Explaining language change: an evolutionary approach [1. ed.] 9787510031113, 7510031117

Table of contents :
Preface and acknowledgements
Contents
Chapter 1: Introduction
Chapter 2: An evolutionary model of language change
Chapter 3: Some theories of language change in an evolutionary framework
Chapter 4: A theory of language and meaning in use
Chapter 5: Form-function reanalysis
Chapter 6: Interference, intraference and grammaticalization
Chapter 7: Selection (propagation) of innovations in language change
Chapter 8: The descent of languages
Chapter 9: Towards an evolutionary linguistics
Glossary of terms
References
Index of authors
Index of populations (languages , communities and species)
Index of subjects

Citation preview

Explaining Language Change An Evolutionary Approach

WILLIAM CROFT

•

An;mprintor Pearson Hariow,

Education

England • London • New

York • Reading,

Massachusetts • San

Francisco • Toronto • Don

Tokyo• Singapore• Hong Kong• Seoul Taipei• Cape Town•Madrid• Mexico City•Amsterdam• Munich• Paris• Milan

Mills,

Ontario • Sydney

Pearton Education Limited Edinburgh Gate Hartow Essex CM20 2JE England

and Associated Companies throughout the world

Visit us on the World Wide Web at: www.pearsoned.co.uk

First published 2000

© Pearson Education Limited 2000 The right of William Croft to be identified as author of this work has been asserted by him in accordance with the Copynght. Designs and Patents Act 1988 All nghts reserved; no part of this publication may be reproduced. stored in retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise without either the prior written permission of the Publishers or a licence permitting restricted copying in the United Kingdom issued by the Copyright Licensing Agency Ltd , 90 Tottenham Court Road. London W1T 4LP

ISBN-10: 0-582-35677-6

ISBN-13: 978-0-582-35677-1 British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library

Library of Congress Cataloging-in-Publication Data

Croft. William Explaining language change : an evolutionary approach I William Croft p. cm. - (Longman linguistics library) Includes bibliographical references and index. ISBN 0-582-35678-4 - ISBN 0-582-35677-6 (pbk.) 1. Linguistic change. 2. Evolution (Biology) I. Title. II Senes

P142.C76 2000 410'1'8-dc21

99-088742

98765 11 10 09 08 07 06

Set in 10/12pt Times by

35

Printed in Malaysia. KHL

To the memory of Keith Denning 1955-1998

Scholar, colleague, friend

Preface and acknowledgements

This book presents a framework for understanding language change as a funda­ mentally evolutionary phenomenon. It is part of a larger research program, outlined in 'Autonomy and functionalist linguistics' (Croft 1995a). In that article, I argued that there

is a

variety of theoretical positions that are unfor­

tunately lumped together under the labels 'formalism' and 'functionalism'. These positions fall into two general classes. The first class comprises positions taken

on the

r

el ationship between

the

conventional

syntactic and

semantic know­

ledge in the minds of individual speakers (their grammar). I argued for a posi­

tion on this issue which I then called 'typological functionalism'. This position

is defended in

Croft (to

appear b), which should be thought of as a pendant to

this book. The second class comprises theoretical positions taken on the nature of the relationship between a speaker's grammar and language use, manifested in the utterances produced by a speaker in discourse. In Croft (1995a), I argued for

a

position on this issue which I called 'integrative functionalism'. Specifically, I

argued that: one cannot deny the existence of grammar, that is, knowledge of linguistic conventions in a speaker's mind; one cannot disentangle this know­

ledge from t11e act of using language; language function influences language form through the dynamics of language use and language

tionary

change.

The evolu­

framework of language presented here is intended to fulfill the promise

of an integrative functionalist model offered in that article.

In fact, the main task of integration in writing this book was to acquaint myself with various aspects of pragmatics, discourse analysis, semantics, lan­

guage acquisition, historical linguistics and several sub-areas of sociolinguistics, not to mention evolutionary biology, in order to work out the consequences of an evolutionary model of language. I hope that I have not committed too

many

scientific gaffes in the

programmatic tho ugh it is,

process,

and that the framework presented

will be of

interest to specialists in those

here,

areas.

Nevertheless, I believe that specialization has gone much too far in contem­ porary scientific research, and I hope that in this book, particular specialists will

find interesting and important connections to ideas and facts in sub-areas of linguistics other than their own, as well as connections to ideas in evolutionary biology and philosophy of science. The central thesis of this book originated with an idea which to me as a doctoral student taking a course

in

first

occurred

sociolinguistics from the late

Charles A. Ferguson - that the proper linguistic equivalent

to the genome was

replacement: the negative cycle

5.4.2

Passives from 3rd person actives and reflexives

5.4.3

Pra gmat ic inference and metanalysis

5.4.4

Constraints on metanalysis

130 131

133

134

Cryptanalysis

134

5.5.1

Pleonastic neg ati on

135

Paratactic negation

135

Pleonastic double marking I reinforcement

137

5.5.2 5.5.3 5.5.4 5.5.5 5.5.6 5.6

-s

Constraints on hypoa nalysis

Metanalysis

5.4.1

5.5

126

Subjunctives and futures from present indicatives

5.3.1

Spread of t h e definite a r t icle to proper names

Crypt a nalys is

+

metanalysis: the evolu t ion of mod als

into subj unctives

Constraints on cryptanalysi s

How far c an we get with form-function reanalysis?

138 138 140 140

Interference, intraference and grammatlcallzatlon

145

6.1

Interference

145

6.2

Intraference

148

6.2.l

148

38

The evoluti on of morphological paradigms

L exical sem an tic change

151

6.2.3

Mechanisms for intraference

154

6.2.4

Constraints on intraference

156

6.2.2

cD

common ground

4.3.l

5.2.l

6

a nd

The openendedness and flexibility of meaning in use

4.3.3 4.4

The function of language and communication

4.2.2

4.2.4 4.3

87

Introduction

6.3

Grammaticalization

156

6.3. l

l 56 159

6.3.2

7

8

G ram maticalizati on processes

Mechanisms for grammaticalization

Selection (propagation) of innovations in language change 7.1 7.2

Introduction Comm unities , societies and the internal/external distinction in

7.3 7 .4

Prop a ga tion and the adoption of conventions Propaga tion and the structure of speaker populations 7.4. l The l oc u s of channels of propa gati on in the population 7 .4.2 Direction of propagation in populations 7.4.3 The time course of propagation: the S-curve 7.4.4 The m acrostruct ure of speaker populations and pr o pagation

166

language change

166 174

178

178 180

183

190

The descent of languages 8.1 8.2

8.3

Phylogeny in biology and language from a plantish point of view

Primary lang u a ge birth: societal diverg e nce and no rmal transmission From contact to merger 8.3.1 Lan gua ge shift: effects on the acquired lan guage 8.3.2 Lan guage maintenance/resistance: effects on the or iginal language

8.4

Hybridization in lan g uage 8.4.1

8.5

8.6 9

166

Hybrid utterances in code-switching and code-mixing

8.4.2

True hybrid l an gua ge s

8.4.3

Mixed languages

Creatio n of a new community or society Language death: selection at the societal level?

196 196 198 200 201 205

209

210

212 213

221

225

Towards an evolutionary linguistics

229

Glossary of terms

233

References Index of

authors

Index of populations (languages, communities and species) Index of subjects

245 262

267 211

iv

39

Chapter 1

Introduction

This book presen ts an evolu tiona ry framework for und erstandin g l an guage

chan ge , interp rets maj or

current theories in the context of this framework, and

makes ce rt ain new proposal s for aspects of the theory of language change .

Alth o ugh this is an ambitious ag en da , there has been much interesting work in

the th eory of l anguage ch an ge in the past two de cades , and I bel iev e the time is

ripe for an attempt to integrate the various advances and insights i n to the

nature of lan gu age change.

1.1

On theories of historical phenomena

Why do l an guag es change? This is a difficult q uesti on to answer. But p art of

the difficulty lies in our vi ew of the thing about which the question is be i ng

asked, namely, a lan guage .

Language change is a historical phenomenon . Hence the study of l an gua ge change - historical linguistics- must satisfy certain basic requiremen t s . The first requirement is th at one must clearly distinguish historical en ti ti es from th e types they represent.

A

his to ri cal entity is a spatiotemporally bounded indi­

vidual, that is, it is a specific entity that exists in a specific place for a specific

period of time. This is basically th e distinction between a t oken of someth ing and the type it bel on gs to. The distinction is Clear when describing p hysic al phenomena: a droplet of w a te r is a col lectio n of to kens of the molecular type H20. Tok en s of water can come .into existence and pass away via various

chemical processes. There are general chemical laws to describe the chemical processes und erg o ne by water. The di s tin cti on is not so clear when dealing with en tities such as languages . Cert ainl y, it is recogn ized that particular langu ages exist during a particular time and place. Proto-Gennanic arose somewhere in northe ast Europe (or

farther east) at a certain time, expanded its range, and came to an end with its breakup into wha t ultimately became English, German, Swedish, etc. More

p recisely , Proto-Germanic was identified as the language of a fi n i te speech

community, whose origin, spread and breakup determined the spatiotemporal boundaries of the histo rica l enti ty of Proto-Germanic. Its daugh ter la ngua ges , that is the languages used by certain oth er speech communities, in turn arose in cert ain locati ons , perhaps migrated or sprea d , and will eventually die ou t or bre ak up themselves.

2

Introduction

But this historical view is often set aside for a different characterization of a language such as German is a system of ru l e s and forms, divi ded into a phonology, m o rphol ogy , syntax, se mantics and so on (in varying ways depending on one's grammatical theory) . This system is treated as an idealized entity, a b s t racted away from any part ic ular speaker's mental knowledge or its uses in particu l ar discourse contexts. Is this en tity a token or a type? It is neither. The G erm an language system is not a token bec au se it has been removed from its historical context: one is not an alyzing the mental know le dge of actual speake rs, or actual occurrences of use. The G erman l an guage system is not a type because it is still a speci fic entity: German is still a specific language; its rules do n ot ap ply to langua ge in gener al but to the part icu l ar l anguage of German. To the e xte n t that its rules apply at all, it is only to an abstract yet p arti cu l ar entity that h as only an i de al existence. One con sequence of the idealization of a particular entity is that it gives the impression t h at there could be laws governing processes of chan ge in the German l an gu a ge system, that is, the impression that one of the t as ks of his­ torical linguistics is to predict what changes to the German language system will occur and at what time, just as one predicts w hat will happen to molecules of H20 when su bjected to electrolysis. Ye t this is paten tly not the case: we c an n o t p redic t what changes will happe n to the German language system. H20 is a type; the German language system is not a typ e. This fact has led some linguists to argue that historical linguistics cannot be a science because it can­ not provide a pr edi ctive exp lanatio n ( Lass l 980:xi). I sugge st that a reason why historical linguistics - and perhaps linguistics itself - as practiced by some might not be a science has to do with what the object of study is taken to be. When l i n guis ts analyze language as an abs tract system, they are not l o oki n g at a h ist orical entity, nor are t hey looking at a type about which predictions can be made. In fact, they are n ot l ookin g at anythi n g that is real at all, e ith er as a type or as a to ken . An emp irical science lang uage . In this characterization, a

must examine real, existing entities, and then construct generalizations about

what

types are involved, as well as principles and constraints go verni n g the structure and behavior of those types. In the study oflinguistics, the real, existing entities are utterances as they are p roduced in context, and speakers and their knowledge about their language as it is actually found in their minds. From these basic tokens, we can describe m o re co mpl ex tokens, such as a specific language or a s peech community; and then we ca n construct types, such as ' l an gu a ge' or ' grammar' (mental repre­ sentatio n) , over which con stra i nts and gen eral izati on s can be made. I am no t ar gu ing that in doing so, we ca n construct a predictive model of language change. In a ll probability we will not be able to make detailed predic­ tions, any more than historical sciences of natural phen omena, such as meteoro­ logy, astrophysics or geology, are able to do. There are two possible reasons for this, one 'o p ti mi stic', one ' pe ssim istic '. The ' optimistic ' one is that we simply do n ot know the facts i n p art icular cases in enough detail to predict the chan ges ; if we did know, we would be able to make pre di ctions (cf. KelJer 1990/1994:159).

On theories of historical phenomena

3

The 'pessimistic' one is that we would never be able to predict the change because there is at least some element of randomness in the process, as is the case with the randomness of mutation in biological reproduction and the random factors for survival of individual organisms. I am inclined towards the pessimistic view with respect to language change, which implies that even with perfect knowledge of the initial state, we would not be able to predict a language ch ange . But not all explan ati on s of histori cal p henomena need predict the outcome of individual cases. With some types of processes, wha t really matters is probabilities of change: the cumulative effect of the probability leads ultimately to an overall change. Historical explanations can be, and often are, probabilistic. Probabilistic explanations are particularly effect ive when the object of study is a population: a gene pool or a population of organisms, for example - or a population of speakers, or of utterances. Both speakers and utterances form populations, and that is what allows pro bab ili s tic mechanisms of language change to be effective explanations. Processes can be roughly divided into two types (see Hull 198 8 :410). One involves INHERENT CHANGE: a single object that exists over time changes in some way or other . An example of i nhe re n t change is human physiological dev elop­ ment over its lifetime. Another example is human li nguistic development, that is, the develop ment of mental structures that we i nterpret as repre sentin g lin­ guistic knowledge, over the lifetime of the speaker. The other type of process invo lves REPLICATION: the creation of a new entity that preserves in large part the structure of its p a re nt enti ty (or enti tie s ) . An example of replication is the creatio n of the DNA of the offspring of an organ­ ism, replicated in MEIOSIS. An other ex ample of repli cati on is the production of a linguistic utterance, which replicates grammatical structures of previously occurring utterances of the language, acco r di n g to patterns of knowledge in the minds of their producers. Another example which is often thought of as replication, but is much more indirect, is the l earnin g of grammatical know­ ledge on the part of a child, compared with the grammatical knowledge in the minds of its parents and/or care give rs . One of the problems with t reatin g language as an idealized abstract system is that it makes language change into an inherent process: a single object - the abstract language system - changes over time. But the re al , e x i st ing entities of linguistics are utterances and spe akers ' grammars. The evolution of both of those entities, pa r ticularly utterances, occurs through replica tion, not inherent c hange. Replication can result in ch a n ge at two levels. At one level, change can occur because the structure of the replicate is not exactly the same as the structure of the original . For e xamp l e , I m ay pronounce bad in an utterance with a sligh t ly hi gher vowel than in earlier utterances which I heard and inter­ n alized . I will call this ALTERED REPLICATION. Al tered replic ation produces variants of a stru ctu re . At the other level, change can occu r by the shift in the frequencies of variants of a s tructure . For example, more an d more utterances occur with a higher vowel in bad than with the lower variants. Perhaps bad

4

Introduction

with the lower vowel variants will eventually die out. This sort of change is called

DIFFERENTIAL REPLICATION

(Hull 1988:409).

The position taken in this book is that the study of language is about empirically real entities, not idealized abstract systems. The real entities of language are utterances and speakers' grammars. Language change occurs via replication of these entities, not through inherent change of an abstract system. In chapters 2-3, I will argue that the primary replicators are in fact utterances, not speakers' grammars; but the point that matters here is that a theory of language change must be a theory of replication of empirically real entities, either grammars or utterances.

1.2

Desiderata for a theory of language change

We are now in a position to outline some desiderata for a theory of language change. First, a theory of language change must avoid the reification or hypostatiza­ tion of languages. If one speaks of 'forces' causing a language to change, such as Sapir's concept of drift (Sapir 1921), then one is speaking as if language change is an inherent change applying to an abstract system. Descriptions such as '[a phoneme's] performance represents an extreme phonetic possibility as when it is an /i/ badly pressed by an invading /e/ with surrounding diphthongs which block all way of escape' and 'Isolated phonemes do not rush into struc­ tural gaps unless they are close eriough to be attracted' (Martinet 1952/1972:147,

159) are examples of reification taken to excess. Languages don't change; people change language through their actions (Croft 1990:257). Keller describes this desideratum as the principle of methodological individualism: 'the explanation [of a language change] is based on acting individuals, not languages, structures, processes, or collectives' (Keller 1990/1994:121) . Second, a theory of language change must explain why languages do

NOT

change in many ways, sometimes over many generations of speakers (see Milroy l 992b: 10-13). Many theories of language change focus their attention on mech­

anisms to bring about change. But if those mechanisms were the only mechanisms around, then languages would

be

changing constantly in all of their respects.

Yet they are not. A theory of language change must provide for mechanisms that act as forces for stability as well as for change, and ideally get the proper balance in order to account for rates of change. Another way of putting this desideratum is to say that a theory of language change must provide mechanisms of NORMAL (identical)

REPLICATION

as well as altered replication (and also mech­

anisms for nondifferential replication as well as differential replication). Third, a theory of language change must distinguish the two processes of change, that is, it must distinguish altered replication from differential replica­ tion. To use the terminology more typically found in linguistics, the two pro­ cesses are - and

INNOVATION

PROPAGATION

or actuation - the creation of novel forms in the language

or diffusion (or, conversely, loss) of those forms in the

Desiderata for a theory of language change

5

language. Both processes are necessary components of the process of language

that both are is very rarely recognized in models of language ch ange (but see Jespersen 1 922: 1 66-7). Because of this, some apparently contradictory p osit i ons have been taken on the nature of language

change. The distinction between these two processes, and the fact necessary components of language change,

change. Some linguists argue that

only innovation is

a language change. Joseph

writes, 'language change always takes place in the present, i.e. it always occurs in some speaker ' s

(or group of speakers) p resent' (Joseph 1992:127 ; see also can only be speaking of innovat ion , as pro pag a­

Joseph & Janda 1988). J o seph

a novel form does take place over long periods of time, exceeding the life spans of individual speakers. Others argue that only propagation con­

tion of

the output of a in the system, whereas of course a change is not a change until it has been ad opte d by more than one speaker' (Milroy l992a:79, emphasis original; see also Lab ov 1 982:46). Others do not r ecognize the distinction, leading to misunderstandings such as that in the fol lowing passage :

stitutes a language change. James

Milroy

writes, 'a change in

s in gle speaker might be regarded as the locus of a change

How can one ' un derstand ' or get at 'the intention behind' or discover the 'mean­ ing' of a shift from SOY to SVO, or a monophthongization, or a vowel-shift, etc., in any 'cognitive' or empathetic way, or 're-enact' them, or attribute them to 'reasons' and 'beliefs'? Especially as linguistic changes ... typically unfold over very long periods of time, most often beyond the lifetime of any human 'actor'. (Lass 1997:339)

Lass is criticizin g hypotheses about the innovation of a language change hypot heses appealing to spea ker intentions - by citing a fact about the pro­ pagation of a language change - propagation typically extends over many generations. Lass may be correct that innovation does not involve intention; but not for the reason he gives. Lass is not alone in this misun derstan di n g ; but neither do the advocates of such exp l an ations (in this case, Anttila 1 989) tend to distinguish i n n ovat i o n and propagation, thereby inviting the erroneous criticism.

One consequence of recognizing that innovation and p rop agati on are dis­ but jointly necess ary processes for language change is recognizing that

tinct

language change is

both

a synchronic

and

a diachronic phenomenon. Innova­

above: it given point in time . Propagati on is a diachronic phenomenon: it occurs sometimes over a very long period of time, e ve n centur­ ies. On the other hand, sociolinguistic research has demonstrated that one can observe propagation occurring in less than the lifetime of a speaker (see e.g. Trudgill 1988). Fourth, a comprehensive framework for understanding language change must subsume structural, functional and social dimensions of langu age change, or their equivalents . Most current approaches to language change address only tion is a synchronic phenomenon, as Joseph notes in the quotation

occurs in speaker action at a

Introduction

6

one or another of these dimensions. Sociohistorical l inguistics examines almost exclus ivel y the social dimensions (although Labov 1 994 discusses the phonetics and phonology of vowel shifts and mergers in some detail). Traditional philo­ logical and structuralist approaches focus on structure and to a lesser extent on function. Generative theories focus on structure exclusively, though largely because they deny function or social factors an y interesting or central role in language change. Granimaticalization t h eory focuses on structure and func­ tion, with a stronger emphasis on t he latter. To the extent that function and social factors play a role in language change - and there is strong evidence that both do - they must be integrated in a single framework. Last, a comprehe n si ve framework for un derstandi ng language change must subsume both INTERNAL and EXTERNAL causes of language change. Theories of internal causes are varied, and most attention has been focused on them. Externally caused changes, that is, changes caused by co ntac t such as borrow­ ing and substratum phenomena, tend to be discussed relatively little in theories of language ch ange , and are typically placed in separate chapters from internal causes in textbooks on historical linguistics. Contact-induced change appears to have an obvious source the other language. Nevertheless, mechanisms for the innovation and propagation of contact-induced change, preferably mechan­ isms co mpar ab le to those posited for internal changes, must be est abl i shed The framework presented in this book satisfies all five desiderat a for a theory of language change, and thus can be counted as a compreh en s ive framework for u nd e rstandin g language change. Howe ver it i s only a framework; it is · hoped that future work will be able to flesh out theories of specific types of ch anges in this framewo rk The last section of this chapter briefly outlines the framework. -

.

,

.

1.3

An utterance-based selectional theory of language change

The framework for understan d in g language change to be presented here is based on a generalized theory of selection for all t ypes of evolutionary phe­ nomena, originally developed in biology and applied to the history of science by David Hull (Hull 1988; see Dawkins 1976). The generalized theory of selec­ t i on described in ch apt e r 2, d i sting u ishe s variation (altered replication) from differential replication . It specifically provides a model of selection, which Hull argues is the mechanism for differential repli catio n for most types of biological evolution and also conceptual evolution in the history of science. Hull also emphasizes that re p licati on and selection involve historical entities (tokens), not types (except for very general types such as 'gene', 'species', 'population' and so on: see ch ap te r 2). The general ize d theory of selection thus satisfies the first three desiderata enumerated in §1.2. The framework presented in c hap ter 2 can be used to inte rp re t the major extant theories of language change; they are surveyed in chap ter 3. However, in ch apter 2, I also present the first of four major theses about the nature of ,

An utterance-based theory of language change

7

language change set forth in this book . I propose that UTTERANCES, more pre cis ely the replication of linguistic structures in u tterance s in language use, play a central role in the theory of language change. A wide ly held view treats language change as occurring in the process of 'replicating' a grammar in child language acquisition. There are however serious empi rical prob lems with this view, discussed in §3.2. For this reason, the Theory of Utterance Selection, which is more or less compatible with most aspects of soeiohis torical linguis­ tics, grammaticalization theory and the invisible hand theory (see §3 . 3), is pur­ sued here In the Theory of Utterance Se lec tio n CONVENTION is pl aced at center st age Normal replication is in essence conformity to con ven tio n in language use. Altered replication results from the violation of convention in language use. And selection is essentially the gradual establishment of a convention through lan­ guage use . Convention does not generally take center stage in linguistic theory, formalist or functionalist. The reason is that conventions are es sential ly arbitrary (see §4.2.4). Saussure considered arbitrariness to be central to the un derstanding of language: '[The arbitra ry nature of the sign] dominates all the linguistics of language; its consequences are numberless' ( Sa uss ure 1916/1966:67-8). Yet neither formalists nor functionalists have shown much interest in arbitrariness per se. Functionalists are ch iefly interested in the nonarbitrary, functi ona l ly motivated aspects of grammatical structure. Formalists find a rbitrariness useful in criticizing functionalist analyses, but they are chiefly interested in general, formal universals that have an innate basis. But a central aspect of a speaker's u se of language is convention. When I say Who did you meet yesterday?, I put the interrogative pronoun Who at the beginn ing of the sentence because that is the convention of my speech community, I know the conventions of my speech communit y , and my use of language will serve its purpose best most of the time if I c onfo rm to the conventions of my speech community. It may be that the initial position of Who is partly motivated by pragmatic universals of informa­ tion s tru cture , or partly specified by an innate Universal Grammar. In fact, on e (or both) of those factors may be the motivation for the origin of the convention. But that is not why I have put it there in that utterance. As Saussure wrote: '[arbitrary con ventions] are nonethe less fixed by rule; it is this rule and not the intrinsic value of the gestures that obliges one to use them' (Saussure 1916/1966:68). Convention - whether conforming to it, violat­ ing it, or establishing it - play s a key role in l anguage use and in language change. (It also plays a key role in our knowledge of lan guage ; see Croft to appear b.) Convention is a property of the mutual knowledge or COMMON GROUND of the speech community. Of course, common ground is found in the minds of speakers, albeit shared with other members of the speech community. Thus, there is an interplay between convention and in d ivi du al speakers' knowledge, or COMPETENCE as it is usually called . There is also an interplay between con­ ventional and nonconventional aspects of language use, which plays a critical .

,

.

8

Introduction

role in the understanding of how replication of linguistic structures in utter­ ances occurs. All of these concepts are discussed in some detail in chapter 4. The second major thesis of this book pertains to the nature of grammatical (and lexical) change. The causal mechanisms for innovation involve the map­ ping from language structure or form to language function, that is, meaning in context (Croft 1 995a; see §4 . 3 for some discussion of other senses of the word). This mapping occurs at two levels or interfaces. One is the mapping from phonological structure to phonetic reality (articulatory and auditory). The other is the mapping from grammatical (morphosyntactic) structure to its semantic/ pragmatic/discourse function in context. Neither mapping has been found by linguists to be simple to represent. But nor is it simple for speakers to represent these two levels of mapping. In chapters 4-6, I argue that altered replication is essentially a result of speakers adjusting the mapping from language structure to external function. I focus chiefly on grammatical change, with a few refer­ ences to sound change. The mechanisms for innovation in language change involve both structure and function . The mechanisms for propagation, on the other hand, are essen­ tially social, namely the various factors discussed by sociolinguists (see §3 .4.3, §7.4.2). In other words, there are two distinct mechanisms operating in lan­ guage change - this is my third major thesis (see also Croft 1 995a: 524; 1 996a:

1 1 6- 1 7). The mechanism for innovation is functional, that is, involves the form­ function mapping. The mechanism for propagation is a selection mechanism, in the evolutionary sense (see §2. 3), and it is social . Thus, the integration of structural, functional and social dimensions of language change is achieved largely by integrating the two distinct processes of change, innovation and propagation/selection . The last major thesis about language change presented in this book pertains to the relationship between internal and external (contact-induced) sources of language change, which is more complex than is usually assumed. In chapter 4,

I argue (following sociolinguistic theory) that a speech community is defined in terms of domains of use, not in terms of collections of speakers. All people in a society are members of multiple speech communities, whether those commun­ ities are conventionally described as representing a single language or multiple languages. In other words, all speakers command multiple varieties or codes, and thus some of the mechanisms for internal sources of change are the same as those for external sources of change. These mechanisms are discussed in chapters 6-7. The blurring of the line between internal and exkrnal changes also allows us to integrate the study of language contact and genetic linguistics. The naturalness of this view becomes more apparent when one recognizes that language 'speciation' is more like plant speciation than animal speciation. This view of language speciation is discussed in chapter 8.

Chapter 2

An evolutionary model of language change

The formation of different languages a n d o f distinct species, and the proof that both have been developed through a gradual process, are curiously parallel . . . Charles Darwin, The descent of man

2.1

Introduction

The relationship between lan gua ge chan ge and b io logical evolution has been debated since the emergence of l ingu i s ti cs as a s c ience in the n i ne tee n th cen­ tury, at around the same time as the eme rge n ce of evolutionary theory. The

deb ate has increased in recent ti mes . One can identify three separate ways in

which biological evolution has been connected to linguistic evolut ion in recent d i sc uss i o n .

First, interest has revived in the: evolution of language, that is, the evolution of the human lin g uist ic c apac ity (Pinker Kennedy

& Bloom 1 990; Hurford, Studdert­ & Knight 1 998; Kirby 1 999) . The evolution of human linguistic

capacity is direct ly a biological process: some ge netic change among ance s tr a l

primates led to the creation of a soc i a l a nd cognit i ve capacity for l anguage or a language-like system for communication, and some process selected those

primates with that capacity, lea ding to hum ans as a spe aking specie s . This

topic , wh i le interesting , is also ex tremely sp eculative , and will not be surveyed in this book: this book is concerned with l an guage change itself, not the evolu­ tion of a certain biological capacity of human beings. Second, interest (and co ntr oversy) has arisen over the so-called geneti c origin of con t e mp o rary human l an gu ages. Here the evolutiona ry connection

is one of histori cal association. The _intern al structure of ge netic families such as Austronesian are c omp a re d to the distribution of biological traits, such as

alleles in mitochondrial DNA, or blood types (B ellwo od 1 99 1 ) . It is assumed that, for the most part, transmission of biological traits through o ffspring is

historically paral J e le d by t ransmi ss ion of language from parents to ch i ldren ,

and hence famil y trees of hum a n communities based on biological traits sho ul d rou ghly parallel family trees constructed on linguistic evidence.

Of course , all know that this parallelism in the history of languages and of

human b i ological traits is not n e cessary. Languages are not transmitted via an

individual 's DNA; biological traits are. Rather, lan gua ges are transmitted to new speakers through exposure to their use . A group of people may abandon 9

l0

An evolutionary model of language change

their language and adopt one of another group to whom they are not biologic­ ally closely related. For more recent families such as Austronesian, where the demographic history is better known, the relative contribution of language shift vs parent-to-child transmission can be more easily sorted out. In Austronesian, for example, it is clear that Melanesians are biologically closer to Papuans, and presumably have shifted to the Austronesian languages that they now speak (cf. Melton et al. l 995; Redd et al. 1 995). Relationships have also been observed between proposals for historically deeper linguistic families such as Amerind (Greenberg 1 987) or even Proto­ World (Bengtson & Ruhlen 1 994) and proposals for historically parallel bio­ logical phylogenies (Greenberg, Turner & Zegura 1 986; Cavalli-Sforza et al. 1 988). These are much more controversial, chiefly because of challenges to the linguistic classification; the classification based on some biological traits (in particular mitochondrial DNA) is also questioned. Again, this relationship between biological evolution and language change will not be surveyed in this book : this book is concerned with the mechanisms of language change, not the origin and spread of specific families of languages . The third connection between language change and biological evolution is found where the theory of biological evolution itself has been adopted, or adapted, in order to construct an evolutionary theory of language change (see e.g. Kel ier 1 990/ 1 994: 1 4 1 -52; McMahon 1 994: 3 1 4-40; Lass 1 990, 1 997 passim; Ritt 1 995). Evolution is recognized as a process that occurs with certain types of entities. The process is probably best understood as it occurs with populations of biological organisms; that i s evolutionary biology. The hypothesis is that language change is an example of the same process, or a similar process, occur­ ring with a different type of entity, namely language. It is this hypothesis that forms the starting point of this book. A number of approaches have been taken to an evolutionary model of language change. The first approach is literal : language is a genetic capacity, and hence obeys certain principles of biology. This approach is associated with Chomskyan linguistics, because Chomsky argues for the biological basis of quite specific linguistic properties (e.g. certain syntactic structures and constraints) . The literal approach also makes developmental claims: for instance, the hypothesis that the human language capacity in all its detail emerges in maturation . However, the main goal of the literal approach is to claim a biological basis for the universal properties of languages. The ways in which contemporary human languages are divergent, and h ave diverged or will diverge in history, cannot be accounted for in the literal approach. A literal approach to language diversity would amount to claiming that the differences among languages reflect genetic differences among their speakers. This is patently false, as can be seen from the aforementioned fact that a person can learn a second language, and learns whatever language is spoken in their surroundings. For this reason, the literal approach generally turns to questions of the evolution of the human linguistic capacity, that is, what gave us the genetic basis for the properties common to all languages, whatever those may be.

11

Introduction

The second approach is essentially analogical: there are analogies between certain biological processes as described by evolutionary theory and certain processes of language c h ang e that call for description . H enc e linguists seeking better descriptions and analyses of those processes can borrow or adapt the descriptions and explanatory mechanisms that evolutionary biologists have proposed. However, no deeper claim is made about the relationship between the theory of evolution in biology and the theory of language change in linguis­ tics. There are simply analogies or metaphors between a process in one domain of scientific study and a process in another domain; in particular, one should not push the analogy too far. In the analogical approach, the relationship be­ tween evolution and language is essentially opportunistic - an opportunity for linguists to utilize some already developed theoretical constructs. An example of the analogical approach appears to be the use of the biological metaphor in creole studies, e.g. Whinnom ( 1 97 1 ) and Mufwene ( 1 996a, to appear). Whinnom suggests that the biological concept of hybridization can be applied to language contact 'provided that the analogies are properly applied' (Whinnom 1 97 1 : 9 1 ) . Mufwene compares languages to species and the factors that determine a language's survival or extinction as ecological factors, but states: ,

I do not want to suggest t h a t language evolution is i n all, or most, respects like

species evolution . . . There are, however, some similarities between the concepts

of language and species, which I find informative and would like to use cautiously to shed light on the process of language evolution. (Mufwene to appear, fn .

I)

I t appears that Lass's adoption of an important concept in recent evolutionary theory, exaptation (Lass 1 990), and its application to historical linguistics, is another example of the analogical approach to the relation between biological evolution and lan guage change (see § 5 . 3 for further discussion) . Lass writes: 'while claiming that the notion of exaptation seems useful in establishing a name and descriptive framework for a class of historical events, I remain fully aware (even insistent) that languages are not biological systems in any deep sense' (Lass 1 990:96). He adds, for instance, that 'There is as far as I am aware no storage or coding mechanism for linguistic transmission equivalent to DNA' (ibid .). However, Lass makes it clear in the following paragraph that he is not taking an analogical approach. It is not that languages A RE biological systems. It is that languages and biological systems are instances of a more general phenomenon: 'rather than extending a notion from biology to linguistics, I arr. suggesting that the two domains . . . have certain behaviors in common by virtue of evolving' (Lass 1 990:96) . In fact, Lass is taking a third, generalized approach. This approach proposes a generalized theory of evolutionary pro­ cesses, which applies to the evolution of species and their traits in biology, to language change in linguistics, and to other domains as well : 'I am convinced . . . that there is such a th in g as a theory of "historically evolved systems" ' (Lass 1 997 : 3 1 6) . In the generalized approach, there is a profound relationship

12

A n evolutionary model of language change

between biological evolution and language change, which is worth exploring in greater detail. The two are not identical by any means. But they both display salient properties that demonstrate that they are instantiations of the same generalized theory that crosses disciplinary boundaries. Even in the generalized approach, the evolutionary biologists are again in the forefront. La ss cites Dawkins in support of this approach. Dawkins' popular book The selfish gene (Dawkins 1 976), besides making specific proposals about the nature of biological evolution, suggests that his model can be applied to cultural evolution as wel l . He proposes the notion of a MEME to represent a sociocultural unit that can evolve via differential replication. David Hull, a biologist who became a philosopher of science but not without continuing to make contributions to systematics and evolutionary theory, develops a gener­ alized theory of selection which subsumes both biological and conceptual evolu­ tion (Hull 1 988, especially chapters 1 1- 1 2). In t his chapter, I will argue that Hull's model can be applied to language change, and allows us to construct the foundations of a theory of language change and the major mechanisms that bring it about . If the generalized appr oach is to be taken seriously, then Lass's claim - that there is no equivalent of DNA in linguistic evolution - raises a difficult ques­ tion. If Lass's claim is true, then are linguistic and biological evolution really instances of the same thing? Does DNA play an essential role in the theory of evolution developed by biologists - developed in far greater detail than the theory of language change in linguistics? If so, then the role in evolutionary theory that is played by DNA in biological systems must have a counterpart in linguistic systems. Lass argues against the literal approach, that is, that there is a genetic basis to the phenomena of linguistic evolution: language change does not occur through biological genetic mutation and selection. But that does not necessarily mean that there is no functional equivalent to DNA in linguistic evolution. Hull writes, 'People reject selection models in conceptual change out of hand because they have a simplistic understanding of biological evolution ' (Hull 1 988 :402) . The trick is making the right instantiation of the theoretical constructs of the generalized theory. In this chapter, I will argue that t he re is an e q uiv alen t to DNA in linguistic evolution, and that it is the utterance. Both the existence of an equivalent to DNA and the entity I am proposing as the equivalent will at first strike the reader as surprising, perhaps even bizarre . But an essential role in the generalized theory of selection is assigned to a function most typically centered on DNA in biological evolution. I will argue that this function is most typically centered on the production and comprehension of utterances in language change. It should be clear from the wording of the preceding sentence that the DNA­ utterance analogy is going to be rather indirect and not the one that Lass rightly rejects. In fact, what will emerge from the application of Hull's theory to language change is a theory that is thoroughly based on what happens to language in use, from the origin of an innovation to its adoption as a conven­ tion of the speech community, and which unifies internal and external causes

Populations and phylogenies

13

change. The remainder of this book will e xplore some of the con­ sequences of this theo ry . But first we mu st explicate the two most fundamental notions in the gene rali ze d theory, the popu lati on theory of species and t h e ge ne ralized theory of selection.

of l angua ge

2.2

Populatlons and phylogenles

maj or advances of the so - c a lled ev olutio nary syn the si s is the re­ of the essentialist theory of species by the population theory of spe cie s (Do bz h a n sky 1 937, Mayr 1 942, cited in Hull 1 98 8 : 1 02; see also M a yr 1 982). The population the ory of speci es is sometimes trace d back to Darwin, but Darwin was not entirely clear i n his defi n i t i o n of species (Mayr 1 9 8 2: 265 69; Hull 1 988:96, 2 1 3 fn . 2; see Mayr 1 982:272 for other precurs ors) . The

One of

the

placement

population theory is also inextricably tied up with the question of systematics,

that i s , the t axo no m ic classification of or gan isms into varieties, species , genera, h i gher taxa. Here the contrast is between classification and phylogeny; we must di scu ss this question as well . Jn the ESSENTIALIST view of a sp ec i es , each species has immutable essential st ructur a l properties that identify it (M ayr 1 982:256). Th at is, the essentialist view is that a s pec i e s instantiates an abstract type . The essentialist view ran families and

into problems due to various sorts of structural variation among species, includ­ ing

high

degrees

of structural variation

among individuals in

a

population

and

also among different life - st a ge s in: an individual in a popul ation (for e x ample, a caterpillar and the butterfly i t t urn s into, or a species that changes sex over its

lifetime; H ull 1 988:430) . The esse nti alist view also ran into pro blem s with pop u l atio ns which could not be distingui shed by structural features b ut were dist i n ct rep ro duc t ive communities (Mayr 1 982:27 1 ; se e below) . But the greate st pr o b l em for t h e essentialist view of a speci e s is that a specie s evolves, and in so ev olvi n g , can lo se 'essential ' structural prope rties . Identifying t his pro b lem is one of the major con tri bu t i o n s of Darwin to evolution ary bi o l o gy . The POPULATION theory of s pecie s is co m pl ete ly different from the essen t ialist theory (Mayr 1 982:272). A spec i es consists of a populati on o f i n te rbreedin g individuals who a re REPRODUCTIVELY ISOLATED from other populations. This property - i n t e rbre e d i n g , a n d l ack th ereo f be twee n species - is the 'essential' property the individuals have in common. Th e re is no essential species type. Individuals can vary in e n o rm ou s ways in phys i cal st ruct u re ( a n d be havior) , but as lo ng as they form a population in the evolutionary sense, t he y are members of the same s pec i es . Conversely, individuals may be s tructurally extremely similar, but if t hey come from two distinct reproductively isolated pop u la tions , t h e y are members of different species . This is a r adic a lly different view of th e species as a conceptual category. The category de fi n i t i o n is based on a speci fic set of i n d i vi dua ls , and category membership is defined in terms of h o w the individuals interact with each o th e r , n ot by an y specific traits ass ociated with all and only the individuals in the category.

14

An evolutionary model of language change

Not every individual need breed with every other individual in order for the set of individuals to form a population; only reproductive isolation is necessary: 'extensive interbreeding with the population system is not an essential property of biological species; non-interbreeding with other population systems is' (Grant 1 98 1 : 9 1 ) . Although reproductive isolation is treated in theory as a sharp divid­ ing line, it is not entirely so in fact (Hull 1 988: 1 02-3). There are cases in which populations which were separated and then brought into contact again developed a stable hybrid region in between the two distinct populations . Hull gives the example of the hooded crow and the carrion crow in Europe: separated by glaciers which then receded, the species populations remain distinct, but there is a stable band of hybrids in a zone not exceeding 75 to l 00 kilometers in width (Hull 1 988: 1 03). Hence there is interbreeding where there is contact, but there is little gene flow between the two populations. Conversely, reproductively isolated populations of plants can merge : 'estimates of the proportion of plant species in general that are of hybrid origin run as high as 30 or 40 percent' (Hull 1 988: 1 03; cf. Grant 1 98 1 : 203). In the essentialist view of species, a species is a type, defined by a set of properties, that is not located in space or time but in an abstract domain of biological traits. In the population view, a species is a spatiotemporal individual, not an eternal essence. The population of organisms constituting a species is circumscribed by the region in time and space collectively occupied by the individual members of the species. The beginning of a species is defined by its branching off in a speciation process, and its end by either its extinction or its fission into two or more new species in speciation: 'Just as the name "Gargantua" [an individual goril la] denotes a particular organism from conception to death, " Gorilla gorilla" denotes a particular segment of the phylogenetic tree' (Hull 1 988:2 1 5) . A species, like a n individual organism, is a historical entity in the popula­ tion view. In the essentialist view, a species is not a spatiotemporal individual: it is a type or kind, whose instantiations may be particular individuals, but the kind is not spatiotemporally bounded itself. But species are not types; they cannot be types. The essentialist view treats a historical entity as if it were a type . In the population view, only entities as abstract as species (in general), and certain theoretically defined subpopulations of a species such as demes (in general) or geographical races (in general) are kinds. Any particular species, deme or geographical race is a spatiotemporal individual. In other words, a population is a fundamentally historical entity. This point is a very important one in understanding both the relationship between populations and classifica, tion and between populations and selection (see §2.4. l ) . If the population theory o f species is distinct from the essentialist theory o f species, then one would expect t o find cases where there are mismatches i n the world between species defined in terms of reproductively isolated populations and species defined in terms of essential structural properties. In fact, this is the case (see e.g. Hull 1 988: 1 04) . SIBLING SPECIES are two reproductively isolated species whose structural descriptions overlap to such an extent that on an

Populations and phylogenies

15

essentialist definition, they would be the same species . For example , five different species of the flower Gilia in the M oj ave Desert ar e so similar that they were once classified as a single species, but the five specie s are h ighly intersterile (Grant 1 98 1 :6 1 -2). PoLYTYPIC SPECIES, on the other hand, are species that are s tructura l ly so hete ro gen eous that an essentialist would be hard put to categ­ orize them as a single species, yet they form an interb reeding popu l a tion (in terms of gene flow; M a yr 1 982:287-92). A p o p u lation may sp li t into two or more parts, often through ge ogra p hic a l isolation . The t e rm VARIETY is generally used for a subpart of a species popu l a­ tion. Varieties may cease to interbreed. In fact the two populations may diverge in structure and behavior such that they could no longer interbree d even if brou gh t together ag ain. At this poin t one would say that the ori ginal species has sp l it into two d au ghter species. (I follow Hull 1 988, and Hennigian system­ atics in gen e ral , by assuming the old species no longer exists after such a split, thereby avoidin g the pse u doproblem of deciding which of the daughter species is 'really' t h e continuation of t h e parent species.) 'Varieties are merely incipient species. Not all varieties become species, but all species a t one time were varieties' (Hull 1 988:96). If one constructs a his to rical account of the s p littin g (and also merging) of p op u lat ions of organisms , the result would be a PHYLOGENY. A phy l o ge n e t ic classification is intended to reflect the history of the organisms being classified. A phyl oge netic classification is not the same as a taxonomic classification, the familiar classification of species into genera, families and so on . A taxo n omic classification is based on similarities and differences among traits. l t is basic­ ally founded on an essen tialist view of species. A phylogen eti c classification is historical. Yet the two h ave often been confused, even by bio l ogical system­ atists (taxonomists). There are two reasons for this confusion . The first is the appare nt similarity b e twee n a phyl ogen etic tree and a taxonomic tree: 'The rela tion s h i p between a b ra n chi n g phy l ogenetic tree and the s uccessive subdivi­ sion s of a hi e r a rchic a l classification could not seem more patent. Yet, it is no t ' (Hull 1 988:98). The second is due to a p ractical problem in determin i n g p h ylo ge ni e s . It is n o t always pr a ct i cally possible to discover patterns of gene flow in populations. In p articular , it is i mp o ssible if the specie s is extinct and one must rely on fossil evidence . Instead, o n e examines traits, and uses th e differences in traits among popu latio ns to hypot h esize a historical scenario of the successive s plittin g (and merg in g) of po p ul a tions. Since one is usin g traits to projec t a p h yloge ny , it is tempting to use a t a xon o mic classification to project the phylogeny. But a proper phyl o geny requires t he differentiation of tr a i ts based on their history. If two taxa share a trait, it could be a retained trait from the p arent population (a SYMPLESIOMORPHY) , or it could be a sh a red innovation of the two taxa (a SYNAPOMORPHY ) . Only a shared innovated trait can justify grouping the two tax a together phylogenetically. A shared re t a i ned tra it si mply indicates that the two taxa may be grouped together phyloge ne t ically with other taxa at a greater time depth . And of course th e shared traits may be shared accidentally, that is,

16

A n evolutionary model of language change

the trait arose i ndepen d en t ly in each pop ulat ion and hence they do not form a phyloge netic group (see Lass 1 997: 1 1 3 -1 4 for a brief introduction t o these terms and concepts in a h i s t orical linguis tic context). In sum, populations and p h yloge nies represent historical entities and evolu­ tion, while the essentialist view of species and taxonomic classification represent a v iew of species and groupings of species that is not historical or evolutionary a n d a t bes t provides indirect evidence for hist ori c al phen omena. In develop ing an evolutionary theory of language change, or of any historical process invol­ ving populations and thei r divergence (or merging), the concepts of po pulations defined by interaction of individuals (or lack t hereof) and phylogenies are the relevant ones All of the phenomena described above are directly relevant to linguistics . A genetic lin guis tic classification , the family tree model, is intended to be a p hylogeny of languages (although some linguists retain certain essentialist views; see §8. 1 ). Shared innovations are crucial to establi sh i ng a gene t ic linguistic grouping; shared retentions represent at best some higher genetic grou pi n g There i s of course the pos sibi l i ty that apparent shared cognates are accidental. Historical linguists have also adopted the conven t io n of gi ving daughter lan­ guages different names from the parent lan guage and generally assume the daughter language is a n ew language (compare the naming of daughter species in Hennigian systematics alluded to above). The mismatches occurrin g between the essentialist and p op ul ati on defini­ ti on s of species are also found in languages. These are the standard examples of the p roble m in defining la nguage and dialect (see e . g. Chambers & T rudgill 1 980, chap te r l ) SIBLING LANGUAGES are two linguistic varieties that are structurally so similar that th ey are considered to be 'dialects of the s ame language' , yet are perceived by the speakers - or at least by one group of speakers - as disti nct languages. E x am p l es of sibling l angu ages (of varying degrees of co n trove rsi a l ity ) include Macedonian and B ul ga rian Danish and N o rwegian Serbian and Croatian, Hindi and Urdu, and M alay and Indonesian. Dixon ( 1 980:33 -40) po in ts out that many i n s tances of neigh b o ring l anguage s in traditional, small, decentral­ ized , nonliterate societies such as Australian a bo ri gi nal societies involve what I call sibling l angu ages In some cases the p ercep tion of the sibl in g l anguages as distinct is no t reciprocal : many Bulgarians tend to see Macedonian as a dialect of Bulgarian, but the reverse does not hold . Of course, this reflects different perceptio n s about the s oci al and p o litical separateness of the communities tha t s pe a k these linguistic varieties. PoLYTYPIC LANGUAGES, on the other hand, are l i ngu i stic varieties that are structurally so diverse that linguists wou l d characterize them as different lan­ gu ages, yet their spe a kers perceive them as dial ects of the same language . E x am p l e s of p olytypic languages include the Chinese so-called dialects: they are mutually unintelligible (Li & Th o m p so n 1 98 1 :2), but the writing system and pol itical unity ten d s to i m pl y identification as a single language (Norman 1 988: 1 -3). Another example of polytypic languages is found in DIGLOSSIA ,

,

.

.

,

'

'

.

,

.

,

Populations and phylogenies

17

(Ferguson 1 959/ 1 972), where there are two related but mutually unintelligible varieties, the L[ow] variety being a vernacular and the H[igh] variety a written standard, as in German-speaking Switzerland or the Arab countries. Speakers perceive H and L as a single l an guage L o ften being perceived as a substandard or imperfect version of the H variety. The same is true of postcreole continua, where the creole basilect a n d s tandard la n guage acrolect are mu tual l y unintel­ l igible, as with Jamaican creole and Standard Jamaican English (Decamp 1 97 1 : 350). The structural d iversi ty of t rad i t io n al dialects of English, German, Italian and other we stern E u r ope a n languages may be instances of a lower degree of polytypy, depending on the de gree to which their speakers id en tify themselves as s pe akers of English, German, etc., albeit nonstandard speakers Turning to the crux of the matter, the confl icting definitions of species, we find a parallel clas h of definitions of a la n gu age The li ngui stic or STRUCTURAL de fin i tion of a language - if two v arieties share enough struc ture in commo n (phonology, grammar or morphosyntax, l exicon) then they should be classified as part of the s ame lang u age - corresponds to the essentialist definition of a spe cie s Comparison of linguistic varieties based purely on struct ura l prop erti e s leads to assessments of l a n g u age vs d i a l ec t based on essentialist criteria. The structural definition of a l a ng uage possesses the same flaw as the essentialist definition of a species. As I arg ued in § 1 . 1 , the structural definition of a language makes a type out of a historical e nti ty As with spec ies languages evolve over ti me undermin ing the structural d efi n iti on ; and sibling languages and polytypic languages demon st ra te further p robl ems with the structural/essentialist defini­ tion of a l an guage What is the lingu istic equivalent of the pop ul a t i o n definition of species? Chambers & Tr u d gi ll ( 1 980) offer a social definit i o n of l a n guage as an altern­ ative to the structural definition (see also Haugen 1 968/ 1 972). T h ey define an AUTONOMOUS VAR IETY as one that is perceived by its speakers as a distinct language no matte r how similar it is structurally to some other variety . A HETERONOMOUS VARIETY, on the other hand, is perceiv ed by its speak e rs as being t he s am e l a n guage as that of an au t o n o m o us standard vari e t y, no mat­ ter how st r uc t u ra l l y distinct th ose va rie ti e s are. We may apply Ch amb e rs & Trudgill's definition to t he cases of sibling l ang u age s and polyt ypic l a n gu a ge s Serbian and C ro at ia n are exam ples of si bl i n g languages. Serbian is autonomous fr om Croatia n b ec a u se Se r b i a n sp e a k e r s pe rceiv e their language as distinct from Croatian and vice versa. M ode rn Arabic is an example of a p olytypic lan­ guage. The colloquial v a r i e tie s of Modern Arabic are heteronomous, because speakers of the modem colloq uial varieties perceive their col l o q ui al variety as a ve r si o n of Arab ic Chambers & Trudgill's soci al definition of l a n gu a ge closely corresponds to the popul ation definition of species . However, their definition is based on speaker perceptions (and the existence o f a st a n dard v a r i ety) The genuine equivalent to the popu lati o n th e ory of spe c i es for a language must be in terms of actual communicative interaction . This does not imply that every speaker of a socia l ly defined langua ge speaks with every other speaker of that language, any more ,

.

.

,

.

.

,

,

.

,

.

.

.

18

A n evolutionary model of language change

than every organi s m of a s peci es mates with every other organism of that in its l i feti me (see above ) . I t merely implies that every speaker perceives eve ry othe r spea k e r as someone he or sh e should be able to commun icate with by using what t he y perce i ve as the same l angu a ge . The last remark illust r a t e s a parallel with one of the qu al i ficat i ons of the population t heo ry of species: not all members of a population must n ecessar i l y interbreed in order to be deem e d members of the population. Not surprisingly, there are pa rall el s with the phenomenon of different degrees of re productive isolation (the European crow example descri bed above) . C h a m b ers & Trudgill note that there is a hyb rid area in East Anglia between t he No rthe rn Englis h vowel [u} and the Southern English [A], one of the most salient p honologic a l m a rkers o f Northern vs Southern English (Chambers & Trud gi l l 1 980: 1 29-37). In the hyb rid area, one finds MIXED a nd FUDGED variet i es (lects in their terms; see §7 . 4 . 4) . Mixed varieties possess [u] in some words and [A] in other words; fud ge d v arieties use a phon eti c ally intermediate vowel such as ['\'] in so m e words. The mixed and fudged va ri e ti e s form a t rans i ti o n zone between the N o rth e r n and Southern English varieties. Similar transition zones are found with bundles of ISOGLOSSES (geographical boundaries between one l i ngu is t ic feature and another). For example, t he boundary between French and Occitan is defined by a number of lexical and grammatical features. But t h e i sog l osse s for each feature do not match p erfectly: there is a transitional zone across the middle of France where varieties possess some ' French' features and some 'Occitan ' featu res (Chambers & Trudgi l l 1 980: 1 1 1 , Map 7-6, after Jochnowitz 1 973). If we purs u e an evo lutionary theory of language following the lead of the evolutionary theory of biology, then we must take the population (social) defini­ tion of a language as the basic one. A s t ru c tu ra l i st linguist may feel uncom­ fortable about the social definition of a l a n g uage But the fact is that the social definition is the correct one from a historical per s pec tiv e , in terms of ca u s a l mechanisms of l a n g u age speciation. The social d e fi n i t i on make s p re d i c tions of likely h i s to ric a l d e v el o pm ent s whereas the structural definition does n o t Si b l i n g s peci e s are l i k e l y to diverge morphologically as their reproductive i solati on continues ( see Hull 1 988:66-7, d i sc u s si n g Mayr's t he o ry of speciation). Like­ wise, sibling l ang ua ge s a re l i kely to d i ve r ge structurally as their commun icative i solation p e r si s t s A po ly t ypi c s pe c i e s may break up if the ge n e fl ow is inter­ rupted, or possibly become more homogeneous or at least maintain itself as a single language. A pol y t yp i c language may break up if its social un i ty is b rok e n - this appears to be wh a t is happening in the distinct modern Arab nations. Or it may survive as a single l an guage as in Ch ina, possibly be co m i ng m ore homo­ gene o us as with the loss of the tradit i ona l dialects of western European lan­ guages. Social and communicative isolation leads to structural dive rgen ce; social and communicative intercourse leads to a maintenance of the s t at u s quo, or even convergence (w h i c h itself is a result of tighter so c ia l cohesion and mobility ) . The l a st few paragraphs have surreptitiously i n t r od uce d the l i n gu i s tic equi val­ ent t o re prod u c t ive i so l at i on : COMMUNIC ATIVE ISOLATION . C o n verse l y , l in g u i s ti c spec i es

'

'

.

.

.

