This book is a collection of papers that explores the ways in which bilingual children cope with two language systems. The chapters address issues in linguistics, psychology, and education, that bear on the abilities that bilingual children use to understand language, to perform highly specialized operations with language, and to function in school settings. The chapters provide detailed analyses of how specific problems are solved, how bilingualism influences those solutions, and how the social context affects the process. Finally, the implications of these findings for policy setting and the development of bilingual-education programs are explored.
Language processing in bilingual children
Language processing in bilingual children
Edited by
ELLEN BIALYSTOK Professor, Department of Psychology York University, Ontario
I CAMBRIDGE UNIVERSITY PRESS
PUBLISHED BY THE PRESS SYNDICATE OF THE UNIVERSITY OF CAMBRIDGE
The Pitt Building, Trumpington Street, Cambridge, United Kingdom CAMBRIDGE UNIVERSITY PRESS
The Edinburgh Building, Cambridge CB2 2RU, UK http://www.cup.cam.ac.uk 40 West 20th Street, New York, NY 10011-4211, USA http://www.cup.org 10 Stamford Road, Oakleigh, Melbourne 3166, Australia Ruiz de Alarcon 13, 28014 Madrid, Spain © Cambridge University Press 1991 This book is in copyright. Subject to statutory exception and to the provisions of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published 1991 Reprinted 1992, 1994, 1996, 1998, 2000 British Library Cataloguing in Publication data Language processing in bilingual children. 1. Bilingual children. Language skills. Acquisition I. Bialystok, Ellen 404.2019 Library of Congress Cataloguing in Publication data Language processing in bilingual children / edited by Ellen Bialystok. p. cm. Includes index. ISBN 0 521 37021 3 (hardback) - ISBN 0 521 37918 0 (paperback) 1. Bilingualism in children. 2. Psycholinguistics. 3. Education. Bilingual. I. Bialystock, Ellen. P115.2.L36 1991 404/.2-dc20 90-37707 CIP ISBN 0 521 37918 0 paperback Transferred to digital printing 2003
BS
Contents
List of List of tables List of contributors Preface
figures
page ix x xi xiii
Introduction ELLEN BIALYSTOK
1
Language modules and bilingual processing MICHAEL SHARWOOD SMITH
2
90
Metalinguistic dimensions of bilingual language proficiency ELLEN BIALYSTOK
7
70
Giving formal definitions: a linguistic or metalinguistic skill? CATHERINE E. SNOW, HERLINDA CANCINO, JEANNE DE TEMPLE, a n d SARA SCHLEY
6
49
Interdependence of first- and second-language proficiency in bilingual children JIM CUMMINS
5
25
Second-language learning in children: a model of language learning in social context LILY WONG FILLMORE
4
10
Phonological processing in two languages IAN WATSON
3
1
113
Translation skill and metalinguistic awareness in bilinguals MARGUERITE MALAKOFF a n d KENJI HAKUTA
141
Contents 8
Towards an explanatory model of the interaction between bilingualism and cognitive development RAFAEL M. DIAZ a n d CYNTHIA KLINGLER
9
167
Constructive processes in bilingualism and their cognitive growth effects JANICE JOHNSON
193
10 Language, cognition, and education of bilingual children ELLEN BIALYSTOK a n d JIM CUMMINS
222
Index
233
vni
Figures
1.1 2.1 2.2 6.1 6.2 6.3 6.4
Forster's models English six-year-old percentage Igl responses Classic discrimination/identification function Domains of language use Oral uses of language Literate uses of language Metalinguistic uses of language
IX
page 15 42 43 122 125 128 131
Tables
2.1 5.1 5.2 5.3 5.4 5.5 7.1
7.2
7.3 7.4
Averaged V.O.T. production values page 40 Means on English and French communicative adequacy (C.A.) and conversational features (C.F.) scores 97 Scores in English and French for children with home vs. school exposure to French 99 Effects of status as a bilingual on scores in English and French 100 Within-language correlations for English and French definitions scores 101 Between-language correlations for English and French definitions scores 102 Percentage of responses that were non-literal or literal translations in different translation directions and conditions 156 R2 values obtained through regression predicting translation times for words and sentences on English and Spanish proficiency, and the word-identification (W.I.) task 157 Examples of errors coded in translation tasks 158 Comparison of frequencies of error types in the story162 translation task across subjects from study 1 and study 2
Contributors
Ellen Bialystok Department of Psychology York University, Ontario Herlinda Cancino Graduate School of Education Stanford University Jim Cummins Modern Language Centre Ontario Institute for Studies in Education
Cynthia Klingler Universidad de las Americas Centro de Exposiciones y Convenciones, Mexico Marguerite E. Malakoff Department of Psychology Yale University Sara Schley Graduate School of Education Harvard University
Jeanne De Temple Graduate School of Education Harvard University
Michael Sharwood Smith Instituut voor engelse University of Utrecht
Raphael M. Diaz School of Education Stanford University
Catherine E. Snow Graduate School of Education Harvard University
Kenji Hakuta School of Education Stanford University
Ian Watson Phonetics Laboratory University of Oxford
Janice Johnson Department of Psychology York University, Ontario
Lily Wong Fillmore Department of Education University of California, Berkeley
XI
Preface
I have always felt somewhat schizophrenic about two of my academic interests: second-language acquisition and bilingualism on the one hand, and psycholinguistic descriptions of first-language proficiency on the other. It is therefore particularly satisfying to be able to unite them in the production of this volume. The tension between these two interests has not only reflected an inner personal conflict between two of my own areas of research but has signified as well an external conflict between two academic subdisciplines. Editing this volume has resolved much of my personal tension. My hope is that the volume contributes as well to the goal of integrating theory and research across the subdisciplines. Most of the chapters in this volume were originally presented in the invited symposium "Language acquisition and implications for processing in bilingual children" at the meeting of the Society for Research in Child Development in 1987. The symposium was co-chaired by Catherine Snow and myself. I am grateful to Catherine for her easy style of collaboration in that symposium and for her support and assistance in my efforts to publish the proceedings. All of the papers that were presented in that symposium have been completely revised for this volume and several new chapters have been added. I would like to express my sincere thanks to all of the contributors. In my view, these authors are among the finest scholars working in the field. I am also indebted to the Natural Sciences and Engineering Council, Canada (N.S.E.R.C.) for providing me with a continuing grant to support my research and writing. Finally, I offer my sincere gratitude and appreciation to my friends, colleagues, students, and family, who supported this project and provided an environment for me in which it was possible to carry it out.
