be areas of non-congruence was inevitable and was allowed for differentially (1948:18-19); among the 'languages which are incompletely Bantu' those with an adequate shared common lexical stock but which lack the major typological features are called 'sub-Bantu', and those showing features such as nominal prefix concord but lacking Bantu lexical correspondences are 'Bantoid'.

28

R. H. ROBINS

Such a classification is admittedly a mixed one derived from criteria of different sorts; indeed Guthrie brings in the three sorts of classification referred to in the title of this volume. His later 1967 treatment does not substantially differ from that of 1948 as far as the delimitation of the Bantu languages is concerned (§13.01). The structural criterion of the presence of a class concord system, which was made a condition of membership in the Bantu family in the former work, is retained (§§11.22, 35.01); but, as the subtitle indicates, much more interest is shown in reconstructing the (pre-) history and geographical situation of the 'ancestor of the whole family' (§61.82, cf. §25.01-21). Greenberg (1963) has taken a much more rigorous line, insisting on comparative lexical criteria alone (1: 'the sole relevance in comparison of resemblances involving both sound and meaning in specific forms' leading to 'a complete genetic classification of the languages of Africa'; but cf. here the observations of Lehmann 1958, referred to above, p. 26). Greenberg has classified the languages of Africa into four major genetic families, and the Bantu languages are classed as a single group in the much larger Congo-Kordofanian family. It is to be noticed that in dealing with unwritten languages several nineteenth century scholars had pressed the claims for typology as being criterially relevant to a historical classification (as with the native American languages), but that twentieth century linguists have seen typological and historical comparison as separate methods leading to theoretically separate classifications, even though an 'empirical' mixed classification may on occasion be justified for practical purposes. One observation of general methodological importance arises very clearly from the consideration of the course of comparative Bantu studies: typological classification, inherently indeterminate, admits of scales ('more or less'), enabling people to write of 'Bantu', 'SemiBantu', 'Sub-Bantu', and so on, whereas genetic classification is always a matter of Yes or No. Von Humboldt (above, p. 16) had recognized the essential indeterminacy of typology, and the three traditional categories of isolation, agglutination, and inflection (fusion) are better understood as directions towards which languages tend in different degrees, or as major sets of characteristics which they share in different proportions, so that they are classified by that set which is preponderant, rather than being assigned wholly to one of a number of mutually exclusive classes. Greenberg's (1960a) quantification procedures gave explicit recognition to this (above, p. 25), and so did the treatments of typological classification by Bazell (1958) and Uspensky (1968). Bazell suggested that languages could well be classified by reference to the relative ease of morphological segmentation on the one hand and grammatical identification of the resultant segments on the other. This leads to a reinterpretation of the triad, isolation, agglutination, and inflection, wherein fairly nuclear member languages are, respectively, Vietnamese (slightly less so Chinese) — in which it is easy to segment but hard to classify the segments unambiguously, Turkish — in which it is easy to segment and to classify the segments grammatically, and Latin — in which it is hard to segment unambiguously but easy to classify any segment in

THE HISTORY OF LANGUAGE CLASSIFICATION

29

terms of the grammar. Uspensky, like Greenberg, proposed the development of a formal interpretation of some of Sapir's ideas. He set up classes of root and of formative elements according to their syntagmatic possibilities (1968:39-43). His main proposal was that languages should be classified typologically by reference to a number of abstract 'étalon languages' or standard language models. These would be credited with particular structural characteristics, and actual languages or rather the grammatical structures of actual languages would be compared with one another in relation to these 'étalon languages', and would be shown to be more, or less, classifiable as of one or another type according to the number of conversion rules required to pass between their structures and that of one of the 'étalons' (1968:29). Uspensky's 'étalons', or models, are therefore not representatives of actual languages, but are structures endowed with features inductively abstracted from actual languages and then used as standards of comparison and classification. In this respect they are somewhat analogous to the cardinal vowels set up by D. Jones (1947: chapter 8) as ideal reference points for the phonetic identification and classification of actual vowel sounds. In grammatical structure Uspensky adapted the earlier fourfold system, proposing four models ('étalons'): amorphous (isolating), incorporating, agglutinating, and inflecting, in a rising scale of complexity (1968:55), but he admitted that an absolutely amorphous language not only does not exist, but could not exist (43). Birnbaum (1968) summarises and comments on Uspensky's system of classification, and associates typology closely with the deep structures of transformational-generative grammarians. If historical classification is interpreted in the sense embodied in the quotation from Vendryes cited on page 3 of this article, then it is in principle nonsensical to say that a given language is 'semi-Indo-European' or 'semi-Germanic'. From the standpoint of history Meillet (1937:16) declared: 'La notion de parenté de langues est chose absolue et ne comporte pas de degrés.' Except in the very special circumstances of a pidgin created in the first place as a shared second language by speakers of two different mother tongues (Weinreich 1953:104-6), an unbroken line links speakers of any language of a genetic family to the assumed parent language, just as at no time during the years between colloquial Latin and modern French was there any generation aware of speaking any but the language of the previous generation (cf. Meillet 1948:81 : 'La parenté de langues résulte uniquement de la continuité du sentiment de l'unité linguistique'). Of course this must not be misunderstood. Languages can be more, or less, remote from one another within a genetic family ; Italian and Spanish are more closely related to each other than either is with English, and this can be diagrammatically indicated in the Stammbaum model, but all three languages are equally members of the I-E family. English has clearly a much greater admixture of Romance (French) lexical forms than has German or Dutch, and it has lost a lot more of the inherited Germanic inflectional system than any other Germanic languages, but this does not alter the fact of an unbroken consciousness of'speaking the same language' that links modern English with

30

R. H. ROBINS

Old English, and further back in time with the assumed Common Germanic. English could have taken over still more non-Germanic lexical forms, for example if French prepositions like sans had proved viable or Latin ones like ex had moved beyond specific technical phraseology and extended their range so as to oust without, out of, etc.; but genealogically English would still have remained one of the Germanic languages in terms of historical classification. Certain theoretical consequences of this must be recognized. It is clearly possible for a language to lose all recognizable or recoverable evidence of a genetic relation through the gradual processes of linguistic change (Meillet 1948:94: 'La parenté de deux langues peut donc être, et est souvent, indémontrable, même alors qu'elle est réelle'). In fact this has quite certainly happened over and over again. There is nothing special about Indo-European and the other major language families or about the Ursprachen 'reconstructed' with different degrees of confidence, except that they go back in history and prehistory as far as it is possible to go back on the evidence that is available and that is likely to be available. On certain borderlines there is indeterminacy i n the evidence, as for example on the question of an even earlier genealogical relationship between I-E and Finno-Ugrian or as in the repeated attempts to establish some historical links between an American-Indian family and a family on the Asiatic mainland. The situation is quite different from that prevailing in typology, where the relationships themselves, not just the evidence, are inherently indeterminate and potentially quantifiable. In genealogical classification the evidence is often probabilistic (strictly speaking it is always so) and at a certain point becomes indeterminate and unreliable, but the question of relationship remains a Yes or No one, apart from the exceptional cases of genuine language mixture referred to above (cf. Collinder 1948; very recently [King 1969:152-3] it has been tentatively suggested that the techniques of generative grammar may enable certain very basic syntactic rules to be cited as evidence of genealogical relationships not otherwise recoverable, but at present this can be no more than an indication of possible research directions, subject to extreme caution). In this light the question of monogenesis versus polygenesis in the ultimate history of languages is not one of theory, still less is it one of faith, as many mediaeval and later thinkers were forced to maintain by a literalist interpretation of the Biblical Babel tradition. It is simply the result of the disappearance of evidence, in the absence of linguistic equivalents of geological fossils, at a date in history far too recent to permit the reliable recovery of genetic relations remoter than those assumed for I-E and the other principal language families. The potential value of quantifying probabilistic evidence and the problems facing genealogical classification in illiterate language families gave rise in the 1950s to lexicostatistic and glottochronological methods (the two terms are not exact equivalents). On the basis of an examination of controlled evidence it was held that under culturally normal conditions the basic vocabulary of languages tends to lose words and to replace them by noncognate (not directly inherited) near-synonyms at a fairly

THE HISTORY OF LANGUAGE CLASSIFICATION

31

constant rate (Hockett 1958: chapter 61, Hymes 1960). If all the theoretical and practical problems could be overcome, these methods might provide a useful tool in the uncovering of genealogical relationships and paths of divergence at time depths beyond the reach of history, especially where written records are scanty or non-existent. They have been so used, e.g. by Dyen (1962, 1965a, 1965b). However, in the light of objections that have been made and problems that still remain to be dealt with satisfactorily (Hoijer 1956, Chrétien 1962), some people have preferred to regard the findings of lexicostatistics as an independent and quantifiable measure of the degree of relationship between languages by reference to the number of cognate word forms that they share, expressly excluding time calculations and other diachronic considerations (cf. Cowan 1962). Much of the brief survey made in the last few pages has been concerned with developments in linguistic classification that have been associated with languages outside the I-E family and have in part been conditioned by the actual circumstances of these languages and the nature of the material available. However, as might be expected, the insights and theoretical developments of structural descriptive linguistics have had their effects on the thinking of some linguists on the whole status of I-E and the corresponding families and on the nature of comparative-historical studies within general linguistics. This movement of thought is to be distinguished from the incorporation of structural concepts (e.g. the phoneme) into the continuing genetic classification system of historical linguistics (Jakobson 1931; Hoenigswald 1950 and 1960; cf. above, p. 23). Trubetzkoy 1939a is illustrative at this point. Facing certain well known, but in fact not fatal difficulties in the traditional conception of the I-E family, in particular that beyond a certain time depth it becomes impossible to distinguish loanwords and inherited words and that one cannot assume a necessary single period when all the 'reconstructed' forms of I-E were copresent in a single état de langue, Trubetzkoy was led to redefine family of languages so that it required both a minimum of inherited lexical material shared between the languages and also broad agreement in structural type {Ähnlichkeit des Sprachbaues). For I-E he set up six criteria, which, he claimed, were individually found outside I-E but nowhere else occurred all together. To be counted as an I-E language, a language must exhibit all six. The criteria were both phonological, for example the absence of vowel harmony, and grammatical, for example the absence of the ergative construction (in I-E languages, he said that universally the subject of transitive verbs shared the same syntactical and morphological features as the subject of intransitive verbs). Benveniste (1952-53), in discussing Trubetzkoy's proposed I-E typology and other typologies, pointed out that the American-Indian Takelma language possesses all six features, though, of course, without any I-E lexical component. Trubetzkoy was in fact reverting in a more sophisticated and informed manner to the assumption found in earlier centuries, that genealogical families would share a general typology. He did, however, accept the I-E family in its traditional form,

32

R. H. ROBINS

presumably selecting those structural features that in his opinion did characterize the whole family. But he made the important concession to typological classification that under his system a language could, at a determinable period, become I-E, and equally could cease to be I-E, when it either acquired or lost the set of six criterial strukturale Merkmale. Presumably Trubetzkoy would have accommodated Allen's (1950-51:69-70) criticism that in a branch of the accepted I-E family, Hindi and some other Indian languages, an ergative construction is found, by saying that these languages had now ceased to be I - E despite their stock of inherited morphemes. The recurrent difficulties of a double-based classification such as was here proposed, arising from the non-congruence of structural classes and etymological, genealogical families, seem at least as formidable as those that led Trubetzkoy in the first place to seek the reintroduction of typology into the criteria for membership of the historical I-E family. Trubetzkoy's proposed classification was based on a combination of shared lexical inheritance and typological resemblance. From an examination of the distribution of the excluded typological features he concluded that the I-E family was originally constituted as a group of dialects spoken over a wide area between the North Sea and the Caspian, between the Finno-Ugrian and Altaic languages on one side and the Caucasian and Semitic languages on the other. Trubetzkoy's hypothesis rested in part on the recognition that geographically contiguous languages tend to share in the diffusion of typological features, at both the grammatical and the phonological levels. This has always been the basis of the linguistic relevance of areal classifications, of Sprachbunde, noticed earlier in this essay ; and, as we have seen, the general idea of contiguous languages bearing resemblances to one another has been commonly accepted as borne out by observation. But the sharpening of genealogical language classification in the nineteenth century directed specific attention to geographical areas in which contiguous languages belonging to different families, or to distinct subfamilies within a family, displayed an obvious sharing of grammatical features. Nineteenth century scholars like Miklosich (1862:1-13) referred to such an area in southeastern Europe, and an areal classification of Balkan languages was fully recognized by Sandfeld 1930. Sandfeld made an exhaustive survey of the features shared by these languages, which comprised members of the separate Slavic, Romance (Rumanian), Albanian, and Greek branches of I-E, and by this he justified the treatment of 'les langues balkaniques' as a linguistic unity (1930:213-6), whether the result of substrate influences or, as he much preferred, the product of centuries of diffusion from the Greek language area under the cultural and religious domination of the Orthodox Church. Emeneau's (1956) comparable designation of the Indian subcontinent as an areal linguistic unity has already been mentioned. Sebeok suggested in 1950 such a reinterpretation of the classification 'Ural-Altaic', as a typologically similar group of languages, for the most part areally contiguous and extending over much of Eurasia,

THE HISTORY OF LANGUAGE CLASSIFICATION

33

without the assumption of any necessary genealogical link between the Altaic and the Ural (Finno-Ugrian and Samoyed) families within it. Following the lines of Jakobson's (1936) general account of 'affinités phonologiques', Weinreich (1953) published his comprehensive survey of areal diffusion resulting from language contacts, in which he drew examples from many areas and from all levels of linguistic structure. A rather different interpretation of areal spread among contiguous speech communities has characterized the 'neolinguistic' movement, originating and largely centered in Italy and in part inspired by the antimechanistic outlook on language expressed by such idealists as Croce and Vossler. On lines similar to the objections raised by Schuchardt against a purely genealogical classification of dialects, 'neolinguists' like Bartoli (1925) and Bonfante (1945, 1947) wanted the processes of areal diffusion incorporated in the methodology of historical classification, not only in the recognition of innovatory and conservative (usually peripheral or isolated) areas within a family, but in modifying what they considered the too simplistic classification of some languages as either members of a certain family or branch, or not, thus denying precisely the unequivocal position of Meillet (1937, above, p. 29). Thus Bonfante (1947:350-1), on the basis of facts such as were noticed on pages 29-30, refused to regard English as simply a Germanic language, or Bulgarian as just Slavic; and somewhat on the lines of Trubetzkoy (1939a), though on rather different sorts of evidence, explicitly allowed that languages can shift their historical allegiance. Neither Meillet nor anyone else would dispute the facts here brought to bear on the question, but the accepted classificatory theory that Meillet was defending, rested firmly on the concept of linguistic continuity maintained over the generations, in relation to which areal diffusion of features, though part of a language's history, could not be part of its strictly historical classification. More radical proposals for the reconsideration of lexical comparison such as form the basis for historical classifications of the accepted sort came from Allen (1953), in the article referred to earlier (p. 3). His arguments were expressly concerned not just with I-E, as Trubetzkoy's had been, but with the whole field of comparative linguistics, though his exemplification was largely drawn from I-E. Allen rehearsed the same difficulties mentioned by Trubetzkoy in relation to I-E and by Boas (cf. p. 26, above) in relation to other, less privileged areas of historical linguistic research; but the conclusion of his critique was different. Broadly speaking he was concerned with the accepted methods of systematic comparisons of semantically related word forms, and he maintained that, strictly as a part of general linguistic science, such methods should not need to give rise to a theory of historical relationship. He distinguished between theory and hypothesis ; if in a given field of languages hypotheses of genetic relationships and consequent historical classifications were set up, this, he argued (Allen 1953:54), would be unobjectionable as long as it was realized that factors outside the science of linguistics had to be invoked in additional support: the contingent existence of earlier written records and in exceptionally favourable

34

R. H. ROBINS

cases the survival of something like the unitary source, as in Romance, support material not available everywhere and not, therefore, properly part of a general science of language universally applicable (cf. 1953:86). The postulate of a continuous process of unnoticed and unintended change, referred to by Meillet (1948:81, cf. above, p. 29), is in particular cited by Allen (103) as extralinguistic. In Allen's view expressed here, traditional comparison between two or more languages simply provided the possibility that at the phonological level, on the basis of lexical correspondence series, a system of common units can be established, fewer in number than the sum of the separate unit systems of the languages themselves. The reduction achieved would be the measure of relationship (1953:89-95), a relationship no longer one of historical descent and so making legitimate the question 'How I-E is a given language?' (91). Traditional comparative theory was based on such phonological correspondences in lexical items, but Allen suggested (99-100) the possibility, given adequate means of linguistic control, of lexical systems themselves and grammatical systems being similarly treated. Allen concluded that, within linguistics as an autonomous science, comparison is not concerned with historical truth or with history as such (87-8), and that linguistic relationship and therefore linguistic classification derived from such relationship is 'not of languages but of systems', and 'is not either-or but more-or-less' (92). In this respect he lined up the methods of lexical comparison with those of typological comparison. Allen's ideas were taken up and discussed by Levenston and Ellis (1963). Despite some alarm expressed at the time, particularly after the delivery of the spoken version of this paper at Oxford, it must be kept in mind that Allen was expressly and deliberately not trying to exclude or to invalidate the historical, genealogical classification of languages where the evidence was available, but just to distinguish between particular hypotheses of genetic relations (which is precisely the term used four years later by Greenberg [1957:35]) and what he was putting forward as a universally applicable theory of linguistic comparison, and to discriminate between what, in his opinion, belonged strictly within linguistics as a science and what fell partly outside it (cf. Milner 1963). Certainly there has been no diminution of historical and historical-classificatory work in linguistics since 1953, either within the I-E field or in the application of this type of linguistics to other superficially less favored areas, for example native America, as evidenced by M. Haas (1966). At the present time it is clear that typological classification has returned to scholarly favour, but without displacing interest in historical classification (a bibliography of recent work on the typological classification of languages up to 1965 may be seen in Allen 1965; for a bibliography of earlier work see Meillet and Cohen 1952:xvii-xxxv). Following Sapir (1921) linguists have been insisting on the importance of typology. Uspensky 1968 has already been mentioned; Bazell, in addition to his 1949 article, devoted his inaugural lecture in London to typology (1958); and both Hjelmslev (1966:123-59) and Martinet (1962: chapter 3) within their own attitudes to linguistic studies have stressed the necessity for typological classifications in contemporary

THE HISTORY OF LANGUAGE CLASSIFICATION

35

linguistics. Hjelmslev saw as a principal objective the delimitation of the possible forms that language structures could take (1966:128-9), though he did not go further than to suggest the lines on which a glossematic treatment of typology would move. Martinet has expressed the essential task of language typology as the classification of the different ways in which the comprehension of a complete statement about our experience may be made possible within the confines of the necessarily linear succession of language (1962:101-2), and he has further declared (67) that the present position of linguistics has made it now 'high time to tackle typological problems'. Within the limits of the present essay an attempt has been made to follow the course of the theory and practice of language classification from antiquity to the present day. History as a narrative and an evaluation of the past must terminate with the present, but movements of thought and applications of theory go on. Future developments will arise from current work now in progress and will have to be interpreted in this light. The examination and exemplification of the contemporary state of language classification is the basis for the subsequent chapters of this volume.

REFERENCES

ALLEN, W. S. 1950-51. A study in the analysis of Hindi sentence structure. AL 6.6886. . 1953. Relationship in comparative linguistics. TPhS 1953.52-108. . 1956. Structure and system in the Abaza verbal complex. TPhS 1956.127-76. . 1965. Classification and language. The Classification Society Bulletin 1.13-21. ALSTED, J. H. 1650. Thesaurus chronologiae. 4th ed. Herborn. ARENS, H. 1955. Sprachwissenschaft: der Gang ihrer Entwicklung von der Antike bis zur Gegenwart. Freiburg and Munich. BARTOLI, M. 1925. Introduzione alla neolinguistica. Geneva. BAZELL, C.E. 1949. Syntactic relations and linguistic typology. CFS 8.5-20. . 1958. Linguistic typology. London. BENVENISTE, E. 1952-53. La classification des langues. Conférences de l'Institut de Linguistique de l'Université de Paris 11.33-50. Paris. BIRNBAUM, H. 1968. On reconstruction and prediction; two correlates of diachrony in genetic and typological linguistics. FLing 2.1-17. BLEEK, W.H.I. 1862-69. Comparative grammar of the South African languages. London. BOAS, F. 1929. Classification of American Indian languages. Lg 5.1-7. BONFANTE, G. 1945. On reconstruction and linguistic method. Word 1.83-94,133-61. . 1947. The neolinguistic position. Lg 23.344-75. BOPP, F. 1833. Vergleichende Grammatik des Sanskrit, Zend, Griechischen, Lateinischen, Litauischen, Gotischen, und Deutschen. Berlin. BORST, A. 1957-63. Der Turmbau von Babel. Stuttgart.

36

R. H. ROBINS

BRUGMANN, K., and B. DELBRÜCK. 1886-1900. Grundriss der vergleichenden Grammatik der indogermanischen Sprachen. Strassburg. BUCK, C.D. 1933. Comparative grammar of Greek and Latin. Chicago. BUFFON, COMTE DE. 1785a. Natural history (tr. by W. Smellie), vol. 3. London. . 1785b. Natural history (tr. by W. Smellie), vol. 8. London. CARROLL, J.B., ed. 1956. Language, thought, and reality: Selected writings of Benjamin Lee Whorf. New York. CATFORD, J.C. 1957. Vowel systems of Scots dialects. TPhS 1957.107-17. CHRÉTIEN, C.D. 1962. The mathematical models of glottochronology. Lg 38.11-37. COLEMAN, R . G . G . 1963. The dialect geography of Ancient Greece. TPhS 1963.58126.

COLLART, J. 1954. Varron: grammairien latin. Paris. COLLINDER, B. 1948. La parenté linguistique et le calcul des probabilités.

UUA

13.1-24.

COWAN, H.K.J. 1962. Statistical determination of linguistic relationship. SL 16.5796.

DIDEROT, D., and J. LE R. D'ALEMBERT. 1772. Encyclopédie, ou dictionnaire raisonnée des arts et des métiers, vol. 9. Geneva. DOKE, C.M. 1961. The growth of comparative Bantu philology. Contributions to the history of Bantu linguistics, 70-3. Johannesburg. Reprinted 1966 in Portraits of linguists, ed. by T.A. Sebeok, 2.110-4 (Bloomington). DOKE, C. M., and G.P. LESTRADE. 1946. Meinhof's contributions to our knowledge of African languages. AfrS 5.73-7. Reprinted 1966 in Portraits of linguists, ed. by T. A. Sebeok 2.115-22 (Bloomington). DREXEL, A. 1917-18. Beiträge zur Grammatik des Bantu-Typus. Anthropos 1213.89-133, 9 1 0 - 5 7 .

DUPONCEAU, P. S. 1819. Transactions of the historical and literary committee of the American Philosophical Society, vol. l.xvii-xlvi. Philadelphia. DYEN, I. 1962. The lexicostatistical classification of the Malayopolynesian languages. Lg 38.38-46.

. 1965a. A lexicostatistical classification of the Austronesian languages. IJAL 31/1, supplement. . 1965b. Lexicostatistics in comparative linguistics. Lingua 13.230-9. EMENEAU, M.B. 1956. India as a linguistic area. Lg 32.3-16. FINCK, F . N . 1909. Die Haupttypen des Sprachbaues. Leipzig and Berlin. GABELENTZ, G. VON DER. 1901. Die Sprachwissenschaft: ihre Aufgabe, Methoden, und bisherigen Ergebnisse. Leipzig. GALLATIN, A. 1836. Archaeologia Americana. Transactions and Collections of the American Antiquarian Society 2.1-7. Cambridge, Mass. GOROPIUS, B. 1569. Origines Antwerpianae. Antwerp. GREENBERG, J. H. 1957. Essays in linguistics. Viking Fund publication in anthropology 24. New York.

THE HISTORY OF LANGUAGE CLASSIFICATION

37

. 1960a. A quantitative approach to the morphological typology of languages. IJAL 26.178-94. . 1960b. The general classification of Central and South American languages. Men and cultures, selected papers of the Fifth International Congress of Anthropological and Ethnological Sciences, ed. by A. F. C. Wallace, 791-4. Philadelphia. . 1963. The languages of Africa. IJAL 29/1 (2). GREENE, J. C . 1 9 5 9 . The death of Adam. Ames, Iowa. Reprinted 1 9 6 1 , Mentor Books (New York). (Page references from 1 9 6 1 . ) GUTHRIE, M. 1948. The classification of the Bantu languages. London. . 1967. Comparative Bantu, vol. 1 : An introduction to the comparative linguistics and prehistory of the Bantu languages. Farnborough, England. HAAS, M.R. 1966. Historical linguistics and the genetic relationship of languages. CTL 3.113-53. . 1969. Grammar or lexicon? The American Indian side of the question from Duponceau to Powell. IJAL 35.239-55. HAAS, W. 1956-57. Of living things. German Life and Letters 10.62-70, 85-96, 251-7. Oxford. HAINSWORTH, J . B . 1967. Greek views of Greek dialectology. TPhS 1967.62-76. HJELMSLEV, L. 1 9 6 6 . Le langage. Paris. (Tr. by M . Olsen from Sproget, written 1 9 4 3 , published Copenhagen 1 9 6 3 . ) HOCKETT, C . F . 1 9 4 8 . The implications of Bloomfield's Algonquian studies. L g 24.117-31.

. 1955. Manual of phonology. IUPAL 11. Baltimore. . 1958. A course in modern linguistics. New York. . 1965. Sound change. Lg 41.185-204. HOENIGSWALD, H. M. 1950. The principal step in comparative grammar. Lg 26.35764. . 1960. Language change and linguistic reconstruction. Chicago. . 1963. On the history of the comparative method. AnL 5.1-11. HOIJER, H . 1954. Some problems of American Indian linguistic research. Papers from the Symposium on American Indian Linguistics (UCPL 10), 3-12. Berkeley and Los Angeles. . 1956. Lexicostatistics : a critique. Lg 32.49-60. HONEY, P. J., and E . H . S . SIMMONDS. 1963. Thai and Vietnamese: Some elements of nominal structure compared. Linguistic comparison in Southeast Asia and the Pacific, ed. by H.L. Shorto, 71-8. London. HUMBOLDT, C.W. VON. 1822. Ueber das Entstehen der grammatischen Formen und ihren Einfluss auf die Ideenentwickelung. Die sprachphilosophischen Werke Wilhelm's von Humboldt, ed. by H. Steinthal, 67-101. Berlin (1883). . 1827. Lettre à M. Abel-Rémusat sur la nature des formes grammaticales en général et sur le génie de la langue chinoise en particulier. Paris.

38

R. H. ROBINS

. 1836. Ueber die Verschiedenheit des menschlichen Sprachbaues. Berlin. Reprinted 1949 (Darmstadt). (Page references from 1949.) HYMES, D.H. 1960. Lexicostatistics so far. CAnthr 1/1.3-44. Ivié, M. 1965. Trends in linguistics. The Hague. JÄGER, A. 1686. De vetustissima lingua Europae. Stockholm. JAKOBSON, R. 1931. Principes de phonologie historique. Trubetzkoy 1939b ( = 1949:315-36). Reprinted in Jakobson 1962:202-20. Previously published in German as Prinzipien der historischen Phonologie, TCLP 4.247-67. . 1936. Sur la théorie des affinités phonologiques entre les langues. Report for the fourth international congress of linguists. Reprinted in Trubetzkoy 1939b ( = 1949:351-65) and in Jakobson 1962:234-46. ——. 1957. Typological studies and their contribution to historical comparative linguistics. PICL 8.17-25. Reprinted in Jakobson 1962:253-32. . 1962. Selected writings, I : Phonological studies. The Hague. JEFFERSON, T. 1784. Notes on the State of Virginia. Reprint of original edition, ed. by P. L. Ford, Historical Printing Club, Brooklyn (1894). JOHNSTON, H.S. 1919-22. A comparative grammar of the Bantu and semi-Bantu languages. Oxford. JONES, D. 1947. An outline of English phonetics. 6th ed. Cambridge. JONES, SIR WILLIAM. 1786. Third anniversary discourse on the Hindus, before the Asiatick Society. Reprinted in Lehmann 1967:10-20. . 1799. Works, vol. 1. London. KANT, I. 1785. Die Bestimmung des Begriffs einer Menschenrace. Versammelte Schriften 8.89-106. Berlin (1912).

KAULEN, F. 1861. Die Sprachverwirrung zu Babel. Mainz. KING, R.D. 1969. Historical linguistics and generative grammar. Englewood Cliffs, N.J.

LEHMANN, W. P. 1958. Review of Essays in linguistics, by J. H. Greenberg. IJAL 24.142-5. . 1962. Historical linguistics: An introduction. New York. . 1967. A reader in nineteenth century historical linguistics. Bloomington, Ind. LEIBNIZ, G.W. 1765. Neue Abhandlungen über den menschlichen Verstand, vol. 2. Reprinted 1961. Frankfurt. LEROY, M. 1963. Les grands courants de la linguistique moderne. Paris. LEVENSTON, E., and J. ELLIS. 1963. Transfer-grammar development of system reduction quantification method. ZPhon 17.449-52. LINNAEUS, C. 1775. The elements of botany (tr. of Philosophia botanica, by H. Rose). London. . 1787. The families of plants (tr. of Genera plantarum), vol. 1. Lichfield. MAHER, J.P. 1966. More on the history of the comparative method: The tradition of Darwinism in August Schleicher's work. AnL 8.1-12.

THE HISTORY OF LANGUAGE CLASSIFICATION MANDELBAUM, D . G . ,

ed.

1949.

39

Selected writings of Edward Sapir. Berkeley and

Los Angeles. functional view of language. Oxford. MEILLET, A . 1937. Introduction à l'étude comparative des langues indo-européennes. 8th ed. Paris. . 1938. Linguistique historique et linguistique générale, vol. 2. Paris. . 1948. Linguistique historique et linguistique générale. (Collection linguistique publiée par la Société de linguistique de Paris 8). Paris. MEILLET, A . , and M . COHEN. 1952. Les langues du monde. Paris. MEINHOF, C . 1899. Grundriss einer Lautlehre der Bantusprachen (Abhandlungen für die Kunde der Morgenlandes 2.1). Leipzig. . 1906. Grundriss einer vergleichenden Grammatik der Bantusprachen. Berlin. . 1916. Sprachstudien im ägyptischen Sudan. Zeitschrift für Kolonialsprachen 7.105-33. Berlin. METCALF, G.J. 1953. Abraham Mylius on comparative linguistics. PMLA 68.53554. MIKLOSICH, F . 1862. Die slavischen Elemente im Rumunischen. Denkschriften der kaiserlichen Akademie der Wissenschaften, phil.-hist. Klasse, Bd. 1 2 . 1 - 7 0 . MILNER, G . B . 1963. Notes on the comparison of two languages (with and without a genetic hypothesis). Linguistic comparison in Southeast Asia and the Pacific, ed. by H.L. Shorto, 28-44. London. PEDERSEN, H . 1931. Linguistic science in the nineteenth century. Cambridge Mass. Translation of Sprogvidenskaben i det nittende aarhundrede: metoder og resultater, Copenhagen, 1924. Republished 1962 as paperback, The discovery of language. Bloomington, Ind. PERIZON, J. 1736. Origines Babylonicae et Aegyptiacae, vol. 1. Trajecti ad Rhenum. PERTZ, G.H., ed. 1879. Monumenta Germaniae historiae, vol. 24. Hanover. POTT, A. F. 1 8 3 3 - 3 6 . Etymologische Forschungen auf dem Gebiet der indogermanischen Sprachen. Lemgo. . 1863. Anti-Kaulen oder mythische Vorstellungen vom Ursprünge der Völker und Sprachen. Lemgo and Detmold. POWELL, J . W . 1891. Indian linguistic families of America north of Mexico (Seventh Annual Report of the Bureau of Ethnology). Washington. PUHVEL, J. 1966. Dialectal aspects of the Anatolian branch of Indo-European. Ancient Indo-European dialects, ed. by J. Puhvel and H. Birnbaum, 235-47. Berkeley and Los Angeles. REIMARUS, H. S. 1751. De assessoribus synedrii LXX linguarum peritis. Sylloge commentationum theologicarum, ed. by D.J. Pott and G.A.Rupert (1801). Helmstedt. . 1781. Abhandlungen von den vornehmsten Wahrheiten der natürlichen Religion. Hamburg. REUCHLIN, J . 1506. De rudimentis Hebraicis. Pforzheim. MARTINET, A .

1962. A

40

R. H. ROBINS

H. 1967. A short history of linguistics. London. Ross, A.S. C. 1950. Philological probability problems. Journal of the Royal Statistical Society, Series B (methodology) 12.19-40. London. . 1958. Etymology. London. Ross, W.D. 1923. Aristotle. London. ROWE, J . H . 1 9 5 4 . Linguistic classification problems in South America. Papers from the Symposium on American Indian Linguistics (UCPL 10), 1 3 - 2 6 . Berkeley and Los Angeles. SANDFELD, K. 1930. Linguistique balkanique: problèmes et résultats. (Collection Linguistique publiée par la Société linguistique de Paris 31). Paris. SAPIR, E. 1921. Language. New York. SAPORTA, S. 1 9 5 7 . Methodological considerations regarding a statistical approach to typologies. I J A L 2 3 . 1 0 9 - 1 3 . SCALIGER, J . J . 1 6 1 0 . Diatriba de Europaeorum Unguis. Opuscula varia, 1 1 9 - 2 2 . Paris. SCHLEGEL, A. W. VON. 1818. Observations sur la langue et la littérature provençales. Paris. SCHLEGEL, F. VON. 1808. Ueber die Sprache und Weisheit der Indier. Heidelberg. SCHLEICHER, A. 1850. Sprachvergleichende Untersuchungen, vol. 2. Bonn. . 1863. Die darwinsche Theorie und die Sprachwissenschaft. Second edition 1873. (Page references from 1873.) Weimar. SCHLÖZER, A. L. VON. 1785. Weltgeschichte. Göttingen. SCHMIDT, W. 1926. Die Sprachfamilien und Sprachenkreise der Erde. Heidelberg. SCHOTT, A., ed. 1603. Hispaniae illustratae, vol. 2. Frankfurt. SCHUCHARDT, H. 1928. Brevier (ed. by L. Spitzer). Halle. SEBEOK, T. A. 1950. The meaning of 'Ural-Altaic'. Lingua 2.124-39. , ed. 1966. Current trends in linguistics, vol. 3: Theoretical foundations. The Hague. SIMMONS, E. J., ed. 1951. The Soviet linguistic controversy. New York. STAN, I. 1963. Problems of linguistic typology. Revue de linguistique 8.229-37. STANKIEWICZ, E. 1958. Towards a phonemic typology of the Slavic languages. S1PR 21.301-19. . 1966. Slavic morphophonemics in its typological and diachronic aspects. CTL 3.495-520. STEINTHAL, H. 1850. Die Classification der Sprachen. Republished 1860 as Charakteristik der hauptsächlichsten Typen des Sprachbaues (Berlin). SUHM, P.F. 1769. Fors0g til et udkast af en historié over folkenes oprindelse. Copenhagen. THOMAS, L.C. 1957. The linguistic theories of N.J. Marr (UCPL 14). Berkeley and Los Angeles. TRUBETZKOY, N. S. 1939a. Gedanken über das Indogermanenproblem. AL 1.81-9. ROBINS, R .

THE HISTORY OF LANGUAGE CLASSIFICATION

41

. 1939b. Grundzuge der Phonologie. TCLP 7. (Tr. 1949 by J. Cantineau, Principes de phonologie, Paris.) UHLIG, G., ed. 1883. Grammatici Graeci, vol. 1. Leipzig. ULLMANN, S. 1 9 5 3 . Descriptive semantics and linguistic typology. Word 9 . 2 2 5 - 4 0 . USPENSKY, B. 1968. Principles of structural typology. The Hague. (First published 1962 in Russian, Principy strukturnoj typologii. Moscow.) VENDRYES, J. 1 9 2 1 . Le langage. Paris. VOEGELIN, C . F . 1 9 4 5 . Influence of area in American Indian linguistics. Word 1.54-8.

. 1954. Inductively arrived at models for cross-genetic comparisons of American Indian languages. Papers from the Symposium on American Indian Linguistics (UCPL 10), 27-45. Berkeley and Los Angeles. WEBB, J. 1 6 6 9 . An historical essay endeavouring a probability that the language of the Empire of China is the primitive language. London. WEINREICH, U . 1 9 5 3 . Languages in contact. Publications of the Linguistic Circle of New York 1. New York. . 1954. Is a structural dialectology possible? Word 10.388-400. WHITNEY, W . D . 1 8 7 0 . Language and the study of language. London. . 1873. Oriental and linguistic studies, vol. 1. New York.

PART TWO

METHODOLOGICAL

INTRODUCTORY REMARKS

HENRY M. HOENIGSWALD

Many of the earlier volumes of Current Trends in Linguistics covered a geographic area each. This tended to be true in two senses: a volume would set out to chronicle linguistic work carried on in the region, but would also not untypically center on the languages of it. This is not to say that a good deal was not also communicated about concern with general linguistics coming from those particular parts of the scholarly world. The present volume, as R. H. Robins' introductory chapter indicates so palpably, may be said to have diversity itself as its unifying theme. Diversity may be synchronic or diachronic, but the relations between the two varieties are intimate and exist on many levels; scholars have always known and now see ever more distinctly that change in time begets synchronic variability just as much as variability leads to further change. It is the aim of this volume to explore this mutual relationship. Section II is given over to expositions of method — most of them to be matched, in the third section, by specific case studies. This plan expresses an equally strong distrust of shadowy abstraction unillumined by example and of raw data uninformed by theory. But then again, the matching scheme itself could not be, and was never meant to be, a rigid abstraction to be carried into execution no matter what. Therefore, no effort was made to confront specific doctrine with faithful practice, or induction with generalization; and none, of course, could have been dreamt of. Instead, the reader will come upon debate and disagreement, and even contradiction. Nor are the topics chosen for separate treatment in Section II neatly exclusive of one another any more than they add up to encyclopedic completeness. Some contributors have tended to take a more strictly circumscribed view of their topic while others have cast their net more widely. Generally they have followed their own bent in providing emphasis. So do their attitudes toward history vary, with respect to both the earlier past and to 'current trends'. Against the background of the historical and critical mood so firmly set by R. H. Robins, the procedures for linguistic reconstruction are discussed in more expository fashion: first the 'Comparative' Method in the technical sense (with other aspects of comparison coming up for debate again and again throughout the volume), and then Internal Reconstruction as presented by Jerzy Kurylowicz. David Sankoff, in his article on Mathematical Developments in Lexicostatistic Theory

46

HENRY M. HOENIGSWALD

deals critically with the core of a fairly recent subdiscipline of historical linguistics. Paul Kiparsky's paper on Comparative Linguistics (Grassmann's Law) is both a theoretical treatise on the generative approach to diachrony and a case study illustrating that vigorous 'current trend'. Werner Winter's purpose, as he writes on Areal Linguistics, is again expository. Joseph Greenberg treats the Typological Method with fullest reference to the climate of lively debate in which typology and the theory of languages universals have existed. William Labov, too, in analyzing the Social Setting of Linguistic Change, has a great deal to say about the antecedents as well as the present-day interdisciplinary aspects of the direction which he has chosen. Ignace J. Gelb's essay on Written Records and Decipherment is essentially systematic in plan but at the same time replete with comment on theoretical controversy and on examples of decipherment. The third section contains a paper by John R. Krueger on Altaic Reconstruction and Culture. This paper was written to exemplify certain aspects of the extralinguistic interpretation to which linguistic reconstructions are open. Quite aside from its colorfulness and its extraordinary interdisciplinary attraction, this line of research has a special methodological interest because it involves the challenge of recovering lost meanings by means that can to some modest degree be described in formal terms. An example first offered by Paul Thieme (1953-54:549-50) may be cited. The Sanskrit word sphya- denotes an archaic object: the wooden implement used in the Vedic sacrifice to trace designs in the ground. The word is nearly opaque, but not quite; it is not impossible to take it as an adjective derived by means of the suffix -a- ('consisting of ...' or the like) from an -i-/-y- stem noun (as in avya- 'consisting of sheep ['s wool]' from avi- 'sheep'). If this is done, we are left with a residual and unique ('huckle' [-berry]) morpheme, *sphi-, which ought to have some such meaning as '(kind of) wood'. At this point we alter the nature of our reasoning. We determine that *sphimay be interpreted as a reasonable allomorph (cf. Sanskrit vi-: Latin auis 'bird') of the name of the aspen tree which recurs in those Indo-European sister languages that were or are spoken near the tree's habitat. 1 In other words, we first reconstruct internally an obsolete pre-Sanskrit term, complete with its meaning. Secondly, we subject this new piece of evidence to the 'comparative' method. Thirdly, we interpret the obsolescence itself archaeologically; in this case we see in it the result of migration from one botanic region to another. Other examples may point to different interpretations. There are innumerable studies in which cultural or technological change is made to account for polysemy. German Feder has two separate meanings — separate presumably by any semantic analysis and not merely as a matter of English translation — 'feather' and 'writing pen'. Of course there was no crass polysemy as long as people wrote with quills and as long as such turns of phrase as mit einer frisch ausgerupften Feder schreiben 'write with a freshly plucked feather or quill' could therefore occur. To what extent we could reach 1 Thieme 1953-54:15-16. See also Janert 1963-64 and Friedrich 1970:51-2. Note objections in Hiersche 1964:164-5.