,

Populations and phylogenies

19

in te rb reeding is commun ic at i on . The p o p ul a ti on definition of a l a n gu age thus appears to be very similar to the notion of mutual i ntell i gi b il i ty used to di s t in ­ guish l an gu age s from d ia lects . However, mutual in telligib i lity tends to be defined in te rm s of potential c o mmun i c at ive interaction , whether or not the s pe a k e rs belong to the s am e speech community. Communicative i n te rac tion d e pends not only on the degree of s truct ura l similarity of the varieties spoken , but also on the social behavior of the speakers. Serbian and Croati an are mutually intel­ l igibl e to a high degree , but many speakers do not communicate with the opposite community due to the recent political changes in forme r Yugoslavia. One must d istin g u i s h the re a l po te nti a l of communicative interaction of two members of t h e same speech community who ha ppe n not yet to have con ­ v e rse d with each other, with the abstract potential of co mmu n ic ativ e interac­ tion of members from different spe e ch communities who w ou l d n o t norma l l y talk with e ac h other, except in a d i a l ec t o l o gist 's ex pe r i menta l situation . A n analog t o the latter situation in bi o l o gica l p o p ul a t i o ns is, fo r ex amp le , plant s p ec i e s which could p ro d uce hyb rids i f t hey interbred, but do not interbreed because one s pec ies is p ollin a te d by certain bees in the e a r l y morn i n g and the other by o t he r bees in the late afternoon (Grant 1 98 1 : 1 1 3). Only th e real poten­ tial of interbreeding/communication matters for the population. definition of a

species/language . It is worth descri b in g the two other sorts of bio lo gi cal populations men­ tioned earlier in this section in a l i tt le more detail. A SPECIES is a p o p u l at io n of interbreeding individuals. A GEOGRAPHICAL RACE is a sub p o pu l at io n of a spe­ cies wh i ch is defined geographically, and often has structurally div erged to a sligh t extent, but presumably not so far as to preven t interbreeding. A DEME consists of o r ga nism s i n s uffi ci e n t proximity to each other that they all have equal probability of m a t ing with each other and prod ucing offspring, p ro vi de d they a re sexually mature, of the opposite sex, and equivalent with re s pe c t to sexual selec­ tion . To the extent that the se condit ions a re met, the o rg a nis ms be lo n gi n g to a deme share in the same gene pool. Of course, in natural populations, s o m e mating occurs between

adjacent

demes, and not all organisms within a single

deme have precisely e q u a l probability of mating, but the isolation between demes

is met often e n o u gh and well e n o u gh for d e m es to play an important role in biological evol u tion . (Hull 1 98 8 : 4 3 3 )

These different types of popu l ation s are also relevant to th e notions of l an guage , dialect and speech community, defined in terms of co mmun ic at ive in teraction and social iden tity rather than in the es sent i a l i st terms of l inguistic str ucture . A lan guage and its spe akers should be defined in p opu l ation terms j ust as spe ci e s gen e r a l l y are . A ge o gr ap h i c al race is a traditional geo gr aph ic al d i a le ct : de fi n e d geo g r aph ic al ly , slightly divergent s t ruc tur a l ly , b ut no t eno ugh presumably to prevent c ommu n i c ati on (i .e. i n te ll i gi b i l i ty ) or to pro vide a se pa r ­ at e soci ol ingui stic id e n ti ty, assu m i ng we are not dealing with s i b l i n g l a n gu age s . A deme is related to one definition of the c omp lex notion of a s peech com­ munity . In fact, the term speech com munity as it is broadly used is the proper

An evolutionary model of language change

20

li ngui s t ic eq u i va len t of a b i ologic a l p o p u l at i on (see §4. 2 . 3 ) . A s peec h commun­ ity can be defined as b ro adly as all of En gl i s h no matter where it is sp oke n , at a n intem1ediate level suc h as H ibemo-English, or as narro wl y as a pa rt i c u la r fai rly cohesive social network such as the ones analyzed by the M i l r oys in Belfast ( M il r oy 1 987). A SOCIAL NETWORK corresponds most closely to a deme: a group of people who are most likely to communicate with each other, a n d not so much with those outside the network . One c an describe the re s u l ts of the Milro ys research in Belfast r at her well by paraphrasing the Hull quo t ation an d m a k i ng t h e ap p r op ri a te substitutions of s ocio l ingui s ti c terms for bi o l o gic a l ones (see ch apte r 7): '

a s oc i a l network consists of s pe a k e rs in s uffic ie nt proximity to each other th a t

t hey all have equal pr o b a b i l ity of c omm un ica tin g with each other, if they have

some reas o n to linguistically interact. To the extent that these conditions are met, the speakers belon ging to a social network share in the same language. Of course, in natural speec h communities, so me communication occurs between adjacent social networks, and not all individuals within a single social network have precisely equal probability of c ommun ic a t in g with each other, but the iso l at i on between social networks is met often enough and well enough for so c i al networks to play an i m p ort an t role in language change .

S pec ies are formed o n l y when a pop ul a ti on rep rod uces sexually. Asexual form s pecies: each asexual or ganism is re prod uct i vely isolated (Grant 1 9 8 1 :64; Hull 1 98 8 : 2 1 5) . Communicative i nter acti o n is se xu al : lan­ gu a ge is p r od u ced when a s pe a k e r communicates wi th a hearer. Even writers pres uppo s e an audience. Hen ce , languages and their speake rs form populations o r ga nis ms do not

'

'

in the biological sense . The metaphor found in the word 'in tercourse ' (sexual or l i n gui stic ) is not an accident. This parallel should make the DN A-utterance equivalence to be introduced in §2.4. l a little more pla u sible But firs t we mus t revie w cer t a i n recent de vel opme n ts in t h e t heo ry of selection in biology. .

2.3

The genera lized theory of selection

The t heory of se l ec tio n pr ovi d e d by

s ubject

of criticism

and

the

neo-Darwinian

syn thes i s

has been t he

mod i fi c atio n in recent decades. The neo-Darwinian

theory of selection is basically that or ganism s that are better a d apte d to the

ecol og i ca l condition s of the e n v i r onment will have a gre a ter likelihood of sur­ vival a n d rep r o duct i o n - i .e. are selected. One cr i ti ci sm directed towards the neo-Darwinian theory of selection is the ro l e of adaptation in selecting indi­ vi d u a ls in the population. Other me ch ani s m s for se lectio n besides the st a n d a rd a d a ptive one have been p rop o sed One such me cha n i s m is e x apta t io n : some trait wh i c h evolved for o ne purpose, or evolved for no apparent purpose at all, is e x a p ted to serve som e other function which bestows a competi ti ve advantage on its possessor (Lass 1 990; see §5.3). This particular application of evolutionary .

The generalized theory of selection theory to historical lingui stics seems qu ite app ropri ate , and

a

strong

Lass

takes it

2l

to be

ar gument for the position tha t each in s t antiates a generalized t heo ry

of evolution .

We wi l l

concern ourselves here with another, pe rh a ps more profound,

cri ­

tique of the theory of selection, that concerning the unit of selection. In the sta n dard view found in the neo-Darwinian synthe si s of evolutionary theory in

biology, it is the o r ganism that is the unit of selection. Se lec tive processes, of whatever sort, operate on the level of th e fitness of th e

organism.

Alt hou gh it is

ge ne tic material that is ultimately repli c a t ed and then generates a new o rg a n ism

in reproduction, it is the o rga ni sm which is

ultimately selected in

the ev ol u tio n ­

ary process, by virtue of its ( suc ce ss ful or unsucc e s sful ) interaction with its environment. This view, the organism selectionist view, was ch a 11 enged by

(am o ng Dawkins ( 1 982a, 1 982b). Dawkins argues that the gene, not the o rgan ­ ism, is the unit of se lecti o n . Selection can be de sc ri bed only in te rms of favoring or disfavoring gene freq u encie s in p opul ation s : 'A cc ordin g to Dawkins, in sexually reproducing o rga nis ms only short segments of the genetic material have what it takes to be selected . Organ i sm s are simply survival machines constructed by gene s to aid them in their single-minded quest for replication' . (Hull 1 988:2 l l ). However, the comp l ic a t i ons in biological s e l ecti on do not end there . Others h ave argued th a t selection may occur at other leve l s as well. It has been argued that selection might oc c u r at the species level, or even at higher taxonomic levels . For e x ampl e, it has been suggested that a s pe ci e s may p o sse s s a p op u l a ­ tion structure that favors its evolutionary s urvi v al (Hull 1 988 : 420- 1 , ci t i n g Vrba 1 984). It has also been sugges t ed that the geographical ra n ge of a higher taxon makes it more lik ely to survive a mass ext inct io n , no matter how many spec i es are contained in the taxon ( H u l l 1 988:220, c iti n g Jablonski 1 986, 1 987). o t h e r s)

Still worse comp lica t i ons ensue when we abandon our zoocentric view of

e vo luti o n and ask ourselves at what level of o rga niza t i o n does natural selection ope r a te for cloned groups of plants and single-celled org ani sms:

botanists distinguish be tween tillers and tussocks, ramets and genets. For

ex­

ample, many sorts of grass grow in tufts (t ussocks) composed of n umerous sprouts

(tillers) growing from the same root system. Which is the 'organism,' each tiller or the entire tussock? More gener a l l y , botanists term each all the ramets

that result from

a single zygote,

physiological unit a ramet,

a genet.

Sometimes all the ramets

that compose a single genet stay attached to e ach other; sometimes

1 988:4 1 7)

not. (Hull

The basic p roblem is that the divisions betwee n the leve l s o f o rgani zat i o n for org ani sm s is not at

all clear, once we go bey ond animals: 'The hierarchical

boundary between o r gani sms and groups of organisms is no sharper than that

between genes and

or ga n isms , in fact much l e s s so' (Hull

1 988:4 1 8). Hence,

one should not base a th eory of selection on a particular alleged l evel of bio­

l o gica l organ iz a ti on .

22

An evolutionary model of language change

But there is anothe r way to look at selection. Hull argues that there has been a convergence in the two approaches in the gene vs organism selectionist debate as they have refined their positions. Hull quotes an organism selectionist, Mayr, and then a gene selectionist, Dawkins: as Mayr ( 1 978 : 52) has emphasized tirelessly, 'Evolution through natural selection is (I repeat!) a two-step process' . . . According to the terminology that Dawkins ( 1 9 8 2 a 1 9 82b) now prefers, evolution is an interplay between replicator survival and vehicle selectio n . (Hull 1 988:2 1 7 ; see also pp. 4 1 2- 1 8) ,

The two steps involve two processes, replication of individuals and selection of individuals through interaction with their environment . However, in the para­ digm case, these two individuals are not the same : it is genes that are replicated and organisms that are selected which ensures the survival of their genes. Hull argues that the debate between gene selectionists and organism selectionists is largely a matter of emphasis as to which process is more import­ ant. But both processes are necessary, and it appears that prominent advocates on both sides of the debate accept this: ' Since the . . . dispute over the units of selection broke out, all sides have come to accept the distinction between replicators and in teractors, albeit not necessarily in the terms I am urging' (H u ll 1 98 8 :4 1 3) . And, in fact, this is the crucial conceptual advance in under­ standing the nature of evolution . Hull himself has contributed to this debate. He borrows the term REPLICATOR from Dawkins and chooses a different term, INTERACTOR , from Dawkins' term 'vehicle' (which Hull believes re n ders the role of the interactor more passive than it actually is, and is a consequence of Dawkins' gene selectionist bias; see Hull 1 98 8 :4 1 3 ) . From this, Hull constructs a general analysis of selection pr ocesses The basic components of Hull's theory of selection are quoted below (Hull 1 988 :408-9; emphasis original): -

.

(1) (2)

(3)

(4)

'

REPLICATOR an entity that passes on its structure largely intact in suc­ cessive repl ications' ' INTERACTOR an en tity that interacts as a cohesive whole with its environment in such a way that this interaction CAUSES replication to be differential ' ' SELECTION - a process in which the differential extinction and prolifera­ tion of interactors CAUSES the differential perpetuation of the relevant replicators' ' LINEAGE an entity that persists indefinitely through time either in the same or an altered state as a result of replication' -

-

-

There are a number of important consequences of Hull's theory that wi ll be described here . The first is that a replicator must not simply replicate its struc­ ture. The replicated structure must also be able to replicate its own structure: ' If al l a gene did was to serve as a template for producing copy after copy of itself without these copies in turn producing additional copies, it could not

The generalized theory of selection

23

function as a replica tor' (Hull I 988 :409) . That is, one must be able to have a replication of a replication of a replication . . . This leads to the creation of lineages of indefinite age . Replication also allows for an indefinite sequence of differences in replica­ tion that can eventually lead to very different structures from the original replicator. Replication must preserve structure largely intact, otherwise we would n ot call it r e p l i cati o n ; but it can involve alterations to that structure. Once the structure is altered in replication, that alteration can be further replicated . The result can be a re p l icato r that is quite di ffe ren t from the original replicator . 'The important principle is that in a chain of replicators errors [alterations WAC] are cumulative' (Dawkins 1 982b:85). The second feature is equally important. Hull emphasizes that causality is involved in the selection process. In fact, there are two different causal mecha­ nisms. The first mechanism, implicit in ( I ) and n o t discussed in any detail by Hull, causes replication, both identical (NORMAL) and ALTERED REPLICATION . T he paradigm example of the mechanism of normal repl i c ation in biology is the process of copying genes in reproduction, with mutation and recombina­ tion being mechanisms for altered replicati on.

The possibility of altered replication gives rise to variation, by creating new The second mechanism, referred to in (2) and (3), causes differential replication, that is, perpetuation of (different) replicators: this mechanism propagates some variants at the expense of others. 1 Differential replication is hypothesized to result from the interac­ tion of interactors as a group with their environment, specifically, the survival of some interactors (and hence their replicators) and the extinction of others. The paradigm example of differential replication is the increase in frequencies of certain genes due to the favored survival and reproduction of the individual organisms possessing those genes in their ecological environment. Hull makes a third important proposal in his theory: that his general analysis of selection processes applies not j ust to the gene-organism level in biological evolution, but may apply to other l eve l s as well : ' Just as genes are not the only replicators, organisms are not the only interactors. Just as variable chunks of the genetic material function as replicators, entities at different levels of the organizational hierarchy can function as interactors' (Hull 1 98 8 :4 1 7) . If the population structure of a species can be heritable, then species might be able to function as replicators. Genes may be interactors as well as replicators, since they interact with their cellular environment at the molecular level. Although Hull expresses some doubts a s to whether organisms and species can function as replicators, he suggests that it is possible, and cannot be ruled out absolutely (for organisms, see H ull 1 988:409, 4 1 5; for species, see ibid . 2 1 9 and 4 1 9). Hull's main point, though, is that a generalized model of selection must be cut loose from the hierarchy of levels of biological organization. On the other hand , change can occur without selection, and selection need not result in change: replicators with a different structure from the original .

24

An evolutionary model of language change Once the distinction between replication and interaction is made, one can dis­

tinguish four possibilities: changes in replication frequencies due to interaction (directional selection), no change in replicator frequencies because the effects of the relevant variations happen, by chance, to balance each other out ( balancing se lection) , chan ges in relative frequencies that are not due to any environmental interactions (drift), and replication sequences in which there are neither changes in rep licator frequencies nor significant environmental interactions ( s tasis ). (Hull 1988:443) This classification of selective and nonselective changes will be relevant

fo r

understanding the mechanisms of language change (see §3 .3 .2) . The

fourth,

and most important, aspect o f Hull's general theory i s that

selection operates only on spatiotemporally bounded individuals: 'only an indi­ vidual has what it takes to be selected' (Hull 1 988:21 5). Spatiotemporally bounded individuals are actual individuals, by definition, and so selection oper­ ates only over actual individuals: 'In selection processes of all sorts, selection takes place among actual, n ot possible, alternatives' (Hull 1 988:473). ' Indi­ vidual' is taken in the broad sense here, so that spatially discontinuous entities (such as populations) are individuals as well, as long as the collection of entities is spatiotempo r ally bounded . Thus, taking the popul ation view of species, a particular species is an individual : it has a beginning and an end temporally and it is also bounded spatially. Particular organisms and genes are also indi­ viduals; so is a collection of plants growing from a single

r oot

stock; so are

other population-based entities such as demes. Hull 's purpose in devising a generalized theor� of selection is not merely to sort out certain controversies in evolutionary b i e, e > e, e > ce > a and a > :> (M arti n et 1 952/ 1 972: 1 45; he includes the E > a shift in the chain) . The result is another symmetric system: ( 9)

e E

ii

0

a

u 0

:l

E

ii

3

Teleological mech an i sms such as Martinet's are problematic in theoretical and empirical respects. How do symmetric systems become asymmetrical in the first place? This objection can be countered by the hypothesis that other changes are nonteleological, and that other parts of the linguistic system can change, affecting the phoneme segment inventory. (One might still ask, why is the l i ng ui s ti c system overall not perfectly balanced so that asymmetry would never arise?) There is a much more serious empirical problem : there is a relatively high tolerance of homonymy, allomorphy and asymmetry in linguistic systems. There is a very high degree of ambiguity of words and especially grammatical mor­ phemes (see chapter 4). For example, there are many languages which do not make a present/future distinction (such as Japanese), hence the potential homonymy of the Greek present and future forms cannot be a strong motivating factor. Also, Lass points out that there is already homonymy between indica­ tive and subjunctive active l st person singular present forms; and intervocalic Is/ survived in the first aorist, although it was not 'needed' to distinguish the aorist from other forms (Lass 1 997 : 3 56) . Lass also points out an example of sound changes in the history of English tell/told in which more allomorphy is created than lost (Lass 1 980: 7 1 -3). (However, it should be pointed out that no systematic cross-linguistic survey of the frequency of homonymy and allomorphy in morphological paradigms has been made, to my knowledge.) Labov surveys a number of studies of variable deletion of phonological segments that also occur as markers of inflectional categories, such as English lt/-/d/ (past tense), Spanish and Portuguese Is/ (noun plural, 3rd singular subject agreement), Spanish /n/ (3rd plural subject agreement), Ladakhi /s/ (perfect) (Labov 1 994, chapter 1 9). According to the systemic functional analysis, the segments are Jess likely to be deleted when they indicate the inflectional cat­ egory in question. The studies that Labov surveys give no consistent evidence for a teleological mechanism. From a typological perspective, it can be observed that many languages lack tense marking, plural marking and subject agree­ ment; hence there is no strong 'need' for a l anguage to maintain these inflec­ tional categories. Moreover, many languages lose tense, number and other inflections. If homonymy avoidance was a significant factor, these inflections would not be lost.

Types of causal mechanisms in language change

69

With respect to symmetry, many phonological systems allow for gaps in their se gment i nventories , as in the following inventory fr om Bej a (Cushitic), taken from Maddieson ( 1 984: 3 1 6) : ( 1 0)

-

t

b

d

f m w

s n

k

t

� f

g

Cantonese pin 'diminish'

PIE b"end"

>

Sanskrit band"- 'bind'

Since hypocorrection and hypercorrection i nvo lve two oppo s i te results, Ohala constraints to differentiate the two ( Oh a l a 1 992, 1 993). Hypocorrect i on

p ro p o se s

of the co nd i tio n i n g environment, since the ph o n ­ of the conditioning environment are re a nalyzed as part of the altered segment. Hypercorrection never results in such a loss, since the prop­ erty fact o re d out of the altered segment is attributed to the other segment. Hypocorrection can cr e at e a n ew segment type but hypercorrection does not, si nce the reason a p h o n et ic property is factored out is because the listener identifies the segment with another segment type ex i s t ing in the lan guage . commonly results in the loss e t i c p r ope rt i es

78

Theories of language change

Hypercorrection is also typically restricted to factoring out of phonetic proper­ ties that require a longer period of time to be perceived. Ohala's model does not call for any intentional processes other than the attempt to conform to the conventions of the speech community. Ohala criti­ cizes explanations of sound change due to economy and other intentional and teleological factors, arguing that they are difficult to justify and moreover unnecessary, at least for the classes of sound changes that he has investigated (e .g. Ohala 1 989: 1 9 1-3, 1 993:260- 1 ) . It i s not clear whether Ohala's research program t o eliminate intentional mechanisms of innovation will succeed. Certainly normal replication - adher­ ence to convention - is an intentional mechanism that nonintentional mechan­

isms cannot do without. Nevertheless, it is worth investigating the possibility that nonintentional mechanisms can account for a number of types of lan­ guage changes, including grammatical changes. Vincent suggests that those features of language that are less in the speaker's awareness or control, such as sound change, will change mechanically, while those more in a speaker's aware­ ness, such as grammatical change, may be teleological ( Vincent 1 978:4 1 6 ). I have argued against teleological mechanisms for innovation; but it may be that Vincent's hypothesis applies to nonintentional vs intentional mechanisms. That is, nonintentional mechanisms for innovation are more likely to be found at lower levels of organization such as sound structure, while intentional mech­ anisms are more likely to be found at hi gh er l ev el s. While Vincent's hypothesis is quite plausible, one might nevertheless suggest that (barring direct evidence to the contrary) a good methodological strategy would be to seek nonintentional mechanisms first, and only turn to intentional mechanisms if those fail. In that spirit, I will propose that many grammatical changes may involve a nonintentional reanalysis of the mapping between gram­ matical form and conventional meaning or function. In chapter 4, I argue that even in normal replication of a word or construction, the mapping between form and function is extremely complex and variable. In chapter 5, I will suggest that reanalysis of the form-function mapping is always possible, and propose a set of form-function reanalyses for the innovation of a number of types of grammatical changes. Table 3 . 1 on page 79 summarizes the mechanisms surveyed above (those argued against above are given in square brackets) . 3.5

The concept of progress in biological evolution and language change

In §3.4, I have argued against teleological mechanisms of language change, and argued instead that nonteleological mechanisms, both intentional and nonintentional, are more appropriate mechanisms for language change, just as they are for biological evolution and conceptual change. In this section, I will address the question of whether there is progress in language change . The idea of progress in evolution is associated with teleology. A teleological mechanism specifies some goal; and progress can be defined as movement

The concept of progress in evolution and language

TABLE 3 . 1

79

Causal mechanisms of language change Intentional

Teleological

Nonintentional

Normal replication

entrenchment convention (being understood)

Selection

accommodation acts of identity pres t ige {including covert prestige)

c h an ge in entrenchment

A ltered [preserve distinctions] replication [preserve symmetry/ gap fi llin g]

expressi venes s n o t be ing misunde rstood ec o n o m y

over/undershoot hyperco rrect i on hyp oco rrec tio n form-function reanalysis [speech errors]

[preserver

isomorphism)

towards that goal. But teleology and progress are independent of each other. If the goal is constantly changing in different directions, then following the goal will not lead to progress, where progress is defined as continuous movement in a single direction. ( Progress is usually defined as movement in an allegedly positive direction; but the judgement of what is positive is so subjective that the more neutral description is better.) Conversely, a non teleological process such as selectio n can be pro gre ss ive if it m ov es steadily in one direction - for example, increased adap t ation to the environment. Hence selection as a process could, a priori, be GLOBALLY progressive, that is, represent continuous movement in one direction . Selection is always LOCALLY progressive, that is, there is a change in a direction defined by the mechanisms of selection . But ' as I understand the cu rrent state of evolutionary biology, a respectable number of evolutionary biologists think that biological evolution is no more than locally pro gressive (Hull 1 988 :464). Hull gives two reasons why biological evolution is only locally progressive. First, the environment is so variable, yesterday s adaptive improvement may be today s dysfunctional trait, and who knows what tomorrow will bring: because so many aspects of th e ir environments change so rapidly and haph azardly, species seem to be forever chasing their chan ging environments' (Hull 1 988:467). One way in which the environment is continually changi ng is the cycle of climate changes Another way is the fact that the environment includes other organ isms For example, there has been extensive coevolution of flowering plants with their insect or avian polli n ators and also trees with mycorrhizal fungi that pass on nutrients to them; as one evolves (or goes extinct) so does the other. The second reason that there is pnly local progress has to do with the nature of the selection process : ,

'

'

'

'

.

.

,

80

Theories of language change Natural selection is

a

p r oce ss capable of only local maximization . . . Compromises

are constantly made in biological evol ution . Chief among these is balancing the need s of personal survival with the needs of reproduction. If either side is shorted

too drastically , the effect is the same - extinctio n . (Hull 1 98 8 :473)

However, Hull suggests that global progress occu rred d u ring an earlier p h a se of the history of the Earth . Hull i s not referring to the evolution of a sin g le org anism as s uperior to other organisms; he vigorously (and entertainingly) criticizes this view (Hull 1 988:46 1 -2, 469-70) . Instead, he is referring t o the overall adaptation of organisms to the environment, a ssum i ng relatively fixed environmental con s traints : Initially in the history of life, when so much of the environment was empty,

selection was n o t very rigorous and everything increased but very slowly . As the

earth filled up, selective forces became more rig o ro u s and species packing in­

crea s ed until there was little ro om for improvement without a c ompen s ati ng loss occurring elsewhere .

.

. (H ull 1 988:46 1 )

Hull quotes Gould a s s tating that after the Cambrian explosion, 'inhabitants change c on ti n u a lly but the roles remain' (Gould 1 977 : 1 9, quoted in i bid ) and himself adds, 'About the only i n s tan ce in w hi ch evolution has had a direc tio n d u ri ng the past 500 milli on years is th e surprisingly late exploitation of air as an adaptive zone' (ibid.). Of course, this only a pp l i es to a very general charac­ terizat i on of e c ological niches ; the examples of coevolution show that at a finer scale, there is still con ti nuous c hange The notion of progress in l an g u a ge was prevalent in the l 9th c en t u ry but is univers ally condemned today by profe ssional ling u i s ts (It survives robustly outside the realm of l i ngu isti cs in the notion of standard languages as supe r ior to vernaculars, and the laments over the ' decline' of the s t an dard whenever it changes towa rds the vernacular.) Part of the reason for the condemnation was the subjective evaluation of p ri mitive cultures as in fe ri or to 'civilized' ones, or of modern l an g ua ge s as i n ferior to c lass i c al ones. However, the preceding discussion of the notion of progress in biol og i ca l evolution should allow us to reopen the q uestion and examine it without prej udice in either direction . First, recal l t hat progress is defined as direc ted change; no evaluative j udge­ ment is assumed here . It seems cl e ar that there is virtually no global progre ss that is, continued change in a single direction i n p honological and gram m ati cal (morphosyntactic) c hange The reas o n for this is p ara l lel to the post-Cambrian state of affairs of life on Earth. All contemporary lang u ages and all earlier lan guages for whi ch we have written records, are general-purpose commu n ica tion systems at a coarse scale of description . I t may b e the case that before rec orded history language-like communication systems had not filled out the space of communicative p o ssibi li tie s and glob al progress was po ssib le But once the space of comm u n ic at i ve possibilities had been filled up (at this coarse view), no further progres s could be made . Instead, what we find is c yc lic changes of linguistic subsyste ms all of which a re wel l enough adapted to effective comm u ni c at i on that no one structural ,

.

,

.

,

.

,

'

'

,

,

.

,

­

,

,

.

,

The concept of progress in evolution and language

81

type has a decided functional advantage over the others. These ch an ges are e ssen ti al l y locally progressi ve, that is, they involve directionality for a short period of time, but then turn back on themselves (completion of the cycle). For example, there is a typological classification of the morphological types of lan guage s in terms of the number of concepts encoded in individual morphemes and words. In the 1 9th century, this classification was pos tul a ted to be globally pro gre ssi ve . L an gua ge s moved from an isolatin g type (one morpheme/concept per word) to agglutinative (multiple morp heme s per word, each m o rpheme expres s i n g a c oncept) to fusional (multiple co n cepts p er m orphem e) . Evidence of the first shift, for example, is the development of compounds in an isolating language such as Mandarin Chinese. The ancient Indo- Eu ropean l anguages were fusional: a form such as Latin cantat 'he or she sings ' comb i nes 3rd sin gular agreement, present tense, indicative mood in the suffix. But further ex amination of the facts demonstrate d th at fus i onal lan guages tended towards reduction of inflections, l eadi ng to a more isola t in g type, as has happen ed i n En gl i sh and French. Also, the process of morphological ch an ge turns out t o be local in another sense, in that so me parts of the grammar of a single language are at different stages in the cycle . For example, the contraction of English auxiliaries and nega t ion is approaching fusion in some cases (won 't, nonstand­ ard ain 't), while many word derivational affixes are more a ggl u tin ati n g (cf. linguist-ic-al-ly). The reasons for these cyclic changes also appe a r to be parallel to the reason s for only local progression in biological evolution. The environment at a fine­ gr ain e d scale i s c o n s tant ly changing. Th at is, p arts of the lan guage system are changing in ways t hat affect o ther parts of the language system, leading to further c h an ge s . The subsequent change s need not be 'compensatory' for the initial change - that would be a teleological explanation - but merely conse ­ quences of the original change in s ome way or another. Research in the interconnections among grammatical systems (that is, cross­ l in gu istic implicational universals in typology) have revealed some inte rconnec­ tions among gram mat ical systems that are man ifested in chains of historical changes. For example, there are well-known implicati onal un i ve rsals l i n ki n g the word order of subject-verb-object, adpositio n -n o u n , adjective-noun and genitive - noun . In the Semitic languages of E th i opia, an original verb-subject­ object word order changed to s ubj ect -obj ect- ve rb order, p resumably due to contact with local languages. Once that change occurred, it appears that there was a gradual shift to adjective-noun order from n o un - adj ective order, then to genitive -noun order from noun-genitive order, then finally to po stpositions from pre positions (Greenberg 1 980; he also reports evidence of a similar pattern in the history of the Iranian languages). These subseq uen t shifts in word order were presumab ly tri gge red by whatever underlying factors account for the impl ica­ tional universals of word order that link these patterns (Greenberg 1 966/ 1 990) . The second explanation for o nly local pr ogress in biology is al s o invoked in d iachronic typological research. Typologists have advocated co mpe ti ng motiva­ tion models of lan gu age ch an ge and the cross-linguistic variation that is the

82

Theories of language change

result of language change (see especially Haiman 1 985). The competing motiva­ essentially local maximizations of communicative value. The commonest example of competing motivation given in typology is that between economy and iconicity. Economic motivation, as discussed above (see 24), is the prefer­ ence of shorter forms, or even zero expression, particularly for expressing con­ cepts that are frequent in use. Iconic motivation is the overt expression of concepts, presumably motivated by the go al of not being mi sund e rsto od (see 23 and §5.6). Competition between economy and iconicity can be illustrated in the devel­ opment and loss of number inflections. Synchronically, one finds l'anguages with economic encoding of the singular number, as zero, and overt (iconic) encoding of the plural (e. g. English); but one also finds languages with overt encoding of singular as well as plural (e.g. Latvian), and languages with eco­ nomic encoding of both singular and plural, i.e. no inflection for number at all (e.g. Mandarin Chinese) . Diachronically, one finds languages both acquir­ ing and losing number inflections, both plural and singular. For example, a pronominal plural marker in Mandarin can be used with nouns referring to humans; and could extend to other nouns; but the plural marking of French nouns and adjectives is disappearing as a result of the loss of final consonants. There is certainly no global directed change towards the acquisition (or loss) of number inflections. The use of language for communication provides an explanation of not only why language change is not globally progressive, but also why it is locally progressive. Language change is not globally progressive because all human languages have already progressed to being general-purpose communication systems, at the coarse-grained level. At a more fine-grained level, the environ­ ment for grammatical structures is constantly changing. This is because all changes are local, yet the language system is interconnected (see §2 .4.3) and the local changes cause other grammatical structures to change. Moreover, even local chan ges are the result of compet ing motivations, which are only local adaptations to the problem of communication. However, there are at least two possible examples of directional evolution at a more global level. If language is adapted to the environment of general­ purpose communication, then if that environment changes significantly and permanen tly , then one might expect some evidence of gl obal progress. One respect in which the environment has changed is the explosion of new objects, properties and processes with the technological advances of the past century. Any general-purpose communication system must involve the ability to name these new entities. The global progress (directional evolution) that has taken place is the massive increase of vocabulary i n the languages used in the speech communities that possess the new technologies. The second example of a possible global directional change is the advent of writing. Writing is a new medium of communication. Exploitation of the dif­ ferent perceptual modalities, time scale, audience interaction, etc. of literary comprehension (re ad in g) and produc tion (wr i ting) may lead to significant tions are

The concept of progress in evolution and language

83

l anguage changes . These differences have been described in terms of oral vs literate style . Two factors that are associated with the oral vs literate style dimensi on are degree of involvement or interaction, and degree of integration (Chafe 1 982). Tannen argues that i n v olvement can vary independent ly of medium (Tannen 1 982:3). However, Tannen argues that integration is more closely identified with the difference in media: cohesion is established in spoken discourse

through paralinguistic and non-verbal

channels (tone of voice, intonation, prosody, facial expression, and gesture), while cohesion is established in writing through lexicalization and complex syntactic structures which make connectives explicit, and which show relationships between

propositions thro ugh subordination and other foregroundin g or backgrounding devices. (Tannen 1 98 2 : 3)

Specific grammatical devices of integration used in writing include the fol­ lowing (Chafe 1 982: 39-45; Tannen 1 982: 8): (26)

a. b. c. d. e. f. g.

nominalizations increased use of participles attributive adjectives conj oined phrases and series of phrases sequences of prepositional phrases comple m ent clauses relative clauses

One might speculate that the advent of the writte n medium led to directed evoluti on in the development of these construction types. Typological researc h indicates, however, that all of these construct ion types are present in most if not all unw ri tten languages. Moreover, Biber ( 1 986) presents a multivariate analysis of 41 l in gu istic features across a wide variety of spoken and written genres, and argues that there are three factors, identified as interactive vs edited text, abst rac t vs situated content, and reported vs immediate style , each of which only partly correlates with the spoken vs written distinction . One possible difference be twee n s poken and written language is the exte nt to which multip le i te rations or embeddings of these constructions are found (Biber did not study this factor). In a st udy of English oral narratives, all of the constructions in (26a-g) were found, but ge nerall y only at one level of embedding (Croft l 995c; comparable evi dence for Wardaman, an Australian aboriginal language, is pres ented in Croft to appear a). M u lt i ple modifiers and nesting of subordinate clauses or phrases are extremely rare in oral conversa­ tion or narrative, and conjoined units, while found, are almost always broken into separate intonation units (an indication of a higher order of com plexity in spo ken l anguage). It is thus possible that there is directed change in the advent of the written register. However, expansion into the new linguistic niche results in the evolu­ tion at most of new degrees of syntactic comple x ity - multiple iterations and ·

Theories of language change

84

embeddings of structures - rather than in developing completely new gram­ matical structures . In this section, I began with a more sop h ist icated analysis of the nature of progress (and lack thereof) in biological evolution than is popularly held. I have then applied this analysis of progress to language change. I have shown that the analysis of progress in biological evolution can shed important light on the nature of language change. These advances in understanding are pos­ sible only by looking at language as it is used in communication.

Notes

For example, a we a ker h ypo t hes is is t h at some though not all processes of lan­ change ar e e xamples of drift, n ot selection, and conversely, some though not all c a se s of absence of la n guage ch a n ge are instances of stasis, no t balanced selection. This hy p othesis is not incompatible with the theory of l angu age ch an ge a dv oca ted he re ; t h i s m i gh t be the case for ce rta i n types of l angu a ge change . 2 The structuralist/generative theory does not claim that this is the only me a n s by whic h innovation can occu r The t he o r y just described a pplie s to internal means for innovation. It is also assumed that there are external means for i nn ova t i o ns produced via l a n gu a ge contact (see §6 . 1 , ch a pter 8). 3 Clark & Roberts 1 993 formulate an al go rithm le adi ng to la ngu age ch a n ge in the l angua ge acquisition process that is s omewh a t different from that used by Li gh t fo o t However, their theory is on ly a th e o ry of innovation and does not address the propagation p roblems that L igh tfo o t addresses. Clark & Roberts examine a set of syntactic ch anges that occurred in M i ddl e French (ea. 1 300-ca. 1 500), whi ch led to a change in parameter se tt i ngs from Old French to Modem French. They empl oy a genetic algorithm model wh ich allows for selection of parameter (re)settings by t he child ba sed on the inpu t Clark & Ro be rt s argue that the various s yn t act ic changes all follow from a single set of parameter re setti n gs F oll owin g t h e gene rat i ve i n n ov ati on m o de l t hey a ss u me t hat t he paramete r reset ti ng process occurs in child language acquisition, which implies that the chan ge occurred in a very sh o rt peri od of time, n o t the two centuries it actu al ly took . And the y , like Lightfoot, still require some exp l an ati on for the variation in the input. They attribu te that vari· ati on to a process external to the syntax, namely the rise of cliticized nom inat ive pronouns (Clark & Roberts 1 993: 3 3 1 , 338) . But th at process has syntactic conse­ quences in their syntactic theory as well (reg ard ing as signment of nom inat ive case, and also whether clit ic subject pro nou n s a re counted in dete rm ining V2 order). Why are those consequences not ca rried out immediately and unive rs ally in the 'next gen er ation ? Lightfo o t s theory can be in te grated with the Clark & Roberts' model of innovation of course, t hough it will st i ll have the problems raised in the last two p aragraphs 4 Thi s is not necessarily the fault of their creators. Grammaticalization theory is not intend ed to be a c om p re hen s ive th e ory of language chan ge The invisible hand theory is l ar gel y programmatic at this point. Even sociohistorical linguistic theory is la rgely a theory of selecti o n al tho ugh a number of its practitioners have grappled with innovation guage

.

.

.

.

­

,

'

'

.

.

,

.

Notes 5

85

A stage in a language in which there is v ar i ati o n between A and B need not lead to a shift to B; it is p ossi b le , and rather common, for a variant B to enter the l an gu age

and then to be lost, leading to the continuat ion of A. Such

a

p rocess can only be

discerned when t h ere are historical records of the A-B stage.

6

Un fortu n atel y , such s t u die s are in t h e i r infancy (de B ot , Gommans & R o s si n g

7

Howe ver , Harris & C amp bell go on to say , 'it is true that children learnin g a

1 99 1 : 87).

language have a special ro le to pl ay in furthering l in guis tic shifts' (ibid.; cf. Jespersen

1 922: 1 77-8). H a rris & Campbell do

a g ain st this position (§3 . 2) .

8

not offer evidence for this claim; I have argued

Keller ar gues that speakers choose

engelhaft

englisch fo r ' angelic' in (Keller 1 990/ 1994:80-3, 93-5),

inste a d o f

order no t to be misunderstood as meanin g ' E n gli sh '

an d speakers choose Dame i n stead of Frau in order to be gallant to women (ibid . 76-7).

Thi s latter account is s ocia lly implausible, and cannot be

extended to other

cases of pejoration, e.g. rac ial terms. Instead, one could propose

a

different invis­

ible hand explanation (Croft 1 99 7 : 396-7}: since most utterances �ith Frau are n eg a t i ve , a different term is

9

This is not always t h e c a se : c on si de r t he many sla n g words rep l ac ing

for 'sex' and 'drunk' . Yet 10

chosen in order not to be misunderstood as maligning

the gro up in qu es t i on by ass ociat io n wi t h previo us neg ative utterances.

taboo words

even these m ultipl e terms can be differentiated socially

and regionally, i.e. are v a ria nts of a sociolinguistic variable.

Keller i n fact does not clearly distinguish between innovation and p ropagation of a

chan ge .

He d i sc usses the distinction briefly but suggests that it is 'not clear' how

it applies to language

ch an ge

(Keller 1 990/ 1 994 : 1 46). In particular, so me of h is

maxims of langu age use are in fact instances of the sociolinguistic mechanisms for

pro pagation , i.e. represent selection as a cause of differential replication (see §3 .4.2).

However, Keller does not investigate t h is aspect of his model, and instead argues

for the central role of the invisi ble h an d in bringing about the propagation of a

11

lan guage change .

I t h a s been suggested to me, by Martin Haspelm.ath and Peter Trudgill in d epe nd ­

e n t ly , that symmetry may p l a y

a

role in cha nges in vowel systems only (cf. Lindblom

1 986).

12

Actually, th e retarding effect of the status o f Standard

has prevented the adoption of

a

E nglish on language change

new 2nd person plural form . Of course, the socio­

lingui s t ic model predicts th a t the establishment and the status of t h e standard

would h ave this effect (see §7 .4.2).

13

However, since Forchheimer and Ingram we re interested in d ifferen t kinds of pro­

their sample overstates the proportion of d ifferen t kinds of sy s ­ tems, particularly the 'asymmetric' ones . An areally and genetically more balanced

nominal systems,

survey performed by M at t hew Gordon (p erso n a l communication)

suggests that a

more accurate percentage of pronominal systems wi th gaps in number is around

14

1 0%

-

still a significant minority.

Jakobson in fact had six functions of l angua ge,

based on a mode l of language with

six c o mp onen ts : t he addres se r , the add res s ee , t he co ntext (i ncluding content), the mes s a ge (the verbal form) , the c ode (the linguistic system) and co n tact ('a p hysical channel and psychological connection between ad d resser and addressee') (Jakobson

1 960/ 1 9 7 1 : 2 1 ) . For Jakobson, the co ntext gi ves rise to the referential function. The

addresser gives rise to the emotive function, i.e. her attitude towards t he context.

As J akobso n himself writes, 'we canno t restrict the not i on of info rm ation to the

86

Theories of language change cognitive aspect of language' (ibid. 22), and so the emotive function can be sub­ sumed under the referential functi on The addressee gives rise to the conative function, which corresponds to the class of speech acts directing the addressee to do something. This function is subsumed under the more general theory of speech acts. Contact gives rise to the phatic function. Jakobson appears to be referring chiefly to speaker-hearer interaction in utterance acts, though he also cites Malinowski' s more general concept. The code gives rise to the metalingual func­ tion, which in his examples involves speaker-hearer interaction in discourse. Finally, the message gives rise to the poetic function (which is in fact the main focus of Jakobson's paper). However, Labov suggests that Martinet's push-chain analysis is the only plausible one known to him for the movement of back vowels to the front, since this process appears to coincide with the creation of a four-way height distinction among back .

15

vowels. Also, Labov proposes further principles (Labov 1 994:280-5).

Chapter 4

A theory of lang uage and m ean i n g i n use

4.1

Introduction

A usage-based theory of language change requires, of course, a model of lan­ guage use. We must answer the question , Why do people talk in the way they do?, in order to discover why the way they talk som eti me s changes - and why the way they talk often does not appear to change. This chapter offers such a theory, based on research in pragmatics, cognitive semantics and a variety of approaches in philosophy. There are two s i gn ifican t aspects of the theory of language use presented in this chapter. First, language use is essentially a joint act between speaker and addressee. Language is a fundamentally social interactional phenomenon . So is language change. It is this hypothesis that allows us to use Hull's generalized theory of selection to develop the Theory of Utterance Selection. Selection of novel variants is a result of interaction of the interactor - the speaker - with her environment - her interlocutors and the knowledge and goals shared between the speaker and her interlocutors. Also, both intentional and nonintentional mechanisms of altered replication make reference to speaker goals in conversa­ tion (§§3 .4.2-3 . 4.4) . Second, meaning in language use is not static or fixed. Meaning is protean and cannot be reduced to simple logical formulae. In this an d the fol l ow in g ch ap te r s I will argue that the slipperiness of meani ng is one of the chief mechanisms for innovation in language change. ,

4.2

4.2. 1

Language, com munication and convention

The function of language and communication

The function of language is often said to be to communicate information . Keller, however, argues that this view of the function of language is not correct. To see Keller's point, we must first define communication . Communication is the transfer of information; but it is not just any transfer of information. A speaker usu a lly transfers more information than she intended, and, d epen d ing on bow successful she is, also less. Grice proposed a now widely used definition of meanin g that captures this difference: spe ak e r A meant something by x" is (roughly) equi valent to A intended the utterance of x to produce some effect in an audience by means of the recognition of this intention" ' (G rice 1 948/ 1 989:220). ' "

"

87

88

A theory of language and meaning in use

If we say that the function of language is to communicate, then we have a circular e x planati o n : language is a mean s of communication, so of course its 'function' is to communicate (Keller 1 990/ 1 994: 87). Keller argues that the func­ tion of an e ntity should be its role in a larger system The function of language should be found in the larger system that language fits into - namely, inter­ pe rsonal interaction in human communities . Hen ce ; the function of language is some ex tral i n gui stic goal of the speaker. K e l l er formulates the function as follows: 'the human be ing has the goal to be socially successfui, and influen­ cing others by means of langu age is an essential element in the e xplanation of social success' (ibid. 86) . Keller emp has izes at various points that social success . comes in many va r ieties and is se n sitive to the particular circumstances of the social context of utterance; hence this is not to be treated as a reductionist view of human behavior. Keller's view is essentiall y not different from that of many analysts of dis­ course who emphasize the interactional nature of conversation . These analysts believe that undue attention has been focused on communication, and not enough on functions such as establ i shing or reinforcing group identity and solidarity. Nevertheless, it can be objected that one cannot use language to reinfo rce solidarity, or carry out other extralinguistic goals, without engaging in the communication of information, even if that information is not particularly informative The apparent conflict in approaches to the function of language can be resolved by recognizing that communication with language is a means to carry out the interactional goals emphasized by Keller and others. Which one is the ' important' function depends on whether one considers the important function to be the ultimate goal of the interlocutors - social-interactional or the omnipresent means to any other goal achieved by language - communica­ tion of information. More generally language in use involves multiple levels of action on the part of the interlocutors, the lower leve ls being the means to achieve the higher levels, and the higher levels achievable only by carrying out the lower levels. .

.

'

'

-

,

4.2.2

The joint character of speech

All utterances are not simply complex symbolic expressions; they are actions on the part of the speaker and also the hearer. This is t rue of all levels of s peech, from pr oduc i ng the sounds to the actions wh ich the speaker wants to accomplish with the aid of the hearer The highest-level goal of language is social. Keller describes it as in fluencing others in order to achieve social success. A less self-interested way to put it than infl uencing others' is to describe language use as engagin g in a joint activity such that the hearer and speaker share some goal. The crucial aspect of this definition is that the function of language is not carried unless both speaker and hearer act jointly. Of course the actions carried out individually by speaker and hearer a re not the same. But this is true of many joint actions: for example, the actions car ri ed out by a fl autist a n d harpsicho rdist playing a CPE Bach -

.

'

,

Language. communication and convention

89

the same. Clark defines a j oi nt action a s foll o ws : 'Ensemble A­ doing joint action k if and o nl y if: 0. the action k includes 1 and 2; 1 . A in te nds to be doing A's part of k and believes that O; 2. B intends to be doing B's pa rt of k an d believes that O' (Clark 1 996:6 1 ). Th at is, each individual is ca rrying out their part of what they recogn ize and intend to be a j o int action. Speaking, or rathe r speaking-and-hearing, i s also a j oi n t action. Most speech act an alysts h ave i gn ored the action performed by the hearer because it appears to be relatively passive. But communication i s not communication unless the h ea rer understands w h at is said, and the extralinguistic goals are not achieved if the hearer does not partake in them or e ve n if the hearer simp ly does not consider partaking in them. That i s l a ngu age is n ot serving its purpose unless the h eare r does his part. A speech act canno t be successful, in fact c ann o t get off the ground, without the u tte r ance the s tri n g of soun ds bei n g properly executed by the speak er. Equally important, the speech act will not get off the ground if it is not properly at te n de d to by th e hearer: if t he hearer i gno re s the speaker, or background noise drowns out the utterance, the speech act has also failed . At the next level up, the formulation of the u t t era n ce in some language, the hearer also plays a rol e The hearer must be able to identify the propositional act p resen ted by the speaker. The speech act would not be completely successful if, for example, the he a re r did n ot know the language or d i a lec t being used by the speaker, or was n ot familiar with some of the jargon in t he utterance. The result of s ucce ss ful production and comprehension of the utterance at these levels is that the intentions of the speaker are mutually kn ow n that is, part of the speak e r and hearer's common gr o und (see §4.2. 3). This is true of even simp le assertions. Consider the following imaginary e xam ple of an asser­ tion followed by two p o ss i b l e responses. sonata are n o t and-B i s

,

,

,

-

-

.

,

(l)

S tuden t A : Student B: Student C:

The t ran scripts need t o b e turned in before the ne x t t ut o rial

.

Yeah . No

they don't!

The response by Student B re cogni ze s the i n ten ti o n of Student A th at B accept A's assertion, and implicitly accepts it. B would not be a b le to accept A's assertion witho ut recognizing A 's intention , for instance if B misunderstood what A s ai d . On t he other hand, the response by S t ude nt C rejects A's propo sal that her as sert ion be made part of their shared knowledge; but C's rej ecti on of A's asse rtion co ul d not be made unless C correctl y understood A's intentions i n communication. Finally, the actions which speakers attempt to accomplish

by their ve r bal Their jo in t character is fairly clear since most of them req u i re e xp licit responses on the part of the addressee: for exampl e a q uesti on calls for a n overt respo nse But there are a variety of responses or upt akes that an addressee can m ake , ran ging from full compliance to altera­ tion of the proposed project to declination or even withdrawal from the project (Clark 1 996:203-5). In any case, one cannot say the joint project is com pl et e d statements are themselves joint.

,

.

90

A theory of language and meaning in use some

until

versio n of

it

is complied wi th, or wi th d ra wal

is accepted by th e

proposer. A j o int project may take a number of turns to negotiate and finally

achieve, as in the following example, taken from a conversation recorded and transcribed by a student at the Unive rsity of M anchester: (2)

N: T: N: T:

N:

[pre-request] Have you got anyth i ng I c an borrow? erm .. well I don ' t like you wearing any of my trousers (inaud . ) [declination of anticipated requestj 'ave you got any ba ggy ones? [alteration ofpre-requestj erm . . why don't you wear them black and white check leggings and then a smart blouse and my black blazer [alteration of project: suggestion j Yeah that's alright [compliance with altered joint project}

The j oin t character of language use i n vo l ves not only language as acti o n b u t als o lan guage as symbol or sign (meaning), as will be seen in the rest of this chapter. 4.2.3

Speech communities and common ground

Language fundamentally involves people - p lural , collective. We cannot talk about language use - speech acts in conversation and the nonlin guistic acti ons they serve - without talking about the peopl e that use them. In particular, we must examine m o re closely the fact that the l anguage that people use defines a speech community . The proper model of a speech community is, or should be, a central issue in linguistic theory . A language, in the sense of a population of utterances as defined in §2 .4. 1 , is the central defining concept; the delimitation of this po p u­ lation is by a speech community . Linguistic conventions, whose rise and fall must be accounted for by a theory of language chan ge, are defined with respect to a s peech co mm u nity . Finally, the distinction between so-called internal and external causes of langu age change uses the notion of a speec h c ommuni t y to provide the b o u n d a ry between in temal and external. Yet the analysis of sp ee ch communities found in s ociol i ngu i stics (which has examined the problem in greatest detail) and by Clark makes it clear that the notion of a speech com­ munity is much more c omp lex than the naive view, and this has important consequences for theories of language and language change. In particu la r , the analysis of speech communities underlies the analysis of propagation in chap­ ter 7 and of phyloge netic rela ti o ns am ong languages in ch apte r 8 . The naive view o f a speech community - immortalized b y a n oft-quoted passage from Chomsky - is that it is a collection of individuals who speak a single lan guage among themselves, and are all native speakers of that language: 'Lin g ui stic theory is concerned prim arily with an ideal speaker- listener, in a completely homogeneous speech-community, who knows its l anguage perfectly' (Chomsky 1 965:3). If we apply the naive view to the u sage-based model of language change being proposed here, then the p opu l a t i on of utterances are ·

Language, communication and con vention

91

the utterances p ro d uced by the members o f t hat s peech c omm un ity the con­ ventions of lan gu age are those that are found among those m embers and the internal-external boundary is d efi n ed by the individuals of one speech com­ munity vs those of an o the r The first obvious p r o b lem with the naive view is presented by multilingual societies, i.e. societies where two or more distinct (often, mutually unintelligible) varieties are used by (at least so m e of) the same people. Generally when people use two (or more) distinct varieties in a society, sociolinguists use the term cooE. In multilingual societies, one can no longe r use individuals to define the internal-external b o u n da r y Instead, soci o li n guis t s define a code as belongi n g to one or more social DOMAINS ( Fi sh m an 1 972:50). Typi cal i m por tan t social domains referred to by sociolinguists are family, friends, work, school, religio n (Fishman 1 965/1 972) . However, there are many subdomains: one knows differ­ ent friends depending on the different sorts of recreational activities one does; different sort s of work may involve differen t codes. We wil l m ake the notion of 'domain ' more precise below; but it is clear that an empirically adequate de scr i p t i on of a si n gl e language based on the no t io n of a speech com mu n i t y will have to de fine a speech c o m m u n i t y relative to domains an d not si mp ly individuals. Of course, the code for each domain is u sed by the people who operate in that d omain (often a subset of the society as a whole). The first important consequence of this change in the definition of a speec h community is th at individual members of society are mul tilin gu al : they s peak d ifferent codes by virtue of the different social domains they operate in. It may be that some members of multilin gua l s peech communities are themselves mono­ lingual; but this indicates that they do not participate in the social domain(s) which require use of the language(s) they do not speak . If ( fo llow in g sociolin­ guistic practice) we continue to use t he term COMMUNITY to describe the spe a kers of a single code the n most members of multilingual s ocieti es are also members of multiple speech communities. The second important consequence of this ch ange in definition is that a code has a social meaning, determined by the context in which it is used, as well as a referential m eanin g i.e. meaning in the usual sense of that word . In other words, conventional lin gui s ti c units such as words or c on s tructi o n s have not only a form (s i gnifi er) an d me an i n g (si gn i fied) but a third di mension , the s ocial do m ain/comm un ity in which they are used. Another significant fact abo u t codes in multilingual societies is that no code has the complete ran ge of communicative power; there are thi n gs one cannot talk about in o ne or the other l ang uage s For example, in the classic sociolinguistic study of the Bua n g in Papua New Guinea (Sankoff 1 972) , three languages were foun d in the c ommu ni ty : Buang (the local language), Yabem (a language spo ken by a larger group, into which the Bible was translated for all the tribes in the area), a nd Tok Pisin (a p i d gin used for a variety of pur­ p oses ) Buan g was used for di scu ssing traditional culture, and Yabem in the realm of Christian religi on A Buang could not really discuss Christian re l i gi on in Buan g or traditional culture in Yabern. Another example is cl assic DIGLOSSIA , ,

,

.

.

'

,

,

.

.

.

92

A theory of language and meaning in use

where two mutually u nintelligible varieties are used, the spoken L(ow] variety and the written/formal H[igh] variety (Ferguson 1 959/ 1 972). In a diglossic situation, neither H nor L can properly serve in the domains for which the other is u sed. Of cou rse, the codes could be mixed, and words and construc­ tions could ultim-ately be borrowed, allowing the modified code to be used in other domains. But that would imply that the domains in which the langu­ age (say, Buang) is used is changing (mixing) or has changed (borrowing), and so the speech community (or communities) have changed. In a multilin gual comm unity, each code or variety is specialized for the domains in which it is u sed . Multilingual communities are very common in the world; yet the norm for a speech community is u sually taken in linguistic theory to be a monolingual speech community, and so a community is defined in terms of individuals rather than domains, and lin gu istic units are treated as two-sided signs, not three-cornered signs-in-a-community. Yet a closer examination of a monolin­ gual speech community reveals that it is in fact much more like a multilingual community. T he profound fact is: every language is a multiplicity of codes. Not everyone knows the language of sociolinguistics - or the language of classical music, or of wine-tasting, or of carpentry. Of course, there is a lot of overlap between the language of sociolinguistics and the language of wine­ tasting - we call them both ' English' (or 'French' , etc.). But what we call ' English' is really a multiplicity of codes, specialized to varying degrees, and known to 'English speaker s' only to the extent that they are members of the relevant community: There is no limit to the ways in which human beings league themselves together for self-identification, security, gain, amusement, worship, or any of the other purposes that are held in common; consequently there is no limit to the number and variety of speech communities that are to be found in society. (Bolinger 1975:333)

To sum up: a linguistic code belongs to a (speech) commu nity, and a com­ munity is defined by a domain. Every person is a member of multiple (speech) communities, hence every person speaks multiple codes, depending on wh ich communities he or she belongs to. This is called his or her REPERTOIRE (Gumperz 1 968/ 1 972:230; F ishman 1 972:47). Everyone has a slightly different repertoire. We still need a term to describe a group of socially interacting speakers as a whole, including all the communities they belong to. I will use t he term SOCIETY to refer to a socially inter acting group of speakers; hence, 'society' refers to the group commonly described as a 'speech community', even in much sociolin­ guistic research ( see §7 .2 for further discussion). Another way of stating the fundamental point is th at every society consists o f multiple communities. An effect of t he massive degree of overlap in codes used in the communities of a single society is to break down the internal-external boundary in the theory of language change. T his is one of the main reasons why l anguages are fu nda­ mentally variable. These facts will be discussed in chapter 6. Here, we will

Language. communication and convention

93

further explore the nature of communities and how that affects how individuals interact in using language. A community is defined in terms of a domain of SHARED EXPERTISE: 'a cul­ tural community is really a set of people with a shared expertise that others lack' (Clark 1996:102). S hared expertise is involved no matter how technical or how ordinary the expertise is. If you are a professional linguist, you share expertise with other profess io nal linguists about the kno w led ge and discoveries and methods of the field. But you can have expertise in so met hin g le ss techni ­ cal . For example, I have some shared expertise with my colleagues about how the University of Manchester works (at least from a professor's point of view). Another basic expertise which is shared among many people is living in a par­ ticular location, such as Manchester, or the San Francisco Bay Area . A n o ther important shared expe rtise is our language: we know its grammar and vocabul­ ary and how it is used to communicate. Yet another shared expertise of ordinary life is cooking: this allows you to recognize and name kitchen implements, various kinds of fruit and vegetables, spices, and how to use them all to create meals. How do we divi de shared experti se into dom ai ns , other than by the indi­ viduals who possess it? The shared expertise is o bviousl y interconnected in v arious ways. Ultimately, h ow ever , it is defined by a social institution in the society (F is hman 1965/1972: 19), for example the i n stitution of the University of Manchester, a po litical entity such as San F rancisco, or a less explicitly defined but nevertheless discrete cultural activity such as cooking in the home. Of course, there is also a l ot of shared expertise that is common across domains: for example, shared expertise about sp at i al relations between objects will be shared across all domains which make reference to objects in the phy sical world. I will call this common shared expertise CORE EXPERTISE (see §7.2).