xin
Introduction ELLEN
BIALYSTOK
The study of the cognitive and linguistic achievements of bilingual children has recently become a respectable, even popular, area of inquiry in psychology and education. The issue is not new, but its history as a serious topic of research is somewhat speckled. Researchers have indulged in speculation for some time about the consequences of bilingualism for children's cognitive development, school achievement, linguistic processing, and metalinguistic abilities, but clear theoretical connections based on sound empirical data have been rare. One factor that has made the problem so difficult to study is the enormous diversity that accompanies children's bilingualism. Consider, for example, some of the conditions under which children can become bilingual. Children can learn both languages simultaneously in the home; the second language can be learned through submersion in a foreign culture (and here the relative status of the first and second language becomes critical in determining outcomes); or the second language can be learned through immersion or foreign-language classrooms with the majority-language environment. These differences undermine most attempts to identify precise conditions for second-language acquisition, the psychological factors that accompany bilingualism, and the implication of bilingualism for academic and other achievements. Where research has been attempted, the enterprise has often been rendered uninterpretable by the failure to account for, and sometimes even to acknowledge, the critical differences among these situations. Much of the early research in bilingualism was motivated by educational needs and policy but biased by particular prejudices against bilingualism (and possibly against immigrants in general). About thirty years ago, for example, the general wisdom held that bilingualism was a disorder that could be corrected through ruthless instruction in a standard majority language, pushing out of the inflicted child all traces of the invading language. This remedy was imposed despite the fact that the unwanted language was often the language of the child's home, heritage, and tradition. But the evidence was clear: immigrant children in North American schools were less successful academically than were their monolingual peers, and 1
Ellen Bialystok the only explanation that entered the imaginations of those researchers was that bilingualism was to blame. It came as a surprise, therefore, when anglophone Canadian children in the early 1970s who had been sentenced to French immersion programs and educated entirely or partially through the medium of French did not display this bilingual pathology. Clearly, bilingualism,perse, was not the culprit. There were, of course, some notable exceptions to this early style, such as Leopold's (1939-1949) seminal study of his daughter learning two languages, but these were rare. Most of the details we have about secondlanguage acquisition and the use of two languages in specific contexts, especially educational contexts, come from research paradigms developed in the 1970s. This research marked a departure from the earlier psychometric studies in that it took bilingualism as a phenomenon worthy of scientific interest in its own right and not simply as an educational complication. But like the attitudes that shaped educational policy towards bilingualism in the earlier studies, the research has been characterized by great and sometimes sudden shifts of attention between models that are in some cases antithetical to each other. Three main research approaches were initiated during the 1970s that were responsible for a proliferation of studies that had consequences for understanding language processing by bilingual children. These correspond roughly to linguistic, sociolinguistic, and psycholinguistic perspectives on language. Educational concerns remained relevant, but educational implications were more likely inferred from research undertaken in these other traditions. Although there was considerable work in evaluation studies, such as Canadian French Immersion and American Bilingual Education of various types, for example, these studies were in some ways parasitic on the more basic research that was setting norms of achievement for children learning a second language against which the results of those programs could be compared. Another important research orientation undertaken during that time was the neurolinguistic perspective, but that approach will not be dealt with in the present volume. It is interesting to note, nonetheless, that it was on the basis of a (possibly false) neurolinguistic argument concerning a "critical period" for second-language acquisition that the Canadian French Immersion programs found their strongest validation and subsequently rallied their greatest support (but see Johnson & Newport, 1989, for a revitalization of the critical-period view). The majority of the research relevant to investigations of bilingualism is that focused on the linguistic, or more precisely, applied linguistic perspective. The principal issue is the linguistic documentation of the course of second-language acquisition. Each linguistic description, however, entails different psycholinguistic processes to support that model. The past two decades have witnessed striking changes in the linguistic interpretation
Introduction of second-language acquisition, and correspondingly, in the psychological and educational implications that accompany each interpretation. The prevailing view at the beginning of the 1970s was based on the linguistic approach of "contrastive analysis," developed from Lado's (1957) early work and reported in Linguistics across cultures. On that view, a linguistic comparison of two languages reveals the sources of potential difficulty for the learner by identifying the aspects of the two languages that differ from each other. The psycholinguistic basis of this interpretation was the claim that language learners learned a second language by substituting targetlanguage forms and structures into what they already knew about their first language. Hence, transfer was a complete explanation for secondlanguage learning. This view was abruptly replaced in the 1970s by a claim called "creative construction," a term coined by Dulay and Burt (1975). On their view, there was no place for transfer. One learned a second language by starting all over again using the same processes that had guided first-language acquisition. It is only recently that a more balanced view has emerged, in which the obvious role of transfer is acknowledged but not offered as a complete explanation of second-language learning. A linguistic explanation of second-language learning which has achieved currency recently is based on the Chomskian view of government and binding (Chomsky, 1982). On this view, languages are learned by setting parameters in the "Language Acquisition Device," and second-language learning involves the resetting of some parameters. Transfer is explained as the influence of the parameter settings established for the first language on the values necessary for the second. The bulk of the empirical work undertaken in this linguistic tradition documented the acquisition of specific linguistic features. The fourteen English morphemes made famous by Roger Brown (1973) in his seminal longitudinal study of three children in the early stages of learning English (as a first language) achieved celebrated status. In addition, the journals were full of detailed descriptions of how learners of a particular language X mastered the Y (negation, question formation, pronominalization, past tense, etc.) of language Z. Although some of this work continues to the present, the legacy of the research of the 1970s is a meticulous body of data documenting the course of second-language learning. A second perspective for research in the 1970s was based on sociolinguistic analyses of language learning and use. An important contribution of this approach is that it documented and analyzed the critical contextual facts that distinguish one bilingual situation from another. Studies examining the social context of bilingualism led to such crucial distinctions as high- versus low-status languages, additive versus subtractive bilingualism, full versus partial control of language, and frequent versus infrequent use of the language. All of these factors were shown to be instrumental in
Ellen Bialystok the second-language acquisition process. A strong version of the sociolinguistic perspective led some to attribute the bulk of the explanation of second-language learning to social factors. John Schumann (1978) adopted this view and developed the acculturation model for secondlanguage acquisition in which he argued that social factors such as integration into the community and personal feelings of assimilation provide an explanation of second-language learning. A third type of research on bilingualism established in the 1970s was addressed to the perennial problem of why some people learned a second language more easily and more thoroughly than others. To this end, many studies of the characteristics of good and poor language learners were undertaken, assessing the role of such variables as aptitude, motivation, personality style, form of instruction, and the like on the ultimate level of proficiency achieved by the language learner. One particularly thorough example of a study of this type was conducted by Naiman et al. (1978). While no single profile of the "good language learner" ever emerged, one could no longer ignore such individual differences in making general statements about second-language learning. If nothing else, second-language learning was complex in ways that seemed irrelevant to first-language acquisition, and the contribution of the research on that problem was at least to identify some of the individual difference factors that conspired to produce that complexity. These three perspectives constitute the research heritage for the study of bilingual children. There are, however, two obvious incongruities between the problems to which this research was addressed and the problems presented by the study of bilingual children. First is the problem of age. Although some of the studies carried out since the 1970s which contributed to the existing data base examined the second-language acquisition of children, the vast majority were studies of adult second-language learners. It is still an unsolved empirical question whether or not the course of language acquisition is the same for conceptually immature children as it is for adults. The most widespread commonsense belief about secondlanguage learning, namely that children are more successful than adults, has been severely challenged with both theoretical and empirical arguments (e.g. Snow & Hoefnagel-Hohle, 1978) although new evidence has begun to emerge again in support of the critical-period view (Johnson & Newport, 1989). A correct description of the similarities and differences between child and adult second-language acquisition is still pending. Second is the problem of effect. What is the consequence of having learned another language? What are the implications for the child's cognitive, linguistic, social, and educational experiences? It is primarily to this second issue that the present volume is addressed. Two recent changes in the field of language research have made the direct and explicit study of the linguistic and cognitive development of
Introduction bilingual children a timely issue. First is the "legitimization" of secondlanguage studies into mainstream linguistic and psycholinguistic theorizing. Second is the shift in emphasis in psycholinguistic research from product descriptions to process models. For a long time, there was a pervasive attitude among researchers studying language acquisition that the main problem to be solved, the dominant issue, was first-language acquisition. Second-language acquisition was perhaps a variation on a theme, but of far less interest and relevance than the child's acquisition of a first language. That view, in the context of the times, may have been correct. It may be that no serious attention could be paid to something that was considered as a twist on a theme when the theme itself remained such a mystery. But we now know a great deal about first-language acquisition, and the insight that is occurring to some researchers is that we could know a great deal more if we included information about second-language learning into our theorizing. An interesting development in this regard is the completion model of language processing developed by Elizabeth Bates and Brian MacWhinney (e.g. MacWhinney, 1987). By taking seriously data about the way in which adults process a second language, they have developed a model which they offer as a valid description of language processing in general. In this way, research with second-language learners has contributed to an advance in mainstream psycholinguistic theorizing. Such developments enhance the status of studies of bilingual speakers. The second change is that research in children's cognitive and linguistic development has shifted from product descriptions of their accomplishments to process analyses of the causes of development. It is only about twenty-five years since psychology shed the shackles of behaviorism and allowed researchers to indulge in speculations and theories that included constructs such as "mind." In some current models, for example those of Robbie Case (1985), Robert Sternberg (1984), and Robert Siegler (1984), processing is attributed to hierarchically related "schemes" that the child develops as a function of specific experiences. Throughout development, the schemes continually become elaborated and restructured. These schemes, then, have the responsibility for guiding and determining the child's performance on specific tasks. Some of these models attribute excessive autonomy to such schemes (Sternberg may be a case in point) and leaves much of the important work in the hands (or mind) of some hypothetical homunculus. The important point, however, is that children's experiences determine critical aspects of their cognitive organization, and that organization is instrumental in influencing the kinds of intellectual achievements that children can attain. In these terms, it is not implausible that bilingualism is an experience that has major consequences for children's intellectual development.