INTRODUCTORY REMARKS

47

the cultural inference if we had no direct knowledge of quills and pens is the kind of fascinating question which tests the efficacy of certain kinds of semantic analysis in the same way in which for instance sound change has been said to test the appropriateness of certain phonological theories. Suggestions are not lacking that fruitful work is possible. Most semantic analyses operate with recurring semantic features. On the other hand, historians of given languages and language families who are in the position of having to piece together a plausible picture from both language and nonlanguage evidence, have always insisted on the systematic rather than atomistic nature of their conclusions. Now, to describe a conclusion as 'systematic' means, among other things, that it contains elements that recur elsewhere; that it forms part of a pattern. As Thieme (1964:594-7) points out, not only can Indo-European numerals be (phonemically) reconstructed, but there emerges a decimal system, with suggestions of another more archaic mode of counting in a special way from one to four. Basic kinship terms exist not only for father, mother, husband, wife, brother, sister, son, and daughter, but also for many of the woman's relatives by marriage. It has long been shown that this is consistent with the institution of a patriarchally organized joint family. Religion, in Thieme's opinion, centered around the family, with the 'father' of the house 'functioning as the family priest, in his ... role as butcher ... and host [at the sacrificial feast].' There seem to have been neither temples nor idols, but a rather thoroughgoing separation of the 'celestial' from the 'mortal' and 'possibilities of divinization of natural phenomena' recognizable from the presence of 'abstract methods of grammatical personification' and from the partial existence of a parallel vocabulary, animate and inanimate, for certain natural phenomena. Clearly the system has to be seen as a whole. The testimony of the tree names and of other designations for stationary objects must have a cumulative, convergent effect to be believable; to the extent that such things ever work out without posing further problems, such an effect has been found. Not one, but many, formally independent etymologies contribute to give the Indo-European kinship terminology its characteristic coloring. Recently, Emile Benveniste (1970) has given excellent reasons why he thinks that the Indo-European word *peku (the antecedent of English fee, but also of Latin pecu 'cattle' and of Modern German Vieh 'cattle') had the more general meaning, 'certain kinds of property' first and was only secondarily specialized to designate livestock; he could have added, among his parallels, the fact that neat 'animal of the ox kind' is known to have gone the same way (the word belongs with German geniessen 'enjoy', Nutzen 'use') under, presumably, like circumstances. Just what makes a reconstruction of this sort plausible is not an easy thing to say. No doubt there are cultural universals, and there are certainly cultural expectations of a typological kind specifying what environments, institutions, or technologies are reasonable to reconstruct for a given (?) place and time. No doubt, also, there are similar universal and typological features that limit the linguistic choices, favoring the positing of one semantic change over another, though, in the words of one of the foremost language typologists (Greenberg 1957:58) 'it is often difficult to know what is

48

HENRY M. HOENIGSWALD

retention and what is innovation, for a semantic change that takes place in one direction can often just as easily occur in reverse fashion. A term for "day" often becomes "sun", but likewise a term that means "sun" frequently comes to mean "day"', or one denoting 'livestock' might, on general principles, just as easily go to 'possession' as the reverse. 'Innovation' and 'retention' are indeed the crucial concepts to apply when it comes to judging competing linguistic phenomena within a language family. As will be pointed out later, it is the special strength of the otherwise severely limited 'comparative' method in the narrow, phonological sense to include criteria for judgment within itself (57). Semantic reconstruction rests on a basis which is at once more fragile (methodologically speaking) and richer (because of the greater affinity of its material to the 'natural world'). Its linguistic side is essentially internal reconstruction, though, to be sure, multiple internal reconstruction. This became clear in connection with Thieme's aspen tree. In general, the linguistic properties that provide a lead for useful inferences are those well known in internal reconstruction: degrees and patterns of productivity and isolation, of analyzability and opaqueness. 'Attempts to infer the culture of the speakers of a reconstructed parent language wholly from the forms and meanings of the daughter languages are always dangerous,' says Charles F. Hockett (1948:117-18), but 'the danger is less ... if the forms compared are morphologically simple than if they are compounds.' From many older writers, and from such current ones as Benveniste, and Georges Dumézil, Manu Leumann, Oswald Szemerényi, and Thieme one learns how to reduce the danger even further, through a many-faceted approach which partly invites and partly resists formalization. 2

REFERENCES 1 9 5 4 . Problèmes sémantiques de la reconstruction. Word ( = Problèmes de linguistique générale, 2 8 9 - 3 0 7 . Paris, Gallimard). . 1969. Le vocabulaire des institutions indo-européennes. Paris, Les éditions de minuit. . 1970. Valeurs économiques dans le vocabulaire indo-européen. Indo-European and Indo-Europeans, ed. by George Cardona, Henry M. Hoenigswald, and Alfred Senn, 3 0 7 - 2 0 . Philadelphia, University of Pennsylvania Press. D U M É Z I L , GEORGES. 1 9 5 8 . L'idéologie tripartite des indo-européens. Collection Latomus XXXI. Brussels, Latomus. FRIEDRICH, P A U L . 1 9 7 0 . Proto-Indo-European trees. Chicago, University of Chicago Press. GREENBERG, JOSEPH. 1 9 5 7 . Essays in linguistics. Chicago, University of Chicago Press. BENVENISTE, ÉMILE. 10.251-64

* It is a matter of great regret that no more than these brief notes could be devoted to this important subject as the volume goes to press.

INTRODUCTORY REMARKS

49

1964. Untersuchungen zur Frage der Tenues Aspiratae im Indogermanischen. Wiesbaden, Harrassowitz. HOCKETT, CHARLES F. 1948. Implications of Bloomfield's Algonquian studies. Lg 24.117-31. Also published in Readings in linguistics, ed. by Martin Joos, 1.117-31, Washington, D.C., American Council of Learned Societies; and A Leonard Bloomfield anthology, ed. by C. F. Hockett, 495-511. Bloomington and London, Indiana University Press. JANERT, L . L . 1 9 6 3 - 6 4 . Zur Wort- und Kulturgeschichte von Sanskrit sphyä- (Pali phiya). KZ 7 9 . 8 9 - 1 1 1 . LEUMANN, M A N U . 1 9 5 9 . Kleine Schriften. Zurich and Stuttgart, Artemis Verlag. SZEMERENYI, OSWALD. 1 9 6 2 . Innsbrucker Beiträge zur Kulturwissenschaft. Sonderheft 1 5 . 1 7 5 - 2 1 2 . Innsbruck. THIEME, PAUL. 1 9 5 3 - 5 4 . Die Heimat der indogermanischen Gemeinsprache. Akademie der Wissenschaften und der Literatur. Abhandlungen der geistes- und sozialwissenschaftlichen Klasse 1 9 5 3 / 1 1 . Wiesbaden. . 1964. The comparative method for reconstruction in linguistics. Language in culture and society, ed. by Dell H. Hymes, 585-98. New York, Evanston, and London, Harper & Row. HIERSCHE, ROLF.

THE COMPARATIVE METHOD

HENRY M. HOENIGSWALD

The term 'comparative' has been applied to linguistics in a number of ways. If it were being coined nowadays it would presumably describe an approach under which languages are classified, according to some taxonomic framework, by similarities and differences. It is well known that there is no general agreement on any such framework; as different criteria are selected different classifications are obtained. Still, a presumption is forming that certain principles may be more natural than others. One could, for instance, ask by what means different languages live up to the requirement for the possession of such overt properties as may be thought to have been established as universal. Or one might wish to classify the systems of rules which lead from one, possibly universal, concealed structure to the multiplicity of surface structures that is in fact encountered. These and other possible instances of 'comparison' are not only worthwhile but central both to linguistics as a whole and to what has come to be called, through the accidents and vicissitudes of the history of scholarship, 'comparative' linguistics in the special, technical sense of the word. To avoid confusion it is best to distinguish typological comparison of languages from applications of the 'comparative method'' in the more technical sense just alluded to. We shall here be concerned with this 'comparative' method. Although this distinction has not always been sharp, the idea of using the comparison, that is, the matching of items in two or more languages with a view to reconstructing their common ancestry, has been present ever since Friedrich von Schlegel, in analogy to comparative anatomy, first spoke of comparative grammar in 1808. It is perhaps ironic that the subsequent development including the work of Bopp, Schleicher, and the practitioners and theorists of the last hundred years like Antoine Meillet, has been contrary to the original bias implied by the choice of the word 'grammar' and has led ever more firmly in the direction of phonological work. Comparative grammar, where the term is still employed, on the whole means 'comparative' (that is, reconstructive) phonology, all kinds of disclaimers to the contrary. While there are methods for the reconstruction of both grammatical and lexical meanings, there is no 'comparative method' of a kind that would simply extend or duplicate the steps of which the comparative method proper consists. As we shall see, there are deep reasons why this should be so. Like all systematic work in historical linguistics, the comparative method rests on

52

HENRY M. HOENIGSWALD

the feasibility, up to a certain point, of separating the effects of one change process called 'sound' change from those of all others. The so-called regularity of sound change has been variously presented as an empirical finding or as a 'postulate'. Perhaps it is wiser to think of it simply as part of a definition. Some replacements (for instance that of an older Latin ae by a classical Latin I in all non-first syllables — conquaerd>conqulro 'I procure'; elaeva>oliva 'olive' [incidentally, a loanword]; tutudae > tutudl 'I have pushed' 1 ) are such that they can be formulated in phonological terms (see the preceding parenthesis); others require listing (the older English participle ending -ende, -inde has been replaced by -ing(e): the old English word for calf, where not retained — that is, replaced by itself — in those utterances where it referred to the living animal, was replaced by veal, a 'loanword' from French, with the meaning referring to the meat). The latter are not 'sound' changes, while some of the former are, namely those for which origin by certain analogic processes, and by dialect borrowing, can be excluded. These concepts need some discussion. As earlier Latin and classical Latin are confronted, it is found that ae in non-first syllables is replaced by I in a significant number of cases (see above for examples). It is, however, also found that ae is replaced by ('remains') ae in a significant number of cases also (e.g. concaedes 'barricade', praesaepe 'stable', exsaeuid 'I cease to rage'). Further investigation turns up the fact that the latter instances all exhibit morpheme boundaries before the syllable with -ae-, and that the morph with -ae- in it can occur also at the head of a word (where it is then no longer in a non-first syllable: caedd 'I fell', caedes 'slaying', saepes 'hedge', saeuio 'I rage'). Some instances of ae I occur under the same circumstances (conqulro : quaerd 'I seek') but not all do (tutudl, ollva are not so compounded). From this it is possible to conclude that ae in non-first syllables > I (conqulro, ollva) and that cases of ae in non-first syllables constitute new compounds postdating the sound change (prae-saepe). If such cases are extant from both before and after the sound change the continuity is only apparent: actually it must be assumed that the remade form has successfully competed with a changed (but unattested) form. That this is the correct interpretation is suggested by doublets (concaedes but concldo 'I cut to pieces') including precious instances in which the two competing shapes are not only preserved but commented on (both the satirist Lucilius and Cicero mention the pair per-tisus, per-taesus 'disgusted' with the implication that per-tlsus is affected and archaic and per-taesus is current usage). This history may be summarized by saying that a sound change has affected all instances of ae in non-first syllables, thereby creating alternations between, say, simple verbs (with unchanged ae) and compounds (with i), and that a subsequent analogic change has created new forms with non-initial ae. These new creations frequently displace their older, phonologically altered synonyms. Since this process can go very far, 'isolated' words like ollva are of decisive importance. 1

The spellings in these examples are normalized in ways which do not necessarily agree with accepted styles. The notion that there are minor sound change processes which are sporadic rather than regular but still sound changes is ill-defined; see Hoenigswald 1966.

THE COMPARATIVE METHOD

53

There are, however, less telling instances in which two potential 'sound' changes may also appear to be at work in contradictory fashion. Older Germanic material with sk- appears in English in one of two, or in two, shapes: as sh- in shaft, shroud, shine, and as sk- ('sc-') in scold, sky, scrape. Occasionally, again, there are doublets: shirt, skirt, shriek, screech. The background is well enough known, as it happens: the items with sk- are in many cases loanwords from Scandinavian while those with share not. But in the absence of such explicit extra information it may be difficult to assert more than that there has been more than one channel of transmission. To interpret these in terms of which is the primary or native one and which represent borrowing from sister languages or dialects, requires a belief in certain generalities about borrowing, such as that borrowings remain inferior in number either in the general or in the 'basic' vocabulary — a particular belief which it is not easy to maintain especially in the case where the general historian would like to see intimate borrowing between closely related dialects. It is of course true that where 'learned' borrowing (as contrasted with the 'popular' primary transmission) occurs from an artificially cultivated earlier stage of the language, the effect is visible in the 'tatsama' or phonologically archaic appearance of the borrowing as it contrasts with its 'tadbhava' and only covertly identical counterpart (e.g. French captif 'captive' borrowed, vs. chétif 'wretched' descended from, Lat. captivus). The uncertainty need not be indicative of any great theoretical weakness. It is not at all clear that the processes of sound change and borrowing deserve to be distinguished as different in kind at least not when the borrowing is total rather than selective, and when the speech forms in contact are closely related dialects to begin with (Hoenigswald 1960). These are the provisos to be added to the statement made earlier, p. 52. Besides, each channel of transmission, and each set of phenomena that results, is amenable to treatment without any one having primary status. The core vocabulary of Albanian may be matched directly with that of other Indo-European languages for the purpose of reconstructing proto-Indo-European; the Turkish loanwords in Albanian may, in theory, be matched with other forms of Turkic for analogous purposes. As it happens, the body of borrowings from otherwise unknown Balkan Romance sources into Albanian can be used with profit toward the reconstruction of Vulgar Latin (proto-Romance). The French loanwords in English could be treated as representing a Romance language. One of the reasons for the efficacy of the comparative method lies in the fact that it is essentially independent of notation, since it is based on processes which must appear invariant under any notation chosen, be it bi-uniquely (autonomously) phonemic, distinctive-feature-oriented, morphophonemic, systematically phonemic, or what not. If, upon confrontation of an earlier stage with a later stage of the same language (the term will be examined later), it develops that two consistently different items have been regularly replaced by what an actual or ideal pair test establishes as one and the same (homophonous) item, and if the process is indeed a 'sound' change as just characterized, the merger process must appear recorded in some way under any

54

HENRY M. HOENIGSWALD

system of notation. 'Verner's law' states that the IE sequences ata adha adha all went, in Germanic, to ada, while ata has become apa. There are various ways of describing this state of affairs. First of all, since the same has happened to apa abha abha to the exclusion of apa, a distinctive feature notation which extracts the voiceless stoppage from both t and p, and the aspirated voiced articulation from both dh and bh, is of course particularly appropriate. As is well known, there are a number of matrix-like devices to accomplish this. Similarly, it would be best to extract the accent (and the datum that there are unaccented vowels as well) from the 'segment' accented (unaccented) vowel — something that happens to be conventional practice. It may further be of importance to assign markedness to one member of an opposition rather than to the other according to particular criteria. No matter what the choice, however, the formulation (unless inaccurate) will somehow show that what phoneticians would describe as voiceless consonants, if preceded by unaccented syllables (but, as it happens, not also by sp k), become homophonous with voiced aspirated ones, while those after accented syllables do not; and also, that the contrast between accented and unaccented syllables is wiped out in all kinds of environments, including many not involving a following t dhp bh. In this special context, there is no need to be concerned about whether the merger of the weakened t with dh occurred before or after the general loss of the accent contrast, nor whether that loss should indeed be thought of as 'one' event (a notion which lacks definition in any case; the microchronology of such histories, when known, always turns out to be highly complicated). However this may be, we have reason to regard phonological merger as the central process of sound change, and therefore as the central factor to enter into any theory of the comparative method. For one thing, mergers may or may not entail their converses, namely conditioned sound change or split, while conditioned sound changes of any importance presuppose (or rather: form an aspect of) merger. Suppose (1) an element (a phoneme, distinctive feature, ...) t and an element d (or, for that matter, an element t and V , that is the absence of an element) merge (into '?', or 'd\ or T \ or 'x' — or, for that matter, 0). There will be potential homonymy and a reduction in the inventory (the 'alphabet'), but otherwise no effect. If on the other hand, (2a) t at the beginning of a word goes to (in this case: 'remains') t while t in word-interior position becomes d and merges with a d that occurs there already, the consequences are different. Among other things, it is possible to describe the merger of some t's with d as a split, conditioned by position in the word, of the element t. Consider also (2b) the following case: k goes to c before i (but not before a and u), while i and a merge into e. This will transform the sequences ki ka ku ni na nu into ce ke ku ne ne nu respectively. No matter what the notation, the 'split' of k into k and c will appear as a corollary of the vowel merger the reality of which is guaranteed, as it were, by the homonymy of ne < ni and ne < na. We may refer to (2a), where a positional variant separates to the point where it merges with another phone, as a primary split; and to (2b), where the merger affects the conditioning environment rather than the splitting element itself, as a secondary split.

THE COMPARATIVE METHOD

55

In either case, the split clearly has not taken place until the merger is a fact. It is only weaker cases than these where the notation may seem to make a difference; but the effects remain superficial. An IE b becomes p, and IE p (on the whole) becomes / in Germanic, except that sp becomes (remains) sp. But there is no IE sb; consequently, sp, in appearing as later sp, has nothing to merge with. To say, then, that the older p has split into p (after s) and/(otherwise) is only a concession to 'mere' phonetics. The alternative would be to say that p>f unconditionally, but that the 'allophone' of / after j (at the later stage) is stop-like and 'resembles' the p's of the language. If we construct a history in the course of which a speech community divides so that different sound changes affect the same language in each offshoot, we may obtain relationships such as the following. Suppose the primary split described in (2a) takes place in one 'daughter' language (B), while the other (A) remains (in this particular regard) unchanged. We shall then obtain the following correspondences: (3) A t / B t\ (4) A t / B d; (5) A d I B d. It will furthermore be true that (unless other mergers in the environment occur as well) tjt and tjd will be found in settings that are mutually exclusive but which in the aggregate make up an entity that intersects with those settings in which dfd is found. In our case, did is both word-initial and word-interior; //i is only word-initial, and i/rf only word-interior. If these three correspondences were projected back into the hypothetical ancestor as *x, *y, and *z, respectively, and *z would fail to contrast and would have to be treated as positional variants. Since the correspondences on which they are based share one component, namely the t of language A, it is easy to think of them of sharing a distinctive feature or of being 'similar' in some analogously defined way. Hence the maxim that reconstruction by the comparative method is attained by assembling correspondences, examining them for partial likeness and for 'complementarity' and reconstructing elements (for instance, phonemes) of the ancestor by grouping together partially alike and mutually exclusive correspondences. The principle continues to function in the face of considerable complexity in the data. Consider, for example, the following situation. In descendant C, k > 0 after u, thereby merging with extant 0 (the language has always had the sequences uu ui ua) by primary split, while after # , k > g. In descendant D we have the secondary split described in (2b) above. The result must consist in the following correspondences between C and D : (6) aje\ (7) u/u; (8) i/e; (9) 0/0; (10) 0/k~, (11) g/k; (12) 0/c; and (13) g/c. Clearly, (6), (7), and (8) contrast (for instance, after vowels corresponding to each other, say in ua/ue uujuu iti/ite). Consequently, something like a proto-*a, proto-*/, and proto-*w are to be assigned to the ancestor, with a merger of *a with *i into the e of language D. Then, (9), a correspondence possessed of necessarily extremely wide but by no means indefinite privileges of occurrence, 8 contrasts both with (12) — u0i (that is, ui) / u0e contrasts with wi/iice — and with (10) — u0u/u0u, uoa/u0e contrast with mujuku, wa/uke — as well as with (13) and with (11). Thus, the 0 of (9) belongs to the ancestor in all these cases. But (10), (11), (12), 2 See Lg 35.409-20 (1959).

56

HENRY M. HOENIGSWALD

and (13) are all complementary with one another, (10) and (11) occurring only before u/u and a/e, (12) and (13) only before i/e; (10) and (12) only after w/m; and (11) and (13) only after # . This being so, the correspondences (10), (11), (12), and (13) lead to the reconstruction of one element, thus recovering our starting data. It is obvious that the alphabetic choice, ik\ for this element *x, with all its phonetic implications, is in a sense arbitrary; such choices are often made on the grounds of some belief held in certain universals of sound change, or upon consideration of typological factors which make some phonetic developments more probable, especially in certain conditions, than others. What the comparative method establishes is, however, the contrastive identity of '*&' — it is one element, not two or four — as well as the reality of the split processes which have taken place separately in C and in D. The procedure here described not only has limitations that can be pinpointed clearly, for instance in the well-known shape of independent duplication, in descendant languages, of given change processes, which make it appear that the change had already occurred in the ancestor; or in the form of one and the same correspondence arising from intrinsically unconnected processes (for instance, since IE dh > Germanic d, IE t > Germanic ¡> but > Germanic d after unaccented syllable, and IE dh > Greek th but > Greek t if another 'aspirate' follows in the next syllable, while IE t > Greek t, the correspondence d/t comes into existence twice, once from dh before another aspirate and once from t after unaccented syllable; only under certain conditions is it possible to partition the occurrences of these correspondences in such a way as to establish the complementarity of each portion to some other correspondence, in this case d/th and p/t). There are other difficulties as well, especially those which mirror the imperfections of the notion of complementarity. It is shown elsewhere in this volume (115-134) how considerations of internal reconstruction and of generative rule ordering can be introduced to do full justice to the workings of history. Suffice it to say here that the comparative method and the method of internal reconstruction, however intimately they may be intertwined in theory and in execution, curiously supplement each other in one particular respect: internal reconstruction serves to retrieve split processes but is powerless with respect to mere (unconditional) mergers, while the comparative method is particularly effective with the latter. While it has been said that reconstruction, aiming, as of course it should, at 'realism', should hesitate to assign to the ancestor a typology that is significantly different from that which characterizes the descendants, this principle can have only limited validity; after all, descendants of one ancestor can differ typologically from one another (and change is not always typologically superficial but may also lead to serious alteration); and who is to say that serious typological alteration could not also sweep a whole area after an ancestor has become differentiated? Many Algonquian languages have, in their phonemic inventory, the items and The correspondences are often straightforward: n/n, tit as, say, between Fox and Cree. Between these two languages, there is, however, one additional correspondence, nit. All three correspondences contrast in non-complementary fashion, so that it is necessary to set up three, rather

THE COMPARATIVE METHOD

57

than two, proto-elements, and thus to assign to the ancestor a typological trait foreign to the descendants mentioned. Such reconstructions are frequently resisted until substantive confirmation turns up. In this case, a southwestern Algonquian language, Arapaho, furnishes such confirmation in the shape of a '0' in those precise instances in which Fox and Cree exhibit n/t. Arapaho is, in this respect, archaic, while most other descendants have reacted to a trend aiming at a reduction of dentals. The fact that they responded to this uniform trend through different machineries — that is, Fox merging 6 with n, and Cree merging 9 with t — makes the matter accessible to the historian (Gleason 1961:448). We have already alluded to the real source of strength of the comparative method when we referred to the 'reality of the split processes' that can be reconstructed as having taken place in descendants (56). These split processes, together, of course, with the merger processes on which they depend (54 f.), constitute part of the famous 'innovations' which are so important for the business of establishing trees of descent with subancestors dominating subfamilies within families. From what was said it is clear that if an element x of a language A corresponds sometimes to a b and sometimes to a c of language B, the comparative method is sufficient to let us decide whether the ancestor, X, was, in this particular respect, like A in having no contrast, or like B in having contrast. The decision is made according to whether or not the correspondences, a/b and ale as wholes, are in contrast or not. If they are, A has 'innovated' in eliminating original contrast; if they are not, B has innovated in carrying out those mergers which have the effect, analyzed earlier (54f.), of splitting two erstwhile positional variants into b and c. Phonological merger is by no means the only kind of innovation that needs to be considered in judging subancestry: it is, as has frequently been pointed out, not even the most characteristic in that others (e.g. grammatical innovations, or certain innovations of a lexical sort) are less likely to be independently and 'accidentally' duplicated. But it has the unique virtue of being demonstrable through the matching technique of the comparative method itself. Lexical innovations, as a rule, need to be identified as such by internal reconstruction (e.g. a compound may be judged more recent than, hence possibly a replacement of, an opaque term; a phrase utilizing a relic morph may be judged relatively old; etc.) or simply through direct knowledge (some Romance languages have a word for 'mare' which continues Latin equa; others have one based on caballa: it happens to be known that caballa is, on the whole, a late loanword within parts of the Latin-speaking area. 3 Family trees are rooted graphs which represent filiations of languages in time in terms of innovation and retention. According to what was just said, we may think of family trees as based on the recognition, through the comparative method, of phonemic mergers, although this restriction gives a quite unrealistic picture of the terms in which such filiation has in fact been discussed. Still, let us consider the following model: 3 Thieme 1964 offers an admirable discussion of the problems of reconstructing morphological and semantic traits. See also Benveniste 1954.

58

HENRY M. HOENIGSWALD

1 Language A Language B Language C

e e

2 3 e a e a a a

4 5 6 7 u u 0 0 o o 0 a o 0

There are three 'related' languages, that is, languages which exhibit correspondences (in addition, always, to other, non-phonological marks of relationship). There are 7 such correspondences; 1-5 are to be thought of as contrasting, and so are 6 and 7 (absence vs. presence of distinctive vowel length). Evidently, the ancestor, X, must be assigned five vowels ( T 'e' 'a' V V ) and a contrast between short ('0') and long (':') quantity; this is a function of contrastive status. However, as the comparative method is applied pairwise, it develops that the special ancestor ('Y') of A/B is different from the other two special ancestors A/C and B/C which are in fact alike, to each other as well as to X itself. Y has four vowels (correspondences 4 and 5 not being distinguished in A/B). This to say that A and B form a subfamily characterized by one shared innovation, namely, the merger of *u into one vowel (whatever notation one wishes to devise for writing it). In addition, each of the three languages carries out at least one individual innovation: A loses vowel length, B merges *i with *e, C merges *e and *o with *a. The result is that A/C and B/C may be said to share 'characteristics' to the exclusion of B and of A, respectively (namely, for A/C the difference between the two original front vowels, and for B/C — synchronically more concrete — the presence of distinctive length). But quite unlike the 'characteristics' common to A and B, these are retentions rather than innovations, and, being phonological in character, they can be proved so without going beyond the resources of the comparative method. Note, also, that our model is constructed in such a way as to show how typological grouping may freely intersect with subfamily grouping without necessarily interfering with our ability to recover either. There is a typological bond between languages B and C to the exclusion of A, in the sense that B and C both possess (1) a three vowel system (with mid and low qualities, if the implications of the alphabetic choices made here were to be accepted literally), and (2) distinctive vowel length. Students of linguistic geography and of typology may wish to interpret this picture by reconstructing (a) a twofold separation of the ancestral speech community (X) into Y and C (Y X

A

1 ! B

r C !

THE COMPARATIVE METHOD

59

with the two rounded vowels merging; C with the three non-high vowels merging); (b) a subsequent separation of Y into A (with elimination of vowel length) and B (with merger of the two front vowels, that is, with a development which is consistent with the idea (c) that B came to form neighborly contacts with C within the framework of a 'language area'). Ideally, this is not a question of amassing evidence. If the tree is a true depiction of actual history, and if the shared innovations are significant rather than trivial (that is, if they are not open to the suspicion of independent duplication) one example is as good as a hundred. In other words, if 'one' shared innovation links A and B significantly against C, there should be none to link A with C or B with C. Accumulating evidence can therefore only have the purpose of excluding accidental duplication. In actual fact, overlapping linkages are not rare; in some language families they occur between languages which are now, or at some slightly earlier known period in history, still relative neighbors in the sense that no other dependent languages are spoken in intermediate territory. This is the rule in Indo-European; it may be interpreted as showing that while the speakers of Indo-European dialects moved radially from a center there were no violent displacements of their mutual geographic position. Other language families behave differently. If Hungarian and the Ob-Ugric languages of Siberia form a possible subgroup within Finno-Ugric — if, that is, they share innovations significantly — it is remarkable that they should be separated from each other by the entire width of Finno-Ugrian territory on the language map. The historical background for this is none too hard to understand if we consider the manner in which the Magyars appear on the medieval European scene. The phenomenon of isoglossic overlapping of innovation areas has been cited as an argument against the validity of the notion of the family tree. It would be truer to say that trees depict certain types of history (involving clear breaks without subsequent contact) and overlappings according to the 'wave' pattern depict certain other types. It is even conceivable to retreat from the general claims made by the champions of the family tree to a position where it is simply maintained that trees describe the relative chronologies and geographies of particular competing (that is, retained and innovated) features. It is certainly true that relationship (that is, ultimate ancestry) is easier to prove than particular subrelationships. While the proof of subrelationship can never become emancipated from the necessity of performing all pairwise reconstructions (an enormous task, though one which is in principle amenable to computerization) and ordering their results, proof of mere relationship reveals a more interesting play between the principle that all 'comparative' operations are binary by nature even when extendedjto whole columns of languages (quite beyond the modest use of A/B/C made at the outset of our discussion of the model above) and the important statistical fact that only 'mass' comparison allows us to exclude pseudo-sharings of a kind which universal and typological factors will always introduce into mere two-language 4

Greenberg 1957:38, 46-55 ("The problem of linguistic subgrouping"); 1963:1-5 ("The methodology of language classification"). See further Allen 1953 and Hall 1950.

60

HENRY M. HOENIGSWALD

material. In fact, for purposes of ordinary language classification by family, 'mass' comparison has rightly been called indispensable. 4 The comparative method has one rather paradoxical use which is based on the following consideration. Let two related languages, A and B, of unspecified (relative) chronology, be given. By applying the comparative method to all observable correspondences, we obtain a reconstruction of their common ancestor X. In the most general case, as we know, X is different from both A and B, both A and B having gone through their respective innovations. It is however possible that only one, say B, has innovated in any way. We know this to be the case, whenever X is found to be identical with A (except for the choice of phonological symbols and the like). Since change is, however, a never-ceasing condition, this must mean that A is older than B, and that, specifically, A is the ancestor or 'older stage ' of B. Thus, contrary to what is often held, relationship is a more primitive notion than descent; descent is a special case of relationship, discoverable by the comparative method. Nor is this an abstract formulation: actual controversies on the subject of direct as against collateral descent (e.g. on whether the Romance languages are or are not daughter languages of Latin as known from the documents) can be shown to have been argued on these very grounds. A word may be also said by way of hinting at certain points of contact between the comparative method and glottochronological lexicostatistics in a wider sense (93 ff.). On the assumption (which this is not the proper context to question) that there is a correlation between the extent of change (measured by vocabulary replacement or in some other way) and the lapse of time, it is clear that calculations of the relative extent of identity and diversity (such as would result from change) could be used to construct trees. Attempts of this sort have of course been presented and are analyzed elsewhere in this volume. If successful, these attempts could replace the comparative method, not as a means of reconstructing features of older stages but as a means to determine lines of subancestry and descent as such; being independent of the comparative method, glottochronology could at least be used for purposes of mutual confirmation. As may be suspected, the comparative method is more powerful in one important respect than a glottochronological scheme, even if statistically valid, for the determination of subrelationships. The reason lies in the fact that the comparative method, deciding as it does between what is retained and what is changed, gives us information on matters of direction, whereas lexicostatistics does not. Along a given line of descent (that is, along what the comparative method establishes, as we have just seen, as a given line of descent) we may find that the descendant has retained, perhaps randomly, a given percentage of 'basic' vocabulary (say, 80 % per time unit, e.g. per millennium). If the chronology is unspecified, all that is observable is the datum (if indeed it is a datum) that eighty per cent of the basic vocabulary for the language pair is shared (except for the effects of sound change, etc.); there is no way of determining

THE COMPARATIVE METHOD

61

which is ancestor, which descendant. Furthermore, suppose the lapse amounts to two time units (say, millennia). Under the assumptions made by some writers, the effect will be a shared vocabulary portion of 80 % of 80 %, or 64 %, again without indications of direction. If, on the other hand, we are dealing with two contemporary daughter languages, A and B, descended from an ancestor, X, one time unit (one millennium) in the past, they will share the product of the two applicable percentages, namely likewise 64% of their basic vocabulary. What an effective lexicostatistic method could determine is thus the distance (in metric time, that is, geometrically, in terms of length; not in terms of edge counts between vertices) along the branches of a family tree, regardless of their direction in time. Concretely, the respective lexicostatistic distances of A from B, and of A from C (expressed in percentages of shared vocabulary) would be the same in this instance

62

HENRY M. HOENIGSWALD

whereas the comparative method would not only provide a decision, but also permit us to infer the properties of X at the apex of the first and the third tree.

BIBLIOGRAPHY

The following is a selected list of recent publications bearing on the subject. 1 9 5 3 . Relationship in comparative linguistics. TPhS 1 9 5 3 . 5 2 - 1 1 2 . E. 1 9 5 4 . Problèmes sémantiques de la reconstruction. Word 1 0 . 2 5 1 - 6 4 . (Reprinted 1 9 6 6 , Problèmes de linguistique générale, 2 8 9 - 3 0 7 , Paris.) BHAT, D.N. S. 1 9 7 1 . Hoenigswald's patterns of sound change. Indian Linguistics 1971.295-300. Poona. GLEASON, H. A. 1961. An introduction to descriptive linguistics. 2nd ed. New York. GREENBERG, J. 1957. Essays in linguistics. Chicago. . 1963. The languages of Africa. IJAL 29/1. HAAS, M A R Y . 1 9 6 6 . Historical linguistics and the genetic relationship of languages. ALLEN, W . S . BENVENISTE,

C T L 3.113-53.

HALL, R.A., JR. 1950. The reconstruction of proto-Romance. Lg 26.6-27. (Reprinted 1957, Readings in linguistics, ed. by Martin Joos, 1.303-14, Washington, D.C.) HOENIGSWALD, H . M . 1960. Language change and linguistic reconstruction. Chicago. . 1965. Phonetic reconstruction. Proceedings of the 5th International Congress of Phonetic Sciences, 25-42. Basel. . 1966. Graduality, sporadicity, and the minor sound change processes. Phonetica 11.202-15. K I N G , ROBERT D . 1 9 6 9 . Historical linguistics and generative grammar. Englewood Cliffs, N.J. KIPARSKY, PAUL. 1968. Linguistic universals and linguistic change. Universals in linguistic theory, ed. by Emmon Bach and Robert T. Harms, 170-202. New York. LEHMANN, WINFRED P . 1972. Historical linguistics: An introduction. Second edition. New York. MEILLET, ANTOINE. 1 9 2 5 . La méthode comparative en linguistique historique. Paris. (Reprinted 1 9 5 4 . ) (English translation 1 9 6 7 , The comparative method in historical linguistics, by Gordon B. Ford, Jr., Paris.) SANKOFF, DAVID. 1 9 7 3 . Mathematical developments in lexicostatistic theory. CTL 11.93-112.

THIEME, P. 1964. The comparative method for reconstruction. Language in culture and society, ed. by Dell H. Hymes, 585-97. New York. W A N G , WILLIAM S.-Y. 1969. Competing changes as a cause of residue. Lg 45.9-25.

INTERNAL RECONSTRUCTION

JERZY KURYLOWICZ

0.1 Linguistic reconstruction is a procedure which may be applied either to a single language or to a group of genetically related languages. Either there is one set or several sets of data at our disposal. Yet the goal we pursue is in both cases the same. Relying on the data we try to establish a chronological sequence of changes which have produced the attested linguistic state. Such an approach is different from that of generative grammar whose analysis may accidentally coincide with the results of historical reconstruction. 1 0.2 The term internal reconstruction, used for over thirty years,2 refers to new methods and aspects of reconstruction applied to the material of traditional comparative grammar. It means application of theoretical insights and practical experience accumulated in the last decades owing to the rapid development of general linguistics, chiefly of the so-called functional (or classical) structuralism. In this sense already the neogrammarian school of comparative grammar had recourse to certain tenets concerning linguistic evolution in general, embodied in H. Paul's Prinzipien der Sprachgeschichte (1st ed. 1880). However, general linguistic considerations seem to play an ever increasing role in historical reconstruction. We become more and more convinced that the apparently infinite variety of linguistic changes, whether phonemic or morphological, can be broken down into elementary relevant stages and reduced to a certain number of types. Reconstructions based on modern linguistic theories are still the most reliable when there is an absence of sufficient historical material. Thanks to the progress of linguistic theory not only have hitherto unsolved problems of comparative grammar been successfully tackled, but also traditional solutions replaced by more convincing ones. 0.3 Because of their more than thirty-centuries-old history and the abundance and variety of their written documents, the Indo-European languages provide a domain of study particularly apt to contribute to the theory of universals and to the methodology of reconstruction. 1 Especially in cases where relatively short and recent prehistorical or preliterary periods are concerned. 2 Cf. "On the methods of internal reconstruction", PICL 9.9-36.

64

JERZY KURYLOWICZ

1.1 Just as in the nineteenth century, it is the recent progress in the study of sounds which has paved the way for a more rigorous study of function (meaning). The firmly established notions of phoneme, allophone, phonemicization, and so forth, caused comparatists to reconsider many traditional views taught and accepted even today. 1.2 The problem of the so-called Indo-European 'gutturals' may serve as a classic example of an obsolete scientific approach. Three 'guttural' series, transcribed k, k, qv (or kv, kw) have been reconstructed up to now in comparative grammars and dictionaries. But no historical language gives evidence of the original existence of more than two, either k and k (in the 'satem' languages) or k and qv (in the 'centum' languages). There is of course no obstacle to using all three symbols in the 'asterisk language' of reconstruction, 3 especially for didactic purposes. E.g. *leiqu 'leave' (Got. leiha) implies that the corresponding forms of the satem languages will offer k (e.g. Lith. lieku), whether the labiovelar qv is of IE origin or only dialectal. And vice versa the k of *kei 'lie' (Ved. saye) implies a A; in the centum languages (e.g. Gr. K£i|iai). But the fact that out of a dozen linguistic groups not one testifies to a distinction of three points of articulation and that the existence of qv automatically presupposes the absence of k and vice versa (jeu de balance), has led to an early reconsideration of the traditional solution. According to H. Hirt and many others, palatalization of velar stops, i.e. their assimilation to a following e, I, i, being a common enough phenomenon, one may ascribe the ^-series of the satem group to a dialectal development occurring in a number of IE languages. On the other hand, since a dissimilation ke>qve, ki>qvi (preventing the merger of Are, ki with ke, ki) is also possible, the labiovelars of the centum languages could also be of secondary, dialectal origin. But whichever point of view is adopted we cannot account for individual exceptions to the rule, like *sueks 'six' or *oktou 'eight', etc., in the former, *agvnos 'lamb', *qvas 'cough', etc., in the latter case. 1.2.1 The traditional explanation postulating a secondary introduction of k (or qv for that matter) before back vowels and consonants by way of morphological 'analogy', contains an inherent contradiction since the original absence of the allophone k (or qv) before back vowels and consonants impedes their morphological spread in these positions. As long as the k of *spek-ie/io- 'look' is an allophone of A:, the k of *spektos cannot be replaced by k. 1.2.2 The chief problem which has not been clearly envisaged, is the phonemicization of k (or qv). Whereas phonetic palatalization, producing allophones, is a general and trivial enough phenomenon, the rise of the phoneme k (or alternatively qv) presupposes a merger of the phonetic surrounding of the allophone k (qv) with the phonetic surrounding of the chief allophone k. 8 As a matter of fact this is a good example of what E. Pulgram calls reconstruction of a 'diasystem', cf. Linguistics 4.66-82 (1964).