Shared e xperti se is the ba sis for mutual knowl ed ge , which C l ark calls COM­ I share as, say, a classical m us ic ian or a cook, is of various types. The common g round is not just shared kn owledge about general concepts, or specific practices (such as how to saute vegetables), but also particular individuals that are important or salient to the community. Common ground also involves shared attitudes to a significant degree, certainly in the case of communities based on rel igious or political beliefs (Catholi cs, Republicans) but al so in m any other communities: shared attitudes strongly reinforce co mmunity solidarity. Common ground is the fundame n tal basis of our social lives. We cannot interact without assuming some common grou n d with other people. Even with someone from the most distant and exotic culture, we can assume some common ground with our being human - ph y sical properties, emotions, etc. and living on the same pl anet - topography, climate, etc. (although we might erroneously assume too much commo n ground). It is common ground that e na ble s us to communica te at all, as will be seen in the remainder of this ch apter . But there has to be some shared basis for common ground . Community MON GROUND. The common ground

94

A theory of language and meaning in use

membership - i.e. shared expertise - is one of the most important shared bases for common ground. Communities are defined by f unctional domains - the functional domain is the shared expertise. The mutual knowledge that comes from communi ty membership is called COMMUNAL COMMON GROUND. A speaker 's knowledge must be specified by the community with which she shares that knowledge . The most obvious linguistic manifestation of communal common ground is its COMMUNAL LEXICON. Communal lexicons are specialized vocabulary for a particular domain of shared expertise, such as the words crotchet, sonata and quartet in classical music (Clar k 1 998; see §7.2 for further discuss ion). Again, the three-cornered representation of a linguistic unit, describing it in terms of community, form an d meani ng, is necess ary to describe communal lex icons. . There is al so common ground shared more directly between individuals , based on their shared experience. Clar k calls t h is PERSONAL COMMON GROUND (Clark 1 996: 1 1 2- 16). It h as two shared bases: what we experience together and what we tell each other. What we experience together is called the PERCEPTUAL BASIS by Cl ark ( 1 996: 1 1 2- 1 3). The perceptual basis is simply that we are l ooking at (hearing, smelling, experiencing in general) the same thing, and we mutually know we are looking at that thing. That mutual knowledge is achieved via JOINT ATTENTION (see Tomasello 1 995), and also via the common ground that yo u and I are human beings with normally func tionin g percep tual o rgan s and brains - a person cannot establish joint attention with a mannequin, for ex­ ample. What we tell each other· is called by Clar k the 'actional basis' (Clark 1 996: 1 1 4) - a rather pecul iar term, which will be replaced here by DISCOURSE BASIS. Discour se basis involves joint attention - to what each of us is telling the other - and community membership (sp eakers of the same language), as well as the common gr ound of being human and rational . Just as commun a l common ground is specific to p a r tic ular cultural commun­ ities, personal common gr ound is specific to the persons with whom we shared the experience/conversation : e.g. a brother, a friend, a professional colleague, a fellow student, a par tner . Not only are there communal lexicons, but there are PERSONAL LEXICONS as well. They a re an indication of intimacy, and so are more likely to be found among family members, close fr iends, or partners. In other words, p er sonal relation s also have linguistic effects, creatin g tiny sub­ communities, even of just two peop le (e.g. partners) . The redefinition of speech community revealed by sociolinguistic research will cause us to reexamine the notion of a language (see §7 .2). The definition of speech community p roposed by sociolinguists and elabor ated by Clark links linguistic units more closely to their social context of use and also to the body of shared knowledge which is the foundation of the meanings t hey commun­ icate (see §4.3.2). The codes u sed in the communities we belong to overlap, thus breaking down the boundaries between internal and external change is another p otential source of language change (see chapters 7- 8).

Language, communication and convention 4.2.4

95

Convention

The definition of language as a population of utterances in §2 .4. l is not the more common definition of a lang uage. The more common definition of lan­ guage is as a system, more p recisely, a conventional system for comm unication (where communication refers to the lower three levels described in §4. 2.2) . Conventions are manifested in the ling ueme pool of a lang uage (see §2. 4 . l). Conventions, commonly called 'norms' in sociolinguistics, are central to the understanding of language as a system and language in use. Convention is also central to any model of language change, because innovation is essentially language use beyond conventions (§3 . 4 .4), and p ropagation is essent ia lly the establishment of a new convention in a language (see §§7.2-7.3). In this section, we will present a defi nition of convention based on Lewis ( 1 969) and Clark ( 1 996). L anguage is a joint action (see §4.2.2). Speaker and hearer share the same goal (at the lowest three levels of speech acts): communication. Speaker and hearer each do their part : t he speaker communicates some meaning and the hearer understands the same meaning (or something like it; see §4 . 3 . 3). These are the two distinct individ ual actions that make up the joint action of com­ munication. That is, speaker and hearer believe they intend to have the same, shared meaning when the speaker utters certain words in a certain grammatical construction . How does joint action succeed (when it does)? T here is a problem here: the hearer cannot read the speaker' s mind, and she can't read his. This is what is called a COORDINATION·PROBLEM (Lewis 1 969:5-8; Clark 1 996:62-5). In speaking and understanding , speaker and hearer are trying t o coordinate on the same meaning . To see how this is accompli shed, w e can first look at simpler kinds of coordination problems. A simple type of coordination problem is a third-party SCHELLING GAME, a type of cooperative g ame from game theory (Schelling 1960, discussed in Clark 1 996:62-3). In a third party Schel ling g ame, tw o people are presented by a third party with a set of stimuli such as the following set of numbers, and are asked to choose one, with the idea t hat both people should choose the same number: (3)

23 4 5 99 57

The players may succeed in a number of ways. They might both choose 99 because it is the highest two-digit number and it h as the same two digits. They might both choose 23 because it is the 23rd of September. Or they might choose 57 because they both k now that is the age of the person who set the game. What ma tters, however, is that each player believes that the solution each chooses is the most salient choice for both of them . I f one player knows that the other doesn't know what day it is, then 23 is not a j ointly salient choice. The essential strategy here is that the two players choose the jointly most salient possibility, that is, the choice that is maximally salient in the two pl ayers' common ground.

96

A theory of language and meaning in use

Third-party Schelling games can be tricky, since the players do not know how well the third party set up the game . Fortunately communication is a first-party Schel ling game: by choosing the words and the constructions, the speaker sets up the game of having speaker and hearer settle on the same meaning for the utterance. The participant who sets the problem is the speaker. The speaker wants the hearer to end up with the same meaning as the speaker has in mind. (We will take meaning to be some thought pattern in the mind; see below.) The speaker produces an utterance which she thinks will allow the hearer to u nderstand the same meaning that the speaker intended. The hearer knows this. That is, the hearer knows that the speaker will have done her best to help him solve the coordination problem of understanding the same mean­ ing that the speaker intended. The speaker and hearer can assume what Clark calls the solvability premi ses (Clark 1 996 :6 8): ,

(4) The speaker: chose the problem (communicating something); (i) (ii) designed its form (chose the words and grammar of the utterance she produces); (iii) has a particular solution in mind (the pa rticular meaning i ntended); (iv) believes the participants can converge on that solution (that is, the meaning intended will become part of the common ground of the participants) . In addition, the participants in a conversation can assume two other premises, according to Clark (ibid 69): .

(5)

sumcIENCY PREMISE:

the first party has provided all the information needed

to solve the problem (6)

the participants can solve the problem immediately, that is, with no delay

IMMEDIACY PREMISE:

The sufficiency premise corresponds to the function of avoiding misunder­ standing (§3.4.4) . The immediacy premise is necessary because conversation is a time-constrained sequence of coordination problems: it is a string of words and sentences that comes at the hearer in very rapid succession. The imme­ diacy premise partly motivates the economy principle (§3.4.4). There are a variety of ways to solve coordinati on problems. In the third­ party Schelling games, the solutions that I suggested were b ased on natural perceptual or cognitive salience: the perceptual distinctiveness of the two 9s in 99, or its cognitive salience as the highest tw o-digit number. More speciali zed cogn itive s alience is found in the date, or the age of the third party (assuming these facts are mu tually known and contex tually salient) . N atural perceptual or cognitive distinctiveness is a COORDINATION DEVICE that can be used by people in order to win Schelli ng games. Of course, perceptuaVcognitive salience is not perfect, but that's the way the world is, since people cannot read each other's minds. No coordination device is guaranteed to be foolproof - and thi s is one of the cracks in the system that gives language change a foothold.

Language, communication and convention

97

C h oos i n g one number out of a set is an e x tre mely limited sort of coordina­ ti on problem . Trying to figure out which of the infinitely possible range of ex­ per iences that a spea k er is t ryi n g to comm u n icate is a participant coord i nati o n problem of an entirely different ord er of magnitude. To say it aga i n: peo ple cannot read each other's minds (most of the time anyway). There are just too many things someone might want to communicate. Moreover, almost any gesture , particularly any string of sounds , could be used to mean anything . This is the arbitrariness of t he lin guisti c sign. So there is no p ercept ual l y natural salient solution to the coordination pro blem of communication. What do pe ople do? Fortunately, pe opl e do want to communicate the same things, or similar thin gs , over and over a gain . That is, co m mun i ca ti on involves recurrent coor­ d inati o n p ro b lems . And speec h c omm u ni tie s generally end up c onverging on a regul ar s olution to a recurrent c oordination pr o b lem . S peec h communi t ie s arbi t rari ly pick one solution, say the string of sounds butt erfly to me an the in­ sect, and s t i ck with it, so that everyone recognizes that butterfly is the solution to a recurrent coordination pr ob l e m . In this way conventions ari s e . A CONVENTION is defined by Clark as follows (Clark 1996:71):

(7)

l

2 3 4

5

o f sounds bu t terfly) could have used Schmetterling i ns tead ; that's what the German speech comm uni ty did) that is common gr ound in a com m uni ty (we in the English language c o mm unit y all know we use butterfly ... ) as a c oordination device for a rec ur re nt coordination pr o b l em (talking abo ut butterflies). a regu l arity in beh avi o r (e.g. producing the s trin g

that is

pa rtl y

arbitrary (we

Lewis (1969) p ro vides a more detailed definition of conven ti on . How can a

coordination device in a community? Lewis argues that it has to be a regularity in behavior that almost everyone in the community c on forms to , almost everyone expects almost everyone else to conform to, and almost everyone would prefer any new member of the c o mmu n ity to conform to. Lewis defines a con ve n ti o n as arbitrary when another s olution (s uch as Schmetterling. or papillon) is of approximately equal preference to the m em bers of the com­ munity. In fact, this ex am ple is not perfect, since Schmetterling and papillon are not of appro xi ma tely equal prefer ence because th e ir ph o n olo gy is not English. A better exampl e would be to compare any of the myriad British English vs Ame ric an English vocabulary d ifferen ces , such as intersection vs crossing or tissue vs kleenex . In ad dition , a co nve ntion is arbi trary to the extent that if almost everyone con form e d to the othe r solution, almost everyone would pre fe r that any new mem be r of the community conformed to the o ther sol u tio n as well. Th a t is, p e o ple don't ca re which solution is chosen - the so l uti o n is arb itrar y - as lo n g as almost everyone conforms to the solution chosen. It is the co n form ity that allows a co nvention to be a coordination device. Hence, the combined Lewis & Clark d e fin i ti on of (linguistic) convention is as fol l ow s (the numbered parts of the definit i o n in (8) are from Clark, while the convention be a

qualifications under the letters are from Lewis):

A theory of language and meaning in use

98 (8)

1 2

3 4

5

a regularity in behavior (producing a string of sounds) that is partly arbitrary a. other regularities in behavior would be approximately equally p ref­ erable by almost everyone in the community that is common ground in a community as a coordination device a. almost everyone in the community conforms to it b. almos t everyone expects almost everyone else to conform to it c. almos t everyone would prefer any additional member of the com­ munity to c onform to it if almos t everyone in the community already conforms to it d. almos t everyone would prefer any new member of the community to conform to another regul arity if almos t everyone in the com­ munity were already conforming to it for a recurrent coordination p roblem (communicating a meaning).

It is imp ortant to note that following a c onvention pres uppos es prior occur­ rences of the behavior. That is , the lingueme has already been used in the community to communic ate the meaning in ques tion. Conventions are only es tablis hed after a lingueme has been p ropagated through a speech commun­ ity. Innovation mus t always involve nonconventional coordination devices (see §4.3.l)� Propagation is essentially the establishment of a convention . Linguistic conventions are coordination devices; they must still be used by the interlocutors to solve the par ticipant coordination problem of fixing on the meaning intended by the speaker. Lewis (and Clark following him) argue tha t a language system i s a special type of joint action (or system o f joint actions), which Lewis calls a SIGNALING SYSTEM (Lewis 1 96 9, chap ter IV). In a s ignaling s ystem, the speaker's action is to produce a signal based on the state of affairs she intends to communicate, and the hearer's action is to produce a response to the s ignal produced. The signaling system can be defined as the foll owing individual comp onents of a joint action: (9)

Speaker: Hearer:

state of affairs �signal signal� res ponse

The s tate of affairs and the response are in some systema t ic relat ionship which we may describe as the spea ker and hearer converging on the same meaning. This is not so obvious in the philosophical rea list version of the joint actions in ( 9) given by Lewis (and Clark). More accurately, a state of aff airs gives rise to a mental s t ate in the speaker, and that menta l st ate is the basis on which she produces a signal. Conversely, the hearer enters into a mental state based on the signal he understands, and then responds accordingly. This psy­ chologi cal version of t he s ignaling system can be defined as follows:

(10) Speaker: [ state of affairs�] mental state� signals s ignals �ment al state [� response] Hearer:

The openendedness and flexibility of meaning in

use

99

I n the psychological version, it i s usually assumed that the speaker and hearer are coordinating on gettin g into the same mental state, which is usual ly spoken of as ' the meani ng' of th e signal , that is, the mental representation of the mean i ng of the signal This is i n fact not a necessary i nterpretati o n of this model . The mental s tates of speake r and hearer need not be identical, though it is assumed that they are systematical ly related. In §4. 3, I will give reasons why the psychological version is more accura te but also why a weaker link bet ween the mental states of speaker and hearer is necessary. At this poin t we now have a precise definitio n of how a language system is a conventional sys tem for communication : langu age is a conventional s ignaling system. Ling uistic convention is ce n t ral to the theory of language change . Normal replication, altered replication, and selection are all defined in tenn s of conven­ tion (as conform in g to convention, not conforming to convention, and estab­ lis hi ng a convention, re specti ve ly). The detailed definition of convention given here is imp o r tan t as a starting point for understanding the processes involved in language change. It a l lows us to define nonconvent ionality in language use in the rest of this chapter , and to describe the establishment of conventions in chapter 7. .

,

,

4.3

The open-endedness and flexlblllty of meaning in use

I n this section, I

will

argue that" mean ing in u se is not fully captured by

the

conven tions of a language. All language use, but particularly in no va tive lan­

guage use, involves nonconventional coordination devices. The conventional meaning of a li ng u i st ic expression cannot be as specific as a small fin ite set of semant ic propert ies. Meaning is encyclopedic and subject to the concept ualiz a­ tion of the speaker. Both speaker's meaning and the meaning of an expression in a speech community are better described as a lineage of rich , context- specific meanings for wh ich the expression has been used. Finally, meaning is negotiated join tly between speaker and hearer, j ust as other speech acts are (§4.2 .2). All of these asser tions about the nature of meaning are controversial in some theories of semantics, although cog nitive linguistic approaches to seman­ tics have adopted most of them. However, in chapters 5-6, I argue that the openendedness and flexibility of meaning in use motivate speakers to reanalyze the relations hi p between form and meaning, and that much innovation in grammatical ch ange involves form-function remapping. Hence, the l en gthy discussion in the following sections is intended to persuade the r eader that these assert ions represent an accurate description of the nature of meaning. 4.3. 1

Nonconventionality in language use

In order to succeed in their goals, speakers must comm unicate success fully; that is to s ay, speakers and their hearers must s uccessfu lly solve a coordination

100

A theory of language and meaning in use

problem when a speaker chooses an expression, conventional or nonconven­ tional, to describe a situation. Nevertheless, convention is not sufficient to describe the relationship between form and meaning in language use. In fact, as I will argue in later sections of this chapter, there is no sharp distinction between conventional and nonconventional coordination . To begin with, how does a linguistic convention get established in the first place? Since a convention involves conformity to a community behavior, there has to be a community behavior in place already. Hence the first use of a word or phrase with a particular meaning is not a conventional use by definition: it is not conforming to an existing community behavior. After the first use, or the first few uses, the word - the coordination device for this meaning - can spread through the community, that is, it is propagated. The first use, or the first few uses, are not conventional though - they do not satisfy the subconditions in clause ( 4a-b) of the Lewis & Clark definition given in (8). But it is still communication. Hence it is a coordination problem, and the new word, or new use of a word, is a coordination device used to solve the coordination problem. The coordination problem must be solved by NONCON­ VENTIONAL COORDINATION DEVICES. Any coordination device will do, as long as it works, which means as long as it is jointly salient to the interlocutors. Lewis (and Clark) suggests three nonconventional coordination devices that are rel­ evant to language use. In the Schelling game in (3), I suggested that the coordination problem could be solved by using JOINT SALIENCE, perceptual and/or cognitive. This is the most important nonconventional coordination device available to human be­ ings, used to solve all sorts of coordination problems other than linguistic ones (Lewis 1969:35-6; Clark 1996:81). To give one linguistic example: I coined the compound name CD room to refer to a small bedroom in our house used to store files, books, records and compact discs. The last are jointly most salient to us in that there are more of them than any other object in the room; they are found in that room only, thus allowing for a rapid solution to the room refer­ ence problem; and they are used the most often (i.e. it is the most common reason for my going into the room). Hence the choice of CD room instead of file room or record room, due to the perceptual and cognitive salience of that description of the room. This phrase has now become part of our personal lexicon, a convention conformed to by the two of us.1 Of course, this example involves conventional as well as nonconventional coordination. Success in solving the coordination problem with the phrase CD room requires conforming to the conventions for the words CD and room and the compound noun construction [Nounmod Nounheadl· But it also requires joint perceptuaVcognitive salience in order to establish reference. Only if a com­ pletely new word is coined, or if one is learning a language in the first place, is coordination completely nonconventional. Another nonconventional coordination device is EXPLICIT AGREEMENT (Lewis 1969:33-5, 83-88; Clark 1996:80-81). For example, in the Schelling game, if

The openendedness and.flexibility of meaning in use

101

the pl ayers h ad explicitly agreed always to choose the second number i n the l ist, then they would have won every t ime. In la nguage, the most obvious examples are with technical terms, as in this examp le : 'Word knowledge, p rop­ erly viewed, divides into wh at I will call commu nal lexicons, by which I mean sets of word conventions in indi vidual communities' (Cla rk 1996: 107). Assum­ ing we rea ders accept this proposal, we have explicitly a greep to the regularity of using the string of sou nd s/words communal lexicon to � lve the recurrent coordination problem of tal king about set s of word conventions in individual communities. As with perceptuaVcognitive salience, explicit agreement requires conventional coord i nation as well, for the exp lici t agreement is c ouched in conventional ly established language. The last example of a nonc onventi onal coordination device is PRECEDENT (Lewis 1969:36, 119-20; Cla rk 1996:81 ) Participants may solve a coordination problem by appealing to a preceden t without that precedent h aving been explicitly agreed u pon, and without that precedent h aving evolved into a con­ ven tion For example, in a Schelling game in one of my classes where I asked the students to choose a place to meet in downtown Manchester, a pair of players succeeded by choosin g the HMV record store, bec ause they had met t here before. Precedent operates only with the second and l ater u ses of a word or form it requires a precedent, af ter all - b ut before the use becomes estab­ lished as a convention (if it ever does). The role of nonconventional coordination devices in lan guage use may appear to be a m arginal one, found in the first novel uses of words and con­ structions, or in the language foam ing process. But, in fact, nonconventional coordination devices pervade the apparently ordinary , conventional use of language . Clark illustrates a large ran ge of linguistic expressions where non­ conventional coordinati on devices are nec essary for the hearer to un derstand the m eaning. Th ere is good reason to believe that virtually all language use involves nonconventional coordination devices. 2 One of t he most pervas ive sources of nonconventional coordination in lan­ guage is index icality and refe rence. Consider the instruction Sit in that seat, please. The noun seat is a conventional device to indicate a certain class of object. This convention does not however help the hearer ide n ti fy which seat he is supposed to sit in. The demonstrative that is a conventional dev ice to identify some object away from the speaker. At this point, j oint perceptual salience (aided in part by a gest ure) must en te r the picture for the hearer to succes sfully estab li sh re fe rence. A similar story must be to ld for s o ca l led a naph oric pronou ns, inclu ding nul l anaphora, definite reference and proper names, such as the italicized phrases in (11): .

,

.

-

-

(11)

A fe w p a ges of Deacon's prose can be exha u sti ng let a lone 500 That is a pity, s ince he presents ideas tha t really sho uld be aired as the ideas of evolutionary psychology bed themselves down in the wider culture. (Inde­ pendent, 5.x.97) ,

_

.

102

A theory of language and meaning in use

To be successful, pron ouns and null anaphora must refer to entities that the speaker believes are highly access ible to the hearer (Ariel 1 990) . Joint cognitive salience, usually determinable from the discourse basis of the interlocutor's personal common ground, is required to successfully establish reference. Even joint perceptual salience may suffice, on those occasions where a pronoun can be used as a first mention of a referent, where joint attention is already directed to the referent. Likewise, definite reference requires the combination of con­ ventional information of the category of referent - provided by the noun and its modifiers - wi th cognitive salience of the unique instance of the category the speaker is referring to (e.g. which wider culture in ( 1 1)?). Even proper names may refer to different individuals ( there are many people named Deacon), so j oint perceptual/cognitive s alience is necessary to establish proper reference. Clark also notes the existence of indirect reference, as in the following ex­ ample I overheard: ( 1 2)

Our flat is going to the cinema - do you want to come along?

The expression our flat refers not to the physical part of a building, but instead to a join tly cognitively salient entity associated with it, namely the people inhabiting the fi at. This is an example of metonymy, one of several concep­ tualization processes that is pervasive in language and thus requires pervasive use of nonconventional coordination devices. Example ( 1 2) is in fact probably a conventional extension of the meaning of our flat in English, from buildings to their inhabitants. If both meanin gs are conventional, then the hearer is still posed a coordin ation problem: choosing the literal or the metonymic meaning. The resolution of ambiguity also requires nonconven tional coordination devices. Ambiguity occurs when a regularity in behavior ( the linguistic expression) is used to solve more than one recurring coordination problem (the mean ing). A typical example of ambiguity is illus­ trated in ( 1 3), with some competing definitions given in ( 14) : (13)

'But what about Toad?' asked the Mole anxiously, as they set off together. 'We can't leave him here, sitting in the middle of the road by himself, in the distracte d state he's in!' (The Wind in the Willows, p. 37; emphasis added)

( 14)

state

n . 3. A men tal or emotional condition or disposition 6. A social pos ition or rank; estate. 7. Ceremony; pomp; formality: robes of state 10. A body politic; specifically, one constituting a nation : the states of Western Europe. (American Heritage Dictionary, New College Edition) .

.

.

.

.

.

.

.

.

I have selected just one word in the passage in ( 1 3). But in any utterance, most if not all of the words will be ambiguous, as in spection of the entries in a good dictionary for all the words in ( 1 3) would indicate. If many or all words are ambiguous, then the hearer is faced with the par adoxical situation described in the following passage (see also Croft 1 993: 362-4): ·

The openendedness and flexibility of meaning in use

I 03

The polysemy of words is such th at, in a given sentence, one can say that the

word put back into its context, not the word itself, has a univocal meaning. But the context itself is constituted by other words, which also have many meanings. Thus an interaction between words is produced which ends up by attributing to each one the meaning which is compatible with the oth ers . In tum, one attributes to these other the meaning compatible with that of the first. (Merleau-Ponty 1964/1973:91; emphasis original)

Merleau-Ponty does not mention in this passage that there is further c ontext than just the other words in an utterance: there is the shared extralinguistic context, which will u ltim ate ly disambiguate the mean in gs of the words in the

utterance (in most cases; see the followin g sections). But thi s c ontext is avail­ able to the interlocutors only via nonconventi onal c oordination devices, namely joint attention and shared expert ise - the bases for c ommon ground . Finally, there are m any construct ions in English ( and other languages), in­ cluding quite basic constructions, which Clark call s 'contextual constructions', whose c onventi onal meaning seems to incl ud e a requirement that the hearer use nonc onventional coordination devices. Perhaps the best-known E nglish c onstruction of this sort is the compound noun construction and its close relative, the denominal adjective, both illustrated in this newspaper headl ine :

(15) Blood money setback

for

Saudi nurses (Guardian headline , 7.xi.97; em­

pha si s added)

The range of m ean ings linking the two n ouns in the c ompou nd noun c onstruc­ tion is almost completely free ( D owning I 977). I n ( 1 5) , blood money is a c on­ ventionalized compound noun, but the c ombination with setback is novel and requires nonconventi onal coordin ation. Lik ewise, the nurses are not of Saudi n atio n ality ; they were working in Saudi Arabia. Joint salienc e solves this coor­ d inati on problem - they were the only nurses with a S aud i c onnection who were in the British news at the ti me . To say that the meanings of the phrases are 'nurses with some connection to something Saudi' and 'a setback with som e respect to blood mon ey ' does not tell us what these phrases mean, that is, the linguistic coordination problem they are inten ding to s olve. The hearer has not solved the problem unless he h as th e right rel a tionship between the two terms . It is not only c omp ound n ouns that require c ontext, so too do ma ny other m odifier constructions, such as the genitive c onstruc tion and even the adjecti ­ val construc tion, as i llustrated in the ex amp le red pencil (color produced by p encil, color of surface of pencil, c olor of stripes on surface of pencil, c olor of the eraser on the pencil, etc.). Other contextual constructions described by

Clark i nclude indirect descriptions (You'll have to ask a zero, that is, a person who you reach by d i alin g zero), denominal verbs and ot her derivational pro­ ce sses (She Houdini'd her way out of the closet , that is, cleverly escaped), and highly general verbs (Since you're working on the spoon, I'll go ahead (Carol T, 23.xi.97), that is, licking off the last bits of food on it).

104

A theory of language and meaning in use

Another aspect of modification that requires nonconventional coordination, not discussed by C lark, is pinning down the degree of a property (expressed by an adj ective ) possessed by an entity (expressed by a noun). How big is a big book? Is it bigger than a small house? Is a small house for a millionaire the same size as a small house for a working-class person? Gradable properties are expressed by pairs of antonymous adjectives, which implies the existence of a neutral reference point or region from which the antonymous adjective s depart in opposite directions. But the reference point is not absolutely fixed: it depen d s on our common knowledge of the range and distribution of possible values of the property for the particular object, and also on salient properties of the current context. In other words, perceptual and/or cognitive salience are necessary in order to establish the reference point which the hearer uses to determine what degree of a property the speaker is communicating about an obje ct . The d e gr e e to which ordinary language use, apparently conforming to lin­ guistic convention, requires nonconventional coordination devices makes it clear that virtually all language use involves nonconventional coordination. In all but timeless gen e ric sentences, people are using conventional expressions describing types or classes of objects, properties and events in order to talk about specific objects, properties and events. In other words, virtualJy every noun, verb and adjective in virtually every sentence requires nonconventional coordination in order to establish reference to the specific object, property and event being talked about. Moreover, the ambiguity of words requires the hearer to jointl y resolve the ambiguity of most words in a senten ce based on extralinguistic context, that is, again using nonconventional coordination de­ vices. Finally, the context dependence of many constructi ons , in parti cular modifier constructions, constructions involving denominal derivation or noun compounding, and constructions with general verbs, requires nonconventional co ordin ation simply to determine the exact content of the utterance. 4.3.2

Context, categorization and conceptualization

No word means the same thing twice. Robert Musil, The

man

without qualities

In the preceding section, I demonstrated that convention, the chief property of human linguistic behavior that inhibits ch ange , does not account for all linguistic behavior. Nonconventional coordination devices are called for in any innovative use of language, and in language learning . Most innovative use of l anguage is also partly conventional, because it is based on the recombination of conventional coordination devices (words and constructions). But ordinary language use that does not appear to innovate also requires the pervasive use of nonconventional coordination devices. There is no sharp distinction between conservative and innovative language use. All language use is innovative, to some degree.

The openendedness and flexibility of meaning in use

1 05

The chief reason why even conven tio n al lan gua ge use is innovative is that there cannot be a word or ph rase to describe every experience that p eop le wish to communicate. Every object and action and concept has a unique ident­ ity, alth ough they share similarities with other obj ects , actions and concepts. Language - the co nvention al sign aling system - attempts to solve this p ro blem by iden tifyi ng recurrent situations that peopl e wish to communicate; that is clause (5) of the Lewis & C lar k definition of convention. It is assumed that it is recurrent situation types that linguistic conventions describe. The full richness of the mea ni n g to be conveyed in a particular context of use must be achieved by the interlocutors using nonconventional coordination devices. But this account of linguistic meaning is serious l y flawed. The line between categorie s of recurrent situation types and th e full richness of specific ex periences to be communicated simply cannot be drawn. Hence convention is itself fluid to some degree - a degree that must be agreed upon by the interlocutors in particular occasions of use. 3 Grice distinguishes two types of me aning on the basis of the supposed dif­ ference between the infinite richness of experiences that sp eake rs wish to com­ municate and the available conventions of the l an guage they use. The full meanin g conveyed in a particular context is cal led the utterer's occasion mean­ ing (Grice 1 9671 1 989:90) or SPEAKER'S MEANING (C l ark 1 996: 1 26). The recur­ rent situ ation type for a linguistic exp ression is the time less meaning or sentence/ word meaning (G ri ce 1 967/1 989: 89), or, to use Clark's term, SIGNAL MEANING (Clark 1 996: 1 26) . The distinction between s ign a l meaning and speaker ' s meani ng is related to another distinction used to characterize lingui s tic meaning, that between the DICTIONARY and th e ENCYCLOPEDIA. The dict i onary approac h to li n gui s ti c mean­ in g is to argue that there is a smal l subset of information that is part of the l in gu istic (conven ti onal) mean in g of a word or construction ; it is roughly equival ­ ent to the standard view of signal meaning . The encyclopedic approach is that the meaning of a word (or construction) is all that we know about t he situ­ ations for which the word or construction is us ed . The e ncyclopedic view of word meaning is not equivalent to the speaker's meanin g , but rather to the sum total of the h is t ory of spe ak er' s meanings of a word or co nstruc t i on in occasions of use. In the remainder of th i s section, I will argue that the descri p­ tion of sign al meaning as a subset of p roperties is too simple, and instead offer an encyclopedic view of the signal meaning based on an individual's history of encounters with sp ea k er ' s mean ings . This model of meaning reveals a furth er source for variability and flexibility which may lead to lan guage change. Clark argue s that although speaker's mean i ng is established in part by sig­ nal meaning - the conventions of the language - speaker's me an in g is logically prior. For e xamp le , the meaning of contextual co n s tructions is fixed in use. All novel use in language, such as metaphor, metonymy and indirect reference, has its meanin g established only in context . But there is another more important way in which speaker ' s mean i n g is p r io r . A lan gua ge learner acquires sign al meaning on the basis of the history of speaker's meanin gs that they encounter, ·

l 06

A theory of language and meaning in use

including their own uses of the signal (expression) . And this acquisition process is not simple. There is evidence from a variety of psycholinguistic experiments that the relevant features for a concept denoted by a word are variable across contexts (Barsalou 1 98 7 1 993: 3 1 -5; Gibbs 1 994: 3 3 -4; and references cited therein) . For example, piano used in the context of producing music had different features as optimal cues for retrieval than piano in the context of moving furniture. When subjects provide definitions of a category, there is a remarkable degree to which definitions do not overlap between subjects and even in the same " subject on di fferent occasions. Moreover, the features and their ranking can be manipulated by manipulating context. The variability of concepts in context found in these experiments suggests no subset can be specified as 'the dictionary meaning' . Barsalou argues that what is much less variant is the sum total of knowledge of the obj ect, which is presumably stored in long-term memory - in other words, the encyclopedic view of word meanings. Similar evidence comes from language change. Proper names do not appear to have any meaning except to identify individuals, yet they can evolve to possess meanings drawn from contextual properties of those individuals, e.g. napoleonic (Haiman l 980a: 350). A well-known example of peripheral contex­ tual information coming to be a central part of the meaning of a word is English bead from bede 'prayer' , since rosary beads were used in prayer. A more grammatical example is the shift of English since from an originally temporal meaning, with a contextual causative inference, to a causal connect­ ' ive (see §3 . 3 .2, §5.4.3). Content - the 'dictionary' meaning - and context can shift and be interchanged, based on occasions of use . Haiman ( l 980a) argues that all attempts to isolate a subset of our total k nowledge of an entity and identify it as the dictionary meaning are ultimately untenable. Perhaps the most plausible division is based on essential informa­ tion about a concept (dictionary meaning) vs accidental information (encyclo­ pedic meaning; a distinction dating back to Aristotle) . However, this distinction is valid only against a background theory of the entity in question, which determines what is essential and what is accidental (see Murphy & Medin 1 985; Keil 1 989; Gelman & Coley 1 99 1 ) . Such a theory would b e built o n o u r ency­ clopedic knowledge; moreover, theories differ from individual to individual and over periods of time. Consider for example the meaning of meaning: who among linguists and philosophers has the same essential concept for this word? Likewise, objective meaning (dictionary) vs subjective meaning (encyclope­ dia) presupposes that there is a universal unchanging theory of what the object­ ive properties of an entity are; and there is not. Lastly, Haiman draws on Quine's famous argument that the distinction between analytic statements ('true by definition') and synthetic statements ('true by empirical assessment') cannot be used to differentiate dictionary meaning from encyclopedic meaning (Quine 1 951/196 1 ). As Haiman summarizes Quine's argument: ' analyticity within a language can be defined only in terms of synonymy and definition, which in turn, however, can be defined within a language only in terms of analyticity' ,

The openendedness and flexibility of meaning in use

1 07

(Haiman l 980a: 349) . Another way of putting it is that it is only in terms of some theory we have about phenomena that some statements describing those phenomena can be definitional . Quine makes the further point that meaning/knowledge of words and state­ ments is interconnected (cf. Gadamer 1 972/ 1 976:84-S). In our terms, the theory is in essence the organization of our encyclopedic knowledge taken as a whole; one cannot isolate the definitional part of it, let alone identify it as the diction­ ary meaning which characterizes the recurrent situation for a conventional linguistic expression . Thus, the meanings of individual words and construc­ tions cannot be isolated mental representations, as is assumed in the· dictionary view of linguistic semantics, but access points into a single network of encyclo­ pedic knowledge ( Langacker 1 987 : 1 63). The flexibility of what knowledge is activated on particular occasions of use is manifested in the results of the experiments on concepts reported by Barsalou and Gibbs. Even the encyclopedic knowledge model of meaning does not exhaust the richness of meaning. Searle ( 1 979), Winograd ( 1 980) and other cognitive sci­ entists have argued that the sort of knowledge that is relevant to a linguistic expression's meaning cannot be enumerated in such a way that all contexts of use can be predicted. Searle argues that a basically infinite set of background assumptions is required to characterize the literal meaning of an utterance, and hence its appropriate use in context. Consider example ( 1 6): ( 1 6)

Give me a hamburger, medium rare, with ketchup and mustard , but easy on the relish . (Searle 1 979: 1 27)

We assume we understand what the literal meaning of this request is. Specify­ ing it is much more difficult , however: Suppose for example that the hamburger is b rou g ht to me encased in a cubic yard of solid lucite plastic so rigid that it ta k es a jack hammer to bust it ope n , or

suppose the h am burge r is a mile wide and is 'del ivered' to me by smashing d own the w al l of the restaurant and sliding the edge of it in. (Searle 1 979: 1 27)

Langacker makes a similar observation with a comparable example ( Langacker 1 988: 1 6): ( 1 7)

He is barely keeping his head above the water.

imagine a race over t he ocean by helicopter, where the contestants must tra n spo rt a severed hea cl , su s pe nded by a rope from the helicopter, from the s ta rti ng line to the finish; a co ntest a nt is disqualified if the head he is carrying ever d ip s below the water's surface . (Langacker 1 98 8 : 1 6- 1 7)

Searle points out that refining the semantic representation of the utterance to exclude all possible qualifications is impo s sib le : one can always create new contexts in which the utterance is inappropriate and other minimally different contexts in which it is appropriate (Searle 1 979: 1 28).4 For example, the philo­ sopher's favorite sentence The cat is on the mat seems inappropriate when cat

1 08

A theory of language and meaning in use

and mat are floating weightless in space; but if we are in a spaceship with cat-mat pairs floating around in just two configurations, mirror images of each other, the sentence could be an appropriate answer to the question, Which is it now? (Searle 1 979: 1 22-3) - if the interlocutors agree to it. Searle's cat-and-mat example demonstrates the problem in describing a re­ current situation for a conventional expression . Is outer space context a recur­ rent situation for The cat is on the mat? Only if the interlocutors choose to conceptualize it as such. A situation does not simply present itself as a recur­ rence of some prior situation type, that is, as belonging to the same category as some prior set of situations: it has to be conceptualized as such by the inter­ locutors jointly. It is a truism that situations can be categorized in many different ways. For instance, a speaker can categorize an individual whom she wishes the hearer to identify in.a potentially infinite set of ways: Tony Blair, neighbor, Labour leader, man, prime minister. Each category is defined by the string of prior coordina­ tion problems which the string of sounds has been conventionally established to solve. For example, the string of prior coordination problems for which Labour leader h as been used includes reference to Michael Foot, Neil Kinnock, John Smith; the string of prior coordination problems for which prime minister has been used includes reference to James Callaghan, Margaret Thatcher, John Major, not to mention a host of prime ministers in other countries. A s in gle coord in at i on pr oblem has several different solutions, that is, every­ thing we talk about can be categorized in different ways. Whi ch solution (cat­ egory, word) the speaker chooses depends on j oint salience . Britons in Britain can use the phrase the prime minister in certain contexts to refer to Tony Blai r because their common ground makes Tony Blair - not the prime minister of Turkey or some other place, or the prime minister at some other time - the jointly most salient prime minister. . However, joint salience is not just there to be exploited by speaker and hearer; it can be created. By choosing a word to refer to something, we identify the current coordination problem as being the recurrence of certain coordina­ tion problems that h ave been solved before, and the hearer knows those prob­ lems and solutions as much as we do (clause (3) in the Lewis & Clark definition of convention). The speaker causes the hearer to compare the current situation to the prior ones, and thus makes ( jointly) salient certain aspects of the situ­ ation by the choice of words. In the cat-and-mat-in-outer-space example, speaker and hearer may or may not define the situation as a recurrence of the situation type conventionally denoted by The cat is on the mat. Once they choose to use the sentence or not in the outer space context, however, the interlocutors have altered the meaning of The cat is on the mat for themselves, by including or excluding the outer space context as an instance of the recurring situation type for that sentence. The clearest cases of conceptualization is when there is overt dispute as to categorization, and hence meaning, of expressions. For example, by choosing to refer to it as a fetus or an unborn baby, one has identified the current

The openendedness and flexibility of meaning in use

l 09

coordination problem as a recurrence of two different situation types or cat­ egories. The term fe tus can be used for any conceived-but-not-yet-born offspring of any animal (and thus dehumanizes it); while baby is used for already-born and specifica lly human offsprin g (see Bolinger 1 980: 1 38). Either term will suc­ cessfully establish reference, that is, taps into jointly salient properties that allow the hearer to k now what the speaker is talking about. But they are different properties and the p roperties chosen thereby become more salient. Speaker's meaning is flexible and highly dependent on context. Inferring signal meaning from prior occurrences of speaker's meaning is possible only by building an encyclopedic representation of meaning that is, re fl ecting the full range of common ground in the community using the conventional signal. Hence we cannot differentiate speaker's meaning from signal meaning in any particular context of use. Instead, we must define meaning in two ways with respect to a different distinction . These two types of meaning are based on the distinction between a language as a population o f utterances produced by a speech community, and a grammar as an individual speaker's knowledge about the conventions of the speech community. The COMMUNITY'S MEANING of a lingu i stic form a lin gueme - is the line age of replications of its use, in their full encyclopedic, cont extual value. The INDIVIDUAL'S MEANING of a lingueme is a mental structure that emerges from the individual's exposure to (necessarily partial) lineages of the community's mean ing includi n g of course the use of the lingueme by that same individual . This is not to say that an individual speaker stores separ­ ately every individual meaning-in-context she has heard or produced Particular meanings-in-context can crystallize into types in an individual speaker' s mind, or be joined in networks of related meanings in her mind ( polysemy; cf. Gibbs 1 994:40-7). But an individual's meaning can be thought of as a sedimentation of the history of uses to which she has been exposed ( i nclud i ng her own uses) . Both the community's meaning and the individual's meaning are always subject to change. Every new use alters the lineage of the community's mean­ ing. Every new use to which a speaker is exposed (or which the speaker creates) alters the individual's meaning as well. A speaker's encyclopedic knowledge is based on finite prior situations of use, which includes an indefinitely large number of background assumptions about the conditions of categorization. These background assumptions do not all apply to the new situations we wish to communicate. Hence, the individual's meaning - our prior encyclopedic knowledge of the situation type - is not obviously or easily extendable to new situations where communication is desired. In fact, applying old words to new situations is often a creative, and even con trove rsial , act, conceptualizing the new situation as a recurrence of an old one. And when the conventional signals are extended - or not extended - to new situation types, the ind ivi dual s mean­ in g is changed. The hearer is expected to recognize the situation being communicated by the speaker as a recurrence of prior situations. Yet the hearer has his own history of uses of the lingueme, comparable to the speaker's only to the extent that -

-

,

.

'

1 10

A theory of language and meaning in use

they share common ground . And the speaker is conceptualizing the current situation in a particular way, so that the hearer is not only recognizing the situation, using conventional and nonconventional coordination devices, but also concurring with the speaker in her partially creative act of categorization. As we will see in the following section, the hearer's role in this process is not entirely passive.

The final level of complexity in relating the meaning of the current utterance to prior uses has to do with the structure of the utterance itself. An utterance is a recombination of replicated linguemes. The recombination of linguemes chosen by the speaker on a particular occasion of use represents her attempt to conceptualize the current situation as a (partial) recurrence of prior situations. Likewise, the hearer must also recognize I share in the conceptualization of the current situation as a recurrence of prior situations based on the lineages of those linguemes in his own histories of their uses. This final level is the locus of much grammatical change, as I w i l l argue in chapters S-6. 4.3.3

Joint construal and the evolution of interpretation

In §4.3.2, I emphasized the flexibility and even fluidity of meaning in use: 'language is made up of significations in the state of being born . . . language is in movement and is not fixed; and perhaps because one must recognize in the last analysis that there are "flowing significations" ' (Merleau-Ponty 1 964/ 1 973:88) . Nevertheless, I have tacitly assumed that for any particular occasion of language use, the me an in g -'-- speaker's meani ng - is fixed, no matter how complex it is and how subtle the coordination problem is. In this section, I suggest that speaker's meaning in any particular occasion of use is not entirely fixed - thereby allowing a still greater degree of flexibility in language use that opens the door to language change. There is a common model of communication which presupposes fixity of speaker's meaning that has been called the CONDUIT METAPHOR (Reddy 1 993). In essence, the conduit metaphor assumes that the speaker packages her thoughts and/or feelings - which Reddy abbreviates as RM (repertoire member) - in a linguistic expre ssi o n and the hearer unpackages them upon hearing t he expres­ sion. Linguistic expressions are conduits by which the RM moves from speaker to hearer. There is a vast number of everyday English expressions that use the conduit metaphor, some of which are given below (all examples from Reddy 1 993): ,

( 1 8)

Language is a conduit for transferring an RM from one individual to another a. You know very well that I gave you that idea. b. Your real feelings are finally getting through to me .

( 1 9)

Humans place their RMs within the external signal a. It is very difficult to put this concept into words. b. He crammed his speech with subversive words.

The openendedness and flexibility of meaning in use

(20)

111

Signals convey or contain RMs a.

His wor ds carry little i n the way Your words seem rather hollow.

b.

(2 1 )

o f recognizable me an i n g

.

Humans .find RMs within the signals and take them into the{r heads

a.

Please pay a ttenti o n to w h a t s there in t he words! We will see this t h o u ght several times again in the sonnet. '

b.

Speakers/writers eject their RMs into an external 'space ' M ary p o u re d out her sorrows. b. T h e essay brings out u n us ua l thoughts on the m atter .

(22)

a.

There is an obvious error in the theory of communication behind t h is meta­ pho r and a number of i m p o rt a n t consequences that follow from rectifying this error. The obvious error is that thoughts or feelings cannot g o anywhere outside of the minds of humans, whether it is 'into' words or i nto an external sp ace . R e ddy proposes what he calls th e toolmakers parad igm (so cal l ed after the fanciful example he uses to i l l u s t rate it) that av o ids this error. The Lewis & Clark m odel of convention an d communication also avoids this err or . The whole idea behind a coordination p roble m is that getting s pe a k er and hearer to c onve r ge on the same meaning is a p r o bl e m precisely because our tho u gh ts cannot leave our he ads A le s s obvious error that follows directly from t he pri m ary one is that li n gu isti c expressions do n o t cont a i n mea nin gs Meaning is s o met h ing that o ccurs in the interlocutors' h e a d s at the point of l a n gu a ge use ( s pe ak e r s mean­ ing), or s o me t h i ng that represents a me m o ry of a history of uses available to a spe ak er albeit organized i n t o senses and sense relations all embedded in a ne two rk of encyclopedic kn o wled ge Reddy e m phasize s this fact that for i n stance a bo dy of texts does n o t h av e mea n i n g ; there must be a b as is for reade r s to evo k e a m e a n i ng in their heads through shared k n o wl e d ge with the cu l tu re that produced the text s . It is not so obvious that the Le wi s & C la rk model of c o nve nt ion avoids this error. Instead, in the preceding secti on l argued that one m ust interpret the notion of a recu r re n t situation in the d efi n i ti on of co n ve n ti o n quite l oosely in order to accommodate the fluidity of m e a n in g Li ngu istic conventions are n o t th at ri gid Another error that follows from the c ond u it me t a pho r model of commun­ ica t i on is that s u c c e s s ful communication is a perfect match of the spea k e r s i n t e n ded m e a n i n g and the hearer' s u n d ersta n d i ng via the 'transfer' of tho u gh t s. Unsuccessful communication is a m i s m a t c h res u l t i n g from t h e hearer's mis­ understanding or possibly the speaker's improper pack a g i n g of her thoughts. Since t h o ugh ts do not t ravel success i n co m m u ni c at i o n has to be c onstrued differently. In the s imp l e S c helling game in (3), success seems to be e a s i ly assessed: the pl ayers perceive that they chose the same number, or th ey did n o t . But how do interlocutors fig u r e out they h ave succeeded in communication? After the ,

'

'

'

'

,

.

.

'

,

.

,

.

.

'

,

,

1 12

A theory of language and meaning in

use

utterance is produced and the hearer has u nderst o od it, in his own way,

how

do speaker and hearer know they h a ve the same meaning for the utterance, in

particular the meaning intended by the speaker? They have no direct means of knowing. They have o nl y an indirect means: if what the hearer says and does in response appears to follow n aturall y from a recogn i tio n of the m ean in g that the speaker intended. And the hearer kn ows that the spe ake r understood his response, and h ence he understood her origi nal utterance, on the basis of what her next response to him is. And so on. But Clark obse rves that this process need not require a fixed interpretation of the speake r' s first utterance as the speaker's original intended m ea n i ng for her utterance. First, a n umber of turns in conversation may be required to establish the jointly accepted meaning, and the hearer himself may contribute to the meaning being constructed (example (23) from Cheepen & M onaghan 1 990: 1 57): ( 23 )

l K: 2C: 3K:

4C:

once t hos e cameras start flash i ng partic ul a rl y with the infants it puts t h e m off [hearer co ntinuatio n of 1 j it p u t s them off [acceptance of 2 by repetition j [acceptance of 3 by a continuerj yeh

M ea n in g in use is defined by JOINT CONSTRUAL: ' Fo r each signal, the speaker and addressees try to create a joint construal of what the speaker is to be taken to mean by it' (Clark 1 996: 2 1 2). Of cou rse a joint construal, like any other joint activi ty , is not always easy to achie ve . The standard case, where the speaker's o ri g inal intended m e an in g is ultimately accepted (more precisely, is t ake n to be accepted , by con t in uing onwards), is found in what C l ar k calls verified construal and corrected misconstrual (Clark 1 996:2 1 5): 5

(24)

Verified construal I C:

2A: 3C:

Buck's Fizz, is t h a t Champagne . Very classy [etc.]

a mixture of wine

and

orange j u ice ?

(25)

Corrected misconstrual l S: what do you have to wear? 2B: wh,at . . for football 3S : yeah 4B: em . . lin gerie SS: (laughs and ends up coughing) what do you have to wear seriously for football . . what do you have to wear?

(26)

Corrected misconstrual I N: Oh . . didn't you notice I put the dining room table back? 2P: Dining room table? Why where's it been? 3N: The k i tche n table I m ea n l cl eared i t yes t erd ay . .

4P:

No

The openendedness and flexibility of meaning

in

use

1 13

In (24), C's simple continuation after A's response implies that A's answer to C's question is sufficient evidence for C that A understood C's question. Examples (25)-(26) indicate misconstruals that are corrected by he arer and spe ake r respectively. In (25), B's re spo n se in 4 to S's question ind icates that B d id not take S's question seriously, and ( after laughing) S asks the question again, add i n g seriously to her utterance. I n (2 6) , P's response to N's question causes P to correct herself. Co rrecte d misconstru als are less infrequent than one might expect. Labov notes with re gard to his Cross-Dialectal Comprehension project, ' [this] s tu d y has demonstrated th at the actual number of misunder­ st a ndin gs in eve ryday life is much greater than casual observations lead us to believe' (Labov 1 994: 565). Note that in (25), B's joking response is probably a deliberate mi sco nstr ual ; not every misconstrual is accidental. Nor is every 'misconstrual' rej e cted . In the more interesting cases, there is an alteration of the speaker's original inten­ tion which is accepted by the speaker as the basis for the continuation of the interaction. Clark describes two types, which are more or less ill ustr ated with the following examples : (27)

Revised construal I just ope n ed i t [film] like you were going to. l K: 2 M : In the dark? 3K: Yeah. And it had rewound so 4M: D i d n t like my idea then, that's fine. 5K: I'm sorry . .I just . . I was n erv ous and I thought about it and decided it would be a good idea rather than tryi ng to figure out about sending some t h i ng to M ilton Keynes and sp e n din g a lot of money and (inaud .) '

(28)

Narrowed construal You need some kitchen roll I H: 2L: Can I have that as well? 3H: Yeah just take it

In (27), K's utterance in 1-3 (including M's contribution to the utterance in the dark) is const rued by M in 4 as an insult; in 5, K accepts that construal and apologizes. Here the revised construal pe rt ain s to an emotive aspect of the speaker's meaning. In (28), L construes H's general observation as an invita­ tion to take the kitchen roll, prob a bly a narrowed construal of H's orig ina l intention, and in 3, H acce p ts that construal by agreeing to L's reque st in 2. In all of these examples, construal of the meaning of an utterance is n ot necessarily fixed by the speaker' s original intention, but is jointly nego ti ated and agreed upon over the course of th e conversation: unders tand i n g is not reconstruction but mediation' (L in ge l 976:xvi) . What may seem like marginal cases in the se attested examples can be multiplied in other contexts. Any teacher who asks students to write essays based on class lectures and sees what is written quickly realizes the scope for altered construal in un dersta n d i n g . One '

1 14

A theory of language and meaning in use

might att ri b ute this to students' uncertain grasp of the common ground of the com m un i ty in to wh ich they are being introduced in the class. But any scientist who has given a lecture to colleagues or d iscus sed his or her own work with colleagues, will recognize that revised construal is a very common phenomenon : ,

Science is a conversation with nature, but it is also a conversation with other scientists . Not until scientists publish their views and discover the reactions of other scientists can they possibly appreciate what they have actually said. No matter how much one might write and rewrite one ' s work in anticipation of possible responses, it is impossible to avoid all possible misunderstandings, and not all such misunderstandings are plainly 'misunderstandings' . Frequently sci­ entists do not know what t hey intended to say until they discover what

it is that o ther

scientists have taken them to be saying. (H ull 1 98 8 : 7 ; emphasis added)

In other words, a speaker's orig inally intended meaning may not end up be ing t he j oi nt ly accepted meaning of the utterance . Other writers in other d i sc i pl i nes have made similar observations: For the speaking subject, to express is to become aware of; he does not express just for others, but also to know himself what he intends . . . Even we who speak do not necessarily know better than those who listen to us what we are express­ ing. (Merleau-Ponty 1 960/ 1 964: 90, 9 1 ) The fact is that i n most cases an author does not understand all the meaning of his work . (Eliade 1 9 52/ 1 99 1 :25) language does not yet contain its meaning . . . all communication supposes in the listener a creative re-enactment

or' what

is heard. ( Merleau-Ponty 1 962/ 1 964 : 8)

Successful communication is not easy. But successful communication involves not t he recovery of an o ri gi nal, 'correct' i nt erpretati o n of the speaker's original intention, but instead an inte rp re t atio n that e v ol ves over the course of the conversation , and is as se s s ed by the success or failure of the highe r social­ interactional goals that t he interlocutors are st ri v i n g to achieve . One re a son wh y this effort is not doomed to fail u re from the begi n n i ng is the fact that although we cannot read each other's minds, we do inha b it a shared world. Even with the Sche llin g game examp le in (6), assessment of success is due to joint attent io n in the shared perceptu al field (the players po i nt to wh i ch number each chose). Reddy reco gn i ze s that there exists 'an "a p rio ri shared context", a pre re qu isi te for achi evi ng any communication whatsoever' (Reddy 1 993: 1 8 1 ). Interlocutors can achieve success in communication by means of j oi n t attention to phenomena i n the shared world and by virtue of shared expertise acq u ire d t h r o u gh interaction with others in th e shared world. Never­ theless, the e st ablishmen t of a joi n t s h a re d mean i n g of an utterance does not happen automatically but involves an evolving interpretation negotiated by the i n terlocutors th rou gh the i r con t inu i ng conversation. And one effect of the e vo l v i n g in terpretatio n is e v olutio n of the mapping between form and functi on in language. ,

,

Conclusion 4.4

ll5

Conclusion

This chapter outlines a theory of l an guage

and me aning in use which can be u sed in the Theory of Utterance Sel ection fo r l a n gu age change. In p art ic u l ar , the t h e o ry shows a number of ways in which l an gu a ge u s e g i ve s speakers opportunity, indeed re qu i re s them, to innovate. First, communication is subservient to other social go al s Qf the interlocutors. Achieving those goals often means conforming to linguistic co nventio n as closely as po ss i ble. But sometimes ac h i evi ng t h ose g o a l s , or even ach ieving just the goal of communication, is be t te r done by s tretc h in g or b rea k i ng l i n gui s tic conventions. This is one source of language c h ange , albeit n ot a subtle or o m n ip rese nt o ne . Language use, and the go a ls it serves, is a joint activi t y. It is facilitated by the interlocutors inhabiting a shared world, but h ampere d by the fact that the interlocutors c an n o t read each o the r ' s minds . (I s a y 'hampered', but one can readily perceive t he advantages of this sort of world by imagining what it would be like if we could all read each other's minds.) Thus, the hearer may interpret the meaning of the speaker in a s l i gh tly novel way, and there is no direct check on the hearer's inte rp re t a t i on ( alth o ugh there are indirect checks in the form of n e go ti ated construals and the natural course of the conversa­ tion). The interlocutors utilize various coordination devices, including joint perceptuaVcognitive salience, precedent, explicit a g re em ent and conve n tion ; but no coordination device is fo o l p r oof, and so the h e ar er may end up with a s l i gh tl y di fferent interp ret a tion of t he utterance. Convention is the p ri m ary means by which l i n gu ist ic communication is achieved. But c onve nti o n is not sufficient to evoke for the hearer t he full r ic h ne s s of the experience be in g communicated by the speaker. This is because every situation is n ovel in some way or another. The novel ch a r act er of the situation may be subtle or it m ay be more dramatic (as in the cat-and-mat-in­ outer-space examples) . But the novelty of a si t u at i on means that all l a n gu age use innovates to some d egree . Also, convention is not equal to competence. Members of the speech com­ munity have a different grasp of co n ve nti ons, for instance in the ran ge of situ­ at i on types for wh i ch an e x p re ssi on is conventionally used. These interindividual differences may lead to i n n o va t i on s as well. Fin all y , communities are d e fi n ed i n te rms o f domains of shared e xpe r ti se . Any society is made up of multiple communities. There is a large degree of ove rl ap in the codes used in any s o ci e ty , e s pec i a l ly s o-c alled m o no l i n gu a l soci­ eties. In t h i s way, internal and external fact o rs in language ch a n ge are inter­ twined and this may facilitate la ng u a ge ch a nge . All o f these opp ortun i ti e s t o innovate p e rt ai n t o the expression o f m ean i n g in grammatical form. Hence, our theory of in no vat ion in l a n gu age chan ge must be cast in terms of an alteration of the relationship between form and m e a ni n g . This theory will be outlined in the following chapters .

1 16

A theory of language and meaning in use

Notes It may be objected that after the first uses of CD

room,

the phr ase is still not

conventional because it is not sufficiently arbitrary: no other solution to the coor­

dination pro blem is approximately equal. The claim is in effect that the two of us compute the phra se CD room on the basis of joint perceptual salience every time we use it. This alternative is the pragmatic analysis, in co ntrast to the semantic anal ys i s where CD room is simply stored in our minds as a learned convention of our per­ sonal lexicon. There is little motivation for the pragmatic solution beyond initial uses of a term . For later uses of a word or phrase, the co mmuni ty behavior to which ,

we are conforming is generally much more jointly salient than the perceptual and cognitive factors underlying the choice of the term in the first place.