Ellen Bialystok In recent years, a growing body of literature has emerged in which process-oriented analyses of these types are applied to specific problems in children's linguistic development. An important part of this interest are the studies of children as they develop the ability to deal with complex uses of language, in particular the literacy skills of reading and writing. These skills have been shown to depend on specific types of language processing and the availability of a set of metalinguistic insights. Skilled readers, for example, are different from less skilled readers because they have more efficient executive schemes for processing language and they have a more elaborated understanding of language structure. The origins of early reading have been traced to types of language processing in which children can pay attention to such features of language as rhyme, alliteration, and paraphrase. Writing, too, is a complex process involving attention to specific linguistic features and integration of those features into a higher-order plan or scheme. More recently, considerable attention has been devoted to uncovering the metalinguistic skills developed by children in the early years, and the ways in which those skills interact with the child's progress in learning to read and write. In all these models, the accomplishments of reading and writing that had formerly been investigated as product descriptions (e.g. what the stages are in learning to read, what features of text make learning to read easier, what errors children make in early reading and writing) are recast as issues requiring explanation. These process models are invariably integrative, incorporating many aspects of children's competence. The contribution that the child's bilingualism might make to such linguistic processes has barely been investigated. Some research explicitly addressed to the influence that bilingualism might have on cognitive and linguistic development has in fact been undertaken throughout the decades. The early studies in the 1950s and 1960s which argued for cognitive deficit to be the consequence of bilingualism can be largely dismissed on methodological grounds. (An excellent review of these studies is presented by Kenji Hakuta (1986) in Mirror of language.) In contrast, a number of studies in the 1970s linked certain cognitive and linguistic advantages to bilingualism. Many of these studies, however, seemed to be overenthusiastic in their effort to identify positive effects for bilingualism, in spite of the fact that results were sometimes equivocal and sometimes even negative. But the optimism was a necessary political tool: only forceful argument and convincing data could dethrone the predominant view of bilingualism as a liability. The current theoretical and political ethos permits a sober and balanced examination of the influence of bilingualism on children's development. Theoretically, the models and research paradigms from which we study children's development are now sufficiently sophisticated that the complex and elusive issues involved in children's bilingualism may be addressed.
Introduction Politically, it seems less necessary to "prove" the acceptability of bilingualism for children. We are free, that is, to discover that being bilingual may in fact bring no special cognitive or linguistic benefit to children, and that finding will not threaten the existence of children in our educational system who happen to be bilingual. The chapters in this volume report research in this new tradition which examines basic processes in language acquisition and use for children who are bilingual. The ultimate aim is to develop a cognitive description for the acquisition and mastery of a second language, and to determine implications of becoming bilingual on children's language processing and language awareness. The relevant questions are whether or not bilingualism affects the way in which children process language and the insights which they subsequently derive about the structure of language. Basic questions such as the relation between first- and second-language acquisition have long been entertained, but only within these recent process models can they be properly addressed. How is the second language acquired? What role is given to the child's first language during that process? And how are the child's cognitive and linguistic resources modified (or not) as a result of having mastered two linguistic systems? The focus throughout is on the child's cognitive resources and their role in the child's development of specific types of intentional language-processing. In all cases, too, the contextual factors that define the bilingual situation are addressed. It is our belief that intensive study of the cognitive and linguistic development of bilingual children is essential, not only because of our obligation to understand the language development of those children who are bilingual, but also because of the incredibly rich source of knowledge gained from this understanding that we can apply to our study of first-language acquisition and cognitive development. The volume begins with a linguistic description of language processing that can accommodate processing in two languages. This chapter, by Sharwood Smith, sets out a clear statement of the linguistic resources recruited in ordinary language use (such as conversation) and points out how those resources might be organized for bilingual speakers (since his focus is not exclusively on children). The second chapter by Watson examines one aspect of linguistic competence, phonological ability, and reports empirical evidence for the course of the phonological development of bilingual children. These data are used to evaluate alternative models of phonological processing. The third chapter addresses acquisition. Wong Fillmore bases her discussion of the acquisition of a second language by young children on the incredibly rich longitudinal data base she and her colleagues have collected studying the language acquisition of immigrant children. She presents a model for acquisition that has three components - social, linguistic, and cognitive - and demonstrates the essential role each plays
Ellen Bialystok in a description of language acquisition. In the fourth chapter, Cummins applies one of the distinctions made by Wong Fillmore concerning the difference between individual and environmental sources of influence in acquiring language proficiency to the problem of how the first language influences the development of proficiency in the second language. He adds a further distinction between contextualized and decontextualized uses of language and demonstrates that the greatest influence is for decontextualized school-related uses of language. Snow, Cancino, De Temple, and Schley pursue this distinction in their investigation of the ability of bilingual and monolingual children to provide definitions, an important linguistic task that is quintessential^ a "school problem." Substantiating Cummins' claims, children perform comparably on this task in both their first and second languages. Bilingualism seems not to influence performance. Other kinds of language tasks do, however, show variation in performance as a function of bilingualism. These are reviewed in the sixth chapter by Bialystok. Malakoff and Hakuta are interested in the problem of translation, clearly a linguistic task available only to children who are bilingual. They ask how children do translations and what that ability reveals about their linguistic organization and processing. Diaz and Klingler, in the eighth chapter, readdress the old problem of the effect of bilingualism on intelligence. They demonstrate that more precision in the description of the child's bilingualism is necessary to make much progress with this question. They also bring a new perspective to the issue by interpreting the question from a Vygotskian viewpoint. An attempt to provide some explanatory constructs for these effects of bilingualism on cognitive growth is proposed by Johnson in the ninth chapter. A final chapter by Bialystok and Cummins summarizes the main findings concerning the language processing and language awareness of bilingual children and assesses the contribution of the last decade of research. Three themes run through these papers. First, most of the papers include reports of original research, but the emphasis in each case is on the larger issues of language processing that are relevant for bilingual children. Thus the data are used to support the more general enterprise of theory building. Second, the linguistic process or accomplishment examined in each case is in the domain of what one might call "intentional language use." These are the aspects of language that impact most directly on cognition - using language in the service of specific goals, as in reading or understanding metaphors, reflecting on language as a structured body of knowledge, as in developing linguistic awareness, and manipulating language to achieve specific ends, as in writing and translating. Third, the approach in all cases is on processing. The tacit assumption is that we are beyond the stage of inquiry in the field in which product descriptions will advance our knowledge. Further progress depends on understanding how the system works, 8
Introduction how it changes, and how it adapts. These questions can only be answered by examining mental processes.
References Brown, R. 1973. A first language: the early stages. Cambridge, MA: Harvard University Press. Case, R. 1985. Intellectual development: from birth to adulthood. New York: Academic Press. Chomsky, N. 1982. Some concepts and consequences of the theory of government and binding. Cambridge, MA: M.I.T. Press. Dulay, H. & M. Burt 1975. Creative construction in second language learning. In M. Burt & H. Dulay (eds.), New directions in second language learning, teaching and bilingual education. Washington, DC: TESOL. Hakuta, K. 1986. Mirror of language: the debate on bilingualism. New York: Basic Books. Johnson J. S. & E. L. Newport 1989. Critical period effects in second language learning: the influence of maturational state on the acquisition of English as a second language. Cognitive Psychology, 21, 60-99. Lado, R. 1957. Linguistics across cultures: applied linguistics for language teachers. Ann Arbor, MI: University of Michigan Press. Leopold, W. F. 1939-1949. Speech development of a bilingual child: a linguist's record. Evanston, IL: Northwestern University Press. MacWhinney, B. 1987. Applying the competition model to bilingualism. Applied Psycholinguistics, 8, 315-327. Naiman, N., M. Frohlich, H. H. Stern & A. Todesco. 1978. The good language learner. Toronto: Ontario Institute for Studies in Education. Schumann, J. 1978. The acculturation model for second language acquisition. In R. Gingras (ed.), Second language acquisition and foreign language teaching. Arlington, VA: Center for Applied Linguistics. Siegler, R. S. 1984. Mechanisms of cognitive growth: variation and selection. In R. J. Sternberg (ed.), Mechanisms of cognitive development. New York: Freeman. Snow, C. E. & M. Hoefnagel-Hohle 1978. The critical period for language acquisition: evidence from second language learning. Child Development, 49, 1114-1128. Sternberg, R. J. 1984. Mechanisms of cognitive development. A componential approach. In R. J. Sternberg (ed.). Mechanisms of cognitive development. New York: Freeman.
1
Language modules and bilingual processing MICHAEL SHARWOOD
SMITH
This chapter deals with the potential contribution of recent linguistic and psycholinguistic research to our understanding of the ways in which bilinguals (and multilinguals) process language data. The term "processing" takes on two meanings here, a developmental meaning and an on-line use meaning. Firstly, in the building up of knowledge representations (development), learners' acquisitional mechanisms need to process relevant linguistic information coming in from the environment. This is developmental processing. However, once knowledge representations become available to learners as users of their knowledge then those knowledge representations in turn need to be processed for the various uses to which learner/users want to put their language system at given moments in time. This on-line control may be called knowledge processing. In considering various aspects of developmental processing and knowledge processing in bilinguals, extrapolations will be made from some current views on how language is represented in the mind of the learner/user and reference will be made to theoretical notions from sister disciplines which have not, at least not until very recently, been developed and demonstrated in a second language context. The theoretical status of various different kinds of linguistic knowledge will be considered, ranging from the more implicit knowledge inherent in the Chomskyan notion of linguistic competence through pragmatic knowledge, to characteristically explicit knowledge about language that is freely accessible to introspection, namely metaknowledge. At each stage, the focus will be on the implications for research into bi lingual ism. The underlying theme in discussion will be the application of a "modular" approach to bilingual processing. If knowledge in general, as Fodor (1983) has argued, and even language knowledge itself, is compartmentalized into modules containing fundamentally different types of mental representation, how might this affect our understanding of how the bilingual mind stores and manipulates information relating to the different language systems available to it? Both sequential and simultaneous bilingualism will be discussed. For the former, the second language (L2) is acquired after the 10
Language modules and bilingual processing first (LI); for the latter, the child grows up acquiring two or more languages at the same time.