INTERNAL RECONSTRUCTION

65

Typical examples are provided by later developments occurring in the majority of Indo-European languages : a) Disappearance of i, i.e. its merger with zero explains the following cases : In Slavic ka, ku:kia, kiu become ka, ku:k'a, k'u (>ca, cu); hence also phonemicization of the palatal allophones of k(e), k{i) (>k'e, k'i > ce, ci). But k is always preserved before consonants. Similar developments occur in Old English, Swedish, many Danish dialects and in Romance, cf. Uapophonie en indo-européenne, p. 368 f. b) The North Slavic merger of word-final and syllable-final & and & (into zero) creates a difference between 'hard' and 'soft' ( = non-palatal and palatal) consonants, e.g. pçtb 'fifth' > pçt, but pçtb 'five' > pqt\ the soft pronunciation being then carried over to consonants followed by front vowels. For a similar development in Old Irish and Albanian cf. Uapophonie..., p. 370. c) In Indo-Iranian the merger of IE e and o accounts for the rise of the opposition ca.ka. Similarly the SI. merger of IE ë and oi, both resulting in ê, explains the difference between cë (first palatalization) and ce (second palatalization). The mergers in question are therefore mergers producing a split. As a rule the new phonemic difference (opposition) is neutralized in a zone of complete merger, showing the unmarked member of the opposition. E.g. in Slavic only k, never c, c, appears in preconsonantal position. The same is true for Indo-Iranian. In Old Irish -cht has no palatal counterpart etc. 1.2.3 On the strength of the above examples one may be tempted to look for a merger explaining the rise of a third guttural series in Indo-European, whether k or qv. But let us stress once more that whatever the general solution, a certain number of individual forms will necessarily remain unexplained. The rise of the third guttural series goes back to such a chronologically remote period that subsequent dialectal sound changes, loans, contaminations, etc., must necessarily have blurred the original distribution. The disappearance of i between consonant and vowel (at least between velar consonant and vowel) would have led to a difference: ki (+vowel) in some languages, k ( + vowel) in other languages. There is one word which could suggest such a hypothesis : *ghiem- 'winter' (Gr. xirôv, Lat. hiems, but Gmc *gôm, Olr. gam). In the satem languages, however, the i is well attested, cf. Av. zyà, vocalized in gen. zamô for *zimô, Ved. instr. himâ, Balt.-Sl. with secondary full grade *zeimâ. But this distribution (ghi>gh in some centum languages) is contrary to what is to be proved, viz. ghi >gh in the satem languages. Moreover it is evident that it would be impossible to derive the k in *kër{d) 'heart', *kel 'to hide', *ker3(s)- 'horn', *krptom 'hundred', etc., from an old *ki with an i nowhere attested. Nor is there any indication of a phonemic merger of two vocalic qualities, velar and palatal, comparable to IE e — o in Indo-Iranian, which could have phonemicized the difference between the allophones k\e) and k(o). 1.2.4 But one possibility still seems to be open. One could attribute the difference

66

JERZY KURYLOWICZ

k : k in preconsonantal position to the disappearance of the reduced vowels e and o which must have taken place in early Indo-European. This would correspond to type b) of phonemic palatalization (cf. above). Then of course the preconsonantal opposition *kru 'blood': *kleu 'hear', etc., would have to be ascribed to the disappearance of o in the former, e in the latter example. But attempts to prove the difference in vocalic quality result in a vicious circle. Although such an origin of the palatal series is phonemically admissible, the arguments necessary to prove it are transcendental, inaccessible for internal reconstruction. 1.2.5.1 However the merger we are looking for can be found, though not for & but for the labiovelar (q v ). I.e. assuming that the phoneme^" is due to a relatively recent, dialectal development of the centum languages, it can simultaneously be shown that it is the result of a phonemic merger. Let us therefore suppose that Indo-European distinguished only between velars (k) and palatals (k), the problem of the origin of the palatals themselves being transcendental, i.e. outside the scope of reconstruction. Then the qv of the centum languages is a continuation both of k and of ku (g and gu, gh and ghu), kept apart in the satem languages. IE k and ku meet halfway, u losing its status as phoneme and becoming a simple phonemic feature. The change k > q " takes place before e, ? (that is under the same conditions as the alleged palatalization k > k). As regards satem velar + u = centum labiovelar, cf. e.g. Lith. kvapas 'breath': Got. af-fvapjan 'smother'; Skr. pakva- 'ripe': Gr. tietccov; Skr. sata-gvin-: Gr. ¿kcitoh-Pti; SI. gvozdb 'forest': G. Quaste 'foliage'; SI. gvozdb 'nail': W. both 'tail, penis'; SI. gvezda 'star' (*ghuoig): Gr. s and s>s are different, the former change occurring between any vowel and stop, the latter after i, u, r, k, and any following phoneme. Cf. on the one hand not only Skr. pi?fa- d (after unstressed vowel; Verner's law) • There is no necessity to speculate on the phonetic realization of IE dh, alternations with related phonemes being the only relevant criteria of its position in the system. In this respect the incompatibility of root-initial t (dh) with root-final dh (f) may be mentioned suggesting that between t and dh there was a difference of one phonemic feature (not of voice and aspiration), whereas roots of the structure d-dh or dh-d were possible. These facts prove that the Indo-European phonemic oppositions were t:d and t:dh, but not d:dh. 7 There would be a merger of th and dh after initial s- and (probably) in word-final position. The possibility of IE initial clusters s + 'voiced' aspirate is borne out by examples like Gr. atpe < *sebh or axelv < *segh.

69

INTERNAL RECONSTRUCTION

Now 1) and 2) are in reality one phase. The Indo-European relation t (voiceless): d (voiced) is shifted to th (tense, aspirated): t (lax, non-aspirated). This shift was a consequence of the merger of IE t and d after s, e.g. *esti 'is' = G. ist like *ozdos 'branch' = G. Ast. Neutralization in favour of t ( = IE d) brought about an inversion of the original phonemic opposition with the Germanic phoneme t becoming unmarked and its counterpart th (from IE t) marked (tenseness, aspiration). The subsequent development of th to p is only a phonetic corollary of the fundamental inversion of the Indo-European phonemic relation t: d. In the second place we know neither the phonetic value of IE dh nor its phonetic fate in Proto-Germanic 8 prior to the action of Verner's law. What we do know is the phonemic relation of the continuants of IE dh and IE t (Gmc p) during and after the working of Verner's law. The merger of IE t (Gmc p) with the Gmc continuant of IE dh after unstressed vowel (in favour of dh>d) is responsible for the relation p . d in word-internal position. 9 Thus instead of the above four phases only two mergers are phonemically pertinent. They are to be distinguished from changes like t >p or d>d (in some positions). 1.7 To quote a more recent example, the Great Vowel Shift of Middle also best regarded as the shift of a relation. The Middle English change four items: e>i; i>ai; g>u\ u>au, whereas the fate of ME f, p, and considered separately, not only for chronological reasons.10 The Great Vowel Shift in the narrower sense can be explained as a phonemic relation:

English is comprises a is to be change of

long open ( = non-diffuse): long close (diffuse) > l o n g close: diphthong. The first question to be answered is whether there had been a phonemic relation e : i, o :u in Middle English. Now e and i had a common point of neutralization in ME i, the result of their shortening in closed syllables and in antepenults. Similarly both o and u alternated with u resulting from their shortening under identical conditions. E.g. sick ( d ) after nasal, in gemination, and in word-initial position may be a later phenomenon. 10 Notice especially the lack of parallelism between the development of # (merging with the continuant of f ) and that of g (collapsing with the continuant of the ME diphthong ow as in low). ME a merges with the continuant of the ME diphthong ey as in prey. 9

70

JERZY KURYLOWICZ

diphthongs (ai, au respectively). From the phonemic point of view the qualitative merger of e with i, of o with u, is therefore prior to the diphthongization of ME i, m.11 1.8 Occasionally a phonemic merger, postulated on comparative grounds, is masked in some languages by subsequent changes so as to make us hesitate whether to ascribe the phonemic fact to the reconstructed protolanguage or only to a later dialectal development. This is e.g. the case of the so-called Bartholomae's law, which may be symbolized by the formula IE dh-\-t>ddh (gh+t >gdh; bh+t>bdh) with an unusual progressive assimilation of stops (gdh x 9 in A f o r m like S e k t o , a n a l y s e d * S e x + c t t o > * 8 e % + t o > S e k t o , g e n e r a t e s a n e w

morphonological rule of internal sandhi, viz. assimilation of the aspirate to the following voiceless stop, hence also £kto ks-). And Gr. apKTOQ 'bear' = Skr. rksa- could be explained in the same way, if Hitt. hartagga- means 'bear' (i*h{tko-). But in the majority of examples, like Gr. (pSivto: Skr. ksindti 'destroy', there are no grounds for a parallel explanation. 12

INTERNAL RECONSTRUCTION

71

languages must be inherent in the fundamental model of linguistic communication, with a speaker and a hearer at a certain place and time. 2.2 The orientation of the speaker is based on local and temporal coordinates intersecting in the zero-point hie (ego): nunc, i.e. here where I am, now when / a m speaking. The relation of / to here, both referring to the same place, needs further elucidation. Within the personal pronouns ego, tu, ille, there is a triple opposition. Moreover, the 3rd pers. may be represented by a demonstrative or interrogative pronoun or a noun. Putting aside grammatical gender, it is evident that the opposition ille : ego, tu is a contrast between neutral and pers. (human) since the pronoun of the 3rd pers. may refer both to humans and to non-humans whereas the 1st and the 2nd pers. denote only humans. Restricting the opposition to humans the chief axis of the system of pers. pronouns will be 3rd pers.: 1st pers. since tu may be replaced by the 3rd pers., cf. G. Sie, It. Lei, in earlier N H G er instead of du when addressing dependants, etc. 13 In relation to the 1st pers. the 2nd functions as the so-called negative member of the opposition: ille (neuter)

/

\

tu ego (ego — the only speaking person). (negative) (positive) In this way the 1st pers. contrasts with the 2nd and 3rd as the marked member of the opposition. The relative affinity of the 2nd and 3rd pers. is also evidenced by the fact that the respective demonstrative pronouns are in many languages identical (cf. E. that as against this). But there are also languages with three different sets of demonstrative pronouns and adverbs, cf. Lat. hie (hie), iste (istic), ille (illic), Sp. este, ese, aquel, etc. The fact that as a rule language distinguishes three persons whereas the corresponding localization by means of demonstrative pronouns may be reduced to two members, proves that localization (e.g. hie) is subordinate to the personal index (ego). Person implies localization, not vice versa. Thus E. that may refer to the 2nd or to the 3rd person. The zero-point of the coordinates of speech situation is therefore first and foremost egojnunc, not hic/nunc. 2.3.1 What are the corollaries of this system of grammatical person and of the corresponding localizations? They are reflected in the grammatical classification of the noun (gender etc.), in its number, and its inflection. 13 This is something different from the use of the 3rd pers. instead of the 1st in cases like thy servant instead of I, where the replacement of the pronoun by a noun necessarily entails the 3rd pers. of the corresponding verbal form. But a semantic merger of the noun with the old pronoun of the 1st pers. may well result in a stylistic distribution of noun and pronoun, cf. Persian banda 'slave' > T (stylistically different from man 'I') with banda governing the 1st pers. of the verb.

72

JERZY KURYLOWICZ

The opposition ego, tu : non-personal ille implies the contrast pers. {human): nonpers. (non-human) as the basis for all systems of the classification of the noun, i.e. of nominal gender in the wider sense. The opposition quis : quid (E. who : what) permits the division of all nouns into two groups, human and non-human. Subdivisions on the basis of sex (he : she : it) etc. are subordinate to this fundamental dichotomy. In Persian where the distinction he : she is lacking, the fundamental opposition who : what is represented by ke : ce. The category of grammatical gender in the current sense, deprived of morphological exponents in English, 14 virtually exists in this language because of the pronominal opposition he : she15 directly employed in derivation to denote sex: he-bear, she-bear etc. Subdivisions of the opposition who : what into animate : inanimate, male : female, old: young, big: small, etc. (cf. the noun-classes of the Bantu languages), are the result of formal differentiations due to derivational procedures. But the primary basis must still be looked for in the opposition who (ego, tu, and certain nouns): what (the remaining nouns). This distinction is important. It proves that in spite of the commonly accepted terminology (non-human \ human or non-pers. : pers.) human (pers.) is to be considered as the unmarked or fundamental term of the opposition. The range of human comprises ego, tu, and a part of ille, that of non-human only a part of ille. 2.3.2 The unmarked character of human accounts for the anthropocentric nature of language. To quote an instance it is well-known that affixes serving to form agent nouns are finally also used for nouns denoting instruments (nomina instrumenti), e.g. Lat. -tor in the Romance languages (Fr. -eur etc.), Gr. -xfip (icaSe-crip, acpvyKTrip), -er in E. (e.g. receiv-er of the telephone) and so on. Cf. also the above example of the use of pers. pronouns he, she in forming names of animals (she-bear etc.). In Slavic the genitive in -a served originally also as the accusative of masc. pers. nouns. Afterwards its use has been extended to masc. animate nouns. In Modern Polish it has already begun to encroach on the inflection of masculine inanimate nouns. The spread of derivational and inflectional procedures from human to inanimate via animate may be considered a diachronic universal and an important chronological criterion in internal reconstruction. 2.4.1 The plural forms of the noun can often be traced back to old collectives, i.e. to derivatives or compounds in the singular form. This seems to be a diachronic universal. Cf. the type lis tie 'leaves' ( A — X (replacing A') i A' (derivative)

la) A in its secondary syntactical function may be replaced by B which then becomes an inflectional form of A (grammaticalization of B). 2) On the other hand A', ousted by X, may be lexicalized becoming a simple derivative of A/X.26 Notice, however, that in lb) the grammaticalization of B is closely connected with the lexicalization of A'. " As regards derivation within a part of speech, procedures like the formation of diminutives (Lat. lupus > lupulus) or of feminines (lupus > lupa), being limited to the paradigmatic shift, present a reduced pattern. But this is not always the case. Fr. arc 'bow' -marcher 'bowman' and similar examples are to be analyzed into a 'horizontal' shift bow -+(man) with a bow (i.e. a concrete case-form of bow) plus a 'vertical' shift (man) with a bow ->•bowman (Fr. arc -+lhommel a I'arc -marcher). 25 An inflectional category embraces a greater number of lexical items belonging to a part of speech than does a derivational category. However, to consider grammaticalization — as against lexicalization — simply as a process of extension would be to simplify matters since on the other hand lexicalization means augmentation of the stock of autonomous lexical items.

INTERNAL RECONSTRUCTION

79

Grammaticalization and lexicalization entail a change of part of speech. 2.9.1 A detailed analysis of the Lithuanian and of the Germanic adjective may serve here to illustrate the shifts adjective -> noun and vice versa. This will permit us to limit ourselves to a more succinct treatment of other types of changes. First of all let us explain the penetration, in Baltic, of the pronominal endings into the inflection of the simple adj. This is a morphological feature shared by Baltic and Germanic (not by Slavic). The fundamental identity of the inflections of the types jis' and gèras in Lithuanian, of the types sa and gops in Gothic, etc., is evident. The declension of the simple adjective in Baltic, of the 'strong' adj. in Germanic is thus a serious deviation from the Indo-European model with its morphological identity of noun and adjective. The case-forms of the adjective morphologically assimilated to those of the pronoun are in Lithuanian the following: sg.dat. loc. plur.nom. dat.

geràm geramè gerì28 geriems

like jam like jamè like jiè like jiems

as as as as

against against against against

vyrui vyre vyrai vyrams

In Gothic all three genders of the adjective have undergone modifications of their inflection: masc.sg.dat. acc. plur.nom. gen. dat. neuter sg. fem.sg.gen. plur.gen. dat.

godamma godana godai godaize godaim godala godaizos godaizo godaim

like pamma like pana like pai like pize like paim like pata like pizos like pizo like paim

but but but but but but but but but

daga dag dagos dage dagam watird gibos gibo gibom

The pronoun did not exert its influence in all case-forms where it would have been possible. Thus we find in Gothic the dative singular of the feminine godai as against pizai. In North and West Germanic, however, there is an original identity of desinences between the pronoun and the adjective. Cf. O. Icel. spakrar, spakre like peirar, peire; OE. blindre (gen. and dat.) like pare] O H G . blintera, blinteru like dera, deru. 2.9.2.1 In Baltic the declension of the corresponding compound or 'definite' adjective is moreover characterized by the enclitic forms of the pronoun -*io- attached to the inflected form of the adjective, e.g. Lith. gero-ji. The original autonomy of jis, ji is still attested by Old Lithuanian forms like zaisla pa-jo-prasta 'the usual toy', visi su-jie-spausti 'all the oppressed ones', pra-jis-puolqs 'the perished one', with the 2e

Shortened from *gerie, cf. the corresponding compound form gerie-ji.

80

JERZY KURYLOWICZ

pronoun (jo, jie,jis) inserted between the preverb and the personal verb (E. Fraenkel, Die bait. Sprachen, 82 f.). The corresponding modern forms would be paprastôji, suspdustieji, prapuolçsis. The status of jis as a regular enclitic, always attached to the adjective, seems therefore a relatively recent phenomenon. 2.9.2.2 A form like Lith. *gërasjis (geràsis) could represent either a 'syntactical' or a 'semantic' nominalization2'' of the adjective, if under 'nominalization' we understand an autonomous use of the adjective, not accompanied by the determined substantive. In the case of syntactical nominalization the adjective refers to a word (viz. a noun) already mentioned in the narrative or dialogue. Semantic nominalization takes place if the adjective provided with the demonstrative pronoun refers to an object (of external reality), not to a word of the text as in the case of syntactical nominalization. Cf. e.g. Lith. raudonàs-is (raudonô-ji) 'gold coin' < raudônas 'red', baltô-ji 'white i.e. colourless brandy (vodka)' < bâltas 'white', kruvinô-ji 'dysentery' < krùvinas 'bloody', greitô-ji 'haste' < greïtas 'hasty'. 2.9.2.3 The distinction between syntactical and semantic nominalization is important. It is founded on the double function of the demonstrative pronoun : anaphoric and deictic, respectively. As long as a form like Lat. bonus ille or SI. dobn + jb refers to a noun and not to an external object, the adjective is in agreement with this noun (occurring somewhere in the text) and keeps its inherited inflection, bonus ille and dobn jb being only a secondary syntactical function of the adjective (normally used as an attribute). But if a form like Lith. raudônas + jis becomes a noun, the sign of an external object, the relation between raudônas and jis changes. Henceforth the deictic pronoun represents the designed object and the inflection of the adjective will be in agreement with that of the pronoun. The fact that raudônas determines jis, and not a noun, is responsible for the partial change of the inflectional endings of the adjective. This change was rendered possible by the inherited agreement of several case-forms of the pronoun jis and of the adjective in -o-. Thus jo : gero,jH : gerî{, jaîs : gérais, etc. (case-forms with identical nominal and pronominal endings) = jâm : gerâm, jamè : geramè, jiems : geriems (instead of *gërui, *gerè, *gerâms). 2.9.2.4 The Baltic development is characterized by the tendency to use such 'semantically nominalized' adjectives first as appositions, then as normal attributes, and as syntactically nominalized adjectives. Besides the old group *gerui vyrui (dat. sing.) a new construction comes up: gerâm jam vyrui, compare distinctions like G. der große Karl : Karl der Große, or Fr. le grand Charles : Charles le Grand. 2.9.2.5 At this point, however, the chief problem turns up. Why hasn't the contrast *gerui vyrui (dat.sing.): gerâm jam vyrui been maintained? In actuality the only attested form is not *gërui vyrui but gerâm vyrui. The opposition between géras vyras (dat. *gërui vyrui) and géras jis vyras (dat. gerâm jam vyrui) became, owing to the gradual weakening of the pronoun (tending to adopt the synsemantic value of an article), a contrast between a simple and a compound or 'definite' attribute. The latter, i.e. the form geràsis invaded the original "

This instead of the awkward neologism

substantivalization.

INTERNAL RECONSTRUCTION

81

semantic zone of géras, especially that corresponding to the use of the definite art. of many other languages: 1) the function of determination (geràsis vyras 'the good man' : géras vyras 'a good man'); 2) the function of generalization (pilkàsis genys 'the grey wood-pecker' in the meaning of species, not only of individual). The simple or indefinite form géras became the marked member of the opposition geràsis : géras. Ousted from some of its old functions and undergoing the pressure of the compound form geràsis the paradigm of the attribute géras was proportionately transformed as follows: gèro-jo : gëro, ger%-jt{ : gerq, geraïs-iais : gérais = gerâ(m)-jam : geràm, etc. The final result of this generalization of pronominal endings was the rise of a morphological distinction in the nominal inflection (adjection versus substantive). Cf. dat. sing, gerâm (adj.): gërui (subst.), labâm (adj): lâbui (subst.). The adoption of pronominal endings by the adjective was facilitated by the simplification of declension the adjective had undergone in Baltic. It had been practically reduced to one type, that of the -(j)a-stems. Even in the (productive) -u- adjective the only inherited desinences still extant are -us (nom.sg.), -aus (gen.sg.), -q (acc.sg.) and -iis (nom.plur.) of the masculine gender. In Lithuanian where the enclitic status of jis is due to a relatively recent development, the inflection both of the enclitic and of the preceding adjective are well preserved. There are some sporadic simplifications which can be accounted for on purely phonetic grounds, thus the dissimilatory suppression of m in : sing.dat. -àm-jam > -âjam, loc. -amè-jame > -ajame, plur.dat. -iems-jiems > -iesiems, and correspondingly in the feminine. 2.9.3 The hypothesis proposed here to explain the Baltic simple and compound adjective may be tested by the analogous evolution in Germanic. The historical stage Got. sagopa, dat. pamma godin, versus gops, dat. godamma (with pronominal ending) continues an older contrast *sa gops, dat. *pamma godamma opposed to gops, dat. godamma, itself a successor of dat. *goda (like daga). In the initial stage the Germanic development was parallel to that of Baltic, putting aside the difference between the pronominal elements, proclitic sa as against enclitic jis; Got. godamma : *pamma godamma like gerâm : gerâ(m)-jam with the generalization of pronominal inflection both in the compound and in the simple form (if we call *pamma godamma a compound form). Now in Gmc *sa gops was replaced, in the case of semantic nominalization, by a derivative form. Just as e.g. in Greek (cf. oôpdvioç : où pa via» v), secondary substantives were formed from adjectives by means of the nasal suffix. Got. pamma godin ousted both the attribute *pamma godamma and the syntactically nominalized adjective, just as in the previous stage *pamma goda had been ousted by *pamma godamma in both these functions. In Germanic, again as in Baltic, the influence of the pronoun upon the inflection of the 'strong' adjective has been facilitated by the uniformity of the adjective declension : apart from some forms of the nominative singular of -/- and -u- stems (Got. hrains, masc.-fem. hardus, neuter hardu) all the other attested case-forms go back to -(¡)o/âstems.

82

JERZY KURYLOWICZ

2.9.4.1 To sum up it seems that the pronominal inflection of the 'strong' or the simple adjective (in Germanic and Baltic, respectively) is finally to be ascribed to forms provided with a pronominal element (or 'article'), originally functioning as semantically nominalized adjectives, afterwards as appositions, as attributes, and as syntactically nominalized adjectives. It is only in conjunction with the pronoun that the adjective could have been governed by the pronoun (as borne out by agreement), and thus adopt its special inflectional endings. Having become an attribute it imposed its pronominal endings upon the 'strong' or simple adj. Lithuanian forms like dat. gSrui, nom.plur. gèrai represent old adjectives semantically nominalized before the spread of the pronominal desinences. 2.9.4.2 From the standpoint of general linguistics the syntactical circulation between the noun and the adjective is rather instructive. Cf. for Lithuanian:

1. 2. 3. 4. 5. 6.

I syntactical function attribute (*gerui) syntactical nominalization (*gSrui) semantic nominalization (geram-jam) apposition (geram-jam) 'definite' attribute (gerdjam) 'indefinite' attribute (geram)

part of speech adj. adj. noun noun adj. adj.

Procedures of word-formation are as a rule motivated by syntactical functions ('horizontal' shifts). The above scheme accounts for substantives derived from adjectives (stage 2-3, syntactical -* semantic nominalization) and for adjectives derived from substantives (stage 4-5, apposition->'definite' attribute, i.e. syntactical semantic 'adjectivalization'). 2.10.1 Besides syntactical nominalization the adjectives, whose chief or primary function is to act as attribute, may undergo another major change, viz. syntactical and semantic 'verbalization'. The difference between the adjective used as predicate noun (whether accompanied or not by the copula) and the attribute is greater than that between attribute and its syntactical nominalization. This is frequently borne out by morphological facts, cf. e.g. the contrast between the short form of the Russian adjective in the predicative function versus the long form functioning both as attribute and as syntactically nominalized adjective. Or compare the lack of distinction of gender and number in the German predicative form gut as against its full inflection elsewhere, both in the attributive and in the nominalized function. The history of the Slavic adjective proves that in the predicative function preservation of the old petrified forms of the adjective may be expected, forms completely ousted from the paradigm of the attributive adjective. 2.10.2.1 Another important syntactical function of the adjective, midway between

83

INTERNAL RECONSTRUCTION

attribution and predication, is the so-called co-predicative function. Constructions with copredicate, which at first glance may sometimes look like attributive constructions, presuppose an originally predicative articulation of a nominal group. Most frequently there are, however, formal criteria differentiating the adjectival copredicate from a normal attribute, e.g. word-order. Two kinds of copredicative function must be distinguished, one where the adjective refers to the subject of the sentence, the other where it refers to a complement. In case of inflectional difference one may a priori expect the following distribution : the copredicate referring to the subject of thé sentence (i.e. only to a nominative) will have the same form as a predicative adjective. On the other hand, the copredicate determining a complement, variable as regards case (accusative, dative, etc.) will be more apt to undergo the influence of the inflected attributive form. 2.10.2.2 A category whose primary syntactical function is to act as copredicate is the participle. In its secondary function as attribute it behaves like an adjective. In case of a differentiation between the copredicative and the attributive function the attributive form becomes an adjective (verbal adj.), whereas the copredicative form continues to belong to the conjugational system of the verb. Cf. Sp. escrito, frito (adj.) as against escribido, freido (part.) or the split between some English participles and the corresponding verbal adjective (drunken adj., drunk part., learned dissyllabic adj. versus learned monosyllabic part.), also Russian participle in -ënnyj (pecënnyj) but corresponding verbal adjective in -ënyj (pecënyj). Within the copredicative function of the participle the previously mentioned split may occur, viz. the differentiation between the participle determining the subject and the participle determining a complement. The latter must be inflected whereas the former, necessarily insensible to case-variation (referring exclusively to the subject), easily loses the distinctions of gender and number as well and becomes a gerund. 2.10.3 The predicative use of an adjective, e.g. Russ. on bel 'he is white', is a secondary syntactical function (attributive function being primary). We may call it syntactical verbalization of the adjective. Semantic verbalization would be the derivation of the verb from the adjective, e.g. Russ. belyj 'white' > beletb. Its participle (belejuscij) functions primarily as copredicate; secondarily it may be used as an attribute (syntactical 'adjectivalization' of the participle). The corresponding semantic stage is implemented by a formal split like that between Russ. pecënnyj and pecënyj or Sp. freído and frito. But on the other hand the source of the renewal of the participle is always the verbal adjective, cf. e.g. the Indo-European verbal adjective in -to- or -no- which gradually became a member of the conjugational system in many historical languages, though not in Sanskrit or Greek. The interrelation between the adj. and the verb may be represented by this diagram: II syntactical function 1. attribute

part of speech adj.

84

JERZY KURYLOWICZ

2. 3. 4. 5. 6.

predicate (syntactical verbalization) predicate (semantic verbalization) copredicate attribute (syntactical adjectivalization) attribute (semantic adjectivalization)

adj. ( + copula) verb (personal) verb (part.) verb (part.) adj.

This scheme accounts for verbs derived from adjectives (stage 2-3, syntactical-» semantic verbalization) and for adjectives derived from verbs (stage 5-6, syntactical -*• semantic adjectivalization). 2.11.1 The above schemes (I, II) show on the one hand the circulation between the adjective and the noun, on the other hand that between the adjective and the verb. Syntactical function and part of speech are interdependent, but only indirectly. When speaking of syntactical nominalization or verbalization of the adjective we mean its syntactical function, the part of speech remaining the same (adjective), though sometimes provided with a morphological exponent, the change being inflectional, not derivational. The inflection may be 'synthetic', e.g. Russ. belyj, syntactical verbalization bel, bela, belo, belyi, or 'analytic', e.g. E. white, syntactical verbalization is (are) white\ syntactical nominalization the white one. When speaking of semantic nominalization or semantic verbalization of the adj. we mean a derivational procedure, e.g. the red, the reds (Indians) or to redden, Russ. krasnetb. It is only by making a sharp distinction between the syntactical and the derivational aspect that the relations between the members of the sentence and the parts of speech become clear. 2.11.2 Diagrams I and II permit the establishing of the hierarchy and the chronology of the syntactical and semantic functions of adjective/noun and adjective/verb. It must be pointed out that some of the contiguous stages are synchronous. In I the stages 1 and 2 represent the inherited syntactical functions of the adjective and the sequence 1-2 denotes functional, not chronological, hierarchy. So does the sequence 3-4. Again, in II the stages 1-2 and 3-4-5 are chronologically 'blocks', since any adjective could be used in Indo-European as predicate and every verb had a participle which could also function as attribute. To put it shortly, inflectional forms of the word corresponding to different syntactical functions, although hierarchically ordered (primary and secondary syntactical functions), are to be considered as synchronous. Relative chronology must be established within a syntactical slot, e.g. I, 2-3, 4-5; II, 2-3, 5-6. 2.12.1 A given part of speech is characterized by the hierarchy of syntactical functions q)j (primary) -> cp2cp3 ... (secondary). Extension of the range of f 2 (form of cp2) exceeding that of f t may entail an inversion of hierarchy, with original cp2 (or ip3 etc.) becoming (pt. This phenomenon is called grammaticalization of the form f 2 (corresponding first to the secondary then to the primary function).

INTERNAL RECONSTRUCTION

85

The Lithuanian type gerás-is became an adjective owing to the change of -is from a derivational to an inflectional suffix (grammaticalization). As inflectional suffix it is proper to all adjectives, i.e. to a part of speech, as derivational suffix only to some adjectives (as raudonásis, cf. 2.9.2.2). The fate of the Latin suffix -tor (forming agent nouns) in Spanish has been similar. The Spanish derivatives in -or, fem. -ora are in the first place adjectives (cazador, -a, encantador, -a), although like any other adjective they may be used autonomously) (syntactical nominalization), and in certain cases the suffix -or serves to derive nouns (semantic nominalization, e.g. matador). The IndoEuropean verbal adjectives function in Slavic as personal forms of the verb (perfect), cf. 2.6.2.1. A secondary (viz. predicative) function of these adjectives became primary since in the predicative function the form in -h was incorporated with the system of verbal inflection by spreading to all verbs whereas in the attributive function it continued to be represented only by a certain number of derivatives. 2.12.2 There is a second type of grammaticalization, already mentioned in 2.8: a derivative of a word A may be used in order to express a secondary syntactical function of A or the replace an inflectional form of A, thus becoming itself an inflectional form of A. Cf. the replacement of the secondary function of the present (viz. future) by a desiderative, etc., in English {he will, I shall write). The replacement of 'synthetic' case-forms by prepositional expressions giving origin to 'analytical' cases in English or in Romance. The replacement of the plural form by a derivative with collective value, henceforth belonging to the declensional paradigm. The compound tenses ousting partly or totally the old simple tenses in many European languages. 2.12.3 The other possibility, viz. lexicalization of inflectional morphemes, is as a rule a consequence of the renewal of an inflectional form, with the old form becoming a simple derivative.28 The natural successor of an inflectional form is the part of speech whose primary syntactical function is identical with the syntactical function of the inflected form. Thus as regards the attributive function, both the noun (as apposition) and the verb (as participle) may be lowered to the status of adjective (cf. 1,4 > 5; II, 5 > 6), whose primary syntactical function is to act as attribute. The desinence -a of the Latin neuter plural was replaced in Italian by -i, on the model of the masculine. But the old forms are in a number of cases preserved as derivatives, viz. as collectives distinct from the plural in -i, cf. budello 'intestine': budella 'bowels'; muro 'wall': mura '(city-)walls'; osso 'bone': ossa 'osseous frame, etc.', or with a different meaning (complete lexicalization), e.g. membre 'members of the body' versus membri 'members of a society, etc.', coma 'horns (of an animal)': corni 'horns (musical instruments)'. 2.12.4 The scheme lb) of 2.8 is complicated. It involves both grammaticalization " The term 'derivation' must be taken here cum grano salís. 'Derivation' may be productive or nonproductive, cf. the Russian type pecenyj (2.10.2.2), but also result in a set of completely lexicalized items. E.g. the Spanish replacement of the inherited Latin past participle of strong verbs by the 'regular' formation in -ido engendered a series of verbal adjectives of varied structure: beodo < beber (pp. bebido), estrecho < estreñir (pp. estreñido), quisto < querer (pp. querido), teso < tender (pp. tendido), tonto < tundir (pp. tundido), tuerto < torcer (pp. torcido).

86

JERZY KURYLOWICZ

and lexicalization. The replacement of the inflectional form A' by the derivative B leads to an inversion of the original hierarchy. Instead of A' -*B we get B (new inflectional form)->/4' (derivative). Now it is evident that the replacement in question is a semantic merger which does not necessarily involve 1) all specimens of A'; 2) a simultaneous replacement of the morpheme of A' by the morpheme of B (that is to say the functional merger does not nessarily imply a structural one). In Arabic the inherited Semitic plural forms (-una for masc., -at1"1 for fem.) were ousted by new forms (originally collectives), but some nouns or rather groups of nouns have kept the old plural and certain others make a semantic difference between the old and the new form of the plur. The hierarchy of these types is to be established on the basis of their respective ranges. The form of the old plur. is marked since the range of its occurrence can be positively established by enumerating the groups of nouns and the individual nouns still forming the ancient plural. Nouns forming the new plural can be defined only negatively as 'all the rest'. Therefore the endings -una, -atm represent marked allomorphs as against the various devices of vowel-gradation and affixation characterizing the new plural. Hence in nouns with a double plural the old form, semantically different from the new one, must be considered as marked, as a kind of derivative. E.g. katibun 'writer': plur. kuttab, but katib"" 'writing': plur. katibuna.29 2.13.1 One of the pitfalls to be avoided in reconstruction is the confusion of the functional and structural levels when establishing the hierarchy of the members of a grammatical category. It is e.g. clear that in French, just as in many other languages, the masculine form of the adjective is unmarked (neuter) in opposition to the marked feminine. Neutralization of the distinction masculine : feminine takes place in favour of the masculine, e.g. les noirs in the meaning of 'black men and women'. But the structural hierarchy is different. If there is a formal distinction between the genders, it is the masculine which is as a rule derived from the feminine by dropping the final (pronounced) consonant: blanche > blanc, fraiche >frais, lourde > lourd, basse >bas, and so on. The hierarchy to be established is therefore the following: blanc functional 1 blanche structural I blanc The relation imperfective : perfective in Slavic may serve as another example of such a discrepancy. Apparently it is the imperfective which is derived from the perfective 29

Only some typical semantic shifts can be mentioned here. The difficulties of reconstruction increase in proportion to the degree of lexicalization, parallels from other languages being scarce or lacking in the case of special semantic developments.

87

INTERNAL RECONSTRUCTION

since the productive grammatical procedure 30 consists in attaching the suffix -aje- to the root of the perfective ( + lengthening of the root vowel), e.g. sb-metetb > sbmetajetb. Another (though non-productive) grammatical procedure is the use of the suffix -i( + o-ablaut of the root vowel), thus e.g. do-vedetb : dovoditb. This fact is apt to lead the linguist astray, making him consider the imperfective sbmgtajetb, dovoditb as the marked members of the morphological opposition. They are marked from the structural, but not from the functional point of view. What is decisive is the over-all functional distribution, the relation of ranges : the semantic opposition imperfective : perfective existing in the past tense is neutralized in the present in favour of the imperfective ( = the unmarked or fundamental member). 2.13.2 Undoubtedly such a discrepancy between structure and function is responsible for a tension the language tries to reduce by adapting the structural to the functional hierarchy. It is the inversion of the structural relation which contributed to the rise and spread of a special perfective suffix -ne/o-.31 It offers the possibility of deriving perfective forms from simple imperfective verbs, e.g. kapljetb 'drips': ka(p)netb. 2.14.1 Still, in the majority of cases, both in derivation and in inflection, structures and functions are from the point of view of hierarchy parallel. This together with the fact that in languages (Indo-European among others) having recourse chiefly to suffixation the sequence of morphs reflects relative chronology and ensures a certain amount of probability in linguistic reconstruction. A relation like A : A + m + n (with m, n recognized as suffixes) is to be traced back either to: basic form A->\. derivative A + m ->2. derivative A-\-m-\-n — or else to A hence A :A+n A-\-m

= A-\-m:A

+ m+n

A+ n

2.14.2 Let us consider the problem of the Indo-European verbal system. Greek points to an old opposition of aspect, e.g. e^eiTte (imperfective): eXirre (perfective), neutralization in the present tense (XeiTtco). In Vedic, however, it is the relation present (simultaneity or non-anteriority): aorist (anteriority) which is fundamental, just as in English {I am writing : I have written), whereas the Indo-European imperfect becomes in Vedic a simple preterite (narrative tense), cf. E. I wrote. Consequently, the difference between these two archaic verbal systems is the following: Greek present Xeircei imperfect sA.ei.jre (perfecting) j. aorist M y i t i s

Vedic present rinakti aorist araik J. (anteriority) imperfect arinak

A direct structural opposition exists between the Greek present and the Greek imperAs against lexical or rather semigrammatical procedure, viz. the use of various prefixes in individual cases like piig (imperf.): napiSg (perf.). 31 Cf. The inflectional categories of Indo-European, p. lOOf.

88

JERZY KURYLOWICZ

feet (simple difference of endings) and between Greek imperfect and Greek aorist (simple difference of stem). In Vedic the poles of the chief axis rinäkti: draik differ both by endings and stem. Therefore it is advisable to consider the Greek system as older and to ascribe the Vedic innovation to a change of the category of aspect into the category of anteriority. 32 This shift corresponds to the pattern A -* A + m -*• A + m + n becoming A -* A + m + n. A 2.14.3 The pattern explains e.g. the Modern German suffix -lein, A + m A+n a conglutination of -(i)l + in-, both serving originally to form diminutives. Such a cumulation of semantically similar or nearly identical morphemes is not uncommon in diminutives, cf. Russ. kus 'bit' > kus-ok > kus-oc-ek. The same cumulation occurs in another German suffix, -chen (Bäum-lein and Bäum-chen). In both cases the old diminutive suffix, -(i)l- or -k-, has undergone a weakening in certain contexts, maybe after roots denoting small objects or individuals, by a kind of 'semantic dissimilation'. Hence the reinforcement by a more recent and productive suffix -in- appearing in diminutives like O H G zick-in 'kid (young of goat)' < ziga 'goat'. But the enlarged suffixes -(i)l-in-, -k-in- themselves proved dangerous rivals of the productive -in- on account of the redundant -(i)l-, -k-, respectively. The upshot was the total ousting not only of the suffixes -(f)/- and -k-, but also of -in-, in favour of the compound suffixes -{i)lin-, -kin-. Cf. Mod. G. Zicklein versus O H G zickin etc. 3.1 Two conclusions may be drawn from this analysis: 1) redundant features show a tendency to spread in founded (derived) forms, 33 since the morphonological swelling of the derivative (the semantic function remaining the same) renders it more expressive', 2) in morphological analysis special attention must be paid to the functional load of the morphs which in certain cases may be a semantic zero, viz. when the morph is semantically redundant. 3.2. Morphs with semantic zero-value ( = morphonemes) 34 play an enormous role in the history of language. Now a morph has zero-value semantically if within a derivational or inflectional series it accompanies the morpheme only in some of the items. Cf. G. Blatt: plur. Blätter versus Brett: Bretter. The plural is formed by means of -er, umlaut being predictable and proper only to a certain type of roots, viz. those containing a back vowel (a, o, u, au). Within the transformational chain Blatt > * Blatter > Blätter the last stage (* Blatter > Blätter) is from the morphological point of 32

For details cf. The inflectional categories of Indo-European, p. 122ff. As a matter of fact redundant features are either generalized within the respective morphological categories or else totally discarded. 34 For this definition of morphoneme see Phonologie des Gegenwart (Phonologie-Tagung in Wien, 1966), (Graz, Wien, and Köln, 1967), p. 168, and Directions for historical linguistics, a symposium, ed. by W.P. Lehmann and Yakov Malkiel (Austin and London, 1968), p. 80. 88

INTERNAL RECONSTRUCTION

89

view redundant, cf. Brett > Bretter. A proportion like Brett: Bretter = Blatt: Blätter is morphologically, though not phonemically, correct. Hence the fundamental tenet: just as in phonological proportions we take no notice of allophones, even so in morphological proportions redundant features are not considered. 3.3 In order to establish the morphological zero-value of a morph we must have recourse to functional hierarchy. Thus e.g. the system of the present tense is by definition the fundamental (unmarked) form of Indo-European conjugation, the marked member being the aorist expressing perfectivity in Greek, anteriority in Vedic (cf. 2.14.2). Both the present and the aorist were usually characterized by affixes. Let us represent them by A (root) + m, and A (root) + n respectively. Since, however, m as against n was the sign of non-perfectivity (or non-anteriority) versus perfectivity (anteriority), the relation m : n was morphologically equal to zero : n. Instead of the grammatical rule that the aorist (and other tenses) were originally formed directly from the root, we may substitute the formula: the characteristic of the present is redundant in relation to other tenses.35 Thus the proportion Ved. ririakti 'he leaves': araik 'he has left', where the present has a nasal infix, and the aorist the -i-suffix (araik ai), must be reduced to rik : araik, exactly parallel to pres. chant-ti 'seems': aor. achän > achänd-s-t, where the present form lacks a special characteristic. 3.4.1 The importance of distinguishing morphemes, i.e. morphs with semantic or syntactical function, from morphonemes, i.e. semantically empty morphs, is manifest in internal reconstruction. To cite an example, Indo-European knew an opposition between active and mediopassive endings of the personal verb. In Greek and in Indo-Iranian a proportional relation is attested between the endings -t and -to (-ta) in the 3rd pers.sg., and -nt and -nto (-nta) in the 3rd pers. pi. The difference, viz. o, is identical with the simple o occurring in the 3rd pers.sg. mediopassive of a few archaic forms in Vedic as well as in Hittite (-a besides -ta). The diffusion of this o of the 3rd pers.sg. as the characteristic of the mediopassive is to be accounted for by two proportions, one of them morphological (Junctional), the other morphonological. 1) Morphological proportion: -t (3rd pers.sg.act.): -o (3rd pers.sg.mediopass.) = -nt (3rd pers.pl.act.): -nto (3rd pers.pl.mediopass.) The -t of the 3rd pers.sg.act. being semantically empty (zero person, zero number, zero voice) the above proportion must be reduced to zero :o = -nt : -nto. " Its redundancy is therefore relative. Among the different formations of the Indo-European present some had positive morphological exponents, denoting various modes of action, in comparison with other presents. Similarly a plural ending like Lat. -a in verb-a is positive versus the redundancy of -um in verb-um (zero number) but redundant (zero case) as against the ending -orum in verb-orum. The redundant or positive character of a morph may therefore depend on the morphological proportion, i.e. the relation it enters into.