2

Most of the foll owing examples of the need for nonconventional coordination in ordinary lan guage use correspond to the need for explicatures in specifying meaning in rel evan ce theory (Sperber & Wilson 1 995).

3

See Gibbs 1 994, especially chapter 2, for a critique of conventional (' l ite ral ') mean­

4

Winograd ( 1 980) makes a similar point in criticizing standard views of meaning as

ing along similar lines. fixed entities, adopted by artificial intelligence researchers (see a lso Clark & Clark

1 979:807) . 5

Clark does no t provide any attested examples of different sorts of joint construals. Examples (24)-(28) are taken from conversations collected and transc ribed by students in classes on discourse analysis at the University of Manchester.

Chapter 5

Form-function reanalys is

5.1

Introduction

In chapter 4, I presented a theory of meaning in use that allows us to locate cracks in the system of conventions to which speakers are attempting to con­ form when speaking . These cracks are the points at which innovation can occur. The most important crack is the impossibility of separating conven­ tional from nonconventional aspects of meaning in a particular context of use. In §4.3.2, I argued that one consequence of this is that the 'meaning' of a grammatical form (word or construction) is essentially the history of its uses by the members of the speech community, in other words, its line�ge. Individual members have internalized knowledge of a partial history of the form, namely its contexts of use that they have he ard or used themselves. This kn ow led ge includes crystallization into senses and subsenses and networks of polysemy relations, and the slice of encyclopedic knowledge that the meaning represents. Hence, when a speaker applies · her grammatical knowledge to a particular context of use, she must se lec t words and constructions that will conceptualize the current meaning to be conveyed in the way desired. More precisely, a speaker combines words and structures that h ave been used for a variety of prior situations in the hope that they will convey what she intends to commun­ icate to the hearer in the current situation . The hearer in turn must rely on his knowledge of the words and structures in prior combinations in prior situ­ ations, and his experience of the current situation, and thereby construct what he thinks the speaker intended to commun icate. The recombination of words and constructions clearly adds a further layer of complexity onto the process of using language in a particular context. (This complexity of course is the result of the flexibility of recombining e xisting forms-cum-meanings to efficiently express an essentially open-ended set of con­ textual meanings.) For example, in producing a sentence such as the following, the verb rob and the argument-structure construction {SBJ VERB OBJ of OsL] are combined, along with the argument phrases, the tense-aspect-mood forms, etc. : (1)

He robbed her

of

her b r acel et

.

Speakers put together the verb and the argument structure construction (choosing the argument structure in ( l ) instead of He robbed the bracelet from her). But the verb and the argument-structure construction never occur alone: the verb is always found in some argument-structure construction, and the

117

1 18

Form-function reanalysis

argument-structure c on s truct i on is always i n s ta n ti ate d with a sp ec ific verb (rob, deprive, relieve, etc.). In other w ord s the re gul a rit ies in spe a k e rs and listeners' in te rn al re p res e n tati on s o f gramm ati c al knowledge m u st be abstracted from histories of lan­ guage uses in th e i r rich communicative context. ABSTRACTION and ANALYSIS are the primary grammatical processes in language use. This is self-evident for hearers (and learners) . It is true fo r spe a ker s as well: spe a ke rs produce new utterances based on a bs t ract ion and analysis th at they have done on prev iou s utterances. We are pre se n ted with gr amm ati cal wholes and must analyze the m into their component units, sy ntac t i c and semantic, in the process of l e arni ng and (re)using la ngµ age A central aspect of this process is abstracting a set of meanings for words and constructions, given the wide range of c om mu n ica t ive contexts in which they can and do occur. The result of this process is a mapping from synt actic units onto components of meaning in the speak e r s mind (see Croft to appear b ) Nevertheless, because of the un i q ue n e ss and novelty of communicative situations, the recombination in utterances of pre e xi s ting gram­ matical units and structures will involve some degree of novel ty in the form-to­ function mapping in each use of la n guage as was argued in ch apte r 4. Moreover, a grammatical construction, that i s, a comp lex syntac t ic unit, is a fixed combination of sma l ler syntactic units. The construction as a whole is mapped onto a co m p l e x semantic structure (directly in c onst ru cti on grammar and co gn i tive grammar, indirectly via mapping rules in other syntactic theorie s) There i s of course a complex form-func t i o n mapping for the construction as a w ho le But since the construction is a fixed whole, and so is the semantic s t r ucture it d e no te s the syntactic units and their se m an t i c denotata repeatedly cooccur in the same confi gur ati on across c ontexts of use. F o r this reason, there is a certain degree of pot ent ia l i ndete rmi n acy or am b igu ity in the attribution of semantic components to s yntacti c components in an utterance. That is, there is some leeway for speakers/listeners to re an a l yze the form-meaning mapp i n g in a gramm at i c al construction . I This is FORM-FUNCTION REANALYSIS. F orm fu nc ti o n re an a lys i s is a non i n tenti onal mechanism for innovation . S pea kers i n te n de d act io ns a re t o wards co n form i ty to convention, but the re­ sult is innovation, an un inte n d e d consequence. The un i nten ded consequence of individual actions is due in part to the po tent ial d i sc re panc y between indi­ vidual competence, which is co n s tan tl y responding to use (§3 . 3 . l , §4. 3), and the c onven ti on s of th e speech commun ity Most of the time, the grammatical forms produced are e sse n ti a lly the same as those p roduced before, albeit in novel combinations, in novel meanings in context, and also with variable pronu nci ations (see the discussion of the inn ov ati on of sound change in §3.4.4). The innovations may be due to random low-level neural processes, as would be modeled by interactive ac tiv at i on n etwo rk s ; o r b y hi ghe r l evel res truct urin g of the k no wl e dge of form-function mapp i ngs in the grammar. Either way inno­ vations result fro m spe ake rs attempting to conform to convention. An illustration of how innovations m ay occur randomly in language use can be found in an e xp e ri m entall y induced type of language use: ,

'

­

.

'

.

-

,

.

.

,

-

'

.

­

-

,

Introduction

1 19

A list was compiled of 90 irregular [English] verbs, mixed in with three times that many regular verbs; thus two, three, or four regular verbs separated the i rre gular verbs . The subjects were asked to listen to the experimenter read the base form of the verb, and then to produce the past-tense form of that verb as quickly as possible . The subjects were told that the object of the experiment was to see how

fast they could go through the list; and the experimenters were instructed to put as much time pressure on the subjects as possible, since the purpose was to induce errors. (Bybee & Slobin 1 982b : 268)

The result of this experiment, as reported by Bybee & Slobin, was various sorts of regularizations and also novel vowel-stem forms, as well as of course many correct productions. Thus, speakers produced innovations. But the in­ novations were a result of speakers attempting to produce 'correct' forms conforming to linguistic norms - in (accelerated) real time. They were not the consequence of speakers trying - intending - to make past tense forms more similar to base forms, although that was in fact the result in many instances. The innovations are likely to be a stochastic outcome of spreading activation and interactive activation in the network of linguistic knowledge, in this case, the knowledge of morphological forms and their meanings (see §6.2. 1 ). Sim­ ilarly, the network of relations holding together the units of a grammatical construction and the semantic structures they conventionally denote can also be subject to interactive activation , leading to production of an altered form of the construction . Form-function reanalysis is abductive (see Andersen 1 973). That is, the representation of the form-meaning mapping in the speaker's and listener's heads is abducted (inferred or reconstructed) from prior and current experi­ ences of the form-meaning mapping found with similar utterances in similar situations. This abduction occurs in language use, however, not in language acquisition (pace Andersen 1 97 3), thereby conforming with what we know about how language change actually occurs (see §3.2 and §3 . 3 . 1 ) . The mechanism for innovation in form-function reanalysis is based on the form-meaning mapping in grammatical constructions. The form-meaning mapping clearly fits in with language use, since in communication , the form­ meaning mapping plays the central role in jointly construing the information communicated by the speaker (§4. 3 . 3). The mechanism is based on a very simple model of grammatical structure. An utterance represents a construction that is made up of its component syn­ tactic, lexical and morphological units; the grammatical units have semantic values, or at least are expected by the speaker to have semantic values; and the semantic values of grammatical units contribute somehow to the semantic value of the syntactic constructions of which they are a part. There are of course more complex formal models of grammatical structure. and different degrees of abstraction that speakers might achieve. Functionalist linguists have questioned the need for the added complexity found in formalist syntactic models and also the maximal degree of abstraction such models assume (see Giv6n 1 979; Langacker 1 987; and Croft to appear b inter a/ia). I will

1 20

Form-function reanalysis

return briefly to this issue in §5.6. The simple model described in the preceding paragraph represents a least common denominator that both functionalists and formalists r ecognize as c onstitu tin g the gr ammatica l structure which must be represented in a speaker's mind. It is certainly the minimum necessary to characterize the grammatical knowledge a speaker has/needs in order to com­ municate successfully . I wi ll argue in the rest of this ch apter that this simple model of grammatical kn owl e dge actually accounts for many types of gram­ matical changes that occur in langu a ge s. The types of innovations to be d e scrib ed as form-function reanalysis par­ tially overlap with other definitions of reanalysis. Many authors assume that reanalysis requires a precondition of alternative surface structural analyses based on the existing structures of the grammar, e.g. am biguity of case mark­ ing or alternative constituency analyses (Langacker 1 977; Hankamer 1 977; see §6. 3.2.4). Form-function reanalysis is not of this type, since remappings between gr amm atic al form an d conventional function may violate existing unambi guous conventional mappings. Harris & Campbell also argue that the precondition of formal grammatical ambiguity for reanalysis is too strong (H arri s & Campbell

1995: 7 (}-2) .

Although Harris & Campbell have a broad definition of what sorts of gram­ matical changes can be defined as reanalysis (Harris & Campbell 1 995:6 1-5), they exclude what other historical linguists would call later stages of a reanalysis, namely actualization, from their definition . ACTUALIZATION is the p roduction of a previously unattested grammatical structure based on the reanalysis of the structure of the construction (whether syntactic rJr syntactic + semantic). For Harris & Campbell, reanalysis involves only the in trod uc tion of an alternative syntactic structure for a construction, even if there is not any visible change to the construction (i.e. an innovated structure). Haspelmath argues that one cannot determine when an unactualized re analysis might have occurred, and so it is impossible to constrain the postulation of an unactualized reanalysis (Haspelmath 1 998a:74-5). In the Theory of Utterance Selection for language change , replication of a lin gue me occurs with each utterance emplo ying that li ngueme . Hence the time span from reanalysis to actualization is probably extremely short, probably to the point of being negligible for theoretical purposes. I include actualization as part of the reanalysis process, since I am concerned with the production of novel forms which occurs only with actualization. Form-function reanalysis is syntagmatic: it arises from the (re)mapping of

form-function relations of combinations of syntactic units and semantic com­ ponents. The process may nevertheless have an apparently paradigmatic resu l t, for example, a ch an ge of me ani ng of a syntactic unit (see for example §5. 4). One of the central goals, if not the central goal , of a language user's process­ ing of grammatical structures is determinin g the semantic contribution each syntactic unit makes towards the meaning of the whole in a particular context of use. The argument I have j ust presented is t h a t innovation in lan gua ge change in many cases represents a reanalysis of this mappin g , possibly as a

Hyperanalysis (overanalysis)

121

stochastic low-level p rocess i n mental representations, emerging i n language use. The next four sections will describe four types of form-function reanalysis: hyperanalysis, hypoan alysis, metanalysis and cryptanalysis.

5.2

Hyperanalysis (overanalysis)

Form-function reanalysis involves the remapping of the relationship between form and function in a grammatical construction . For any given construction made up of multiple (morpho-)syntactic units, the hearer attempts to identify what component of the content or function communicated in context is attrib­ utable to each syn tactic unit, using prior knowledge of construction and word uses and the overall meaning communicated. Of course, all of these processes are o ccurrin g in parallel: a hearer has knowl ed ge of the current context, a partial history of the construction's use, and partial histories of the uses of the words in the utterance . The hearer attempts to identify what component(s) of the meaning are conventionally denoted by the syntactic unit in question and what component(s) of the meaning are contributed by the context. The context for a syntactic unit includes other syntactic units in the construction, the con­ struction itself (to the extent that it contributes conventional meaning not associated with any sin gle unit) , and the nonconventional context. As argued in §4 . 3 , there is no s h arp line between conventional (signal) meaning and the nonconventional components of . speaker's meaning in an utterance. Hence, the meanings may be reassigned from one component to the other, or from nonconventional mean ing to conven ti onal meaning or vice versa. Jn HYPERANALYSIS, the liste ner reanalyzes an inherent semantic/functional property of a syntactic unit as a contextual property (usually, a property of another syntactic unit of the construction). Jn the reanalysis, this inherent property of a syntactic unit is then attributed to the con text (often another unit in the construction), and so the syntactic unit in question loses some of its meaning or function . Hence, hyperanalysis is a major source of semantic bleaching and/or loss in ge neral . Hyperanalysis is analogous to Ohala's notion 2 of hypercorrection in his listener-based model of sound change ( see §3.4.4). In §§5.2. 1 -5.2.2, J give two examples of plausible contexts for syntactic hyperanalysis. 5.2. 1

Loss of governed oblique case

In many languages, the argument structure of certain verbs includes a governed oblique case or adpositional phrase. That is, the semantics of the verb requires an argument which is exp ressed as a (usually obligatory) oblique NP or PP. A good example is Russian, where verbs may govern 'objects' in the genitive, dative or instrumental case. Examples of verbs governing the genitive are given in (2) (Pulkina n.d.: 82-7, 89-99):

Form-function reanalysis

1 22 (2)

Genitive trebovat' prosit ' iskat ' idat' s/u§at 'sja stoit '

lisat 'sja

'demand s.t. ' 'ask s.t . ' 'seek, look fo r s.t. ' 'wish (for) s.t.' 'listen to s. t . ' 'be worthy of s.t.' 'lose s.t.'

'be afraid of s. t . ' 'be frightened of s.t.? 'fear s.t.' 'be shy of s.t.' 'avoid/shun s. t.' 'deprive of s.t.' 'escape , avoid s . t . '

bojat 'sja pugat'sja opasat 'sja stesnjat 'sja storonit 'sja lisat'

izbegat '

English also has a number of governed PP arguments, three of which are illustrated in (3): (3)

a. b. c.

I was looking at the redwoods. She was listening to the quartet. We were thinking about the election .

To some extent, there is semantic motivation for the choice of oblique case with particular verbs or semantic subclasses of verbs. For instance, the Russian

genitive is used with arguments in which there is lack of physical contact or lack of complete affectedness of the genitive NP's referent, particularly as found in the oblique objects of verbs of seeking/requesting and the stimulus argument of emotion verbs. These patterns presumably reflect the semantic contribution of the genitive case, as extensions of its use in other contexts. However, in many languages including Russian, governed oblique cases are highly conventionalized. There' are instances in which the governed case does not fit the semantic motivations. The verbs deriat 'sja and prideriat 'sja 'keep, hold (onto) s.t. ' and kasat 'sja 'touch s.t.' involve contact in their physical meanings yet govern the genitive. The verb dobivat 'sja 'seek s.t. ' can also mean 'obtain, achieve s.t. ' , but governs the genitive in the latter senses as well although they represent successful completion of the act. These facts blur the link between syntactic unit (the genitive case) and its semantic contribution to the meaning of the whole . Also, the inherent meaning of the verb also specifies the semantic role of the governed oblique argument. That is, there is overlap between the semantic contribution of verb and oblique case. These two conditions, semantic irregularity and semantic overlap, provide the conditions for hyperanalysis. The inherent semantic value of adposition/ case can be reanalyzed as belonging sol ely to the verb meaning. The adposition/ case is devo i d of meaning, or rather is semantically unnecessary, and eventually is left out (actualization of the reanalysis). This appears to be a relatively com­ mon phenomenon, and can be observed in Slavic and in Germanic languages. The genitive object in Russian is governed by particular verbs and also by particular constructions, such as the negative and the partitive (Timberlake 1 977: 1 57-8). Jakobson ( 1 936/ 1 984) argued for a general semantic characteriza­ tion of the genitive/accusative distinction in terms of a limi tation on the extent to which an object participates in an event' (Timberlake 1 977: 1 58). Although this semantic analysis is rather vague, it probably reflects the semantic distinction '

Hyperanalysis ( overana/ysis)

1 23

between accusative and genitive in these contexts. Tim b er l ake notes that the gen itiv e is be in g replaced by the accusative in Russian, particularly in Jess formal regi sters This is a hyperanalysi s of the o b l ique case. Timberlake also argues that this shift is occurring gradually, and the path of shift can be ac­ counted for by the degree of nonlimitation of the object, e.g. higher individuation (Timberlake 1 977 : 1 62-8). I would ar gue that the gradual shift to accusative marking reflects a gradual attenuation of the semantics of the governed genitive case by hyperanalysis. Verbs govern nonaccusative obj ects in Germanic as well. Comparative indi­ rect evidence suggests that nonaccusative objects are gradually being hyper­ analyzed, and there are limited examples of shift to accusative case in English s hortly before the breakdown of the case system. In languages such as German and Icelandic, the case system is still in place and many verbs still govern obl iq ue cases. However, there is indirect evidence for object status from syntactic b ehavi or associated with objects, in particular passivization. The evidence is somewhat obscured due to the existence of .

impersonal passives in both languages, where the active subject is deleted or encoded as an ob li q ue , but the active o bj ect I governed oblique is not encoded as subj ect (Zaenen, Mating & Thrainsson 1 98 5 :476):

(4)

Ihm wu rde geholfen. he.DAT was help. PASS-FTC

(5)

Honum var hjal p a� he.OAT was help .PASS-FTC . .

Zaenen, Maling & Thrainsson ( 1 98 5 :476-8), following Cole et al. ( 1 980:7278), argue that the German construction in (4) is not a true personal passive, because the p assi ve subject' Ihm fails various behavioral tests for subjecthood, such as EQUI co nt ro l , relative clause r eduction and c o njuncti o n reduction. Thus it is an oblique, or at least not a subject. On the other hand, the Icelandic construction in (5) is a personal p a s siv e, albeit with a so-called q u i rk y case p a s s i ve subject, since it p as ses similar behavioral tests for subjecthood. Thus it is a subject, and following the standard formulation of the passive rule, it should correspond to an active object. Hence there is evidence that the Icelandic object of he lp and other governed o b l iq ues has been partly hyperanalyzed as a direct object. Zaenen, Maling & Thrainsson hint at an exp lan ation of this difference between German and Icelandic in a footnote (ib id 47 1 , fn . 22). Ge rman has a rich case system and relatively free word order, so word order does not inheren t ly code for subject status. Icelandic retains the rich case system but has rigid word order. Hence word order can encode subjecthood, and in fact hyperanalysis of the grammatical relation must have taken place, allowing for the q ui rky dative NP to pass the behavioral tests for subjecth ood in this pas­ sive construction. In Modern En gl ish the o bj ect of help i s encoded as a direct object, and the passive formed from it is a personal passive (English lacks impe rson al passive s ) '

'

'

.

,

.

1 24

Form-function reanalysis

However, the reason for this is as much the loss of the case system as hyper­ analysis. Evidence for hyperanalysis would be a recoding of nonaccusative objects as accusative objects before the loss of the case system in Early Middle English. There is some evidence of this, in that governed genitives are occasionally found in the accusative in Old English before the loss of the case system: 'although there are not a very large number ofexamples with accusative objects, they are nevertheless too frequent to be mistakes' (Allen 1 995: 1 35; examples are from Allen 1 995 : 1 33, 1 3 5):

(6)

Mice! lrecedomes wund behofaO micles great.NOM would.NOM needs great.GEN leechcraft.GEN 'A great wound requires �reat medicine' ( [COE] Bede 4 26.350. 1 9)

(7)

. . . swa heo maran lrecedom behofaO . . . so it greater leechcraft.ACC needs ' . . . so it requires greater medicine' ( [COE] JECHom I, 33 496. 30)

Allen remarks that 'it seems rather surprising that genitive objects should have disappeared so early on, because the genitive was the most clearly marked of the cases in M[iddle]E[nglish]' (ibid. 2 1 7) . A hyperanalysis account renders this fact less surprising. 3 English also had passives that look like those of Germanic or Icelandic. Genitive objects rarely occurred in passives, leaving mostly examples with dative objects. These constructions were replaced by passives with nominative objects. Interestingly, the passives with �ominative objects arose even when the dative/ accusative distinction was still present, in at least some dialects (Allen 1 995: 35764) . Allen treats this as problematic, because this change would complicate the grammars of those speakers (ibid. 375). However, in a hyperanalysis account, it is a plausible remapping of the form-function relationship in these construc­ tions; and grammar simplification is not an aim of language users (see §3.4. l ).4 5.2.2

Evolution of impersonals from locative agreement

second example of hyperanalysis is in the evolution of impersonal construc­ tions from locative agreement in the southern Bantu languages (Croft 1 995b) . Most Bantu languages have a rich system of agreement with noun classes, so that for instance in the Chichewa example in (8), the verb Ii 'be' agrees with chitsime 'well' (Bresnan & Kanerva 1 989:3):

A

(8)

·

Chi- tsime chili ku- mu- dzi CL7- well CL7. SBJ- be CL 1 7- CL3- village 'The well is in the village.'

In addition to the inherent noun classes, there are a set of noun class markers that originated as prepositions, traditionally labelled Classes 1 6- 1 8 ; cf. kumudzi 'in the village' in (8). Not only may the locative class markers be prefixed to nouns, yielding a locative phrase, but if the locative phrase is fronted, the verb will agree with it (ibid.):

Hyperanalysis ( overanalysis) (9)

1 25

m u - dzi kuIi chi- ts ime CL3- villag e CL } 7.SBJ- be CL7- well 'In the vi l l a ge is a wel l . ' Ku-

CL } 7-

This is a ge n u i ne case of a g r ee m en t , not some sort of impersonal exp ress io n . The locative prefix on t he verb must m a tch the l oca tiv e prefix on the l oc ative phr ase (see ( 1 0); ibid . 1 2), and a different class p re fix is used for 'true' im per ­ sonals (see ( 1 1 ) ; ibid. 1 0):

( 1 0)

Ku-

m u - dzi mukuganiz -a kuti k u -/ * p a vi l l age 2 . PL . SBJ- PROO- th i n k -IND COMP CLI 7 . SBJ-/*cL 1 6.sBJ-

CL l 7- CL3-

bwer -a al e n do come -IND CL2- visitor Lit. : 'To the vi ll age , you think that there came visitors. '

na-

REC. PST-

(1 1)

Zikutsibweni a - nga ndi g an izi r -idw - a kuti a CL l O.SBJ- PROG- think -PASS - IND COMP cL2- u n cle CL2- my COP a- fiti CL2- witch ' It is th ough t that my uncle is a practitioner of witchcraft.'

While it is true that locative agreement in C h ic hewa is genuine agreement, othe r southern Bantu lan guage s beh ave diffe re ntl y , and one can i den ti fy a gradual pr oces s by whi c h one l oc a ti ve agreement prefix is re anal yze d as an imperson al . The detai ls of this p r ocess are de scrib ed in Cro ft l 995b; for present purposes, the relevant process i s w hat licensed the first step from locative to impersonal. In Chichewa, as in a num be r of other Bantu languages, a s ubj ect p h rase can be clause-final a s well as clause-initial . This is true of locative subjects as well (B re sn an & Kanerva 1 989: 3):

( 1 2)

Mwakhal -a any a n i mmi- tengo CL 1 8 .SBJ- PERF- sit -IND CL2- baboon CL 1 8- CL4- tre e ' In

the trees are sitting baboons.'

The effect of this option, however, is to place the locative phr ase in what appears to be the same p o si ti o n as a normal n o n su bj ect lo cati ve phrase (compare e xample ( 1 0) ) . This allows hyperanalysis to take place: the locative mea ni ng is attributed the oblique function of its supe rfi ci al po si ti on , the theme NP is in a n o nsubject po s i tion , and the locative agreemen t on the verb now has no func tion , having been hyperanalyzed away . 5 This hyperana lysis has led to a com p l eted chan ge in Zulu. In addition to an inva rian t Class 1 7 locativ e verb form for e xi ste nt ial contructions ( 1 3; D oke 1 930:296), the s ame Class 1 7 form is used to resolved ge n de r conflicts in conj o ined su bj ect NPs wh ich have n o t h i ng to do with locative meaning ( 1 4; Doke 1 930:298): s u pe rfi cia lly

( 1 3)

khona izi nja emazwen i on ke be.present dogs countries.we all 'The re are dogs in all countries. '

ku-

CL l 7-

1 26

( 1 4)

Form-function reanalysis ku- yokubhubha aha- ntu nezinkomo CL 1 7- wil l perish CL2- person and:CL 1 0- cattle 'The people and cattle will perish. ' ,

5.2.3

.

Constraints on hyperanalysis

A de sir abl e feature of any theory of innovation in language change is the postulation of constraints on the occurrence of the innovation. This is espe­ ci ally critical here, since as will be seen in the fo llowin g sections, some of the me ch a ni sm s for form-function reanalysis go in oppo si te directions. Fortunately, the nature of form-function reanalysis permits us to construct a hypothesis as to what contexts would lead to hyperanalysis. In other words, a prediction can be made as to the contexts in which semantic b leachin g and loss shoul d occu r In hyperanalysis, the conventional meani n g or function of a less contentful syntac tk" µ n it is reanalyzed as belonging to a more contentful syntactic unit with overlapp ing con ven tional meaning but not vice versa. For example, the meani n g associated with governed oblique case marking is h yp e ran al yzed be­ cause of the overlapping but more contentful verb m e ani n g Sem a nt ic b le ach ing or gr amm ati cal loss in other situations, if they exist, must be attributable to other mechanisms for innovation . For example, grammaticalization., w hi ch gen­ erally does not involve this ki1'1d of semantic overlap , has sometimes been described as loss of mea ning (e.g. Heine & Reh 1 984: 1 5) . H oweve r it is better described as a CHANGE of me an in g rather than a loss of meaning (see §6. 3.2 and Trau gott 1 98 8 , 1 989), and so · does not constitute a counterexample to the proposed constraint o n h yperana l ys i s This hypothesis of a constraint on hyperanalysis can be tested in three ways. Fi rs t , other examples of language change of this type may be examined to see if they satisfy the conditions j ust offered . Indeed, the constraint proposed here is inferred from occurring cases of language change . Second, a large sample of innovations in l an guage use should demonstrate a prep o nderan ce of innova­ ti o n s of this type, where hyperanalysis is a plausible explanation. F inall y a psycholinguistic e xpe ri ment can be constructed, alon g the lines of the past tense experiment of Bybee & Slobin described in §5. l . In such an experiment, it is p redicted that innovations due to h yperan a lysis should occur significantly more frequently in the c o n te xts which favor it than in the contexts which do not. It is beyond the scope of this book to construct the second and third tests. In principle, however, we should be able to test theories for innovation both syn chro nically a n d experi men t al ly .

.

­

,

.

,

.

5.3

Hypoanalysis (underanalysls)

In HYPOANALYSIS, the listener re analyzes a contextual semantic/functional p rop erty as an inherent property of the syntactic unit. In the rean alys is the inherent property of the context (often the grammatic al context, but see §5.6) is then ­

,

Hypoanalysis (underana/ysis)

1 27

attributed to the syntactic unit, and so the syn tact ic unit in question gain s a new meanin g or fu nction Hypoanalysis is an alogous to O h a la s notion of h ypocorrect io n Hypoanalysis is the source of a process that has recen tl y gained attention in h i sto ric al linguistics. The process is cal led EXAPTATION by Lass (Lass 1 990: 989) an d REGRAMMATICALIZATION by Greenberg (Greenberg 1 9 9 1 : 303 ). Greenberg describes the process thus: .

'

.

increasing extension [of a grammaticalized element] leads to zero intension, so that the item has become desemanticized. There is, however, with expansive lexicalization a further possibility, namely its reinterpretation in a new function. We may call this process regrammaticalization. (Greenberg 1 99 1 :30 1 )

I n the rest o f this section, I gi ve four ex am ples of p la u sib le contexts for syn tac tic h ypo a nalysi s ( see §5 . 6 for another example) . 5.3. 1

Subjunctives and futures from present Indicatives

Bybee, Perkins & Pagliuca discuss a case in which a verb form with subj unctive meaning has evo lved from one with present indicative mean i ng (Bybee, Perkins & Pagliuca 1 994:230-6). In Armenian, the progres si ve verb form h as spread fr om present ongoing action contexts to ha b itual con te x ts thereby ousting the simple in dic ati v e verb form in main clauses. The simple indicative form has survived in subordinate clauses. Many subordinate clause contexts are sub­ junctive, that is, the meaning of the construction implies a su bj un ct i ve me a nin g for the subordinate clause . In Mo dem Armenian, the simple indicative has been re analyzed as a subjunctive and can be fo un d as a su bj u nct ive in main clause contexts (ibid. 232, from F ai rb ank s & Stevi ck 1 958: 1 1 8, 1 50): ,

( 1 5)

p'ayman6v vor usadrutyamb varek mekenen condition th at carefully drive 2so car 'On condition that you drive the car carefuJJy. ' .

( 1 6)

me lezvov g;irem h a scen what l an guage write. I sa address 'In what language s hould I write the address?'

This ch an ge appears to have begun as a hypoanalysis. I ns tead of at tri b u ti ng to the sentential context, speakers have reanalyzed the su bjunctive value as inherent to th e simp le indicative form, which o ccurs only in that context. Haspelmath describes several other cases of subjunctives arising from old present indicatives where a new present indicative has sprea d to al l main clause present functions (Tsakonian Gree k, Modern ln dic, Persian and Cairene Arabic; Haspelmath l 998c:4 l -5). Haspelmath also describes a number of cases of futures a rising from old present indicatives, al so th rou gh what I would c all hypoan alysis (Welsh Hebrew, Lezgian, Tu rkis h, Udmurt and Kannada; ibid. subjunctive functi o n

,

1 28

Form-function reanalysis

36-41). Again, a progressive is extended from ongoing action contexts to habitual contexts, leaving the future context for the old present (the original present is more like a nonpast tense form). What was originally a contextual property of the present or nonpast in future contexts has been hypoanalyzed as the function of the former present/nonpast. For example, the Hebrew imper­ fect form, accompanied by person agreement prefixes, is a future in Modem Hebrew ( ?e-xtov 'I 'll write') but was used for present time reference in Biblical Hebrew (ibid. 37). H aspelmath argues that the evolution of old presents to future and/or subj unctive meaning is not a case of grammaticalization. The hypoanalysis account agrees with that claim; semantic changes in

grammaticalization

are

brought about by metanalysis instead (see §5 .4.4, §6. 3 . 2). Haspelmath further argues that telic present tense verbs are the first to be restricted to future meaning, because a telic event is more likely to be interpreted with future time reference in a present tense than atelic or stative events (cf. the English future use of the present as in The train for London leaves at 9:30, Haspelmath 1 998c:49-5 1 ) . This argument fits in with the hypoanalysis account: one would expect speakers to attribute the contextual meaning to the form in question first with verb classes where the contextual meaning is almost always associated with use of that form. 5.3.2

German umlaut

German umlaut is a classic case of exaptation (Lass 1 990:98-9) or hypoanalysis, in that it began as a phonological process and then was hypoanalyzed to gain a plural function (among others) . A history of the umlaut or i-mutation is found in Keller 1 978. The late Germanic i-mutation fronted back vowels (Keller 1 978:80-1 ). The i-mutation is phonemicized by the Carolingian period of Old High German (ibid. 1 59-64; thi s is an example of Ohala's hypocorrection, incidentally), and thus associated with i-plurals (ibid. 1 80). By the Middle High German, the i-mutation is morphologized (ibid. 279). In the Early New High German period one observes the 'analogical extension' of umlaut (ibid. 4 1 2, 560). It is this last stage that involves hypoanalysis. The semantically never­ before-meaningful umlaut is attributed the contextual property of plurality and reanalyzed as a marker of plural inflection; the 'analogical extension ' is the actualization of the hypoanalysis. Lass gives examples such as Modem German Baum/Biiume 'tree', as a novel umlaut; cf. OHG boumiboum-e (Lass 1 990:99). Another example that Lass gives is Yiddish shvits-shop, which comes from English sweat-shop and hence is not descended from OHG, and has the plural shvits-shep-er (ibid.). 5.3.3

Nominalizers from Stage Ill articles

Another sort of hypoanalysis is Greenberg's regrammaticalization. Greenberg has argued that a grammaticalization process leads to the spread of etymological

Hypoanalysis (underanalysis)

1 29

demonstratives to defin ite NPs (h is S ta ge I articles), then to specific in defi nite NPs as well (Stage II article s) , then to nonspecific NPs (Stage III article s ; Green be rg 1 978/ 1 990) . By S tage III, t he articles are found on virtually every occurring noun or NP. Hence it is s imp ly associated with nouns. An example of a S tage III article is the so-called 'movable k-' of Nilo-Saharan (Greenberg 1 98 1/ 1 990). In some Nilo-Saharan languages, such as Ngambay Mundu, 'ver­ bal roots be ginn in g with vowels form verbal n o un s meani ng " act of . . . " by p refixin g k-, as so often in Nilo - Sah a ran An examp l e is usa "eat", k-usa "act of e a ting (Green be rg 1 9 8 1 / 1 990:479, citing Vandame 1 963:66, 75). I sugges t that this occurred due to hypo an alysis The k- p refix is attributed the contextual property of nominality that is inherent to the nouns it is invariably found wit h , and then comes to be used as a nomin al izer .

.

.

" '

.

.

5.3.4

Multiple exaptation/hypoanalysis: do and English 3rd singular present

-s

Lass observes the minimal remnant of Engli sh subject agreement found in the Standard English 3rd person singular present tense suffix -s, and writes , 'Now however this relic inflection [-s] not only has no "communicative" function . . . it is a sy stemic excrescence . . . But the ex ap tive impul se is strong (Lass 1 990:99). In fact, the e x apta ti o n/hypoanaly si s has a lre ady occurred, in combin ati on with the hypo an a lysis o f periphr astic do, in the traditional d i alects o f Somer se t and Do rset (l h a l aine n 1 99 1 ; Trudgill 1 990). Commonly in earlier Standard En gl ish (and las tin g into the 1 8th ce n tury) ' [periphrastic] do is simply used as an un stressed tense marker' (lhalainen 1 99 1 : 1 48). It may be that hypo analys i s has led to its emphatic function in contemporary Standard E n gl i sh presumably due to the attribution of a feature of the discourse context to the perip h rastic do . In Somerset/Dorset trad i ti on al dialects, however, we find a combined rea n a lys i s so that t he s imple present -s is used for s pecific or single events, while do is used for iterative or h ab itu al events (Trudgill 1 990:95; Ihalainen 1 99 1 ): '

,

,

,

( 1 7)

a. b.

I sees the doctor tomorrow [specific/sin gle event, present] I do see h im every day [iterative/habitual, presen t]

It appears that the semantically min imal -s acquires the contex tual semantic property of m ark in g present tense; this hypoanalysis leads to its u se beyon d 3rd person sin gular (as in ( 1 7a) ) . In addition, the seman t ica lly minimal do is attributed contextual p rope rties of verbal aspect, and is reanalyzed as an iterative/hab itu al marker. As a conse q uence present tense -s is a ttribute d a semelfactive/nonhabitual semantic function. Again, we do not really know the u sage contexts which mi ght have led to thi s p articular division of semantic labor between -s and do. But the use of do in 1 8 t h cen tury E n gl i sh plus the marginal gramma tical status of -s probably gave rise to just the sort of com­ ple xi ty in the form-function mappin g in language use th at would invite hypoan alysis of these forms. ,

-

1 30

Form-function reanalysis

5.3.5

Constraints on hypoanalysis

Hypoanalysis is the opposite of hyperanalysis. If both processes operate unconstrainedly, then innovation can lead anywhere . However, the eitamples of exaptation and regrammaticalization that have been identified suggest that hypoanalysis can be constrained, just as I proposed that hyperanalysis can be constrained. In hypoanalysis, it is hypothesized that a conventional meaning/function of context is attributed to a syntactic unit by virtue of the syntactic unit's restricted distribution or apparent redundancy with an ov er l apping more con­ tentful expression . The crucial aspect of this constraint is the distribution of the syntactic unit across constructions. The syntactic unit simply does not have content in other constructions. Either it doesn't occur in other constructions, as with the Armenian simple present indicative; or it occurs in all contexts, and hence has no discriminatory semantic value; as in the Stage III article (noun marker) and other examples given above. 5.4

Metanalysls

METANALYSIS

is the simultaneous occurrence of hyperanalysis and hypoanalysis. In metanalysis, the listener swaps contextual and inherent semantic values of a syntactic unit. 6 It appears that these two events occur simultaneously; that is, that there is no stage in w�ich hypoanalysis has occurred but not yet hyperanalysis, or vice versa. For this reason, a distinct type of form-function reanalysis is proposed. Metanalysis does not seem to be analogous to any type of phonological change in Ohala's model. Ohala does not describe such a process, and I cannot think of _any clear examples of phonological analogs to metanalysis. Metanalysis appears to be the mechanism for innovation of the sort of grammatical changes that have gone under the name of invited or pragmatic inference (Hopper & Traugott 1 993, chapter 4; see §5.4.3 and §6. 3.2). 5.4. 1

Reinforcement > replacement: the negative cycle

One particular type of reinforcement appears to be caused by metanalysis. It can be illustrated by the well-known negative cycle (Jespersen 1 9 1 7) . In the negative cycle, an emphatic element comes to be used with the negative marker and ends up usurping its negative function; in the final stage of the process, the original negative marker disappears. Jespersen illustrates the negative cycle with its pattern in the history of French (Jespersen 1 9 1 7 :7-8). In Classical Latin, there arose a negative marker non, as in non dico. In Old French, the parallel construction was syntactically the same (plus the subject pronoun): jeo ne di. However, at that point a variety of nouns came to be used with emphatic function associated with the negative construction, such as poin t , pas, personne, etc . These forms, which originally had a positive semantic value, came to cooccur

Metanalysis

l3l

obligatorily with ne, and hence became nonemphatic elements. This function came to be restricted to pas, as in Standard French je ne dis pas. The process by which the emphatic forms became obligatory negative markers is caused by metanalysis. Before metanalysis, the negative meaning is attributed to the preverbal pa rt i cl e ne, and the emphatic function is attributed to the postverbal element (pas, personne, etc.). From the point of v iew of the postverbal element, negation is a contextual feature while emphasis is an inherent feature. However, as has been argued by a n u m ber of linguists (e.g. Schwegler 1 988 :36; Giv6n 1 979, eh. 3), negative utterances are more likely to be emphatic in actual use than positive ones; that is, there is a high degree of correlation between negation and emphasis. This correlation sets the condition for metanalysis: since the emphatic element is found frequently in negative contexts, and negat­ ive contexts are frequently emphatic, there is a swapping of the two functions: the negative functi on is attributed to the emp h atic element, while the emphatic function is attributed to the nonlinguistic context. 7 The actualization of the metanalysis is the use of the formerly emphatic element in nonemphatic negat­ ive contexts. The final stage in the negative cycle, manifested in colloquial French je dis pas, is caused by hyperanalysis. The inherent negative value of the preverbal particle is attributed to the postverbal element alone, since its (newly acquired) negative function overlaps with that of the preverbal p article ; and the preverbal particle is dropped. It is worth pointing out here that given the preceding stage of the process - two syntactic units mapping to the same semantic function the theory does not predict which of the two units will undergo hyperanalysis. Nor should it, most likely. If in fact the etymologically emphatic marker un­ derwent hyperanalysis and disappeared, then the result would not have been a change in this aspect of grammatical structure. Abortive changes certainly take place; we only know of them in cases where direct historical records are available (an example, the aborted shift of seem to an experiencer subject verb, is descri bed by Harris & Campbell 1995:88-9) . Finally, we may note that French provides examples of metanalysis outside the context of the subsequent hyperanalysis of pas as the sole (colloquial) French negator. Two of the other emphatic elements that formerly occurred as negative emphatics survive as negative pronouns: rien 'nothing' < ''" th i ng ' , personne 'no one ' < * ' pe rson ' . Having been metan a lyzed by speakers, these forms were no longer restricted to use as emphatic markers; having acquired ne gati ve meaning, they came to be used in o ther n e gative contexts. 5.4.2

Passives from 3rd person actives and reflexives

Another example of metanalysis is fo un d in the development of pa ssiv e s from 3rd person active forms. This is one of several common sources for passive constructions ( Haspe lm a th 1 990) . An ex am p le of this development can be seen in Maasai (Greenberg 1 959/ 1 990). In Maasai, there are subject and object agreement p refi xes , illustrated in ( 1 8) and ( 1 9) ( Green be r g 1 959/ 1 990:4 1 3) :

1 32

Form-function reanalysis

( 1 8)

d:.)l al SG.SBJ- see 'I see him/her/them.'

( 1 9)

aad:.)l l SG.OBJ- see 'He/she/they see me.'

The passive subject uses the object agreement forms but with a suffixed -i that indicates that the verb is in the passive voice (ibid.): (20)

aad:.)l -i l so.oBJ- see -PASS 'I am seen. '

The suffix - i i s a former 3rd person plural subject marker, so that etymologically, the passive in (20) is 'They see me' (ibid. 4 1 5). This develop­ ment can be interpreted as a case of metanalysis. As many linguists have argued (e.g. Giv6n 1 98 1 ; Shibatani 1 985), the function of the passive includes backgrounding of the agent, including elimination of the agent's specific identity. Other constructions than the passive have this function as well. For example, 3rd person plural subjects of active transitives are often interpreted imperson­ ally: in English I can say I called the ticket office and they told me that Friday 's performance was sold out, even if it was only one person who told me. In Maasai and other languages, the 3rd person subject affix was metanalyzed as a passive affix. The passive contextual function of 3rd person plural impersonal sentences was attributed to the 3rd person plural subject suffix, and the inde­ finite agent that the suffix originally denoted was attributed to the context. It is possible that passives derived from reflexives via middles are also caused by metanalysis. In a reflexive such as Harriet washed herself or Theodore killed himself, the reflexive morpheme signals that the subject and the object are the same referent, which means that the semantic role of the referent is both agent­ like and patient-like (or experiencer-like and stimulus-like, depending on the verb). In a number of languages, the reflexive construction develops passive functions, as in Spanish Se cerraron /as puertas a /as diez 'The doors were closed at ten' (Schevill 1 970:66). Croft, Shyldkrot & Kemmer 1 987 argue that the extension of reflexive con­ structions to passives follows a stage when the reflexive construction is used for a variety of intransitive counterparts of transitive verbs where the intransitive subject is an undergoer. Of course, only a subset of transitive events are such that one can perform them on oneself: one cannot build oneself, sing oneself, etc. The reflexive is first found with events typically undertaken on oneself (unmarked middles), such as vestirse 'dress (oneself)' and banarse bathe (one­ self)' - note the optionality of the reflexive in English, suggesting a construal of the event as intransitive. The reflexive middle is found both in verbs where the intransitive subject is under its own control (so-called unergative) Spanish '

-

Metanalysis

1 33

arrodillarse 'kneel' - and in verbs where the subject appears to undergo the action spontaneously (so-called unaccusative) quebrarse 'break' . In both cases, the action could be carried out by an external force, and in fact may be implied even for the intransitive : saying The window broke does not commit me to a belief that the action occurred spontaneously. Finally, the construction Se venden libros aqui 'books are sold here' also implies an external agent and it is a short step from this construction to an agentive pas sive At some point in this process, certainly by the point that the reflexive was used for unaccusatives such as quebrarse and for passives, the reflexive mor­ pheme no longer denotes a participant coreferent with the subject� The for­ merly reflexive morpheme came to denote a verbal semantic property, something like 'subject both controlling and affected by the event' and eventually 'subject affected by the event, whether controlling or not'. The semantic shift can be analyzed as metanal ysis. The contextual property associated witn the reflexive is a class of events where acting on oneself is possible, and even typical (the unmarked middles). In metanalysis, the coreferential subject denotation of the reflexive morpheme is swapped for the contextual property of affected-subject verbal semantic type. In fact, the coreferential-subject property, now merely a contextual factor, is weakened and removed by the time the former reflexive is extended to unaccusatives and pas sives . -

,

.

-

5.4.3

Pragmatic inference and metanalysls

Metanalysis is basically an account of the innovation of invited inferences. PRAGMATIC INFERENCE represents a large class of semantic changes where some

nonlinguistic contextual factor comes to be part of the meaning of the unit in question, and (a fact not emphasized in the literature) the former meaning of the unit is lost. A typical example is the connective since in the history of English (Hopper & T r augott 1 993:74). Since began as a temporal connective meaning 'after' . In actual use, speakers frequently einploy an 'after' clause when a causal as well as a temporal semantic relation holds between the main clause and the adverbial clause. Since underwent metanalysis: it is now largely used as a causal connective, although it retains some temporal uses (She has been out of the country since February). A nother typical example of pragmatic inference is the evolution of aspect markers into tense markers (see for example Bybee & Dahl 1 989). Pe rfectiv e aspect generally occurs in use with past time reference: it can be used for future time reference, but it is semantically incompatible with present time reference. Present tense usage is thus necessarily imperfective, though imperfec tive is compatible with past and future tenses. In many languages, the aspect markers have become tense markers, a change caused by metanalysis. The contextual past ten se value of the perfective aspect becomes the inherent value · of the marker, which loses its aspectual value. Similarly, the imperfective is metanalyzed to a present tense marker. 8

1 34

Form-function reanalysis

5.4.4

Constraints on metanalysls

It should be clear from the examples that metanalysis is predicted to be possible when there is a high correlation between the relevant inherent and contextual meanings in language use. Under those conditions, the inherent and contextual meanings may be reversed. This is not a sufficient constraint to describe the facts of semantic change for which we are invoking metanalysis. The changes illustrated in this section are unidirectional. Is it possible to cons train metanalysis so that the chan ge s are uni d irectional? In some cases it can pl ausi b ly be argu ed that the correlation is unid i recti o n al , and hence the change will be unidirectional. For example, negative utterances usually have emphatic force, b ut emphatic sentences do not usually have negat­ ive force. Hence the emphatic force can be metanalyzed in negative utterances. Likewise, present tense is always imperfectiv e ; but impe rfe ctiv e aspect can be . fo und with all tenses. Hence the im perfective meaning can be metanalyzed in p resent tense utterances. Passive utterances gene r ally have 3rd person agents and affected subjects, but not all utterances with 3rd person agents or affected subjects are passive in discourse function ( agent backgrounding, p atient to p ic) . Hence the 3rd person o r reflexive affix can b e metanalyzed in utterances with a passive discourse function. Finally, in the evolution of future tense markers from modals, future time reference is almost always irrea1is in modality, but irrealis modality does not always imply future time reference. Hence certain irrealis modality markers can be metanalyzed as futu re tense markers. Un fo rtu n ately, not all examples of unidirectionality in (p utative) examples of metanalysis represent a unidirectional correlation. For example, the unidir­ ectional change perfective aspect > past tense would imply that past tense correlates with· p e r fective aspect more than perfective aspect correlates with past tense; but it is not obvious that this is true . If the contextual correlations in these exam pl es are bi di rect i onal , then metanalysis does not prevent the unattested change from past tense to perfective aspect. If so, an indepe n den t theory of the unidirectionality of semantic change will be necessary to account for these changes. Nevertheless, it may be that metanalysis will account for the unidirectionality of most if not all of the semantic changes found i n grammat ical izat i o n (see §6. 3 .2).

5.5

Cryptanalysis

is the result of re analysis of a more subtle relationship between syntactic units and their seman tic function. In cryptan alysis, the listener analyzes a covert semantic/functional property of a syntactic unit as not grammatically marked, and inserts an overt marker expres sing its seman ti c value. Cryptanalysis, like me tan alysis , appears to have no analog in Ohala's theory of phonological change. Cryptanalysis is the source of man y cases of pleonasm and reinforce­ me n t , several of which are illustrated in the fo l lowin g subsections.

CRYPTANALYSIS

Cryptanalysis 5.5. 1

1 35

Pleonastic negation

A gramm at ical phe n omenon that has att racte d attention in both the standard literature and electronic discussions is ple on astic n e gati on . A discussi on on the on an innovation based on the cliche phrase an d re l ate d p hr ases (see Hom 1 97 8 : 1 76; Lawler 1 974: 3 5 8 , 3 72). The (a) a n d (b) s en te n ces i n (2 1 )-(22) mean the same thing, a l th o ugh the (b) sentences contain a negative th at the (a) se n tence s l ack : LINGU I ST list in 1 99 3 focused

That 'll teach

(2 1 )

you

That'll te ach you to come early. That'll te ach y o u not t o come e arl y a,b You came early, some unhappy con sequence ensued, and you should not come e a rly in the future. (Laurie Bauer, LINGUIST 4.873) a.

b.

.

=

(22)

I really miss ha vin g a phonologist around the h ou se I re al l y miss not having a phonologist around the ho use a,b = There used t o be a phonologist around t he h o u se there isn't any more, and I wish there were o n e around the ho u se n o w . (Dale R usse ll LINGUIST 4. 859)

a.

.

b.

.

,

,

Horn made the following remarks on this class of exam pl es : Now the negation that shows up pleonastically i n miss (not) VPing, surprised if it does(n 't) . . . , prendre garde de (ne pas) tomber, etc . , i s NOT, unlike the one in so don 't I, ironic or sarcastic; [it is] attributable to the difficulty of processing multiple negations ESPECIALLY WHEN AT LEAST ONE IS NON·OVERT. (Horn on .

LINGUIST

4.898;

my

.

.

emphasis)

The (a) sentences in (2 1 }-(22) h ave in common a covert n ega tive as Horn notes. Now, p ol arity is s ome thi ng that is normally expresse d ove rt l y and gram­ matically in other sen tences in Engl i sh (an d other l an gu age s ) In (2 1 a) and (22a), a li s te n e r recognizes the n e gati ve semantic component but the re is no overt syntactic unit to wh ich the neg a t ive semantic value can be li n ked ; instead it is an en tailme nt of some other syn t acti c unit. What has ha ppe ned in (2 l b ) and (22b) is t h at the hearer has reanalyzed the construction as ' needing' an overt syntactic unit to convey the semantic value, and actualized. the reanalysis by in se r ti n g a pleonastic ne gat ive marker. Th i s is cryptanalysis . ,

.

5.5.2

Paratactlc nega tion

A very similar sort of cryptanalysis has occurred in the com p lements of ce rtai n negative-entailment verbs, s pecific ally the translation equivalents of 'deny' , 'forbid' , 'hinder' , d oubt etc. (Jespersen 1 9 1 7 : 7 5 , cited in Hom 1 978 : 1 72) . In Engl i sh the complements of such verbs do not have an ov e rt mark of nega­ tion, a l tho u gh t he s p e cial c omp leme nt form from Ving is ge ner ally required '

',

,

with these verbs:

(23)

She prevented him from

r o om ] '

entering the room. [entails 'he didn't en ter

the

Form-function reanalysis

1 36

In a n umbe r of l a n gu a ge s , the comp l eme nts of th ese verbs actually do con­ tain an overt ne gator . Horn gives examples from L ati n ( (24); Horn 1 978: 1 73, from Lakoff 1 968) and Basque ( (25); i b id . , from Lafitte 1 962): (24) (25)

Potuisti pro hibe re

ne fi e re t 'You could have prevented it .

from

h appen in g

.'

Debekatu diot ez dezan holakorik egin. 'I forbade h i m to do suc h a th i n g [lit. . . . so that he '

'

not do . . .']

Example (24) d oe s NOT mean 'You could have prevented it from not h appen in g ; nor does (25) m e an ' I forbade him t o not do such a thing'. Espinal 1 992 de scr i be s the phenomenon i n Catalan as optional, which places t he complements of these verbs at the same stage of this process as the English clicbes in (2 1 )-(22) ( Espi n al 1 992: 3 36, 333): '

(26)

Tine por que (no) arribin. 'I'm afraid that they will arrive. '

(2 7)

Impediu que en Joan (no) surti elegit. 'Stop Joan from be i n g elected. '

Pa rat ac tic n egati on i s also found i n adver b i al subordinate clauses introduced by befo re , 'until' and 'unless' . There is n o covert entailed negation in these adverbial clauses. However, other adverbi al subordinate clauses presuppose the proposition in the adverbial cl a use, while these adverbial clauses do not do so. This i s illustrated by the following minimal pair offered by Levinson ( 1 983: 1 87): '

(28)

'

a. b.

Sue cried before she fi n i sh ed her t hesi s S u e died before she finished her thesis.

.

Semantically, the state of affair� in the subordinate clause has NOT occurred at the time of the state of affairs described in t he main clause (and may never occur) , unlike other adverbial subordinate clauses. This is the covert ne ga tiv e value found in these adverbial clauses. Vincent 1 993 ( amon g othe r s) p o i n ts out the presence of ne gatio n in the se adverbial clause types in F re nc h (Vincent 1 993 : 1 53): (29)

a van t que Pierre ne vienn e ' be fo re Pie rre comes '

Paratactic negation is also found opt i on ally in adve rbi al clauses in Catalan (Esp inal 1 992:333):

(30)

H aurem d'intervenir abans que (n o) arribi el nou ge rent 'We'll h a ve to take part in the discussion before the new man ager arrives. ' .

Vincent also observes the existence of a special complementizer for 'before'

clauses in Maltese (Vince nt 1 99 3 : 1 52-3 , citing Sutcliffe 1 936:207):

Cryptanalysis (3 1 )

(32)

1 37

war a Ii te l qu afte r COMP they. left 'after they left' qabel

t el qu

ma

b e fo re NEG.COMP 'before

th ey. l eft

they left'

comp l em ent izer ma is derived from the negative marker and possibly also the re l ativi zer (V i nce nt 1 993: 1 62, n. 1 1 ) . That is, the specia l comple ment izer is another ex ampl e of paratactic negation . O t he r e xampl es of etymo lo gical negative markers in these adverbial clauses are described for West F l em i sh (33) and Standard Dutch (34) by Haege m a n (Haegeman 1 995: 1 60):

The 'before'

(33)

tenwoare 'unless'

(3 4) te nzij 'unless'


concessive in con­ nectives ( i b i d . 1 93). Heine, Claudi & Hiinnerneyer ( 1 99 1 ) argue that all grammaticalization processes involve p r agma tic inference, which they called co n text -ind uced re­ interpre t a t io n . They h ypothesize that p ragmatic inference represents the micro­ structure of the gramm aticalizati on process, while metaphor represents the macrostructure . They convi n c in gly demonstrate that the sort of changes that Traugott & Konig describe as being only metaphorical involve pragmatic inferencing. Their method is to show that for any apparent quantum leap between two meanings or situation types, there is a continuum of situation types that bridge the gap between the two such that pragmatic inference will allow a speaker to extend the meaning gradually from o n e situation type to the other. They give examples of intermediate situation types for allative > purposive > future (H ei ne , Claudi & Hiinnemeyer 1991 :70, given in ( 1 5) below); c omit ative > instrumental (ibid. l 04, from Schlesinger 1 979: 3 1 0); Ewe gbe 'back' > 'back of [object]' > ' b eh in d ' (ibid. 65 - 8); Ewe kp:J ' see ' > ' re al ly' > counterexpectation (ibid. 1 94-7):

Grammaticalization ( 1 5)

a. b. c. d. e.

161

Henry is going to town. Are you going to the library? No, I am going to eat. I am going to d o my very best to make you happy. The rain is going to come.