Modularity in the representation of knowledge The perspective to be outlined crucially involves the theoretical notion of modularity (basically following Fodor, 1983; see also Jackendoff, 1987), whereby an individual's knowledge is viewed not as a single entity but as one which involves a number of separate compartments or modules. Some of these modules are highly independent systems, the prime example being the mental grammar which lies at the heart of human linguistic ability. The precise nature of modular structure in the mind is a matter of theoretical debate. However, the advantages of viewing the mind as a modular entity are becoming clearer as the debate ensues and inevitably touches upon the interpretation of bilingual behavior. Grammatical competence, in Fodor (1983), is taken to be different in kind from more generalized "encyclopedic knowledge" such as our knowledge of history and physics, as well as our idiosyncratic knowledge of events, people and places. This important qualitative difference has to do with the fact that the grammatical processing mechanisms (which is what Fodor is principally interested in here) handle the processing of relevant sensory information independently, that is, without any help from the other knowledge systems. These grammatical mechanisms must then format that information for subsequent integration into a more global act of language comprehension or production. Moreover this processing is "rapid and mandatory," that is, it is essentially a reflex response to the input. The standard example of this sort of processing is vision, where people cannot decide to not see what is in front of them unless they actually block their field of vision by closing their eyes or turning away. In the same way, one cannot be exposed to comprehensible speech without automatically comprehending it. As has been discussed at length by Chomsky, with whom Fodor's ideas are very broadly compatible, the actual development of mental grammars, is also independent of general cognitive growth. These grammars develop by means of acquisitional mechanisms that are to a great degree constrained by principles that are biologically endowed and not available via experience. In this way, the child, in interacting with primary language data (the input) solves "Plato's problem" and comes to know much more about the grammatical structure of the target language than it has evidence for (Chomsky, 1980). This process is not at all akin to the processes involved in developing knowledge about, say, the history of the planet, or how the local library works, where development, unlike the acquisition of grammatical structure 11
Michael Sharwood Smith of a given language, can be explained by general inductive principles working on the much richer relevant evidence that is typically available via experience. This crucially involves "negative evidence," that is being continually informed about what is not possible or not permitted in the system being learned. By way of contrast, language acquisition, at least firstlanguage acquisition and quite possibly, despite appearances, a fair amount of second-language acquisition (see White, 1987), has to proceed, where the grammar (basically syntax and phonology) is concerned, with little or no negative evidence to speak of (see Lasnik, 1982, for qualifications of this position). Children do not go to school to learn to speak their mother tongue; second-language learners (who have previously become native speakers of a first language) sometimes do but the instruction they get can be fragmentary, not structured according to known principles of language acquisition and as a result, possibly very misleading, and devoid of constant and consistent individualized corrective feedback. Fodor (1983) discusses the use of grammatical knowledge at a given moment in time. Grammatical performance involves a system like the visual system which is specifically geared to processing particular kinds of information, that is grammatical information and none other than that, coming in from the environment. This system processes input automatically without deliberate activation: if someone addresses us, we cannot choose not to interpret the utterance except, as mentioned above, by blocking off the sound wave. The speech signal, once perceived, triggers an automatic response. We begin interpreting it in the same way as we automatically interpret familiar visual patterns that we perceive by means of our eyes. And, in order to speak ourselves, we normally have to simply think about what we have to say and not about the fine phonological and grammatical detail of the shapes our utterances will have to take. In this way, grammatical utterances (give or take a few slips of the tongue) are formed automatically without any conscious manipulation on our part. In fact we have no choice since we have access to only a few of our utterance-formation mechanisms. Encyclopedic knowledge is a type of "knowing about" knowledge which is more amenable to deliberate conscious control by the individual. A native speaker knows his or her language but without knowing much about it. One can deliberately select an area of encyclopedic knowledge, and reflect upon it, analyze it, test its validity, revise it and consciously learn more. In Fodor's terms, whose conception of modularity is actually much less fine-grained than others who have discussed this issue subsequently (see, in particular, Jackendoff, 1987), such knowledge falls more directly within the domain of the central processor. If encyclopedic knowledge itself can be seen as compartmentalized in some way, the thrust of the argument is that the basic building blocks are the same: the knowledge is developed 12
Language modules and bilingual processing with the same learning mechanisms and it is used in the same way. Linguistic knowledge itself, in the sense used above (as subconscious and relatively inaccessible to introspection), can also be seen as internally modular, although Fodor with his focus on knowledge processing does not discuss this. Certain aspects of that knowledge may also fall into this more general category, for example the choice of words and style to suit situations and particular speaker intentions in the production of utterances as also a whole range of morphological phenomena specific to any particular language. This implies that some parts of morphological, lexical, and pragmatic knowledge may not be part of the grammar module per se, although they are certainly part of what most of us mean by "language" (see Foster, 1985). If some aspects of linguistic knowledge share the same type of mental representation with other more general kinds of knowledge, this also means that they can be acquired and used following more general cognitive principles involving, presumably, inductive learning and some conscious manipulation during actual performance. This, again, would appear to be particularly true of metalinguistic activities. Our reflections on the formal properties of language (grammar or other aspects of a language) are most vividly represented by the case of a linguist trying to work out the rules and principles behind a particular natural language. Typically, the better we get at inductive learning and conscious manipulation of ideas, the better able we are to develop and use linguistic knowledge that falls under this category. By the same token, our ability to develop other kinds of linguistic knowledge will not reflect the same improvements. What we were good at when acquiring our first language as young, cognitively immature learners, we will be perhaps just as good at later, but certainly no better by virtue of being better at reasoning coherently. Research will no doubt cast much more light on this issue by demonstrating the clear influence of general learning principles in specific areas of the language, especially the beneficial effects of corrective feedback. It already seems intuitively reasonable to suppose that corrective feedback will be readily available with regard to the idiosyncratic use of the lexicon, where the learner refers to objects or concepts with the "wrong" words (as in "that's a pretty cat" while pointing at an ugly dog) or uses the referentially appropriate words but violates the pragmatic norms of the community (as in "Hi there, Your Highnesses" or "Salutations, guys"). To sum up, it seems that the learning conditions for different aspects of language acquisition are radically different in kind. The fact that the immature and uninstructed child does so well suggests that the learning mechanisms are specialized not only in contrast to non-linguistic learning mechanisms but also internally. In some areas, notably syntax, corrective feedback is not required and may well not be processed even when it is 13
Michael Sharwood Smith available. In other linguistic areas, it may be processed and lead to a revision of knowledge. In other words, modular treatment is also called for when developing a theory to account for language behavior itself. Modularity in on-line processing. If processing in the actual development of linguistic knowledge over time appears to require modular explanations, it is not clear that this should also be true for processing in the on-line language use sense of the word. Our hypotheses about the knowledge modules may or may not be of relevance with respect to the processing "machinery" available to access this knowledge and manipulate it actually using language on a given occasion. This aspect of language ability will be called the processing-control dimension ("control" for short): how knowledge is controlled by the performance mechanisms (see Bialystok & Sharwood Smith, 1985, for instance, for an application of this idea to second-language acquisition theory). In principle, the on-line control of qualitatively different kinds of knowledge could be non-modular, that is there could be uniform access and manipulation across different knowledge modules. Modularity in this sense would be a matter of knowledge representation, not knowledge use. And if processing control is modular, then the modules do not have, again in principle, to be the same as the knowledge modules. Recent thinking on language processing has, in fact, employed the notion of modularity. Frazier (1988) in particular argues that the evidence to date strongly suggests that language processing should be seen in terms of autonomous modules associated with but not necessarily identical with the modules defined in current generative theory, that is, modules within the grammar. This means that the speech signal is analyzed by separate autonomous modules that cannot bring information to bear from other areas to disambiguate the message on-line. In other words, the evidence suggests that until all the processing modules have done their work on the input, there is a certain level of "garden-pathing" leading to alternative-candidate interpretations being available at given points in the process. To this extent, Frazier's approach follows the seminal paper on processing by Forster (1979) in which the speech signal is analyzed in a manner as represented in figure 1.1. A crucial feature of Forster's model are the three linearly ordered processing stages represented vertically in figure 1.1, that is the phonological, syntactic, and message processors respectively. Each processor operates independently without reference to information from outside except the input from the preceding stage. Where there is ambiguity, then the particular processor currently operating must assign alternative-candidate analyses. These ambiguities may be resolved at a later stage. For example, the syntactic processor can reject, say, one of two phonologically acceptable 14
Language modules and bilingual processing message processor
general problem solver
world knowledge
i[
syntactic processor
lexicon
j
phonological processor
Figure 1.1 Forster's models
candidates (Do you like to roam/*Do you like to Rome). The message processor can reject anomalous alternatives passed on from the syntactic processor and real world knowledge can select the most suitable of any semantically well-formed alternatives passed on from the message processor. Frazier (1988) details the empirical evidence that supports the Forster model, showing how, for example, people (reading) can be temporarily garden-pathed (led towards an incorrect interpretation) even where preceding sentences contain disambiguating information to prevent this happening (Ferreira & Clifton, 1986; see Frazier, 1988). As Frazier (1988) points out, research adopting quite different views of processing has strongly suggested, in contradistinction to what is predicted by Forster's model, a rapid use of different information types in some stage or other of processing. Frazier's model is an attempt to combine this idea of rapid use of different information types (phonological, syntactic, etc.) with what she sees as a strength of the Forster model, namely the autonomous, encapsulated processors. Furthermore, she tries to exploit insights into the structure of the grammar provided by recent generative linguistic research. In her autonomy model, she departs from Forster in that she claims that the evidence shows that processing cannot be serial, that is, going from one module to the next. Serial processing would require the system to generate an enormous number of alternative interpretations and hold them in store until everything was resolved in the last stage. Rather, she says, processing is done in modular fashion with structure assigned by separate subsystems with regard to very specific and welldefined principles. Hence, even within the domain of syntax itself, a string (linguistic input) can be individually analyzed with regard to, respectively, its phrase structure and the extent to which it obeys case theory (constraints on how case is assigned), binding theory (e.g. how anaphors are linked 15