90

JERZY KURYLOW1CZ

2) Morphonological proportion: -nt (3rd pers.pl.act.): -t (3rd pers.sg.act.) = -nto (3rd pers.pl.mediopass.): -to (3rd pers.sg.mediopass.) explaining the spread of a redundant trait, the -t of the 3rd pers.sg.act. A further spread of o, in the 2nd pers.sg., takes place independently, in Iran, (-anha), Gr. -e(cj)o > -ou, and Got. (-aza). 3.4.2.1 Indo-European apophony (ablaut) or vowel gradation is a classical example of morphs deprived of semantic function. Whatever the vocalic change accompanying suffixation, it takes place only in a part of the forms provided with the given affix. Thus the change called nil grade (ei: i, eu :u,er\ r, etc.) is not possible if the basic form already contains i, u, f . The qualitative ablaut e: o is not applicable to roots with o-vocalism, nor the long grade e : e, o : d, etc., to roots with a long vowel. Hence in Indo-European — as opposed to Semitic — ablaut is only subordinate to suffixation and is therefore as a rule morphologically redundant, i.e. has zero value. One of the productive procedures to form adjectives, patronymics, etc., from nouns was in Sanskrit the suffixation of -a- entailing the lengthening (vfddhi) of the vowel of the initial syllable, e.g. gairikfit-a- 'descendant of giri-ksit- (proper name)'. But if the basic form already contained a long vowel or long diphthong in the first syllable, no lengthening could take place, hence e.g. cayamana- (family name) < cdyamana(proper name), oxytone accentuation being the only differentiating feature of the derivative. In spite of the relative rarity of such cases the morphological redundancy of the lengthening is incontestable. 3.4.2.2 Apophony must not be confused with (phonemic) alternation. The merger of word-final -d with -t in Polish, Russian, or German, i.e. the phonemic replacement o f / d / by jt/, is an alternation independent of morphological conditions. The replacement of e (from *ai i like oi > i; er >r like or >f and so on). 36 " Cf. the author's Indogermanische Grammatik 11.257 (Heidelberg: Carl Winter, 1968). The lengthened grade (e:e etc.) originated in a morphological inversion of the relation normal grade:nil grade, cf. i:ei — u:eu = e: (e + e=) e, ibid., p. 298 if.

INTERNAL RECONSTRUCTION

91

In Semitic the apophony both of ü: & and ofl:ä was originally rooted in respective alternations based on the merger of ü I with ä in the neighborhood of laryngeals (Ph 'h); the apophony ü : i goes back to an alternation characterized by the merger of these timbres before i. Finally the ablaut short: long, e.g. Ar. qatala 'he killed': qätala 'he fought', is still attested in Classical Arabic as a phonetic alternation based on the shortening of long vowels in closed syllables, e.g. iaqülüna 'they speak' masc. versus iaqülna fem. 3.4.2.4 A phonemic alternation becomes an ablaut because of the morphological opposition between a basic word and its derivative or its inflectional form. Let us assume that the base contains the phoneme phx and the derived form (enlarged by an affix) shows in the same place the 'archiphoneme' PH, viz. the result of the merger of phl withph 2 - Phonemically PH may be equal to the unmarked member pht. Then, by opposition to phr, PH will be appreciated as ph2 and will tend to spread within the derived forms outside the point of neutralization. The Indo-European perfect distinguished (just as the present) between strong forms accented on the root (1st, 2nd, 3rd pers.sg. of the active) and weak forms stressed on the desinence (all the rest). In the present the nil grade of the weak forms (/, u, r, etc.) was interpreted as a reduction of ei, eu, er, etc.(appearing in the corresponding strong forms). Now the nil grade was a point of neutralization of e : o. Therefore the zero of the weak forms of the perfect was interpreted, by opposition to the weak forms of the basic present, as a reduction of o. Hence the introduction of full o (instead of e) into the strong forms of the perfect, cf. Gr. XeiJt© : Xs^oirca, Got. beita : bait, and so forth. 3.5.1 A final remark must be devoted to the frequent historical change of morphemes into morphonemes and vice versa. The former eventuality may be illustrated by some examples of compound suffixes, cf. above 2.14.3 on the loss of semantic content of the diminutive suffixes -(/)/-, -k- in German. Here belong also the so-called unionconsonants and union-vowels. The Indo-European primary suffix -t- forming deverbative derivatives became a union-consonant, hence predictable, e.g. in Indo-Iranian where it was automatically attached to verbal root-nouns in i, u, or sonorant, thus Skr. rue-< roc-ate 'shines', but mi-t- y.

100

DAVID SANKOFF

N o t only is Brainerd's work pertinent to the standard theory which deals with cognation relationships between words having the same meaning, but it also applies in a m o r e general context. In Brainerd's theory any type of word feature may be used as a criterion of word relationship, for example whether or not b o t h have Latinate stems. In this case y becomes, more or less, the p r o p o r t i o n of Latinate stems in the general vocabulary. Some attempt is being made to apply this type of theory, but n o results have appeared to date. DRIFT AND CONTACT A problem area crucial to glottochronology is that dealing with the non-independence of evolution of related languages. Where a speech c o m m u n i t y splits in two so t h a t two daughter languages evolve independently, the mathematical considerations (at least in the models discussed so far) are the same as in the comparison of two stages in the same language (except that K must be replaced by 2A or t must be replaced by 21). F o r models based on gradual change processes (to be discussed later), however, there will be an initial period where an effect similar to inertia results in a degree of parallelism between the changes in one language and those in the other. In case this latter effect (drift) should prove to be empirically significant, it is k n o w n how to incorporate it into the theory (see Gleason 1960). F o r example, (iii) and (iv) would read

the expected value of ^

is e

2Mt

K)

, and the m a x i m u m likeli-

h o o d estimator of t is M 2 A

+ K,

where K is an empirically determined constant, or the m e a n of an empirically determined r a n d o m variable.

These formulae would be valid only for time intervals lengthy compared to K. W h e n there is continuous contact between two evolving languages, the mathematics become considerably m o r e complicated. A n u m b e r of ad hoc suggestions for correcting f o r m u l a (iv) were reviewed by H y m e s (1960), but none of these has proved of mathematical interest. Within the context of the deterministic model it is possible to add to the replacement rate parameter kr and time parameter t another, borrowing jl rate, parameter Ah, and letting A = Ar+ Ab a n d 8 = - ^ - t o show t h a t (iii) and (iv) A become (see D . Sankoff 1969)

MATHEMATICAL DEVELOPMENTS IN LEXICOSTATISTIC THEORY

101

the expected value of — after time t is N e~2Xt + 0[1 — e an estimate for t is IM

Ml

-) el

2k

These formulae are identical in form with Brainerd's results for the stochastic model with recurrent and chance cognation. Indeed, with the interpretation of 9 as the probability that given a replacement event in either language, the replacing word will be borrowed from the other language, Brainerd's mathematics carry over to a stochastic model of two daughter languages which borrow from each other.

REFINEMENT OF THE REPLACEMENT PROCESS

The Swadesh, Dyen, and Brainerd models are all based on the possibility of eliciting exactly one word per test list meaning and on the assumption that the basic change process is a sudden replacement of one word for a meaning by another at an instant in time. Such oversimplifications of reality are justified in model-building, by the desire to construct a mathematically tractable model and by the intuition or hope that the model captures the mathematically significant aspects of the real phenomena. There is always the slight possibility, however, that this optimism is misplaced; this possibility, compounded with the inhospitability of linguists to this type of scientific procedure, has engendered some of the heaviest criticism of lexicostatistic theory. Motivated by these considerations D. Sankoff (1972b, 1971) studied the mathematical consequences of dropping the requirements that each meaning be expressible by exactly one word and that word replacement be an instantaneous process. In the more general model used in those studies, a test-list meaning may be represented by a number of partial synonyms. As time progresses, there is a constant probability per unit time that a new word may become usable (at a low frequency) for the meaning. The frequencies with which words for a meaning are actually used are assumed to vary gradually (by means of a one-dimensional random walk12), and occasionally a word becomes permanently obsolete (by dropping to zero frequency). In such a model there arises the problem of how to compare a set of synonyms at one time (or in one language) with a partially different set at (in) another. This may be resolved by " During each small interval of time the word usage frequency has equal probabilities of increasing or decreasing by a small increment.

102

DAVID SANKOFF

always choosing the most frequently used synonym or, if less variability is desired, by computing a metric distance between the frequency distributions on the two sets. Some surprising results pertaining to this model have been proved, some rigorously, and others in terms of large numbers of computer simulation experiments. 13 One such result is that, on the average, the policy of choosing the most frequent word gives the same expectation formula of type (iii) as does computing the metric (which takes into account all synonyms), although the latter method is less variable. More important, these formulae are similar to those in the corresponding Swadesh, Dyen, or Brainerd model and they are not very dependent of the precise nature assumed for the random walk — the only important parameter being the rate of introduction of new words. These results provide partial justification of some of the simplifications inherent in the earlier lexicostatistic models. In particular, the validity of a lexicostatistic model can no longer be disputed on the grounds that it unrealistically assumes exactly one word can be elicited for each meaning; nor on the grounds that it assumes an instantaneous word replacement process instead of a more realistic gradual replacement process. This rather strong statement stems from the fact that each model in a rather wider class (where these unrealistic assumptions are not made) behaves, in expectation, much as if it were the corresponding simple model (i.e. Swadesh, Dyen, or Brainerd model).

A FULLY PARAMETRIZED

LEXICOSTATISTICS

We can now summarize the progress in investigating the mathematical aspects of lexicostatistic change models. The formalization of Swadesh's original theory required a number of oversimplified and somewhat unrealistic assumptions. Dyen and others generalized the theory by postulating a distribution of mean replacement rates instead of a single rate, the same for each meaning. This approach has been empirically corroborated. Brainerd generalized Swadesh's theory by removing the assumption that word replacement always involves non-cognates, and using instead a more plausible model where cognate (or apparently cognate) words could sometimes replace each other. He also investigated the mathematical properties of models where mean replacement rates are different in different languages. Gleason showed how to incorporate into models the phenomenon of parallel drift of two closely related languages, and Brainerd's work proves applicable to a model of two evolving languages which remain in contact. Sankoff studied stochastic models of word replace13 A more complicated model was simulated by Sankoff (1969), where each meaning was not considered individually, but as part of a system of contrasting and overlapping meanings. The relationships between the different meanings were measured in terms of shared lexical items by the same metric used for diachronic comparisons mentioned in the text. This model was similar to the Swadesh model with respect to simulated diachronic behavior and it also resembled natural languages in that the over-all word frequencies inevitably took o n a Zipf's law configuration n o matter what initial configuration was used as input.

MATHEMATICAL DEVELOPMENTS IN LEXICOSTATISTIC THEORY

103

ment more natural than the instantaneous change model of Swadesh, and in which it is possible for a meaning to be represented by two or more lexical items instead of just one. These models turned out to behave remarkably similarly to the simpler model. In the preceding sections, we introduced the following parameters : A - the Swadesh mean replacement rate, a,P - parameters of the T-distribution of replacement rates, y - probability of chance recurrent cognation, K - Gleason's drift time constant, 6 - borrowing probability. Is it possible to construct a stochastic model incorporating all the the historical linguistic phenomena represented by these parameters? It is, and by straightforward extensions of the methods already mentioned. This model results in the lexicostatistic and glottochronologic properties (for times greater than K) : M the expected value of — is [l + y f l - y - 0 ] [l + 2 j 3 ( t - K ) y * + y +

9-yG,

and the maximum likelihood estimator of t is / l - y - f l + yfl \ 'fr

(£-,-.+*) " Tf

+ K

That these results are a logical extension of our earlier formulations can be seen by noting that all the formulae displayed in boxes throughout this paper are either special cases of these (where one or more of the parameters are zero) or limiting cases (where a oo, p0, and fia = X). As t gets very large, the lexicostatistic relationship approaches an equilibrium value y + O—yQ, dependent only on borrowing and chance cognation. Thus the genetic aspects of linguistic relationship gradually become less important than the diffusion aspects. While the 'second generation' of lexicostatistic theory is almost complete, appropriate methodology and improved and systematized empirical results are lagging somewhat. As reported, there are preliminary results about the parameters a and fi. About y quite a bit is known (Swadesh I960; 14 Bender 1969; Brainerd ms.), but it is extremely sensitive to lexicostatistic methodology, i.e. to choice of criteria of similarity. There is no empirical information about K. The borrowing parameter 6 will "

Reviewed by Hymes (1964).

104

DAVID SANKOFF

vary from situation to situation; it is, however, amenable to investigation (G. Sankoff 1968). There is a great need for systematic determination of replacement rates, and secondarily of other parameters where pertinent; it is also crucial to the development of lexicostatistics to know how these rates and parameters may vary over time and/or from language family to language family. That pidginization shows a very rapid rate of lexical change and that a language of literary tradition may show a rather slow rate, may limit the range of validity of particular models, but provide all the more reason for developing a model able to handle these aspects of linguistic evolution. See Dobson (1972) and Kruskal (ms.) for work in this direction. Another topic on which systematic data is lacking is the universality and stability of the test-list (see Hymes 1960 for a discussion of this point). It seems likely that a lexicostatistic theory could be built without a test-list. D. Sankoff (1970) calculated the replacement rates of over a thousand meanings in the Indo-European family. This study confirmed the relative stability of the Swadesh list, but also indicated that it is perfectly feasible, within a language family, to work with any suitably large set of meanings. Criticisms of lexicostatistics based on the untranslatability of 5 or 10% of the test-list into the languages of some family are not statistically well-founded. The assumptions which must be tested are of the existence within a family of some basic set of a fixed number of meanings with lexical representations in almost all the languages of the family, and of the universality of the replacement rate distribution for such sets. RECONSTRUCTION O F GENETIC TREES

Turning now to the other major problem area in lexicostatistics and glottochronology, it is important to note that the inference of genetic history of a language family has close parallels in other sciences,15 expecially chemical paleogenetics, the biochemical study of evolution (Dayhoff 1969; Nolan and Margoliash 1968), and in the investigation of manuscript linkage (Haigh 1970). More general references include Sokal and Sneath (1963), Jardine and Sibson (1971), and Hodson, Kendall, and Tautu (1971). The main problem is as follows. Given that a family of n modern languages all share a common ancestor, and given the cognation data between all pairs of languages for each of the N test-list meanings, how can the sequence of language splits which led to the present n different languages, be reconstructed? The basic assumptions here are that the genetic relationships among the languages can be summarized in terms of a 'family tree', a diagram such as Figure 1 consisting only of points and directed lines connecting them (all directed away from a central point, the root); and that once two languages diverge, they evolve independently. See Gleason (1959) for an early discussion of this problem. 15 Dyen and Aberle (ms.) most recently show the insight into cultural history gained by use of lexicostatistics in the subgrouping of languages.

MATHEMATICAL DEVELOPMENTS IN LEXICOSTATISTIC THEORY

105

Time

I

Language no.

1 22

3 . .

.

.

.

nn

Fig. 1. Family tree summarizing genetic relationships among languages

CLUSTER ANALYSES USING SIMILARITY MATRICES A relatively easy but crude way of handling lexicostatistic d a t a is to reduce it t o a similarity matrix F„. Fn(i,j), the entry in the i t h row a n d jtb column, represents the lexicostatistic

index

M or the proportion of cognate j u d g m e n t s , — ,

between the i t h

language and the jtb language. F„ is an n x n matrix, it is symmetric, i.e. Fn(i,j) = F„(j,i), and the diagonal elements are all 1. T h e problem of constructing the best tree to represent a matrix of relationship indices is pertinent not only to the study of evolution (Cavalli-Sforza and E d w a r d s 1967) but occurs in m a n y contexts (see, e.g., Hartigan 1967). The solution generally depends on the definition of 'best', and this is where the evolutionary type of model is distinguished f r o m other subgrouping problems, a n d where lexicostatistics is distinguished f r o m most other problem areas. T h e first large-scale application of subgrouping techniques in the context of lexicostatistics was m a d e by Dyen (1965), in classifying 371 Austronesian languages. His procedure seems to involve a certain a m o u n t of subjectivity. It has n o t been described in algorithmic terms and it is not clear whether it is applicable to arbitrary d a t a sets. It does seem to have the desirable property that it seldom results in a spurious subgrouping. U n f o r t u n a t e l y it also discriminates against valid subgroupings, and tends to p r o d u c e greatly oversimplified family trees. T h u s D y e n ' s family tree of the Austronesian languages has a n u m b e r of very large subfamilies with little internal structure. O n e subgrouping procedure which is well-defined a n d has been used by D . Sankoff (1969) and by G . Sankoff (1968), is a special case of the hierarchical grouping proced u r e of W a r d (1963). Proceed as follows. Let a, b, etc. stand for the n languages, each of which has weight w(a) = w{b) = ... = 1. Find the two different languages i and j such that F„(i,j) is a maximum. Posit a node (point in the tree) pt connecting only to i and j, to represent the close relationship between them. T h e two rows a n d two columns corresponding to i and j are deleted f r o m the matrix and replaced by a single row a n d single column representing pu where

106

DAVID SANKOFF

F„_ ,(/>„«) = [w(i)Fn(i,a) + w(j)Fn(j,aMw(i) + w(j)] Fn-i(a>P\) = [w(i)Fn(a,i) + wQ)F„(a,j)]/[w(i) + w(j)] added onto the bottom and right end of the matrix, respectively, and w(px) = w(i') + w(j). The rest of the n— 1 x n— 1 matrix Fn^x is carried over from Fn. The procedure continues with F„_lt i.e. search for a non-diagonal matrix entry Fn_ i(k,l) which is maximal. If k and I represent two languages, we proceed as for i and j above, and establish a new node p2. It is possible, however, that k, say, is py. In this case, px and I should be linked to a node p2. Continuing in this way, we derive matrices Fn_2,Fn_3,...,F1. In this process, we find nodespi,pi, --,p n -i, and hence the whole tree. This procedure produces only binary trees (although this is easily corrected) and does not necessarily produce a unique tree. Nevertheless in actual use and in trials on simulated data it performs well even in comparison with the sophisticated programs of Dayhoff (described below). Milke (ms.) experimented with a cluster analysis procedure which is somewhat similar to of the one just described. Since there is little empirical or theoretical basis on which to evaluate various matrix-based clustering procedures for lexicostatistics (and there are many more available), and since the trend seems to be toward more sophisticated methods, we will not dwell on further description of such cluster techniques.

LANGUAGE FAMILIES WITH NEITHER BORROWING NOR CHANCE COGNATION

Cluster analysis based on a similarity matrix does not depend strongly on the evolutionary nature of the lexicostatistic model. The techniques which we now describe do. In a formal split-and-diverge model of linguistic evolution, one can first assume that no borrowing takes place (or that it can be detected) and that judgments of cognate retention can all be made with perfect confidence. In this idealization, the true genetic tree is usually the only one logically compatible with the data, as long as the number of meanings is sufficiently large. This follows from the fact that if languages a and b are more closely related to each other than either is to c or d, then there is likely to be at least one meaning for which the lexical realizations in a and b are cognate, those in c and d are cognate, but the first pair are not cognate to the second. Then any genetic tree containing a subtree topologically equivalent to those in Figure 2 is not logically compatible with the data. This principle has been rediscovered many times (e.g. Quentin 1925; Gleason 1959; Dobson 1969a). Gleason used it to determine the more recent splits in a Niger-Congo language family. Dobson (1969a, 1969b) has used it in devising and programming an algorithm for recovering the complete true tree from lists of cognation relationships holding between the words for the test-list meanings in the different languages. Her algorithm begins by considering the first four of the n languages. These may

107

MATHEMATICAL DEVELOPMENTS IN LEXICOSTATISTIC THEORY

> >

a

c

a

d Fig. 2.

< b

d


skaphtas). Otherwise the aspiration (and the later spirantization, e.g. skaphtas > skaftas) would have affected the p in skapjan and we should have had Germanic ' s k a f j a n ' rather than skapjan. If the restructuring is just partial and affects unproductive forms only, subsequent processes of rule generalization or reordering may leave these unproductive forms intact as 'relic forms'. In German, e's which historically but not synchronically were derived from a (e.g. bet 'bed' from *badi, netz 'net' from *nati) became phonemic, and did not participate in the subsequent lowering to x which in great numbers of dialects affected the e's that were synchronically derived from a. Whereas the productive umlaut rule in dialects that have undergone this simplification turns a to x, the derivation of the relic forms from their historical prototypes requires the application of a no longer synchronically operative version of the umlaut rule which turns a to e. Similarly, the structure of idioms, or idiomatic constructions such as the one exemplified by the more the merrier, may represent older syntactic constructions which are no longer productive in the present state of the language. In general, like flies in amber, evidence for dead rules is preserved in the lexicon, where it is carried along in an inert form which is not affected by restructuring of the system. The lexicon, then, is the one area of grammar where Saussure's model of linguistic structure, and Meillet's model of comparative linguistics which is based on it, are truly applicable. This is why one of the more successful of the practical maxims in the traditional comparative method is that one should reconstruct from the idiosyncratic rather than from the general. In a comparative method which had no place for rules in synchronic grammars, it was quite correct to focus on forms exempt from general processes, since only these forms could, within the limitations of that method, give a reliable picture of the past. They still have a special importance, and often become crucial in comparative arguments. What we can now also do, however, is to begin tackling the evolution of the system of rules itself, and thereby to bring productively derived forms into their rightful role in comparative grammar. Indeed, without explicit grammars as the basis of comparative linguistics, we cannot properly apply the principle that reconstruction should proceed from idiosyncrasies. In the first place, we cannot identify the idiosyncrasies without knowing what the rules are. This is a more serious problem than is commonly realized. For example, in the Greek nouns of the type patir 'father', there are two genitive plurals, patron (in Homeric) and paterdn. Of these, paterdn looks like the exception, since genitive endings normally have end stress, including the genitive plural (as in monosyllables, e.g. khthonSn). We might therefore conclude that paterdn is the older form, which has been regularized to patron. But this would be wrong. The form paterdn is on closer inspection seen to be derived regularly by a rule of Attic. That patron is the older form is confirmed by both the internal Greek evidence and by the comparative evidence (Kiparsky 1967).

120

PAUL KIPARSKY

Even if we somehow succeed in identifying the idiosyncrasies without having an explicit characterization of the regularities to go on, we still cannot project them into the past without further ado. We require not only that they be idiosyncratic in the present system but also that they be regular in the prehistoric system that we are reconstructing. This again cannot be safely determined without reconstructing the relevant portion of the grammar. An alarming number of hypotheses about earlier stages of Proto-Indo-European are based on hapax legomena or otherwise problematic forms in languages which present abundant philological difficulties as it is. Explicit rules for the proto-language constitute at least one check on this potential source of error. It has often been claimed that constructing a generative grammar amounts to internal reconstruction, or at least involves using the 'methods' of internal reconstruction. We can now see why this claim is wrong. The rules of a generative grammar cannot be converted into rules of a historical grammar simply by deleting the shafts of their arrows. The synchronic rules themselves are subject to change in time, and part of the problem of reconstruction is to recover these changes. On the other hand, the historical information that can be obtained from a full synchronic description is far richer than what can simply be read off from the rules directly. A relationship between two lexical items, for example, may be perfectly apparent to the linguist as an etymological fact, although it has no synchronic standing in the language. Historical studies therefore presuppose synchronic descriptions but go beyond them in quite essential ways. In the following sections I will investigate a specific problem in historical phonology: that of Grassmann's Law. The discussion can be limited to two languages, Greek and Sanskrit, and involves a relatively small number of rules, though their interrelationship is anything but simple. It is in the historical analysis of purely phonological rather than morphological rules, and especially of productive ones such as these, that generative phonology has so far made a tangible contribution to historical linguistics. 2.

GRASSMANN'S LAW: SYNCHRONIC DESCRIPTION

In both Greek and Sanskrit, Grassmann's Law (G.L.) is a phonological rule that eliminates aspiration before a following aspirate. E.g.: Greek Sanskrit

thi-the-mi > tithemi, the-throph-a > tetropha dha-dha-mi > dadhami, bho-bhudh-ye > bobudhye.

Although I will write aspirated stops in the usual manner as ph, bh, etc., they are in fact single phonemes distinguished from unaspirated stops by the features of aspiration. Accordingly, G.L. is to be stated in the following form: (G.L.) [—vocalic]

[—aspirated] /

X

—vocalic +aspirated

ON COMPARATIVE LINGUISTICS

121

Bpth languages also have a Cluster Rule (C.R.) that eliminates aspiration immediately before obstruents: Greek Sanskrit

gluph-s-o > glupso labh-sya-te > labsyate (lapsyate by the later voicing assimilation rule).

(C.R.)

[—vocalic]

[—aspirated] /

[+obstruent]

There are additional restrictions on these rules, to which I return directly. First note how the correct ordering 1. C.R. 2. G.L. accounts for the apparent 'throwback' of aspiration to the initial stop in certain roots when the final stop of these roots is deaspirated, e.g. Greek Sanskrit

trepho, Fut. threpso dohmi, 2.P. dhoksi.

The behavior of such roots is a consequence of the two rules and their ordering if we assume that their initial and final stops are both aspirated in the underlying representation. The correct phonetic forms are then derived as follows: Underl. repr. C.R. G.L.

threpho — trdpho

thrdphso threpso —

I now turn to a more detailed inspection of the form of C.R. and G.L. in Greek and Sanskrit, and the way they interact with other phonological rules in these languages. Some quite striking similarities as well as many differences will emerge as the particulars of the picture in each language are filled in. One important restriction in the application of G.L. in both Greek and Sanskrit is that the aspirate in the environment must be part of the root. In neither language does an aspirate in an ending have a dissimilatory effect on an aspirate in a root or a prefix. Cf. Gk. Imper. phathi, phanethi, straphethi, Perf. Inf. pephanthai, Aor. Pass. ephanthert; Skt. vibhubhis, dhestha.4 This is not an easy restriction to state. If we can assume that reduplication is not separated from the root by a morpheme boundary, then the rule, in both languages, must be as follows: (G.L.)

[—vocalic] -> [—aspirated] / X [ + aspirated] where X does not contain a + boundary

The restriction that the second aspirate must be in the root interacts in an interesting way with the cluster rule to explain the fact that in Sanskrit as well as in Greek the initial segment of a double-aspirate root is not deaspirated when the ending begins 4

It should be pointed out that there are a few exceptions to the restriction in Greek: notably the

Aor. Passives etethen, etuthen.

122

PAUL KIPARSKY

with an aspirate. Thus, in Sanskrit dhugh-dhve the cluster rule deaspirates the rootfinal segment, yielding the correct phonetic form dhug-dhve. G.L. fails to apply to it since the remaining aspirate in the ending does not qualify as an environment. 5 Exactly the same situation prevails in Greek. The Aor.Pass. of the root threph is ethrepthen, derived from underlying ethrephthen by application of the cluster rule but not of G.L. 6 In Sanskrit, the relation to G.L. of a well-known change which affects clusters of the form 'voiced aspirate + stop' to yield clusters of the form 'voiced unaspirated stop + voiced aspirated stop' is of importance. Examples: rundh-tas -* runddhas, rundhthas -» runddhas. This change consists of two steps. The first is a progressive assimilation of voicing and aspiration. We will call this step Bartholomae's Law (B.L.), which is the name usually applied to the whole process rather than just the assimilation. +obstruent (B.L.) +aspirated +voiced This is followed by the second step, deaspiration of the first aspirate of the resulting diaspirate cluster. If we assume that this deaspiration is effected by the already discussed C.R., we get derivations as follows:

B.L. C.R.

rundh-tas rundh-dhas rund-dhas

rundh-thas rundh-dhas rund-dhas

Diaspirate roots undergo both G.L. and B.L., e.g. dhugh-tas dugdhas. The mutual order of G.L. and B.L. is irrelevant, but both must precede C.R.: G.L. B.L. C.R.

dhugh-tas dugh-tas dugh-dhas dug-dhas

dhugh-thas dugh-thas dugh-dhas dug-dhas

But this conflicts squarely with our former observation that G.L. has to be applied after C.R. in a derivation such as C.R. G.L.

dhugh-dhve dhug-dhve —

I agree with most of Anderson's (1970) objections to the unpublished treatments of Grassmann's Law in which Zwicky (1965) and I (1965) both tried to get around this 5

The 2.Sg. Imperative ending -dhi poses a problem. Instead of the expected dhugdhi we get dugdhi. The spelling ethréphthën does not indicate a cluster of two aspirates. This is the ordinary Greek convention for writing a cluster of the form 'unaspirated stop + aspirated stop'. 8

ON COMPARATIVE LINGUISTICS

123

problem in different ways. Anderson's proposal, in line with his general theory of local ordering, is that the three rules in question are related by the following ordering conditions: (1)

(2)

(3)

C.R. precedes G.L. Hence /dhugh + dhve/ C.R. dhug + dhve G.L. (inapplicable) B.L. precedes C.R. Hence /labh + ta/ B.L. labh + dha C.R. lab + dha G.L. precedes B.L. Hence, with (2) /dhugh + ta/ G.L. dugh + ta B.L. dugh + dha C.R. dug + dha

Since what I will say about the history of Grassmann's Law does not particularly depend on this ordering problem, I will not pursue the question further here. In Sanskrit, voicing is distinctive for aspirates, and the deaspirated stops retain the voicing of the underlying form: cha-chand-a -> cachanda dha-dhar-a -+ dadhara But in Greek, all aspirates are predictably voiceless. Greek is generally assumed to have had voiced aspirates at an earlier point, which underwent the sound change of Devoicing of Aspirates (D.A.). (D.A)

[+aspirated]

[—voiced]

Presumably this led to restructuring, so that aspirates in Greek are voiceless in underlying representations. However, Greek may still have possessed a synchronic rule like (D.A.). This is because processes of aspiration are accompanied by devoicing when they apply to voiced stops, as in the perfect Skha (from ago), cf. pepompha (from pempo). Another Greek rule which is of importance in connection with G.L. is the rule which turns s into h in sonorant environments: (s-Aspir.)

s-h

o b s t r u e n t ]

} _ [-obstruent]

The root sekh, which retains the s intact in the aorist e-skh-on shows it as h in the future hek-sd. The h that results from this rule drops by G.L. in the same environment as the other aspirates. Thus, in the'present tense of the same verb, the underlying

124

PAUL KIPARSKY

sekh-o becomes hekh-d by s-Aspir. and then ekho by G.L. We do not have to add a new environment to G.L. or change it in any other way in order to accommodate this new case of deaspiration, if we assume that h is aspirated and that its unaspirated counterpart is simply a smooth onset. Derivations in Greek: (s-A). (C.R.) (G.L.)

the-throph-a — — te-troph-a 3.

e-threph-then — e-threp-then —

threph-s-o — threp-s-o —

sekh-o hekh-o — ekh-o

sekh-s-o hekh-s-o hek-s-o —

GRASSMANN'S LAW: ITS PREHISTORY IN SANSKRIT

It is clear that C.R. (in some form) must have arisen earlier than G.L. in both Greek and Indie. To see why this is so, note first that the order 1. 2.

C.R. G.L.

is a marked order by the conditions proposed in Kiparsky (1968), which therefore could not have arisen by reordering. We can also exclude the possibility that C.R. should have been entered in the grammar later in time than G.L. but in position before it. For if G.L. had applied before C.R. it would have deaspirated the initial aspirate of diaspirate roots in all environments. But then there would have been no basis for subsequent reintroduction of the initial aspirate. Internal reconstruction can be pushed still further back in Sanskrit if other facts are taken into consideration. It has long been known (Wackernagel 1896:126) that double-aspirate roots in Sanskrit occasionally lose both aspirates before endings beginning with s, in direct conflict with the rules of Section 2. Cf. grtsas 'dexterous' (probably historically but hardly synchronically related to ghrdh- 'be eager'), the 3.P1. form bapsati (bha-bhs-ati, from the root bhas 'chew'), etc. These anomalous forms, which are most frequent in the oldest texts (Rigvedic), are evidently relic forms reflecting an earlier version of Indie phonology. Our problem is to find a grammar that meets the following two conditions: 1) it must generate the relic forms (grtsas etc.) from their etymological representations (ghrdhsas etc.); 2) it must be a possible ancestor of Sanskrit, i.e. Sanskrit must be derivable from it through bona fide forms of linguistic change. The traditional explanation for relic forms like grtsas is that B.L. and C.R. (it will be recalled that these are combined into a single step in the usual formulation) originally preceded Grassmann's Law, with derivations such as the following: (B.L.) (C.R.) (G.L.)

ghfdhsas ghfdhzhas ghrdzhas gfdzhas

dhoghsi dhoghzhi dhogzhi dogzhi

ON COMPARATIVE LINGUISTICS

125

Subsequently a sound change deaspirated and devoiced the clusters dzh, gzh, bzh to ts, ks, ps, yielding grtsas, doksi. By analogy, doksi then became dhoksi (Thumb and Hauschild 1958:297, Wackernagel 1896:127). The simple but fatal flaw in this solution is that G.L. would not apply to the third line in the above derivation to yield the bottom line, for the same reason that it does not apply to dhugdhve to give 'dugdhve' or to vibhubhis to yield 'vibubhis\ namely because of the restriction, well motivated both in Greek and Indie, that the aspirate in the environment must be in the root, as pointed out in 2. The solution is thus based on an incorrect formulation of the facts and must hence be rejected. Application of G.L. to clusters like dzh not only fails to explain relic forms like grtsas; the existence of such clusters can not be considered as established by internal and comparative reconstruction. They have been proposed by Wackernagel (1896: 239), following Bartholomae. One of his pieces of support is the false argument from G.L. Another concerns the development of underlying ja-ghs-ta to jagdha. He supposes that ja-ghs-ta first became jagzdha by B.L. and C.R. and that the sibilant subsequently dropped between stops by the well-known rule. In the first place, this would be, if correct, an argument for an optional s in the environment of B.L. rather than for the clusters dzh etc., which are at issue. In the second place, it is false. We may just as well suppose that ja-ghs-ta first lost the sibilant and the resulting ja-gh-ta subsequently became jagdha by B.L. and C.R. Wackernagel's remaining arguments are comparative in nature and concern Pali and Iranian forms. They indicate the former existence of voiced clusters but are, so far as I can see, neutral as to where the aspiration was. Only the Iranian evidence for voicing is clear, but here aspiration was lost anyway. If we instead assumed a former order 1. 2. 3.

B.L. G.L. C.R.

the traditionally reconstructed pre-Sanskrit forms would be correctly derived in this particular case: (B.L.) (G.L.) (C.R.)

ghrdhsas ghj-dhzhas gfdhzhas gfdzhas

dhoghsi dhogh?hi doghzhi dog?hi.

This account must assume that (C.R.) was then reordered to precede (G.L.) to yield the Sanskrit system. But as was pointed out at the beginning of this section, this is an impossibility, because (C.R.) can never have preceded (G.L.) in Sanskrit, as the root type with two aspirates would in that case have been irrevocably lost. Nor would the difference between s and the other environments of the (C.R.) be accounted for. This solution, too, must therefore be rejected.

126

PAUL KIPARSKY

Neither of these changes of rule order, then, will give us the desired form of preSanskrit: one does not generate the relic forms and the other is not a possible ancestor of Sanskrit. Apparently no simple ordering of the rules can be made to work, and we have to say that the applications of (C.R.) before s followed Grassmann's Law (and, if clusters like dzh did exist, also Bartholomae's Law), whereas the applications of (C.R.) in the other environments (e.g. dh and # ) preceded it. The Sanskrit ordering represents a simplification of this system, by which all subcases of (G.L.) came to be treated on a par. 4.

GRASSMANN'S LAW: ITS PREHISTORY IN GREEK

What historical connection, if any, is there between G.L. in Greek and G.L. in Sanskrit? This question was raised already by Grassmann himself in 1863. His conclusion, reached with obvious reluctance, that the Greek and Indie dissimilations were historically unrelated, has remained unquestioned ever since. Neither neogrammarians nor structuralists have seen any need to re-examine the argument that underlies this dismal conclusion in the light of subsequent insights into linguistic structure and linguistic change. The classical line of argument was concisely summarized by Bloomfield as follows: A further step in the reconstruction of the historical events proceeds from the facts that the loss of aspiration results in Sanskrit in [b, d, g], but in Greek in [p, t, k]. This implies that the Primitive Indo-European [bh, dh, gh] had already become unvoiced [ph, th, kh] in pre-Greek when the loss of aspiration took place. Since this unvoicing does not occur in Indo-Iranian, we conclude that the de-aspiration in pre-Greek and the de-aspiration in pre-Indo-Iranian took place independently (1933:351). The argument has two steps: 1. The deaspiration took place after the Greek aspirates became unvoiced, since it yields voiceless stops in Greek. 2. Since the deaspiration took place after the Greek aspirates became unvoiced, it took place independently in the two languages. The same argument, with both steps, could be repeated with respect to G.L. and i-Aspiration: 1. The deaspiration took place after Greek s became h, since it applies to this h. 2. Hence it took place independently in the two languages. I want to argue in the following that both steps of these arguments are false, and that, furthermore, the second step cannot even be defended within Bloomfield's linguistic theory. Finally, I want to propose an account which not only relates the Greek and Indie dissimilations but also sheds light on certain apparently anomalous phonological developments in Greek. Taking the second step of the above argument first, let us inquire after the rationale that underlies it. By what general considerations could one support the contention that the specifically Greek sound changes (D.A. and s-Aspir.) preclude Greek and

ON COMPARATIVE LINGUISTICS

127

Indie from subsequently borrowing a sound change (G.L.) from each other? There is no hint of this in Bloomfield, and other statements on the subject, though wordier, are no more informative. What immediately comes to mind is some strong form of the stammbaum theory, say the contention that dialects which differ in some respect cannot undergo joint innovations, i.e. that isoglosses cannot intersect.7 From this, the second step follows directly. It is entirely possible that this is what underlay Grassmann's reasoning in 1863. Yet very few linguists of the twentieth century would subscribe to this view; certainly not Bloomfield: The comparative method presupposes clear-cut splitting off of successive branches, but the inconsistent partial similarities show us that later changes may spread over the isoglosses left by earlier changes; that resemblance between neighboring languages may be due to the disappearance of intermediate dialects (wave-theory); and that languages already in some respects differentiated may make like changes (1933:318).

The discussion preceding this passage appears to indicate that Bloomfield considered only the spreading of innovations over closely related dialects, and not over language boundaries. Such a restriction is quite unmotivated and contradicted by the facts.8 But whether or not Bloomfield had some such restriction in mind is not relevant, for it can in any case not have played a role in the case of G.L., since the two specifically Greek changes cannot obviously have split Greek and Indo-Iranian into two separate languages. But there is another possible, quite distinct reason why the Greek devoicing of aspirates should have been thought to preclude a subsequent common Greek-IndoIranian deaspiration, even when the falsity of stammbaum theory is admitted. Sounds have commonly been regarded as unanalyzable entities rather than as complexes of distinctive features. But then there is nothing at all in common between G.L. in Greek and G.L. in Sanskrit. The only thing that can be said is that Greek changes ph, th, kh into p, t, k respectively, whereas Sanskrit changes bh, dh, gh into b, d, g ' The history of the notions 'wave theory' and 'stammbaum theory' is a confused one. The point of Schmidt (1872) was twofold: that isoglosses could intersect; and that isoglosses between the IE languages in particular did intersect. Both these claims came to be called 'wave theory' and their denial 'stammbaum theory'. In many of the early discussions around Schmidt's monograph the attention centered primarily on the specific IE implications. Thus, Leskien (1876) all but ignores the general question and seeks to criticize 'wave theory' by citing evidence of a historical kind that IE tribes which migrated became isolated. Later, as interest in IE dialectology subsided, the general question of intersecting isoglosses came to the fore. Later, stammbaum theory also came to denote not, as originally, the denial of wave theory, but instead the much weaker (and totally undisputed) assertion that dialects can split into separate branches. In this sense it is correct to say that wave theory and stammbaum theory are 'compatible' and 'reconcilable'. 8 Boas (1911), Jakobson (1931), Sommerfelt (1962), Emeneau (1956). The very languages we are dealing with provide examples in support of this view. The Greek-Indo-Iranian change of syllabic nasals to a — a change so unusual that the possibility of independent development in each of the languages is highly unlikely — took place after s-Aspir. in Greek. Thus *dnsus ->-Gk. dasus (Lat. densus); *nsis ->-Gk. asis (Lat. ensis, Skt. ash), cf. Thieme (1958). As these rules are in marked order and the change of syllabic nasals to a led to partial restructuring, we can be certain of their relative chronology.