Examples ( 1 5a-b) can be interpreted as a pure allative situation type, but with a human subject, it is plausible to infer an intention on the part of the subject. If intention is interpreted as an inherent part of the meaning of to, then the complement of to can be an intended action as well as a spatial destination, as in ( 1 5c) . In ( 1 5c) , the motion meaning for go remains if we assume that it is an answer to ( 1 5b) . In ( 1 5c), future time reference is a contextual inference but . in ( 1 5d), the motion meaning may be absent and the (intentional) future mean­ ing is then metanalyzed as an inherent aspect of the construction. Finally, while prediction is a contextual inference in ( 1 5d), it is metanalyzed as inherent and applied to an inanimate subject in ( 1 5e). Having given several such microstructural analyses, Heine et al. argue that one can still interpret grammaticalization as metaphorical because if one selects certain points along the continuum, the process then appears to make a quantum leap from one conceptual domain to another. But the metaphorical interpretation is superfluous and even misleading. It is superfluous because the microstructural analysis as metanalysis is sufficient to describe grammaticaliza­ tion. Worse, the metaphorical intepretation of grammaticalization is misleading as an empirical description of the · process and hence of the cognitive processes underlying grammaticalization . Metaphor involves a quantum leap from one conceptual domain to another (LakotT & Johnson 1 980), yet grammaticalization is a gradual process. There is good evidence for metaphor in lexical semantic change, but not so clearly in grammaticalization. Nevertheless, metaphor may constrain the types of metanalysis that constitute grammaticalization . There is one reason why one might want to invoke metaphor in gram­ maticalization, and that is to motivate unidirectionality. Heine et al. propose the foJlowing ranking of domains for metaphorical transfer: ( 1 6)

person < object < process < space < time < quality [state]

of these catego ri es are rather loosely defined, and not all lex­ ical metaphors obey this hierarchy. Consider for example the metaphor TIME IS MONEY (LakotT & Johnson 1 980:7-9), as in I have no time for this. Here, possession is the source domain transferred to time. Possession is a quality in Heine et al. 's scheme, because it is a stative relation; but it is lower in the hierarchy than time. Thus, the hierarchy in ( 1 6) is specific to grammaticalization processes, and so requires explanation. (Besides, the hierarchy still does not account for the intermediate situation types in grammaticalization paths.) Pragmatic inference or metanalysis, on the other hand, offers the possibility of a general explanation of unidirectionality: the contextual property is usually entailed by the inherent property but not vice versa. I believe this hypothesis However, some

1 62

Interference, in traference and grammaticalization

has a chance of success for accounting for unidirectionality in grammaticaliza­ tion, based on the examples given in Heine, Claudi & Hiinnemeyer ( 1 99 1 ) . For example, allative motion with a human subject usually entails intention , but not vice versa (many intended actions do not imply directed motion on the part of the intender), and inten tion usually entails prediction, but not vice versa ( predicted events without human subjects and sometimes even with them often are unintended). Of course, this hypothesis cannot be entirely persuasive without a system­ atic survey of grammaticalization paths and their intermediate situation types, which is beyond the scope of this book. Likewise, there may be some gram­ maticalization processes in which a quantum-leap metaphorical change does take place, rather than a gradual shift driven by metanalysis . The directi onali ty o f semantic ch ange i n grammaticalization may not always be predictable from the metanalysis of utterances in context. Whether it is or not, the fact of the continuous, gradual nature of the extension of function in many attested cases of grarnmaticalization dem o nstr ates that the structure of conceptual space plays a significant role in defining grammaticalization paths (see e.g. Haspelmath 1 997), as it does with intraference. 6.3.2.2

Fusion and morphosyntactic 'freezing' of a construction

Once a pe rip hr astic expression has

been chosen to express a novel meaning, it then undergoes FUSION, that is, it is perceived as a fixed unit (Ludtke 1 986:273 1 ; Keller 1 990/ 1 994: 1 1 0) . Ludtke and Keller present this as a psychological phenomenon, that is, en t r en c hment (see §2.4.2) . However, it is also a social phenomenon, namely the conventionalization of the periphrastic expression with a particular meaning. In other words, the fusion phase involves propaga­ tion of th e construction as a variant lingueme for the new function. As will be seen in chapter 7, conventionalization involves the reduction of variation in forms to express a particular function, that is, the reduction of variant linguemes . One way to reduce variation is, of course, to eliminate the older form. However, there often are variants of the new construction, such as the variety of postverbal negative emphatic markers in the history of French (pas, point, etc.; see §5 .4. l ) , that compete to be th e conventional expression of the function. Among the ways to reduce this sort of variation are to: (i) fix the word order of the construction - i.e. rigidification; (i i) eliminate optionality i.e. obligatorification; (iii) reduce the range of elements that fit into a slot in a construction to a closed class or an invariant element - i.e. paradigmaticizatitln. These strategies correspond to the morphosyntactic processes of grammatical­ ization in Table 6. 1 . It may not be obvious that two alternative forms o f a construction are actually in competition. Consider the following example of paradigmatidzation eventually resulting in an invariant element. Many languages in the Austroasiatic family have numeral classifiers, used in combination with a numeral modifying a noun (Adams 1 989). In some Austroasiatic languages such as Wa, the classifiers

Grammaticalization

1 63

are specific to particular classes of nouns (Adams 1 98 9 : 1 80). The classifiers appear not to be in competition with each other. But in other languages of the family such as V ietnam es e one classifier starts to be used with other noun classes, i n which case it is competing with other classifiers (ibid. ) . The existe nce of intraference entails that a form can become a competitor for any neighboring form in conceptual space, espec ial ly if th e intrafering form is more freq ue n t (see §6 .2). The same remarks apply to variants that differ in word order o r in the presence/absence of an element of the construction (i.e. optionality). 6 ,

6.3.2.3

E rosion , (mostly) phonological

The last stage in the grammaticalization cycle is EROSION of the conventionalized expression (Liidtke's usure phonique; Lud tke 1 986: 1 5 -23; Keller 1 990/ 1 994: 1 08). E rosio n is ge n er ally treated as phonolo gical, that is, reduction, coalescence and adaptation. However, as noted in §6. 3. 1 , coalescence and erosion may apply also to morp h e mes as well as ph on eme s Hence the correlation between periphrasis, fusion, erosion and functional, morphosyntactic and ph o nological processes respectively is not quite perfect. The principle of economy is invoked by Ludtke and Keller to account for erosion. As was argued in §3.4.4, it is p o s si b l e to interpret the e ssen tia l ly psychological p r i ncip le of economy as a social-interaction!!-1 phe nom en on , by exploiting the principle of immediacy in coordination by interlocutors on a single m ea n in g and the principle of joint salience in the interlocutors' common ground (§4.2.4) . In fact, since the' ran ge of p o s s ibl e ph o n ol ogical reductions of a co n struction is quite wide (everything from full and distinct articulation of every phoneme to a minimally distinct mumble), it is clear that the speaker chooses degree of reduction of a constructional form with the hearer in mind . Only certain elements in t he construction will be p hono logica lly eroded, namely the specific morphemes that are associated with the construction, the more invariant the better. To use the same example from §6. 3 . l : in the E n gl i sh go-future, [sBJ be going to VP;nJ], be is always present but in its usual range of forms, while the st rin g going to is alw ays present in just that form. Thus, it is going to that reduces to [gAna]; n ext certain forms of be reduce to enclitics on the subj ect in this construction as they do elsewhere, d ue to the high token freq ue ncy of those forms co mp ared with the variable p arts of the construction (subject, infinitival ve rb phrase). Of course, it is precisely the specific, especially invariant, morphemes associ­ ated with the construction that are interpreted by the interlocutors as en cod in g the meaning characteristically associated with the c on s truction as a whole (cf. §5. 1 ) . It is this fact that gives the impression that gramm aticalizat i o n is a process affecting individual morphemes (an d the lexemes they are derived from) . However, phonological erosion occurs only when the individual m orp hem e s have the meaning that is conferred on them by the construction. For t his .

,

reason, grammaticalization theorists now recognize that grammaticalization

applies fo whole constr uct ions not j us t lexemes and morphemes. ,

Interference, intraference and grammaticalization

1 64

As erosion proceeds, the need not to be misunderstood may assert itself, and the periphrasis-fusion-erosion cycle may repeat itself for the particular function in question. Also, the grammaticalized construction may itself be extended to a new function, that is, it may be drawn into a periphrasis-fusion­ erosion cycle for a new function . 6.3.2.4

G ram maticalization and structu ral reanalysis

Structural reanalysis plays an important role in some theories of grammatical change (e.g. Harris & Campbell 1 995), and is often implicated in grammaticaliza­ tion . STRUCTURAL REANALYSIS is the reanalysis of the syntactic structure of a construction, and thus differs from form-function reanalysis as described in chapter 5. A typical example of genuine structural reanalysis is the reanalysis of an external possessor as an adnominal possessor in some varieties of German (Haspelmath I 998a: 59):

( 1 7)

[Da zerrissv [dem Jungen]N r [seine Hose]Nr1s 'Then the pants tore on the boy.'

( 1 8)

[Da zerrissv [ [dem Jungen] NP seine Hose]Nr]s 'Then the boy's pants tore . '

Haspelmath ( 1 998a) argues that although a number o f grammaticalization theorists assume that structural reanalysis plays a significant role in gram­ maticalization, it is a distinct process. According to Haspelmath, genuine struc­ tural reanalysis does not involve the loss of syntactic autonomy or phonological substance; the syntactic relations or dependencies change in an abrupt manner; and the process is 'potentially reversible' (ibid .). Grammaticalization, on the other hand, does involve loss of autonomy and substance; the process is gradual, and it is unidirectional . The phenomena within grammaticalization which have sometimes been in­ terpreted as structural reanalysis are word class changes, alleged restructuring of the syntactic tree, reversal of head-dependent relations, clause fusion, and changes in grammatical relations. Haspelmath argues that these processes have the three aforementioned properties of grammaticalization - this is not really in dispute - and that the reanalysis interpretation depends on certain assump­ tions about syntactic representation that need not be the case. Haspelmath's arguments can be summarized as follows. Word class mem­ bership is a matter of degree, and the changes in word class are gradual. Using dependency relations instead of constituency for syntactic representation elim­ inates the alleged syntactic restructuring in grammaticalization. Some theories of heads would lead to absence of reversal of head-dependent relations in grammaticalization. Clause fusion is actually change in word class (syntactic category) membership (e.g. verb > auxiliary). Finally, grammatical relation ch anges can be interpreted as gradual extension (intraference) of the semantics llf an argument structure construction (case frame) to a new event class .

Notes

1 65

Haspelmath's arguments su ggest that str uct ural reanalysis is not as comm o n a syntactic change as has previously been believed. This section concludes the survey of mechanisms for in n ovat io n . I have argued that a wide range of types of grammat ical change involve changes in the form-function mapping, in sy nta gm a tic p roce sses (form-function re an alysis and grammaticalization) and in paradigmatic processes (interference and intraference) . Change in the form-function m appi n g is not the sole mech an ­ ism for gr ammat ical innovation . Processes such as phonological erosion and genuine structural rean alys is in grammatical constructions cannot be interpreted as involving the form-function map ping . Nevertheless, change in the form­ function mapping of words and constructions plays a major ro le in innovation of grammat ica l changes in languages.

Notes The past tense went(e) is actually a 1 3th-century innovation for the earlier past tense wende (David Denison, personal communication).

2

Harris & Campbell use the term 'borrowing' for all contact-induced change, in contrast to Thomason & Kaufman 1 988; see §8 . 3 .

3

Allen argues convincingly against Janda's hypothesis that the transi ti on from affixal to clitic

-s

occurred because of the convergence of affixal

-s

possession construction . Allen's proposal, that the affixal

with a pronominal -s

(h) is

became a clitic after

spreading (at least optionally) to the entire nominal paradigm makes it even more

clear that th e change was from an affix to a clitic, rather than from a word (the pronoun his) to a clitic.

4

It might be suggested that avoiding misunderstanding is the mechanism underlying intraference as well. However, intraference does not generally involve the asymme­ try between a longer (periphrastic) expression and

a

shorter one . One situation in

which avoiding misunderstanding mi gh t motivate intraference is the repl acem en t o f a form with a wide range of functions by a form with a narrower range of functions, as suggested by Bybee (personal communication) for the extension of the progress­

ive into the ongoing present function.

5

Traugott & K o n i g argue that metanalysis is actuated via an intentional mechanism ,

namely the Gricean principle of informativeness: be as in form ati ve as possible

(Traugott & Konig

1 99 1 : 1 9 1 ).

However, given the framework of the indeterminacy

of the form-function mapping defended in chapters 4-5, it is not necessary to appeal to an intentional mechanism here. 6

It would appear that it is at the fusion stage that idiomatization of the expression

for the meaning in question takes place. In fact, idiomatization takes place at the periphrasis stage, that is, at the innovation, not propagation, of the construction with its new function. Any novel use of language is partly nonconventional (§4 . 3). The nonconventionality of the novel use is reflected in the (non)compositionality of the expressiori . Further steps in the idiomatization process (e.g. loss of analyzab­ ility) occur with further semantic extensions of the construction, i.e. new cycles of periphrasis-fusiono-erosion .

C hapter 7

Selection ( p ropagation) of i n novations i n lang uage change

7 .1

Introduction

Altered replication and selection, as defi n ed in §2.3, are the two fundamental processes involved in the framework for anal yzi ng e vol ution ary processes. Al­ tered replication of the replicator is innovation in language ch ange . Selection of interactors is the propagation of an innovation in l anguage change. It has been ar gu e d in the p rece d in g chapters that the mechanisms for innovation are functional: more precisely, the mechanisms invo l ve rem app in gs of the link between form and function in a conventional l i ng uistic s i g n or l i ngueme - t he rep licator . Chapters 5 a n d 6 discu ssed a range of mechanisms for innovation, and the processes of lan gua ge change that appear to be the result of tho se mech ani sms . In contrast, the mechanisms for propagation of innovations in l anguage change are social, that is, they i_nvolve the rel ation s h ip between the speaker the inte ract o r - and the society to which she belongs. Pr opagati on is essenti�lly the ad o pti o n of a convention . Thus, two fundamental prerequisites to an un de r­ standing of mechanisms of p ro p a gati o n are the structure of so c iety (rel ev an t to the l an guage used in the soci ety) and the nature of c on vention s . These two concepts were given an init i a l exp lic ati o n in §4. 2 . 3 and §4.2.4. In §7 . 2 and §7 . 3 , I will e l a bo ra te o n these concepts i n order t o lay the gr o undwo rk for a theory of the mechanisms for p ro paga ti on . In §7.4, I witl i n te rp re t curre nt s o ci o lin ­ gui st i c theories in the ev ol u tio n a ry framework. There I will argue that the basic mech an i s m for propagation is the speak er id e ntify in g with a social group, a nd that patterns of propagation in social populat io n s that are paral le l in si gnificant respects to patterns of selection in biological p opulations .

7 .2

Communities, societies and the internal/external distinction In language change

In §4. 2 . 3 , I argued against the naive view of a speech community as a homoge­ b oun ded group of individual speakers. Instead, a speech com­ m unity is a gr oup in g of indiv id uals by their participation in a social domain. The social d omai n can be defined by the shared ex pert i se of the members of the c o m m unity , by virtue of w hich the members of the community sh are common ground (mutual knowledge and beliefs). Shared ex per tise in a commun ity is

neous, sharply

1 66

Communities and internal/external change

1 67

not the only basis for common ground between individuals. Individuals can common ground th roug h experie nce s h a red d i rec t ly (perceptual basis) and i nd i rectly , via conversation (discourse basis) . Finally, different commun­ iti es have distinct codes (in the s ociol i n guis t i c definition of a code). The chief consequence of the sophisticated a p p ro ach to speech communities is that all individuals belo ng to m u lti ple communities, and hence all individuals are to some de gre e multi l i n gual . This latter fact is obvi ous when there are two or more codes that are u su a l l y described as d i ffe ren t lan guage s . It is n o t so obvious when the various codes are us ually described as parts of the same l anguage . On the other h and , soc iol i n gu i s t ic analys i s has made it amply clear that all speech communities (in the naive sense of that word), including so­ called monolin gu a l communities, possess variation in l i ngui s tic form to some de gree. S oci ol in gui sts describe this vari at i o n as structured heter ogen eity (see §3.2), orde red by social domain - that is, commun i ty in the soph istic ated sense. In this s e ction, I will d ev e l op a d e scripti on of soci ol i n gu ist ic h e te ro gene i t y in terms of the social st ruc t ure and the l i n gu i sti c behavior of communities in the s op hi sti cated s en s e . Clark ( 1 996: 1 03) gives the fo llowi n g ex amp le s of what he calls 'comm on ' communities defined by shared ex perti se . I h ave reordered Clark's lis t , g ivin g his examples of communities:

ga i n

(I ) Nationality Re sidence E thn i city Language

Cohort [Age] Gender Educ ati on

Occupation Employmen t [Place] Subculture Hobby Religion Politics

American, Canadian, Dutch New Ze alan de rs, Californians, Glaswegians Blacks, Hisp an i cs , Japanese-Americans English speakers, J apan e se speakers, German spe ake rs teenagers, senior citizens, thi rty -year- ol ds men , wome n u nive rsi ty students, law students, h igh school grad u at e s ophth almo lo gi sts, p l um bers , used-car dealers Ford auto workers, Stanford faculty, Newsweek reporters rock musicians, d r ug addi c ts , teenage gan gs p i an i sts, baseball fans, p hilate l i s ts Protestants, Bap ti sts , M u s lim s Democrats, libertarians, Fabians

The reader familiar with socioli n gu i sti c s will immediately rec ognize com­ monly used social variables in soci o l i n gu ist ic research, in p art icular gender , age cohort, education a n d occup at io n (m ore broadly defined as social classes in class-based research; see Labov 1 972b). Also, most s o ciol i n guis tic rese arch fixes the value of certain other social variabl e s which de fine other types of com­ munities on the li s t in ( 1 ): the language of the sp eake r s and a combination of nati onality, residence and et hnici ty (e.g. white Phi lade lp hia English speakers). Hence, we may interpre t correlations of l i n gu i s ti c vari ant s w i t h s o ci al v ari abl e s

1 68

Selection of innovations in language change

as correlations of those variants with communities. We will return to this point below. Clark also points out that communities are nested in degrees of greater or lesser inclusiveness, and gives these examples (among others; Clark 1 996: 1 04): (2)

Residence

North Americans ::> Americans ::> Westerners ::> Californians ::> Northern Californians ::> San Franciscans ::> Nob Hill residents

Occupation

middle class ::> professionals gists ::> ophthalmic surgeons

Language

English speakers ::> speakers of New Zealand English speakers of Auckland English dialect

::>

physicians

::>

ophthalmolo­ ::>

The nesting of communities can be compared with the nesting of populations of biological organisms referred to in §2.2 (see Grant 1 98 1 : 8 1 , Fig. 7 . 1 , for a more fine-grained nesting of biological populations): (3)

Organisms

species ::> geographical race ::> deme

The nesting of biological organisms is defined in terms of degree of reproduct­ ive isolation: species are more or less completely isolated reproductively from other species, while geographical races and demes are relatively isolated re­ productively from other races in the same species and other demes in the same race respectively. The relevant nesting of individuals in comm unities for the evolutionary model of language, then, is in terms of their communicative isolation in that community. And the nesting of communities in (2) intuitively does provide an approximate guide of the degree of communicative isolation. Nob Hill residents, other things being equal, are more likely to talk with other Nob Hill residents than with other San Franciscans taken as a whole, and so on up the scale. Ophthalmic surgeons, other things being equal, are more likely to talk with other ophthalmic surgeons (on the job, at conferences, etc.) than with other physicians taken as a whole, and so on up the scale. Of course, if an ophthalmic surgeon is a neighbor with a lawyer on Nob Hill, she may talk more with the lawyer than with another ophthalmic surgeon living in Chicago . However, that is by virtue of the fact that they are neighbors, and so are in the same exclusive community based on residence, not (only) the fact that they are both in the more inclusive occupational community (professionals). We may now relate community exclusivity and communicative isolation to other significant sociolinguistic concepts. Milroy & Milroy ( 1 985) and Milroy ( 1 992b) argue that the strength ofan individual's tie to other individuals in the community is the most significant factor in the maintenance of linguistic con­ ventions ('community norms' in their terminology) and the transmission of change. They adopt the following definition of TIE STRENGTH: 'The strength of a tie is a (probably linear) combination of the amount of time, the emotional intensity, the intimacy (mutual confiding) and the reciprocal services which

Communities and internal/external change

1 69

characterize a tie' (Granovetter 1 973: 1 36 1 , cited in Milroy 1 992b: l 78) . I suggest that the strength of a tie is relevant to language evolution because it correlates with degree of communicative interaction: we talk more to those with whom we have strong ties, and in fact our strong ties are strong in si gnificant part due to the degree that we talk with those people. The weaker the tie is to someone, the greater our communicative isolation from that person . The deme is an important unit of organization in biology (Hull 1 988 :433): as noted in §2.2, it is the unit where individuals have equal likelihood of mat­ ing, and where one can say with confidence that the individuals share the same gene pool. In §2 .2, I suggested that the equivalent unit of organization in society is the social NETWORK , the group of people which is most likely to talk with one another, and less likely (or less frequently) to talk to other people outside the group. I suggest that what distinguishes the social network from more inclusive communities as defined by Clark is a high degree of personal common groun d shared by in d ivid u al s in the network, in contrast to common ground derived from shared expertise in the community as a whole. Personal common ground is based on directly shared experiences and experiences shared through conversation. Personal common ground is closely related to the first criterion proposed by LePage & Tabouret-Keller ( 1 98 5 : 1 87) for the focusing of a group of individuals, a concept closely related to tie strength, namely 'close daily interaction in the community'. Thus, personal common ground follows as a consequence of communicative interaction; the more communicative inter­ action, the more personal common ground. Persons in more exclusive conim unities in the nested scales in (2) are more likely to share substantial personal common ground . To continue our example from above, two ophthalmic surgeons probably have attended the same con­ ferences, heard the same papers, know the same fellow ophthalmic surgeons, and talked to each other about their own professional experiences; this is less likely if the two are physicians with different specialties. The more communities shared among individuals, and the more exclusive the shared communities are, the more personal common ground will be shared. If in addition our opthalmic surgeons are also neighbors on Nob Hill, they will have shared the same events (a heat wave, a crime wave) and talked to each other about their own experiences living on Nob Hill (e .g. remodeling their houses, traffic problems, a fire down the street). Strong ties correlate with the two chief measurements of social networks, network DENSITY and network MULTIPLEXITY. Network den sity is the number of connections of any kind between individuals in a group (see e.g. Chambers 1 995:7 1 ), that is, whether they know each other or not. Network multiplexity is the number of different ways in which individuals in a network know each other, e.g. family relations, workmates, classmates, shared activities. Multiplexity corresponds to the number of communities which are shared among individuals in the network, and density corresponds to the number of links among indi­ viduals in any community which they may share. As we noted above, the more communities shared among individuals in a network, and the more exclusive

1 70

Selection of innovations in language change

those communities are, the stronger the ties among the individuals in the . network . The preceding paragraphs illustrate the fact that groups of human bein gs in social terms differ from groups of organisms in biological terms in a signific an t respect. Human be in gs can and do be l o ng t o mu lti ple comm un it ies , while o rganisms be l on g to just one deme. One consequence of this fact is that two people may bel o n g to a very exclusive c om mun i t y and y,et not be members of other communities at all . For example, a mother and her teenage son may belong to a very exclusive community in terms of residence (the same house) and family (n u cle a r famil y members), but are in different communities in other ways, e .g. gender (female vs male) , age (teenage vs middle aged), hobby (ga r­ de n i n g vs classical pi an o ) , etc. As a result, there may be a high degree of communicative interacti on amo n g individu al s who bel on g to different com­ munities. Of course, to some extent the communicative interaction is domain­ specific: mother and s o n will tal k a lot about household chores and family members, but not much about classical music. But there will nevertheless be si gnifi can t transmission of language, that is, repl i cati on of lin guem es, from one community code to another. An d a large part o f the reason for th a t is the relationship between community codes. One of the most obvious ways to differentiate commun ity codes is by com­ paring their communal lexicons. In th e communal lexicons of tw o communit­ ies, one can find di ffe rences in fo rm, meaning or both. Clark ( 1 998) gives the fo l lowi ng examples of di ffere n ces between the co m mun ity of Americans in general (defined in terms of natio n ality) and the commu n ity of American miners (defined in terms of occupati on):

(4)

Specialized form, same meaning

[Miners: hoist, 'conveyance between levels of mine ' ] [ Amer ica n s : (mine) elevator, 'conveyance between levels

of

m in e' ]

(5) Specialized meaning, same form

[Miners: raise, ' shaft between 2 levels of a mine'] [Americ an s : raise, 'increase in salary']

(6) Specialized form-meaning pairing

[Mi ne r s : stope, 'type of mine cavity']

However, the fact of the matter is that there are an enormous number of lexical items, gramm at i cal elements, and grammatical constructions whose form and meaning are identical across these communities: (7)

Same form-meaning pairing [Miners: walk, 'move on foot'] [Americans: walk, 'move on foot']

(8)

Same form-meaning pairing [Miners: be, 'predication of a nonverbal concept'] [Americans: be , 'pre dica tion of a nonverbal concept']

Communities and internal/external change (9)

171

Same form-meaning pairing [Miners: +PL, 'more than one object'] [Americans: +PL, 'more than one object']

( 1 0)

Same form-meaning pairing

[Mi ne rs: DEM precedes NOUN, 'deictic specification of location of object'] [Americans: OEM precedes NoUN, ' deictic specification of location of object']

In other words, there is a large amount of shared expertise, n amely core experti se about the world and about human beings' conceptualization of the wor l d for communicative purposes, that is la rgely the same across d omai n s (see §4.2. 3). This cor e ex pertise is often expressed by the same fo rm-meaning pai rin gs acr oss commun ity · codes. When it is, we generally speak of the d iffer­ ent community codes as being part of the same lan gu a ge (in sense of language as a conventional si gna l ing system) . In doing so , we ignore differences in com­ mu nity lexicon - and in commu n ity grammar. The latter sort of variation is what sociolinguistic analysis typically focuse s its attention on . As mentioned above , variationist sociolinguistic analysis focuses on certain types of community. It ge ne ral l y holds invari ant res iden ce (defined quite exclus­ ively), nationality , and language; and e x ami ne s different communities within this group based on age cohort, gende r and so metime s occupatio n (cl ass ), education and/or eth n icity . This sort of analys is impli c i tly assumes that com­ munities based on residence in p a rticu l ar (nationality gene rally follo w ing from residence) have a different status from communities based on age cohort, gender, occupation, etc. In fact, this is a reasonable assumption, at least for the geographically relat­ ive ly nonmobile groups typically studied by modern variationist sociolinguists: inner-city communities, particul arly lower-class communities including street gangs (Labov I 972a, Milroy 1 987); adolescent students (Eckert 1 989); geo­ graphically isolated groups such as in Martha 's Vineyard (Labov 1 963/ 1 972); to name a few classic studies. That is, communities based on residence of these types are more communicatively isolated from other communities than com­ munities based on age cohort, gender or occupation . People belonging to dif­ ferent age cohorts, genders and (to a lesser extent) occupation are more likely to talk with each other because they tend to live together, work together and engage in social transactions with each other. In other words, there is a higher degree of continuous contact among different age cohorts and gender than among different neighborhoods or l arger geographical units. Indeed, sociolinguists still use the word 'comm un ity ' to describe a group of individuals l argel y based on residence, regardless of the heterogeneity of other communities (in the strict sense used in this book) to which these same individuals belong. As noted in §4.2. 3 , I will use the term SOCIETY to describe a set of individuals among whom there is a certain degree of communicative interaction, that is, rel atively strong ties, and - more important - a certain degree of communicative isolation (weak ties or no ties) with respect to other

1 72

Selection of innovations in language change

societies. In that communicative isolation is more important than communicative interaction, a society is analogous to the species concept in biology (Grant 1 98 1 :9 1 ; see §2 . 2). There is enough relative communicative isolation among communities based on age and gender within a society as defined in the preceding paragraph for differences to exist. But there is enough communicative interaction - contact between these communities in a society that these differences will be manifested in proportions of variants of a linguistic variable, that is, variant linguemes. In other words, societies - groups of individuals held together by relatively strong network ties - typically based on geographical residence are linguistically vari­ able because they are socially heterogeneous in terms of other community types. But the variation is probabilistic rather than categorical because these other community types are not communicatively isolated; they involve the same indi­ vidual speakers. If we think of the variation as representing a lingueme pool in the population of utterances created by speakers in the society, then the hetero­ geneous character of a language - defined as a population of utterances - and the structured nature of that heterogeneity (in terms of correlations between variants and communities), is to be expected in the evolutionary framework adopted here. One consequence of the evolutionary model of language, and of the socio­ linguistic model of language that it accounts for, is that speakers must be able to represent knowledge of the structured heterogeneity of the makeup of the lingueme pool. This sort of speaker knowledge can be reasonably represented using SPREADING ACTIVATION MODELS of human cognition (Collins & Loftus 1 975; Elman & McClelland 1 984; Cruse & Croft to appear eh. 1 2) . Spreading activation models allow for the representation of quantitative values associated with the relevant unit of knowledge (such as a lingueme) in terms of degree of activation. Degree of activation in turn is primarily determined by exposure to tokens of the un it in question in the environment (such as uses of a lingueme in. utterances) . Activation of one node or set of nodes in a network can activate (or inhibit) other nodes. Degree of activation corresponds to knowledge of the proportions of the variants of a lingueme in the community. Activation of a lingueme involves activation of the nodes or sets of nodes corresponding to the lingueme's form, meaning, and the community in which the utterance occurs. (The latter is determined not only by the speakers but also the situation and topic of the utterance: a lingueme occurring in a conversation about classical music at a concert will activate a different community than the same lingueme occurring in a conversation about cooking dinner at home between the same two people.) Although I cannot develop this model further here, it should be clear that the spreading activation network model can in principle represent the sort of lin­ guistic knowledge that the evolutionary and sociolinguistic theories require of speakers. Of course, it need not be the case that the core expertise shared across· com­ munities is expressed by the same form-meaning pairings. Instead, a different

Communities and internal/external change

1 73

set of form-meaning pairings may be used. In this case, we usually speak of a multilingual society. One might expect there to be competition or mixture between the two sets of form-meaning pairings, that is, the two li n gueme pools, that are spoken in the same society, indeed by largely the same indi­ vidu a l s The evolutionary frame wo rk predicts that competition or mixture between two lingueme pools will be resisted to the extent that there is separa­ tion (communicative i so l ati on) and equali ty of the communities, that is, the domains of sp.ared expertise found in the society. This in fact appears to be the case (see §8. 5). We may summarize the model of com m un ities as follows. Communities are defined by shared expertise. Communities exist in greater or lesser degrees of inclusiveness/exclusiveness. An important boundary in the degree of exclusivity of a community is the boundary inside which members share a significant degree of personal common ground, which can be obtained only through d irec t contact, in p ar t icu l a r d i rect communicative in teraction. A significant degree of personal common ground corresponds to a social network with reasonably strong ties in sociolinguistic theory, and a deme in evolutionary biology. Unlike the organization of biological organisms in demes, however, all individuals in a society are members of multiple communities. Each com­ munity has its own code, hence all individuals in all societies are multilingual . However, much of the shared expertise in a community is common across communities, and the code is often common as well. Certain communities, particularly those based on residence (and ethnicity or nationality), tend to be relatively communicatively isolated . Other communities, such as those based on age, gender and occupation, tend not to be as communicatively isolated. A society is defined as relatively communicatively isolated from neighboring societies, and typically is defined in terms of geographical region and (where releva n t) ethnici ty or nationality. With respect to other community types, such as age, gender and occupation, a society is heterogeneous. The language - the population of utterances and the lingueme pool it defines - of a society is structured in a corresponding fashion . Variant linguemes are correlated with different communities in the society, but the correlation .is not categorical, because of lack of commun icative i s o l a tio n between different communities in the same society (multilingual societies will be discussed in chapter 8.) I n fact , much of the variation in sociolinguistic studies has been shown or is hypothesized to represent language change in progress. That i s, the variation represents the propagation of a novel variant in the community, that is, t he selection of a vari a n t of a lingueme as the conven tional sign for t hat concept in that community. In order to situate the results of sociolinguistic research on the propagation of variants in the evolutionary framework, it is necessary to return to another theoretical co n struct from chapter 4, convention, an d li n k it to sociolinguistic research on the one hand and the evolutionary framework on the other. .

1 74 7.3

Selection of innovations in language change Propagation and the adoption of conventions

The propagation of a novel lin guistic vari ant i s essentially the adoption of a new lin gu i stic convention by the c ommunity . This process takes time, and in fact may never go to completion, that is, complete replacement of the old convention by the new one. Moreover, many innovations are not selected, or do not survive very long. In descri bin g a lin gui stic convention in §4.2 .4, I as s umed (as d.id Clark and to a l esser extent Lewis) that a sin gle conventional s i gnal would be used for speaker and hearer to coordinate on a particular recurrent situation type (ignoring for the time being the problems described in §4.3 in identifying a recurren t situation type) . Lewis, however, did allow for degrees of convention­ ality in his definition ( Lewis 1 969: 76-80). The various 'almost's in the Lewis & Clark definition of convention in §4.2.4 correspond to different ways in which a s i gnal could be a convention of the communi ty to varying degrees . The Lewis & Clark definition of convention is reproduced below, this time with the 'almost ' s indexed by the parameters by which convention can vary in degree : ' (1 1)

1 2

3 4

5

a regularity in behavior (pr oduci n g a string of sounds) that is partly arbitrary a. other regu larities in behavior would be approximately eq ually preferable by a lmost 3 everyone in the comm u nity that is comm on ground in a community as a coordination devi�e a. almost , everyone in the co m munity conforms to it b. almost2 every one expects almost , eve ryone else to conform to it c . almost4 everyone would prefer any additional member o f the community to conform to it if almost, eve ryo ne in the commun ­ ity alread y conforms to it d. almost4 everyone would prefer any new member of the commun­ ity to conform to another regularity if almost 1 everyone in the community were already conforming to it for a recurre nt coordination problem (commun ic a ting a meaning)

The different ' al most ' s in the definition can be defined as a proportion or percentage. In fact, the two si gn ificant 'almost's in the development of a conven­ tion are ' almost1 ' , which defines the proportio n of use of the signal as a linguistic convention in the commun i ty , and 'almost4', which defines the proportion of mem bers of the community that prefer the use of one convention over another. I will use two s i mple examples, from the term i nology of hist orical l inguistics and l 9th -centu ry French avia tion te rminology , to illus trat e this point. The class of lin guistic changes discussed in §6. 3 is called grammaticalization by some historical lin guists and grammaticization by others. It is clear from the literature that both t erms are current and are used roughly for the same recur­ rent coordin a t i o n pro b lem . Let us assume for the sake of argument that in fact the tw o terms are used equally, th a t is, in about half of the cases when a

Propagation and the adoption of conventions

175

historical lingu i s t wants to refer t o the p henomenon , gramma t icaliza t ion is used, and in t h e other half, grammaticization is u se d . It is re as o n ab l e to assume that clause 2 o f the definition is la r ge ly i nv a r iant. It is approximately equally arbitrary to choose either grammaticalization or grammaticization as t he convention of the hi st o ri ca l li ngui stic co mmun i ty . That is, almost3 all m em be rs of the commun i ty would a pri ori accept either word. To be sure, t he re m i ght be s o m e who find one somewhat preferable (e . g. gramma t icaliza tion because of Meillet' s p receden t, grammaticalisation in Fre nch) ; but there is no m aj o r difference between the two . In fac t , it app ears that a fair degree of d iffer e nc e in functional utility of a l t e m at iv e conventions must be allowed, in order to allow fo r th e typol ogical diversity of conventional li ngu i s­ tic structures u sed for a given function across the world's languages. This i s to say that in general, differences in functional utility do n ot pl ay a role in the pro p ag ati on of a variant; only diffe ren ce s in social util i ty d o (see §2.4.2, §7 .4.3). The crucial part of the definition is clause 4. ' A l m o s t , e ve ryo ne ' is in fac t only about half: about half1 of the commun ity uses grammaticalization and half, grammaticization . However, the two half-conventions are known t o almost everyone in the community . For example , al th ough I us e the term gram ­ maticaliza tion, when I re ad an article on grammaticiza tion, I recognize what the a u th o r is talking about , and I recognize that the author is using a partly conventional alternative to the half-convention that I follow. (If the au t h o r used smtntification i ns tead , then he or she wo u l d not be co n forming to any c o n ven t i o n of the c omm u n i ty . ) So do most of my colleagues, that is, a lmo s ti everyone in the community expects about half1 of the community to conform to gramm at ica liza t io n (or grammaticization, m utat i s m u tandi s ) . It is p o ss i b l e that s o me members of the c o m m u n i ty who read very se l e ct ive l y may not k n ow of the e x i s te n ce of t he other convention; but this is pro b a b l y a small minority, and certainly wo uld not in cl u d e the ce ntra l members of th e community. Likewise, the preference for using either gramm a ticaliza t ion or grammaticiza­ tion is cent ral to the propagation of a co n ve n ti o n . How many members of the c o mm u n i ty would prefer that o t he rs u se gra mm a t ica lizatio n if o nl y about half1 of the comm u n i t y are u s i n g it? Presumably, ' alm o s t4' will include the half t ha t uses grammaticalization. But the same group would also p refer an al tern a t i ve regul a r i ty , e.g. grammaticization, if enough of the c o m m un i ty , s ay h a l f1 , we re a l re a dy conforming to it. In o t h er words this pa rt of t h e definition of c onve n ­ t io n i mpl i es that members of a comm u n i t y would prefer almost everyone in a c om m u n i ty to c o n fo rm to the same co n ve n t i o n , whatever it is; that is, ev eryon e would eventually conform to a si n gl e convention . (There wo ul d of course be a few stubb o r n diehards who insist on using an al te rna t i v e form e ve n when almost everyone else has a b an d o n e d it; but we must allow for the e x i s te n ce of such types in a r e a l i stic theory of convention. ) The preference fo r a unique convention does i n d e ed appear to b e a phen om ­ enon of human language . All l an gu ag e s are h e t er ogen e o u s , as n ot ed in the p recedin g section. That is, a language's l i n g ue me pool h as multiple variant lin guemes , that is, multiple partly co n venti o n a l signals for a particular meaning.

1 76

Selection of innovations in language change

The variants typically correlate with communities within a society, but most speakers use multiple variants, albeit in different proportions, in the utterances they produce. Nevertheless, there appears to be a natural human tendency for a community to select one alternative as the conventional signal for a recurrent coordination problem. That is, there seems to be a natural human tendency to increase the conventionality of one variant of a lingueme in a community at the expense of another, albeit over a long period of time in many cases. We may call this the FIRST LAW OF PROPAGATION. A striking example of the First Law of Propagation is the development of technical terms for newly discovered or invented phenomena. At first, a wide range of terms is used, and this range is relatively quickly reduced, ultimately to one term. This pattern is observed in a number of cases in a study of l 9th­ century aviation terms in French (Guilbert 1 965). The hot-air balloon (without a means to direct its motion) was invented by Montgolfier in 1 783. He called it a machine aerostatique, but it was also called aerostat by the end of that year. Other terms used in 1 78 3 - 1 784 were statique, globe aerostatique, ballon aerostatique, and the truncations of the latter two, globe and ballon (Guilbert 1 965 : 54). There was even the feminine form aerostate along with masculine aerostat (ibid.); the Journal de Paris assumed that the latter referred to the activity and the former to the machine itself (ibid.). 2 But the masculine aerostat was quickly established; by 1 798 the Academie's dictionary included only ballon aerostatique and aerostat. The truncation ballon and aerostat were the only terms used in the l 9th century and were essentially synonymous throughout this period (Guilbert 1 965:57). Yet as early as 1 787, it was noted that there was a difference between the two terms, which held through the 1 9th century: aerostat was more learned or scientific, ballon more popular (Guilbert 1 965:545, 57). Aerostat was also used to describe any machine that hovers in the air, not just a balloon in the period 1 86 1- 1 89 1 studied by Guilbert; ballon was never used in this context (ibid. 58). Later the terms were largely replaced by montgolfiere, which was used sporadically in this period (ibid. 588). It is not immediately obvious why the First Law of Propagation holds. After all, human beings have the memory capacity to handle multiple variants for many linguemes; bilingual individuals have two complete sets, for example. And members of a community can expect other members to know all of the variants, if they know them themselves. If there is more than one solution to a coordination problem, and speaker and hearer know all of �'1e solutions, any solution can be used. Yet there is a strong tendency to avoki complete syn­ onymy, that is, multiple forms with the same meaning and the same social (community) value . For bilinguals, the two languages will have different social values, and so the variants of a lingueme are not completely synonymous. However, there is a plausible social explanation of the First Law of Propaga­ tion, which will be presented in §7.4.2. There are three ways in which increasing the conventionality ofa variant of a lingueme can be done, corresponding to modifying one of the three clauses of the definition of convention that do not involve a qualifying 'almost' . The first two ways include what Trudgill calls reallocation of variants that arise in

Propagation and the adoption of conventions

1 77

contact sitq,ations (Trudgill 1 986: 1 1 O); the third involves replacement of one variant by another (e.g. through processes of leveling and simplification; see Trudgill 1 986 and §7.4.4). The first way is to modify clause 5 of the definition of convention. That is, speakers will divide the coordination problem so that the alternative forms are used for distinct functions. This corresponds to elimination of synonymy and subsequent functional specialization or division of the meaning of the com peting forms. For example, in my speech I distinguish the meanings of gram­ maticalization and grammaticization. JJ,eine Claudi & Hiinnemeyer ( 1 99 1 : 1 3 , 20- 1 ) note that the terms have been used for two relatively distinct stages of the diachronic process: the conventionalization or syntacticization of a discourse­ pragmatic strategy, and the further reduction of the conventionalized strategy. I restrict grammaticalization for the latter stages and would define grammatic­ ization as the first stage. If this functional division of labor were adopted, then each form would be much more conventional in its restricted ftinction than it was in its broader function. (In fact, grammaticization would t-heti compete with conventionalization and syntacticization.) This particular innovation has not itself been propagated. But there are many cases of a functional division of labor for competing conventions, such as the split between the two plurals brethren and brothers mentioned in §2.4.3. A partial separation of meanings was described above for aerostat vs ballon, where the latter was not used for a hovering machine other than a balloon (e.g. a helicopter). The counterpart change in phonology would be reallocation of variants as allophones occurring· in complementary phonological variants, as has happened with the variants [i] and [a] of the variable III in South African English (Trudgill 1 986: 1 6 1 , citing Lanham & Macdonald 1 979). The second way to increase the conventional status of alternative forms is to modify clause 3 of the definition of convention . That is, speakers will divide the community or set of communities and associate the distinct forms with distinct communities. For example, I heard a historical linguist suggest that grammaticalization tends to be used by European-trained historical linguists and their students, while grammaticization tends to be used by American-trained historical linguists and their students. Setting aside the validity of this observa­ tion (it is only partly true), it represents an attempt to divide the usage by community based on education . If it were true, or if speakers shifted to con­ form, then each form would be much more conventional in its restricted com­ munity than either was in the wider community. This phenomenon is also familiar to historical linguists aqd sociolinguists. Otherwise synonymous forms come to be associated with different stylistic registers (corresponding to different situation types associated with different communities), or with different communities within a society, such as age cohort, gender, or class/occupation (see §7 .4. 1 ). An example is the social or stylistic difference between aerostat and ballon: the former was used in the scientific community, the latter in the wider French community. The phono­ logical counterpart is the reallocation of phonological variants as stylistic and/ or social variants (Trudgill 1 986: 1 08-26). ­

,

1 78

Selection of innovations in language change

The third and last way to increase the conventional status of alternative forms is to modify clause l of the definition of convention. That is, speakers will increasingly select one form over the other. The end result of this process is of course the propagation and increased conventionality of the selected form, and ultimately the loss (extinction) o f the alternative form. This is the type o f change that is focused o n b y sociolinguistic analyses of language changes in progress. In fact, combinations of the three ways to increase the conventional status of alternative forms can occur. For example, there was a partial functional split of aerostat and bal/on as well as a split of the terms into two different communities (and their eventual replacement by montgolfiere) . And complete replacement may never occur due to a functional or social specialization of the losing variant. However, the first two processes do not involve differential per­ petuation of replicators (the variants); each variant survives in its newly defined niche. For this reason , we will now focus on the third way, direct competition and eventual replacement. Competition leading to perpetuation/extinction is the one process of the three that corresponds directly to the propagation of a gene allele in biological populations. 3 The propagation of a linguistic variant is a selection process: one variant is selected over another one. Selection is a process that takes place among interactors, however. A speaker does not produce one linguistic variant in preference to another in an utterance because of its linguistic properties. A speaker identifies herself with a community or a subset of a community and that causes her to produce one linguistic variant in preference to another. In order to understand this process, one must look at the social structure of communities, that is, populations of speakers.

7.4

Propagation and the structure of speaker populations

A number of proposals have been m ade by sociolinguists and dialectologists about the structure of speaker populations and the types and paths of changes that occur. These proposals are supported by a wide range of empirical evidence. There are certain striking parallels between certain of these proposals and the structure of biological populations with respect to selection. As was discussed in §3 . 3 . 1 , sociolinguists generally assume that the two competing variants already exist, and that one of them has entered into the society or network 4 via contact (cf. the Kerswill quotation in §3. 3 . 1 ). We will begin with that assumption, and then examine the situation with innovation proper, that is, altered replication of a lingueme . 7.4. 1

The locus of channels of propagation in the population

The first question to be dealt with is how a novel variant is introduced into a community and then spreads. Milroy & Milroy ( 1 985) and Milroy ( 1 992b)

Propagation and the structure of speaker populations

1 79

offer a theory based on the structure of the speaker popu l at io n , the weak-tie theory, and argue for it in favor of an alternative theory proposed by Labov ( 1 980) . Labov argues that the channel for a novel variant to enter a community from outside is an individual with high prestige and extensive contacts in the community but who has many contacts outside that community (Labov 1 980:26 l ; cf. Milroy & Mil r oy 1 9 8 5 : 343; M il r oy l 992b: 1 72). M ilroy questions the existence of such an individual: 'it is doubtful whether an individual can be a central member of a close-tie co mm uni t y and at t h e same time have large numbers of close-tie outside contacts' (Milroy l 992b : 1 74). The Milroys argue instead that it is an individual with weak ties, to both the network in question and to outside networks, who introduces a novel vari a n t to a n e two rk Th is is the pattern that they find in their Belfast work: the hi g h est users of novel v a ria nts are the individuals (males or females) for whom the v ari ant is not correlated with strong netw o rk ties: 'in the case of both /rJ a nd /a/ i t is the persons for whom the vowel has less significance as a network marker who seem to be leading the linguistic change' (Milroy & Milroy 1 985:36 1 ) . This is true whether the variant has high prestige in the outside society (/E/ raising) or low prestige (/a / backing; ibid.). The same pattern is found according to the Milroys in a range of anthropological and sociological studies (Granovetter 1 973, 1 982; Rogers & Shoemaker 1 97 1 , cited in Milroy l 992b: 1 7 7- 8 , 1 84). In contrast, individuals with strong ties to each other - the ce ntr a l members of the network - maintain the linguistic conventions of the network, and tend to resist changes from the outside. The weak-tie theory can also account for the fact that it is the social classes in the middle of the hier­ archy that lead linguistic changes: the hi ghe st and lowest classes in a society tend to have the strongest ties in their network, whereas the intermediate classes (upper workin g cl ass, middle class) are more mobile and have mostly weak ties to each other (Milroy 1 992b: l 8 l ) . The wea k tie th eo ry for t he introduction of novel variants into a network or community can be desc r i be d in terms of t h e structure of a society in §7 .2. Indi­ viduals with strong ties to each other form a network and generally are relatively communicatively isolated from other networks. Individuals with weak ties to other networks do not have as strong ties to their own network (s ince some of their links are to other networks). Moreover, they are therefore not as com­ municatively isolated as the mem be rs with strong ties to each other. Weak-tie individuals are significantly exposed to the external variant; they come to use it, and thus introduce it into their own network. The Milroys call such indi­ viduals innov ato rs However, such individuals have not actu al ly created the novel variant (via altered replication); they have replicated it from an outside network and thereby introduced it into their own network. Since I have used the .term 'innovation' for altered rep lica t io n , I w i l l use the term INTRODUCER instead for the Milroys' 'innovator'. The Milroys argue that a strong-tie network functions as a convention (norm) enforcer (Milroy & Milroy 1 98 5 : 363): 'many studies, both urban and rural, have shown that a close-knit network structure functions as a conservative ·

.

-

'

' .

1 80

Selection of innovations in language change

force, resisting pressure for change originating from outside the network' (Milroy l 992b: 1 77). Why might this be so? It may be that being part of a strong-tie network gives an individual a sense of social identity. In contrast, weak ties to different networks do not give an individual a sense of social identity: he or she feels that he or she belongs only partly to any one group, and perhaps that he or she does not really belong to any group. Hence the weak-tie speaker will be more variable in her usage, in particular, using the outside variant in the net­ work to which she n ominally belongs and thus introducing it into the network. I have suggested that a mechanism based on a desire for a social identity operates to select variants from the outside for weak-tie individuals and replic­ ate them in the network to which they belong. Although the social mechanism of selection does not operate in biological evolution, of course, one can find a population structure in biological species that is equivalent to the structure that channels innovations from one linguistic network to another. The phe­ nomenon of contact-induced change in language is equivalent to hybridization, rare in animals but commonly found in plants and other so-called lower organ­ isms (see §8. l ) A common pattern of successful hybridization is INTROGRESSION : 'introgressive hybridization . . . is the repeated backcrossing of a natural hybrid to one or both parental populati ons It results in the transfer of genes from one species or semispecies to another across a breeding barrier' (Grant 1 98 1 :205). The individual with weak ties to two or more strong-tie networks is equivalent to the hybrid plant: repeated interaction with other members of one or both networks leads to the transfer of a lingueme from one social network to the other. .

.

7.4.2

Direction of propagation In populations

The population structure described by the pattern of strong and weak ties in societies locates the channels for variants to pass from one group to another in weak-tie individuals. However, the weak-tie model does not account for the directionality of the transmission of variants, as the Milroys are aware (Milroy & Milroy 1 98 5 : 368; Milroy l 992b: 1 94, 207-20). A further factor must be intro­ duced in order to describe the selective advantage of the variant. This factor is often analyzed as an asymmetric relationship between the two social networks, particularly in class-based approaches such as Labov's. This relationship is usually described as a POWER relationship (Brown & Gilman 1 960/ 1 972; Bro wn & Levinson 1 978/1 987). Brown & Gilman describe power as the ability of one person to control the behavior of the other (Brown & Gilman 1 960/ 1 972:255). The power asymmetry holds between different communities of the same type. For example, the social structure of Western societies generally ranks older over younger age cohorts, men over women, and upper class over middle class over working class in the power hierarchy. Brown & Gilman contrast power with SOLIDARITY, a symmetric relation between individuals equal in the power ranking that ranges from intimate to formal (Brown & Gilman 1 960/1 972:257). Solidarity thus involves social

Propagation and the structure of speaker populations

1 81

distance as well as equal status in the power hierarchy: highly solidary indi­ viduals are both intimate (distance) and equal (power) . Brown & Levinson separate the two variables of power and distance in their discussion of polite­ ness phenomena in general (Brown & Levinson 1 978/ 1 987:74; Brown & Gilman examined one sort of politeness phenomenon, the usage of 'formal' vs 'familiar' pronouns of address in European languages). Brown & Levinson argue that the degree of politeness in language is dependent on power, distance, and the culturally defined degree of imposition of the illocutionary act carried out by the speaker (Brown & Levinson 1 978/ 1 987 :76). Power and distance can vary independently: in addition to intimate and distant equals, there can be intimate and distant unequals (an example of intimate unequals is parent and child). Distance can be identified with tie strength in a social network. Power cor­ responds to PRESTIGE in a class-based model such as that used by Labov (§3 .4.3). The power/prestige asymmetry implies that the variant used by the more pow­ erful community will be propagated into the less powerful community. In most sociolinguistic studies, those of urban varieties in North American and Britain, the variants that are associated with the powerful groups form the standard, while the less powerful grou p's variants are called the vernacular. Hence, many studies document the gradual intrusion of the standard into a l�al vernacular. However, this pattern of propagation is not always the case: speakers some­ times propagate variants from groups that are lower in the power hierarchy. This apparent inversion of the role of power in propagating language change has been called 'negative prestige' (Labov 1 966:499-50 1 ) and COVERT PRESTIGE (Trudgill 1 972; the latter term is how more commonly used). The existence of covert prestige clearly indicates that power/prestige in itself does not define the direction of propagation of variants in communities. Some other factor sometimes correlated with power is the relevant factor. Many sociolinguists analyze this other factor as an ACT OF I DENTITY (LePage & Tabouret-Keller 1 985; see §3.4.3), the desire of a speaker to identify with one social group over another. Speakers wish to identify with a particular group and in so doing replicate the linguistic variants associated with that group. In many cases, speakers wish to identify with a more powerful group, particularly speakers with weak ties to a local network who may have some doubt about their social identity. In other cases, speakers may wish to identify with a less powerful group, particularly speakers in a strong-tie network with a local group speaking a vernacular (recall the observation that strong ties create a social identity that individuals might want to associate themselves with). Hence, the factor determining the direction of change is social identification, not power per se. The theory that choice of a variant is a result of an act of identity with a social group, that is, social identification, can also subsume certain aspects of the theory of ACCOMMODATION (Giles & Smith 1 979; Trudgill 1 986). The theory of accommodation in sociolinguistics was developed by Giles and his colleagues, based on theories in social psychology (see Giles & Smith 1 979:47-53). The theory of accommodation argues that a speaker alters her speech in order to

1 82

Selection of innovations in language change

accommodate to the hearer, leading to the convergence in the form of the speech of the interlocutors. There are in fact two types of accommodation phenomena examined by G iles & S m i th (they do not note thi s d i stinc ti on ) The first type represents adjustments of speech in order to compensate for the fact that the hearer is from a different community (e.g. Canadian vs English in the experiment re­ ported in Giles & Smith 1 979) . These adjustments include slower speech rate, elaboration of content (to adjust to ab sence of certain common gr oun d with the outsider), and simplification of grammatical constructions (Giles & Smith 1 979:45-6). These adjustments do not necessarily involve shifting to the con­ ventions of the hearer's community; instead they represent a style of commun­ ic at io n with individuals outside one' s community . When the adju stm ent s do represent convergence, for instance in the case of language directed towards children or foreigners, the speaker is in fact adjusting to the incomplete and/or hesitant acquisition of the speaker's language by the hearer. The second type represents adj u stment s of speech that result in the speaker shifting towards the conventions of language of t he community to which the hearer belongs. These adjustments include changes in pronunciation, pause and utterance lengths and vocal intensity (Giles & Smith 1 979:46). Of these, the latter two do not represen t conventions of the signal sytem, but rather conventions of conversational interaction, or perhaps even characteristics of j ust the individual hearer. However, shifting one's pronunciation is an instance of shifting the linguistic conventions adhered to by the speaker towards those of the hearer. In this case, the speaker is a dj usting her conventions of speaking in order to identify with the community of the heare r . 5 Another phenomenon that can be accounted for by the principle of social identification is the stylistic dimension in sociolinguistic research. Many socio­ linguistic studies elicit linguistic behavior in a range of interactional settings that are ranked on a sty le h ie rarc hy The settings t yp ic ally used are (from most to least formal style): read in g minimal pairs in a word list; reading a word list; reading a passage; oral interview; and casual conversation (see e.g. Chambers 1 99 5 : 6). Chambers states that 'the essential difference between speech styles is the amount of self-monitoring people do when th ey are speakin g (ib i d ) The results of study after study demonstrate that there is a shift of p ropo r ti on of varia nts in the same direction as is found in groups ranked by social class: the more formal styles elicit a greater proportion of the higher (standard) variants. This shift can be analyzed in terms of social identification. A higher degree of self-monitoring brings greater attention to the language used by speakers. Linguistic awareness taps into attitudes about language, more precisely, atti­ tudes about the social groups using language. In an interview context with a str an g er who is stu dy in g language, speakers will most likely want to identify themselves with the 'correct', that is, standard, v a ri ety and shift their language production accordingly, to the extent that they are able to do so. The result is style-shifting parallel to shifting to the variants associated with the higher social group. .

.

'

.

.

Propagation and the structure of speaker populations

1 83

The existence of covert prestige and acco m m o d a t i o n demonstrates that acts

i de n t ity do not always i mp ly identification with a si n g le social group in a so ciety . The existence of style-shifting further demonstrates th at acts of iden t­ ity are not invariant for individual s peake r s . On p a r ti c u l a r occasions of use, a s pe ak er wi l l want to i d ent i fy first with o n e grou p , th e n wi t h ano t h e r , depend ­ ing on the c onver s at ion al s i tu at i on Finally, acts of identity may be ab le to account for the First Law of Pro p aga­ tion, that s pe ak e rs in a p a r t i c u l ar c om m uni ty tend to c o nverge on a singl e variant for a gi ve n coord i nat i on prob lem , that is, for con v eying a p a rtic ul ar meanin g (see §7 3 ) A re gul a rity in b eh av i o r , such as using a p art i c ul ar li n guem e, does more than convey a particular m ean in g. It a l s o serves as an act of ide nt i ty with a co mmun i ty by the speaker. The best way that a l i n guem e can serve as an act of com m un a l identity is if there is a si n g le variant associated wi th the c om munity . As n o t e d in §7 .3, multiple variants divide community iden ti ty , so to sp e a k , and tend to be resolved by d i ffe re n t ia t i o n of the two conventions, or a b an d onm en t of one co n ve n ti on for the other. Another way of p u t t in g it is that a l i n gu i s t ic co nve n tion s i mu ltane ous ly so lves TWO coordination p ro ble ms : est ab lishing a mutual understanding of the m ean ing to be c on v e yed, and e s t ab l i s h i ng or confirm i ng a c om m un a l identity. E s ta blish ing a co m m un a l id enti ty is also a coordi n a ti on problem among the members of the c om mu n ity ; it suc ce ed s only if the me m be rs of the c o m m uni ty all c onform to it. W h e n a s pea k er uses a l i ngu em e associated with a particular c o m m u n i t y , she is defining her communal id ent ity ( a t least for the p a rti cu l a r c o n ve r sa t io na l e xch an ge ) , as well as b ri n g ing her communicative i n te n t i o n s into the interlocutors' common g r o un d (compare the description of focusing in LePage 1 992:78-9). The con se q u en ce of s o l v in g t he co o rd i nati o n p r o b lem of e s ta b l i s h in g c o m m u n al i de n t it y is c o n ve r ge nce on a si n g l e convention , i . e. the First Law of Propagation. 6 In sum, the factor in language use gr anti n g selective adv ant a ge to an indi­ vidual speaker (and thus to the way she ta lks ) is the desire of hearers who in teract with her to i de n t i fy with the c ommuni ty to which she bel ongs. This selective adv an ta ge ensures the d ifferen t i a l pe r petuat i o n of the re plica t ors she pr o d u c e s , that is, the pr o p a ga t i on of the l i n guis ti c variants associated with the of

.