Michael Sharwood Smith to given antecedents), the theta principle (roughly assigning one and one theme only to given arguments) and so on. Direct reference can thus be made to those highly constrained subsystems which, in current Chomsky an theory, apply in different combinations to produce a wide variety of natural grammars and never any humanly impossible grammars. Language processing in Frazier's model can be concurrent since the analysis of the string by one processing module is never going to conflict with the analysis of another; presumably some phonological preanalysis has to be done first in order to present candidates for concurrent processing. After this, input is presented simultaneously to different processors for analysis. A simple analogy would be the identification of a particular object, say an egg in a bed of rocks, whereby one would not first check the egg shape, using a "shape processor" against the object present in the visual field, and then the color, and then the size, using a "size processor," but rather all these parameters would be checked out concurrently - the various processors would get to work on the basic visual input provided by the first stages of processing at the same time. A final decision about the interpretation of the visual signals would thus be arrived at much more quickly. It is instructive to note, in passing, that conscious problem-solving does seem to be serial, and consequently slow and laborious and that the conscious meta-analysis of linguistic information, like a serial analysis of the visual field is also slow and laborious.
Implications for bilingual processing
Knowledge implications In simultaneous bilingual acquisition, bilingual developmental processing, that is, processing input for the development of competence, must, if we accept the arguments mentioned earlier, at least be modular in one sense. The linguistic subsystems that, by hypothesis, obtain for one language must surely obtain for the other language as well. It seems hardly likely that a second language system that is mentally represented in a quite different way should exist alongside, or be integrated into a first language system, that is, in someone who has always been bilingual. Whether second languages learned later have quite a different structure and developmental history is a matter of debate in current second-language research. However, whatever the way in which different languages are represented in the bilingual mind, evidence from crosslinguistic influence in language loss and language acquisition (see Dechert & Raupach, 1989; Gass & Selinker, 1983; Kellerman & Sharwood Smith, 1986; Obler & Hyltenstam, 1989; Seliger & Vago, in press; Weinreich, 1953) would suggest that LI and L2 knowl16
Language modules and bilingual processing edge is not rigidly separated and that knowledge in one system can indeed affect the shape of knowledge in the other. What this precisely means with respect to the idea of autonomy or encapsulation is, of course, theory-dependent. Since knowledge is only "visible" in performance, any crosslinguistic influence between LI and L2 could still be the on-line accessing of what are still separate systems, as in codeswitching. In other words, crosslinguistic influence could be a purely realtime control phenomenon. For example, sequential bilinguals, just like simultaneous bilinguals, producing utterances that exploit their knowledge of more than one language may simply be doing this for reasons of expediency. At the same time, they may be perfectly aware of what aspects of their utterance belong to which language. The systems remain separate but their output can be combined during performance (Meisel, 1989; Sharwood Smith, 1986). If crosslinguistic influence really does take place at the knowledge level and if L2 knowledge is influenced by the structure of resident LI knowledge, then explanations about where and when crosslinguistic influence can and cannot take place should make reference to modular theories of language. It becomes vital in that case to ask how each language is represented and used. If the second language in a non-native user is encyclopedic in nature, then one would expect crosslinguistic influence to operate only between modules that shared the same general cognitive structure, for example, pragmatic knowledge. To take the way anaphoric reference works, for example, one would not expect, under this scenario, the way the firstlanguage grammar required anaphors and antecedents to be bound to be transferred to the second-language system. Since the first-language binding system is, by hypothesis, one of the autonomous processors characteristic of syntactic but not encyclopedic knowledge, one would not expect its properties to be copied on to a second-language system that was of a totally different kind. However, if the second language has the same cognitive make-up as the first language, if it is modular and non-encyclopedic in the same ways, then, in principle, crosslinguistic influence might be expected to occur at all levels (see discussion in Perecman, 1989). Control implications As far as processing control is concerned, rapid code-switching behavior in bilinguals makes it clear that different linguistic systems are often activated concurrently in production. If we consider receptive processing, and specifically how bilinguals process the speech stream, the old question arises of whether or not this is done by a bilingual system requiring a pre-setting of the language system (LI or L2) or at least a very early identification with regard to which language system was to be accessed (see Albert & 17
Michael Sharwood Smith Obler, 1978; Macnamara, 1971). It will be assumed that the clear majority view is the right one, namely that there is no neat division in bilinguals such that for each language, a totally separate, autonomous and encapsulated system is available in reception. Rather, the incoming speech is initially processed by rapid and mandatory mechanisms that make no distinction regarding what language system is expected, that is there is no preattentive language switch. This means that at the outset, at the very least, both/all language systems in the bilingual are activated. In a listening/ speaking situation, this means speech-processing and lexical access. However, it is clearly the case that different rules or routines have to be activated to continue processing efficiently once this initial stage is passed. This stage would seem to come when the phonological analysis triggers a search in the lexicon prior to the parallel, modular syntactic processing that Frazier talks about. Lexical representations in bilinguals are presumably tagged for which language system they belong to (whether or not there is any cross referencing with equivalents in the other language system) so that a search can be carried out efficiently. Lexical representations will thus contain information that triggers access to the L2 syntactic processor. It is clear that in a sequential L2 context, problems encountered at an early stage in processing the speech stream may have serious implications for later stages. Even if initial analysis of the speech stream results in the activation of a particular language-specific set of processors, those phonological processors need to be developed enough to cope with the task. If not, the processing program simply crashes - the learner registers a failure to process the utterances and perhaps attempts to guess from the context using real world knowledge. Another possibility is that the learner automatically co-opts fully developed phonological processors that are not actually appropriate, that is, "belong" to another language system. Sajavaara (1986) suggests that Finns processing English will have serious problems segmenting the speech stream if they deploy Finnish processing mechanisms that assume no consonant clusters in syllable-initial position. If segmentation runs into problems due to Finnish interference, then everything else that follows suffers. Following Frazier, this means that the learner's lexicon cannot be accessed efficiently and that subsequent stages of processing will suffer. The input will be difficult or impossible to handle. This means that Finnish learners of English will have to become aware of the syllable structure of English and, following this development in their knowledge of Finnish, devise appropriate ways of implementing this knowledge in processing incoming utterances. It also means that Finns will be considerably more successful processing graphological information where word boundaries are marked by easily perceptible spaces in the visual field. A somewhat similar illustration, this time with respect to simultaneous French-English bilinguals, was provided recently by Cutler et al, (1989). 18
Language modules and bilingual processing French, being a syllable-timed language, has clear syllable boundaries which enable French native speakers to segment the speech stream accordingly. English is stress-timed. The syllable boundaries are unclear and attempting to segment the speech stream using syllable boundaries is problematic. Cutler and her colleagues provide evidence that adult bilinguals, classifying themselves as French-dominant (albeit generally recognized by others as being native in both languages) have no problem under experimental conditions segmenting in either language. English-dominant bilinguals, on the other hand, apparently do not process the French speech stream like natives under these conditions. Cutler etal. make the intriguing point that perhaps syllable-timing is the marked option in languages in general. That is, the French-dominant learners learnt an exceptional method of segmentation and had no difficulty also segmenting via the default means when necessary (in this case with English). The English-dominant speakers had begun with the unmarked, default option and then found it difficult to switch to the exceptional, marked technique when coping with French. Cutler et al. (1989) therefore suggest that perfect bilingualism may well be a myth. In this specific example, bilinguals simply learn one method of segmenting speech and stick with it. Marked options include unmarked option, but not vice versa. While it is difficult to assess the validity of this claim without further investigation, it does at least provide a useful example of how, even at the control level, no general qualification such as "is accepted as a native" can be taken as interesting until, in modular style, a careful examination of all the various subsystems is conducted. It also shows that even in simultaneous bilinguals, who are apparently fully native in both languages, there seems to be a non-trivial degree of crosslinguistic influence. If preliminary analysis of the speech stream by bilinguals runs into problems, as in the Finnish and French examples above, we may assume that the phonological processor(s) (or presumably the graphological processors in the case of reading) represent the stage at which the system decides whether to engage one set of syntactic processors or another. A clear example would be bilinguals who have knowledge of, respectively, a configurational and non-configurational language, in other words, a language with a syntax based crucially on word order (like English) versus one which is not (like Russian, Japanese, Chinese). If processing the signal at the phonological/graphological stage results in the identification of English, then the syntactic processors to be engaged obviously have to follow configurational principles and take word order not only as stylistic but as relevant to the identification of object noun phrases and so on, that is the assignment of phrase structure. At a simple non-technical level, one could say the more fixed word-order languages require immediate attention to the order of identifiable items in a string whereas a freer-order language with a rich morphology would require that attention should be initially 19
Michael Sharwood Smith directed towards, say, inflections. If attention, for one reason or another, is misdirected, then the subsequent assignment of higher status to inflections in a fixed word-order language (probably poor in inflectional morphology anyway) would exhaust processing capacity better directed elsewhere. The result should then be an empirically observable delay in processing. Clearly, a well-defined theory of processing can be of great benefit in contributing to problems of a specifically bilingual nature.