128

PAUL KIPARSKY

respectively. Under this view, these are entirely separate events which could not in principle be related to each other. We may say, then, that in general a corollary of denying the feature structure of sounds is that a sound change can be borrowed from one dialect into another only if the domain in which it applies functionally is the same in both. This, in effect, trivializes the wave theory by restricting it, for practical purposes, to the spread of lexical items and the like. Again, this argument may have played a role in the early days, but clearly not for Bloomfield, to whose conception of phonology distinctive features were crucial. On p. 353 of Language he specifically remarks that features ('habits of articulation ... common to several phonemes') are themselves subject to change. And obviously the whole discussion of G.L. on pp. 349-51 shows that he considers it to be the same sound change in both languages. There was thus no reason at all for Bloomfield to deny that G.L. could have 'spread across the isogloss' created by D.A. and s-Aspir. in Greek. The fact that he did must be simply due to the venerable status of G.L. as a paradigm case of independent innovation. Gotze (1923) considered it a 'Musterbeispiel' of an isogloss that could be shown by the methods of linguistic reconstruction to be spurious. Traditional conclusions remain standing long after their foundations have been exploded. Let me now return to the first step in the argument, the one purporting to show that G.L. could not have arisen in Greek before the aspirates were unvoiced and s became h. It furnishes a good illustration of an application of the principle that I claimed in section 1 to underlie traditional reconstruction, namely that the phonetic representations of the descendant language should be derivable from the phonetic representations of the ancestor language by application of the sound changes in the order of their relative chronology. I would now like to show how in this particular case there is good reason for altogether different conclusions. The two Greek innovations under discussion, D.A. and s-Aspir., are in unmarked order with respect to G.L. It is accordingly a possibility that G.L. originally preceded them and is historically earlier. If so, G.L. may well have been an innovation that took place before the Greek-Indo-Iranian dialect area of Indo-European diverged. At a certain stage of Pre-Greek we then had C.R. and G.L. followed by the two innovations D.A. and i-Aspir.: (la) (lb) (2) (3)

C.R. G.L. D.A j-Aspir.

At this point, the deaspiration by Grassmann's Law still yielded voiced stops which alternated with voiceless aspirates. In place of classical Greek trepho : threpso we still had *drephd: threpso. This alternation between voiced unaspirated stops and voiceless aspirated stops motivated the retention of underlying voiced aspirates and D.A. as a

ON COMPARATIVE LINGUISTICS

129

morphophonemic rule. Also, at this point the aspirate derived from s did not dissimilate: we had *hekho in place of classical Greek ekhd. D.A. and i-Aspir. stood in a marked order with respect to G.L. and both were hence susceptible to inversion with it. The inversion of D.A. and G.L. had the effect of replacing the two-feature alternation in *drepho: threpso by the familiar onefeature alternation of classical Greek, and also permitted further simplification in the morpheme structure rules. Aspirates could now be restructured as voiceless in the lexicon. The inversion of s-Aspir. and G.L. had the effect of making the aspirate h from s undergo the same dissimilation by G.L. as the other aspirates. Together, these two order changes give us the attested situation of classical Greek. Thus we do not, after all, have to admit the extraordinary coincidence of a rule as striking as G.L. arising independently in two IE dialects as closely related as Greek and Indie. In fact, we now must choose between two entirely distinct ways in which the Greek and Indie dissimilations might be historically connected: the wave theory explanation and the reordering explanation. As always in choosing between competing theories, the problem is to find crucial cases that will discriminate between them. A critical difference between the two accounts is that only one of them (the one which supposes G.L. to antedate D.A.) assumes the former existence of forms like *drepho, replaced by the classical Greek trephd. Would relic forms have survived in Greek which have escaped this replacement? Let us look more closely at the stage of Pre-Greek with G.L. and D.A. in marked order which this account postulates. The morphophonemic distinction between the IE root types ghedh and gedh would at this point still have been retained to the extent that these roots occurred before an obstruent or word boundary that caused loss of their second aspirate by C.R. Thus, [drepho] would have been represented as /dhrebho/ with an initial underlying aspirate, and [glupho] as /glubho/ with an initial underlying non-aspirate, because of the distinction in the future forms [threpso], [glupso]. This would have been the case quite generally in root nouns, verbs with aorists and futures in s, etc. But there were also roots that never happened to occur in any such pre-obstruent environment. In these, the distinction between the types ghedh and gedh must have been irrevocably lost for want of morphophonemic support. In such opaque forms the constant operation of G.L. rendered the initial stop ambiguous or neutralized as to underlying aspiration. Whether given the morphophonemic representation /gedh/ or /ghedh/ they obtained the phonetic form [geth] by G.L. and D.A. Let us reconstruct the further development of this hypothetical stage of Pre-Greek. When the devoicing rule D.A. came to apply before G.L. the outcome in derived forms, we recall, was that the morphophonemic difference in the underlying form obtained a phonetic reflex: [dreph5] (cf. [threpso]) became [trepho] while [glupho] (cf. [glupso]) stayed unchanged. In opaque forms, however, the original distinction between initial aspirate and non-aspirate had been lost and could not re-emerge. If the etymological distinction between initial aspirates and nonaspirates in such

130

PAUL KIPARSKY

roots was lost in opaque forms, we should expect these forms to have been restructured in their underlying representations according to their actual pronunciation, that is, with initial nonaspirates. The analog to this happens, for example, when the final devoicing rule is lost in Yiddish and Swiss German. Words that do not take vowelinitial inflectional suffixes, and therefore lack any paradigmatic evidence for the voicing of their final obstruent, turn up with a voiceless final obstruent when the underlying voicing distinction is reestablished on the surface. As soon as the aspirates are devoiced in Greek, both root types, /gedh/ and /ghedh/, should therefore show up as /geth/ in opaque forms, in contrast to their distinct treatment as /geth/ and /keth/ respectively in all roots that were subject to the cluster rule in some of their occurrences. However, in the Greek case additional factors are relevant to the shape which the restructured opaque forms assume in the lexicon. Let us consider a hypothetical case in which the initial consonant is labial, say /bhadh/. The principle that among alternative underlying forms, other things being equal, the closest to the phonetic form is preferred, would indeed choose /badh/ rather than /bhadh/ as the underlying form of isolated [badh] at the time before the aspirates had been devoiced, when both underlying representations would have given the right output. But the restructuring has to be seen in the context of the whole phonological system, including morpheme structure. In the early stages of the Indo-European languages, there were certain restrictions on the underlying consonantal system. The voiced unaspirated labial [b] was missing, or at best extremely weakly represented. This fact would have tended to counteract the preference for /badh/, and motivated an underlying /bhadh/ instead, which after the devoicing of aspirates would have yielded path (and not bath). Systematic considerations of a more complex sort were relevant to the other series of stops as well. Manner of articulation was often predictable on the basis of place of articulation and root structure. Magnusson (1967) has given an elegant argument that the root types T-D and D - D H (where T, D, D H represent the main manner classes of the Proto-Indo-European obstruent system and the dash represents intervening nonobstruents) were in complementary distribution according to the place of articulation of each of the obstruents. For example, k-d occurs (e.g. *kerd- 'heart'), but g-dh does not occur. Conversely, t-g does not occur, but d-gh occurs (e.g. *drtghwa 'tongue'). The same complementarity seems to hold for the reversed cases D-T and DH-D. Magnusson suggests that the underlying forms are /T-DH/ for T-D and D-DH, and /DH-T/ for D - T and D H - D . The outcome can be derived by a rule stating that the 'weaker' of the two obstruents assimilates to the other by becoming voiced or losing its aspiration, whichever is applicable. Weakness is determined by the place of articulation according to the hierarchy Kw T K P

(weakest)

(strongest)

ON COMPARATIVE LINGUISTICS

131

Hence gw has the freest distribution of the voiced stops, and kw and ghw, its sources by the assimilation rule, the most restricted distribution of their respective stop classes. Conversely, b has the most restricted, and p and bh the freest distribution among their stop classes. Note that this explains the well-known gap in the Indo-European root pattern of roots like T-DH and DH-T. These are always subject to assimilation and hence do not appear on the surface. Such systematic considerations, pertaining to the phonological inventory and to the structure of morphemes, would have been relevant if the historical processes in question took place at a fairly early time. Their effect would have been to cause certain cases of opaque D-DH forms to be analyzed with /DH-DH/, in order to retain the otherwise valid constraints on underlying segments and on the combinations. The logical form of the argument does not change because of this. We can still determine whether the distinction between the old root types /D-DH/ and /DH-DH/ was consistently retained in opaque forms. Non-retention of the distinction may because of these systematic factors amount to redistribution rather than constant representation by /D/. What is important is whether the etymological distinction between the initial aspirated and non-aspirated voiced stop was given up or not. This is the crucial test for determining the relative chronology of G.L. and D.A. If opaque forms retain the etymological distinction between the root types gedh and ghedh, D.A. was earlier than G.L.; if not, it was later. As test cases, then, we require words which are synchronically isolated in Greek and which by means of clear cognates in Germanic, Italic, or some other Indo-European dialect that preserves reflexes of distinctive aspiration, can be shown to have had Indo-European roots of the type gedh or ghedh. A list of these words with their proposed Indo-European cognates follows. The majority of the words that qualify as test cases are of the type ghedh. It appears from their treatment in Greek that the etymological distinction between initial aspirate and initial non-aspirate is indeed obscured. 1. brekhmos 'top of the head, sinciput' has since Grassmann generally been regarded as cognate with OE brxgen 'brain' (Gmc. *brayna-). The problem in this etymology has been the Greek b- (rather than p-). Linguists have either assumed a proto-form *mregh-, requiring a change of IE mr- to Germanic br-, which is dubious, as Persson (1912:34) shows, or else set up *b(h)regh-, thereby giving up the attempt to account for the shape of the initial in Greek. An IE *bregh- is in any case highly unlikely, because Indo-European had very few cases of b, and indeed absolutely none in roots with other stops. If we set up *bhregh-, both the Germanic and the Greek forms are derivable from a normal Indo-European root type, provided that we accept the reordering of G.L. and D.A. here proposed. Greek brekhmos then is a relic form showing the result of the application of Grassmann's Law at a point when the aspirates were still voiced. 2. bothros 'pit, hollow in the ground' has generally been related to Latin fodio 'dig', fossa 'ditch', Gothic badi- 'bed', a family of words which is widespread in other IndoEuropean languages as well. Again, the only problem is the Greek b-, where traditional

132

PAUL KIPARSKY

Indo-European phonology would expect p-, since the Latin and Germanic cognates clearly show a root *bhodh-. The etymology is evident enough for Schwyzer (1939: 262) to have been willing to assume an application of Grassmann's Law before the devoicing of aspirates in this one word: 'Griech. bothros bothunos lassen sich nicht trennen von lat. fodere fossa usw (Walde P. 2, 188) ... Zeigt both- fur idg. *bhodh-, dass gelegentlich einmal schon auf der Stufe der med. asp., nicht erst der ten.asp., dissimiliert werden konnte?' If we can replace this isolated pre-(D.A.) application of G.L. by a general reassessment of the chronology, the argument is greatly strengthened. 3. pekhus 'arm' is related to Skt. bahus, and Germanic forms such as Olcel. bogr prove that the form was IE *bhaghus. Here the opaque initial segment turns up as voiceless in Greek, presumably because prior to the devoicing of aspirates [baghus] was represented as /bhaghus/ rather than as the equally possible /baghus/, which would have had the rare /b/. Why this tendency to avoid /b/ was sufficient to counteract the tendency to choose the underlying form closest to the pronunciation only in some words but not in others is not clear. As the following example shows, there was vacillation between the two resolutions of the conflict in some morphemes. 4. buthos 'depth', bussos 'bottom' have often been considered related to puthmin 'base, foundation' and Latin fundus 'bottom of a vessel'. Latin shows that the root was *bhudh; the Greek forms thus constitute a case where the synchronic opacity of the initial segment prior to the devoicing of aspirates led to actual doublets in Greek after the devoicing took place. The form bussos, from *bhudh-yos, together with the next two cognate sets, shows that the assimilation of stops to y, which caused the loss of the second aspirate in this case, came later in time than the devoicing of aspirates. Otherwise forms like *phussos would have resulted. In fact they are only found in derived cases, such as takhus: thasson, where a reordering into unmarked order can have taken place. This chronology is compatible with what we know about the history of Greek phonology. 5. deisa 'sludge' is related by Vasmer to Armenian gej 'wet' and OCS zizda, Russ. ziza 'slop', zidkij 'of runny consistency'. Vasmer does not mention the crucial fact that the Armenian word clearly indicates an Indo-European voiced aspirate. Greek deisa must therefore reflect IE *ghweidhya, which is compatible with the d- only if we assume that Grassmann's Law applied before the devoicing of aspirates. 6. thrupto (aor. di-etruphen) 'crush, spoil' might be related to drupto (druph-) 'tear'. The prehistoric alternation *drubh-yd ~ *edhrup-sa, became druph-yo ~ ethrup-sa by devoicing of aspirates. The normal outcome of the present by the reordering of G.L. was *truph-yd, which developed into thrupto after j-assimilation, because this rule was placed before G.L. But * druph-yo might in some meanings have been dissociated from the rest of the paradigm (ethrupsa etc.) and then been reanalyzed as having the root /druph/. So analyzed, the word would have retained its d- into classical Greek, where it developed a (fairly limited) paradigm of its own, with such forms as aor. edrupsa. 7. agathos 'good' has always been connected with the Germanic words for 'good'

ON COMPARATIVE LINGUISTICS

133

(Gothic gods). Although there are some problems in the vowel correspondences, the major difficulty, again, has lain in the initial voicing of the Greek word, which by the traditional theory of Grassmann's Law conflicts with the double-aspirate character of the root indicated by Germanic. It is crucial to note that agathós is morphologically isolated in Greek, forming no primary derivatives that might have triggered the cluster rule. The word is therefore opaque, and can be added to the list of Greek relic words preserving evidence of the prehistoric ordering of Grassmann's Law before the devoicing of aspirates. The variant akathós, with k, attested in Hesychius, serves to confirm our analysis. 8. kephalé 'head' has a cognate in O H G gebal. This is a case of a historically double-aspirate word which, in spite of being isolated, surfaces in Greek only with the voiceless initial, for reasons that I am unable to explain. Roots of the gedh type seem to have been less frequent in Indo-European than those of the ghedh type. Consequently there are very few test cases of this type available. I know of only two completely clear cases: dolikhós 'long' (cf. Goth, tulgus) and gómphos 'nail, peg', gomphios 'molar tooth' (cf. Engl. comb). In the latter case, a doublet kómbos 'molar tooth' (with the common lenition after nasals, cf. brekhmós ~ brègma, also bregmós, brékhma) is in fact preserved in Hesychius. The evidence is thus not plentiful but nevertheless suggestive. We do not find (nor would we expect to find) doublets, or voiced initial consonants, preserved in every case. But we do find — and this is crucial — that the etymological distinction between the root types gedh and ghedh is lost in Greek exactly where it had no morphophonemic support from the cluster rule. If this is a fact — and it could be refuted by giving better etymologies for the crucial examples — it clinches the early operation of G.L. in Greek, and makes it likely that G.L. was a rule shared by the Greek-Indo-Iranian dialect area of Indo-European before it split up into separate dialects. REFERENCES

W. S. 1953. Relationship in comparative linguistics. TPhS 1953. S. A . 1970. On Grassmann's Law in Sanskrit. Linguistic Inquiry 1.387-96. Toronto. ANTILLA, R. 1969. Uusimman àànnehistorian suunnasta ja luonteesta. Turku. BECKER, D.A. 1967. Generative phonology and dialect study: An investigation of three modern German dialects. Ph.D. Dissertation, University of Texas. BEVER, T.G., and D. T. LANGENDOEN. 1 9 7 2 . The interaction of speech perception and grammatical structure in the evolution of language. Historical linguistics in the perspective of generative grammar, ed. by R. Stockwell. Bloomington, Indiana. BLOOMFIELD, L. 1933. Language. New York. BOAS, F. 1911. Introduction. Handbook of American Indian languages, Vol. I. Washington, D.C. [Reprinted 1963, ed. by C.I.J.M. Stuart.] ALLEN,

ANDERSON,

134

PAUL KIPARSKY

EMENEAU, M . B .

1956. India as a linguistic area. Lg 32.3-16.

GÖTZE, A. 1923. Relative

Chronologie

von

Lauterscheinungen

im Italischen.

I F 41.78-194.

JAKOBSON, R. 1931. Über die phonologischen Sprachbünde. TCLP 4. (Reprinted 1962 in Selected Writings, Vol. I. The Hague.) . 1958. Typological studies and their contribution to historical comparative linguistics. PICL 8.17-25. (Reprinted 1962 in Selected Writings, Vol. I. The Hague.) KING, R.D. 1969. Historical linguistics and generative grammar. New York. KIPARSKY, P. 1965. Phonological change. Ph.D. Dissertation, Massachusetts Institute of Technology. . 1967. A propos de l'histoire d'accentuation grecque. Langages 6.73-94. . 1968. Linguistic universals and linguistic change. Universals in linguistic theory, ed. by E. Bach and R. Harms. New York. . 1972. Explanation in phonology. Goals of linguistic theory, ed. by S. Peters. Englewood Cliffs, N.J. LABOV, W. 1972. The internal evolution of linguistic rules. Historical linguistics in the perspective of generative grammar, ed. by R. Stockwell, Bloomington, Indiana. LESKIEN, A. 1876. Die Deklination im Slavisch-Litauischen und Germanischen. Leipzig. MAGNUSSON, W.L. 1967. Complementary distributions among the root pattern of Proto-Indo-European. Linguistics 34.17-25. PERSSON, P. 1912. Beiträge zur indogermanischen Wortforschung. Uppsala. SCHMIDT, J. 1872. Die Verwandtschaftsverhältnisse der indogermanischen Sprachen. Weimar. SCHWYZER, E. 1939. Griechische Grammatik, Vol. I. München. SOMMERFELT, A. 1962. The development of quantity as evidence of Western European linguistic interdependence. Diachronic and synchronic aspects of language: Selected articles, by A. Sommerfelt. The Hague. THIEME, P. 1955. Review of T. Burrow, The Sanskrit language. Lg 31.428-48. . 1958. Review of Carnoy, Dictionnaire étymologique du proto-indo-européen. Lg 34.510-5. THUMB, A., and R. HAUSCHILD. 1958. Handbuch des Sanskrit, Vol. I. Heidelberg. VOYLES, J . B .

1968. G o t h i c a n d Germanic. Lg 44.720-46.

WACKERNAGEL, J. 1896. Altindische Grammatik, Vol. I. Halle. WATKINS, C. 1962. Indo-European origins of the Celtic verb, Vol I. Dublin. ZWICKY, A. 1965. Topics in Sanskrit phonology. PhD. Dissertation, Massachusetts Institute of Technology.

AREAL L I N G U I S T I C S : SOME G E N E R A L CONSIDERATIONS WERNER WINTER

The contributions to volumes of Current Trends have in the past usually tried to provide surveys of the state of the art in specific subfields by a rather detailed review of articles and books dealing with these subfields; at times, authors have taken a quite different approach to their topic and tried to arrive at an overview of the total complex ofproblems before them. The following remarks are meant to fallin the second category; it should, however, be stated outright that what seems achievable at this point is not a summary of uncontested results, but rather a set of questions, some hints of possible answers, some attempts at revising commonly held views — nothing very definitive then, an essay rather than a summary. All this leads me to favor an eclectic, and not an encyclopedic, approach to the question: a decision which seems warranted in view of the fact that the results hitherto achieved in the field of areal linguistics apparently do not form a coherent fabric or even a somewhat consistent pattern, but merely a patchwork quilt of colorful, but largely unrelated, data and anecdotes. It hardly means overstating a case if one says that the inspection of a wide array of observations made and committed to print leads to the conclusion that in this field nearly everything can be shown to be possible, but that not much progress has been made toward determining what is probable and to what degree, so that the time does not seem to be at hand yet for an empirically based, coherent theory of areal linguistics (provided there can be such a theory for a complex field not amenable to investigation under simplified and consistent test conditions, and not just an ordered set of observations concerning events that can be shown to have taken place). Modern linguistic science started with a strong, almost exclusive emphasis on language history. This emphasis had some extremely positive results: language phenomena could be explained in terms other than system-internal or philosophical, and language could be viewed in conjunction with other human activities. At first, the interpretation of language as a historical phenomenon concentrated almost exclusively on the pure time dimension: the family-tree model favored by August Schleicher and his followers marked only this dimension (with geographic considerations entering more or less accidentally); every language placed on the tree as a node was associated directly only with immediately preceding or following languages within the same line of descent. Even certain aspects of the time dimension were not fully appreciated: the question whether languages equidistant (in terms of intervening

136

WERNER WINTER

nodes) from the reconstructed common parent tongue were contemporary to one another hardly entered into consideration — in the absence of earlier attested stages, Lithuanian of the second half of the second millennium A.D. was given equal status, for comparative purposes, with Vedic Indie of perhaps the first half of the first millennium B.C.; it was not absolute chronology that counted, but relative chronology, and even that only within the bounds of a given sub-branch of the entire group — excluding Modern High German, but not Lithuanian, Old French, but not Church Slavic, Medieval Latin, but not Old Irish, from the basic data used in reconstructive work. The family-tree model served well toward an explanation of what is conventionally called 'inherited' features; as these features do not constitute the totality of a language, but only a (nevertheless significant) part of it, the explanatory power of the model was not unlimited, and it was only natural that other approaches should be proposed that would offer explanations not available in the Schleicherian context. If time is the one major dimension in Schleicher's model, the next model to be proposed in the history of Indo-European studies emphasizes, at least on the surface, the space dimension. While in Schleicher's frame of reference only a transmission of linguistic features along a vertical, downward line within a given sub-branch is accounted for, Johannes Schmidt's 'wave theory' introduced the horizontal dimension: new linguistic features were said to develop in some given place and spread from there sideways to neighboring dialects and languages, losing momentum as they moved further away from their place of origin. There can be no question that Schmidt's proposal is in many ways more realistic than a pure Schleicherian view: the latter, to work fully, would require the continued existence of linguistic communities totally isolated from one another and, at any given time, totally homogeneous internally. For this to happen, human life would have to be spent in small groups on desert islands; history shows that this type of situation, even if it somehow comes to pass, does not prevail — a Swiss-family-Robinson existence may appeal to the romantic mind, but it stands no chance for survival. For all its merits, the wave model could not replace the family-tree model altogether. The latter was better suited for a presentation of some simple facts of language history that could not be handled very well in the context of wave theory: the family tree did depict quite adequately the varying degrees of relationship that existed between, say, Middle High German and Modern High German, Middle High German and Middle English, Middle High German and Old Russian, and it did point up the chain of transmission for features common to an earlier and a later stage of a particular language. To be sure, even for these limited purposes, the family-tree model suffered from grave shortcomings. Above all, the terminology used was seriously misleading: parent tongue, daughter language, inherited items are all metaphors that suggest wrong parallels in the extralinguistic world, and the whole family tree itself is drawn up as though language development was some kind of perpetual parthenogenesis. The most

AREAL LINGUISTICS

137

serious defect is probably the fact that the imagery used seems to imply that languages entered as nodes on a sub-branch of the tree are discrete entities, comparable to individuals in the human world, while it is a characteristic of language development that it is always gradual and without abrupt breaks. However, if the family tree tended to suggest false parallels in the outside world, even the basic notion of wave theory, for which we may use a currently favored term, diffusion of linguistic features, is far from being an adequate representation of the actual events. Like its congeners in the terminology of dialect geography, the notion of spread or diffusion suggests action (though perhaps action initiated by some external factor, as by the rock thrown into the tranquil waters of a pond) on the part of the properties said to affect adjacent languages. However, what does take place is entirely different in nature: linguistic features (just as certainly very many, if not most, cultural features) are not diffused, nor do they diffuse, but given the proper circumstances, they are adopted as speech habits by members of outside speech communities. The picture customarily presented by linguistic geography of a diachronic type is in the nature of a photographic negative: it reverses the active and the passive sides in the transmission process, thereby replacing the proper notion of relatively free, and often even playful, selection by its very opposite, that of imposition by outside forces. While wave theory could not replace the family-tree model, it provided a useful complement. This could be so because both theories concerned basically similar, perhaps even identical, processes occurring under different circumstances. Both propose to describe language transmission; both have to be reinterpreted in terms of the active role of the supposed target of the transmission process: language transmission along the branches of a family tree is adoption of linguistic features within the bounds of speech communities, language transmission as described by wave theory is adoption across such boundaries. Adoption of 'inherited' features into the speech habits of an individual is a learning process akin to that employed in acquiring 'foreign' features; to a child, all language can be said to be foreign, and the term 'native language' deserves a place in a list of misleading metaphors. The historical bias of the family-tree model is obvious; but also of wave theory it can be claimed that its main interest is in diachrony. When it is noted that a feature has spread to (i.e. has been taken over by) a language other than the one in which it originated, wave theory is not primarily concerned with the moment of adoption: borrowed features matter really only as incorporated features, and incorporation can be observed only through time. This attitude does not weaken the effectiveness of the theory; on the contrary: While purely synchronic description may be able to point out a disruption in the network of subsystems of a language, historically oriented investigation will add information about the nature of this break, both by clarifying the source of the phenomenon incorporated and the conditions under which the transfer took place; moreover, such investigation will also identify transferred features with a perfect fit, items which do not show up as disruptions of existing systems and can therefore not be identified as 'foreign' by means of a strictly synchronic analysis.

138

WERNER WINTER

For speech habits to be adopted, certain conditions must prevail; without contact, there will be no transfer, and there will be transfer only if the contact is sufficiently strong. There appears to be no basic difference in this respect between transfer within a speech community (intralanguage transfer) and across its boundaries (interlanguage transfer): in both cases a sufficiently strong contact must exist between minimally one person who uses a certain speech habit and another who is in a position to adopt it. Normally the presence of both the originator and the secondary user in one place and at one time will be required; for thousands of years, though, the invention of writing has made it possible to circumvent the requirement of the physical presence of the originator of a speech habit by replacing his actual spoken utterance with a record thereof, and recently modern media like radio, television, records, or tapes have introduced a new type of vicarious presence of the originator. Even so, it can be stated that no really crucial change in the conditions for the transfer of speech habits has occurred: If we say that an utterance is placed in time and space not by the circumstances under which it is produced, but by those under which it is perceived (a possible, albeit perhaps not intuitively natural, specification which seems called for in view of the all-important role of the adopter), the manifestation of a speech event in written or recorded form turns out to be just a subclass of all speech events as identified at the moment of perception, and we can continue to insist that for a transfer to take place the conditions of contemporaneity and collocality must be met. It has long been known, though, that these two conditions are not sufficient to have a transfer really occur. First of all, channels of communication must exist between the originator and the potential adopter of a speech habit: it is not enough that these two live side-by-side; if their contacts are exclusively with others and not with each other, no transfer will take place between the two. Furthermore, a transfer must be in response to some need; the adopter must be in a position to derive some gain from the transfer. Such gain may be of a very material nature: only by using material from a second language may one be able to do business with an outside group of people; or the inducement may be less tangible, though not less real: the second language may be that of a particularly prestigious group, and by adopting part of their speech habits one might be able to share their glory, if only in one's own imagination. Material gain may be derived from exploiting the socially inferior as well as the superior; there is thus an inducement here to acquire some command of the language of either type of people. The more subtle, non-material gain, on the other hand, stems only from adopting the usage of those one considers superior to oneself; adoption of linguistic features of this type will always be unidirectional (except that of course for some specific purposes a group otherwise deemed inferior will be ranked higher, which explains the adoption of agricultural or hunting terminology by inhabitants of urban centers). Prestige as reflected in the direction of linguistic transfer may occur in combination with other traits of a certain group of people. Thus, it may go with political power in a given region: what is technically

AREAL LINGUISTICS

139

called the 'enrichment' (obogascenie) of the national languages of the USSR involves transfer from the dominant language of the Union and the principal language of intraUnion communication to the languages of only regional importance. But political power is not absolutely decisive: in the Roman Empire, the politically leading language, Latin, adopted much more from the language of the, at least initially, politically subordinated Greeks than vice versa; Latin, in its turn, remained a source language long after it had ceased to be used as the principal language of any political power (disregarding here the role of Latin as the language of the Holy See because recourse to Latin was not limited to the sphere of influence of the Roman Catholic Church). To a certain extent, prestige defies rational definition; in this respect, prestigious and fashionable are almost interchangeable terms. To be sure, prestige usually does not develop without a discernible cause, be it power, economic importance, size, educational level of a group; but certainly in the field of language interaction fashions seem to linger on long after the original conditioning factors have disappeared. Since, for instance, the acquisition of foreign languages is, at least in modern times, intimately connected with the teaching and learning of foreign languages in schools, there tends to be a simple phase displacement conditioned by the failure to rapidly adjust curricula to changing conditions, partly because the scope of what 'ought to be taught' is determined to at least some extent by what the teachers once had to learn themselves. Thus, the replacement of French by English in West European school systems has been slower than one would have expected from an assessment of the relative importance of the two languages in current world affairs (the recognition of American English as a respectable competitor of British English has even been slower to materialize); the role of German in East Central and Eastern Europe remains stronger than present-day channels of communication would lead us to think likely; the place of Russian in Central and Western European school curricula is in no way commensurate with the political and scientific weight of the Russian language. We thus have to recognize at least the extralinguistic conditions of collocality, contemporaneity, existence of channels of communication, and presence of social stimuli as necessary presuppositions for linguistic transfer to materialize. In addition to these extralinguistic factors, at least one language-internal one appears to be equally important. To the extent that speech habits convey meaning, the adopter must be aware, or at least believe to be aware, of their meaning: no meaningful feature can be transferred unless the adopter attaches an interpretation of its meaning to the form he takes over (this interpretation need not agree with the full range of meaning of the form in the source language, and in exceptional cases the interpretation may be quite wrong because the adopter misunderstood the form he selected for adoption). Stated differently, a linguistic condition for the transfer of a meaningful speech habit is the existence of at least unidirectional intelligibility; bidirectional intelligibility may be found, but it is an obligatory condition only for intralanguage transfer. A mapping of the results of interlanguage transfer leads to the determination of the

140

WERNER WINTER

boundaries of linguistic areas. A group of languages forms a linguistic area because of the presence of some shared feature or set of features in all these languages; these features are present because linguistic transfer has taken place. Area boundaries represent the limits of specific transfer processes; they separate areas with contrasting speech habits. Dialect areas are merely a subclass of all linguistic areas, characterized over against broader types of areas by the characteristic of mutual intelligibility found within the bounds of the area. If the extent of a linguistic area is determined by the presence of a shared feature or set of features, it follows that a language or dialect may be part of more than one area at a time. This phenomenon is well known from dialect geography: a given local dialect D will share one set of features with one, possibly neighboring, dialect or group of dialects, a second set with another, and so on — with respect to F ^ D then is part of a linguistic area with D 1 ; with respect to F 2 , with D 2 , with respect to F 3 , with D 3 , etc. In dialect geography, only those groupings receive lasting attention that are confirmed by repetition, that is, the grouping D - D t becomes of real interest only when the shared set F t is found to be matched by other sets F ' ^ F " ! , etc. While strictly speaking any single line surrounding the region in which a certain set of features is found to be in use delimits a linguistic area, only those lines are conventionally assigned such status that occur in combination with other lines as bundles of isoglosses. The agreement between isoglosses tends to be only partial; bundles will fan out in places and converge again in others, which means that it is a characteristic of linguistic areas as determined by the presence of clusters of isoglosses that their boundaries are sometimes neatly defined, but just as often fuzzy. Clustering of isoglosses is an empirically observed phenomenon; that it should be found follows from the fact that linguistic transfer depends on favorable linguistic and extralinguistic conditions and that these conditions will then exert their influence not on isolated items but on the entire languages as used at the time of transfer. If sameness in the conditions leads to homogeneity (varying in degree, to be sure) in the results of transfer, it does, on the other hand, go without saying that a change in the conditions will also affect the nature and distribution of isoglosses, and the assignment of a given language to a specific area will vary accordingly. While it would be correct to say that dialect D in the above example forms part of one area with with respect to the feature sets F l f F ' l 5 F " t , etc., it has become customary in dialectology to single out one of the isoglosses and to take it as a mark of the boundary between the area (here D - D t ) and others. The advantage of this decision is obvious: the fuzziness which prevails with all borders formed by bundles of isoglosses disappears when only one isogloss is selected. As long as the isogloss chosen is part of a massive bundle and does not deviate substantially from the main body of the bundle, no harm is done by limiting oneself to the consideration of just one feature or set of features. However, sometimes the requirement that the isogloss selected should be one of a very substantial cluster is not met, and by far too much importance is attached to a single dividing line. The classic example in this respect is

AREAL LINGUISTICS

141

that of the centum-satam line which for generations was taken to be a clear indication of an east-west division of major rank within the group of Indo-European languages, with virtually no attention paid to the fact that there are very few isoglosses approximately parallel to the centum-satam line, from which the latter could be selected as a representative case; as it is, the line stands for hardly anything but itself and has thus very little value from the point of view of methodology as employed in dialect geography. If nature and size of linguistic areas are the result of historical conditions, both linguistic and extralinguistic, it follows that one can attempt to use observations about the existence and extent of linguistic areas as the basis for conclusions about the conditions that caused them to come into existence. This possibility is of greatest interest where other informations about the history of a region and about groups of people inhabiting the region are scant or even nonexistent. Thus, the results of contact between two languages or dialects are noted, the direction of the transfer is determined, and conclusions are drawn as to extralinguistic properties of the relationship assumed to have existed between speakers of the two languages or dialects. Or, for a relatively widespread linguistic phenomenon, a 'center of diffusion' is found, and this center is then interpreted as the center of a cultural area. In view of the far-reaching deductions from observations concerning linguistic areas, it seems appropriate to turn now to a discussion of the question of the internal structure of such areas, the more so as this question seems to have been largely neglected. All transmission of linguistic habits involves two parties, the original user and the adopter of the habit in question. For these two, the requirements discussed above, such as collocality and contemporaneity, apply. Both the originator and the adopter can be parties to further transmission events, including those involving the very same habit. Each of these transmission processes again is subject to the requirements of contemporaneity and collocality as far as this process itself is concerned; but there is no need that these requirements be met for both the first-mentioned and the second transmission process taken together — it even seems to be the rule that the second process will differ from the first at least with respect to one condition, be it collocality or contemporaneity. If these observations are correct, it appears that linguistic areas will generally lack internal homogeneity, and that for all historical conclusions to be based on the discovery of linguistic areas, the internal structure of these areas will be of the greatest importance. T w o sample cases, formulated in an abstract fashion, will serve to illustrate this point: Inspection shows that a set of features is found in a group of languages A B C D E F ; the shared isogloss indicates that the group can be interpreted as a linguistic area, at least if (as to be assumed here for the sake of the argument) the languages are in use at one time and in a geographically contiguous region. Still, the investigation of the transfer processes in terms of pairs of languages actually involved at a given time may yield totally different patterns, such as:

142

WERNER WINTER

Time levels [1]

[2]

I

Languages involved in feature transfer A -

B

II

V A

¥ B -»

C

III

¥ A

¥ B

V C

D

IV

# A

V B

V C

U D

V

V A

V B

V C

V D

I

B

II

V B

III

B

-»

A D D

«

E V E - > F C

¥ A

• E

C

F «- A

V E

» C


z>r, but not the reverse. From this digression, we turn to developments within the Prague School in their relation to typological problems. In some of the earliest statements of the Prague School as a whole or by individual members of it, the possibility and the importance of structural comparisons is emphasized. Indeed, it is noted as a point of superiority of the new structuralist doctrine that contrary to the limitations imposed by the Neogrammarians, in the words of Trnka (1929:69-70), structuralists employ 'the synchronic method which has no need to limit itself to related languages, but can compare with each other any systems of expression whatever, can go not only from form to meaning, but also from meaning (function) to form....' 2 5 In carrying out such comparisons there was in the beginning an evident reluctance to employ the word 'type' because of the disrepute into which the traditional morphological typology had fallen. Hence such expressions as analytic, synchronic or structural comparison were devised, while Mathesius coined the term 'characterology' for investigations of this sort. Eventually, however, the term typology reasserted itself and became the usual one. While there was thus general agreement that comparison of structures was possible and significant, there was equally evident disagreement as to what this consisted of. In terms of our present framework we may characterize this as a vacillation between the two poles of the individualizing and the generalizing approaches. Thus, in the theses presented by the Prague Circle of Linguistics to the Sixth International Congress of Linguists in Paris in 1948, it is stated: 'This latter [sci. typology] will take account of the total structure of the languages; avoiding a reliance on isolated resemblances or differences, it will examine all its properties while taking into consideration their hierarchy.' 26 However, Skaliôka, one of the members of the school most concerned with typological matters, in his review article of 1958 considers that the distinguishing mark of modern investigation is that languages are no longer assigned to types as wholes.27 Moreover, the connection between types and implicational universals, already emphasized by Jakobson (e.g. 1941, especially p. 59; 1953:313), is specifically pointed out. Indeed, Skaliôka, in an earlier work on the Czech declensional system (1941:4), had stated : 'Type is for me a collection of grammatical characteristics, which are close to 25

'La méthode synchronique, qui n'a pas besoin de se limiter à des langues apparentées, mais peut comparer entre eux n'importe quels systèmes d'expression, peut aller non seulement de la forme à la signification, mais aussi de la signification (fonction) à la forme...' 28 Cited in Vachek and Dubsky (1960:305-6): 'Cette dernière rendra compte de la structure totale de la langue: évitant de s'appuyer sur des ressemblances ou des differences isolées, elle examinera toutes les qualités en considérant la hiérarchie de celles-ci.' 27 However, Skalicka's point, as can be seen from his papers on Bantu and Chinese (1946a and 1946b) is not so much, as it might seem, that typologies refer to only delimited aspects of language, since a typology based on a word morphology is still central for him. It is rather that languages cannot be assigned wholly to one type or another. Moreover, in particular languages, the noun might, for example, be predominantly inflective, while the verb was agglutinative, etc.

174

JOSEPH H. GREENBERG

each other, such that if one of them is in a given language we expect that the second will be present also, as well as the third, and so on.' 28 The actual practice of part rather than whole language typologies was indeed already to be found in the classic work of Trubetzkoy (1929) on vowel systems, although without explicit reference to implicational universals. This work also serves to illustrate what is probably the most important contribution of the Prague School to typological methodology, namely the introduction of relational rather than absolute properties, most completely and clearly carried out in the area of phonology. In this way languages could both be said to exhibit or not to exhibit the relation of opposition between high and low vowels even though the phonetic 'realization' might be different in individual languages which presented this contrast. Relational notions, a basic concept of structuralism, could thus be employed in order to resolve the conflict between the demands of individual language descriptions and of comparability among languages. The phonetic facts of individual languages would be repeated while not sacrificing broader comparisons based on the common presence of particular oppositions. This approach is used in the Jakobsonian development of a universal framework consisting of a quite limited number of binary oppositions by means of which all phonological contrasts in language could be expressed and which has turned out to be a powerful tool for typological comparison. Even a brief review of structuralist typology should take into account the utilization of frequency data for typological purposes which figured with some prominence in theoretical discussion. In a document produced by a group of Czech linguists (Theses, 1957), it is even stated regarding quantitative linguistics that 'so-called linguistic typology may be regarded as one of its branches, and likewise linguistic characterology, which V. Mathesius sought to establish on the basis of the comparison of structural features between a number of European languages, cannot do without an appeal to statistics'. The basic notion applied essentially to phonology was that even languages with highly similar phonologic systems might differ typologically in the way these resources are used functionally and these differences could be investigated by the study of frequency. The work of Kramsky (1946-8, 1959) is along these lines. For example, the degree of text exploitation of phonemic inventory features is calculated by a formula V = where P ; is the proportion of phonemes with a certain feature in the inventory and P, is the proportion of the text frequency of these phonemes to the total text frequency. On this basis, for example, a language is typologically more vocalic if the text frequency of its vowels is proportionately greater than its inventory frequency. As indicated earlier, another source of interest in typological approaches was to be found in the tradition of American anthropological linguistics. The emphasis here, at least in the earlier period, i.e. up to the 1950s, was essentially individualizing. " "Typ je pro nine soubor gramatickych vlastnosti, ktere jsou si navzajem blizke, tiji je-li jedna v danem jazyce, ocekavame, ze bude i druha, tfeti atd.'

THE TYPOLOGICAL METHOD

175

A favorite expression during this period was 'pattern' first introduced by Sapir into linguistic studies and then assimilated into the general anthropological vocabulary as in the classic work of Benedict (1934) on patterns of culture. The notion of type here satisfied the need for some broader conceptual framework than just the portrayal of each individual language or culture as a unique pattern, a need which always seems to assert itself in such cases. In Benedict's work the concept of polar or extreme types played an important role. Whorfianism, a characteristic expression of cultural relativism inherent in much of the anthropology of that period, also made use of the concept of type. Thus a Hopi type was contrasted with SAE (Standard Average European) as exemplified by English. Unlike Sapir, Whorf's treatment was essentially independent of the traditional morphological typology. Also unlike Sapir it did not involve any attempt at the substruction of a general dimension or set of dimensions applicable in principle to all languages. Part language in place of whole language typologies made its appearance in the 1950s in American anthropological linguistics, especially in the work of Voegelin (e.g. 1956, 1962 and Voegelin and Yegerlehner 1956), whose activities contributed greatly to the revival of interest in typology in American linguistics. This approach, basically classificatory, was confined to phonology. It did not treat phonological systems as wholes, but rather as consisting of series, e.g. stops, within which types were distinguished. The quantification of the morphological typology by the present writer (Greenberg 1954) during this period was an attempt, based on Sapir's version of morphological typology, but with considerable revisions, by utilizing the analytical tools of contemporary American structuralism, to provide the rigor and exactitude which had always been a conspicuous weakness of this kind of typology. 29 It was indicative of the influence of structuralism in its American form in the anthropological linguistic tradition that the particular aspect of the typology on which Sapir placed the greatest value, the expression of derivational and mixed relational concepts, was lacking in my treatment, because it involved an approach from the direction of semantics; that is, it asked the question how certain concepts were expressed. It was also an indication of the growing tendency towards a generalizing approach that questions are raised regarding the properties of a universal frequency distribution of languages arranged in accordance with the numerical indices described. This is explicitly stated (76-77): '... non-random distribution of languages among typological classes suggests in the first instance a connection among linguistic traits themselves.' An interest, however, in typology as a means of characterizing individual languages as wholes persists in the American anthropological tradition as evidenced, for example, in Hymes' work (1961) on the typology of cognitive style in language. It seems undeniable, however, that the general trend, both in American and European *" Applications of and critical comments on this typology include the following: Bazell (1958), Householder (1960), Kroeber (I960), Pierce (1962, 1963, 1966), Cowgill (1963), Krupa (1965), Krupa and Altmann (1966).