.

.

communi ty to which she belongs. However, the fact that iden tification with a p a rt i c u l a r

social group varies from s itu a ti on to situation means that n o variant h ave an absolute selective advantage over all other variants. This fact ensures the perpetuation of variation in a language, even as the language cha n ge s in speci fi c d i rec tio n s .

will

7.4.3

The time course of propagation: the S-curve

From detailed qu an ti t at ive h i st o ri c a l l in g u i s t i c studies, it appe a r s that the time c ou rs e o f t h e p ro p a ga t ion of a langua ge change typ ic a lly follows an S - c u rve . Th e s-cURVE h a s been noted b y many researchers (see Kroch 1 989 :203 and Chambers 1 992 : 693-5 for references to e arli e r ob s e rva t i o n s) . However, the

1 84

Selection of innovations in language change

S-curve pattern is not uniform across the grammatical contexts of a particular language change. An equally widely observed phenomenon is that language changes appear to have progressed farther in some grammatical contexts than others. These contexts presumably favor the change for functional reasons (e.g. phonetic reasons for phonological changes), that is, reasons of the sort proposed for innovation (altered replication; see §3 .4.4). This has led to the suggestion that functional factors as well as social factors play a role in th� propagation of change. That is, certain contexts which are functionally favorable are said to accelerate the time course of the change. Kroch ( 1 989) argues that changes that appear to have accelernted in certain contexts have not done so . A unitary language change follows the same S­ curve in all contexts, that is, the S-curve has the same slope, which corresponds to a constant value in the logistic function that mathematically describes the S-curve (Kroch 1 989:204). The difference in absolute frequency of the novel variant in different contexts is due to either a difference in the initial frequency of the novel variant in the different contexts, or a different starting time for the innovation in the different contexts. Kroch offers one piece of evidence, from the history of periphrastic do in English, that the former scenario is the correct one. Periphrastic do is used least frequently in affirmative declarative contexts, compared with negative and/or interrogative contexts. Yet since affirmative declarative contexts are the most common ones, occurrence of periphrastic do in its least-favored context is observable from the beginning of the change. In fact, periphrastic do occurs in affirmative declarative contexts from the begin­ ning of the change (Kroch 1 989 ; 230). Kroch's analysis has two important implications. First, it suggests that func­ tional factors are involved at only the innovation of a change, determining the relative frequency of the novel variant in different contexts: the pattern of favoring and disfavoring contexts does not reflect the forces push­ ing the change forward. Rather, it reflects functional effects, discourse and process­ ing, on the choices speakers make among the alternative available to them in the language as they know it; and the strength of these effects remains constant as the change proceeds. (Kroch

1 989 :238)

If the functional factors do indeed represent constant facts of human commun­ icative interaction, then one would expect them to remain constant over time. These observations confirm the hypothesis in §2.4.2 that the propagation of a change is purely social, and its innovation is functionally motivated. Second, the S-curve pattern must represent some fairly fixed and regular process, if a speaker is able to maintain the rate of propagation of a change in differe.nt contexts even when those contexts have different absolute frequencies of occurrence of the novel variant. We will return to this point below. Kroch points out that the logistic is only an approximation of the time course of a language change. The logistic function has a zero value (the abso­ lute beginning of a change) at -oo, and the value of l (the completion of change) at +oo. Kroch writes, 'Of course, actual linguistic changes have starting

Propagation and the structure of speaker populations

1 85

and endi n g points, so the model can only approximate real data, and this appro x im a tion falsifies the ch an ge process preci s el y at the begin n in gs and ends of changes' (Kroch 1 989:204). Kroch is correct of course, but nevertheless it should be noted that the beginnings and e n din gs of changes are farther away than one might think. When does a language change begi n? The basic premise of this book i s that language change consists of two processes: altered replication of a replicator, i.e. innovation; and differential replicat i on of repl icato rs leading to survival/ extinction, i.e. propagation. That is, change innovation + propagation. The terminology in sociolinguistics and historical l inguistics is quite confusing on this, as noted in § 1 .2. For instance, Kroch appears to describe innovation as the beginnin g of the propagation phase, not the initial production of altered replicators: 'at the begin ning of a change p [the frequency of occurrence of the change] jumps from zero to some small positive value in a temporal discon­ tinuity, which Weinreich, Labov and Herzog ( 1 968) dubbed the "actuation" of the change' ( Kroch 1 989:205). Milroy uses innovation in basically the same sense as I do in this book, but describes propagation as 'change' (Milroy 1 992b:20 1-3), and uses the term 'innovator' to describe the person in a net­ work who gives a change a social value (th i s is why I have chosen the term 'introducer' instead). Milroy, however, recognizes that there are two distinct stages in the process. The first stage, innov atio n proper, is the isolated production of altered repl icators by speakers, for the reasons described in ch apter 5 and §§6. 1 -6.2, before they have acquired any social value. As noted in §3.3.2, close analysis of both his­ torical corpora and sociolin guistic field recordings reveals that random occur­ rences of an altered replicator occur, sometimes for a very long period of t ime , before the altered rep licat or acquires a social value and thus takes off, that is, starts to be propagated . If we treat the period of random altered replication a s the begi nn i ng of th e S-curve, as Milroy does (Milroy l 992b:202), then the begi nn i ng may extend quite far back in time. Conversely, one cannot assume that a chan ge ever entirely ends, if the re is a residue of relic forms or vesti gial variants (Trudgill 1 999). It may be that the relic forms will, over a long period of time, eventually be s ubjected to the chan ge in proce ss . The chan ge cannot be said to have been completed until every last relic form i s gone. One reason why a change appears to end abruptly is that other chan ges interrupt its progress. For example, Kroch notes that the process of change for periphrastic do undergoes an abrupt change of direction in the mid 1 6th century: what appeared to be a unitary change of do in various grammatical contexts breaks up, with do goin g in different directions at differ­ ent rates in different contexts (e.g. affirmative declarative do declines at this point). Thus, the tail of the S-curve may be very long but never is achieved due to other interfering changes in the language. In other cases, the vestigial vari­ ants survive in surprising ways, as stereotypes of older or more traditional speakers, in remembered phrases, in passive community knowledge of the ves­ tigial variant, and in the sporadic occurrence in one or two unusual speakers =

l 86

Selection of innovations in language change

(Trudgill 1 999). In sum, the S-curve may fit historical linguistic facts more closely th an Kro ch su gge sts How does a random process acquire a social value? Innovations at their source do not conform to pre-existing conventions of speech and may also be nonintentional (see §3.4). One innovative utterance does not entrench the in­ novation, even for the producer of the innovation . One cannot assume that a speaker's gr ammar has changed just because the speaker has produced an innovation . The innovations must be reinforced by use in order to become part of the mental knowledge of a speaker. 7 If innovations are perceived as lacking social or stylistic value, then they are perceived as 'errors' (and often described as such). A simil ar ar gume nt to that put forward in § S . l can be used to describe the transition from innovation to propagation . In §S. l , I argued that the experi­ ence being communicated by the speaker i s very complex, and so is the rela­ tionship between the experience and the grammatical structures used to convey it. But the complexity does not end there. Speakers are trying to achieve many other things besides communicating inform at ion As Keller say s : .

.

Under no rmal circumstances, we do not cho o se our linguistic means according to

exactly one maxim . When we are talking, we try to kill several birds wit h one stone: we try to conform, attract atten t i on , be understood, save e ner g y . It is

extremely rare that someone wants nothing but to be understood . (Keller 1 990/ 1 994: 1 05)

M oreover, the goals that the speaker is trying to achieve differ from context to context (ibid. 1 06) . The hearer must disentangle the complex matrix of social intentions of the speake r from each other and fro m the intention of the speaker to be under­ stood (that is, to be understood to convey the information she intended to communicate) . It is thus quite possible that an innovation, which does not adhere to prior conventions of the community, is reanalyzed by the listener as a social indicator (in Labov's sense of that term). At that point, the innovation passes fro m an ' error' - that is, a form l ac ki n g any social value - to a soc ial ly defined variant in the mind of the hearer. The hearer later replicates the innova­ tion in another context, intending it to be understood by his interlocutors as having a social value. If his interlocutors understand his intention in this context, then the propagation of the innovation has begun . This descrip tion of the transition from innovation to p rop ag ation applies most clearly to genuinely novel variants produced by speakers. When an intro­ ducer produces a variant from one social group that is new to her own social group, often that social group is already aware of the existence of the variant in the other social group and thus the variant already has a social value (e.g. it i s the standard form) . N everthe less, it is possible for the social identity of an introduced variant to be reanalyzed by the social group to which it is intro­ duced. Two of the variables studied by Milroy and Milroy, 1£1 raising and /a/ backing, have their source in rural varieties of Ulster Scots (Milroy & Milroy

Propagation and the structure of speaker populations

1 87

1 9 8 5 : 3 54, 358). Yet ' a comparison of the diffusion mechanisms and distribu­ tion patterns of /a/ and le/ in Ulster shows that elements originating from the same (rural) dialect can take on, apparently arbitrarily, entirely different social values in their new urban context' (ibid . 38 1 ). 8 Proceeding beyond the inception of propagation , at least the introduction of a variant by weak-tie individuals, Milroy and Milroy propose that the next stage of the propagation process is use of the novel variant by individuals who they call the EAR LY ADOPTERS (Milroy & Milroy 1 98 5 :367, taken from Rogers & Shoemaker 1 97 1 ). Early adopters are more central members of the social network, linked by strong ties to the other central members of the social group. Milroy & Milroy further propose that since early adopters are central members of a close-knit group, and thus are predisposed to maintain the network's conventions rather than shift, they must be exposed to many occurrences of the novel variant, presumably on the part of introducers, in order to adopt it (Milroy & Milroy 1 98 5 : 368) . Once early adopters adopt the novel variant, it can then spread rapidly through central members of the group, representing the middle, rapid increase, stage of the S-curve . Finally, the novel variant will diffuse to other peripheral members of the society (Milroy l 992b : 1 84), rep­ resenting the final stage of the S-curve . There are two problems that the Mi l r o ys analysis of the time course of a change must address, both based on the assumption that the S-curve pattern of the time course of a linguistic change is regular. The first problem is that there are also societies that are characterized by large numbers of weak ties and relatively few strong-tie netw orks. Typical examples are larger political­ geographic units with a great deal of geographical and social mobility (see §7.4.4). The Milroys state a principle governing the differences in patterns of change in the two types of society: 'linguistic change is slow to the extent that the relevant populations are well established and bound by strong ties, whereas it is rapid to the extent that weak ties exist in populations' (Milroy & Milroy 1 98 5 : 375). Yet presumably the S-curve pattern applies to both types of society. In fact, all of the examples of S-curve patterns described by Kroch are of lang­ term changes found in a wid e range of written records, probably representing a relatively cosmopolitan, weak-tie society One possible solution to this problem has to do with the slope of the S­ curve. Kroch's examples demonstrate that although all of the changes he de­ scribes follow the S-curve, the S-curve itself may be stretched over different lengths of time, represented by differences in the constant s, the slope of the logistic function (Kroch 1 989:204). It could be the case that societies with a greater proportion of strong ties will h ave a l ow slope to the function, meaning that the change takes longer to be propagated, while societies with a greater proportion of weak ties will have a high slope to the function, meaning that the change takes less time to be propagated. Such a hypothesis can be empirically tested . However, there is a second, more in tr act abl e problem with the Milroys' analysis. Kroch presents evidence that different frequencies of a novel variant '

.

,

1 88

Selection of innovations in language change

in different con te x t s re p re se n t the sa me c ha n ge at different stages of the S­ cu rve in the different contexts. If Kroch 's a n a lysi s is correct, then the s ame language ch an ge is at different st a ge s of the S-curve for th e same s pe ake rs . Yet the Milroys' model suggests that d i ffe re n t stages of the S-curve are associated with different speakers, d efi n ed by t he ir position in the social network (i n t r o­ ducers, e a rl y a dop te rs, o ther cen tra l memb e rs oth e r periphe ral m embers) . These two m odel s of t he S- cu rve are i n co n s iste nt . Labov has proposed a mechanism of chan ge underlying the S-curv� th a t is co mp a tibl e with Kroch's model (Labov 1 994:65- 7; Labov does not mention Kroch's model in this context). Labov models the S-curve with a binomial distribution, wh i ch like other S-curve functions 'can be generated by a model in which t he p rob a b il ity of contact between the two [variants] governs the rate of c h ange ' ( Labov 1 994:66) . To put it in evo l u tionary terms, Labov argues that the rate of ch an ge is a function of the pro po rtion of th e l ingueme v a riants av a i la bl e to e ach sp e ake r b a se d o n t he h i s t o ry of her communicative interac­ tion with other s peak ers . Labov also a ssume s a re l a tiv el y low selection pressure and h ence a sm a l l shift in variant frequencies in the l in gueme pool at each stage of the process (ibid .). Thus, if interlocutors in a s peech situation use d ifferen t variants of a l in gueme there is a small constant prob a bi li ty that there will b e a shift toward s the v a ri an t associated with the social group fav o red for an act of iden t it y on the part of the in terl ocutor s in fu ture replications of the lingueme. Summin g ove r t he tota l i ty of such c om m un icati o n e nco u n t er s y i e l ds the S-curve. At the b e g i n ­ ning of the ch a n ge few speakers are exposed to both variants and in re l at iv e ly few i n teracti on s ; so the change in variant frequencies is slow. In the middle of t he change, most speakers a re exposed to most variants in many interactions, and the ch ange in variant fre q ue n c i e s is t he re fore fastest. At the end of a c h an ge ag ai n few speakers are exposed to both v ari an t s and in re l a t ive ly few i n t er act i ons ; so the change in vari an t frequencies (i n this case, eliminating the last occ urre nces of the losing variant) is again slow. The last question to be addressed in th i s section is the nature of the chan ge that procee d s by an S-curve. Kroch 's ex amples are all cases of g rammatic al ch an ge occurrin g in a v ari ety of grammatical c o n texts . As we have see n , K roch' s an alys is allows for two possible sce n a ri os: t h e gramm a t ical change starts at d ifferent times in different gr am ma t ical contexts, but follows the same S-curve; or th e grammatical change starts at the same time in all gramma tic al contexts, but at d i ffe ren t ini tia l fre q uencie s, and each context follows the same S-curve from its initial frequency. T he first scenario wo ul d co rre spon d to grammatical spread of a change from one c on text to another; the second to simul taneo u s ch an ge (but with d i ffere n t favoring factors in d iffe re nt contexts) . Kroch argues for the second scenario for Engl is h pe rip hrast i c do up to the mid 1 6th ce n tury . On the oth er hand, grammaticalization is a lw ays c h aractf" ized as a s pre ad i n t o new gramma t ic a l con te x ts, typi ca lly with a fairly abru, : sem an tic c h an ge (such as met a nalys i s) on the part of t he constructi o n un de r­ go i n g gram m at icaliz a ti o n . However, this d escr i pti on of grammatical iza t i on is ,

,

,

,

,

·

Propagation and the structure of speaker populations

1 89

orthogonal to the sort of grammatical chan ges described by Kroch. The time course of a change is the selection process. In analyzing the time course of a grammaticalization process, one must compare the competing variants. For instance, in examining the time course of the grammaticalization of since to a causal connective, one must examine the selection of since over other causal connectives of English. The emergence of causal since from temporal since is an instance of a branching lineage via altered replication, not an instance of selection . In phonology, however, there is a long-standing debate over two alternative implementations of the S-curve, neogrammarian change and lexical diffusion. NEOGRAMMARIAN CHANGE p roceed s by incremental phonetic shifts in all words in which the phoneme occurs. Proponents of LEXICAL DIFFUSION argue that sound change proceeds by abrupt phonetic changes but gradually across the lexicon. There appears to be a consensus that both neogrammarian and lexical diffusionist changes occur, but some argue that lexical diffusion is extremely widespread (Wang 1 969; Chen & Hsieh 1 97 1 ; Chen & Wang 1 975; Wan g 1 977; Chambers & Tru dgill 1 980: 1 74- 80; Chambers 199 2: 694 - 5 ; Milroy 1 9 9 2b : l 6 1 2 ; and other references cited in Labov 1 994:438), while others argue that neo­ grammarian changes are the more widespread (Labov 1 98 1 , 1 994: 4 1 9-543). It is also possible for a change to be both lexically gradual and phonetically gradual. Chambers & Trudgill describe transitional areas between the northern English and southern English varieties, involving the variables (u) and (a) ( Chamb ers & Trudgi l l 1 980: 1 32-42; Trudgill 1 986: 59-62). Some transitional areas are acquiring the southe rn variant of (u), [A], in a lexically gradual but phonetically abrupt fash i on , i.e. lexical diffusion . These are called MIXED LECTS. Other areas use a phonetically intermediate value, [Y], but also in a lexically gradual fashion; these are FUDGED LECTS (see §2 .2, §6. 1 ). Fudged lects are thus both lexically and phonetically gradual (Trudgill 1 986:6 1 ) . Bybee (2000) also argues that t/d-deletion in English is both phonetically gradual (duration of lingual gesture) and lexically gradual (dependent on the token frequency of individual words) . We can describe both types of sound changes in the evolutionary frame­ work . Linguemes exist at sev era l levels of inclusiveness; both phonemes and lexemes are linguemes. Phonetically gradual change is altered replication of a phoneme lineage, and lexically gradual change is a l te red replication of multiple lexeme lineages. Once again, however, we must identify the competing variants in order to characterize the selection process in neogrammarian and lexical diffusionist change. Both types of sound change are gradual in another sense. Whether the ch an g e is phoneticaJJy gradual or lexically grad ual (or both), both the older and newer variants can occur in the speech of the same speakers (Trudgil l & Foxcroft 1978/ 1983). Selecti on occurs in the choice of the older or n e we r form, wh et he r that form is a phoneme or a le xi cal i tem. It is more disputable whether the selection mechanism is social, of the sort described in §7 .4.2, in both types of sound change. Labov argues that neo­ grammarian changes are correlated with social variables, but that there is no

1 90

Selection of inno vations in language change

social conditioning of lexical diffusion (Labov 1 994: 527). Milroy on the other hand argues that the shifts in Belfast /a/ and other variables involve lexical diffusion (Milroy 1 992b: l 6 1 ), but they are also associated with soc i a l network structure (ibid. I 09-22). If so, then the same selection mechanism is involved in both types of change. The major difference is that the selection process operates one word at a time in the lexical diffusionist approach, selecting one variant over the other, and also word by word, such that some words co mp l ete the change before other words. Lexical diffusionists have argued that token frequency of words plays a major role in diffusion of a change word by word, t h a t is, across the lexicon (e.g. Phillips 1 984; but see Labov 1 994:483-5) . If so, then word token fre­ quency must be posited as a mechanism of selection. However, in t rackin g the course of the change, word token frequency may be an a l yzable in the same way as Kroch analyzes the differential gramm atica l conditioning in grammat­ ical changes. (I assume that Kroch would analyze the differential phonetic conditioning of neogramm a ri a n sound changes in the same way.) That is, one of the two scenarios applies. The first scenario is that word token frequency governs the initiation of the S-curve, but all words follow the same S-curve. The second scenario is that token freq uency of a word may govern initial p rop ortion a l frequencies of the novel v ari a nt but when the change takes off, a l l words follow the same S-curve . However, the lexical diffusionist analyses cited above do not provide quantitative detail for individual word forms to test Kroch's hypothesis. It would however be a striking confirmation of Kroch's analysis if lexical diffusion of sound change followed a similar pattern to gram­ matical change. ,

7.4.4

The macrostructure of speaker populations and propagation

The preceding sections dealt with what could be called the microstructure of speaker po pul a ti o n s : the relations holding between individuals in relatively small populations. Historical dialectologists have observed significant patterns at the mac rostru ct u re of speaker populations, representing large geographical regions with large populations . The main pattern observed resembles a widely observed pattern in evolutionary biology, and may have a s imi l ar ex p l a n a ti o n. The main pattern observed by historical dialectologists was formulated by M a tteo Bartoli in the following two principles (Be rto n i & Bartoli 1 925, cited in Manczak 1 98 8 : 349) : 9

( 1 2)

I.

II.

The more isolated area usually preserves the earlier stage. If one of two linguistic stages is found in peri ph e ral areas and the other in a central area, the stage occurring in the peripheral areas is usually the earlier one.

Manczak argues that the second principle has been improperly formulated . Using lexical evidence from Ita l i an French (central areas), Spanish and Roma­ nian (periphe ral areas), he argues that innovations can occur in peripheral ,

Propagation and the structure of speaker populations

191

areas a s much as i n central areas (Mailczak 1 988). This i s because peripheral areas (unlike completely isolated areas) may innovate by virtue of contact with other languages (ibid .). What appears to distinguish central from peripheral and isolated areas are the types of changes that occur, not whether language change occurs. In cent­ ral areas, there is a ten d en cy towards leveling of distinct forms used in the geographical communities making up the central area. Leveling may involve adoption of one geographical community's conven tion by the other commun­ ities in the central area, or it may involve convergence onto a new convention that is in some way a compromise, typically a simplification, of two or more communitie s ' conventions. This process, now called KOINEIZATION because it represents the creation of a koine for communication in the central area, was proposed by Jakobson 1 929/ 1 962, and is documented by Andersen ( 1 988) and Trudgill ( 1 986, 1 989, 1 996a) . In this cespect, the retention of archaic features in isolated areas probably represents the nonparticipation of the isolated com­ munities in the koineization and diffusion of novel variants arising i n the cen t­ ral area. Andersen ( 1 988) demonstrates that in contrast, in peripheral and isolated areas, certain changes occur which rep resent a distinctive, often more complex, grammatical structure . Andersen offers evidence of phonological elaboration, including tpe development of palatalization from a secondary articulation to separate glides and then independent fricatives in peripheral areas of Polish, and the occurrence of parasitic consonants in isolated and/or peripheral di a­ lects of Rornantsch, Proven�al, Frenc h, the Dutch-German area and D enm a rk (Andersen 1 988: 56-66). These processes are all phonetically motivated, and so represent natural phonetic de vel o pme nt s taken to a high degree of elaboration. Trudgill has documented further examples in papers leading to a forthcoming monograph on language in isolation (Trudgill 1 989, 1 992, 1 996a). The equivalent p henomen on in evolutionary biology is the fo under popula­ tion theory of speciation, originally proposed by Mayr in 1 942 (Mayr 1 982:6006). Mayr argued that speciation, that is, the divergence of a population to the point that it is reproductively isolated from the rest of its species, occurs most fre q uen t ly in an isolated p opul a t i on , the foun der p opulation . Later res earch suggests that in fact it is small populations, not simply isolation , that is th e primary engine for speciation: 'There is little doubt observationally that rapid specjation is most easily accomplished in very small populations' (M ayr 1 982:603) . In contrast, large populations tend to be evolutionarily inert: 'new alleles, even favorable ones, require very long periods of time to spread through the entire species range . Genetic homeostasis . . . strongly resists any changes in a large, undivided gene pool ' (Mayr 1 982:602; see also Grant 1 9 8 1 :24-6). The reason for this pattern .appears to be that the isolated population 'con­ tains only a small fraction of the to ta l genetic variability of the parent species' (Mayr 1 982:602). This fact allo ws for drastic chan ges at the genetic level (not fully understood; ibid. 605-6) which would not occur in the large parent popu­ lation. The genetic changes in turn lead to significant changes in the phenotype

192

Selection of innovations in language change

of the offspring of the founder population, and eventually speciation in the successful cases. The geographical patterns for language change and speciation are similar except in one apparent respect: changes occur in the central areas of a lan­ guage, while change does not occur (or occurs only quite slowly) in the large central population of a species. The difference can be accounted for in looking at the explanation for the linguistic pattern. The linguistic pattern as described above represents existing smaller populations that are in relatively high contact with each other (Trudgill

1989,

1996a), not only through geographical contigu­

ity but also by means of greater socioeconomic interaction in general (note that most of the studies of this phenomenon apply to European languages repres­ enting large-scale political and

economic entities). It is the linguistic result

of the high level of contact - the levelled or simplified variety - that should properly be compared to a large biological population. And the leveled or simplified variety is relatively more stable (other than further simplification or leveling). Trudgill argues that the explanation for the linguistic pattern is not merely high contact but also loose-knit social networks, that is, weak-tie networks (Trudgill

1992,

1996a). These two social phenomena frequently coincide, as

noted in §7.4.1 (see Trudgill

1996a : 4).

In fact, high contact on a large scale

would be extremely difficult to maintain. The opposite combination, low con­ tact and close-knit (strong-tie) networks, also frequently coincide; these are peripheral or isolated areas (Trudgill

1992:201;

1996a:4,

11). Moreover, a close­

knit network is of necessity small in size: it is difficult for everyone to know and talk to everyone else in a large society. Large size societies tend to be loose-knit overall, although, of course, there may be close-knit networks within the society - and, as one might expect, the latter tend to be socially marginalized and hence peripheral groups (lower class, street gangs, etc.; see §7.2). The significant factor with respect to high contact is the degree of adult second-language (or second-dialect) learning involved (Trudgill

1996a).

1992: 197-9;

In most high-contact situations, there is a high degree of adult second­

language learning, which contributes to the leveling and simplification observed in those situations. Some close-knit societies may also have extensive contact Trudgill gives the example of the native American languages of the Pacific Northwest and the Caucasus - but they also have bilingualism from childhood, and hence less adult second-language learning (Trudgill

1989:234).

The small, close-knit networks are those in which linguistic conventions can

be maintained in resistance to outside contact, which is typically relatively minimal anyway. Trudgill also proposes that isolated close-knit networks are those that allow a grammatical change to follow through to completion, lead­ ing to the elaborated structures that seem to be characteristic of such groups 1988; Trudgill 1996a : 8) including phonological fortition processes

(Andersen

,

and complex allophony and allomorphy

.10

In sum, linguistic changes that are difficult for the nonnative speaker assimilation, fast speech phenomena, longer words - and changes that are

Propagation and the structure of speaker popl!lations

193

more lik ely to occur or survive in close-knit networks - e la borate c h an ges , allophonic complexi ty , complex ph oneme segment inventories - will be found in small i solated socie ti es (T rudgi ll l 996a: 12). Conversely, lin gu istic changes that are e as y for the nonnative speaker - mergers, simplification, sh orter words and ch anges th at ar e m ore like l y to su rvive in loosek nit networks - so-called 'unmarked' or 'natural' grammatical structures - are more likely to be foun d in l ar ge societies characterized by high contact. The equivalence of the l in gu is tic and biolo gical patterns should now be cle are r. Gene flow throughout the biological populati on ensures that the genetically prev alen t al leles remain p rev alent in the entire populati on, l eading to rel a t iv e homogeneity and stasis in a large pop ul ation . Likewise, high con tact leads to compe t i tion between lin gueme variants, and the resulting leveled vari­ ant then predominates over

the central are a . Only a subset of alleles is found in

a small, i sol ated fou n der popul at ion of biologi cal or ga ni sms , and significan t

changes can then occur. Likewise, lin guis tic isolation aJiows for pro­ to evolve to an elaborate deg ree that would otherwise be cur­ tailed by le vel i ng or simplification in a larger, more loose-knit soc iety. In additi on to the geograph ical linguistic pattern which paralle l s the geo­ gr a ph ical bi o l ogical pattern, there is a social pattern of the propa ga t i o n of a l an guage change that follows from the community structure of a society. This is the observ atio n that diffusion (pr opagation ) of novel variants j umps from popul ati on center to population center. Trudgill (1974/1983:52-87) gives evid­ ence of th is pattern with the diffusion of uvular /r/ acr oss the l angu ages of

genetic

cesses of ch an ge

Western Europe, diffusion of changes to lad on the Brunlanes peninsula in

southern Norway, and the replacement of interdental fricatives by labiodental ones in Norwi ch ; Chambers and Trudgil l give a further example of /re/ raising in northern Illinois (Chambers & Trudgill 1980: 189-92). Uvular /r/ appears to have originated in Paris, but it then

jumped to other well as diffusing to the rest of France in the usual geographical pattern. The pattern is clear when on e looks not simply at the presence of uvular /r/, but its presence in some educated speech, most educated speech , and u ned ucate d speech (Trud g i l l 1974/1 983:58, map 3.3). This div i sio n demonstrates that uvular /r/ is spreading from educated speech to uned ucated speech partially i n depen de ntly of geography. Likewise, diffusion of ch anges to /re/ jumps from city to city in B run lanes ; Norwich c hanges in fricatives originate in London and are transmitted to Norwich, and the per­ ce n tage of /re/ raising in northern Illinois is a function of the size of the town as well as d istan ce from Chicago (the origin of the in novat ion) . Trudgill uses a formula from geography, the gravi ty model, which measure s influence as a function of both population size and geographical distance of the influe n ci n g cen te r(s) from the area potentially being infl u enced (Trudgill 1974/ 1983:73-4). It appears that the gravity model fits some data very well (Cham­ bers & Trudgill 1980:200-2). Nevertheless, it c an not be simply the size of the center that determines l i n guistic influence. There has to be commun icative

major Western

European

cities,

as

interaction between speakers in the centers which jumps across intervening.

194

Selection of innovations in language change

areas (or at l e ast is less pronounced between the center and the intervening areas). Fo r exampl e, the influence patterns in Brunlanes have cha n ged over time, app a re n t l y due to a shift in imp o rt a nce from sea to land transport links (Trudgill 1 974/ 1 983:80-2). It seems clear from these examples that there is communicative interaction among members of the same co mmuni ty (the educated elite, or city dwellers in general) across geographically se para ted centers at least as much as there is . communicative interaction between members of different communities (the elite vs other classes, the urban vs the rural) in th e same geographical area. This pattern is a manife st a t i on of one of the basic differences between linguistic communities and biological popul a tions: an o rgani sm belongs to only one deme, defined largely geogra phical ly (within its e co l o gica l niche), whereas a spea ker b elo ngs simultaneously to mult iple communities, defined geographically, socially and in other ways (§7 .2). Th e analysis of the s t r ucture of a spe a ke r population is in fact richer and more c omple x than that of a bio lo gical pop ulat ion As will b e seen in the next chapter, this is also true of the p h ylogenet i c structures describing the descent of languages. .

Notes The numbers co rres pond to the numbers indexing the various degrees

( l 969:79).

d;

in Lewis

I have left out two further dimensions by which a signal could be a

convention to some degreed: the deg ree of occurrence of the recurrent coordination

problem for wh ich the signal is used

(Lewis' d0),

and the degree to whi ch an indi­

vi d ual cannot simultaneously conform to the reg ularity in questio n and an altern­ ative regularit y (Lewis' d5). 2

3

The term

aeronef,

first mentioned in 1844 (Guilbert 1 965:46), was also used as both

masculine and feminine at first ( ibid .

47).

The first two ways resemble one of t he processes of selection between spe ci es compet ing for the same ec ological niche (biotic sympatry): the t wo species may survive by ecological differentiat ion , that is , specializing into slightly different niches

(Grant 1981: 119-20). This is select ion at the level of species in biology , whereas I am d escr i bing selection at the level of organisms (spea ke rs) . However, recall that Hull's generalized theory of selection is independent

of

levels at which selection

occurs (§2.3). It so happens that a mechanism operating at the interspecific level in bio l ogi cal evolution is si milar to two mechanisms operating at the speaker level in

4

lan guage evolut i on

I

.

am using 'society' as it is defined in §7.2; as noted t here, the term 'community' is

gene rally used for this construct by sociolinguists. 5

In fact, this may not always be welcomed by the hearer. In the experiment with

English subj ects evaluat in g a Canadian's speech repo rt ed in Giles & Smith 1 979, for instance, shift in pronunciat ion (gen uine convergence) was evaluated negatively by subject s, while shifts in speech rate and content (the first type of accommodati ng behavior) were evaluated positively (Giles & Smith 1979:60). 1n·other words, social iden tification is n ot always id entifi cation with the hearer ' s community , at least with respect to conformi ty to linguistic norms.

Notes 6

195

There is one significant exception to the First Law of P r opagation In some abori­ ginal cultures in Australia (see for examp le Heath l 981) and New G ui nea (Bernard Comrie, personal com m uni cat ion), words identical or similar to the names of re­ cently decease d persons are tabooed for some period of time. One of the st rate gie s for deali ng with the dea t h taboo on words is 'to maintain e la bo rate stocks of synonyms for common nominal c o ncepts , so that one no un may be temporarily .

taboo without creating problematic lexical gaps' (Heath 1981:361). Clearly, this is a c ultura l context in w hic h the ex istence of sy nonymy

has a pos itive value. How­ death taboo in some cultures does not invalidate the First Law of Propagation in other social contexts and in other societies. (I thank Be rnard Comrie for bringing this fact to my attentio n ) It is possible th at an innovation is particularly s alient in some respects and may thus become part of the linguistic system qui t e rapidly. For example, when I innovate a word such as intraference for t he first time, it can become a part of my lin guistic system in a re l a tively small number of replications (uses) . T he re still remains the question of how a genuine innovation, not a variant from a neighboring social group, is taken up and propagated. One plausible hypothesis is that only the innovations of weak-tie members have a chance of being propagated ; innovations by stron g- tie members are not propagated because of the stren gth of the exi sting conventions of language in the strong-tie group. One migh t also argue that innovations by strong-tie members would be pro pagated precisely because they be gan in the center of the network. This may be true of explicitly legi slated changes by the power elite; h oweve r these changes are a tiny minority of the changes that ever, the ex is tence of the

.

7

8

,

.

occur. 9

Bartoli formulated

five principles;

he later

withdrew the fourth and fifth

principles.

The th ird p rinc ipl e , that earl ier " forms are found in larger areas, does not seem to . hold, based on the evidence discussed in this

10

section. fact that certai n social s itua tio n s favor t h e opposite sort of linguistic chan ges that are considered to be opti mal or 'natural' on functional grou n ds is additional evidence that functional considerations do not play a ro le in selection (propagati on) Instead, the more com plex older forms, or elaborated 'dysfunctional' innovations, are propaga ted through the speech community

The

'

'

.

.

Chapter 8

The descent of languages

8.1

Phylogeny in biology and language from a plantish point of view

In a num ber of places (e.g. Hull 1988:416), Hull compl ai ns of the zoocentric orientation of much thinking about ev o luti on (e specially views of evolution outside biology). We have already encountered o ne context where lo oking beyond animals to plants has had a libera ting effect: recognizing that levels of o rgan izati on in biolo gy are not well defined, and h e nce they sh ould be se parated from the generalized theory of selection (§2.3). This chapt e r will illustrate another aspect of the evolutionary framework w here taking a plantish point of view offers some insights into the nature of langu age evolution, namely ph yl o ge ny or the historical descent of species. The zoocentric view of phy l o geny is that all new species come into being by d i v e rg e nce . A speci es is a popula tion of interbreeding indivi d u a ls that is reproductively isolated from other species . The p o pul ation itself usually is d iv i ded into races and demes, which are defined by partial reprod uct i v e isola­ tion (i.e. a lower probability of interbreeding) from sibling races or demes. O ver time, races can become more reproductively isolated. O n ce the former race is more or less fully reproductively isola ted, it has diverged to the point of becomin g a new spec ies . The zoocent r i c view of ph ylo gen y correspon d s to the family tree model of l anguage families in l i n gu istic s . A language (a popu l at ion of utterances) is produced by a p o pu l ation of spe a kers who talk to each other but are commun­ icatively isol ated from oth er l anguage speakers. The lan guage , or more pre­ cisely the popu latio n of spea k ers, can be divided into geog r aph ical dialects, more precise l y subpopulations of speakers, wh i ch are defined by parti a l com­ municative isolation from si b ling dialects. Over time, dialect speaker populations can become more c ommuni ca tively isolated. Once the former dialect subpopula­ tion is mo re or less fully communicatively isolated, it has diverged to the poi nt of be c omi ng a new l angu age . Patterns of divergence lead to tre e structures in phyl o genies . E very s pe cies has a s i ng l e ( former) species as a parent. Every language has a singl e (former) language as a p a rent. But this mo d el has ofte n been chall e nged in l in g uisti c s . Some changes diffuse across language and dialect boundaries, ignoring bound­ a r ies of past divergence. This pattern i nspired the wave model in 19th�century histo ri ca l lingu ist i cs , w here lingui stic features move thr o ugh a set of lan gu ages/ 196

Phylogeny in biology and language

l 97

dialects to different degree s of ex tent and in different directions, leadi ng to div­ varieties. Also, the results of lan gu age contact - bo rrowin gs and substratum structures - ch al len ge s the integrity of langu a ges as h is tor i ca l entities originating from a single parent. Finally, pidgins , creoles and mixed lan­ guages challenge the family tree model of l a n g u age birth by divergence in a number of ways. The facts of l an g u age contact and its resu l ts appear to challenge any attempt to develop an evolutionary model for l angu a ge s based on speciation via t he divergen ce of populations in biology. One might be a ble to ignore bo r row ing and su bstr atum structures if the bulk of the linguemes of the la n guage have lineages traceable to a single parent . But this is much more difficult to do in the case of p id gin s, creoles and mixed l an guage s . The pro b lem s with the latter sorts o f lan gua ges have led Thomason & Kaufman to argue that such languages simply d o not have any ancestors, that is, th ey do not have a place in the phylogeny of languages:

ergent linguistic

a claim of genetic relationship entails systematic correspondences in all parts of the language because that is what results from normal transmission: what is tran smitted is an entire language - that is,

a

compl ex set of interrelated lexical,

... a language can not of normal transmission. To be su re , mixed

phonological, morphosyntactic, and semantic structures

have multiple ancestors in the

course

languages in a nontrivial sense exist, but by definition they are unrelated genetic­ ally to the source(s) of any of their multiple components. (Th o m a son & Kaufman

1988:1 1 , emphasis original)

Thomason & Kaufm a n make two assum p ti on s, neither of which is valid in the evolutionary framework for language chan ge proposed in this book. 1 First, Thomason & Kaufman's th eory of p hylo genetic relationship assumes an essen­ ti al ist theory of a l angua ge , nam ely, that it has a c oo rd ina ted set of structures, phonological, morphosyntactic and semantic, s uch that if its apparent descend­ ants are lacking in enough of these structures, then it no l onge r can be consid­ ere d a member of that language group. To be precise, Thomason & Kaufman do not argue that there is a specific linguistic trait that is essential; instead they argue p rob abili stical ly that enough linguistic traits in en o ugh aspects of the language are replicated in normal tran smiss ion to classify that l an gu a ge in the gene t ic fami ly in que&tion. Still, this is a we a ker variant of an e s se nt ia list model. In the pop ul a tion model ad op ted in this book, the li n gueme pool can ch an ge radically without a l angu age ceasing to be a ge neti c descendant of its pa ren t. Or parents. The second assumption made by Thomason & Kaufman is that the fa m ily tr ee model is the only correct one for describing (phylo-)genetic re l at ion s among languages; th a t is, a l a ng ua ge can have only one phylogenetic parent.2 This assumption lea ds them to describe mixed languages as having no linguistic genetic p aren t s instead of having multiple linguistic genetic parents . Thomason & Kaufman desc1ibe the devel opm en t of mixed langu age s as 'nongenetic' (Thomason & Kaufman 1988:108). This leads to the second po i nt .

198

The descent of languages

If we take 'genetic' as referring to the process of evolution described by the generalized theory of selection as proposed by Hull and as applied to language change in this book, then we can see the development of mixed languages as genetic in the sense used in this book, that is, mixed languages can be placed in a phylogeny. The difference is that they do have multiple parents, contrary to the family tree model (cf. Trudgill 1996b). In fact, in biology there are very similar phenomena to language contact once one leaves the animal kingdom, moving no further than to the plant kingdom (Grant 1 98 1 ; see also Hull 1988:450). Genes can be transferred from one plant species to another by the process of introgression, described briefly in §7.4. l (Grant 1981, chapter 1 7) . If the two species hybridize, then the hybrids can backcross repeatedly with the first species·, thereby introducing genes of the second species into the gene pool of the first species. The process of HYBRIDIZATION itself is extremely common in plants, occur­ ring in a wide variety of ways in genetic terms (Grant 1 981; more than half the book is devoted to the processes of hybridization) . Sometimes hybrids lead to the partial or complete merging of the formerly independent divergent species into a new species subsuming both of the former species. The merging of for­ merly divergent phylogenetic lines is called RETICULATION. In other cases, the hybrids can themselves create a new, third species, if they become reproduct­ ively isolated and succeed in reproducing themselves (Grant 1981, chapters 19, 20 and 35) . Thus, even in the biological ..yorld it is not a necessary fact of evolution that new entities are created solely by divergence: 'The process of evolutionary divergence is not inexorable. Within wide limits it is subject to reversal' (Grant 1981:72). Thus even in biology the tree model is inadequate: 'If a phylogenetic tree is the extension of the normal pattern of animal speciation, plant speciation has often led to the formation of a phylogenetic web' (Grant 1981:76; see for example the phylogenetic diagram for species in the genus Clarkia in Grant 1 98 1 : 332, Fig. 25 . 3) . The following sections will survey the sources of linguistic diversity from a plantish point of view.

8.2

Primary language birth: societal divergence and normal transmission

Grant retains the term PRIMARY SPECIATION to describe speciation by divergence (Grant 1 98 1 : 1 53). I will use the term PRIMARY LANGUAGE BIRTH to describe the evolution of new languages by divergence. More accurately, primary language birth must be described as a consequence of societal divergence. The linguemes of a language are only the replicators. The interactors are the individual speakers of the language. The speakers form a population, that is the society, which is communicatively isolated from other languages. (Communicative isolation is never entirely complete; see §8.3.) The utterances they produce also form a population, that is, their language. The

Primary language birth

l 99

l i n guem es of the language form a lingu e me pool. It is th e behavior of the inter­ actors, however, which leads to any changes in the constitution of the lingueme

is, selection of var i an ts and thus the p ro paga tion of a ch an ge As argued in §7.2, a society is h etero geneou s in two ways: it is made up of mu l ti p l e communities defined by domains of shared ex perti se, and individuals are linked to other individuals by ties of different strengths, with a limiting case of zero (no tie at all). The pattern of tie s tr en gths defines subpopulations of the society, which are relatively communicatively isolated. To the extent that subpopulations of the so ciety are communicatively isolated, the lingueme poo ls of each su b p opu lati o n are distinct. The structure of each li n gueme pool will be s li ghtly different, but the di ffe re nces may i n crease over time through different social selection processes o pera t i n g in each sub po pul ation. B iol o g i sts describe a va ri et y of ISOLA TING MECHANISMS for o rgani s m s. The biol ogical isolating mechanisms are divided in t o three types (Grant 1981: 11112): geographical, ecological and reproduct ive Equivalents to all three types of i so latin g mechanisms are foun d in l anguage as wel l The c las s i c isolating mechanism i n l a n guage as in bio lo gy is geographi ca l sep arat ion Geograp h ical s ep aratio n is the sp a t i al separatio n of speak ers in two po pulat i ons to a di s t an ce b eyo nd the normal geograp h i ca l mobility of the sp eak ers . Two populations of highly mobile speakers would have to be qui te

pool, t h at

.

.

.

.

wide l y separated in order for geo graphical separation to lead to communic­

isolation . Nevertheless, even in the con temporary hig hly mobile, long­ communication world, a l arge enough proporti on of the popul ation is sufficiently isolated for di fference s between American and British E ngl is h to

a tive

d i s t ance

survive and flourish (in particular, accent and vocabulary, but also some sig­

nificant grammatical differences). A s econd linguistic isolating mechanism is social separation and relative lack of social mobility. This i so la tin g mechani sm is equivalen t to eco lo gical

separation, treating social differences as e qu ivalent to ecolog ical differences (see §2.4.2). S ome individuals may not talk to each o ther even if t hey live in the same ge o graph ic al area for social reasons, e.g. th ey belong to different ethnic groups or different castes. In fact many cases of social s eparat ion in a single geog raphical area involve sep a rati on by communities of a single society. That is, the different communities play different but interconnected roles in the s oci ety and hen ce there is still s ign ific an t communicative interaction between ,

,

the communities (see §8.3.2).

Rep roductive isolation is the lack of gen e exchange via

repro du c t i o n due to

differences in the rep ro ductive organs or reprod ucti ve behavior. The equivalent

reproduction is conversati ona l intercourse between individuals. There is at least one case of reproductive isolation in language. Deafness con­ sti tu te s a barrier to using voice and ears for commun ica tive interaction, t h ereby isolating deaf s peak ers from the speak in g la n gua ge community. Deaf commu n ities have deve l oped sign languages using hands an d eyes for communic ati on, instead of voice and ears. Of course, he arin g individuals who are sighted can acquire a sign language ; and deaf individuals can use l i p reading to understand

of biological

­

200

The descent of languages

spoken language, and learn to produce spoken language themselves. Neverthe­ less, the isolation from spoken language caused by deafness, even if it is not complete, has been sufficient to spawn sign languages which are conventional signaling systems as much as spoken languages are. The family tree model is based not only on divergence as the primary mech­ anism of language birth, but also on the normal transmission of language. Thomason & Kaufman define normal transmission as follows: 'a language is passed on from parent generation to child generation and/or via peer group from immediately older to immediately younger, with relatively small degrees of change over the short run, given a reasonably stable sociolinguistic context' (Thomason & Kaufman 1 98 8:9-10). Normal transmission is a description of a social structure. There are no significant communicative interactions with individuals outside the society, and the only significant changes in the mem­ bership of the population are through biological reproduction and death, not through individuals from other societies entering the population. If these social conditions hold, then the lingueme lineages will be traced back in time through that society alone, and back through that society's unique parent society, and so on. Any new linguemes will arise only through altered replication of the existing linguemes, and hence will belong to lineages confined to that society of speakers. Of course, no society is purely communicatively isolated in the way that the hypothetical society described in the last paragraph ·is. There is always some degree of communicative interaction as a result of social contact. The follow­ ing sections describe the chief 'linguistic consequences of that contact. The sections are organized by social parameters, those being the matrix in which lingueme transfer occurs. There are a number of social parameters that appear to play a role in determining the linguistic consequences of social contact. These parameters include: whether or not the societies merge into one group, or a new group is created from parts of the old groups; the degree of multilingualism in the community; the number and type of communities in the society through which contact takes place; and the stability of the social structure.

8.3

From contact to merger

Any linguistic change that is due to contact requires some de�ree of social contact and bilingualism on the part of at least one speaker of the language in question . The contact and the bilingualism may be minimal: one speaker acquiring (even imperfectly) a word spoken by a speaker of the other language is sufficient to introduce that word in the first speaker's original language, from which point it may be propagated. Of course the degree of bilingualism could be much more substantial, involving most speakers of the original language and good acquisition of the other language; and there will be all sorts of intermediate possibilities. But the contact is the result of some degree of corn-

From contact to merger

201

municative interaction between speakers of the two languages, which requires some minimal knowledge of both languages on the part of the speaker of the original language.

If a linguistic change through contact is propagated, what has happened in effect is that a lingueme - anything from a word, an inflection, a phoneme, a

grammatical distinction, or a syntactic schema - has been transferred (more accurately, replicated) from the Jingueme pool of one language to the lingueme pool of the other language. As was mentioned above. the equivalent process in biology is introgression:

a

hybrid backcrosses with one of the parent species

and thereby introduces the transferred gene into the gene pool of the parent species. In identifying the equivalent process in language change, one must clearly distinguish the interactor and the replicator. In biology, the hyb rid

plant is an organism, that is, the interactor; it possesses genes (replicators) from species The hybrid interactor in language contact is the spe ak e r who possesses some knowledge of both languages that she speaks, that is, possesses l inguemes (replicators) of both languages. In other words, the linguistic equivalent of a hybrid organism is a bilingual speaker, not a mixed language ; this misunderstanding has plagued other evolutionary accounts of language contact (e.g. Whinnom 1971). Communicative interaction of the (possibly only partially) bilingual speaker with other members of her society is the equivalent of backcrossing: the interaction replicates the acquired lingueme in the language of the s ociety and further differential replication of that lingueme can propagate both parent

.

,

it in the language until it is established as a convention. The nature of the contact between two societies can then be described in terms of the nature of the lingueme flow from one language to the other, via bilingual speakers. 3 Lingueme flow is

a

function of two factors, associated with

innovation and propagation. The first factor is the degree of bilingualism of speakers in the two societies; this will determine the amount and direction of lingueme flow. The second factor is the degree of group identification of the speakers in the two societies. The second factor is particularly important when there is a high degree of bilingualism: will the speakers of the original language shift com pletely to the acquired language, or will they resist, and to what extent will they resist?

B.3.1

Language shift: e'ffects on the acquired language

Thomason & Kaufman divide language contact phenomena into those that have linguistic effects on the original language of a society coming under the influence of another society, and those that have linguistic effects on the acquired language. Thomason & Kaufman describe the effects of language contact on the original language as BORROWING: 'Borrowing is the incorpora­ tion of foreign features into a group's native language by speakers of that l an gu age: the native l a nguage is maintained but is changed by the addition of the incorporated features' (Thomason & Kaufman 1988:37). Borrowing will be discussed in §8.3.2 and §8.4.3.

202

The descent of languages

Thomason & Kaufman describe the effect of language contact on the ac­ quired language as SUBSTRATUM INTERFERENCE, which is t h e effect of language contact on the acquired language (if there is any). Substratum interference occurs when a speaker population SHIFTS to the language of another speaker population with which they are in contact. More specifically, substratum inter­ ference 'results from imperfect group learning during a process of language shift' (Thomason & Kaufman 1988:38). The chief social factor characterizing substratum interference is that the society speaking the original language does not resist merging with the society speaking the acquired l anguage ; in fact, the speakers often wish to merge wi t h the acquired language society. This con trasts with borrowing, where the speakers of the original language wish to maintain their distinct social identity, includ­ ing their language, albeit sometimes not too successfully and sometimes under great pressure from the influencing society The chief linguistic effect of substratum interference is alteration of phonological and syntactic patterns (Thomason & Kaufman 1988:118-19). In particular, the alterations do not involve the actual morphemes, that is, form­ meaning pairings or signs but instead abstract or schematic patterns. For example, many members of the Austronesian language family spoken in or near Papua New Guinea have undergone a number of syntactic changes, including changes in basic word order and types of morphology, which make them resem­ ble the non-Austronesian languages of Papua New Guinea. It is likely that these changes are due to shift. The case for shift is supported by the physical appearance of Melanesians (closer to Papuans) and by mitochondrial DNA studies of the Melanesian, Papuan and other Austronesian-speaking peoples (Melton et al. 1995; Redd et al. 1995). Comp a re for example the following sentences from Yimas, a Papuan language spoken in Papua New Guinea; Manam, a Melanesian language spoken on an island just off New Guinea (about 200 km from the Yimas territory), and Tongan, a Polynesian (Austrones­ ian) language spoken well away from Papua New Guinea: .

,

(1)

Yimas (Foley 1991:309): yara yaka- kra -ua -t -akn tree them- I- cut BEN P AST -him 'I cut trees for him.' -

-

(2)

Manam (Lichtenberk 1983:165): tam6ata ntag a -i -au -1o man I.will- follow - him -BEN -you 'I will follow the man for you .'

(3)

Tongan (Churchward 1953:112): na'a nau langa 'a e fale mo'o Siale PST 3PL build ABS ART house for Charlie 'They built a house for Charlie.'

From

contact to

merger

203

Manam and Yimas have SOV word order, whereas Tongan has SVO order. Man am and Yimas have bound person m ar ke rs and app lic ativ e suffixes such as the benefactive to encode oblique arguments, whereas Tongan has inde­ pendent pronouns and prepositions. But Manam and Tongan are Austronesian languages, while Yimas is a Papuan language . This pattern is in contrast to borrowing: 'while borrowed morphosyntactic structures are more often expressed by actual borrowed morphemes, mor­ phosyntactic interference through shift more often makes use of reinterpreted and/or restructured [ ori gi n al language] morphemes' (Thomason & Kaufm a n 1 988: 1 1 4- 1 5). Likewise, the phonological effects of shift m a y affect the ph o n ­ etic realization of acquired language phonemes or phonotactic and metrical patterns, but they do not alter the basic p hon e mic content of morphemes. Of cours e, these p attern s are linguemes j ust as much as actual words and m o rphe mes are (see §2 .4.3). SOV word orde r is a schematic syntactic lin g ueme , and a CVCV syllable structure is a phonotactic lingueme. I will call these SCHEMATIC LlNGUEMES, contrasting with SUBSTANCE LlNGUEMES, wh ich are words and mo rph em es with actual p honem i c substance . Thomason & K aufman state that the correlations substratum i n t e rferen ce I schematic l in gue me s and borrowing I substance linguemes ' are too we ak to have predictive value' (Thomason & Kaufman 1 988: 1 1 5). However, in §8 . 3 . 2 I will argue that structural borrowing of schematic linguemes, as opp osed to the borrowing of grammatical substance linguemes, is more like substratum inter­ ference than borrowing proper. Thomason & Kaufman also note that borrow­ ing of lexical substance linguemes often occurs if the shifting population is socially superordinate (Thomason & Kaufman 1 988: 1 1 6), giving the example of Norman French and English . However, it is quite possible that this is a case of borrowing by the subordinate population, especially if the loanwords are in d omain s of expertise dominated by the superordinate shifting population (as they generally are in the Norman French I English case4). After all, as Thomason & Kaufman repeatedly point out, it is easy to oversimplify t he social patterns of each contact situation . Thus, the correlations may be stronger than Thomason & Kaufman imply. We may now de scribe the process of substratum interference at the level of a single utterance. For example, a speaker may produce an utterance with a different word order (say, SOV instead of SVO) but the same native language words. The speaker's utterance is based on a combination of knowledge of the original language (with SOV order) and knowledge of the acquired language (its words). The speaker is combining a schematic l in gueme from the original language and substance linguemes from the acquired language. The utterance is an innovation from the perspective of the lingueme pool of the acquired language: a new schematic lingueme has been introduced into the lingueme po ol . But this innovation occurs through a lineage crossing from one language (the original language) to the other language (the a cquired language). 5 According to Thomas on & Kaufman, the primary factor in substratum interference is the degree to which shifting language speake rs do not perfectly

204

The descent

of languages

acquire their second l an guage As we have seen , Thomason & Kaufman incor­ porate this factor a s part of their definition of substratum interference. How­ ever, this is only half the story: i nnov a t i on There must be an explanation for how the innovation is successfully p ropagate d in the language that is, how it becomes a convention of the language. Thomason & K a u fm an identify se vera l soci a l fac tors that affect the degree of (successful) su b st r at um interference : the relative size of the shifti n g popula­ tion, its social status, and the le n g th of ti m e for the shift (Thomason & Kaufman 1 98 8 : 1 1 9 -20) . A l arge shifting p op u latio n relative to the native po pu l a tion of the acqu i red language, a socially superordinate shifting population, and a rapid shi ft all fa v o r substratum interference . However, it is rare that a large socially supe r ord i n a te p o p ulat i o n would shift (instead t h e su bordinate popu l at i o n typ­ ically shift s) ; and if the su pe rord i nate shift i ng group is small, few interfere n ce effects are found . Thomason & Kaufman ar g u e pe rs uasively that the infl ue nce of Norman French on English and Fr a n k i sh on French is smaller t han is u s u al ly s up pose d ( Th o ma s on & K a u fman 1 98 8 : 1 22-9). Th u s t h e m o s t lik e ly scenario for significant su bstrat u m interference is wh e n the shi fting popu la tion is larger t h an the native population , and the shift is rel a tively r ap i d Another i m por t a n t feat ure of this scenario is that most of the population speaking the acq uired language - n ame ly the shifting s pe a k ers - is bilingual . The native spe ak e rs may be mostly m o nol i ngual but in this scen a rio the y form a small m i n o rity of the total p opulat i on of the speake rs of the langu­ age . The l arge r the num ber of bilingual speakers, the greater p ropo rtio n of innovations will occur in the acquired language due to i nte rfe rence from the shifting sp e ake r s native language. The underlying mechani s m in the successful establishment of substratum interference is the degree to which native speakers of the ac q uired lan gu age accept the shifting speakers' er r or s as new conventions of the acquired lan­ guage. That is, n a ti ve speakers are willing to acc o mmodate to the s h i ft i n g speakers, even though the sh i fti n g speakers are not native s pea k e rs Two of the three social factors described by Thomason & Kaufman p rov ide motivation for native speakers to accommodate to nonnative speaker s (The third factor, rapidity of shift, is a ro ug h index of the l i kelihood that sh i ft i ng speakers ac q u i ri n g the language will succeed in doing so pe r fec t l y and thus pe rtains to innovation rather than propagation.) In th e most common scenario, the native speakers are outnumbered, and the innovation is p ropagate d through accom­ modation by the native speakers, or by a lack of access to e nough n ative sp eaker s on eno u gh occasions of u se Or the s hifting speakers have a so ci a l status which the native speak ers wish to identify with, ma k in g them willing to acco mmod a te to the nonn a ti v e sp eake r s The above scenario shows how (typically schem atic) lingue me s from one language enter a nothe r language An other, milder, consequence of shift is the simplification of the acq u ire d lan guage in certain respects (see, e . g. Tru dgill 1 979/1 983; Trudgill 1 989:232-3 ; Holm 1 988: I 0). In the evolutionary framework, simp lific at i o n is altered replication of the ex i sti n g l in guemes of the acquired .