M eta mode
So far, little has been said about the metalinguistic dimensions of knowledge and control amongst bilinguals. Children, be they bilingual or monolingual, develop language sensitivity at an early age, allowing them to appreciate and make observations and jokes about the words and structures of the language(s) they possess. This manner of thinking about language may be called operating in the meta mode. It would seem, then, that bilingual children just as monolingual children (and perhaps more so: Ben-Zeev, 1977; Cummins, 1976) operate in the meta mode from an early age although the nature of this operation differs from that of more mature learners by virtue of their cognitive immaturity. Older and more mature learners can make use of the more elaborate (encyclopedic) knowledge about the formal properties of language gained during formal education. For this reason it may not be helpful to distinguish metalinguistic awareness as one end of a scale which ranges from sporadic insights into aspects of language up to the highly formalized understanding of language entertained by academic linguists, that is, metalinguistic knowledge. If the modularity hypothesis as sketched above holds, then the development of metalinguistic ability from the early years onwards must be seen as separate from, and parasitic upon, the growth of intuitive linguistic ability. The child becomes able to reflect on linguistic properties only when these properties have been acquired. Both kinds of ability may be seen in terms of mental representation versus the real-time mechanisms for acting on those representations (see Bialystok, this volume). The separate nature of metalinguistic awareness is a view held by Karmiloff-Smith (1986), who sees it as an optional final stage in LI development. The idea that metalinguistic knowledge is part of what Fodor very generally calls the central processor helps us to see metalinguistic awareness/knowledge as allowing the language user to consciously reflect on and analyze language in the same way that a game of tennis or the act of riding a bicycle can be analyzed without any necessary connection between the degree of explicit knowledge of the person doing the analyzing and the actual degree of skill he or she has in the activity being analyzed. As has 20
Language modules and bilingual processing been indicated above, pragmatic knowledge, the interface between realworld, encyclopedic knowledge on the one hand and, on the other, linguistic competence, would have the same kind of open architecture as a metalinguistic knowledge. The functional aspects of these two kinds of knowledge will of course be difficult since, in order to function efficiently as a language user in given situations, it is vital to know the pragmatics of a language. Being in possession of metalinguistic knowledge, however, is not likely to play the same crucial role in one's day-to-day activities. Notice, however, that a degree of metalinguistic awareness may be important for successful social functioning: to understand the full significance of a speech or a television commercial or indeed jokes based on word-play will be a distinct advantage in dealing with fellow members of a given speech community. In any case, the development of pragmatic and metalinguistic knowledge will, by hypothesis, follow the course characteristic of encyclopedic learning, with the appropriate development of control itself typified by a slow and laborious first stage then later, smooth and automatic provided there has been enough practice, and, crucially, corrective feedback. To the extent that sequential L2 acquisition does not proceed as does first language acquisition, we also expect this pattern to emerge in other areas of the language. The metalinguistic abilities of bilin^uals may well be, as some have suggested (Ben-Zeev, 1977; Cummins, 1976; Kessler & Quinn, 1982; Leopold, 1939-1949) superior to that of monolinguals. In the process of acquiring and using different languages, they may have much more occasion to reflect consciously upon the ways languages differ in all kinds of ways. Looking upon this from a modular point of view, metalinguistically superior language users will not necessarily be superior in their command of other levels of language: the English-dominant French speakers were still not able to segment syllabically as monolinguals do. There is good reason to suppose that the benefits will, again as some have suggested, be in terms of general cognitive development, since in both cases the relevant cognitive domain is not the language-specific module but the central processor (see Diaz & Klingler, this volume; Johnson, this volume). By the same token, metalinguistically superior people, in being more sensitive to all aspects of language that fall outside the language-specific module(s), will, ceterisparibus, be more likely to show a flexible and creative use of language in areas accessible to conscious manipulation. This throws some interesting light on the traditional insistence that learning classical languages, a metalinguistic activity par excellence, is good for mental development. Presumably, it also facilitates relevant aspects of foreign-language learning (within the domain of deliberate manipulation). From a real-time processing point of view, in that metalinguistic processing will by definition be linear and, for the average language user at least, relatively slow and 21
Michael Sharwood Smith laborious, going into the meta mode will disrupt fluent performance and, to the extent that bilingual speakers may have a more enhanced metalinguistic ability, they will have more problems with fluency where they are unable to switch the meta mode off. High-speed metalinguistic processing may be achieved, but this will only arise amongst individuals who have had a great deal of practice, and, where sophisticated structures are concerned, only occur amongst the more creative members of a speech community.