176

JOSEPH H. GREENBERG

linguistics since the advent of structuralism, has been towards the generalizing approach with its concomitant linking of the topics of typology and universals. Thus Diebold (1965:210), in his review of developments in psycholinguistics notes that 'Questions about language universals and language typology have been paired in recent decades, especially now that new trends and interests in the study of language (such as psycholinguistics) have revived a flagging concern with typology.' Finally, it should be mentioned that there has been a considerable revival of typological interest both in Slavic and general linguistics in the Soviet Union, as evidenced by a number of collections of articles and individual works (e.g. Lekomtseva 1968, Moloshnaja 1963, Uspensky 1962). Here also, particularly in the most recent work the study of typology has been connected with that of universals (see especially Konferencija, 1968). IV. TYPOLOGY A N D LINGUISTIC THEORY

One of the conclusions that might be drawn from the historical review of the previous sections is that the manner in which language is conceived by a particular school or during a particular historical period bears a significant relation to the attitude taken towards typology. Thus the Neogrammarian view of language as a totality of individual items, each of which is to be explained as the outcome of a historical process, led to a negative attitude towards typology in a period in which typology was virtually synonymous with the attempt to characterize languages as wholes. The structuralist concept of languages as an organized synchronic system possessing a certain unity naturally led to a reversal of this attitude insofar as the historical no longer appears to be the only legitimate mode of explanation. Insofar as the notion of a linguistic system constituting a unity prevailed, the interest in individual language characterization revived, while, on the other hand, the extension of comparison to include synchronic structural forms led to an emphasis on universally valid descriptive categories and generalizations about language. The rise to prominence of the generative approach in American, and, indeed, in world linguistics in the last decade or so, as well as the existence of differing models for description, both outside of generative grammar (e.g. tagmemics, stratificational grammar) and within the school as of 1970 (e.g. Chomsky, Lakoff, Fillmore, McCawley) once more raises serious questions regarding the role of typology within linguistics. In particular, this would appear to be the case in regard to the generative approach, because it involves a different concept of the nature of language itself than that which prevailed in the previous historical period. Language is now defined as a set of internalized rules rather than a set of statements concerning surface phenomena. Furthermore, since as has been seen, questions of typology have in recent years become increasingly involved with those regarding universals, particularly implicational universals, the status of universals in this new setting will obviously have central relevance.

THE TYPOLOGICAL METHOD

177

It should be clear from both the theoretical discussion and the historical review of the role of typology in linguistics, that the concept of type has sufficient breadth and flexibility to accommodate any one of a number of theoretical approaches to languages. As has been seen, the domain of the typological function within linguistics is not even confined to languages as its individual objects. Even where languages do constitute its domain, typology is always possible as long as languages have properties in terms of which they can be compared. The viability of the concept 'property of language' is thus the essential prerequisite for the construction of typologies in which languages are the objects to be typologized. In generative theory the existence of a common vocabulary and common principles in regard to the form of rules guarantees the possibility of assigning common properties to different languages. However, while typologies are thus always possible under certain conditions, they would have little interest. Such would be the case if, under any overall approach to typology, the result of the application of any criterion would be either that all languages belong to the same type or that each language would belong to a unique type, that is, one of which it was the only member. This situation would indeed result from the following assumptions: 1. There are two kinds of universals, formal and substantive, pertaining respectively to the general organization of the grammar and its vocabulary. 2. Universals are only found in deep structure. This is the view enunciated in the earliest published treatment of the topic of universals by members of the generative school (Katz and Postal 1964, esp. 159-66; and Chomsky 1965, esp. 27-30 and 118). There is, correspondingly, a complete ignoring of the topic of typology. The first of these doctrines, that of existence of two kinds of universals only, formal and substantive, would seem to be highly restrictive in that even substantive deep structure rules, in distinction from the form that such rules take (i.e. the metarules concerning such rules) and the vocabulary of items that figure in them, would not be included. Thus if all grammars contained a rule S ->• N P + VP, a statement to that effect would not be a universal. That this was felt in some quarters to be unduly limiting can be seen from the following statement of Bierwisch incorporating the substance of such rules as universals (n.d. : 47—8). 'This means, however, that outside of the elements with which the grammars of individual languages operate, a part of their rules also derives from a general reservoir and as substantive universals determines the human faculties which we have called faculté de langage.'30 We now consider the second of the two assumptions; namely that all universals are in deep structure. If we admit that substantive deep structure rules are possible material for universals as suggested by Bierwisch, the consequences will be different, depending on whether we assume the same deep structure for all languages or not. The former assumption has been called the hypothesis of a universal base. 30

'Das heisst aber, das ausser den Elementen, mit denen die Grammatiken der einzelnen Sprachen operieren, auch ein Teil ihrer Regeln aus einem generellen Reservoir stammt und als substantielle Universalen die menschlichen Anlagen bestimmt, die wir als faculté de langage bezeichnet haben.'

178

JOSEPH H. GREENBERG

If we assume, then, that there is a universal base, and that all universals are in the base which constitutes the deep structure, then we have precisely the situation outlined earlier. All languages belong to one type that is defined by the universal base which they all share. If all universals are found only in the base, then there are no surface universals, nor presumably are they to be found in the transformational component which produces these surface structures. Under these circumstances, typology, while permissible, is indeed pointless. Here again, a broadening of the concept of universals in the thinking of some members of the school has occurred, and in this instance with repercussions in the area of typology. Bach (1965) in an article, "On some recurring types of transformation", dealing with relative clauses in a number of languages, tries to establish that there are a limited number of transformations which occur in this aspect of language. Depending on the choices among these possibilities, language will fall into types. Moreover, some, at least, of these transformational rules are themselves ordered through a relation of presupposition. Consider two rules, for example, such that Y presupposes X. This will mean that if any language has rule Y, it also has X, which applies earlier in the grammar. But there are languages which lack Y, although they do have X, and these will, of course, fall under a different type. It is not difficult to see here, as is noted by Bach himself, that the familiar implicational kind of universal is here being stated in terms of implicational relations among the rules themselves. Another well-known transformationalist who takes into account the possibility of lawlike limitations on the mapping of deep into surface structures and the associated notions of typology and implications universals is Fillmore (1968, esp. 1-2 and 61-70). Thus one of the three questions to which, according to him, scholars have addressed themselves in the search for common features in the syntax of the world's languages is the following: 'Are there universally valid constraints in the ways in which deep structure representations of sentences are given expression in the surface structure?' In answer to this, he points to studies of markedness and notes: 'If such studies can be interpreted as making empirical assertions about the mapping of deep structures into surface structures, they may point to universal constraints of the following form. While the grammatical feature "dual" is made up in one way or another in all languages, only those languages which have some overt morphemes indicating "plural" will have overt morphemes indicating "dual". The theory of implicational universals does not need to be interpreted, in other words, as a set of assertions on the character of possible deep structures in human language and the way in which they differ from each other (p. 2)'. Bach's discussion seems to be neutral in regard to the hypothesis of a universal base, while Fillmore, in the concluding section of the passage just cited, asserts that a theory of implicational universals with empirical content need not depend on there being systematic differences in the bases of different languages, i.e. is compatible with the theory of a universal base. If, contrary to the hypothesis just assumed, the base is not universal, additional

THE TYPOLOGICAL METHOD

179

possibilities for typological applications present themselves. The question arises whether those parts of the base which are not universal are merely random deviations or whether they take the form of recurrent alternatives, thus providing once more a basis for typological classification. For example, as a further elaboration of the word order typology presented by Greenberg (1963) it has been proposed that languages in their deep structure ordering fall into a very small number of fundamental types (e.g. McCawley 1970, esp. 298-9; Ross 1967). On this view, of course, typologies based on such differences assume fundamental importance in linguistic theory. However, the future of typology within linguistics is unlikely to be decided in the long run by which of a number of alternate views of the base and its relationship to universals prevails. Indeed, after a further period of theoretical development, the present framework of discussion may, as has happened in the past, become quite obsolete as a basis for discussion. There is here an empirical question. Presumably everyone will agree that there are features that are found in all languages, i.e. that are universal. There will also be agreement that all languages contain many features that are not universal and that any pair of languages differs in some respects, at least. The hypothesis that typology is of theoretical interest is essentially the hypothesis that the ways in which languages differ from each other is not entirely random, but show various types of dependencies among those properties of languages which are not invariant, differences stateable in terms of the 'type'. The construct of the 'type' is, as it were, interposed between the individual language in all its uniqueness and the unconditional or invariant features to be found in all languages. If this is empirically true, then any overall theory of language must ultimately be prepared to deal with these questions in a systematic way if it is not simply to ignore basic problems. As noted previously, the rise of generative grammar and of other at least partly similar contemporary theories, like every major development which introduces ideas about the nature of language, raises once again a question basic for typology, namely, what is it to be a property of language. The chief contribution it can make to typology is presumably to reconsider typologies in terms of new concepts concerning language. Typology will still be possible and, in fact, many of the classifications and criteria will merely be restatements of familiar classifications and criteria. 31 However, the use of rules rather than directly observable properties and of deep structures as theoretic constructs is capable of producing some novel results. For example, a typology of relative clauses, as already discussed in at least preliminary fashion by Bach, would, it is true, result in types which would be, by and large, the same as those based on surface observations. However, what has been called earlier the range set, would have imposed on it a novel logico-mathematical 81

For example, Fillmore (1968) restates Sapir (1917) in which a typology is proposed regarding syncretisms in the expression of pronominal subject and object in relation to both transitivity and voice.

180

JOSEPH H. GREENBERG

structure based on how each member resulted from the application of different sets of ordered rules. One other way in which contemporary models, and in particular that of generative grammar, raises questions about typology, issues from the very fact that such models are considerably superior to earlier approaches in their degree of explicitness. That is, present metatheory exercises far more constraint on what will figure in the grammar of a particular language as a 'fact' about the language than earlier more 'free-wheeling' procedures. As a result, it was easier in the past simply to equate for typological purposes 'property of a language' with the content of some statement in the grammar of a language, precisely because in the absence of theoretical constraints on the form of the grammar, a wide variety of statements could be made without any great importance being attached as to whether it actually did or could appear in the grammar. However, the powerful and carefully formulated constraints being put on grammars bring with them a situation in which certain very common sorts of statements commonly made about languages for typological purposes in the past will not figure in any immediately discernible way in the grammar. Thus, a grammar of standard French written on the generative model will probably not state anywhere that French has two genders, or that it has a certain number of phonemes. No doubt it will be possible to deduce such 'facts about French' from a generative grammar, but presumably, since the overall purpose is to incorporate in explicit form the linguistically significant generalizations about French, perhaps such statements are simply not linguistically significant generalizations. On the other hand, it would seem wiser not to rule out a priori the possibility that such properties might not be typologically fruitful. One might go still further in pointing out that the concept 'property of a language' is considerably broader than 'generalization stated in a grammar' by noting that there are, in the common usage of the term, many properties of languages neither overtly stated in grammars nor logically derivable from grammars, e.g. the statistical properties based on texts which have been used so often in the past. Again, it would seem to be a somewhat incautious move to wish to exclude such properties a priori, since it is clear enough that some of them exhibit lawful interconnections with properties which do figure in grammars. For example, the theoretical importance of hierarchies involving an opposition of marked and unmarked members is widely recognized. It is also a fact that in general there are frequency relations in languages in terms of which the unmarked is the more frequent. It seems arbitrary to omit considerations of such frequency properties in attempting to develop a more general theory of marking. V. TYPOLOGY A N D DIACHRONY

The definition of typological investigation by Marouzeau, cited in the introductory section of this paper as 'that which defines their characteristics in abstraction from

THE TYPOLOGICAL METHOD

181

history', represents what may be presented as a widespread, but far from universal contemporary consensus regarding the relation between the typological and historical study of languages. It is on this basis that there is fairly general agreement that, given the independence of typological and genetic criteria, there is no contradiction in the fact that closely related languages might be separated in some particular typological classification, while languages only remotely or not at all related are classed together. Moreover, there is no reason why a typological characteristic should not itself involve an historic fact about the language as long as no assumption is made that the properties found in the languages are themselves historically connected. For example, in Stewart's (1962) sociolinguistic typology, historicity is itself a criterion, one which is present in English and Krio, but absent in Volapük. Again, in the sort of dynamic typology to be discussed in greater detail later in this section, languages with the phonetic segment Py (a laryngealized high front glide) could be typologized in accordance with whether this segment arose from a former palatal implosive as part of an implosive series consisting ofPb,Pd, etc., or as part of the general phenomenon of glottalization of sonants, in which case it is usually a member of a set containing Pw, Pm, Pn, etc., and accompanied by glottalic pressure stops pp, tP, etc. It is true enough that the presence of other members of the series, or of morphophonemic alterations between Pj and Py are synchronic phenomena which could be utilized in most cases for typological assignment. However, this would obscure the basic insight that generalized processual comparison of a large number of languages shows that Py arises in two distinct ways and that the strictly synchronic trace, if any, that is left depends in general on the age of the change and other contingent phonetic developments within the overall systems. The point to be made in the present connection, however, is that if we decide to classify two languages together because both have Py, which can be considered to have come from Pj, this does not imply that the phenomena are historically connected with each other in these languages. The changes themselves may very well be, and in fact usually are, historically unconnected events in different language families or chronological periods. The view just outlined, although now so generally accepted that it might be called 'orthodox', did not always prevail. During the nineteenth century the assumption was generally made that languages which were classified together typologically could ultimately be shown to be related genetically. Schleicher seems to have been almost alone during this period in insisting on the absence of any necessary connection between the two. He asks (1859:37-8): 'Would it not be possible that while one language said varkasas adanti (Proto-Indo-European), the other expressed the same thing by vark ad? In that case the two languages would be related, but, viewed morphologically, belong to two different classes. The relationship of languages is not necessarily connected with their agreement in morphology and the latter cannot serve as proof of relationship.' 32 It has been noted earlier that 32

'Wäre es nicht möglich, dass, während eine spräche sagte varkasas adanti (indogerm. Ursprache), die andere dasselbe etwa durch vark ad ausdrückte? Dann wären die beiden sprachen verwant und

182

JOSEPH H. GREENBERG

until ca. 1930 the word 'typology' was not in use in linguistics, and the only widely employed typology was called morphological classification. According to Benes, Schleicher's viewpoint was considered 'heretical' (Ketzerei) at the time, and was answered by Steinthal. 33 The preponderant majority of linguists later in the nineteenth century were Neogrammarians insistent on concrete evidence of relationships of the kind that were found in Indo-European studies. It seemed to them that the assumption of genetic relationships among typologically similar languages could not be sustained. Since they generally saw no aim for typology beyond this one, the failure to prove this case was an important factor in the decline of morphological typology and led to the situation, alluded to earlier, in which Pedersen, an historian of linguistics with a strongly Neogrammarian point of view, simply sees typology as an attempt to classify languages on the basis of incorrect premises. It was structuralism which, by rehabilitating typology as structurally synchronic comparison in contrast to the diachronic comparison of genetically related languages of traditional historical linguistics, ushered in the modern view of the two as separate and independent approaches. This conclusion was strengthened by the concomitant development of a whole series of new typologies, for example, phonological, in place of the virtually unopposed dominance of a single typology in the earlier periods. Since these all gave different language classifications, it was clear that not all of them could coincide with a 'true' genetic classification and indeed, the conclusion that none of them did was rather obvious. Yet the relationship between typology and diachronic considerations is far from exhausted by the single question whether typological and historical classifications necessarily give the same results, which it now seems clear, they do not. The general type of question is one with an obvious parallel in the discussion of the anthropological problem of the relations among race, language and culture. At the time when Boas wrote his classic work, The mind of primitive man (1911), many scholars made casual assumptions about the identity of racial and linguistic classifications and even used such terms as 'Germanic race'. Boas essentially established the independence of these modes of classification, yet, as has now been realized for some time, although classifications need not coincide, this does not mean that there are no causal connections among the phenomena concerned. The basis for the connection between typological and genetic phenomena in languages can be stated in the following manner. Every language is typologically identical with itself in any typological classification, just as it is genetically identical with gehörten doch, morphologisch betrachtet, in zwei verschidene classen. Die Verwandschaft der sprachen ist mit irer morphologischen Übereinstimmung nicht notwendig verbunden und die letztere kann nicht als beweis der Verwandschaft dienen.' Of course, varkasas adanti is 'wolves eat' in Schleicher's reconstructed PIE. " In his positivistic approach to typology, he resembled earlier Indo-Europeanists who had been likewise concerned with typological problems, e.g. Bopp and Pott.

THE TYPOLOGICAL METHOD

183

itself in any genetic classification. Closely related languages are more likely to be typologically similar than remotely related languages or those not related at all. The extent to which this holds will be relative to the particular typology involved. Given two typologies concerned with the same data, it is obvious that the application of the 'finer-grained' typology, that is, the one with the larger number of typological classes, will produce results less in agreement with genetic classifications. There being more classes, the probability that closely related languages will have come to belong to different classes increases. Another basic factor is the relative historical stability of the criteria employed. Let us hypothesize two typologies, both of the simplest possible type logically; that is, monocriterial with two values, a and p. Suppose that in typology A the states a and (3 are very unstable, e.g. on the average, languages will change from one to the other in approximately 100 years. In typology B, the states are far more stable, e.g. a language will change from one to the other in about 3,000 years. On the basis of typology A, closely related languages will differ from each other almost randomly and there will be scant congruence between such a typology and genetic relationship, while in the latter case, the resemblance will be far greater. A third factor is the differential stability of the states themselves within a given typology. Let us once more posit a typology with two classes, and let us suppose that one state is far more stable than the other; such differences appear to be very common and to constitute an underlying factor in synchronic universality. Let us call the very popular state a and the unpopular one p. Then if the proto-language of a particular group was in state a, the probability is great that most or all of the descendent languages would agree typologically by still remaining in this state, while no such result would ensue if the proto-language had been in p. By the same token, however, the popularity of a will bring it about that languages only distantly related or that unrelated languages will agree typologically with each other in having a. What has been considered up to this point in the discussion is the attempt to use typological methods to assist in the discovery of more distant genetic relationships than those discoverable by genetic methods proper. There is, however, the opposite side of the coin. Given the literal infinity of possible typologies, there develops a search for unarbitrary or 'natural' typologies. The criteria of naturalness can be sought in two directions; synchronic implicational laws or some kind of historical significance. In a science which already has a strong historical component or is essentially historical, the attempt to distinguish 'natural' types from 'artificial' types by some test of historical relevance is easily understandable. For example, Krieger (1944:273), in regard to archaeology, a thoroughly historical discipline, characterizes the 'true typological method' as that 'wherein types are taken to be specific groupings of structural features which have proved historical significance'. The corresponding method in linguistics is to consider those classifications as 'natural' which arise from the application of the historically most stable criteria. Such attempts in the past have not been based on empirical studies of the actual

184

JOSEPH H. GREENBERG

stability of the criteria employed. Rather, a classification is carried out on the basis of criteria assumed intuitively to be significant on some notion of synchronic functional importance, e.g. word structure, and the assertion is then made that these characteristics are highly stable and their assumed historical stability is taken as further confirmation of their true significance. For example, Sapir states that his most significant axis of classification, that which is based on the presence or absence of derivational and mixed relational concepts, is also historically the most stable. The question at issue is really the following. Are there eternally stable linguistic traits which are not universal? These two requirements must be fulfilled. For if they are universal, then they cannot lead to a typology, while if they can change, then to the extent to which they are changeable they lose historical significance. The assumption that there are eternal non-universal traits is inherent in a certain type of reasoning which has sometimes been employed by linguists. The following may serve as an illustration. Maspero denied that Vietnamese could be an Austroasiatic language, because Vietnamese was tonal and the other languages generally agreed to be Austroasiatic, were not. The basic assumption here is that a certain trait, in this case tonalism, or its absence, cannot change over the course of time. Otherwise, there would be nothing incompatible between the tonalism of Vietnamese and Austroasiatic affiliation. Hence, there are two traits, tonalism and its absence, both non-universal and both eternal. Haudricourt later showed convincingly how Vietnamese became tonal. In general all traits concerning which such assumptions have been made turn out to be changeable on the basis of reliable direct or reconstructional historical evidence. The Austroasiatic example illustrates another and related point. Maspero is here characterizing a particular genetic family, Austroasiatic, by the presence of a particular typologic trait, nontonalism. This approach has sometimes been carried out in greater detail for particular language families, most commonly Indo-European. The attempt is made to define or delimit a family by the possession of a cluster of typologic traits. A wellknown example is Trubetzkoy (1939) in which Indo-European is defined by a number of typologic traits. Since none of these are completely stable, there is the strange result that if an historically Indo-European language loses one of these, it ceases to be Indo-European. Moreover, as pointed out by Benveniste (1952-3:110), on the basis of Trubetzkoy's definition, Takelma, an Oregon Penutian language, would be classified as Indo-European. This approach easily leads to the assumption that languages within the family as defined on the usual genetic grounds, which do not exhibit all of the posited typological characteristics, are in some sense deviant or atypical, and this deviance is to be explained by mixture or substratum. For example Hartmann (1956: 17) in his work on the typology of Indo-European, states that obvious mixed languages like Hittite, Tocharian and Irish are not considered. Thus certain Indo-European languages, e.g. Greek, Sanskrit, are typical and others are not. Is it possible to give some objective content to this notion even if we reject the foregoing approach? There are indeed here a number of significant and empirically

THE TYPOLOGICAL METHOD

185

researchable kinds of phenomena. One of them is the following. Within a language family are some languages more conservative than others when compared to the ancestral language, i.e. have changed more or less from this historically determined base? If so, then the more conservative languages may be considered more typical. However, a clear-cut objective answer on an overall basis is not easy to attain. Languages conservative in certain respects are innovative in others. Moreover, there is a strong tendency to take certain languages as 'typical' on unproved assumptions, and even to use its supposed typicality as a basis for reconstruction, thus producing a circularity. For example, traditional comparative Semitic studies tended to consider classical Arabic as highly conservative and then reconstruct Proto-Semitic in such a manner that it was very close to classical Arabic. This could then be used as an argument that classical Arabic was the most typical Semitic language. A second variant of this kind of interpretation raises another problem of great interest. Rather than solely taking resemblance to an ancestral language as the criterion of typicality, one may also consider widely shared resemblances within a single family based on similar parallel processes of change. This is the phenomenon called 'drift' by Sapir. On this view perhaps, since the widespread tendency in IndoEuropean is to lose the case system, Persian and English become more typically IndoEuropean than Lithuanian or Icelandic! We see, then, that there are a number of empirical problems of considerable interest which concern processes of change in related languages. The upshot of the preceding discussion is that the old approach to the relationship between typological and genetic linguistics, which either uses typological criteria to arrive at supposed genetic connections or seeks to choose among variant typologies by choosing historically stable, as it were 'quasi-eternal', traits in order to characterize genetic families typologically, is theoretically unsound. It is possible, however, while maintaining a clear distinction between typological and genetic criteria and classifications, to investigate the very real problems which gave rise to these attempts. Moreover, a basic tool for these studies turns out to be an aspect of typological studies which has been relatively neglected, both in theory and practice, namely change of type. Such study of change of type may be called dynamic or processual, since a process is simply a class of similar changes. In a fundamental contribution to the question of the theoretical relationship between typological and historical linguistics, Jakobson (1958) urged that the results of synchronic typology be used as a touchstone for the validity of historically reconstructed linguistic systems. This is an important and legitimate point. It is possible, however, to shift one's center of interest from synchronic states as such (and a proto-language is a conjectured synchronic state) to the process of change from one state to a chronologically later one in consonance with de Saussure's fruitful notion of the diachronic as the succession of synchronic states. In such an approach, the domain set for typology becomes not a set of languages, but specific linguistic changes whether these changes are historically connected or not. Moreover, just as synchronic typology is connected with synchronic universals via the

186

JOSEPH H. GREENBERG

empirically discovered limitations to the logically possible types, so diachronic typology turns out to be connected to diachronic universals by the empirical verification of limitations on the logically possible transitions among states. In order to illustrate more concretely the forms that these possibilities may take, it will be convenient to distinguish four methods whose differences are chiefly pragmatic in that they vary to some extent in aim and in the basis for selection of relevant languages. These methods may be called the dynamicization of typologies, the dynamicization of subtypologies, the intragenetic and the intergenetic method. Each of these will be briefly characterized and illustrated. 34 In the dynamicization of typologies, we take as our point of departure a typology theoretically applicable to all the languages of the world. The n classes of languages defined by such a typology may be called language states. The question raised by dynamicization is the existence and manner of the n («-1) logically possible transitions from any state to any other state in either direction. We may call this a state-process model. Since ex hypothesi all languages of the world are covered by the typology, strictly speaking all languages are relevant and in practice sampling becomes necessary. However, limitations on a basis other than random sampling are possible by the selection of particular kinds of typologies which require a far smaller relevant sample for studies of change of type or by limiting the study to a transition of the typology. An example is the study of vowel systems. Since all languages have vowels, a typology based on vowel systems will embrace all of the world's languages. Of course, many typologies of vowel systems are possible and each defines state and hence process differently. Suppose we choose a very simple typology such as that defined by the presence or absence of oral and nasal vowels respectively. There will be only two empirically found states of the four theoretically possible ones, namely, languages with oral and without nasal vowels, and those with both oral and nasal vowels. There are no languages with only nasal vowels and none without at least one of the two types, since in that case it would have no vowels at all. There is clearly a hierarchical relation of marking here in that nasalized vowels are the marked category whose presence implies oral vowels. This means that it will be natural to view the two logically possible transition states in terms of the acquisition and loss of the marked feature of nasality. This also has the consequence that a large number of the world's languages, even though all are included in the overall typology, may be considered irrelevant to the investigation, namely, those which by internal or comparative evidence do not appear to have acquired or lost nasality recently. Another way in which the relevant sample can be reduced to more manageable proportions is to single out a particular subtype within a complex typology and investigate transitions to and from this type. The connection between the dynamicizations of the typologies and diachronic universals can be illustrated by a consideration of one of the topics mentioned previously, the change from the type of languages which has oral but no nasal vowels 81

The following exposition is largely a summary of Greenberg (1969).

THE TYPOLOGICAL METHOD

187

to the one in which both oral and nasal vowels are found. The study of a number of instances in which languages have acquired nasal vowels reveals a typical course of events. A previously oral vowel becomes nondistinctively nasalized by a preceding or following nasal consonant. The nasal consonant, the former conditioning factor, is lost and the oral and nasal vowels are now in contrast. If this can be shown to be the only way in which nasal vowels arise we may call it an exclusive origin theory. This can be restated as a diachronic implicational universal. The existence of nasal vowels in a language implies that at some time in their history, if their origin can be discovered, they developed from the loss of nasal consonants in the environment of nondistinctively nasalized vowels. The sequence of events can be schematized as follows, where sequential symbols are meant to apply in either order: VN->"VN-» V. Such a theory in effect accounts for changes from one type to another by a further elaboration of the original typology. It does this by interposing one or more types as transitional, for example, in this case languages with VN. However, since the original typology was by definition exhaustive, this new type must be either a subtype of one of the original types or cross-cut them. In the present instance, if we define our types phonetically, then languages with allophonic V adjacent to N belong to the type of languages with both oral and nasal vowels. Hence we have subdivided the original type into these languages as against all the rest. In the new typology a structure has been imposed on the range set, which grew out of the diachronic process investigation. This new structure is one of processual priority of the subtype with VN. The existence of such sequentially ordered subtypes is, of course, a strong limitation on the possibilities of type change. Just as in synchronic studies limitation on the set of empirically attested types as compared to the logical possibilities leads to synchronic universals, so the limitation in possible transitions between states leads to diachronic universals. The second method is the dynamicization of subtypologies. In this method we take just one subtype within a typology of universal application. Such a subtype will usually be defined by the presence of a marked feature. For example, it would seem more natural to choose all the languages in the world with numeral classifier systems or glottalized consonants than those without them. By comparison of changes in historically independent examples, whether such changes are deduced from internal reconstruction, the comparative method or direct historical attestation, we seek to construct a dynamic life history of the changes which such a type undergoes from its origination to its disappearance. Unlike the previous method all the languages of the subtype will be relevant. Corresponding to the imposition of further typological subdivisions in the original type in the first method, there will result from this method further subdivisions of the original type and of a more elaborate kind. Moreover, even in the simplest instance it is unlikely that a single set of ordered subtypes will result. The third method, intragenetic comparison, differs in significant ways from the first two methods. Our example is drawn from a genetically defined family or sub-

188

JOSEPH H. GREENBERG

family and the linguistic traits studied often involve sound and meaning simultaneously, unlike the 'normal' situation in typology. This may be illustrated from the history of the Slavic noun declensional system. The vowels I and u (the so-called jers) were always lost in word final and under certain conditions in other positions within the word. The result of this phonetic change was that a fairly large proportion of masculine nouns now had zero in the unmarked category of the nominative and for inanimates in the accusative singular, where on a synchronic basis, zero is very commonly found, but also, for the vast majority of nouns, in the genitive plural, a marked category which usually has an overt mark. However, two declensional classes, one of them very small, the old masculine u stems, and one minor but wellestablished class, the i stems, still had overt endings after the change, endings which may be schematically represented as ov and ej, respectively. There were thus three alternants of the genitive plural: zero ~ ov ~ ej, and we would predict that for a marked category the latter two will gain at the expense of zero. On the other hand, we will expect that the zero of the nominative singular will maintain itself or even spread at the expense of overt affixes in this category. A number of additional hypotheses can be developed, for example that -ov which started with masculine nonpalatalized ('hard') u stems will be acquired by other masculine declensions sooner than by palatalized ('soft') declensions. Such an assumption leads to diachronic implicational hypotheses, for example, that the spread of -ov to the masculine -jo stems (a 'soft declension') implies its previous spread to the masculine -o stems (a 'hard declension'). Such hypotheses, while framed in terms of the specific phenomena of a genetically delimited group, are intended as localized representatives of universal hypotheses applicable to any language family in which the same general conditions are fulfilled. The last method, that of intergenetic comparison, involves the testing of reconstructions not in terms of the typological plausibility of the proto-system as a static entity, but in regard to the plausibility of the dynamic succession of types posited by the same reconstruction. This involves comparison with similar developments in historically independent cases. An instance is the testing of Indo-European laryngeal theory by means of the hypothesized changes of the laryngeals compared to other documented or hypothetical cases. For example, it is assumed in IE that when a laryngeal is lost a preceding vowel is lengthened, but often this does not occur in final position. A comparison with Coptic, which had lost laryngeals found in the earlier stages of the language, shows similar developments in these and other instances. The intergenetic method is, in the final analysis, identical with the first method, the dynamicization of typologies, in that both are studies of change of type involving the comparison of historically independent cases. The difference is a pragmatic one of basic aim. The former seeks to verify and perhaps modify reconstructions in the light of the comparisons undertaken, the latter seeks general principles by the confrontation of independent cases of same change of type. All of the methods under discussion may be considered as the application of the comparative-historical method on a higher

THE TYPOLOGICAL METHOD

189

plane in which comparisons, confined in the traditional manner to single genetic groupings, become single instances within the broader dynamic type. Mr. Reid's paper elsewhere in this volume, on the vowel systems of the Philippine languages, may be considered an application of the intragenetic method. Additional studies carried out in similar care and detail would be required, of course, to arrive at diachronic conclusions of general scope regarding even developments from systems of the postulated a, i, u, 3 type. Detailed discussion of this and similar cases is reserved for a more specialized study. One basic result of this study, however, may be noted. Whenever the original system is reduced, it is always d which merges with some one of the three remaining vowels. The reduced system which results is a, i, u, by far the commonest three-vowel system on a world-wide basis. Similarly, if the four vowels are retained as distinguishable entities, but with some important phonetic change, a phonetic shift of 3 is always involved. This 'instability' of the central or back unrounded vowel has parallels in many other instances in other parts of the world.

REFERENCES 1 9 5 5 . Sprachwissenschaft; Der Gang ihrer Entwicklung von der Antike bis zur Gegenwart. Freiburg/Muenchen, Alber. BACH, EMMON. 1 9 6 5 . On some recurrent types of transformation. M S L L 1 8 . 3 - 1 8 . BAZELL, CHARLES E . 1 9 5 8 . Linguistic typology. London, School of Oriental and African Studies. BENEDICT, R U T H . 1934. Patterns of culture. Boston and New York, Houghton Mifflin. BENES, BRIGIT. 1958. Wilhelm von Humboldt, Jacob Grimm, August Schleicher, ein Vergleich ihrer Sprachauffassungen. Winterthur, P. F. Keller. BENVENISTE, EMILE. 1 9 5 2 - 1 9 5 3 . La classification des langues. Conférences de linguistique de l'Université de Paris, 11. [Reprinted in Problèmes de linguistique générale, pp. 9 9 - 1 1 8 . Paris, Gallimard. 1 9 6 6 . ] BIERWISCH, MANFRED, [n.d.] Strukturelle Linguistik, Geschichte und Methoden. [Mimeographed.] BOAS, FRANZ. 1 9 1 1 . The mind of primitive man. New York, Macmillan. BRONDAL, VIGO. 1 9 3 6 . La structure des systèmes vocaliques. TCLP 6 . 6 2 - 7 4 . . 1938. Les oppositions linguistiques. Proceedings of 11th International Congress of Psychology, Paris, 25-31 July, 1937, pp. 237-40. Paris, Alcan. CHOMSKY, NOAM. 1 9 6 5 . Aspects of the theory of syntax. Cambridge, M I T . COWGILL, WILLIAM. 1963. A search for universals in Indo-European diachronic morphology. Universals of language, ed. by Joseph H. Greenberg, pp. 91-113. Cambridge, MIT. DELBRÜCK, BARTHOLD. 1 9 0 1 . Grundfragen der Sprachforschung. Strassburg, Trübner. ARENS, HANS.

190

JOSEPH H. GREENBERG

DIEBOLD, A. RICHARD. 1965. A

survey

of psycholinguistic research

1954-1964.

Appendix to Psycholinguistics: A survey of theory and research problems, ed. by Charles E. Osgood and Thomas A. Sebeok, 2nd ed. Bloomington, Indiana University. DILTHEY, WALTER. 1883. Einleitung in die Geisteswissenschaften.

[Reprinted as

Vol. 1 of Gesammelte Schriften. Leipzig and Berlin, Teubner. 1922.] FIESEL, EVA. 1929. Die Sprachphilosophie der deutschen Romantik. Tübingen, Mohr. FILLMORE, CHARLES J. 1968. The case for case. Universals in linguistic theory, ed. by Emmon Bach and Robert T. Harms, pp. 1-90. New York, Holt, Rinehart & Winston. FINCK, F.N. 1899. Die deutsche Sprache als Ausdruck deutscher Weltanschauung. Marburg, Elwert. . 1910. Die Haupttypen des Sprachbaues. Leipzig, Teubner. FOLSOM, JOSEPH K. 1931. Social psychology. New York, Harper. FREI, HENRI. 1948. De la linguistique comme science de lois. Lingua 1.25-33. GABELENTZ, GEORG VON DER. 1891. Die Sprachwissenschaft. Leipzig, Weigel.

GRASSERIE, RAOUL DE LA. 1890. De la classification des langues. II. partie: C l a s s i f i cation des langues non-apparent6es. ZAS 5.296-338. GREENBERG, JOSEPH H. 1954. A

quantitative

approach

to

the

morphological

typology of language. Method and perspective in anthropology, ed. by Robert F. Spencer, pp. 192-220. [Reprinted IJAL 26.178-94. (I960).] . 1957. The nature and uses of linguistic typologies. IJAL 23.68-77. . 1963. Some universals of grammar with particular reference to the order of meaningful elements. Universals of language, ed. by Joseph H. Greenberg, pp. 58-90. Cambridge, MIT. . 1969. Some methods of dynamic comparison in linguistics. Substance and structure of language, ed. by Jan Puhvel, pp. 147-203. Berkeley and Los Angeles, University of California. HARTMANN, PETER. 1956. Z u r T y p o l o g i e d e s I n d o g e r m a n i s c h e n . Heidelberg, Winter.

HEMPEL, CARL G. 1952. Typological methods in the social sciences. Science, language and human rights, pp. 65-86. American Philosophical Association, Eastern Division, Philadelphia, University of Pennsylvania. [Reprinted in Philosophy of the social sciences, ed. by Maurice A. Natanson, pp. 210-30. New York, Random H o u s e . (1963).] HEMPEL, CARL G . , a n d PAUL OPPENHEIM.

1936.

D e r Typusbegriff im Lichte

der

neuen Logik. Leiden, Sijthoff. HÖRNE, KIBBY M. 1966. Language typology: 19th and 20th century views. Georgetown University, Institute of Language and Linguistics, Washington, D.C. HOUSEHOLDER, FRED W., JR. 1960. First thoughts of syntactic indices. I J A L 26. 195-97.