.

,

.

,

'

'

'

.

.

,

.

.

.

From co ntact

to

merger

205

transfer. Alth o ugh the society has undergone a ch an g e in me mbe r s h ip t h rough shift, t h e l i n g uem e l ineages can still be traced back into the ancestral po p u l a t i �n of the original members of the so ci e ty . language, not li n g u e m e

8.3.2

Language maintenance I resistance: effects on the original language

The ch ief social d i fference bet ween borrowing an d su bstratum interference is t h at the society spe a ki n g the b o rrow i n g language ma i n t ai ns or eve n resists merging with the society speaking the source language of the borrowed l i n g ue m e s . M uch of the interest in borrow i ng has centered a ro u n d the s o rts of l i n gui st i c features - in our terms, linguemes -- that are borrowed , an d any ranking of t hose fe a t ure s in terms of relative ease or difficulty of borrowing. Thomason & Ka ufman argue that at least some earlier proposals for ra nk i n g of likelihood of borrowing suffer from the failure to distinguish between b o r r owi n g and sub­ stratum interfe rence . T he y also a rgue that descriptions of bo rr ow in g patterns should be c onst r u c t ed with respect to degree o f social c ontact, and p r o pos e a borrowin g s c a l e based on five d e g ree s of contact . The first degree o f contact is ca l l e d casual c on t ac t , defined as 'little bil i n ­ gu ali sm am on g borrowing-language speakers ' (T h om a s on & Kau fman 1 988:50, Table 3). At this stage, only c ont e n t words are borrowed. Specifically, n on ba s i c voca bulary is borr o wed, for ' c u l t ural and functional . . . reasons' (ibid . , 74). The fact that n o nbasic vocabulary are the first lin g ue mes to be transferred has long b ee n observed. They arc typically associated with domains of expertise found in the society spea king the s o u rce language but not in the society spea k­ ing the b o rr o wi ng l a n g u a ge . This process can be described in terms of the model of speech communities deve l o ped in §4 . 2 . 3 a nd §7 . 2 . The domai n of expertise of the source soc iety defines a c o m m unity. Members of the borrowi n g so ci e t y share in the domain of e x pert i se , and use the cod e of t he loaning society in that d om a i n . Those linguemes in the code that are not already p re se n t in the b o r r o wi n g s o c i e ty' s language are the more likely to be transferred to t h e borrowing lan guage, and those are more l i kely to be nonbasic vocabulary . Basic vocabulary, on t h e other hand, ge n e r ally denotes the core ex pe rti se that is common across many communities in a soc i e ty. Si nce t ha t core expe rtise is present in any socie t y , v o ca bul a ry for the core expertise is fairly strongly entrenched and is m ore likely to prevail over foreign basic vocabulary . (These are all probabilistic s t atemen t s , of co u rs e . ) H i g he r degrees of co n tact involve 'much bilingualism am o n g borrowing­ language spea ke rs ove r a long period of time' (T homa son & Kaufman 1 988: 50, Table 3). In the lexico n , function words begin to be borrowed, st a rt i n g with conjunctions and various adverbial pa rt ic l es (stage 2 bo rrowi n g) , and then other function words such as adpositions and then some basic vocabulary (e.g. some pro n ou ns and n umeral s ; stage 3 b o rrowi n g). The structural modificatio ns

206

The descent of languages

described by Thomason & Kaufman for stages 2-3 borrowing are to a great extent grammatical patterns abstracted from the increasing amount of foreign vocabulary, such as new phonemes, new phonemic distinctions, metrical and phonotactic patterns (Thomason & Kaufman 1 98 8 : 74-5). Matras ( 1 998) proposes an explanation for the category of the first function words that are borrowed . He argues that the conjunctions and adverbial par­ ticles that are borrowed perform discourse-regulating functions, that is, they are used by interlocutors to structure the flow of discourse in communicative interaction. Matras analyzes the b o rrowin g of discourse-regulating function words as 'fusion' (Matras 1 998:29 1 ). Fusion represents the employment of a single discourse-regulating grammatical subsystem, namely the one found in the loaning language . Matras argues that 'fusion . . . is motivated by the need to reduce overload in the mental monitoring of hearer-sided language-processing activities in bilingual communicative interaction' (ibid.). This processing explanation motivates the reduction to a single discourse-regulating system, but it does not explain which system is chosen. There are social reasons motivating the choice of the source language, in v o lvin g the relati on s hip between the source and the borrowing society. Thomason & Kaufman observe that two-language contact situations commonly involve only one-way bilingualism: only the borrowing society is bilingual (Thomason & Kaufman 1 988:95). In one-way language contact situ­ ations, one society presumes that it does not need to learn the other society's language; and so it is in that society's language that intersocietal communicat­ ive interaction takes place. That communicative interaction is structured by the discourse-regulating grammatical subsystem, which then becomes the con­ ventional discourse regulating subsystem for all of the bilingual borrowing­ language speaker ' s conversations. This represents a first step in the process of the absorption of the borrowing-language society by the source language society: the latter now defines the regulation of discourse for spe akers in both societies. The next stages in Thomason & Kaufman 's borrowing hierarchy (stages 4 and 5) involve 'moderate to heavy structural borrowing' (Thomason & Kaufman 1 98 8 : 83). There is an even higher degree of multilingualism and/or a longer period of contact with the source-language society than with slight structural borrowing. Structural borrowing comes in two types, however, which are not clearly distinguished by Thomason & Kaufman. One type is the transfer of substance linguemes, that is, actual inflectional morphemes and grammatical particles such as classifiers. The other type is the transfer of schematic linguemes, that is, grammatical (syntactic or phonological) patterns. As was pointed out in §8 . 3 . 1 , transfer of schematic linguemes does not involve the borrowing of any actual forms, but just transfer of a word order pattern, or a calque of foreign construction using native morphemes, or a foreign allophone or metri­ cal pattern onto native words. All of the examples that Thomason & Kaufman give of moderate to heavy structural borrowing involve mostly transfer of schematic linguemes and only a few cases of grammatical substance linguemes.

From contact to merger

207

Kaufman reserve the substantial tr an sfer of gra mmatical to the case of extreme bo rrowi ng {see §8.4. 3). This differ en ce in type s of structural b o rr ow ing is signi fican t because the transfer of schemas is gene ra l ly found with substratum interference, but the transfer of grammatical forms ge ne ra ll y is not {see §8 . 3 . I ). I s u ggest that i t is not ( j ust) intense contact between the two societies that leads to th e transfer of schematic li n gu eme s. Intense contact involves a high degree of multilingualism , just as is found in t h e cases of successful substratum interference . A s we have seen, substratum interference ge n e ra lly requires a sub s tan t ial group of second­ language speakers me rg i ng with the acquired-language society, in other words, a large prop o rti o n of the p o pula t i on is bilingual. In t he case of stru c tura l borrowing, a large pr opo rt ion of the borrowing language population is bi l i ngual . But there must also be a si gn i fican t degre e of openness to nonnative pat te rn s of speech by n ati ve speakers of the borrowing langu age . In the case of struc­ tu ra l borrowing, however, the source l an guage speakers are not b i lin gu al in the borrowing l a ng u a ge . Hence it m ust be the nonnative speech patterns of the native speakers themselves that are transferred, n amely the speech patterns of their kn ow led ge of the source language. Of cou r se , the social reasons for the one-way bilingua l i sm i n d i ca t e the socially dominant status of the source lan guage soc i e ty, and {as in the case of s ub s t r a t u m influence) the sociall y dom­ inant status can be a mechanism for the adoption of their speec h patterns. It is t hese two social fe a tu res - e x ten sive m ult i l i n gu al i sm and openness to accep ti n g n o nnative speec h patterns - that the moderate t o he avy structural borrowin g languages have in common with - substratum interference. Multilingualism an d openness to innovations m ay also underlie the CON VER­ GENCE in ph o n o l o gical and syntactic structure of the local v arietie s of U rdu , Marathi and Kannada in t h e Indian vil lage of Kupwar (Gumperz & Wilson 1 97 1 ), and the lesser d e gree of co n vergenc e found in the lar ger linguistic area s such as the Balkans , the Pacific North we st and M e so ame r ic a . Thomason & Ka ufman place the Kupwar varieties in t h e category of moderate structural b orr owin g. But unlike the other cases of moderate s t ru ct u ral bo rrowing, there has apparently been relatively little lexical bo rrowing in K up wa r . Thomason & Kaufm an state 'structural b orr o wing is i n v ari ab l y preceded by lexical borrow­ in g' (Thom as on & K a u fm an 1 988 : 1 1 3). Yet Gu mperz & Wilson write that 'o nly morphophonemic differences (di ffe rences of le x i ca l shape) remai n ' (Gumperz & Wi lso n 1 97 1 : 1 5 5) - i .e. the actual words and gr amm atical morp hemes are native to each language (see al so ibid. 1 6 1 -2). The following examples illus­ tra t e the parallel schematic linguemes an d different substance linguemes (i bid . In fact, Thomason &

substance l i ng u emes

1 57, 1 58 ) : (4)

Kupwar Urdu : Kupwar Marathi: Kupwar Kannada:

g;)C te tu twa k:m;) gel hotas yalli h o g idi ni where gon e were (2so) you 'Where did you go?'

kha

208

The descent

of languages

(5) Kupwar Urdu : K upwar Marathi : Kupwar Kannada :

ye tumhar-;:i gh;:ir gh;:ir he tumc-� m;:ini id nim-d this your-SUFF house 'This is your house . '

h;:iy hay eti

is

These three languages have coexisted i n the village fo r centuries, but the speakers have not abandoned their traditional languages during that time ( in fact, they live in separate neighborhoods; ibid. 1 53). There is also extensive symmetrical multilingualism (ibid .), not one-way bilingualism, in Kupwar. The languages have converged on a single set of schematic linguemes which are drawn from all t h ree languages. In other words, there is no clear favored directionality for the flow of schematic linguemes from one society to another. The extensive multilingualism may have led to the presence of many utterances by second-language speakers in the pop ulation of utterances constituting each language. But there must also have been a moderately high degree of accom­ modation so that native speakers of each language propagated schematic linguemes from the other languagcs. 6 LINGUISTIC AREAS probably represent a less dramatic example of the same phenomenon, partly due to the diffuseness of the broad geographical areas they represent. There was a great deal of movement and large ly symmetrical multilin­ gual ism in the Balkans over a long period of time, for example (Thomason & Kaufman 1 988:95). The variou s ethnic groups maintained their languages, but accepted a number of schematic linguemes from other languages, to the extent that linguists have observed the areal patterning of linguistic features. Substratum interference through shift, convergence / linguistic areas, and struc­ tural borrowing all result in the transfe r of schematic linguemes to another language . What all three have in common is a high degree of multilingualism and an open ness on the part of native speakers to accepting nonnative schematic linguemes . In the case of substratum interference through shift, the native speakers may be o utnumbered or outclassed by the nonnative speakers, and accommodate to them out of necessity. In the case of convergence and linguis­ tic areas, the native speakers do not have to accommodate to the nonnative speakers but do so anyway, for reasons not entirely clear (beyond the extensive multilingualism). In the c a s e of structural borrowing, the native speakers adopt the schematic linguemes of their own second language, probably due to the increasingly overwhelming domin ance of that language's society. We may now consider the rel a tions h ip between social structures and the types of linguemes transferred in borrowing vs substratum influence. Borrowing appears to be a result of the partial, gradu al adoption of the source-language social structures by the borrowing language, including the substance linguemes that go with them . It begins with adoption of expertise unique to the source­ language society (nonbasic vocabulary), then, as multilingualism increases, the adoption ( Matras' fusion) of the conversational structuring of the source-

Hybridization in language

209

language society (discourse-regulating fun cti on morphemes), and then the adop­ tion of some of the core expertise of the source-language s o cie ty ( some b as ic vo cab ulary and other function words) . T h e adoption of the s ou rce- l a ngua ge spheres of action is g radu a l : despite the in c re a s i n g extensiveness of the borrow­ ing, the borrowing socie ty still attempts to maintain some degree of in d e pen d ­ ent i den t i ty . This appears to be the case w i t h what Thomason & Kaufman call extreme bo rrowin g, the resu lts of which are more generally called mixed langu­ a ge s ; these will be discussed in §8.4. 3 . Substratum interference on t h e other hand i s a result o f a wh ol e s a l e (even if d raw n -o u t) shift o n t h e part of the s pea k e r s . In oth e r words, as many substance l i n gue mes as possible are ad opted from the acquired language (i n the case of shift) or retained from the o riginal lan gua ge (in the case of convergence, l i n gu i s t i c areas a n d structural borrowing) . In addi ti o n , the native speakers of the acquired lan gu age are ope n to accep t i n g nonnative schem at i c linguemes. A possible exp l ana ti o n for this l atter fact is that the social identity of the native spe ak e rs is being eroded by t he influence of the s oci ety wit h which they are in c onta ct , whether it is nonnative s peak e rs surrounding them (shift, convergence, linguistic areas) or sp eak e r s of the source l an g ua ge whose society is grad u al ly i m po s i n g itself on them (structural borrowing).

8.4

Hybridization in language

In the bi o l o gical world, hybridization only occurs between relatively closely populations. In §8 . 3 , I argued t hat the linguistic equivalent of a hybrid o rgan i sm is a bi- or multilingual ind ivid ual . But an individual may be b i l i n gual in distantly rel ated languages. This suggests that either distantly related lan­ gu ages are m uch more similar to each other than d i s tan tly related organisms a re, or that the restriction on li n gu ist i c hybridization is much loose r than th at for organisms. The answer is probably a little bi t of both . All l an gua ge s per­ form r o u g h l y the same functions, that is, express the same range of concepts, and utilize remarkably similar conceptualization processes for construing the world (a l t ho ugh they may v ary in wh i c h construal is applied to which real­ world p he nom en on ) . Biological organisms, in contrast, occupy a wide and va ry i ng range of e co l og i ca l niches, and thus vary s t ru c t ura l l y much more than l ang uages do. On the ot he r hand, there is no restri ct i o n on the combi n at i o n of languages that a multilingual can master, whereas there are sometimes rather su rpris i n g restrictions on the natural hybridization of ecologically and mor­ phologically similar speci es (Grant 1 98 1 : 1 97-8). Hybridization of org an i sms requires a merger of the genotypes of the two paren t speci e s in the hybrid, since the phenotype - the hy b ri d individual - is an expression of the gen otype (as a function of the organism's environment) . As was noted in §2. 5, the re latio n ship between speaker and utterance is not the same as the re lati o ns h ip between the genotype and the p hen o type . Nevertheless, re l ate d

210

The descent of languages

multilingual speakers can in principle produce hybrid utterances and even hybrid languages. In this section, we will discuss some of the circumstances under which linguistic hybridization occurs. 8.4. 1

Hybrid utterances in code-switching and code-mixing

A bilingual speaker in a bi liIJ gual society employs two distinct languages in her linguistic repertoire. But if the hearer is also bilingual, utterances from either language can be produced, if socially appropriate . Moreover, utterances made up of linguemes from both languages can be produced, again if socially appopriate . All degrees of hybridization of utterances are found, without sharp distinctions. However, it is not clear what status the population of hybrid utterances has in terms of a language. The various degrees of hybridization of utterances are called CODE­ SWITCHING and (in more extreme cases) CODE-MIXING (Romaine 1 995). The strictest degree of separation of codes in a multilingual society is the use of only one code in each social domain. This obtains in the case of diglossia (Ferguson 1 9 59/ 1 972), in which a literary standard (called H for 'high') and a mutually unintelligible vernacular (called L for 'low') are used in discrete domains. Ferguson lists typical H domains as a religious sermon, political speech, news broadcast or personal letter, and typical L domains as instruc­ tions to servants or waiters, conversations with family, a radio soap opera and folk literature (Ferguson 1 959/1 972:236). In this case there is no merging of the lingueme pools of the two languages. In bilingual communities, however, one also often finds what Gumperz ( 1 982) calls conversational code-switching, where both codes are used in the same con­ versational encounter, to varying degrees of integration of the codes in single utterances. Gumperz describes code-switching as 'meaningful juxtaposition of . . . strings formed according to the internal rules of two distinct grammatical sys tems (Gumperz 1 982:66, emphasis original) . In other words, the interlocutors construe the codes as distinct, and exploit the social value of the two codes in a single conversational encounter. When larger segments of discourse, such as whole utterances or whole sentences, occur in a single code, it is generally clear that the two languages (utterance populations) are still separate. In particu­ lar, the two codes represent alternative ways of saying the same thing: 'it is the juxtaposition of two alternative linguistic realizations of the same message that signals information, not the propositional content of any one conversational passage' (Gumperz 1 98 2 : 84) . For there to be true hybridization of a language, there would have to be a single population of utterances defined by some social domain, with a random mixture of linguemes from both parent languages. Individual sentences or clauses would involve a mixture of linguemes. Each lingueme in the lingueme pool would be accepted by the speech community as the conventional signal for the relevant function in the code. There are a number of situations in which various subsets of these criteria appear to hold . '

Hybridization in language In most c on ve r sat ion al co d e sw i t c h i n g , -

21 1

there are also utterances that involve

code-mixing at all gramm atical levels. R omaine discusses a passage produced

by a Panjabi-English spe ake r, part of wh ic h is g iven bel ow (Romaine 1 995: 1 22) : (6)

mean, ma khad eana ma ke, na, j ado Panjabi bolda £, pure Panjabi bola cosi mix karde r£n e a . I mean, unconsciously, subconsciously, bri jan ee, par I wish, you know k e ma pure Panjabi bol �ka. ' I mean , I my se l f would like to s pe ak p u re Panjab i whenever I sp ea k Panj abi . We keep m i x i ng I mean u nc o nsc i ou s l y subcon sc i ou sly, we keep doing it, you know, but I wish you know, that I co ul d speak pu re Panjabi.'

I

.

,

,

Romaine suggests that no lin gu i st believes that there are no syntactic con­ straints on c o de swit ch ing (Romaine 1 995: 1 25). But in a critical review of the prop o s als for synt acti c c ons tr a i n t s it appears that there are no con strai n ts that cannot be violated. Eve n t he t ig h tes t combination, a stem from one language with an affix from another, occurs, as in Ge rm an/ Engl ish That's what Pap s ch i me i n s to say (cf. German mein-t 'means' ; Romaine 1995: 1 48-9, from Clyne 1 987:756; Romaine al so ci te s e x amp le s from Dutch/Turkish, Maori/ E ngl is h Ja pa n e se / E n gl i s h an d F i nn i sh / E ngl i sh ) A number of soci oli n guis ts argue that utterances which appear to v i o l ate · p ro p ose d grammatical constraints on co d e swi tch ing are cases of 'nonce bor­ rowing' , that is, the y represent utterances of the bor r ow ing la n g uage and not code-switches (see Romaine 1 995 : 1 39-40, 1 45 , 1 56). But i f a bor rowi n g is a nonce form, th a t means it is not i n co rpo r a ted as a conventional l i ngueme of the lan guage . To be sure, borrow i n g b egi n s w i th some bil ingual speaker u s in g a fore i gn word, which then is propagated through the speak er s communi ty incl ud ing to other m o n o l i n g u a l speakers. Th i s is also a gradual p roce s s , with variation in the degrees of integration of the fore i gn word into the system of the l angu a ge until it is fu l ly conventionalized (see Romaine 1 99 5 : 56, 59-62; Pfaff 1 979; see also the conventionalization of innovations described in §7. 3). But when the so-called nonce bor ro w ing occurs in the m idd l e of a highly code­ m ixed utterance, it h a r dl y differs in status from code-mixin � itself. Instead, po si t in g a category of nonce b o rrowi n g appears to b e used only to save the cons t rai n t i n ques tion : 'there are no u n am bi gu ou s criteria which will decide in all cases what type of lan guage co ntac t p he n omen a we are d eal i ng wi th ' (R omain e 1 99 5 : 1 57). In fact, some utterances are so mixed that one cannot clearly decide whether the utterance is bas i call y in o n e language (the so-called base l an gu age) with elements of the other language added, or vice versa ( R omaine 1 995; LePage 1 992: 76-7) . Again, gramm a ti cal criteria proposed to determine the 'base lan­ guage' give amb i guous or unintu itive re su l ts (Ro mai n e 1 995 :44-9) . In these cases, if not in c ode - mi xing more generally, utterances appear to be a genuine mixture of l i ngueme s from the two source codes, to almost any degree of grammatical in t imacy Romaine suggests that the mixed utterances may form their own code (see R oma i ne 1 995: 1 60 and references cited the re). That is, the mixed utterances may -

,

-

,

.

-

'

.

,

The descent of languages

212

population of utteran ce s distinct from b o th parent codes/languages . If codes would satisfy the first criteri o n for hybridization. One q ue stion that arises is how stable this code is. is there an in de pe nd en t l ingueme pool of l i n gu em e s that conventionally de s i g n a t e various concepts? Or i s the mixed code simply an unconventionalized mi xt ure of l inguemes from the two p ar ent languages, used to perform ce rtain social and pragm atic functions? The d e gr ee of stabilization around a set of conventions has been described as the degree of focusing in a con tinuum from FOCUSED to DIFFUSE (LePage & Tabouret-Keller 1 985; see also Milroy l 992b:6 l ; Trudgill 1 986:85-6). As Trud gi l l no te s , Le P age & Tabouret-Keller d e scri b e societies as varyin g in focusin g de­ pe nding on how close-knit the ne twork is ( L ePa ge & Tabouret-Keller 1 98 5 : 5), but then shift to describing the code itself as v aryi n g in focusing de pendin g on how much variation is foun d (LePage & Tabouret-Keller 1 98 5 : 1 8 1). The assumption is that the number and frequency of variants of a sin gle l in gue me in a lan guag e is dependen t on the social s t ructure . A more focused community or soci et y d e fine s a more sh arply bounded population of utte rance s and a more u niform set of conventions. C om pl e te un iform i t y of conv e ntion s is an i mp ossi b i l ity because of the het­ erogeneous structure of societies (§7 . 2 ) . C omp lete diffuseness would be a totally ad hoe means of com m unic ation . But the d e gree of focusin g of soci e ti es , and of t he i r linguistic conventions, varies si gnific antly within the bounds of those extremes. When e xam i ning hybrid lan guages , we mu st assess the d e gr ee of foc u si n g of the society using the hybrid lan g u age as well as th e contributions to the m ixe d lingu em e pool from the p arent la n gua ges .

form this

a

is true, t hese

8.4.2

True hybrid languages

cle ar examples of true HYBRID LANGUAGES fittin g the criteria given in §8 .4. l . Like biological h ybri ds , they occur when the parent l anguages are closely related varie t ies . These are the cases of dialect mixin g and koineization d iscu sse d in §7 .4.4 (Trud gi ll 1 986, ch a pte rs 3--4). For example, the urban v arietie s in Western European countries such as Bri tain developed in the past two centuries as a result of immigration from the rur a l areas to the cities in the Industrial Revolution . The pop u l a t i on of utterances produced by the new urban speakers represented a mixt u re of the linguemes in the rural dialects spoken by the immigrants. Presumably the con­ tac t gave rise to a state of affai rs in which lin gu emes from different rural dialects were used in individual utt e ran c es , or intermediate v a r iants were cre ­ ated (see §7 .4.4). Finally, the urban dialects have become relatively focused, that is, n ew co n ven t ions were established . These new conventions result in the loss of va r ian t s or their reallocation to social/stylistic functions in the urban society (Trudgill 1 986: 1 1 �26) .7 The urban socie ty is separated and delineated from the rural societies which the urban spe a kers came from. Of course, the focusing process is a gra d u a l one, whic h can be measured by q u anti t ative techniques a p p l ied to u t terances produced by the community (see for i ns t ance Thelander 1 979, de scri bed in Trud gil l 1 986: 9 1 -4).

T he re are some

Hybridization in language

213

Another example of hybridization between dialects i s the development of st an dard varieties in many societies. Standardization can develop naturally (see Haugen 1 966/ 1 972 for a description of the process), when a particular city or region becomes politically, economically and/or socially domina nt. This occ urre d in the case of English, French, German, S p ani sh and Italian in the pe riod 1 200-1 500. and in Persian in the mid first millennium AD (Thomason & Kaufman 1 98 8 : 209- 1 0; see also Scatton 1 98 3 : 9- 1 0 o n the recent develop­ ment of the standard Bulgarian language). In all of these cases, there was also i n fl uence from speakers of other varieties, n o t to mention the mixture of dialects in the l e adi n g cities, resulting in a standard variety that contained elements from a range of dialects (and forms intermediate to those found in the source dialects; see §7 .4.4) . Hybrid languages may occur where there is contact between two closely related languages as well , such as Spanish and Po rtuguese in Latin America. Uruguayan Spanish speakers are in contact with Brazilian Portguese speakers on the Uruguayan-Brazilian border. The mixed u tterances produced in contact have focused in some isolated communities in No rthern Uruguay: the mixed variety is the only language spoken in those communities (Trudgil l 1 986:83---5). There are other instances in which two relatively closely r�lated varieties come into contact and produce hybrid utterances, but which do not appear to focus on a third variety which i!1 di s t i n ct from the parent variety. These are the creation of s oc i a lly defined continua, as in decreolization (see the descrip­ tion in Trudgil l 1 986: 86-9 1 ), the s pre ad of a sta ndard, and of some diglossic situations . Decreolization occurs when a creole (see §8 . 5 . 2) c om e s into cont act with the language from which the bulk of the creole ' s vocabulary is d rawn (the lexical source language), as in Guyana. Speakers of the creole who have access to the lexical source lan gu a ge to some degree produce utterances that represent a mixture of l in gu e me s from the creole and the lexical source language; the range of i mme di a te utterances is called a POSTCREOLE CONTINUUM. These utterances occur in social situations in which an intermed iate identity between the creole society and the lexical source-language society is desired. These intermediate varieties are also hybrids. However, in diffuse s ocie t ies the hybrid varieties have not stabilized as a distinct code. Instead, variants a re chosen (or blended) from the creole and the lexical source language to express different degrees of social distance from the two extreme points in the social structure. Similar phenomena are sometimes found when a vernacular varie ty comes into con tact with a standard v a rie ty and in diglossic societies where a continuum of utter­ a n ce s is produced between the H and L v a ri e t i e s (see Heath 1 989: 1 0, 30-32 for discussion of code-mixing between Moroccan colloquial Arabic [L] and Classical ,

Arabic [H] ).

8.4.3

Mixed languages

The question remains, can a hybrid language arise whose parent la nguages are not c losely related? That is, can a code that is mixed between distantly

214

The descent of languages

related languages become focused as a third code distinct from the two parent languages? There are a number of cases of MIXED LANGUAGES reported in the literature . However, there is relatively little information on many of them, of either a social or linguistic nature. Nevertheless, it appears that there is a range of social situations and a range of linguistic mixtures. However, the mixing is always linguistically asymmetrical, albeit in different ways. There is a wide range of theories about the origins of mixed languages, but it appears that different mixed languages arose in fundamentally different ways (Matras 2000). One question that often is raised about all mixed languages is the phylogenetic question: given a language A that appears to be a mixture of languages B and C, can we conclude that language A is 'really' a descendant (albeit dramatically altered) of language B or language C? In the evolutionary framework, we may break this down into two questions, based on the two populations involved in language evolution , the speakers (interactors) and the linguemes (replicators) . We may compare the social contexts in which mixed languages arose, and then examine the lingueme pool defined by the mixed language in relation to the two parent languages. The following discussion is very cursory, and does not do justice to the complexity of the social and linguistic facts of the situations described. Never­ theless, it attempts to divide mixed languages by the presumed historical social context of the speakers, and examines the nature of the lingueme pool of the mixed language spoken by the society in question. I suggest that there are at least three distinct social contexts with three different types of mixed lan­ guages: mixed marriage languages, death by borrowing, and semi-shift. 8.4.3. 1

M ixed marriage languages

There are two mixed languages which appear to have arisen through marriages where the father came from one society and the mother from another, Mednyj Aleut (also called Copper Island Aleut) and M ichif. These MIXED MARRIAGE LANGUAGES are precisely the two cases known by Thomason & Kaufman in which the grammatical system of one part (nominal) is from one parent lan­ guage and of another part (verbal) from the other parent language (Thomason & Kaufman 1 988:233). However, the make-up of the lexicon is different in the two cases. Mednyj Aleut is spoken on Mednyj or Copper lsland. 8 The Aleuts moved to ' Copper Island in the early l 9th century (possibly moved there by the Russians), and Russian seal hunters moved there soon after. The Russian population was approximately l 0% of Copper I sland, and they married Aleut women. (This contrasts with neighboring Bering Island, where there was a much lower proportion of Russians, and the language remained basically Aleut.) The ori­ ginal populations were probably mutually bilingual (Thomason & Kaufman 1 988 :234) . There was a period of Russian withdrawal, but now Russian is rapidly taking over on the island (ibid.).

Hybridization in language

215

The Mednyj Aleut language has Aleut noun stems, nominal derivations and nominal inflections, and Aleut verb stems and derivational morphology, but Russian verb inflections and as socia ted forms, such as the imperfective future auxiliary (Golovko 1 996) . Some of the Aleut verbal affixes in Mednyj Aleut include subordinate verb forms and aspectual-modal concepts ( Go l ovk o 1 996 :66), though sometimes both Aleut and R ussi an constructions are used (ibid. 72-3) . The contributions from both parent languages are largely preserved intact; in particular the Russian verbal constructions are perfectly ordin ary morphologically complete R ussian pattern s (Thomason & K aufman 1 988:237). A more subtle mixture of Russian and Aleut linguemes is found with verbal agreement, in which Russian number/gender forms agree with the subject pos­ sessors as in Aleut (Golovko 1 996:68-70) , an d word order, in which largely Aleut nominal and verbal forms follow Russian word order (ibid., 72). Michif is spoken in a number of Metis communities in Canada and the US, p articul arly in the Turtle Mountain area in North Dakota (B akke r & Papen 1 997) . The ancestors of present-day Michif speakers are bel ieved to be the descendants of French trappers and (mostly) Cree women in Canada (Bakker & P ape n 1 997:296-7) . The Metis have been multilingual du ri ng most of thei r history (ibid. 30 1 , 352), but more recent generations include native speakers of Michif, and French and Cree are being lost in favor of English (ibid . 3 0 3) . Michif is e s se n tially restricted to the domain of the home (ib id ) . Bakker & Papen hypothesize that the Michif language must have developed in the early 1 9 th century, although it is first reported in the 1 930s (ibid. 352-3). The M ichif l an guage has l argely Cree verbs inc lud ing essentially their full complement of (complex) verbal inflections (ibid. 35 1 ). The noun phrase is largely French, albeit with some simplification (e.g. no count/mass distinction), and w ith Cree demonstratives, post positions (in addition to mo s tly French pr epo s i tions) and even Cree obviative markers (ibid. 348, 3 5 1 -2) . There are some French verbs which take French inflections, though again reduced. The French and Cree items re t ain their i ndepe n dent phonologies (ibid. 349-50) . As Thomason & Kaufman have noted, Mednyj Aleut and Michif have in common the fact that nominal and verbal morphosyntax come from different parent l an gua ge s . However, in Mednyj Aleut, the verb stems remain Aleut, while in Michif the lexical items in each category come from the corresponding parent languages. In both cases, the language of the presumed ancestral mothers is domi nant : the lexicon of Mednyj Aleut is almost entirely A l eut while the French noun phrase grammar is reduced compared with the Cree verbal gram­ mar. Interestingly, in both cases the two components are close to the parent languages and not well integrated with each other (in the case of Michif' s phonology) . This is probably due t o the persistence o f bilingualism i n both parent languages during most of the mixed marriage languages' history. 9 The mixe d marri a ge lan guage s p re sente d in this section appear t o be extreme cases of the merging of substance as well as schematic linguemes between the distinct languages of the parent speakers. A more common situation in tradi­ t i onal societies was exogamous marriage patterns, where a marriage partner '

'

.

,

216

The descent of languages

was required to come from a different society, which may speak a different language . Exogamy is one of the important sources of multilingualism in tradi­ tional societies. However, it normally does not lead to as extensive mixing as in Mednyj Aleut and Michif. Exogamy may involve speakers of more than one other language, unlike Mednyj Aleut and Michif, where only one additional l anguage was involved (Russian and French respectively) . Th i s lessens the in­ fluence of any one external language. Also, the l an guage of the society which the new spouse joined tends to be the dominant language; this was also true of Mednyj Aleut and Mich if, where the mothers' lan gu age is dominant in the mixture. One case of exogamy and language mixture has been examined in some detail, that which formerly held in eastern Arnhem Land in Australia (Heath 1 978, 1 9 8 1 ) . In this area , languages that are quite distant genetically have been in contact through exogamous m a rri age patterns and through seasonal gather­ ing for ritual purposes (Heath 1 978: 1 4- 1 6) . The social l in ks between two pairs of societies, R ith ar ngu and Ngandi, and Nun ggubuyu and Warndarag, were particularly strong (ibi d . ) . The societies thus became bilingual, and the bilin­ gualism was relatively stable. From this sociolinguistic description, one might expect the mixing i n Amhem Land to exhibit properties of convergence and mixed marriage languages. This appears to be the case . There has been convergence of schematic linguemes, as one mi ght expect in stable multilingual societies. In particular, there has been a convergence of phoneme segment inventories and surface ph onotactic patterns across the l a n gua ges (Heath 1 978:66), independP.nt of lexical borrowings (see below). Morphosyntactic sharing of schematic linguemes is also found, such as enclitic pronominals and the re t reat of the infinitive in subordinate clauses· in Ritharngu under Ngandi influence, a nd the restructuring of the usage of the Nunggubuyu case affixes under Warndarag influence (Heath 1 978, eh. 4) . In addition, however, there has been substantial diffusion of substance lin­ guemes, including grammatical substance l inguemes, as one might expect i n a mixed marriage language situation . Both pairs of language e xhi b i t a substantial influx of lexical items from the other language in a wide range of categories, the number of borrowed words i n Ngandi approaching 50% (He ath 1 98 1 : 3 56). There has also been borrowing of substan ce linguemes for inflectional and derivational morphology, including many case affixes, noun class prefixes and a negative suffix (Heath 1 978, eh . 3 ) . As with the lexicon, more diffusion has occurred between Ritharngu and Ngandi than between Nunggubuyu and Warn­ darag. Ngandi has been more receptive of grammatical substance linguemes than Ritharngu (Heath 1 978 : 1 43 ; he does not indicate overall directionality of lexical borrowings) . Heath compares the Arnhem Land situation to the stable multilingualism that led to convergence in Kupwar (see §8 . 3 .2) . On the one hand, the Amhem Land language speakers were not as bilingual as those in Kupwar (Heath 1978: 1 42). On the other hand, bilingualism in Amhem Land is intrafamilial, and t he distinct languages were not as sharply focused as those in Kupwar (Heath

Hybridization in language

217

1 978: 1 42-3; H eat h 1 98 1 : 3 6 1 indicates that the languages are still quite focuse d) . I suggested in §8 . 3 . 2 that a coverge nce si t uation requ i res some degree of openness to the i ntroduction o f schem atic l inguemes as well as a high degree of b i l i ngual­ ism . It m ay be that the more intimate re l at ion shi p of marriage between spe akers of different languages in Arnhem Land led to a high rate of d iffusi on of sub­ stance linguem e s as we ll . Unlike Me d nyj Aleut and Michif however, th e re is not the massi ve introduction of whole integrated grammatical subsystems (such as the Russian verbal inflections in M ed nyj Aleut) . Th e difference in degree s of mixing of s ubst a n ce linguemes may be due to the fact that the entry of the Russians and the French resulted in a more substantial d is rupti on of traditional social patterns co m pa re d with the exogamous marri a ge patterns in Arnhem Land (and possibly elsewhere).

8.4.3.2

Death by borrowing

S o c ieti e s un de rgoing DEATH BY BORROWING are thos e wh i ch a r e d e fyi n g cul­ tural assimilation but gradually adopt more and more of the substance linguemes from the surrounding language, in wh ich t hey are bilingual. The last phase in . this process is what Matras calls FUNCTIONAL T URNOVER (Matras 2000: 1 7-23): the basic vocabu lary and sometimes some grammatical affixes of the original language are restricted to a secret or i n-gr oup regi ster of the now-acq u i red langu age of the external group. One well-documented example of extensive borrowing no t yet appro ach ­ ing l an guage death is the Asia M inor Greek dialects from before W o r ld War I ( D awk i ns 1 9 1 6; see Thomason & Kaufman 1 988:2 1 5-22). These dialects were spo k e n by Greek c o mm u n it ies surrounded by Tur k i sh communities. The Greek speak e r s were commo n l y bilingual, and some Gree k spe akers shifted to Turkish (and also to Islam; Thomason & Kaufman 1 988:2 1 5). The Asia M i no r Greek dialects exhibit extensive borrowing from Turkish of both non­ basic and ba si c vocabulary (Dawkins 1 9 1 6: 1 97- 8), fu nction words including discourse-regulating function words, some derivational morphology and in some Cap padoc ian vi l l a ge s even nom inal and verbal in fl ectio n morphology (see D aw k i ns 1 9 1 6 : 203 for a summary). There has been some a dopti on of schematic linguemes from T urkish , including genitive and re l ative word order, b u t not extensively (Daw k ins 1 9 1 6:20 1 -3 ; Thomason & K au fm an 1 988:22022; we would predict t hi s from th e lack of universal b ilingu al i sm ; see §8.3 .2). Another example of extensive borrowing is the Arabic of Kormakiti in Cyprus. The s peakers are universally bilingual in Greek: massive borrowing of Greek vo cab u lary has bro ugh t in Greek m orph osyn tax as well , while the Arabic words largely retain their Arabic mo rphosyn tax (Newton 1 964; see al so Thomason & Kaufman 1 98 8 : 1 05-7). The speak ers are Maronites who left Lebanon some time after 1 1 9 1 , thus o riginal ly speakers of Arabic (Newton 1 964:43). A t h i rd example of extensive borrowing , albeit not as focused, is Pol i s h Romani (M atras 2000): much nonbas i c v ocabul a ry and some bas ic

218

The descent of languages

voca bul a ry , as well as d i sco u rse- regulating grammar and Slav ic as pectual affixe s , are bo rrowe d fr om Po li sh . Ag ai n , the spe akers are Roma who migrated to Poland. If more and more of the source language were borrowed, there would re­ main little more of the original lan g uage than a c o re o f b a sic v ocabu l a ry and some grammatical inflections and constructions . Yet the borrowing society may still resist assimilation into the surround i n g s ocie ty . Matras argues that for such societies, at some point 'the need to retain a speci al v ariety is str ong er th a n t he abi l ity to transmit a coheren t lingui stic- grammatica l system' (Matras 2000) . At this point , there is a functional turnover: the original langu a ge is abandoned as a potentiall y general-purpose communication system. 10 The rem ai nin g elements of the ori ginal language - mainly its basic vocabulary - are then ava il a ble as a register to employ in order to exclude outsiders from the communicative act. Since the remaining elements include basic vocabulary, they prove useful for t hi s purpose . Matras ar gues that functional turn ove r occurred in the Romani varieties of We stern Europe (including Britain and Scan dinavi a) , called Para-Romani. There is no currently e xisti n g P a ra-R om an i l an gu a ge (Matras 2000), but instead an a bili t y to insert Romani vocabulary into utterances in the local language (e.g. English or S pa nis h) . Matras describes the process of employing Romani lexical items as lexical reorientation, and the retention of some (often fo ssil ized) gram­ matical inflections as selec ti ve replicat io n . 1 1 Matras a lso suggests that the current status of Ma'a (Mbugu) i s a result of functional turnover (Matras 2000). Ma'a is a la n guage possessing a basic vocabulary, including personal, possessive and demon strative pronouns, that is l a rgely S o u t he rn C u s h i tic (Th om as on & K a u fma n 1 988:225), though no t from any one existing Southern Cushitic language (Greenberg 1 999:627). But Ma'a a lso possesses much nonbasic voca bulary and almost all grammat ica l inflec­ tions that agree closely with i t s ne ighbo rin g Bantu languages, Pare and S h ambaa (Th o m a s on & Kaufman 1 988:223). Th omason & Kaufman note that ' di fferen t ( an d indepe ndent) sources present a pict ure of t he Ma'a people as resisters of total cultural assimi l a tion ' (Thomas on & Kaufman 1 988:225; see also refer­ ences in Thomason 1 997:472) . Historically , Ma' a represents a case o f extensive b o rrowin g of Pare/Shambaa grammatical substance linguemes by a Southern-Cushitic-speaking s ociety (Thomason 1 997:478-82); for ex am p l e , early 20th -ce n t ury sources indicate that the Bantu inflections were not obligatory at that time, especially in in­ group communica ti on (Greenberg 1 999:629). However, the contemporary Ma'a spe ake rs are all bilingual in a Bantu language which they call Mbugu which is very similar to Pare (Thomason 1 997:469) . The Ma'a l a n guage appears to be an in-group re gister (Thomason 1 997:473, ci ti n g Mous 1 994), though Thomason notes it is still learned as a first language (Thomason 1 997:476). Hence it is possible that Ma'a has undergone functional turnover with lexical (re) orien tation - more accurately, retention - of mostly Cu shi tic vocabulary and selective re pli cat i on of a limited amount of Cushitic grammatical structure.

Hybridization in language

219

The result o f death by bo rro wing is q u i te different from that of mixed ma rr ia ge languages. In d e a t h by b or ro w ing a n d ulti mately functional turnover, much of the basic v ocab u l a ry i s retained, with perh aps some grammatical inflections, while the rest of the g r am m a tic a l structures are of the e xte r n a l s ociety s l angu age After functional turnover, the original language linguemes represent a secre t or i n gr o up register, only one c o de of the society . 1 2 '

.

-

8.4.3.3

Semi-shift

The third social situation b ree d i ng mixed l angu ages is semi-shift. In SEMI-SHIFT, society appear to shift only part way to the external society's lan gua ge The se m i s h ift may be due to lack of full acce ss to th e external society's langu age or may be a marker of a distinct soci al i de n t ity The lin guist ic result of semi-shift is the mi rr o r im age of functional turnover: the vocabulary is that of the external soci e ty s language, while many gram mat ical inflections (substance linguemes) and most g ram m a tical constructions (schematic linguemes) are at least in part those of the n at iv e l angu age soci ety Semi-shift is thus a sort of extreme case of subs tratum interference, which is the result of a more complete shift to the ex terna l socie ty s lang u age (see §8 . 3 . 1 ). An e xa mp le of semi-shift is Media Lengua, a mixed language in Ecuador (Muysken 1 997). T he Media Lengua c omm u n i t ie s were pro b ab l y mo n ol in gua l Quechua a�ound 1 900 ( M uysk en 1 997 : 374). Then men bega n to work in the S p anis h spe aking cities and became fluent in Sp ani sh Media Len gu a arose as a home lan gu a ge in t h e se communities. Media Lengua is clearly a stable code (i bi d . 407-8). These communities are now mostly trilingual, a l t hou gh the oldest people may be native Quechua spe a k ers and the youngest, native Span i s h speakers ( ibid 374). M ed i a Lengua employs almost en tirely Spani sh vocabu­ lary (89%; ibid. 3 7 8) but almost en tirel y Quechua grammatical inflections (sub­ stance linguemes) and con struction s (schematic linguemes) . Media Lengua does not look like an in te rl a ngua ge : interlan gu age does not u su a ll y contain native l anguage grammat i cal inflections (Muysken 1 997:4057). The Spanish words are regularized and adapted to Quechua phonology for use with Q uechu a inflections, not unlike Spanish borrowings i n to Quech u a In o ther words, semi-shift differs from shift proper: semi-shift is more like a deliberate h alfway shift, in this case by using Spanish vocabulary in a Quechua framework, rather than the stabilization of an inte rlangu a ge form. Another possible example of s em i shi ft is Petj o , forme rly spo ken in Indonesia (van Rheed e n 1 994) . Petjo is spoken by the descendants of Du tch fathers and Malay mothers, a group called Indos (van Rheeden 1 994:224). Thus, Petj o has a social origin not unlike a mixed marriage l an g u age However, it appe ars that Malay was alway s sp o ken alongs ide Petjo , and Dutch was only acquired at a later age, and often not completely ( i b i d ) Petjo is an in group la n guage ; Malay is used for outsi de communication (i b id , 225) . Petjo has approx imately 83% D utc h lexical items, in cludi ng prepositions, conjun cti on s numerals, articles and personal pronouns (ibid., 226--2 7). There

th e speakers in a .

-

,

.

'

-

.

'

-

.

.

.

-

.

.

.

.

,

-

220

The descent of languages

are a significant number of M a lay nouns and adjectives, mos t ly related to local natural and cultural objects, and some po ssessi ves , demonstratives, and the relative clause marker. There is also occasional m ixi n g of verbal p refi xes , e.g. ge-goejoer 'be po u red over' and di-scheiden 'be se par ated ' ( ibid . , 232; Malay in bold). However, most of the syntac tic constructions and word order is calqued on M al ay , as is the ph on ology . Thus, Petjo seems to be almost e n t i rely Dutch in substance linguemes, except fo r the usual sorts of lexical borrowings and a small number of function words, but al most ent i re ly M a lay in schematic linguemes. In fact, Petjo looks l ike an arrested examp l e of a fairly advanced state of second language ac q ui sition of Dutch (van R he eden 1 994: 235), u n l i ke Media Lengua. Given its social ori gi n , Petjo is perhaps more like Mednyj Aleut, wit h most of its lexicon from one langu a ge and a few grammatical substance linguemes from the other language. On the other hand, the minimal presence of M a la y grammatical substance linguemes in Petjo may be due to the minimal presence of grammatical substance linguemes in (Low or Bazaar) Malay ( i b i d . , 229, 2 3 3). Unlike Med n yj Aleut and M ic h if, the second parent language , Dutch, was o n l y rarely a home language. Thus, Petjo may be b etter described as a case of sem i -shi ft towards Dutch, with a deliberate use of what little gr amm a t ical morph ol o gy M al ay m a kes ava ilabl e . In mixed marriage languages, d ea t h by borrowing (and functional turn­ over), and semi-shift, the l in gueme pool contains substan tial numbers of lin g uemes from two parent l a n g u age s . Nevertheless, in each case , the cont ri b ut i on is asymmetrical, and can be motivated by social factors. Semi-shift represents a shift of a society partway but not completely towards another social group and its language, acqui ring its vocabulary but not its gramm ar (neither schematic linguemes nor substance linguemes). Death by borrowing and functional turn­ over represent the cultural de fia n ce of a soc i e t y being overwhelmed by the surro un d i n g soci e t y , re tain i n g its basic vocabulary t i l l the last. M ixed ma rriage lan guage s are the cl osest to true hybrid languages, with grammatical substance linguemes functionally split between the two parent l angu age s ; bu t in the two well-documented cases, there is a predomin ance of the ancestral mothers' 0ri­ ginal l anguage. These three types of social mixing probably do no t exhaust the possibilities, but they appear to acco u nt for the bes t- k nown examples of mixed languages. A model of language p hyl oge ny must allow for reticulation of ph yl ogenetic l ines of descent for whole languages. But the phylogenetic model must also represent the different ways in which the two p a rent l angu a ges contribute to the l ingueme pool of the daugh t e r language, and the asymmetry of the con­ tribution of each parent. It is a l so important to n o te that for all of the mixed languages d i scussed above, many if not all spe a k ers were multilingual with one or both of the parent languages during much of the history of the mixed language . This prolonged multilingualism without (comple te) shift, combined with the development of a social identity for a group distinct from the parent­ language societies, appears to have played the major role in the stabilization of the mixed code in all three types.

Creation of a

new

community

or society

22 1

I s t h ere a way to d e s c ri be the a symmetry of the lin g ueme contributions of the two parents across all mixed language types, such that one parent is primary? If the soc i al classification of mixed languages proposed here is valid , then de a th by b o rrowi n g and se mi - s h ift can be considered a s s pec i a l cases o f e x t re m e bo r ro w i n g and incomplete shift respectively. If so, t h e n the PRIMARY PARENT LANGUAGE is the one contributing the majority of the basic vocabulary li n gu e mes . In d ea th by borrowi ng, t he original l a n gua g e would be primary, as in cases of l e ss extreme b o rro w i n g . Even for functional turnover, one c an describe the secret or in-group register as a code where the or igin a l language w o u l d be pr i mary . I n semi-shift, the acquired l a n g u age would be pri mary , as it is in c a se s of a more complete shift . This leaves the mixed m arri age languages . Mednyj Aleut has a basic voca­ bulary that is largely Aleut, h en ce Aleut would be pr im a ry (in fact, Thom ason & Kaufman consider Mednyj Aleut to be an example of extreme borrowing; Thomason & Kaufman 1 988:237). Michif has a spl i t in basic vocabulary be­ tween French nou n s, adjectives, n um e ra l s and p re po s i t i on s and Cree verbs, demonstratives and postpositions . Cree appears to be dominant, b ased l a rge l y on gr a mmatica l evidence. But the soci al evidence of the mak e - u p of the speaker population also points to primacy of Aleut and Cree : 90% of the population of Copper I sl a nd were Aleut at the time when Mednyj Aleut was formed , and ' the ethnic h eri ta ge of most M i c h if speakers is Cree' (Th oma son & Ka u fm a n 1 988:23 1 ) } 3 Thus, t h e social and l i n g ui s tic patterns point towards Aleut and Cree as pri m a ry parents. The mixed mar ri a ge languages are therefore m o re like b o rrow i ng than s h ift (cf. Thomason & Kaufman 1 98 8 : 2 3 7) .

8.5

Creation of a new community or society

and language types described in §8.4 al l represent the creation of one sort of another; so the headi n g of this section would subsu me those la n g u ag e s as well . The ex a m p l es in t hi s section are different from the preceding examples in that the social si t u a ti on is gene ra l ly much more com plex . The languages desc ri be d in this s e ctio n are those known as lingua The languages

a new so ci ety of

francas , pidgins and creoles. Despite the extreme variation of the soci a l con­ texts in which they arise, the linguistic re s ul t is remarkably s i m i la r , hence the j ustification fo r discussing these languages together . Essentially, the lin gu istic result is th e same as in sh i ft and sem i - shi ft : most of the v oc a bu l ary is d rawn from a single parent language. B u t unlike shift or e v e n semi-shift, the gram­ matical elements -- s ubstance linguemes and sche matic linguemes - are not drawn from the same parent l anguage, and in fact have disputed lineages. Societies of individuals are never com p l etel y communicatively i sola te d (§8.2). Some individuals from one society will interact with other individuals in other s oc ietie s for a variety of reasons, one of the most common rea son s being trade (M iihlhausler 1 986:75). Such con tacts may be re l at ivel y brief or random, but if

222

The descent of languages

they are reasonably stable, then an intersocietal community will develop, in the sense of community defined in §4.2.3 and §7.2. The community's code will develop at the same time, of course . In some cases, the intersocietal code will simply be the language of one of the societies. That is, one society's language will function as a LINGUA FRANCA (Samarin 1 962/ 1 968), and the language is thus extended for use in a new domain . For example, English is a global lingua franca in a wide range of intersocietal communities; individuals from non-English-speaking societies learn English in order to communicate with other individuals. The use of a lingua franca presupposes that members of the oth� society or societies have mastered the lingua franca at least to some degree. This fact in turn presupposes that the members of the other society or societies have suffi­ cient access to the language of the society contributing the lingua franca, that is, there exists the opportunity to master the lingua franca on the part of the former. If this presupposition fails, then PIDGINIZATION is likely to occur. One reason for insufficient access on the part of the other societies using the lingua franca is that the lingua franca is in fact the language of a third society, not the language of the interacting societies. For example, Swahili is spoken natively on the coast of Kenya and Tanzania, but is used as a lingua franca much farther inland. Some of these inland areas use Swahili because Swahili traders settled there but others use Swahili as a second language lingua franca among other Bantu peoples (Holm 1 989:564-6) . The form of Swahili used is considerably simplified from the native language on the coast, a characteristic of pidgins (see below) . Another reason for insufficient access to the language drawn on for the intersocietal code is unwillingness on the part of the society speaking that language to give the other societies access to it. Thomason & Kaufman give a number of examples of withheld languages (Thomason & Kaufman 1 988: 1 757; see also Holm 1 989: 5 1 4). The members of the withholding society sometimes give the impression that the language they are using with outsiders is their own language, or they deliberately avoid using their full language with the other group(s) . These social factors create and maintain social distance between the two societies in question. Social distance between two groups who otherwise inter­ act is the underlying motivation for pidginization: 'what is common to [social contexts of pidginization] is the original geographic, social and linguistic dis­ tance of the parties involved. In most cases, there remains a desire to maintain the social distance, and it is this desire that keeps jargons and pidgins alive' (Miihlhii.usler 1 986:8 1 ; JAROONS are not yet stabilized pidgins). Jargons and the PIDGINS that develop from them are characterized by their simplified form. There are two reasons for simplification of a pidgin. The first is that the pidgin is used for only a single domain, such as trade or tourism (see Miihlhii.usler 1 986:75-84 for typical domains of pidgin use), and hence only a fraction of the entire communicative resources of the language from which its lexicon is drawn (the so-called lexifier language) are necessary for successful communication.

Creation of a new community or society

223

second reason is that since one or both of the interlocutors do not know language, nonconventional coordination devices (see §4 . 3 . l ) are fav ore d over conventi onal coordin ati on . Shared ex perti se in th e domain of trade, etc. , j oint attention in cludin g gestures, joint salience o f common gro un d i n the context, and other nonconventional devices are used to establish joint me anin g (cf. Miihlhiiusler 1 986: 1 37-8; Holm 1 98 8 :73). The con ven tion al gr amm a tic al inflecti on s an d gr amm ati cal constructions of the lexifier language are largely di spen sed with - up to a point. The conven tions of the le xic al source language are preserve d in the pidgin if they more or less coincide with the counterpart co nventi on s in the language(s) o f the other inter locutors . That is, interlocutors will use co nvention al coordina­ tion devices if they coincide . Tho maso n & Kaufman give a nµmber of e xamp l e s of pidgins, includin g many pidgins from non- E uropean lexical source languages, contain i ng rather complex or 'marked' l i n guemes . One of their e x amp l es is Chinook Jargon, used in the Pacific Northwest coast (see als o Thomason 1983). In addition to a range of unusual pho nemes , C hinoo k J argon also possesses p le ona s tic pronouns, a variant VS order, a NEG V S order, and a questi o n p a rticle . All of th e se features are unusual in pidgins, but common to all of the N at i ve American lan gu ages of the area. And, of course, it should not be for­ go t te n that the languages underlying most European-based pidgins (Western Eu ropean languages, West African languages, and Austronesian lan guages) share si gn ifi can t schematic lin guemes, i ncl u ding SVO word order, prepo siti on s, and p reve rb al auxiliaries derived from verbs, all of which are used in the pidgins based on these l an guages. Thomason & Kaufman 's arguments fit in with the analysis of coordination in c o mm un ic a ti on in §4.2.4: converge on the most sa lient coordination device, co n v e n t i on al or n o nco n vent i o n a l . In fact , Thomason & Kaufman's description of the mutual accommodation u nd e rl yi n g pidginization is e sse n tia l l y a de sc r ip ­ tion of a firs t- par ty Schelling game : 'members of the new co n tact c ommu ni ty make guesses about what their interlocutors will understand, and 'right' gue sse s are i n co rp o ra ted into the grammar of the d eve l opi ng co n t ac t l a ngu age' (Th om aso n & Kaufman 1 988: 1 74). The remaining conventional coordination devices in a jargon and a pidgin are its v oc ab u l a ry . The striking fact about almost all pidgins is th a t the major­ ity of the v oca bu lary of each pi d gi n is taken fro m o ne lan guage . Miihlhausler argues that this is a res ul t of stabilization of a j argon into a pid gin (Miihl hau sle r 1 986: 1 43). A l t h o u gh j a rgon s may consist .of a m e l ange of vocabulary, a pidgin will h ave most of its lexicon fr om o ne l an guage for two reasons. If one of the groups of interlocutors is highly mobile, such as European sailors, then the only contact language in common wi th all of the gro up s in the c omm unity is that of t he mobile population. If the two (or more) groups of interlocutors are not mobile, then other factors intervene: the soci al l y dom in an t gro up m ay impo se its vocab u l ary (as in the Europe an slave p lan ta tions in the C ari b be an) , or the gro up in whose territory the c o n t acts are made will i mp o s e its vocabu­ lary (as in the j argo n s of medieval p il grims and modem tourists; Miihlhausler 1 986: 1 43-5). One of the very few exce pti o n s to this prin cip le is Ru ssenorsk , a The

the full

224

The descent of languages

pid gi n used between N orwegian and R ussian traders of the 1 9th century (Broch & J a hr 1 98 1 ; Holm 1 9 89:62 1 -4; Jahr 1 996; Trud gi l l 1 996b : 9- 1 0). Its v ocab u­ lary is a pp ro x im at el y half from each language, reflecti n g the fact that the two social groups were essentially equal in power. 1 4 T h e se tt l in g o n v oc ab u l a ry lar ge ly from a single source l a n gu age makes the pidginization pro cess look like shift . Thom aso n & Kau fm a n argue that the lan guage that is being learned is not the lexica l source la ngu age but a s i mpl ified version of it (Th o m a s on & Kaufman 1 98 8 : 1 7 8). Moreover, the simp lificati on is not due to im pe rfect learning but is a product of the lexifier language's n at i ve speakers (ibid. ) . 1 5 Nevertheless, it is st i l l the resu l t of shift to the simplified code by nonspeakers, and has similar linguistic effects: the nonspeakers acq uire the vocabulary and some schematic linguemes ( g r ammat i c a l st ruc t u re s) of the sim­ plified code, and i m port some of their own schematic l in gu emes (Thomason & K aufman 1 98 8 : 1 93 ; cf. also the wide spread occurre n ce of E urope an words in A tlantic creoles with A frican semantic r an ge , po lyse my , syntactic behavior or ,

calques; H olm 1 98 8 : 82-9 ) .