Conclusion
As the above discussion has hopefully made clear, much is to be gained from explicitly adopting some version of modularity theory in investigating psychological aspects of bilingualism. The fact that research so far has already indicated differential effects in code-switching and mixing with regard to different levels of language (Perecman, 1989) suggests an implicit modularity in some approaches anyway. Work done more directly within the domain of generative grammar, where the notion of modularity gains greater precision, turns out to be extremely relevant for understanding the bilingual mind, how bilingual knowledge is represented and how it is used. This, on reflection, is hardly surprising, since Chomskyan metatheory is essentially an attempt at illuminating problems of language acquisition. The usefulness of making a distinction between knowledge and real-time processing should be fairly clear. However, it has simply not been obvious how abstractions such as binding theory, theta theory, and case theory, for example, bear on real-time use of language. One important example here is Frazier's work, which makes it clear that the grammatical subsystems used to explicate linguistic universals may also be implicated in what has here been called knowledge processing, and, by extension, bilingual knowledge processing as well. Modularity, in combination with the knowledge/control distinction, should help us to formulate questions about all areas of bilingual ability more precisely. Depending on the kind of knowledge or processing-control mechanisms we are looking at, we can make reference to particular welldeveloped theories in sister disciplines. We are also protected against rejecting some potentially useful theory of development or processing prematurely on the grounds that it did not work for, say, the development of one particular, such as the lexicon. In addition to this, as the syllabletiming segmentation example suggested, findings in bilingual research may have interesting repercussions for the ways theoretical linguists view relationships between language types and the implications this variation has for learnability. 22
Language modules and bilingual processing References Albert, M. & L. Obler 1978. The bilingual brain. New York: Academic Press. Ben-Zeev, S. 1977. Mechanisms by which childhood bilingualism affects understanding of language and cognitive structures. In P. A. Hornby (ed.), Bilingualism: psychological, social and educational implications. New York: Academic Press. Bialystok, E. & M. Sharwood-Smith 1985. Interlanguage is not a state of mind: an evaluation of the construct for second language acquisition. Applied Linguistics, 6,101-107. Chomsky, N. 1980. Rules and representations. Oxford: Blackwell. Cummins, J. 1976. The influence of bilingualism on cognitive growth: a synthesis of researchfindingsand explanatory hypotheses. Working Papers on Bilingualism^, 1-43. Cutler, A., J. Mehler, D. Norris, & J. Segul 1989. Limits on bilingualism. Nature, 340,229-230. Dechert, H. & W. Raupach, 1989. Transfer in production. Hillsdale, NJ: Ablex. Ferreira, F. & C. Clifton, 1986. The independence of syntactic processing. Journal of Memory and Language, 25, 348-368. Fodor, J. 1983. Modularity of mind. Cambridge, MA: M.I.T. Press. Foster, K. 1979. Levels of processing and the structure of the language processor. In W. E. Cooper & E. C. T. Walker (eds.), Sentence processing. Hillsdale, NJ: Erlbaum. Foster, S. 1985. Taking a modular approach to universals of language acquisition. Paper presented at SLRF, Los Angeles. Frazier, L. 1988. Grammar and language processing. In F. J. Newmeyer (ed.). Linguistic theory. Cambridge: Cambridge University Press. Gass, S. & L. Selinker 1983. Language transfer in language learning. Ann Arbor, MI: Newbury House. Jackendoff, R. 1987. Consciousness and the computational mind. New York: Academic Press. Karmiloff-Smith, A. 1986. Stage/structure versus phase/process in modelling linguistic and cognitive development. In I. Levin (ed.), Stage and structure: reopening the debate. Norwood, NJ: Ablex. Kellerman, E. & M. Sharwood Smith 1986. Crosslinguistic influence in second language acquisition. Oxford: Pergamon Press. Kessler, C. & E. Quinn 1982. Cognitive development in bilingual environments. In B. Hartford, A. Valdman, & R. Foster (eds.). Issues in international bilingual education: the role of the vernacular. New York: Plenum Press. Lasnik, H. 1982. On certain substitutes for negative data. Paper presented at the Workshop on Learnability and Linguistic Theory, May 1982. University of Western Ontario. Leopold, W. 1939-1949. Speech development of a bilingual child: a linguist's child (4 vols.). Evanston, IL: Northwestern University Press. Macnamara, J. 1971. The linguistic independence of bilinguals. Journal of Verbal Learning and Behaviour, 10, 480-487. Meisel, J. 1989. Early differentiation of languages in bilingual children. In L. Obler 23
Michael Sharwood Smith & K. Hyltenstam (eds.), Bilingualism across the lifespan. Cambridge: Cambridge University Press. Obler, L. & K. Hyltenstam (eds.), 1989. Bilingualism across the lifespan. Cambridge: Cambridge University Press. Perecman, E. 1989. Language processing in the bilingual: evidence from language mixing. In L. Obler & K. Hyltenstam, (eds.), Bilingualism across the lifespan. Cambridge: Cambridge University Press. Sajavaara, K. 1986. Transfer and second language speech. In E. Kellerman & M. Sharwood Smith (eds.), Crosslinguistic influence in second language acquisition (pp. 66-79). Oxford: Pergamon Press. Seliger, H. & R. Vago, in press. Language attrition: structural and theoretical perspectives. Cambridge: Cambridge University Press. Sharwood Smith, M. 1986. The competence-control model, crosslinguistic influence and the creation of new grammars. In E. Kellerman & M. Sharwood Smith (eds.), Crosslinguistic influence in second language acquisition. Oxford: Pergamon Press. Weinreich, U. 1953. Languages in contact. The Hague: Mouton. White, L. 1987. Against comprehensible input: the input hypothesis and the development of second language competence. Applied Linguistics, #, 95-110.
24
2
Phonological processing in two languages IAN WATSON
Phonology is to a large extent the Cinderella of bilingual studies. The most immediately obvious aspect of speech is ironically that which has been least intensively researched with respect to bilingual speech behavior in general, and bilingual acquisition in particular. One reason for this may be its very peripherality in language processing. For the non-specialist, phonology is so predominantly concerned with the mechanics of language realization, so far removed from the broader psychological issues of cognitive development and processing, that it fails to grab the attention (even of the linguist who is a non-phonologist) in the same way as matters semantic, pragmatic, or even syntactic. These others may appeal as worthy of study not only in their own right, but also for their links with non-linguistic cognitive phenomena. The subject of phonology is in itself so complex and ridden with controversy that it is difficult to know where to start to find a conceptual framework which will permit bilingual acquisition to be investigated. Even the analysis of the adult sound system of a single language involves a plethora of theoretical choices. Is the aim a taxonomic phonemic account, and if so how does one deal with any instances of neutralization or overlapping? Should one seek underlying forms, and, if so, how abstract should they be? What about suprasegmental phenomena - will one try to give a metrical account of rhythm, what about tones/intonation, how do these interact with segments? Are there segments? How many dimensions are there? What to do about feature geometry, etc? Each of these choices will fundamentally affect one's view of how phonology works, and therefore how it can be acquired. Worse still, any one of the former may leave open a number of possible views of the latter, a point returned to below. It is thus unsurprising that, even with respect to the acquisition of phonology for monolingual children, the number of studies is very restricted, especially compared to the massive theoretical literature; there are far fewer studies on bilinguals. One of the major problems in acquisition studies lies in drawing the line between phonological and phonetic processing abilities. In a general sense, the term phonology is, of course, often used to refer to all the 25
Ian Watson sound-related aspects of language knowledge and behavior. For the generativists (e.g. Chomsky & Halle, 1968), especially, the phonological component of a grammar is that module which converts a string of lexical/ grammatical items, duly arranged into an acceptable surface structure, into what a speaker actually says. It further, by some ill-explained process, allows the hearer to decode the utterance back into a string of linguistic units. Close attention to the processes of real speech behavior, however, requires recourse to the traditional, but problematic, distinction between the underlying phonological aspects of linguistic expression and the phonetic aspects of how phonological units are realized in perception and production. In this schema, what remains under the heading true phonology is concerned with the structural use of sounds to build linguistic units. Here, one would expect, minimally, (i) information about the number of different contrasts between sounds a particular language makes use of in its lexicon; (ii) enough information about the physical identity of the sounds themselves to allow them to be kept distinct at various levels of processing; (iii) (depending on theoretical orientation) how the structures of different forms of the same lexical item vary. A child's acquisition of phonology, in this sense, is the acquisition of knowledge about the patterning of sounds in his or her native language. Phonetics, in contrast, is concerned with the processes by which a linguistic, phonological representation is turned into an acoustic signal by means of physical gestures, and with how this acoustic signal is reconverted to a linguistic representation by the perceptual system. Clearly, it is closely linked to phonology as the linguistic representation is expressed in phonological units, in phonemes or allophones. But the fact of knowing what the phonemes of a language are is no guarantee of phonetic proficiency in that language; each language has specific and different phonetic ways of realizing its phonemes in different contexts (Nolan, 1989). Thus both German and English have systems which contrast the phonemes /p/ and /b/; these contrast in word-initial position primarily because the former is aspirated and the latter is not. However, aspiration is typically longer in German than in English (Kohler, 1977). In French a similar contrast is normally realized by the presence of closure voicing in /b/, and its absence in /p/ (see, e.g., Simon & Fourcin, 1978). Furthermore, in each language, the same phonemes may be realized and distinguished by different means in positions other than word-initial. A traditional way to express this has been to associate each phoneme in a language with a set of allophones which realize it in different circumstances, but even this approach fails to yield a presentation which captures every single articulatory or acoustic parameter used by the native speaker in continuous speech, with its range of coarticulatory and stylistic variation (see, e.g., Nolan, 1982). Nor can 26