HUMBOLDT, WILHELM VON. 1825. Ueber das Entstehen der grammatischen Formen,

THE TYPOLOGICAL METHOD

191

und ihren Einfluss auf die Ideentwicklung. Abhandlungen der hist.-phil. Klasse, pp. 401-30. Berlin, Königliche Akademie der Wissenschaften. . 1836. Ueber die Verschiedenheit des menschlichen Sprachbaues. Vol. I of Ueber die Kawispache auf der Insel Java. Berlin, Königliche Akademie der Wissenschaften. . 1903-1920. Gesammelte Werke. 15 Vols. Berlin, Königliche Preussische Akademie der Wissenschaften. HYMES, DELL H . 1 9 6 1 . On typology of cognitive styles in language. AnL 1 . 2 2 - 5 1 . JAKOBSON, ROMAN. 1 9 4 1 . Kindersprache, Aphasie und allgemeine Lautgesetze. Uppsala. . 1953. Remarks. An appraisal of anthropology today, ed. by Sol Tax, Loren E. Eiseley, Irving Rouse and Carl Voegelin, pp. 312-13. Chicago, University of Chicago. . 1958. Typological studies and their contribution to historical comparative linguistics. PICL 8.17-25. KATZ, JERROLD J . and PAUL M. POSTAL. 1964. An integrated theory of linguistic descriptions. Cambridge, MIT. KLUCKHOHN, CLYDE, and HENRY A . MURRAY, eds. 1 9 4 9 . Personality in nature, society and culture. New York, Knopf. KLÜVER, HEINRICH. 1 9 2 4 . The determination of types with an experimental study of the eidetic type. Doctoral dissertation, Stanford University. Konferencija. 1968. Konferencija po problemam izuöenija universal'nyx i areal'nyx svojstv jazykov. Moscow, izd. Nauka. KRÂMSKY, JIRÎ. 1946-48. Fonologické vyuziti samohlâskovych fonemât. Linguistica Slovaca IV-VI, 39-43. . 1959. A quantitative typology of languages. L&S 2.72-85. KRIEGER, ALEX. 1 9 4 4 . The typological concept. American Antiquity 9 . 2 7 1 - 8 7 . Washington, D.C. KROEBER, ALFRED L . 1 9 6 0 . On typological indices I : Ranking of languages. I J A L 26.171-77. K R U P A , VICTOR. 1 9 6 5 . On quantification of typology. Linguistics 1 2 . 3 1 - 3 6 . K R U P A , VICTOR, and GABRIEL ALTMANN. 1 9 6 6 . Relations between typological indices. Linguistics 2 4 . 2 9 - 3 7 . LAZARSFELD, PAUL F. 1937. Some remarks on the typological procedures in social research. Zeitschrift fuer Sozialforschung 6.119-39. LEKOMTSEVA, MARGARITA I. 1 9 6 8 . Tipologija struktur sloga v slavjanskichjazykach. Moscow, Izd Nauka. LEROY, MAURICE. 1 9 6 3 . Les grands courants de la linguistique moderne. Brussels and Paris, Presses Universitaires. LEWY, ERNST. 1 9 4 2 . Der Bau der europäischen Sprachen. Proceedings of the Royal Irish Academy, Volume 48, Section C, nr. 2. Dublin. MAROUZEAU, JULES. 1 9 3 1 . Lexique de la terminologie linguistique. Paris, Geuthner. MATHESIUS, VILÉM. 1 9 2 8 . On linguistic characterology with illustrations from

192

JOSEPH H. GREENBERG

Modern English. Actes du 1er Congrès International des linguistes, Leiden, pp. 1990.56-63. M C C A W L E Y , JAMES D . 1 9 7 0 . English as a V S O language. Lg 4 6 . 2 8 6 - 9 9 . MEILLET, ANTOINE. 1924. Introduction. Les langues du monde, ed. by Antoine Meillet and Marcel Cohen, 1st edition. Paris, Champion. MEINHOF, CARL. 1936. Die Entstehung flektierender Sprachen. Berlin, Reimer. MEINECKE, FRIEDRICH. 1 9 3 6 . Die Entstehung des Historismus. Bd. I . Munich. MENZERATH, P A U L , and WERNER MEYER-EPPLER. 1 9 5 0 . Sprachtypologische Untersuchungen I Teil: Allgemeine Einführung und Theorie der Wortbildung. SL 4.54-60.

ed. 1 9 6 3 . Issledovanija po strukturnoj tipologii. Moscow, izd. Akademia nauk SSR. M U C H O W , MARTA. 1931. Zur Frage einer Lebensraum- und epochaltypologischen Entwicklungspsychologie des Kindes- und Jugendalters. Festschrift William Stern, pp. 185-202. Leipzig. MÜLLER, MAX. 1871-1872. Lectures in the science of language. 2nd revised edition, 1890, New York, Scribner. OSGOOD, CHARLES E., GEORGE SUCI, and PERCY TANNENBAUM. 1 9 5 7 . The measurement of meaning. Urbana, University of Illinois. PEDERSON, HOLGER. 1 9 2 4 . Sprogvidenskaben: det nittende Aarhundert: Metoder og Resultater. Copenhagen, Gyldendalske Bokhandel. Translated by J.W. Spargo as The discovery of language, linguistic science in the nineteenth century. Bloomington, Indiana University ( 1 9 6 2 ) . PIAGET, JEAN, and PEDRO ROSSELLÔ. 1 9 2 3 . Notes sur les types de descriptions d'images chez l'enfant. Archives de psychologie 18.208-34. Geneva. PIERCE, JOE E . 1 9 6 2 . Possible electronic computation of typological indices for linguistic structures. I J A L 2 8 . 2 1 5 - 2 6 . . 1963. Typological indices for written vs. spoken Turkish. IJAL 29.71-2. . 1966. Sampling and typological indices in language. Linguistics 24.43-50. ROBINS, R . H. 1967. A short history of linguistics. London. Ross, JOHN R. 1 9 6 7 . Gapping and the order of constituents. PEGS Paper No. 8. Washington, D.C., Center for Applied Linguistics. SAPIR, E D W A R D . 1 9 2 1 . Language: An introduction to the study of speech. New York, Harcourt, Brace and World. . 1917. Review of Het passieve karakter der possessieve flexie in talen van Noord-Amerika, by C.C. Uhlenbeck. IJAL 1.86-90. SAYCE, ARCHIBALD H. 1879. Introduction to the science of language. 2 vols. London, Paul. SCHLEGEL, AUGUST WILHELM VON. 1 8 1 8 . Observations sur la langue et la littérature provençales. Paris. SCHLEGEL, K A R L WILHELM FRIEDRICH VON. 1 8 0 8 . Ueber die Sprache und Weisheit der Indier. Heidelberg, Mohr and Zimmer. MOLOSHNAJA, T . N . ,

THE TYPOLOGICAL METHOD

193

1859. Die Sprachen Europas in systematischer Uebersicht. Bonn, König. . 1859. Zur Morphologie der Sprache. Mémoires de l'Académie Impériale des Sciences de St. Pétersbourg, 7e Série, Tome 1, No. 7. SEITERICH, EDUARD. 1930. Die logische Struktur des Typusbegriffs bei William Stern, Eduard Spranger und Max Weber. Inaug. Diss. Freiburg. SKALIÖKA, VLADIMÎR. 1941. Vyvoj ôeské deklinace; Studie typologickâ. Studie Prazského linguistického krouzku, 4. Prague. . 1946a. Ueber die Typologie der Bantusprachen. AO 15.93-127. . 1946b. Sur la typologie de la langue chinoise parlée. AO 15. 386-411. . 1958. O souôasném stavu typologie. SS 19.224-32. STEINTHAL, CHAJIM. 1860. Charakteristik der hauptsächlichsten Typen des Sprachbaues. Berlin, Dümmler. STEWART, WILLIAM A. 1962. An outline of linguistic typology for describing multilingualism. Study of the role of second languages in Asia, Africa and Latin America, pp. 15-25. Washington, D.C., Center for Applied Linguistics. Thèses. 1929. Thèses présentées au Premier Congrès des philologues slaves. TCLP 1.5-29. Theses. 1957. Theses for the circle of modern philologists. B. Trnka, J. Yachek and others. VJa 1957.44-52. TRNKA, BOHUSLAV. 1929. Méthode de comparaison analytique et grammaire comparée historique. TCLP 1.33-8. [Reprinted in A Prague school reader, ed. by Joseph Vachek, pp. 68-74. Bloomington, Indiana University, (1964)]. TRUBETZKOY, NIKOLAI. 1929. Zur allgemeinen Theorie der phonologischen Vokalsysteme. TCLP 1 . 3 9 - 6 7 . . 1939. Gedanken ueber das Indogermanenproblem AL 1.81-9. USPENSKIJ, B. A . 1962. Principy strukturnoj tipologii. Moscow, Izd. Moskovskogo Universiteta. VACHEK, JOSEPH, and JOSEPH DUBSKY. 1960. Dictionnaire de linguistique de l'Ecole de Prague. Utrecht, Het Spectrum. VOEGELIN, CARL F. 1956. Subsystems within systems in cultural and linguistic typologies. For Roman Jakobson, pp. 5 9 7 - 9 9 . The Hague, Mouton. . 1962. Methods for typologizing directly and by distinctive features (in reference to Uto-Aztecan and Kiowa-Tanoan Vowel systems.) Lingua 11.469-87. VOEGELIN, CARL F . and JOHN YEGERLEHNER. 1956. The scope of whole system ('distinctive feature') and subsystem typologies. Word 12.444-53. WHITNEY, WILLIAM D. 1870. Language and the study of language. New York, Scribner. WINCH, R . F . 1947. Heuristic and empirical typologies: A job for factor analysis. American Sociological Review 1 2 . 6 8 - 7 5 . WUNDT, WILHELM M . 1900. Völkerpsychologie I I . Die Sprache. Leipzig, Engelmann

SCHLEICHER, AUGUST.

T H E SOCIAL S E T T I N G OF L I N G U I S T I C C H A N G E

WILLIAM LABOV

The study of language change in its social context has been described by some as a virgin field; by others, as a barren territory. A brief examination of what has been written in the past on this subject shows that it is more like an abandoned back yard, overgrown with various kinds of tangled, secondary scholarship. The subject has been so badly treated with voluminous, vacuous and misleading essays that one can sympathize with linguists who say that it is better left alone. But the consequences of avoiding the social dimension of language change are serious. We are then left with such a limited body of fact that we are condemned to repeat the arguments of our predecessors; we find ourselves disputing endlessly about bad data instead of profiting from the rich production of new linguistic change around us. Nevertheless, historical linguists have adopted and vigorously defended a thoroughly asocial policy in the past half-century. To understand why, it will be helpful to re-examine briefly the history of the relations between linguists and society, and how they have succeeded in avoiding each other. This review will raise three questions: whether the expressive and directive functions of language are important determinants of change; whether highly abstract rules of grammar can be affected by social forces; and whether linguistic evolution is entirely dysfunctional. This chapter may be considered the diachronic correlate of "The study of language in its social context" (Labov 1970) which was concerned almost entirely with the synchronic aspects of the first two questions. Data from a number of recent studies of change in progress will be presented to show how linguistic changes may be embedded in a social context, how they are evaluated, and how change may be activated at a particular time and place. These findings will then be applied to the three questions raised above.

1.

THE LINGUISTIC VIEW OF LANGUAGE AS A SOCIAL FACT

Every linguist recognizes that language is a social fact, but not everyone puts an equal emphasis on that fact. When linguists write about language change, we find a very different degree of concern with the social context in which these changes occur. Some broaden their view to include a wide range of facts about the speakers and their extra-linguistic behavior, while others constrict their view to exclude as much as

196

WILLIAM LABOV

possible. We can generally predict from an author's definition of language how much he will be concerned with the social factors in linguistic change. His view of the functions of language will also be a factor: those who focus upon the communication of cognitive or referential information will fall on the individual side, those who become involved with affective and phatic uses of language will deal more with social matters. It is not difficult to find nineteenth century spokesmen for the importance of the social factors in linguistic change. Whitney was firmly committed to the social side of the question: Speech is not a personal possession, but a social; it belongs, not to the individual, but to the member of society (1901:404). In considering the functions of language, Whitney emphasized communicative function in a social sense, rather than the primacy of 'ideas': Man speaks, then, primarily not in order to think, but in order to impart his thought. His social needs, his social instincts, force him to expression (p. 401). For an opposing view, we can look to Hermann Paul, whose individualist approach is well reflected in most current theories of language change (Weinreich, Labov and Herzog 1968). Paul saw the language of the community as a rough mixture of the well-formed speech of individuals. On this basis, he dismisses as quite transparent the problem of explaining the diversification of language: If we start from the undeniable truth that each individual has his or her own language, and that each such language has its own history ... the rise of variations seems a mere matter of course [die Entstehung der Verschiedenheit scheint ja danach selbstverständlich] (Paul 1889:23=1886:37). For Paul, the function of language was to organize 'groups of ideas [ V o r s t e l l u n g s gruppen]', a process which 'takes a peculiar development in the case of each individual'(1889:6 = 1886:22). Sweet studied Paul, and absorbed and endorsed this point of view; he cautions that all general principles of change are subordinate 'to the main function of language ... the expression of ideas' (1900:34). We should not be surprised, then, that Sweet defines language without any reference to social context, as 'the expression of thought by means of speech-sounds'(1900:1). His explanations of language change accordingly revolve about such individual traits as 'laziness' or 'carelessness'. The emphasis on cognitive or representational functions of language was maintained by the Prague school in their synchronic studies. Other functions were of course recognized: following Bühler and Laziczius, Trubetzkoy would set up three divisions of phonology: the expressive, appellative, and representational, or phonology proper. The net effect of this division was to free the linguist from any concern with social factors and non-cognitive functions. After devoting a few pages to an anecdotal account of these matters, the linguist could proceed with the real business at hand (1957:16-17). Martinet (see below) seems to be a direct descendent of this tradition by his manner of dealing with linguistic change.

THE SOCIAL SETTING OF LINGUISTIC CHANGE

197

Bloomfield inherited from Paul the same individualist psychology, though he did object to its subjective character (1933). Bloomfield's S-R model shows language as the property of the individual; his model of sound change imagines a perfectly regular but unobservable process taking place within the individual's speech pattern. Larger social factors are presented as relatively vague and confused processes, under the heading of 'Fluctuation in the frequency of forms', or 'Dialect borrowing'. Chomsky and Halle, who differ from Bloomfield and Paul on so many other points, continue the tradition of speculation on the basis of individual models of the speakerhearer relationship. Chomsky would deliberately exclude all social variation from the subject matter of linguistics (1965:3); Halle (1962) presents a model of linguistic change in which the individual child restructures his parents' speech. Although Paul's individualist views have been followed in the main stream of linguistics, there has been considerable opposition from many who shared Whitney's concern with the social context of language and its wider range of social functions. Witness the well-known position of Meillet: From the fact that language is a social institution, it follows that linguistics is a social science, and the only variable to which we can turn to account for linguistic change is social change, of which linguistic variations are only consequences (1921:16-17).

Meillet's associate Vendryes continues in the same vein, some twenty years later: Language is thus the social fact par excellence, the result of social contact. It has become one of the strongest bonds uniting societies, and it owes its development to the existence of the social group (1951).

Jespersen followed Sweet on many matters, but he laid heavy emphasis upon the role of language in social interaction: The language of a nation is the set of habits by which the members of the nation are accustomed to communicate with one another (1946:21).

Although Jespersen was deeply involved in his 'notional' theory of grammar, we find throughout his writings a concern with the expressive and directive functions of language, which enters strongly into his discussions of linguistic change. Sturtevant wrote in the same tradition: A language is a system of arbitrary vocal symbols by which members of a social group cooperate and interact (1947:2).

Sturtevant's proposals for the mechanism of linguistic change (1947:74ff.) laid a great deal of emphasis on the assignment of social and affective values. It is clear that we shall not understand the regularity of phonetic laws until we learn how rivalry between phonemes leads to the victory of one of them .... Before a phoneme can spread from word to word ... it is necessary that one of the two rivals shall acquire some sort of prestige (1947:80-1).

But Sturtevant represented a late survival of Meillet's fading notion that we might search for an explanation of the fluctuations of linguistic change in the fluctuating

198

WILLIAM LABOV

course of social events. The predominant trend is expressed by many eminent linguists who fiercely resist any such involvement, and insist that we must confine ourselves to purely internal, linguistic explanations. Martinet, for example, declares that it is only the results of outside impact that the linguist is competent to study. In his capacity as a linguist, he will decline to investigate sociological conditioning (1964a: 52). Kurylowicz takes an even harder line : Once we leave language sensu stricto and appeal to extra-linguistic factors, a clear delimitation of the field of linguistic research is lost. Thus, e.g., the physiological (articulatory) aspect may be a consequence of social factors, the latter being themselves due to certain political or economic facts (conquest, migrations involving bilingualism). ... It seems that the field of linguistic explanation in the literal sense must be circumscribed by the linguistic aspect of the change in question, i.e., by the actual state of the system before and after the change ("l'état momentanée des termes du système" — Saussure) (1964:11). Kurylowicz wishes to purify linguistic argument of all contaminating support; he renounces the use of dialect geography, phonetics, psychology and cultural anthropology in the reconstruction of the history of a language, in order to rise to a 'higher conceptual basis' (1964:30). Thus linguists seem to fall into two major groups with regard to their view of the relation of social factors to language and language change. Group A, the 'social' group, would pay close attention to such factors in explaining change ; see expressive and directive functions of language closely intertwined with the communication of referential information; would study change in progress and see on-going change reflected in dialect maps; and emphasize the importance of linguistic diversity, languages in contact, and the wave model of linguistic evolution. Linguists of Group B, the 'asocial' group, focus upon purely internal — structural or psychological — factors in explaining change; segregate affective or social communication from the communication of 'ideas' ; believe that sound change in progress cannot be studied directly, and that community studies or dialect maps show nothing but the results of dialect borrowing; and take the homogeneous, monolingual community as typical, working within the Stammbaum model of linguistic evolution. Without doing too much violence to individual views, we can line up in Group A such linguists as Whitney, Meillet, Vendryes, Jespersen and Sturtevant. In Group B we would find Paul, Sweet, Trubetzkoy, Bloomfield, Hockett, Martinet, Kurylowicz, Chomsky and Halle. It is not the case that Group B linguists would disregard social factors entirely in explaining linguistic change. Rather they define the influence of society as alien to the normal operation of language, and view the operation of social factors as dysfunctional interference with normal development (Bloomfield 1933), or as rare and unsystematic interventions. Thus Martinet develops what we may call a 'catastrophic' view of the relations of social and linguistic events. He argues that extraordinary social upheavals can disturb the linguistic equilibrium at rare intervals, setting off a

THE SOCIAL SETTING OF LINGUISTIC CHANGE

199

wave of linguistic re-adjustments in which purely internal factors govern the succession of changes over 'years, centuries, and millenia' (1964:522). Thus the influence of Norman French on English in the twelfth and thirteenth century had profound effects which are still being felt today through a long series of internally linked re-adjustments. Within their own perspective, Chomsky and Halle (1968) share this point of view. They argue that the underlying forms of English have undergone very little change since Middle English, and the last serious change in the system was probably the substitution of the Romance stress rule for the Germanic stress rule as a result of the Norman invasion. Thus, there are areas of general agreement about the effects of certain violent social changes upon language. No one would deny the importance of conquests, invasions, and massive immigration, with consequent extinction, superposition or merger of whole languages. We can distinguish three sub-types, following Lehmann (1962): (1) an invasion in which the language of the conquered people all but disappears, as with the Celtic of Britain; (2) a conquest in which the conquerers eventually adopt the language of the conquered, with consequent large-scale alteration in a class-stratified vocabulary, as with the Norman hegemony; (3) an invasion which results in an intimate mingling of two populations, with intimate borrowing of vocabulary and even function words, as with the Scandinavian invasions of England. It would be important to specify, if possible, under which conditions one could expect which of these outcomes, but the problem seems to be an historical and political one. In considering the relation of linguistics to social setting, we must take into account one important field where there has never been any doubt about the importance of the social context: the study of pidgins and Creoles. From the time of Schuchardt, Creolists have found it necessary to learn as much as possible about the social changes in which these languages were formed and reformed, and many have brought this knowledge into intimate contact with the linguistic data (Hymes 1971). Valdman's treatment of "Decreolization in Creole French" in this volume is an account of an important linguistic process which is plainly conditioned, almost from moment to moment, by the course of social change. We have good reason to think that a similar process took place in the case of Black English in the United States (Stewart 1967) but we have not been able to observe the historical process in any detail. Many of the systematic processes of language change that we would like to trace in normal linguistic evolution can be observed in accelerated form in Creoles. In the French Creole described by Valdman we observe the vocalization of postvocalic liquids, the loss of other finals, the condensation and loss of function words, the unrounding of front rounded vowels. These processes seem to be the result of variable rules (such as we now observe in modern colloquial French) which went to completion in Creole. Somewhat less regular are the results of re-analysis of articles and partitive elements preceding the noun. Chien incorporates the article, so that 'my dog' is mo listen and so does table: 'on the table' is en latab; but 'mattress' is still simple matla. Mass nouns tend to incorporate de by re-analysis, so we have dilo 'water' and diri 'rice' but

200

WILLIAM LABOV

sik for 'sugar'. The re-analysis of the liaison element into the singular stem of the noun gives use forms such as zozo 'bird', but sometimes this z continues to function as a plural element: kat zanfan, on anfan. These processes affect lexical items one by one, sometimes changing their base form, sometimes assigning to them the irregular and unpredictable use of a rule. One of the best studies of systematic morphological change is that of Bickerton (1971), who shows the rules for complementizer placement gradually shifting through age levels in the English-based creole of Guyana, with tu replacing fu in an implicational series of verb complements. Until recently, Creole studies did not provide us with data on language in actual use, as Valdman indicates. They have largely focused on the place of Creole in the schools, in traditional folklore and literature, and in government policy. But the work of Bickerton and current sociolinguistic work in Hawaii are indications that Creole studies will give us the best view of systematic grammatical change in its social context. Creolization and decreolization are responses to the kinds of ordinary, every-day social interaction which is the major concern of this chapter. Our focus will be on the more systematic areas of grammar and phonology : to what extent are they sensitive to the social and stylistic stratification of speech, and the expressive information conveyed by social and stylistic variation? To what extent do we have to take these factors into account in order to understand the steady process of grammatical and phonological change? So far, we have not yet placed in this dichotomy the most influential linguist of the twentieth century, Meillet's teacher and associate, Saussure. At first glance, Saussure's definition of langue seems to place him squarely in Group A : la partie sociale du langage, extérieure à l'individu ... elle n'existe qu'en vertu d'une sorte de contrat passé entre les membres de la communauté (1962:31). Since Saussure is said to be the most influential linguist of the century, and Meillet is generally held to be one of the most eminent in historical linguistics, and Jespersen is currently read and cited with the greatest attention, it is not at all obvious why Group A should not be the dominant element in twentieth century linguistics. In 1905, Meillet predicted that this century would be devoted to isolating the causes of language change within the social matrix in which language is embedded. But this did not happen. In fact, almost no empirical studies of language change in its social context were carried out in the fifty years following Meillet's pronouncement. It is clearly Group B that is the dominant factor in current linguistic theory and practice, and in practice most linguists agree with Chomsky in taking as the object of linguistic description the 'ideal speaker-listener in a completely homogeneous speech community, etc.' (1965:3). It seems important that before proceeding further, we attempt to grasp the reasons for this outcome. Without good evidence to the contrary, the great achievements and the good authority of the linguists cited under Group B would argue for the correct-

THE SOCIAL SETTING OF LINGUISTIC CHANGE

201

ness of their asocial view. We can isolate four general conditions which favor the predominance of Group B outlook in recent decades, which have to do with the climate of thought rather than the substantive issues involved. 1. The first elements in the success of Group B involves the SAUSSURIAN PARADOX. Saussure argues that langue is a social fact, knowledge possessed by practically every member of the speech community. It follows that one can find out about langue by questioning any one or two speakers of the language — even oneself. On the other hand, parole reveals individual differences among speakers that can be examined only in the field, by a kind of sociological survey. Thus the social aspect of language can be studied in the privacy of one's own office, while the individual aspect would require social research in the heart of the speech community. The Saussurian paradox explains how Bloomfield could analyze the English 'spoken in Chicago' from knowledge of his own speech (1933:90-2). The popularity of Saussure's langue/parole dichotomy was further assured when it was transformed into Chomsky's competence/performance distinction. Both treatments illustrate how linguists can adapt their methodology to suit their personal style of work without abandoning principle. There is no doubt that self-examination is a congenial methodology for most linguists. 2. Linguists have traditionally leaned to the side of psychology rather than sociology. The road between language and thought is a well-traveled one; psychologists of language have always appeared prominently in the linguistic literature, from Wundt and Paul to Buhler and Piaget. On the other hand, Durkheim's influence on Meillet seems to have been a matter of historical accident which has not been repeated. Thus nothing could be more natural for linguists than to explain language change in terms of the parent-child relationship. To explain the acquisition of language, one need only consider the mother as speaker and the child as hearer; to understand linguistic change, one considers the child first as hearer, then as speaker. Linguists are always performing 'thought-experiments' in which they put themselves in the position of an imaginary child grappling with fictional data from an imaginary mother. This image follows naturally from the small amount of data that we have on children's speech, traditionally derived by parents studying their own children. More recent psycholinguistic experimentation confronts the child with the test situation ; just as the linguist traditionally summons the informant to a formal elicitation session, so the psychologist of language summons the child to the laboratory to wrestle with blocks or matrices, searching for meaning in answers to meaningless questions. Since no one follows the child to observe his day-to-day interaction with other members of society, it would be strange if our explanations of his linguistic behavior took such social factors into account. 3. In the latter half of the nineteenth century, historical linguists were quite open to the influence of dialectology. As one striking example, one can cite the impact on Osthoff and Brugmann of Winteler's monograph on the Swiss German dialect of Kerenzen (Weinreich, Labov and Herzog 1968:115). But in the twentieth century,

202

WILLIAM LABOV

dialectology as an institution seems to have lost any orientation towards theoretical linguistics, and dialect geographers have generally been content to collect their materials and publish them. Bloomfield's chapter dealing with this subject is strangely disjunct f r o m his other chapters; as other have noted (Sommerfelt 1930), the most solid a n d startling results of dialectology, such as those of G a u c h a t (1905), could not be fitted into the f r a m e w o r k of neogrammarian thinking. It was not until the 1950s t h a t the work of Martinet (1955), Moulton (1962), a n d Weinreich (1954) demonstrated again the theoretical impact of areal linguistics. 4. Perhaps the most i m p o r t a n t reason for the eclipse of the social g r o u p of linguistics is the limitations of their own work and writings on the social context of language. They relied almost entirely u p o n an intuitive explanation of a few anecdotal events drawn f r o m their own general knowledge. W h e n we read the comments of Whitney, Meillet, Jespersen or Sturtevant we c a n n o t argue that any of these authors knew m o r e a b o u t society's impact on language than anyone else; they were simply willing to talk a b o u t it more. Typical of Whitney's arguments a b o u t the social nature of language is one that he raises to prove that 'external circumstances' are the most i m p o r t a n t factor in linguistic change: While a Swiss Family Robinson keep up their language, and enrich it with names for the new and strange places and products with which their novel circumstances bring them in contact, a Robinson Crusoe almost loses his for lack of a companion with whom to employ it (1901:405). W e encounter hundreds of 'thought-experiments' in these writers, taking the place of actual data. Again, Whitney says: Let two children grow up together, wholly untaught to speak, and they would inevitably devise, step by step, some means of expression for the purpose of communication; how rudimentary, of what slow growth, we cannot tell ... (p. 404). The 'cast-away' thought-experiment occurs over and over: Suppose an illiterate English family to be cast away upon a coral islet in the Pacific and to be left there isolated through a succession of generations. How much of our language would at once begin to become useless to them! (p. 138). W e find to our dismay t h a t Whitney lives in a world o f ' f a c t s ' which are obvious to him but not to us, the kind of commonsense experience which has never been questioned: The fact of variation in the rate of linguistic growth is a very obvious one (p. 137). The mode of argument which the twentieth century inherited to deal with social matters is remarkably similar. We find most often a series of anecdotes which are designed to prove ideas already accepted as true. T h u s Vendryes, expounding the Saussurian notion of the uniformity of langue, reports that We know what misadventure befell Theophrastus of Lesbos in the market at Athens. When he asked the price of some commodity, a woman of the people recognized him as a stranger by his speech (1951:240).

THE SOCIAL SETTING OF LINGUISTIC CHANGE

203

With distressing regularity we raise the familiar desert islands, populated with the cast-aways of thought-experiments: When a solitary French man meets a Persian on a desert island, each will forget the differences between them and they will naturally seek to make common cause (p. 239). We need only look up the word 'social' in the index of any of the authors cited above — in Group A or Group B — to encounter more thought-experiments and more anecdotes. Thus Bloomfield explains the diversification of language by imaginary differences in density of communication, discovered in an elaborate thought-experiment which charts every utterance of every speaker in a community (1933:46). Since the experiment cannot be performed, Bloomfield admits that he is 'forced to resort to hypothesis'. The hypothesis is then enriched with a further hypothesis about linguistic change — that the relative prestige of speaker and hearer determine borrowing and the fluctuation of forms. The two hypotheses — both stated without evidence — are then combined in another thought-experiment which would provide a final accounting for the direction and rate of linguistic change: If we had a diagram with the arrows thus weighted [by gradations representing the prestige of the speaker with reference to each hearer] we could doubtless predict, to a large extent, the future frequencies of linguistic forms (1933:403). It is difficult to believe that Bloomfield, with his fine sense of evidence, would give much weight to such arguments, even his own. In following the traditional mode of dealing with social facts, he was also arguing that this field lies permanently outside the domain of the linguist. Although some of our most precise linguists are still prone to thought-experiments and anecdote, they tend to limit these to fields outside their normal linguistic activity. They are certainly sensitive to such vacuity on the part of others. To understand the strong Group B position taken by Martinet, Kurylowicz and Chomsky, we have to know what they were reacting against. Arguments about race and climate in the earlier literature, which strike us today as too absurd to take seriously,1 are the kind of 'external' arguments that Martinet and Kurylowicz were denouncing.

2.

THREE SUBSTANTIVE QUESTIONS O N LINGUISTIC CHANGE

The preceding section set out some of the historical and strategic considerations which influenced the decision of linguists to resist being drawn into a study of the social setting of change. But there is also a solid basis for their position in the substantive issues involved. If we decide to engage in a study of the social setting of linguistic 1 Even Sweet had a favorite argument about the effect of climate on language that he returns to over and over again. H e attributes the Germanic raising of IE long a to o to the fact that speakers in the northern climate tried to keep the cold, damp air out of their mouths by not opening their mouths as wide as Mediterranean speakers (1900).

204

WILLIAM LABOV

change, we will be confronted with three serious questions that must eventually be resolved. (A) The place of social variation : Does the social and stylistic variation of language play an important role in linguistic change? By 'social', I mean those language traits which characterize various subgroups in a heterogeneous society; and by 'stylistic' the shifts by which a speaker adapts his language to the immediate context of the speech act. Both of these are included in 'expressive' behavior — the way in which the speaker tells the listener something about himself and his state of mind in addition to giving representational information about the world. 2 Social and stylistic variation presuppose the option of saying 'the same thing' in several different ways: that is, the variants are identical in referential or truth value, but opposed in their social and/or stylistic significance. Martinet takes a strong negative position on the question of social variation. Under the heading, 'Communication alone shapes language', he argues: It is therefore the communicative uses of language which must engage our attention if we wish to discover the causation of linguistic changes. What we shall establish and be able to formulate will not necessarily hold good for those linguistic utterances which do not serve the purpose of communication. But we shall be disposed to disregard these since they are modelled on communicative utterances and so offer nothing which will not be found in these (1964b: 170). Of course the term 'communication' could include all kinds of expressive communication, but a reading of the wider context makes it clear that it is intended to exclude any information about the speaker contained in the linguistic form, as well as excluding 'phatic' expression. The non-representational information we are considering under this question is usually delivered simultaneously with other messages, so it is a question of neglecting one aspect and following others. To simplify our analysis, we shall assume that the language in process of evolution is that of a strictly monoglot community, perfectly homogeneous in the sense that observable differences represent successive stages of the same usage and not concurrent usages ... we must disregard these [social and geographical] variations as we did above in the case of descriptive linguistics (1964b: 164). Martinet's position allows us to specify the phrase 'play an important role'. If the answer to the original question (A) stated above is positive, then the Chomsky-Martinet strategy will be defective, and produce vacuous or erroneous statements about linguistic change and its causes. In the body of this discussion, I will assemble evidence to show that this is the case. (B) The level of abstraction : Can high-level, abstract rules of phonology and grammar be affected by social factors? One of the factors that has led to the decline of the Group A approach is that 2

This expressive information can of course be given overtly, as in 'I feel quite relaxed with you', and in that case the reference or representation may be identical with information normally given by supersegmental or non-distinctive alternants. But every utterance can also carry further expressive information which is distinct from 'what is said' overtly, and can therefore be denied.

THE SOCIAL SETTING OF LINGUISTIC CHANGE

205

linguists have steadily shifted their attention away from lexicon, phonetics and inflectional morphology to higher-level rules of abstract phonology and syntax. By 'higher-level' I mean rules higher up in the ordering sequence, which, if changed, would affect the output of many other rules, and containing more abstract information. We are increasingly aware that most rules of grammar are quite remote from conscious awareness. It therefore raises a major conceptual problem to say that social factors bear heavily on the systematic development of a language, since speakers are not even vaguely aware of most of the deeper relations. How are such social factors brought to bear upon the process of language learning? Members of Group A tended to focus strongly upon the role of the word as receiving and reflecting social influences. The contention of dialect geographers that every word has its own history (Meillet 1964:29) fits in with this orientation. At the same time, it was recognized that grammatical particles are less subject to borrowing, more stable in the face of outside impact upon language. This would seem even more true for the rules that relate surface structure to underlying forms. Even if social factors should alter profoundly the phonetics and vocabulary of a language, and possibly the surface formatives as well, we might still argue that linguistic change in higher level rules is purely an internal readjustment, not even remotely related to the immediate social context. We will consider this question in relation to some complex phonological rules; the answer is not at all obvious, but it can be seen to be located on the side of doubt. (C) The function of diversity: Is there any adaptive function to linguistic diversification? Throughout the nineteenth century, there were many analogies drawn between linguistic and biological evolution, not least by Darwin himself who saw the two as 'curiously parallel'. He observed in language and in biological species homologies of form due to community of descent, and he compared the generalization of phonetic change to correlated growth in plants and animals. Languages and species both show the reduplication of parts, the effects of long-continued use, vestigial elements, hierarchical grouping, typological vs. genetic taxonomies, dominance and extinction, hybridization, and variability (1871:465). But Darwin felt it necessary to complete the analogy by arguing for the survival of the fittest in language, quoting as his authority Max Müller: 'A struggle for life is constantly going on amongst the words and grammatical forms in each language. The better, the shorter, the easier forms are constantly gaining the upper hand, and they owe their success to their own inherent virtue.' No linguists would endorse this point of view today, which runs exactly counter to our notion of the arbitrariness of the linguistic sign. Languages do not seem to be getting better and better, and we see no evidence for progress in linguistic evolution (Greenberg 1959). Except for the development of vocabulary, we cannot argue for adaptive radiation in any area of language. The diversification of languages is not

206

WILLIAM LABOV

immediately and obviously functional, as the diversification of species may be. We receive no immediate benefit from not being able to understand the Russians or the Gaels, and the time taken to learn their languages does not seem to help in the survival of our own. We must seriously consider the possibility that the diversification of language is dysfunctional, and that we are worse off than if we all spoke a mutually intelligible version of post-Indo-European. As far as systematic phonetic change is concerned, most linguists, both Group A and B, felt that the process of diversification was entirely negative. It was only the effects of analogy, or even conscious social intervention which restored the balance. 3 The notion that sound change destroys and analogy reconstructs is so widespread that it it almost assumed by most linguists. It is true that Martinet finds some value in the symmetry produced by the hole-filling tendency, and other linguists would find value in the simplification and generalization of rules. As far as grammar is concerned, it still must be demonstrated that analogy can be systematic, or that systematic grammatical change exists (Kurylowicz 1964). As indicated above, Creole studies provide us with our largest body of observations of grammatical change in progress. Many variable phonological rules go to completion in the formation of pidgins and Creoles — the small bits of phonetic substance which represent function words and inflections in the contact languages disappear entirely. When that happens, we can observe compensating grammatical changes; as the Creoles develop, we observe new particles and new morphophonemic rules of condensation. Thus the past tense marker -ed and the irregular forms of strong verbs disappeared entirely in the English of Pitcairn Island. The habitual present was then marked by a new particle; from the habitual past form [justa] was formed a corresponding present [juss]. In Hawaii, the developing Creole introduced a new past tense marker in the auxiliary, went; we now observe the development of new rules of glide deletion and elision, so that they went grab can be [den graeb] and they went look can be [dejAsen luk]. But we are just beginning the study of such ingenious solutions to the problem of reconstructing grammatical systems. On the whole, linguists still seem to be confronted with the view that diversification of language is due to the systematic and destructive effects of sound change (usually laid to the principle of least effort) and the breakdown in communications between isolated groups. This finding not only destroys the parallelism between linguistic and biological evolution, but is strangely conservative as well: it projects the view that the unchanging, homogeneous speech community of Chomsky-Martinet is the ideal towards which we should be striving, and that any degree of heterogeneity subtracts from our communicative powers. Because this is an unattractive and seemingly unrealistic result, I am inclined to reject it; some of the available evidence for this point of view will be marshalled in the final section. 3

In this sense, Darwin's argument was reversed. He thought that the phonetic processes which shortened words necessarily improved them, so that the fittest or shortest survived.

THE SOCIAL SETTING OF LINGUISTIC CHANGE

3.

207

THE STUDY OF SOUND CHANGE IN PROGRESS: THE UNIFORMITARIAN PRINCIPLE

If we are to apply ourselves to the solution of the three questions of linguistic change outlined above, we must immediately face one grave difficulty: we have too little information on the state of society in which most linguistic changes took place. The accidents which govern historical records are not likely to yield the systematic explanations that we need. Some historical linguists have achieved remarkable and insightful results with the texts on hand; H. C. Wyld will be cited as the most striking example. But such efforts have never been more than moderately convincing, and one is always free to disagree on the basis of other fragmentary evidence. The only strong solutions to the problems of language change just posed are through the study of change in progress. Group B linguists have often claimed, in defense of neo-grammarian principles, and on the basis of their thought-experiments, that linguistic change is too slow, too subtle, or too elusive to be studied as it occurs around us. In the work to be cited below, there is overwhelming evidence to the contrary: that the study of on-going change is a practical strategy. This evidence will reinforce and help us interpret the results of historical investigations. To use it in this way, we necessarily have to operate under a uniformitarian principle,4 We posit that the forces operating to produce linguistic change today are of the same kind and order of magnitude as those which operated in the past five or ten tousand years. There are certainly new factors emerging, with the growth of literacy, the convergence of widespread languages, and the development of scientific vocabulary. Yet these represent minor interventions in the structure of languages. If there are relatively constant, day-to-day effects of social interaction upon grammar and phonology, the uniformitarian principle asserts that these influences continue to operate today in the same way that they have in the seeable past. To the extent that this principle is unjustified, our interpretations of the past by the present may be wide of the mark; but from present indications, the principle will be as successful in linguistics as it is in geology. To cite Gauchat, the most brilliant and successful of the earlier workers in this field : ... les dialectes parlés sont les représentants vivants de phases que les langues littéraires ont parcourues dans le cours des temps. Les patois ... pourront nous servir de guides pour arriver à une meilleure compréhension de l'histore des langues académiques (1905:176). It may immediately be argued that we do not literally observe the change 'in progress'. In most of the studies to be reported here, the investigator observed distribution in 4

A term borrowed from geology; the concept introduced into nineteenth-century geological theory by Charles Lyell that the mountains, volcanoes, beaches and chasms we now have are the result of observable processes still taking place around us, rather than violent convulsions at some remote time in the past ('catastrophism'). The uniformitarian doctrine is one of the accepted principles of current geomorphology — perhaps its fundamental tenet.

208

WILLIAM LABOV

apparent time — that is, the differential behavior of speakers in various age levels. We distinguish this behavior from regular and repeated age-grading by obtaining at least one measurement at some contrasting point in real time. In the case of Martha's Vineyard, we had the Linguistic Atlas records of thirty years previous. Gauchat's study of Charmey (1905) was oriented to real time twenty-four years later by the supplementary observations of Hermann (1929). But even if we did repeated studies of the same areas every few years, it might still be argued that we were only studying discrete stages and not the change in progress. Such an argument is based on a vision of linguistic behavior as a set of uniform, homogeneous rules which can change uniformly, much as a beam of yellow light can gradually be changed to orange; the assumption is that we could theoretically observe the change in the same way that we watch the sunset turn to various colors. But this view is based on a faulty model of the homogeneous speech community. As Gauchat demonstrated, 'l'unité phonétique de Charmey ... est nulle'. We find instead differentiated behavior, in which there is a gradual change in the frequency with which certain rules are utilized in various environments (Labov 1972). The internal evolution of linguistic rules involves shifts in the ordering and prominence of certain 'variable constraints' — too abstract an object to be directly observed in the ordinary sense. We can take measurements of this process at discrete points in time, but we cannot observe the rules change from one moment to another. Consider the rule which diphthongized the low back /a:/ in Charmey. In our current notation:

This is a variable rule which states that a back glide appears variably after a low tense vowel, and that the rule applied more often to stressed (final) words, more often if the final consonant was not an /r/ (i.e. +central), and as a lesser effect, more often if the final consonant was a labial. The reflexes of Lat.porta, corpus, were rarely diphthongized from [pworta, kwo] to [pwa°rta, kwa°]. 5 When Hermann visited Charmey in 1929, he found the rule in a more advanced state: it had gone to completion in all environments except before /r/, where it was now quite common.

The rule now states that the vowel is diphthongized variably, except that wherever it is 5

Though the vowel before r is given by Gauchat as open o [q], and the main diphthongization rule affects a, these two phones can be analyzed as conditioned a variants, and so we are dealing with the same rule. A later rule deletes final r.

THE SOCIAL SETTING OF LINGUISTIC CHANGE

209

not followed by an /r/, it applies without exception. 6 It is clear that we could not follow the change of this rule from one utterance to another, for a discrete period of time would have to elapse before the new ordering of the variable constraints would be evident even for an observer who never left the scene. The qualitative changes shown in the rule involve (1) the appearance of the 'categorical' symbol *, for all but noncentral environments, and (2) as a consequence, the disappearance of the stress and labial constraints which are now obliterated in the general advance. Finally, something must be said about the distinction which is often drawn between the 'origin' and the 'propagation' of a change (Postal 1968:284; Sturtevant 1947; Sommerfelt 1930). Speaking on behalf of those who investigate change in progress, we do not find such a distinction coherent. What is the origin of a linguistic change? Clearly not the act of some one individual whose tongue slips, or who slips into an odd habit of his own. We define language, along with other Group A linguists, as an instrument used by the members of the community to communicate with one another. Idiosyncratic habits are not a part of language so conceived, and idiosyncratic changes no more so. Therefore we can say that the language has changed only when a group of speakers use a different pattern to communicate with each other. Let us assume that a certain word or pronunciation was indeed introduced by one individual. It becomes a part of the language only when it is adopted by others, i.e. when it is propagated. Therefore the origin of a change is its 'propagation' or acceptance by others. From that point on, we only have a continuation of the same pattern. We do not rule out the possibility of independent simultaneous innovation by a number of speakers; but we do find absurd the notion that an entire community would change simultaneously without reference to each other, without a gradual movement of the pattern from speaker to speaker. Every empirical study, beginning with Gauchat, shows such systematic differentiation in even the most isolated and closely knit communities. In this chapter, I will utilize data from six of these empirical studies of speech communities, each of which provides a reasonable amount of data on the social setting of the change under observation. 1. Louis Gauchat's investigation of phonetic diversity among three generations of Swiss French speakers in the village of Charmey (1905), with Hermann's following report (1929). The main sound changes observed in progress are summarized in Table 1, showing (a) the palatalization of 1' y, proceeding variably in the middle generation, completed in the youngest; (b) the monophthongization of /aw/, variable in the oldest generation, completed in the middle; (c) the diphthongization of back o - > a ° beginning in the middle generation, completed in the youngest except before /r/; (d) the diphthongization of e e1, beginning in the oldest generation, variable in the middle age group, 8 The asterisk notation indicates a feature in a variable rule which, when present, causes the rule to apply categorically, without exception (Labov 1970).