The origin of the l inguem es i n E uro pe a n la n gu age- ba sed pid gins i s h i ghly varied . In add i ti on to vocabulary fro m the lexical source language, there is vocabulary from substratum la nguage s, a n d sch e m atic l i n g ueme s (gramm a t ica l patterns) from a variety of sources. Even in an apparently s trai ghtfo rwa rd bil ateral trade situation, such as t h a t which gave rise to p id gi n s such as Russenorsk, the linguemes come from a v ar iety of sources: 'be side s a few words from L app i sh, Finnish, Swedish, and French , there are a number of words from Dutch or Low German (e .g. gro t 'big' , junka 'boy') and En gl ish (e . g . jes 'yes' , ju 'you ' , verrigod 'very good') and international nautical jargon (e.g. skaffom 'to eat')' ( H olm 1 98 9 : 62 1 , citin g an unpu b lished MS by J . A . Fox). From a sociolinguistic point of view, this is not surpri s in g . The t rad ing and colonial communities which used these pidgin s , and pidgin communities in general, were and are pe rhaps the most loose-knit social ne tw ork s i n existe nce. S peak e rs are often very mobile. with con t acts with m any societies; ne w slaves and laborers were co n s t an tl y and rapidly be ing added to (and d ying off in) the colonial plantations. The pidgin speakers are also s ociall y quite d i s ta nt from each other, as noted above, typically w i t h un i p l e x links to other spe akers (e.g. only trade). In accordance with the M i l roy s ' theory of introducers be ing weak -tie i ndividuals (§7 .4 . l ), virtually all p idgi n speakers are in a position to i n t rod uce and propagate l i nguem es from almost every code with which th ey come into contact. Thus, there is proba b l y some truth to all of the theori e s proposed for the ori gin s of Eu ropean - b ased p id gi ns - mo noge nesi s from an Atlantic nautical j a rgo n , substratum influences from Africa and Oceania, 'relexification ' (sh i ft) from one Eu ro pe a n l e xic al pidgi n to an other, not to mention m o tiv ated uni­ versal de ve l o pments (see Holm 1 9 8 8 : 27-70 for an o vervi ew) . Nevertheless, for all pi dgi n s, on e c an i d e nt ify the lexical source l anguage as the primary parent in the same way as for the mixed languages in §8 . 3 . 3 : it is the parent providing most of the b asic v oca bul a ry linguemes.

Language death: selection

at

the societal level?

225

In some cases, the extraordin arily loose-knit network (or part of it) may coalesce into a new society, with a full array of domains . In th i s case, the pidgin expands, becoming the code for more and more social domains, and undergoes CREOLIZATJON, that is, the language becomes a native language for the younger generation of users - the most intimate social domain. A sharp line is generally drawn between an EXPANDED PIDGIN (still a second language for its users) and a CREOLE (a first language for its users). However, a sharp line cannot be drawn in reality. No clear cases of ABRUPT CREOLIZATION - nativization of a jargon - have been attested (Holm 1 989:63 1 , Miihlhausler 1 986: 8), and rather few examples of nativization from an unexpanded pidgin (Muhlbausler 1 986:8). In the most commonly attested case, nativization of an expanded pidgin, 'competent adult non-native speakers always coexist with the children growing up as native speakers, making it impossible to determine with any certainty which group is contributing what to the structure of the emerging creole' (H olm 1 989:63 1 ). Also, creolizing societies are multilingual societies (after all, the parents' generation are not native speakers), and the children may grow up multilingual in the pidgin/creole and other languages of the society, thereby eroding the distinction between native language and second language. The evolutionary status of a creole developing from an expanded pidgin is fairly straightforward . Its linguemes generally come from the pidgin (although some may be borrowed from other languages spoken in the society), or are created from the pidgin's linguemes via altered replication, particularly gram­ maticalization (Mufwene 1 996b). The problems in the phylogeny of a creole are essentially inherited from its pidgin parent. 1 6

8.6

Language death: selection at the societal level?

Language death is a result of the differential survival of whole languages. It is possible to analyze language death as selection at a higher level of organiza­ tion. In order to do so, one must identify the replicator, the interactor, the environment and the relevant causal mechanisms for differential replication and selection at the level of a whole language. ln this analysis, the interactor is the society as a whole . The relevant organic structure that the society possesses is the social domains of language use and their interrelationships (Fishman 1 965/ 1 972; see §4.2. 3). Fishman ( 1 972) sug­ gests a con crete mechanism for selection . He proposes that language main ten­ ance in multilingual communities is supported by a sharp differentiation of social domains (and presumably, their stability over time; Fishman 1 972: 5 1 ) . If there is no sharp differentiation, or if the structure of social domains is dis­ rupted (e.g. by the impact of European invaders and colonizers), then one language may invade the social domain of the other and ultimately replace it. The interaction of the society with its environment - other societies with which it comes into contact mostly, but possibly also its physical env i ro nment - thus leads to the survival/extinction of the replicator, which is the language's

226

The descent of languages

lingueme pool taken as a whole (cf. Mufwene to appear). The interactor - the society - can also cause altered replication. For instance, a society can alter its language through borrowing, coinage, calquing and creation of new construc­ tions, and development of a written style in order to extend the language to new domains or stylistic registers - or by not doing so, thereby allowing an­ other language to occupy that social niche instead. This is a process of social evolution . The processes described by Fishman are essentially social: contact between societies (including immigration and conquest), and changes in the organization of domains of social interaction . These social processes happen to have linguistic consequences because of how languages are identified with the societies that speak them, or are identified with particular social domains in a multilingual society. This may not be the correct analysis. After all, in the standard neo-Darwinian model, competition between species is usually reduced to competition between organisms belonging to those species, and selection is said to operate at that level only: '[Species] compete, but probably competition between organisms of t}J.e same and different species is more important than competition between one species and another species' (Ghiselin 1 98 7 : 1 4 1 , cited in Hull 1 98 8 : 2 1 9). It may be that the social processes in choosing which language to use in a society should be analyzed as choices made by speakers: they do not replicate the linguemes of their traditional language in utterances in a community, or they even abandon the community altogether (e.g. abandoning a traditional reli­ gion). Both o f these choices would result in the extinction of the language of the community. 1 7 Nevertheless, some biologists argue for selection at the species level, arguing that species have a population structure that can be replicated. Likewise, the proliferation of European standard languages at the expense of indigenous languages may be due in part to their social structure. That is, European standard languages are already fully developed for use in the full array of domains of expertise of the modern world, with vocabulary and writing sys­ tems and the technology to support them (cf. §3 .5). The sociolinguistic struc­ ture of the European standard languages is replicated when indigenous groups join the modern world (so to speak). That is, the European standard languages proliferate in the indigenous groups' society and the indigenous languages go extinct. It should go without saying that the analysis of the facts of language death and language shift in terms of selection at the language level should NOT be given an evaluative interpretation such that the indigenous languages are 'in­ ferior'. That would be social Darwinism. Consider again the biological parallel. European species such as starlings and Mediterranean annual grasses have invaded and eliminated North American songbird species and perennial bunch­ grasses respectively. Biologists do not assume that this fact demonstrates that the native American species are 'inferior' . On the contrary, biologists are in the vanguard of the movement to save endangered species, and linguists should be in the vanguard of the movement to save endangered languages. ·

·

Notes

227

Notes However, as will be seen below , I will adopt . much of Thomason & Kaufman's model of the relationship between society and language in cont a ct situations; their assumptions about genetic descent do not invalidate their social and l ing ui stic analyses of contact situations . 2

Thom a s on & K aufm a n actually split this assumption into two : genetic relation­

ship s can be defined only in cases of n o rmal transmission, and language s transmit­ ted by normal transmission cannot have multiple p a rents (Thomason & K aufm an 1 98 8 : 1 00- 1 1 ) . 3

Bilingual speakers are typically weak - tie members of the society (§7 .4. 1 ) : their bilingualism reflects t heir po s i t ion between two societies. In the case of multilingual societ ies , however, speakers of the s oci et y ' s languages are not necessarily weak-tie members .

4

See also Thomason & Kaufman 1 98 8 : 1 29 , where they suggest that Gaelic loanwords

5

Matras calls this combination 'convergence' (Matras 2000).

into Irish 6

E ngl i sh

were introduced by English speakers.

Tho m ason & Kaufman argue that there was already a fair degree of structural similarity between the three lan g uages in Kupwar , due in fact to the ancient shift of Dravidian speakers to Jndic languages (Thomason & Kaufman 1 9 8 8 : 8 6) .

7

Trudgill considers reallocation a problem for the theory of accommodation (Trudgill 1 986: 1 25). As noted above, no society is completely homogeneous, and th u s no

society will be. completely focuse d , so variants may remai n

to

signal

the heteroge­

neity of the s o ci ety .

8

The sociohistorical information in this para gr aph is d rawn from Thomason &

9

& Kaufman for further references. T ru d gill 1 996b describes these languages as 'dual-source creoloids' , that is, lan­

Kaufman 1 9 8 8 : 2 3 3 - 8 , in tum taken largely from Menovscikov 1 969; see Thomason

gu a ges which underwent significant ch an ges due to shift ('creoloid') but have two parent lan guages .

JO

M atras argues that functional turnover is a relatively ab rupt process, not the end result of gradual borrowing (Matras 2000); there is no assumption in the passage above that functional turnover is the endpoint of

a

gradual borrowing

process , despite my name for the general case. 11

Matras also cites some ex amples of secret-language regi s ters in Germ a ny such as Leko u de sch, used by Jewish cattle traders until 1 9 3 3 and drawing on Ashkenazic

Hebrew; and Jenisch, a generic term for secret lexicons including items from Rom­ ance, Ashkenazic Hebrew and Romani used by commercial nomads (Matras 2000) .

It appears that these registers originated at least partly through contact with Romani speakers and may represent a sort of adoption or borrowing of Romani, etc. 12

v oc abulary for a secret register (ibid . ) . Trud gill l 996b d escribes a s imilar case as a ' reverse creoloid ' , giv in g the example o f Shetland Island Scots, which retains ' con si d erable amounts of Scandinavian (Norn) influence , particularly in lexis' (ibid . 1 2).

13

Bakker & Papen state that the Metis formed their own distinct cultural identity (Bakker & Papen 1 99 7 : 296) . However, they note that in one the earliest phases of the formation of the Metis culture, during the late 1 8 th century, the children French-Indian unions ' generally continued to live with their Indian mothers '

of

(ibid.

297), and that in the e arly 20th century, 'Cree monolingualism was not at all

228

14

15

16

17

The descent of languages uncommon among uneducated Metis' (i bi d . 30 1 ) . Both of these observations sug­ gest a cultu ral orientation of the Metis more towards the Indians than towards the French. Miihlhiiusler suggests that Russenorsk remained a jargon, since some lexica l items varied between Russian and Norwegian sources (Miihlhiiusler 1 986: 1 43). However, Jahr ( 1 996) argues convincingly th a t Russenorsk is a stabilized pidgin. In fact, this is not always true : in the example of Swahili given above, simplification is partly due to the behavior of nonnative speakers in using it as a lingua franca. However, this may be more like the simplification process that sometimes occurs in shift ( see §8.4. 1 ) than pidginization. Thomason & Kaufman propose that mutual accommodation, the same mechanism as they propose for pidginization, is involved (Thomason & Kaufman 1 988: 1 53). In other words, the process for abrupt creolization (and its problems for phylogeny) is essentially the same as those for pidginization discussed above. In fact, selection at the societal level requires that the replicator be the lingueme pool taken as a whole: one language simply replaces another in the social domain. The cases of mixing described in §§8 . 3 -8.4 would not count as differential replica­ ti on at the societal level .

Chapter 9

Towa rds an evo l utionary l i ng u istics

This book has presented a program for rethin k i n g l an guage change in evolu­ tionary terms. Two fundamental these s underlie this framework. Fi r st there is a general theory of selection, most fully developed in evolutionary biology, which shows that evolutionary processes operate at two leve ls, rep l ication and selection (Hull 1 988). Second, this model can be applied to language by t a kin g the speaker as the unit of selection and the l ingue me as the un it of replication; this is the Theory of Utterance Selection (§2.4. 1 ) . The re sult is a framework for language change enco m pa s s i n g both innovati on and propaga­ tion in language change. The fra me wo rk firmly si tua tes both innovation and p rop aga tion in a uniform model of l anguage change th rou gh language u se (§2.5; ch a pter 3). But m u ch of the presentation has implications for language taken as a wh o le . In fact, any theory of language change gi ves one the o ppo rt unity to look at language as a whole. Language ch a nge cannot be separ ated from lan­ guage structure (pho n ol o gy morphology, syntax), language function (seman­ tics, p ragmatics , discourse analysis, and phonetics with respect to phonology), l an guage in the mind (psycholinguistics, first and second l an gu age acquisition), language in society (sociolinguistics), language variation (sociolinguistics again), or language diversity (typology, ge n e t ic linguistics, l angu age contact studies, pidgin and creole linguistics) . This has been dem on strated by the fact that all of th e se areas of linguistics had to be drawn on in some way or another in the course of arguing for the ev o l uti o n a ry framework for language change in this book. In fact, language change is the g l u e that holds all of th ese facets of language together . The study of language c hange can easily be transmuted into the study of l angu age itself, because language is fun d am e ntal l y a variable, dyn ami c phe­ nomenon . The focus on a l i n gu i st ic system described and criticized in § 1 . 1 has obscured this fact for the l ast century. But, as argued there a lingui sti c system taken in iso l ation is a mythical object, neither a type nor a token, and hence not a prope r object of sc ien tific study The prope r objects of the study of language are actually occurring utterances and an actual speaker's k n owledge about her language. An ev olu ti on a ry analysis of actual language (utterances and speaker know­ ledge) p os i ts the existence of further entities: l i n guemes and their interrelation­ ship s in utterances and in the speaker's mind (§2 .4; chapter 6); the delimitation of p op ul atio ns of speake rs and utterances into communities and societies and the l anguages/co de s associated with them respectively (§4.2 . 3); and co nven tion s ,

,

,

.

229

230

Towards an evolutionary linguistics

of language as a speaker's knowledge of the common ground between her and her interlocutors (§4. 2 . 4). These concepts reintroduce the notion of a li n gui sti c

system, now empirically firmly grounded in existing speakers and utterances. But the linguistic system that the evolutionary framework reveals is quite dif­ ferent from that posited in the structuralist and gene rative traditions. What does the linguistic system look like from an evolutionary perspective? The fact that l anguage is so prone to hybridization (mixing) in utterances, dialects and l anguages offers a hint as to the true nature of the linguistic system. Grant makes an interesting and important observation as to why plants are more prone to hybridization and hybrid speciation than animals are. It is worth quoting him at len gth : An obvious and basic difference between pl a nts and animals lies in

the degree of The devel opm ent of the re l at ively simple plant bo dy is c o mm en su ra te with an open system of growth by which new parts are built up in series. The animal body is a vastly more complex and delicately balanced org aniza ti on and, furthermore, is one wh ich m u st deve l o p as a who le and by a closed system of growth. It is logical to suppose that the gene s ys tem s co n troll in g growth and development differ correspondingly in co mp lexi ty and i n te gra ti o n in the two types of o rgan isms This single premise will account for the two main factors contributing to the pa rti a l or complete breakdown of species o rg anizati o n in many plant groups . . . A sp ecies po ssessing a high ly integ rated and finely balanced genotype will suffer rel at ively worse effects from inters pecific hyb ridizati on than will a species with a simpler and more l oo sel y coordinated set of genes. The relat i vely s impl e physiological-morphological organization of the plant bod y reflecting a cor­ respo nd in gly relatively simple genotype, probably gives plants a greater tolerance for interspecific gene ex ch a n ge than is present in most animal groups. (Grant co m ple xi ty of the i ndivi d ual o rga n is m .

.

,

1 9 8 1 : 7 5 -6)

Grant's description of the organization of the animal genotype and phenotype is reminiscent of the structuralist description of a language system: a complex and finely balanced system, 'ou tout se tient' in the famous phrase. This model of a language system may be derived in turn from the 1 9th-century model of a language as an organism - as an animal in particular, due to zo ocentrism I suggest in stead that a language system is more like a plant than an animal. This suggestion is based on the high receptivity of languages to change by contact with other languages, often ve ry different in grammatical s tructure (Thomason & Kaufman 1 98 8 : 1 4 -20), not to mention their high degree of com­ municative flexibility as described in §4.3. In essence, any lingueme can be borrowed without destroying the communicative power of a language: 'as far as the linguistic possibilities go, any linguistic feature can be transferred from any l anguage to any other language' (Thomason & Kaufman 1 988: 1 4); 'given enough time and intensity of contact, virtually anything can (ultimately) be b orrowed (Harris & Campbell 1 995: 1 49). A language is a loosely coordinated set of linguemes that is relatively simple compared with a truly finely balanced and complex system. Likewise, a speaker's .

'

Towards an evolutionary linguistics

23 1

k nowle dge ab ou t h e r la ngu age is not as finely ba l a nced a n d com p le x a men t al structure as s o me co n tem po ra ry gram ma tica l theories make it out to be. As B o l in ger w ri t e s , Language may be an edifice where everything hangs together, but it has more patching and gl uing about it than it has architectonics . . . a brick can crumble

here and a termite can nibble there without setting off tremors from cellar to attic. I wan t to suggest that language is a structure, but in some ways a jerry-built

structure. (Bolinger 1 976: I )

O f course, th e em p ha s i s should be on the qualifier 'relatively'. Plants are extremely complex organisms co mpared with so-called l ower forms of life, such as singl e - ce lled o rga n isms. Plants are relatively simple o n l y in co m p ar i s o n with animals. Likewise, h um an language systems are extremely complex com­ pared with animal communication s yste ms and various o the r sign sys tem s , such as mu s ic al notation or the iconography of road signs. Nevertheless, the remarkable similarity of the phyl ogen et i c patterns in l an gua ges an d p l an t s suggests that the structuralist model of a l a n gu age is too rig i d . I n st e a d , a s im pl i e d in this book , the linguistic system is not rigid, homogeneous, self­ contained, or 'finely balanced' .

A l t ho ug h li n gue me s a re o rgan i ze d in relationships of inclusiveness, high er­ s pecify their lower-order pa rts i ndep end en t of th e auto­ nomous organization of the lower-order structures (§2.4.3). L ingueme s make up a l i n gu em e poo l replete with alleles (variants) which occur in various fre­ q ue ncie s in a langu a ge (p op ul a tio n of utterances), and in a spe ak er ' s mind re lat ive to her exposure to the l angu a ge (§3 . 3 . l ; chapter 7). Th e se frequencies are dyn amic, changing incrementally with every u tte ran ce spoken and heard . Lin guem e s are sufficiently loosely integrated that they can change relatively independently, and even be transferred from one phylogenetic line to another (chap te r 8). Communities and societies a re not sharply de l i n e ate d . Members of a soc i e t y are simultaneously mem bers of multiple communities (§4.2. 3; §7.2), and no society is totally com m unic ati v e l y i solated from other societies (ch apter 8). Nor are communities and societies homogeneous . Within a s oc i e ty and/or com ­ m un i t y , membe rs themselves are differentiated by network tie stren gth (§7 .4) and by th e amount and type of c ommo n ground the y share with other members of the comm un i ty (§7 . 2) . The he ter o gen eity and interpenetr at i o n of societies and communities is of co u r s e directly reflected in the hete r ogene i ty of the lan gua ge s o rde r l i n gu e m e s often

delimited by them .

for communica tion - communication of social as semantic con ten t (chapter 4; §7.4.2). But convention m u s t alw ay s be supplemented by nonconventional devices for c omm u n i cati on (§4. 3 . 1 ). And the line between conve n tion al and n onconventional communica­ tion is not sh a rp (§4.3). T his fact a l l ow s the interlocutors to reanalyze the re l ati onship between form a n d meaning, thereby allo wing internal innovations to be spawned (chapters 5-6). Finally, conven tions themselves are never 1 00% Con ventions are conventions

ide n t i fi cat i o n as we l l

232

Towards an evolutionary linguistics

invariant; conventions are always in the process of being acquired, replaced or lost (§7 .3). The evolutionary framework for language EVOLUTIONARY LINGUISTICS wi l l req u i re a reth i nking of the model of grammatical knowledge developed by grammatical theory . The result of that rethinking will look quite different from most contemporary grammatical theories (see Croft to appear b). But evolu­ tionary linguistics also offers hope for a reintegration of the now fragmented field of l i ngu i sti cs with the aim of producing a coherent theory of language taken as a whole. -

,

G lossary of te rms

This gl ossary includes terms from evolutionary biology, philosophy , psycho­

logy, historical linguistics and sociolinguistics that have been used in the text. Each entry contains the term, the section(s) where it is introduced and first discussed, a brief definition, an d a reference, usually to a general book which defines the term or (for less widely accepted terms) to the introducer of the term . Absence of a re fere nce indicates terms wi th speci al ized meanings intro­ duced in this book. abrupt creolization see cre o l e accommodation §7.4.2 a mechanism for selection, by wh ic h a spe a k e r alters her utterance to resemble more closely the conventions of her interlocutors (Giles & Smith 1 979) actuation see i n no v a ti o n act of iden tity §7.4.2 a mechanism for selection, by which a speaker alters her utterance to resemble more closely the conventions of a community with which she wants to identify herself ( LePage & Tabouret-Keller 1 985) age-graded variation § 3 2 variation in a speec h community that does not represe nt change in progress but instead patterns of speec h ado pted by each cohort as they age (Chambers 1 995) allele §2.4. I on e of a set of genes that occurs at the same locus in a chromo­ some (Dawkins 1 982b) altered replication see replication analogy § 3 4 . 1 an innovation in a lingueme (typically, a mem ber of a mor­ ph o logical paradigm) that is based on the form of a cl o se ly related lingueme (McMahon 1 994) artifactual phenomenon §3.3.2 a phenomenon that is created by human design, .

,

.

such as a tool

variety §2.2 a variety whose speakers perceive them se lves as be in g li n gu istically distinct from any other speech comm unity (Ch ambers & Trudgill 1 980) borrowing §8.3. 1 , §8.3.2 the result of l angu a ge contact on a s o ciety attempt­ ing to maintain its l angua ge ; a li ngu eme contributed by the encroaching l angua ge (Thomason & Kaufman 1 98 8) child-based theory §3 .2 a theo ry of language ch ange that treats a speaker's grammar as th e replicator; altered replic a ti o n occurs th rough a child learn­ ing a language, and selecti o n occurs th rough the mort al ity of the older gen­ eration of speake rs autonomous

233

234

Glossary of terms

code §4.2.3 a language specific t o a single community o r domain ( Wardhaugh 1 992) code-mixing §8 .4. 1 the production of utterances in which linguemes from two different languages are mixed (usually, within a clause) to the point that it is difficult to specify which language is being spoken (Romaine 1 995) code-switching §8 .4. 1 the process of using two languages in a single conversa­ tion in a single social setting, switching from one to the other typically above the clause level (Romaine 1 995) common ground §4. 2 . 3 mutual knowledge shared by a set of people. Com­ munal common ground is shared by a community by virtue of shared expert­ ise; personal common ground is shared by individuals by virtue of shared experience, direct ( perceptual basis) or indirect (discourse basis) (Clark 1 996) . communal common ground see common ground communal lexicon §4.2 . 3 the lexical items specific t o a particular community (Clark 1 996) communicative isolation §2.2 t he absence of significant communicative inter­ action between populations of speakers community §4. 2 . 3 a population of speakers defined by some sort of shared expertise (Clark 1 996) community's meaning §4. 3 . 2 a lineage of the meanings (including full con­ textual meaning) for a particular word or construction produced on occa­ sions of use in a community competence §3.4.2 a particular speaker's knowledge about the linguistic con­ ventions of the communities to which she belongs conceptual space §6 . 2 . 3 t he organization in the mind of concepts with links of varying kinds and varyin g degrees of closeness; the space over which intraference takes place (Croft to appear b) convention §4.2.4 an arbitrary regularity of behavior (a coordination device) which is common ground in a community, used for a recurrent coordination problem ( Lewis 1 969). A linguistic expression is a linguistic convention. A commonly used term for con vention in sociolinguistics is 'norm' (e .g. Milroy '

'

1 992b) .

convergence §8 . 3 . 2 a linguistic situation arising when two o r more groups speaking different languages cohabit, becoming multilingual but retaining their languages (Gumperz & Wilson 1 9 7 1 ) coordination device §4.2.4 a behavior tha t is employed by individuals to solve a coordination problem (Lewis 1 969). Coordination devices include conven­ tion, joint salience, explicit agreement and precedent. coordination problem §4.2.4 a situation where two people attempt to con­ verge on the same solution to a problem, overcoming the fact that each cannot read the other's mind (Lewis 1 969) core expertise §4 . 2 . 3 that shared expertise which is found in a large number of communities covert prestige §7 .4.2 the type of prestige gained by a speech community in a subordinate position in society with which speakers want to identify by using lingueme variants of that speech community (Wardhaugh 1 992)

Glossary of terms

235

creole, creolization § 8 6 a p idgin that has become the first language of a group o f spe a ke rs , and the process by which ch i l d ren acqui re a pid g in and expand it (Miihlhausler 1 986). If t he creole arises through children acquiri ng a jargo n rather th a n a pidgin, the process is called abrupt cre olizati o n . cryptanalysis §5 . 5 a type o f form-function reanalysi s where the speaker inserts an element overtly expressing a m eani ng that is covertly entailed by another element in the construction death by borrowing §8.4. 3 a process, resulting in a type of mixed language where the spe akers are resisting an en cro ach i ng society but borrowing more and more linguemes from t h e encroaching society's language deme §2.2 a p op ula ti on in a species which has a high likelihood of inter­ breeding; less inclusive than a geographical race (Hull 1 988) density §7.2 see network dictionary §4. 3 . 2 the description of word meanings in the l exi co n accord ing to a view that th e me a n i n g of a word or construction can be re d uced to a small finite set of semantic features (Haiman I 98 0 a) differential replication see re p l icat i o n diffuse §8.4. I t h e description of a society characterized by weak ties; its language is expected to have more variants and fewer fixed conventions (Le Page & Tabouret-Keller 1 985) diffusion see p rop agati o n diglossia §2.2, §4. 2 . 3 the state of affai rs in a society in which two lan gu ages exist but each language is used in different domains. The H varie t y i s typ i c ally a second language and the L v arie ty is a first lan gu age (Ferguson 1 959/ 1 972) . discourse basis §4. 2 . 3 the experience shared by individuals by virtue of b ei ng reported by one individual to the other, which is one o f the shared bases for pe rs on al common ground (Clark 1 99 6 ; Clark calls this a ct io nal b as is ) DNA § 1 . 1 the m o lecu le that contains the genes of an organism (Hull 1 988). domain §4 2 3 an area of human experience, which can be characterized by shared expertise, and gives rise to a co m m un i ty and a code for that community ( Wa rdh augh 1 992) downward specification §2.4 . 3 the view that in fo rm at ion combining features of two different linguistic levels in the h ie ra rchy of inclusiveness of linguemes should be represented at the more inclusive level drag-chain §3.4. l a change which is explained as arising in order to fill a gap in a l a ngu age system (McMahon 1 994) drift §2 . 3 a shift in gene (lingueme) frequencies that occurs through altered replication but without selection (Hull 1 988) early adopters §7 4 3 central members of a social network who are hypo­ thesized to t ran sm it innovations from innovators to the rest of the network (M i l ro y & Mil r oy 1 985) economy principle §3 .4.4 the pri n c i ple that spe a ke rs minimize the effort in­ volved in their l in guistic expre ssi ons (Keller 1 9901 1 994); m otivated by th e immediacy premise and constrained by the extent of the interlocutor's common ground .

­

'

.

.

.

.

'

236

Glossary of terms

encyclopedia §4 . 3 . 2 the description of word meanings in the lexicon accord­ ing to a view that the meaning of a word or construction must access all of the knowledge associated with the concept in question (Haiman l 980a) entrenchmen t §2 .4.2 the psychological routinization of a behavior, such as the behavior of recognizing a linguistic expression and producing it (Langacker 1 987) environment §2 . 3 the complex system with an interactor interacts as a co­ hesive whole, thereby causing differential replication of the relevant replicators (Hull 1 988) erosion §6. 3 . 2 the stage in the grammaticalization cycle at which a gram­ maticalizing expression is phonologically and morphologically reduced (Ludtke 1 986: Keller 1 990/ 1 994) �ntialism §2 . 2 the view that a species is defined by a set of necessary properties of the organisms making up that species (Mayr 1 982), or that a language is defined by a set of structural properties of that language (Chambers & Trudgill 1 9 80) etymology §2.4. 3 a lineage for a word ex a ptation see hypoanalysis ex p anded pidgin §8 . 5 a pid gin that is used in a wide ran ge of domains (com­ munities) in a society ( M iihlhausler 1 986) explicit agreement §4 . 3 1 a nonconventional coordination device which solves a coordination problem by a preset, explicit agreement between the individuals (Lewis 1 969) extension see intraference external change § l . 2 a language change that arises as a result of contact between two societies speaking different languages (Hock 1 986) First I.aw of Propagation §7 . 3 the observation that competition between variants in a community always proceeds in the direction of establishing one conventional variant for a specific meaning in a particular community focused §8.4. l the description of a society characterized by strong ties; its language is expected to have fewer v ariants and more fixed conventions (LePage & Tabouret-K eller l 9 8 5) form-function reanalysis §5. 1 a mechanism for innovation in language change by which a speaker alters the mapping between formal component elements of a n utterance and the elements of the semantic structure conveyed by the utterance fudged form §6. l an intermediate variant of a lingueme arising through con­ tact between two existing varieties (Chambers & Trudgill 1 980) fudged Ject §2. 2 , §7.4 . 3 a language variety possessing an intermediate variant arising through contact between two existing varieties (Chambers & Trudgill 1 980) functional explanation §3.4 an explanation of a language change by appeal to an intentional mechanism functional turnov er §8 .4.3 the process by which a speech community en­ croached upon by another speech community gives up their language as a

Glossary of terms

237

general purpose com m u n i c at i o n system, but retains its lexicon and p os sibl y some con s t r uct ion s as a spec i al ized register (code) (M a t r a s 2000)

fusion §6 . 3 . 2 the stage in the grammaticalization cycl e at which a grammat­ i c alizi n g expression is established as a fixed c o n ve n ti on a l unit of t he language (Ludtke 1 98 6 , Keller 1 990/ 1 994) gene §2.4 the basic unit of he red ity in an organism, con s ti t u ted in the DNA; also the replicator in the basic selection process in bi o l o gical evolution (Dawkins 1 982b) gene pool §2 .4. 1 all of the genes fou n d in a population of orga ni sm s (Dawki n s 1 982b) genotype §2. 5 t h e set of genes that define th e phenotype of a parti c u l a r o rg an i sm (Dawkins 1 98 2b) geographical race §2 . 2 a p o pul a tio n that is less inclusive than a species but is d e fin e d by a geographical s ub-area occupied by the species; more inclusive than a deme (Hull 1 988) grammar §2 . 4 . l a n actual speaker's k now l e dge about her language; the replicator in ch i l d - base d theories of language c h a nge grammaticalization §3 . 3 . 3 , §6. 3 t h e process by which a c onst ru c t i on with full lexical items ac q u i re s gramma tical functions the re by turni n g the l exica l item(s) into g ramm a tic al mor phem es (Ho ppe r & Traugott 1 993). The v a ri o u s gram­ maticalization p ro c es s es are given i n §6. 3 . l , T a ble 6. 1 . Grammaticalization processes can be a r ra n ge d in a grammaticalization cyc l e of pe r i phr a s i s­ fusion-erosion .

grammaticalization chain §2 . 4 . 3 a li n eage of a construction and its gram­ matical morphemes (Heine, C la u d i & H iinnemeyer 1 99 1 ) heteronomous variety §2.2 a variety whose s peake r s perceive themselves as being l in gu i stic al l y a part of another speech commu n i t y (Chambers & Ttudgill 1 980) hybrid language §8.4.2 a lan guage cre a te d by the m ergi n g of the l i ng u eme pools of two other languages. Occurs only between closely re l a te d languages or dialects (cf. mixed languages). hybridization §8 . l the p r o ces s by which a new organism is created by the in te rbreed in g of two o r gan i s m s from different spec ie s (Grant 1 98 1 ) hyperanalysis §5.2 a typ e of form-function re ana lysis wh e re t he s pe a k er re­ moves a semantic feature that has been inherent in an element of t h e co n ­ struction and maps it onto the context (i nc l udin g l i ngui stic context) instead hypercorrection §3 .4.4 a mechanism of innovation in sound cha nge wh ere the hearer fact o rs out a p hon e t i c feature in t he signal th at is part of t h e inherent specification of a p h one me (Ohala 1 98 I ) bypoanalysis §5 . 3 a type of form -function re ana ly s i s where the speaker takes a sem a nt i c feature of the context and m a p s it onto an inherent feat ure of an element of the c o n s t ruct i o n . Same as ex apt a ti o n (Lass 1 990), regram­ maticalization (Greenberg 1 99 I ). hypocorrection §3 .4.4 a mecha nis m of i nn ov a ti o n in s o u n d change where the hearer interprets a contextually determined p hon e t i c feature in the signal as part of the in he ren t s pecifi cat ion of a pho neme (Ohala 1 98 1 )

238

Glossary of terms

immediacy premise §4.2.4 the assumption by interlocutors that the speaker will make the coordination problem posed by communication as quick to solve as possible (Clark 1 996) individual's meaning §4. 3.2 an individual speaker's actual knowledge about the lineage of the meanings (including full contextual meaning) for a particu­ lar word or construction produced on occasions of use to which she has been exposed (including her own uses) inherent change § 1 .2 changes to a single enduring entity over time (Hull 1 988) innovation § 1 . 2 the creation of a novel variant by altered replication of a lingueme in an utterance innovator see introducer intentional mechanism §3.4 a mechanism of altered replication or selection of linguemes by which a speaker produces a variant in order to achieve some other goal in the communicative interaction interactor §2. 3 the entity that interacts with its environment in such a way as to cause replication to be differential (Hull 1 988) interbreeding see reproductive isolation, population interference §6. 1 the process by which a foreign language lingueme is pro­ duced in a language by interlingual identification (Weinreich 1 968) interlingual identification §6 . l the mechanism by which linguemes from two different languages in a bilingual individual are identified as the same in some respect, allowing for interference ( Weinreich 1 968) internal change § 1 .2 an innovation that is produced by a mechanism other than interference by contact with another language (Hock 1 986) intraference §6. 2 the process by which a novel variant of a form with a new meaning is produced by intralingual identification. Basically equivalent to extension (Harris & Campbell 1 995) intralingual identification §6.2 the mechanism by which a form is associated with a closely related meaning in the conceptual space of an individual language introducer §7.4. 1 a speaker who introduces a variant of a lingueme from another community into the introducer's community. Same as innovator (Milroy & Milroy 1 985). introgression §7 .4. 1 , §8 . l the process by which a hybrid organism backcrosses with one of its parent species, thereby introducing genes from the other species into the gene pool of the first species (Grant 1 98 1 ) isolating mechanism §8 .2 a mechanism that brings about the reproductive isolation (or, in the case of speakers, communicative isolation) of a popula­ tion of organisms (speakers) (Grant 1 98 1 ) jargon §8 . 5 a simplified code of a language, used for communication in a restricted domain between two societies that do not have a language in common, that is not yet conventionalized (Miihlhiiusler 1 986) joint attention §4.2.3 the ability of two human beings to focus their shared attention on some entity (Clark 1 996)

Glossary of terms joint construal

§4 . 3 . 3

t he

239

joint un de rs t a nding by spe ak er and hearer of t he

meaning of the speaker's utterance (Clark

1 996)

§4 . 3 . l a nonconventional c oord in a t ion devi ce exp l o i t i n g the p e rce p t ua l or cognitive salience of s ome e n tity or concept that is sh a red by two or more individuals (Clark 1 996) koineization §7 .4.4 the process by which a group of speakers of differe n t dialects creates a new unified soc i e ty with a new lan g uage that is based on the former dialects (Trudgill 1 986) language §2.4. 1 a popu l at i o n of ut t e r ance s . The term lan gua ge sy stem is u sed where nece ssary to d i st ing ui s h the po p u l ation definition of a language from the view that a l an gu age is a sy stem of conventions. lexical diffusion §7.4. 3 propagation of a sound ch a n ge g r ad ua l ly through the lexicon, but invol vin g phonetically abrupt al tered repl icati o n of the lexical item with the new p ho n eti c v a l u e (M c M a ho n 1 994) lineage §2 3 the chain of replications of a gene (or l i ngu e me ) (Hull 1 988) lingua franca §8. S a language used for c o m m un i cat io n among societies that do not h ave a l an gua ge in common; specifically, t he language of one of the societies in a relatively un si m p l ified form, in con trast to a pidgin (Wa rd haug h

joint salience

'

'

.

1 992) §2 .4. l

unit of li n gu i s tic structure, as embodied in p a rt icul a r in re p lica tion; the re p licator in the basic l ing u ist i c selection process; the li n gu i st ic e qu ival e nt of a gene . The term l i n gueme will refer t o a l i n gu em e token; otherwise the te rm l i nguem e ty pe is used. lingueme pool §2.4. l the total number of linguemes (including a l l variants) found in a p opu l ati on of utterances; the li n gui s tic e quiv al e n t to a gene pool linguistic area §8. 3.2 a g e og rap hi cal are a o ccup i ed by s o cie t i es speakin g dif­ ferent l an g u ages, but p oss e ssing some de gr e e of multilingualism such that linguemes are diffused th rou gh the languages in the area (C h am bers & Trudgill 1 980) locus §2 .4. 1 a specific location on a chromosome (or in an u tt e ran ce ) whe re a single gene occurs. Only one of a set of gene alleles (lingueme variants) can occur at a single locus (Dawkins I 982b ) . A l i n g ui s t i c locus corresponds to a l i n gu i st ic v a ri ab l e i n s oc io l i n gu is t ic theory. meiosis § 1 . 1 the process of cell di vi s io n where a cell gives rise to two cells each with half as many chromosomes as the parent; essential to sexual repro­ duction (Hull 1 9 88) meme §2. l a unit of cultural inheritance, an a l ogo us to a gene {or lingueme) (Dawkins 1 976) metanalysis §5.4 a type of fo rm-fu n c t i o n reanalysis w her e t h e spe ake r simu ltan eo u s l y in terprets a semantic feature o f t h e context as an inherent feature of an element of the construction, and i n terpre t s a correl ated i nher­ ent semantic feature of the co nstruction as a feature of t he context . Same a s p r agm atic inference; the chief mechanism for se m a nti c change in grammaticalization .

lingueme

a

utterances, that can be inherited

'

'

'

'

240

Glossary of terms

§6. 3 . 2 a p r oce ss of semantic change that is semantically abrupt, applying a word from one semantic domain ( th e source domain) to another domain (the t a rge t domain) by virtue of certain a bst ract schematic similarit­ ies between the domains (Lakoff & J o h n s o n 1 980) mixed language § 8 . 4 . 3 a language in which there are major p o rt i on s of the linguemes in its lingueme pool from two ( or more) parent languages (Thomaso n & Kaufman 1 988) mixed marriage language §8 .4 . 3 a mixed langu a ge resulting from a high degree of mixed marriages between speakers of two languages; the two best -known cases ( Mednyj Aleut and Michif) involve marriages between exogenous fathers and indigenous mothers mixed lect §2 . 2 a language variety possessing lexical items with different variants of a phoneme arising through contact between two existing varieties (Chambers & Tru dgill 1 980) multiplexity §7 .2 see network natural phenomenon §3 . 3 . 2 a p henomenon that occ urs as a result of natural physical processes, without h uman inten tion involved (Keller 1 990/ 1 994) neogrammarian change §7 .4. 3 a sound change that is phonetically gr ad u a l but occurs sim u l taneously in all lexical items containing the phoneme (McMahon 1 994) network §2 . 2 , § 7 . 2 a group of speakers defined by their social lin ks wi th each other (Wardhaugh 1 992). Networks vary in density (how many indi­ viduals know each other) and multiplexity (in how many different domains the individuals know each other) . Individuals I-ave relatively strong or weak ties, defined in terms of density, multiplexity, and i nt i m acy of links with other individuals in the network . nonconventional coordination device §4. 3 . 1 a beh a vi or other than convention that is employed by individuals to solve a coordination problem (Clark 1 996) nonintentional mechanism §3.4 a mechanism for altered replication or selection of linguemes that is an unintended side-effect of intentionally conforming to convention (i.e. normal replication) norm see convention nor m a l rep li ca ti o n see repl ication perceptual basis §4 . 2 . 3 the experience shared by individuals by virtue of their directing joint attention to the experience. which is one of the sh a re d bases for personal common ground (Clark 1 996) periphrasis §6.3.2 the stage in the grammaticalization cycle at which a con­ struction i s used for a novel function that is 'gramm atical ' (that is, falls into the re gi on s of conceptual space that are typically expressed by grammatical m orph eme s in languages) personal common ground see common ground personal lexicon §4 . 2 . 3 the lexical items specific to a particul ar group or pair of ind ividuals (Clark 1 9 9 6) phenotype §2. 5 the manifested physical and behavioral features of an organ­ ism ( Dawkins 1 982b)

metaphor

·

Glossary of terms phylogeny

§2 .2

24 l

the historical patterns of descent for o rga n is ms (or languages) In linguistics, the term 'genetic ' rather than 'phylogenetic '

( D aw k i ns l 982b) .

is used. pi dgi n , pidginization

code of a lan guage, u se d for com ­ d omain between two societies that do not have a language in com m o n , that has become conventionalized. The p r oc e s s of dev el op i ng such a code i s pidgi nization ( Muh l hii.usler 1 986). polytypic species , la n g uage §2 . 2 a species (or language) whose organisms (dialects) are so d i fferent structurally that they would be divided into mul­ tiple species in a n essentialist v iew ( M ayr 1 9 8 2 ) population §2 . 2 a g r o up of organisms u nite d by t h e possib i l i t y of in terbreed­ ing and defined by t hei r reproductive isolation fr o m other organisms (M ayr 1 982). R e p r od u ctive isolation is relative for populations smaller than a spe­ cies . A gene pool i s a p opu l a t io n defined by the population of organisms that contains them . The population approach to d efi n in g species contrasts with the essentialist approach . I n la ngu age , a population is a gro u p of spea k er s united by the po ssi bi l ity of communicative interaction and defined by their communicative isolation from other speakers. Communicative isolation is relative for populations sm aller than a language . The utterances produced by speakers form a populat ion , as does the l i n gu em e pool contained in the population of utterances. postcreole continuum §3.2, §8 .4.2 a range of intermediate utterance forms that arise when a creole comes into contact wi th its primary parent l an gu a ge . and no focu se d i n termediate variety develops ( Wardh augh 1 992) power §7.4.2 an asymmetrical rel ationship between m e mbe rs of two differ­ ent communities such that one member (commu nity) i s dominant and the other subordi nate i n the re levant social context (Wardhaugh 1 99 2 ) . The dominant power group has greater (overt) prestige. pra gmati c inference §6 . 3 . 2 see metan a l ysis preced ent §4 . 3 . I a n onconventional coordination device by which speak ers solve a coordination problem using an a l ready established preceden t (Lewis 1 969) prestige s e e power pri mary l angu ag e birth §8.2 the evol ution of a new language by divergence of a dialect subpopulation from the parent l a n guage p o p u l a ti o n pri mary parent langua ge §8.4 . 3 the lan guage t h at contributes the majority of the basic vocabulary linguemes of a mixed language pr i ma ry speciation §8 . 1 the evolution of a new species by d i verge n ce of a subpopulation from the parent species pop u l a t i on (Grant 1 9 8 1 ) propagation § 1 .2 the increase i n freq uency of a l i ngueme in a language by selection push-chain §3.4. 1 a c h an ge which is explained as arising in order to av o i d two forms occupying the same position in a language system (McMahon munication in

a

a sim plified

§8 . 5

restricted

1 994) regrammaticalization

see

hypoanalysis

242

Glossary of terms

reinforcement §3.4. l

an in nov ati on by which a new v a ri a n t of a lin g uem e is concatenated w i t h t h e older variant repertoire §4.2 . 3 the set of linguistic codes that a speaker k n ows ( Wardhaugh 1 992) replacement §3.4. l an innovation by which a new variant of a l in gue me is substituted for the o l d er v a ria n t replication §§ 1 . 1 - 1 .2, §2. 3 t he process by which an enti ty (the re p lica tor ) produces a copy t h at possesses a ve r s i on of inherent structure of the ori gi n a l ent i ty (Hull 1 988; Dawkins 1 976). R e pl i ca t i on can be normal (i d entica l w ith t h e struc t u re o f th e pa re n t) or altered (not co m p l e t el y identical with the structure of the parent). D i fferen ti a l rep licat i o n is the replication of a replicator at an i n c re asin g (or d ec reas i ng) re l a t ive fre q uen cy com pared with other re p l i c a t ors . replicator §2 . 3 an en tit y th a t rep l ic ates its structure (Dawkins 1 982b, Hull 1 988) reproductive isolation §2.2 the (relative ) absence of interbreeding between members of distinct pop u la t i on s of organisms (H u ll 1 988) reticulation §8. l the me r gi n g of two in depe n d en t phylogenetic lines; com­ m on in plant species (Gran t 1 98 1 ) S-curve § 7 . 4.3 t h e typ i c al pattern by which a nove l variant is propagated t h r o u gh a po pul at ion of utterances (Kroch 1 989) Schelling game §4.2.4 a type of co o pe rative game in game theory in which two players a ttem pt to converge on the same solution to a coordination problem. In third-party Sc he ll in g games, the game is designed by a t h i rd person . In firs t- p arty Sc h e ll i ng games, including commu n ica tion , the game is d es i gned by one of the players, e . g . the speaker (Le wi s 1 969; Clark 1 996). sch�matic lingueme §8 . 3 . 1 a lin gueme that is only a schema or t emp l ate for a co n s t ruc t i o n, word order, word structure, phonotact ic structure, etc. A sc hem­ atic l i n gue me does not c o n ta i n any actual m orphe mes or phonemes. sel ection §2 . 3 the process by w hic h an interactor i n t e ract i n g with its e n vi ro n ­ ment cause s differential replication of its rep lic ators; also the evo l ut i on a ry model of processes that requires the gene r a t i on of variation by (altered) re plic at ion and the selection of variants (Hull 1 988) semi-shift § 8 .4 . 3 a process by which a group of speakers ad o pt s most. of the vocabulary of the la n gu age of an o t h er group of spe ak e r s, but retains the gramma t ic a l constructions and inflections or function words of its an ce st ral language shared expertise §4.2.3 the shared knowledge, skills a nd prac tice s in a domain of h um a n a c t ivi t y . Shared expertise forms t h e basis for defining a c om mu ni ty (Clark 1 996). shift §8 . 3 . l the process by which a so ciety gives up its a nce st ra l l angu age for the langu age of an o the r soci e ty (T h omas on & K a ufm an 1 988) sibling species , language §2. 2 two species (l angua ge s ) th a t are re prod uc tively (communicatively) isolated but are so similar s t r u c t urall y that they wo ul d be cons i dered a s i ng le species (language) in the essentialist app ro ach (Grant 1 98 1 )

Glossary of terms si gn al m e a n i n g

243

§4. 3 . 2 the conve n ti o na l meaning of a l i n g ui s t ic exp re s si on ( Cl a rk 1 996) signaling system §4.2.4 a system of conventions for evo k in g a meaning in the hearer's mind that is in some sense equivalent to the meaning in the speaker's mind (Lewis 1 969) social network see network society §4.2. 3, §7.2 a group of individuals that are (relatively) communic­ a tive ly isolated. A society includes all the communities which its members belong to. s olidarity §7 .4.2 a sym m etri c al re l at i o nship between two members of a com­ munity such that neither is in a power relation over the other , and the two are socially relatively intimate (Wardhaugh 1 992) speaker's meaning §4.3.2 the meaning conveyed by a linguistic expression on a particular occasion of use, i ncl uding all of its contextual richness (Cl ark 1 996) species §2.2 a reproductively isolated popu l ati on of organ ism s that inter­ breeds (Hull 1 988) speech comm unity see community spreading activation model §7.2 a model of the mental representation and th o ught where all k nowl ed ge is o rgan ized into a network of ne ur a l connec­ tions, and thought processes are rnod e led in terms of the activation of rel­ evant nodes. Activation spreads from one node to other linked nodes (Collins & Loftus 1 975). strong tie see network structural (definiti on of a language) see essentialism structural reanalysis §6 . 3 . 2 the assi gnme nt of a new syntactic structure to an existing construction (contrast with form-function reanalysis) substance Jinguem e §8. 3 . 1 a li n g u eme c on ta in i ng specific phonological sub­ stance, i.e. one or more specific phonemes, mo rph e me s or words substratum interference §8 . 3 . 1 the resul t of l a n gu a ge contact w hen a . society shifts to the language of another society; the linguemes co n trib uted by the ancestral lan gu a ge of the s hiftin g population to the acquired language (Thomason & Kaufman 1 988) sufficiency premise §4. 2.4 the assumption by interlocutors that the speaker will provide enough information in the coordination p ro blem posed by com­ munication to allow the h e a re r to solve it (Clark 1 996) symplesiomorphy §2.2 a shared trait be t w een two species or languages that is retained from the parent species (Lass 1 997 ) synapomorphy §2.2 a shared trait between two species or l an guages that is an i n nov at i o n of the two species (Lass 1 997) systemic functional e xpl anation §3.4 an e xp lan a ti o n of a lan guag e change in terms of the spe a ker in te n t i o nally m a k i n g the langu age sys tem simp l e r , more symmetrical, etc. teleological mec:h an ism §3.4. l a mechanism of altered rep l i catio n or selec­ t i on o f l ing u eme s in which changing (or preserving) the l an gu a ge system "in some way is the intentional go al of the speaker

244

Glossary of terms

tie strength see network upward specification §2. 4 . 3 the view that information combining features of two different linguistic levels in the hierarchy of inclusiveness of linguemes should be represented at the less inclusive level utterance §2.4. 1 an actually occurring piece of language, completely specified at all levels of structure, including its full contextual meaning on the particu­ lar occasion of use (i.e. speaker's meaning). The utterance is made up of linguemes, which are the replicators in utterance-based theories of language change. utterance-based theory §3. 1 a theory of language change that treats the lin­ guistic features (linguemes) of an utterance as the replicator; altered replica­ tion and selection both occur through production of utterances i n language use Utterance Selection, Theory of §2.4. l the theory that the utterance contains the paradigm replicators (the linguemes) in language change, altered replica­ tion of grammar largely occurs through form-function reanalysis, and selec­ tion occurs through social mechanisms variable §2.4. l , §3 . 3 . l a locus for a set of variants of a lingueme variant §2 .4. 1 one of a set of linguemes that can occur in one locus in a given utterance; equivalent to allele in biological evolution variety §2 .2 a population of organisms smaller than a species (Hull 1 988); also a neutral term to describe a language, dialect or code (Chambers & Trudgill 1 980) weak tie §7.4 . 1 see network

Refe rences

ADAMS, KAREN LEE. 1989. Systems of numeral classification in the Mon-Khmer, Nicobarese and Aslian subfamilies of A ustroasiatic. (Pacific Linguistics, Series B, No. 1 0 1 . ) Can­

ber ra : Australian National University. AITCHISON, JEAN. 1 989. The articulate mamm a l:

an introduction to psycholinguistics (3rd

edn) . London: Routledge.

AITCHISON, JEAN. I 99 1 . Language change: progress or decay? (2nd edn) . Cambridge: Cambridge University Press. AITCHISON, JEAN. 2000. Psycholinguistic perspectives on language chan ge .

Handbook of

historical linguistics, ed . Richard D . Janda & B ri a n D. Joseph. Oxford: Blackwell .

ALLEN, CYNTHIA L. 1995 . Case marking and reanalysis: grammatical relatiqns from

Old

to Early Modern English . Oxford : Clarendon Press.

ALLEN, CYNTHIA L. 1 99 7 . Investigating the origins of the ' group genitive' in English . Transactions of the Philological Society 95 . l I 1 -3 1 .

ALLEN, CYNTHI A L . 2000 . Obsolescence and sudden de ath in syntax : the decline of verb­ final order in early Middle English. Generative theory and corpus studies: a

dialogue

from 1 0/CEHL, ed . Ricardo BermUdez-Otero, David Deni son , Richard M . Hogg & C. B. McCully, 3-2 5 . Be rlin : Mouto n de Gruyter .

ANDERSEN, HENNING. 1 973 . Abductive and deductive change. Language 49.165-93 .

ANDERSl!N, Hl!NNING. I 98 8 . Center and periphery: adoption, diffusion, and spread.

Historical dialectology: regional and social, ed . Jacek Fisiak, 39-83. Berlin:

M outo n

de Gruyter.

ANTTILA, RAIMO. 1 989. Historical and comparative linguistics (2nd edn). Amsterdam: John Benjamins. ARIEL, MIRA . 1 990 . A ccessing

noun phrase

an tecedents. New York : Routled ge .

BAKKl!R, P!!T!!R & ROB ER T A. PA PEN . 1 997. M ichi f: a mixed language based on Cree and Fren ch . Contact languages: a wider perspective, ed . Sarah G. Thomason, 295-36 3 .

Amsterdam: John Benjamins. BARSALOU, LAWRENCE W. 1 987. The instability of graded structure: implications for the

nature of concepts . Concepts and conceptual developm en t : ecologica l and intellectual

factors in categorization, ed. Ulric Neisse r , 1 0 1 -40. Cambridge: Cambridge Univer­

sity Press. BARSALOU, LAWRENCE W. 1 99 3 . Flexibility, structure and linguistic vagary in co ncepts:

manifestations of a compositional system of perceptu al

symbols. Theories of memory,

ed. A lan F. Collins, Susan E. Ga thercole , Martin A. Conway & Peter E. Morris, 29 - 1 0 1 . Hillsdale, NJ: Lawrence Erlbaum Associa tes .

BAVIN, EDITH. 1 992. The acquisition of Warlpiri . Slobin 1 992, 309-7 1 .

BELL, R . T . 1 9 7 6 .

Sociolinguistics: goals, approaches and problems. London: Batsford.

BBLLWOOD, P!!T!R ! . 199 1 . The Austronesian d ispers al and the origin of lan guages. Sci-

entific American July I 991, 8 8-93 .

245

246

References

BENGTSON, JOHN T. & MERRITT RuHLEN . 1 994. Global etymologies. On the origin of

ed. Merritt Ruhlen, 277-336. Stanford: Stanford University Press . G. & M . G. B ARTOL I . 1 92 5 . Breviario de neolinguistica. M odena : Societa

language, BERTONI,

Tipografica M odenese.

BIBER, DOUGLAS . 1986.

Spoken and written textual dimensions in English.

62 . 3 84-4 1 4 . BOLINGER, DWIGHT . 1 97 5 . Aspects of language ( 2nd

Language

edn). New York: Harcourt Brace

J ovan o vi ch. BOLINGER, DWIGHT . 1 976. M ea n i n g

and memory .

Forum Lingusticum 1 . 1 - 1 4 .

London : Longmans. Evaluating competing linguistic models with language acq u i s iti o n data : implications of developmental errors with causative verbs. Quaderni

BOLINGER, DWIGHT. 1980. Language, the loaded weapon. BOWERMAN,

MELISSA.

1982.

di semantica 3 . 5-66.

The 'no negative evidence' problem: How do children avoid constructing an overly general grammar? Explaining language universals, ed. John A. Hawkins, 7 3 - 1 0 1 . Oxford: Blackwell. BRESNAN, JOAN & JoNNI KANERVA. 1989. Locative inversion in Chichewa: a case study of factorization in grammar. Linguistic Inquiry 20. 1 -50. BRIGHT, WILLIAM & A. J.