210

WILLIAM LABOV TABLE 1

Four sound changes in progress in the Swiss French of Charmey, 1899

( n

(aw) (ey)

0>)

I 90-60 yrs

II 60-30 yrs

III under 30

r o° ~ (a-) e-(e') 0

l'~j ti-

Q"

j

fi ~ E*

o~ a

E1

0

a0

and completed in the youngest generation except for the word class with underlying /r/ following the vowel. 2. Ruth Reichsteirfs investigation of phonemic variables among Parisian school children (1960), against the background of Martinet's World War I inquiry. Reichstein tested some 570 schoolgirls for phonemic contrast with nine minimal pairs, revolving about / a ~ a / , /E ~ E :/, /E — DE/; these contrasts seemed to be disappearing rapidly, and comparison by area and class structure shows that certain interior working-class districts led the way. 3. My own study of the centralization of (ay) and (aw) on Martha's Vineyard, compared to earlier phonetic records made by Linguistic Atlas interviewers. This study showed (a) progressive centralization of (ay) in the oldest generation, with (b) a later centralization of (aw) in the middle generation, overtaking the first process for younger speakers: see Table 2. The Atlas data shows no centralization of (aw) in 1933. Recent spectrographs studies of the same speakers (Labov 1972) have confirmed the original view of the mechanism, and added considerable detail. TABLE 2

Centralization of (ay) and (aw) in 3 generations of English speakers: Martha's Vineyard, Mass. Generation la lb Ila lib III

(over 75 yrs) (61-75 yrs) (46-60 yrs) (31-45 yrs) (14-30 yrs)

(ay)

index 25 35 62 81 37

(aw) index 22 37 44 88 46

4. My own study of the evolution of New York City vowels (1966a), as confirmed and enlarged by our current instrumental studies (1972). Relation to real time is shown by comparison with four other reports extending back to 1896. The New York City investigations show (a) a rise in the stratification of final and

THE SOCIAL SETTING OF LINGUISTIC CHANGE

Fig. 1.

Four stages in the New York City chain shift /ahr ->ohr -*uhr/

211

212

WILLIAM LABOV

pre-consonantal /r/ among speakers below 40; (b) a tensing and raising of short a (to form the variable (eh)) and raising of long open o (to form the variable (oh), from low to high position, with accompanying merger of mid and high ingliding vowels; (c), a backing and raising of the nucleus of /ay/ and /ah/ in guy and God, with a corresponding fronting of the nucleus of /aw/. Figure 1 shows four stages in the chain shift of ahr -»ohr -> uhr relating the word classes of lard, lord and lured, from the spectrographic measurements of the vowel systems of four working-class New Yorkers. 5. The study of Hillsboro, North Carolina carried out by Levine and Crockett, primarily through questionnaires and formal tests of pronunciation (1966); a study of the black population of the same town, using the same methods, by Anshen (1969). Levine and Crockett have reported so far only the pronunciation of final and preconsonantal /r/. They found a strong shift towards a new norm of r-pronunciation, parallel to (3a), but evidence for an older high-prestige norm of r-lessness surviving also. 6. A recent investigation of the English of Salt Lake City and its environs by Stanley Cook (1969) showing an early stage in the development of an urban dialect. The most prominent feature in the process of change was the fronting of (aw), which was strongest among young college people, and spreading gradually outwards from Salt Lake City. Cook also studied the rural merger (and stereotype of reversal) of (ar) and (or) in far and for-, it was shown to be an advanced change undergoing a certain amount of overt correction. Cook was able to trace the history of the merger in a rural community, its stigmatization and reversal among younger speakers, and a tendency towards hypercorrect raising of (or) to [oV] among younger suburban speakers. The separation of the two word classes among rural speakers is shown in Table 3. TABLE 3

Phones used for (ar) and (or) in casual speech by three sets of speakers in Minersville, Utah a

N Older informants ( 6 0 + yrs) Noncollege youth (ca. 17 yrs) College students (ca. 21 yrs)

5 2 2

(ar) (or) (ar) (or) (ar) (or)

89 40 100 14 100 —

D

oT 11

—

ox

n

o

—

—

—

—

53

—

—

—

29

54

—

—

8

100

4 —

—

—

There are of course many studies of secondary data which give us considerable insight into the social setting of linguistic change. The most useful examination of historical texts, for our purposes, is that of H. C. Wyld in A history of modern collo-

THE SOCIAL SETTING OF LINGUISTIC CHANGE

213

quial English (1936). Wyld had a strong interest in the influence of class dialects on the history of English, and their origins in regional dialects, because he himself was a speaker of a class dialect — 'Received Pronunciation'. We can profit from his close examination of the spelling in such documents as the Cely papers, the Paston letters, Machyn's diary and the Verney Memoirs, by writers with a wide range of social background. We might enlarge the list of sources indefinitely by including qualitative and fragmentary accounts of change in progress. But quantitative studies of actual speech communities must necessarily take precedence. We can draw some conclusions from qualitative observations which are framed in a three category system: that a given form or rule is never found, occurs variably, or is always found. But much more profitable use can be made of linguistic variables ranging freely from 0 to 1, where X represents the proportion of all permitted environments in which the rule actually applies, so that 0 < x < 1. Here we are in a better position to detect the progress of a change or give a convincing account of how the change is correlated with social factors. A review of the tables just given will show that most of the relationships would disappear in a qualitative report. In the case of Gauchat, we will see that the most interesting and important findings add quantitative detail to the qualitative Table 1. In our most recent studies of sound change, we can move to a higher level of precision by the use of instrumental measurements, and this has now been done for two of the six studies outlined above (see Labov 1972).

4. THE EMBEDDING OF LINGUISTIC CHANGE IN ITS SOCIAL CONTEXT

We can identify at least five different problems connected with the explanation of linguistic change (Weinreich, Labov and Herzog 1968) but not all of them are related to the social setting of the change. The universal constraints upon linguistic change are by definition independent of any given community. The question of locating the transition between any two stages of a linguistic change is an internal linguistic problem. The embedding problem has two aspects: the change is seen as embedded in a matrix of other linguistic changes (or constants), and also as embedded in a social complex, correlated with social changes. There is also an important social component to the evaluation problem — to show how members of the speech community react to the change in progress, and discover what expressive information the variants convey. Finally, we can expect that social factors will be deeply involved in the actuation problem: why it took place at the particular time and place that it did. It should be clear then that a full understanding of linguistic change will require many investigations that are not closely tied to the social setting, and other studies which plunge into the network of social facts. Our other studies of the universal constraints on the expansion of mergers (Herzog 1965:211), on universal principles of vowel shifting (Labov and Wald 1969), and the internal transition of rules (Labov

214

WILLIAM LABOV

1972) are concerned with the speech community only as a source of data. In order to assemble the data we need to answer the three questions raised in section 2 of this chapter, we will have to take full advantage of the available data on the social embedding, evaluation, and actuation of the linguistic changes under study. Our first problem is to find out which aspects of the social context of language are most closely connected with linguistic change. We might begin with a complete account of the immediate social setting of the speech event, following the the descriptive program of Hymes (1962). We would consider all social relationships holding among speakers, addressees, audiences, and inhabitants of the social domains of the speech event (school, church, job, family ...). We could then ask whether changes in the language reflect changes in the relationships between these participants and settings. For example, we are now witnessing a steady series of changes in the use of second person pronouns of respect in Spanish (Weinberg and Najt 1968), French (Lambert 1969), Serbo-Croatian (Kocher 1967) and other languages. We suspect that there is some truth to the conventional reaction of older people: that young people don't have the respect for their elders that they used to. But what independent measures of respect behavior would show us that this is more than a change in conventions or surface modes of expression? We have increasingly sophisticated means of recording and measuring linguistic behavior, relating our observations to the language used in everyday social interaction. But we do not have equally well-developed measures of authority, respect, or intimacy. We would therefore be wise to correlate our linguistic data with whatever objective measures of social position or behavior we have — measures which can be repeated reliably by others at other points in time.7 It therefore seems reasonable to connect linguistic behavior with measures of ascribed and achieved status of the speakers. Whereas shifts of linguistic expression can register momentary changes of social attitudes, we will be more concerned with well-established ranges of linguistic expression — the way in which the individual habitually presents himself in various social settings. From one moment to another, our language gives information to answer the listener's question, 'How are you feeling about me?' But the speaker's language provides general information about himself as well, answering the other's questions: 'Who are you?' and 'What are you?' These are matters of ascribed status — ethnic and religious membership, caste, sex, family — and achieved status: education, income, occupation, and possibly peer-group membership. Changes in language can then be correlated with changes in the position of the sub-groups with which the speaker identifies himself. As a matter of empirical fact, most incoming changes show meaningful social distributions before they register any stylistic shifting (see Cook 1969 on (aw)). ' That is not to rule out the possibility of more direct measures of social behavior in face-to-face interaction, closer to the speech act, and therefore with greater explanatory power.

THE SOCIAL SETTING OF LINGUISTIC CHANGE

4.1

Socio-Economic

215

Class

The social status of an individual is determined by the subjective reactions of other members of society, but it is easier for outsiders to use objective social and economic indicators to approximate the position of given individuals. In the United States, we usually obtain the sharpest over-all stratification with various combinations of occupation, education, income, and residential area. In studying historical records, we usually judge upper-class figures by their family connections and title; less prominent individuals are easiest to classify by their occupations and habitual associations. Henry Machyn, the Diarist, seems from his own words to have been a simple tradesman, possibly an undertaker, with a taste for pageants — especially for funerals (as was natural) — and for gossip. Of the great persons whom he mentions, he knew no more than their names and faces, scanned as they rode past him in some procession, and an occasional piece of gossip picked up, one is inclined to think from some other spectator among the crowd (Wyld 1936:141). Machyn's status is an important issue: he provides us with evidence for the lower middle class treatment of several important linguistic variables in sixteenth century London English. For example, most of his -er- words are spelled as -ar-: armyn, 'ermine'; hard 'heard'; sarmon, 'sermon'. Wyld takes this alternation of -er and -aras a classic case of the correlation of social class movement and linguistic change. He raises the question, how did it come about that many words now pronounced with a mid central vowel, [ a r ] , like clergy, heard, etc., were pronounced by good speakers in the 17th and 18th centuries with [as - ]? If the [ar] pronunciation was'wrong', how did it get to be adopted by the aristocracy in the first place? Wyld collects a considerable body of data to show that the [ay] pronunciations entered the London dialect in the 15th century from Southeast dialects, and appear strongly in the private writings of middle-class Londoners like Machyn up to the middle of the sixteenth century. From this point, such spellings appear with increasing frequency in upper-class English, until the end of the 18th century when they begin to recede. We have here a linguistic feature which found its way from a Regional dialect into Middle Class London speech, passed thence into Received Standard, only to be ousted later by a fresh wave of Middle Class influence (1936:11). The ultimate rejection of the [aa - ] pronunciation by middle class speakers in favor of the standard spelling pronunciation was a natural by-product of their upward social mobility. 8 The new bourgeoisie has no way of knowing which words spelled -er- were pronounced [aa1] by elegant aristocrats, and they naturally could only adhere to the solid orthographic standard. 8

An alternate analysis of the New York City data (Labov 1967) based on the social mobility of the speakers provides as good or better correlation with linguistic behavior than the analysis based on the speakers' socio-economic positions.

216

WILLIAM LABOV

We see in this development three processes of social embedding: (1) the transformation of a regional dialect into an urban lower-class dialect; (2) the spread of a linguistic feature upward from a lower to an upper class; (3) the recourse to spelling pronunciation by an upward mobile group. We can easily find further support for Wyld's view of the matter. The [aa - ] pronunciation still appears in Southeast dialects today. 9 And as we will see, there is ample evidence for the upward movement of lower-class innovations or importations into the standard language. Oddly enough, a great deal of the speculative literature on dialect borrowing is based on the notion that all movement of linguistic forms is from the higher prestige group to the lower.10 When Group B linguists deal with this topic, they inevitably assert such a principle. Among his occupational companions, for example, a speaker will imitate those whom he believes to have the highest 'social' standing (Bloomfield 1933:476). This is simply a remark, with no more justification than any of the other general observations in Bloomfield's treatment of dialect borrowing. Studies of current sound changes show that a linguistic innovation can begin with any particular group and spread outward; and that this is the normal development; and that this one group can be the highest status group, but not necessarily or even so frequently. We can find two other examples of incoming changes in Machyn's speech which eventually reached the upper class. One is the high pronunciation of -ea- words from M.E. as [i:], indicated in Machyn's spelling by y in prych, spyking, bryking, brykefast. Eventually, this word class was detached from the mid-vowel position of long à words so that meat fell together with meet rather than mate. Secondly, we note in Machyn a pronounced tendency for the loss of -r before -s, giving Woseter 'Worcester', Dasset 'Dorset', continuing a tradition which ultimately led in the 18th century to a completely r-less pronunciation for all Londoners. 11 In more recent studies, Reichstein (1960) finds that several on-going phonemic changes she was studying were most advanced in working-class districts of Paris. Our own studies of the suburbs of Paris find the most extreme raising and backing of /a/ among working-class youth, so that casser la tête can be raised to [koselatet], and j'sais pas to [spo]. 12 In our other exploratory studies of the dialects of Boston, Rochester, Detroit and Chicago, we find the most advanced forms of many vowel shifts among * As shown in data gathered by Howard Berntsen and analyzed in our current studies of sound change in progress. 10 Expressed as a general principle by Gabriel Tarde in his Lois de l'imitation in 1890, and since k n o w n as Tarde's Law. 11 By r-less, we mean that there is a categorical rule for the vocalization of consonantal r in final and pre-consonantal position: whenever not followed directly by a vowel. This style of speech was adopted in the nineteenth century in all those cities of the United States which looked to England as a center of cultural prestige: Boston, N e w York, Richmond, Charleston and Atlanta, but not Philadelphia. 12 Although the working-class Parisian youth that we have studied preserve a great phonetic distance between front and back a, the lexical distribution of this distinction is different from that of the standard.

THE SOCIAL SETTING OF LINGUISTIC CHANGE

217

working-class youth. In Chicago, it is working-class youth who show the most extreme tensing and raising of short a and fronting of short o, together with the lowering of lax short vowels /i/ and /e/ to mid and low position. Figure 2 shows such an extreme vowel system, that of a 17-year-old working-class girl. 13 In her sentence, That ended that [SiT:9t Bndid 6i T :'t], we observe low vowels raised to high, and a mid vowel opened to low position. Dialect studies based on the speech of college students fail to capture such remarkable transformations of vowel systems until they are already far advanced, and already being repressed. It is not always the working class which is at the leading edge of linguistic change. Often it is a higher status group, comparable to Machyn's. Table 4 gives systematic data on one of the ongoing changes in the New York City vowel system: the backing and raising of (ay) along with the fronting of (aw). The numbers here refer to a scale in which (ay)-OO would represent a low central nucleus for either (ay) or (aw): [a1, a u ], characteristic of the oldest speakers. Consistent pronunciation of (ay) as [D1] would yield (ay)-40, and consistent (aw) as [aeu] would also show a score of 40. Table 4 shows that the movement of (ay) seems to have begun in the middle class groups, but continued most strongly in the lower middle class, spreading gradually to the working class while upper middle class speakers avoided it. The fronting of (aw) is strongest in the lower middle class, but is gradually expanding in this generation to affect the other groups. Innovation by the highest status group is normally a form of borrowing from outside sources, more or less conscious; with some exceptions, these will be prestige forms. 14 The original spread of r-less pronunciation in the United States was modelled on the London patterns of the early 1800s — a 'change from above' spreading outwards from the centers of Anglophile influence. 15 The present reversal of this trend throughout the Eastern U.S. can be seen in Boston and the South (Levine and Crockett 1966) as well as New York (Labov 1966a). This is another form of change from above, reversing the prestige relations. There is a sharp differentiation between older speakers, who still reflect the earlier Anglophile norms, and the younger speakers who have adopted the new broadcast norm of 'general American' /--pronunciation. This change from above can be quite regular; it can affect each subgroup in proportion to its distance from the center of prestige, and the formality of the speech situation. A linguist who is interviewing these subjects will observe these importations from above more clearly in reading texts than in colloquial speech; in the unattended 13

By her own characterization, this girl spent most of her high school years as an integral member of the stable working-class group ('greasers'), but recently has moved into a more upwardly mobile group ('soshes') with her current boy friend. 14 In the same volume that marks Sam Weller and his father with the stigmatized vjw confusion, Dickens also ridicules a foppish young lord with a stereotyped labialization of prevocalic r, all printed as w's in the text. 16 This expansion reached a radius of approximately 150 miles from Boston, Richmond and Charleston, but was confined to the immediate vicinity of N e w York City.

218

WILLIAM LABOV

Fig. 2. Vowel System of Carol Muehe, 16 years, Chicago

THE SOCIAL SETTING OF LINGUISTIC CHANGE

219

TABLE 4

Average (ay) and (aw) values for all white NYC informants by half-generations

Age

5-19 20-34 35-49 50-64 65+

Generation II-B II-A I-B I-A 0

(aw)

(ay) 0-2

3-5

6-8

9

7 5 8 5 0

23 18 17 10

22 24 18 10

12 10 20 15

0-2

3-5

6-8

9

8

20 7 7 7

17 10 8 5

8 4 1 5

—

4 2 0

Index of (ay) and (aw) = average of phonetic ratings x 10

Phonetic scales for rating fronting of nucleus of (aw) and backing of nucleus of (ay)

220

WILLIAM LABOV

speech of everyday life, we observe more dramatically the effects of change from below as in the movements of (ay) and (aw) cited above. 16 Table 5 shows the average indexes of /"-pronunciation in casual style for New Yorkers : it is plainly an upper middle class TABLE 5

Average frequency offinal and preconsonantal [r] in casual speech for New Yorkers by age and class SEC Age

0-1

2-5

6-8

9

8-19 20-39 40-

00 00 00

01 00 06

00 00 09

48 34 09

SEC index: 0-1 2-5 6-8 9

'Lower class' 'Working class' 'Lower middle class' 'Upper middle class'

N: 6 3 10

16 13 25

6 9 8

4 4 7

TABLE 6

Percentage of those using no final and preconsonantal [r] in reading wordlists by age and class SEC Age

0-1

2-5

6-8

9

20-39 40-49 50-59 60-

67 20 33 71

75 18 14 50

00 00 00

20 25 33

—

—

N: 3 5 3 8

17 17 7 4

11 9 1 —

5 4 3 —

phenomenon which has not affected the natural speech of other groups. Table 6 shows the way in which this prestige form affects speakers in the most formal style: the percentage who use the completely r-less norm in reading wordlists. The second highest status group — the lower middle class — is most affected; not one speaker stayed with the older norm. This is one of the many indications of the strong pattern of 'hypercorrect' behavior on the part of the second status group, an important element in the mechanism of linguistic change (Labov 1966b). A parallel example of this 16

In the early stages of a linguistic change, a consistent pattern can be observed in formal styles as well as casual speech. This is the case with most New Yorkers' pronunciation of front (aw) and back (ay). It is not until the more advanced stages of the change that formal styles show correction and irregular distribution.

221

THE SOCIAL SETTING OF LINGUISTIC CHANGE

hypercorrect pattern appears in the quantitative study of /--pronunciation of Hillsboro; as Table 7 shows, the greatest shift from sentence style to wordlist style was on the part of the high-school educated group, rather than the college group. The sentences had blanks, and the subject's attention was drawn to the problem of filling these with a lexical item, rather than the words of the text that contained /r/. In the wordlist, attention was focused directly on the variable. TABLE 7

Frequency of final and preconsonantal [r] in sentences and wordlists by age, education and sex for white speakers in Hillsboro, N.C. Sentence-list

Wordlist

Net increase

Age 21-39 Years 40-59 Years 60 Years and Over

56.6 54.2 44.5

65.1 60.3 49.3

8.5 6.1 4.8

Education Any College High School Graduate Some High School Grade School or None

52.7 54.6 50.0 52.6

58.9 65.6 57.0 57.3

6.2 11.0 7.0 4.7

Sex Male Female

52.3 52.9

57.4 61.1

5.1 8.2

Returning to the age relationships in Table 6, we find that in the working-class groups, the older speakers tend to hold to the earlier norms, and the younger speakers as well; it is only among the middle-aged groups that the new prestige norms are adopted. As we will see below, overt correction tends to be rather unsystematic when it occurs late in life, and it focuses on individual words rather than on general rules.17 We might ask: can the highest status group ever innovate unconsciously? Insofar as any habit of speech is associated with a high prestige group, it is apt to be remarked. Conservative critics will quickly call attention to it, as Gill ridiculed a fashionable pronunciation which raised long a to [i:], giving [ki:pn] for capon (Wyld 1936). The difference between a change in progress and an advanced change may sometimes be seen clearly in the pattern of social distribution. A change may begin first in 17

Speakers who have acquired the norm of formal /--pronunciation late in life show regular and predictable patterns of shifting, but do not achieve consistency. The word classes containing /r/ are fairly well defined, yet there is a fair amount of hypercorrection, in ¡dear, lawr and order and occasionally Gard for God.

222

WILLIAM LABOV

a social group located anywhere in the social hierarchy. As long as it is developing and spreading outward, one can still see the pyramidical pattern through various age levels, with the highest values in the youngest speakers of the original group. But when the change reaches an advanced state, and all social classes are affected, it is often stigmatized, and the social correction of formal speech begins to obscure the original pattern. In this case we get a linear distribution, with the highest social class showing the least amount of the stigmatized feature in ordinary conversation. We can see this clearly in the study of the Spanish of Panama City by Henrietta Cedergren (1970). Table 8 shows Cedergren's findings on the social stratification of five variables in Panamanian Spanish across four social classes. TABLE 8

Social stratification of five Spanish variables in Panama [from Cedergren 1970] Social groups

(R) (PARA) (ESTA) (S) (CH)

I

II

III

IV

1.62 1.11 1.26 2.03 1.88

1.88 1.37 1.56 2.24 2.24

2.29 1.39 1.62 2.31 2.13

2.29 1.69 1.71 2.36 2.00

The linguistic variables in Table 8 may be defined briefly as follows: (R): the devoicing, fricativization, pharyngealization, and deletion of syllable-final /r/, with values ranging from 1 to 6 in the direction of these processes. (PARA): the alternation of the full form of the preposition para with pa, with values of 1 and 2 respectively. (ESTA): alternation of the full form esta with ta, assigned values of 1 and 2 respectively. (S): the syllable-final alternation of [s], [h] and [0], with values of 1, 2 and 3 respectively. (CH): palatal vs. retroflex and reduced stop onset of /c/, with values of 1 and 2 respectively. The figures in Table 8 are the arithmetic means of the values of the variables. The social classes range from the highest (I) to the lowest (IV), with the first four variables showing a linear distribution with lowest scores for the highest social class. But the (CH) variable shows a curvilinear pattern which suggests that the change originates in the second highest status group II. That (CH) represents an on-going change in progress is clearly shown by Table 9, which gives the distribution of scores by age groups for the same five variables. We can see that (CH) is the only variable with a

THE SOCIAL SETTING OF L I N G U I S T I C C H A N G E

223

uni-directional linear progression across age groups; there is a dramatic advance in the youngest groups quite comparable to the values shown for change in progress in Martha's Vineyard and New York City. TABLE 9

Development offive Spanish variables by age groups [from Cedergren 1970] Age

(R) (PARA) (ESTA) (S) (CH)

11-20

21-30

31-40

41-50

61-70

2.28 1.31 1.64 2.34 2.15

1.90 1.34 1.50 2.22 2.29

1.95 1.48 1.67 2.15 2.05

2.23 1.33 1.57 2.18 1.81

1.46 1.39 1.41 2.19 1.31

Not every linguistic change is attached to a particular social group. The raising of M.E. e to [i:] seems to have been 'common to the speech of all areas and classes in London' (Wyld 1936:207). In areas of the United States, the merger of short open o and long open o seems to affect everyone in certain areas. 18 But here I am speaking from general impressions; it cannot be denied that in every case that has been studied closely, one social group or another has been found to lead strongly in the development of a linguistic change. We have thus documented in some detail the embedding of linguistic change in one type of social structure; the socio-economic system of class differentials. Change does not occur without regard to class patterns; instead, the incoming pattern enters like a wedge with one group or another acting as the spearhead. The feature is rarely confined to a particular class (unless it is stigmatized and recedes — see section 5 below). It is difficult to see how Martinet or Chomsky would handle this situation. Would the homogeneous society posited by the abstract linguist contain the new form or not? If not, at what point would the abstractor make the decision to include it in his homogeneous system and for what reason? I can imagine two directions in which the homogenist might go. He might simply abstract from social class, and allow the incoming form to be optional for everyone. He would then argue that its association with a particular class of people is below the level of linguistic significance. In that case, the new option would appear suddenly as a blind fact, without direction or interpretation. Our discussion of the evaluation of such well-developed linguistic changes will illustrate how much significance the option can have in fact. A second alternative would be to abandon the notion of writing a description of 18 There are exceptions to the uniformity of this merger of hock and hawk, don and dawn. In Phoenix, for example, it appears to be more characteristic of the Anglo population than Negroes or Spanish-Americans.

224

WILLIAM LABOV

langue as a community property, but simply describe the speech of one class. If the community is differentiated according to class, we could then seize the homogeneous groups which do exist, and describe their language. As a first step this might be practical, but complications ensue when we discover that there are other social structures intersecting with this one, such as caste or ethnic group.

4.2

Ethnic group and caste

The beginning of the current interest in sociolinguistic studies can be seen in the 1960 publication of Ferguson and Gumperz, Linguistic diversity in South Asia. Some of the papers concern literary-colloquial oppositions, but others are involved with the description of caste differences: in Kannada (Bright and McCormack), Tamil (Pillai), Bengali (Dimock and Chowdhury), and Hindi (Gumperz and Nairn). Bright was primarily concerned with the effect of caste differences on linguistic change. Though his comparisons of two dialects of Kannada are only based on two college students, his conclusions are of the greatest interest. The non-Brahmin dialect [NB] accepted foreign words and foreign phonemic patterns (English and Sanskrit) less readily than the Brahmin dialect [B]; B was more resistant to grammatical change and phonological change from within. For example, B has imported /f, z, o/ from English, where NB has substituted native /p, j, a/, so that B kofi 'coffee' = NB kapi, and B has Sanskrit /§/ in sdnti, 'peace', while NB has sdnti. On the other hand, B shows original /e/ in words where NB has differentiated to /ya/ if a mid or low vowel occurs in the next syllable: B pete, NB pyate. A later secondary study of Tamil and Tulu showed the same general pattern: The Tulu evidence shows the Brahmins as chief innovators in the more conscious varieties of change — semantic shift, lexical borrowing, and phonological borrowing. In the less conscious processes of phonological and morphological change involving native materials, both B and NB dialects innovate (Bright and Ramanujan 1964). Bright and Ramanujan conclude that upper and lower class dialects innovate independently; more conscious importations are regularly the mark of the upper class, while the less conscious changes affect both classes. They then point out that the Brahmins can show overt correction of these changes, as in Kannada and Tamil. These findings from South Asia fit in well with the general principles of change which emerge from our New York City studies, contrasting the behavior of the highest status group with the others. 19 In the development of the New York City vowel system, we find that ethnic identity plays an important role — more important than socio-economic class, for some items. 19

Current studies of Maxine Berntsen in Phaltan are designed to examine the social distribution of non-standard features of Marathi, using many of the same techniques as in the New York City studies (Berntsen).

THE SOCIAL SETTING OF LINGUISTIC CHANGE

225

The ethnic differentiation of the tensing and raising of short a in bad, ask, dance — the (eh) variable — is shown in Table 10. The numbers represent the average degree of openness of the vowel. Consistent pronunciation of bad with [as :] at the level of unraised bat would show a value of (eh)-40. If all such short a words that were tensed TABLE 1 0

Distribution of (eh) by age and ethnic group Age 8-19 20-39 40-49 50-59

Jews

Italians

Negroes

22 23 27 29

20 19

24 28

18

33

were then raised to the level of [e:], the index would show (eh)-20, and for the level of [i:], (eh)-10. Although all groups show a gradual decrease in the openness of the vowel with age, New Yorkers with Italian background show the greatest decrease for each age level. This ethnic differentiation shows up in all but the lowest social group, which has not been drawn into this process. When we speak of ethnic identity as a part of the social context of a linguistic change, it immediately raises the question of a substratum. Is the underlying parent language the cause of this differentiation? The Martinetian view would bend to this tactic, since we are considering outside impact as reflected in 'the pressure of another language.' But the language-contact explanation fails by itself. Italian has no [®], and firstgeneration Italian-Americans tend to use their native Italian low vowel [a] for the class of English short a words. It would then appear that the second-generation Italian tendency to raise short a is not what a structural version of contact theory would predict: it reverses the direct influence of Italian. Yiddish has no [ae] either, but first-generation Yiddish speakers tend to raise English short a to [e]. Second-generation Jewish speakers of English have somewhat less tendency to raise this vowel than the Italians. This result can be interpreted in the light of other sociolinguistic data on second-generation speakers who reach for native status by removing themselves as far as possible from the low-prestige pattern of their parents. This is another version of the 'hypercorrect' or reverse influence that appears to play a major role in linguistic change. We now see that linguistic change may also be differentiated by its association with a particular ethnic or caste group, and that various ethnic groups may treat the same variable in different ways. An abstract treatment of this data might still decide to ignore the association with caste and ethnic group as beneath linguistic notice. The second approach — contracting one's focus to homogeneous groups — is now somewhat more difficult. A group of working-class men will no longer do; we must be sure to insist on working-class Italians.

226

WILLIAM LABOV

4.3

Local Identity

In addition to the cross-classifying assignments of class and caste status, communities often develop more concrete categories by which individuals are placed. In rural communities (or in urban villages), local identity is an extremely important category of membership — one which is often impossible to claim and difficult to achieve. In many small towns of New England, there is a large, low-status sub-category of 'summer people'. Then there are 'incomers' or 'outsiders', people who have moved into the town permanently but are shouldered aside for many decades before they are accepted. Third are immigrant groups, like the 'damn Portagees', Indians, Negroes, and other ethnic groups who are not immediately accepted into the mainstream. The eighth- or tenth-generation Yankees form the core of the local population, but members of other groups may gradually assume local identity. On the island of Martha's Vineyard, Massachusetts (Labov 1963), it was this network of social categories which was most closely correlated with the linguistic change in progress — more important than occupation, geography, education, or sex. In the final analysis, the sound change being studied was associated with the assertion 'I'm a Vineyarder'. For economic or social reasons, many persons felt unconsciously that they could not make this claim, and their language reflected the fact. The Martha's Vineyard study focused on the relation of social factors to linguistic change; it demonstrated that the direction and development of centralization could not be understood without relating it to the basic categories of local identity.

4.4

The Transformation of Regional Dialects into Urban Class Dialects

Wyld observed that a regular pattern in the development of English was the transformation of rural, regional dialects into class dialects in the cities. This process often involves the movement of rural speakers into urban low prestige occupations, concentrated in rapidly growing ghettos. When the rural speaker arrives in the city, he usually finds that his country talk is ridiculed. Even if it was a marker of local identity and a source of prestige at home, he may already have been conscious of the provincial character of his speech before he came to the city. As a result, we often see a rapid transformation of the more salient features of the rural dialects as speakers enter into city life. In the United States, the movement of the Southern Negro population into the northern cities has brought about the creation of a uniform caste dialect — the non-standard Negro English of Harlem and other inner cities. Black speakers in smaller communities, unaffected by this process of dialect swamping, tend to participate in the linguistic changes taking place around them. But in the larger ghetto areas, we find black speakers participating in a very different set of changes, with no direct

THE SOCIAL SETTING OF LINGUISTIC CHANGE

227

relation to the characteristic pattern of the white community (Labov, Cohen, Robins and Lewis 1968). First we find the extreme rural forms stigmatized: lexical items such as tote, verbal auxiliaries such as done or liketo ( = 'almost') are used less often. The characteristic vowel patterns of the regional Southern dialects are modified. These differences are so extreme that the sound [dae] can mean 'die' in Alabama and 'deer' in South Carolina. But such radical shifts are levelled out in the Northern cities. The /ay/ of die is limited to [da e ], [da V ] or [dae1" :], and deer is [di: 3 ] or [de:*]. The basic phonological patterns which intersect with grammar — consonant cluster simplification, r and I vocalization, copula deletion, remain remarkably constant throughout the Northern ghetto areas. Inflectional morphemes which are absent in the original Southern Black English, like third singular -s, remain absent. Syntactic patterns of Southern colloquial speech are maintained or extended, as with negative concord and inversion (Ain't nobody see that). The end result is a consistent caste dialect, relatively constant for speakers from six to twenty years old, with remarkable geographic uniformity and resistance to standard English importations in the school system. Speakers in Boston, Newark, Cleveland, Detroit, Chicago, New Orleans, St. Louis, Houston, San Francisco, and Los Angeles, show the same patterns with startling regularity. Similar processes appear to be operating in other languages, wherever large capital cities are developing at the expense of the hinterland. In the traditional literature of section 1, the social setting of language change is discussed in terms of the spread of the prestige patterns of urban capitals such as London and Paris. The creation of lowprestige working-class dialects is a pattern of equal linguistic interest; it embodies two major linguistic trends of the past several centuries: the decline of local dialects and the growth of vertical stratification in language. This rapid language mixing seems to follow a kind of classic structural reductionism, and it would not be difficult to argue that it is a sub-type of the same process that produces contact languages. As the history of various Creoles shows, the rapid result of sudden contact of two dissimilar structures is frequently the lowest common denominator of both, with a strong push towards inflectional simplification (see section 6 of Valdman's paper in this volume for various theories of the creation of pidgins and Creoles). One of the universal constraints on change seem to be operating here — that in contact situations, mergers expand at the expense of distinctions (Herzog 1965:211). Yet if we are to apply our linguistic insight to predict such mergers, we must first recognize the existence of heterogeneous dialects, as the common, even normal form of language system. Not every linguist is willing to do this. Wyld and Kokeritz tried to explain the re-alignment of the -ea- words as the result of the upward movement of a regional dialect within the London system. But Halle preferred to abstract from any social evidence on this point (1962) and argue the history of mate, meat and meet as if they were elements in the homogeneous speech community constructed by Chomsky and Martinet.

228

WILLIAM LABOV

4.5

The Role of Women

Gauchat's elegant and convincing study established the variability of Charmey patois, the existence of change in progress, and the role of women in furthering linguistic change. In case after case, Gauchat discovered that women used more of the newer linguistic forms than men did (1905:205, 209, 211, 218, 219, 224-6). 1. The palatalization 1' y was found variably among speakers 30 to 40 years old, and regularly for those under 30. Above 40 years old, only women showed this trait. 20 2. 0 - » h variably for post verbal pronouns so that for veux-tu [vuGo->• vuho] among the youngest generation, under 30, 'surtout les filles'. 3. a 0 -* a is variable in the oldest generation, and women apply the rule more than TABLE 11

Comparison of men and women for (eh) and (oh) Style B

A Men (eh) 10-13 14-18 19-21 22-26 27-32 33-39 40-42 (oh) 10-13 14-18 19-21 22-26 27-32 33-40

Women

1 3 4 3 4 1

1 4 10 6 4 4 2

16

33

3 3 4 3 3 1

4 10 14 5 4 1

17

38

—

D

Men

Women

—

—

—

3 7 11 3 —

24

2 9 9 12 5 6 43

4 7 8 5

1 10 7 16 10 2

24

46

—

Men —

Women —

1 1

—

8 4 4

2 5 9 14 16

18

46

—

2 5 5 5 1

3 5 13 4 8 10

18

43

—

Contextual styles: A Casual speech B Careful speech D Reading wordlists 20

Gauchat reports that one woman, 63 years old, pronounced his list of /' words regularly with y:

viyo (veclu), pxare, (plorat), byats3, (blanca), etc.

THE SOCIAL SETTING OF LINGUISTIC CHANGE

229

the men. Laurent Rime, 59, pronounced douce as [da°03]; his wife Brigide, 63, said [daft»]. 4. D->a°, variably in the middle generation, regularly in the youngest. 'Comme toujours, les femmes se mettent plus facilement sur la voie ... que les hommes.' Gauchat reinforced his finding by citing other examples from the history of French in which the women of Paris were portrayed as initiators of linguistic changes. We can point to similar behavior in the evolution of New York City English, and here the pattern of sex differences is even richer. In case after case, we find that women use the most advanced forms in their own casual speech, and correct more sharply to the other extreme in their formal speech. Table 11 compares men and women in the raising of tense short a, (eh), in three styles. In causal speech, women use more of the high vowels around (eh)-10 = [i T : 3 ], and show a modal value around (eh)-20, or [e:°]. But in reading lists of words with (eh) items, women shift to the opposite extreme, with a modal value around (eh)-40, = [ae:]. Men, on the other hand, shift their modal value only one notch, from (eh)-22-26 to (eh)-27-32. Our instrumental studies confirm these impressionistic tables: women are almost a whole generation further along in the raising of (eh) than the men (Labov, Yaeger and Steiner 1972). We find the same pattern in Detroit, where women clearly lead in the more extreme raising of this vowel, and in Chicago as well. Figure 2 shows the extreme rotation of the vowel system in Chicago which we cited as an example of young working-class speakers; but again it is the women in this group who show the more extreme forms. Throughout Shuy's studies of Detroit speech (Shuy, Wolfram and Riley 1967), we find the same pattern — women show more extreme shifting than men. In the Detroit studies, we observe only a shift towards prestige forms in formal style, but again women show the more mobile pattern. Why do women do this? It cannot be only their sensitivity to prestige forms, since that explains only half of the pattern. We can say that they are more sensitive to prestige patterns, but why do they move forward faster in the first place? Our answers at the moment are no better than speculations, but it is obvious that this behavior of women must play an important part in the mechanism of linguistic change. To the extent that parents influence children's early language, women do so even more; certainly women talk to young children more than men do, and have a more direct influence during the years when children are forming linguistic rules with the greatest speed and efficiency. It seems likely that the rate of advance and direction of a linguistic change owes a great deal to the special sensitivity of women to the whole process. We are now faced with data that poses even greater difficulties for those who claim that language change is independent of social variation. To abstract entirely from women's speech is an unattractive possibility ... we would then have to deal with working-class Italian men. The differentiation of men and women cannot depend on weaknesses in communication networks as Bloomfield would have it, or the law of least effort as Sweet would argue. In the communities we have been studying, there is

230

WILLIAM LABOV

no barrier to men and women talking to each other, and no reason to think that women are lazier than men; if anything, they put more effort into speech. We are dealing with some positive factor here — with a subtle set of expressive values. The advance of women over men is not a function of the amount of referential information being supplied, but something more affective than that.

4.6

The Peer Group vs. the Family

The role of women in the progress of linguistic change brings us closer to the mechanism of daily interaction; we cannot complete our view without examining the two social matrices in which young children and adolescents find themselves embedded: the family and the peer group. It is obvious that the child learns to talk from his parents; it is not so obvious why children do not talk like their parents. Dialect geographers prefer informants whose parents and grandparents lived in the same locality. But when they are forced to accept second-generation speakers, we find that these informants are really good representatives of the community in which they were raised, rather than their parents' dialect area. In our New York City studies, there was no noticeable difference between second and third-generation speakers. But if a person had spent more than half of his formative years (4—13) outside of the city, he did not show the New York City pattern. In the northern New Jersey town in which I live, over half of the adults were raised in New York, and maintain an r-less vernacular. With one or two exceptions, their children use the r-pronouncing dialect of New Jersey. In our studies of Harlem youth, we find no correlation between Southern linguistic features, and parents being raised in the South (Labov, Cohen, Robins and Lewis 1968). When children move into a new area in the middle of their formative period, they show a differential selection of dialect features, oscillating between the influence of the family and former peer group, and the current peer group. Some detailed evidence can be cited from an on-going study in a fifth grade in Radnor township, a suburb of Philadelphia in which roughly half of the parents come from other dialect areas. I have studied the phonological patterns of several self-selected peer-groups of elevenyear-old boys; the relevant linguistic patterns of one such group is shown in Table 12. Jim and Charlie have local parents and were raised in Radnor; Alley came from Worcester, Mass., at the age of eight; Tim came from Cleveland at the age of seven. Alley and Tim are two of the most popular and prominent members of the class, and Charlie too is among the best at a number of sports. Tim is the class president, and the group as a whole is at the center of things. Table 12 shows that Alley and Tim have learned the specific Philadelphia rules for fronting /uw/ and /ow/; they show the Philadelphia area's sharp difference between the central allophones used for most vowels and the back vowels before /l/, which are not fronted. Furthermore, Tim has acquired the backed and centralized allophone of

THE SOCIAL SETTING OF LINGUISTIC CHANGE

231

TABLE 12

Acquisition of Philadelphia phonology by four preadolescent boys

Age Born in Moved to Radnor at the age of

Jim Woisbecker

Charlie Campbell

Tim Midgett

Ken Alley

11 Radnor

11 Radnor

11 Cleveland, O. 7

11 Springfield, Mass. 8

—

—

TENSING AND RAISING OF

before: voiceless fricatives (anterior) nasal (anterior) voiceless stops

SHORT A

11/12 19/19 0/9

4/7 4/4 1/7

0/7 3/3 0/2

1/5 6/6 0/1

19/19 0/22

11/11 3/9

12/12 0/4

0/2 0/2

a, a H ox, o T

a, a"1 o \ oT

a

D

0

D

CENTRALIZATION OF / a y /

before : voiceless consonants voiced and final DISTINCTION OF LOW BACK VOWELS

short open /