Phonological processing in two languages this range be adequately represented using the feature notations used in the different branches of generative phonology. A more useful conceptualization is one in which a phonological representation in terms of phonemes, possibly with some (largely extrinsic) allophonic variation dealt with by phonological rules, is realized by an array of routines which each native speaker has developed to translate the phonological units into an acoustic signal (via articulatory gestures) in speech production, and to translate acoustic patterns into phonological units when perceiving it. The learning of speech production and perception thus involves not only learning the phonological system, but also the identification of acoustic patterns with elements of that system and the successful imitation of those patterns. The notion that it is quite possible to achieve one of these elements without mastering the others is crucial to understanding the phonological development of children, and especially bilingual children. A further problem arises with the description of both phonological and phonetic phenomena; they are highly variable. Across different accents, identical lexical items may have different phonemic composition, either because the phonemic systems of those accents are different (compare Brittany French, which contrasts /a/ and /a/, to the modern Parisian accent, which does not), or simply because the phonemes are differently lexically distributed (Southern British English and the Midlands variety have identical monophthongal vowel systems, but in the lexical item one, the former has /A/, the latter /o/). A single phonemic unit may also have different phonetic realizations; for example, the French /p/ phoneme is normally realized without aspiration over most of France, but in some areas of the north and in Belgium is frequently aspirated (Goudailler, 1981). On a more detailed level still, even sounds which are phonetically identical will correspond to different acoustic signals when spoken by different people, or even the same people on different occasions. The task facing any child learning the phonology and phonetics of a single language is therefore highly complex. From an ever varying mass of acoustic input, he or she has to: 1. learn to recognize distinct, but non-invariant acoustic patterns; 2. deduce the set of oppositions which constitute the phonological structure of the language; 3. associate the acoustic patterns with the phonological system, despite the non-invariance of the former; 4. master the correct articulatory routines to produce acoustic patterns which satisfy other native speakers as being adequate realizations of different phonemes. Since native speakers of many, possibly all, languages have at least receptive abilities for a wide range of accents, the child will have to learn to 27
Ian Watson perform step 3, above, in a way allowing many-to-many relationships between acoustic signal and phonological unit; and since we all use different accents in different situations, he or she will have to learn different articulatory routines for various phonemes in step 4. None of the four stages outlined above is remotely well understood; indeed they represent only one of many different interpretations of how phonological competence is acquired (see Best, 1985, for a quite different view; Watson, 1985 for objections to it). It is therefore hardly surprising that modelers of bilingual language processing have tended to avoid this area. For approaching the bilingual acquisition of phonology does not merely add to the problems faced, it multiplies them. At each stage of the acquisition process, the child is faced with many more different signals, but without automatically knowing which language they belong to, while younger children cannot initially even know that they represent two different languages. After differentiation between the languages, both the differences and the similarities between the two systems may lead to interference which hampers the acquisition process. That cross-linguistic differences should cause interference is obvious: the child can easily fail to realize that a feature of one language is not shared by another. Crosslinguistic similarities are problematic when phenomena in the two languages are seductively alike, without being identical. Nevertheless, any problems caused by similarities or differences may not be serious. Provided the child's speech behavior is sufficiently close to that of other speakers of each language to allow lexical distinctions to be maintained and communication to be effected, slight infelicities in pronunciation (by monolingual standards) may not be seen as insignificant. There are certainly some bilinguals who seem to achieve monolingual-like pronunciation in both of their languages, at least as far as a phonetically trained listener can judge. Whether this is true of a small or large percentage is an empirical question, but in any case we should like to know if this group is different from those who do not achieve such a balanced performance. However, a further problem lies in deciding what the standard should be by which a balanced performance is judged. Impressionistic judgments must be supplemented by instrumental measurements, and experimental phonetics allows minute differences in individual pronunciation and perception processes to be measured, at a level of detail which is too fine to be perceptible to the best-trained ears. If there is any way in which bilinguals deviate from monolingual norms, we are in a position to know, and if this is the case, it will give us evidence that the two groups may use alternative processing strategies. Of course, monolingual norms are themselves constructs; they conceal such enormous variation that it is very difficult to be sure that a bilingual 28
Phonological processing in two languages is doing something that a monolingual would never do. We must be aware that we can, at best, note tendencies, rather than make absolute statements. To this end, in this chapter the main findings in the literature will be reviewed, not only those pertaining directly to bilinguals, but also, where relevant, those based on monolingual studies. Some new data will also be presented, from an ongoing study of phonological acquisition by FrenchEnglish bilinguals. Three main questions will be addressed in what follows. First, to what extent do bilinguals actually keep the phonological systems of their two languages separate, rather than using a partially merged system? Second, what is the relationship between production and perception? Third, how similar are the stages by which bilinguals acquire phonology to those followed by monolinguals? A hypothesis will be developed that, at least at some stages of acquisition, bilinguals do, in fact, take advantage of the variability which each speech community will accept, in order to achieve the twin aims of reducing the amount of different processing required while avoiding the pitfalls of sounding unlike most native speakers of the language.
Phonological acquisition
Two apparently contradictory claims have been made about the early stages of phonological acquisition. On the one hand there is experimental evidence that infants' early production and perceptual abilities show signs of moving towards the phonological systems of their target language. On the other, it has often been suggested that when speech begins, phonology is hardly acquired at all; Ferguson and Farwell (1975) suggest that in the beginning words and their meanings are learned whole, with no analysis into smaller phonetic units, and consequently no awareness of individual oppositions. The contradiction is resolved by detailed examination of the experimental evidence, discussed below, which suggests that any tendencies in prelinguistic infants reflect either inherent psychoacoustic limitations or phonetic, rather than phonological development. Several papers show infants either producing particular sounds or performing perceptual tasks which seem to be related to the phonological structure of the sounds around them. With respect to the first of these, although children's early vocalizations include all the sounds in the human phonetic repertoire, Locke (1986) shows an increasing tendency in the babbling stage (ages between six and nine months to eighteen months) to produce the sounds which are found in the target language. While this 29
Ian Watson is undoubtedly strong evidence that children's babbling is partly conditioned by the sounds around them, there is no need to believe that this is anything other than a phonetic phenomenon. As a preliminary to true phonological acquisition, the child directs its articulatory activity towards practicing the sounds it hears regularly; there is, however, nothing linguistic about this activity at this stage. A similar argument obtains with respect to the perceptual data. Results from tests concerned with infants' responses to synthetic voice onset time (V.O.T.) continua have suggested that the use of this measure to make categorical discriminations between phonemes differing in voicing quality is both present at a very early stage, and language-dependent. In a study of six to eight month-old infants from English and Spanish speaking environments, Eilers, Gavin, and Wilson (1979) used a head-turning procedure to identify discrimination ability. As noted above, languages differ in the way they realize voicing contrasts, a major distinction being between those like French and Spanish, which contrast initial prevoiced and voiceless unaspirated stops, with a V.O.T. category boundary in the region of 0-15ms., and those like English and German, which contrast voiceless unaspirated (short-lag) and aspirated (long-lag; category boundary around 25 ms.). The English-environment children showed reliable discrimination only between short- and long-lag stops, in line with the English sound system. The Spanish infants were able to discriminate all three different groups: that is, they were sensitive to the contrasts used in both languages. One interpretation of this apparent imbalance in abilities has been prompted by psychoacoustic studies which show that the English contrast corresponds to a discrimination facility inherent in the mammalian auditory system (Kuhl & Miller, 1987). Non-human animals can be trained to discriminate V.O.T. continua categorically, with a category boundary in the 25ms. region. The possibility thus arises that human infants have an inbuilt predisposition to use such a boundary, which has to be changed by experience if they are exposed to a language which functions differently. Englishenvironment children would never have to make such a change, whereas those exposed to Spanish might do so, initially, by a process of suppletion rather than replacement. Even in the first few months of life, the child can thus be seen as establishing the bases of a future phonological system. However, three types of evidence argue against this interpretation. One is that much older children are far less certain in discrimination tasks, irrespective of their language, than would be expected if their very early linguistic exposure had established the phonological categories used by their language. Simon and Fourcin (1978) showed English speakers unable to label continua categorically until the age of five years, and French speakers not attaining sharp categorical labeling until the age of eight. Clearly, whatever the explanation of young infants' laboratory per30
Phonological processing in two languages formance, it does not translate directly into practically useful phonological knowledge. Besides, evidence from other perceptual tests suggests that the ability to discriminate between all three voicing categories may be less language-specific than Eilers, Gavin, and Wilson's results imply. Streeter (1976) showed that two-month-old Kikuyu infants were sensitive to both contrasts, yet, at the relevant (bilabial) place of articulation, Kikuyu use neither! Furthermore, Eilers, Wilson, and Moore (1979) showed that, although six out of eight six-month-old English-environment infants did not show any sensitivity to the prevoiced/voiceless distinction, two did, and a group of English adults reliably discriminated both. Interpretation of such conflicting evidence is difficult, but Burnham (1986) suggests that the general pattern is an initial ability to distinguish among all three categories, followed by the waning or disappearance of any discrimination skills which are not reinforced by environmental stimuli. Thus, English infants show less consistent ability to make the prevoiced/ voiceless distinction because they are not exposed to these two categories of sound in the way that Spanish children are. Nevertheless, the initial discrimination abilities may not be completely lost, because prevoiced stops do exist phonetically in English, as an allophonic variant of the unaspirated set. Similarly, Kikuyu does have a contrast between prevoiced and voiceless unaspirated stops in some places of articulation. Thus whether or not sounds of a particular phonetic type function phonemically in a language, infants may be exposed to some of them. Even the infants who seemed, according to the Eilers, Gavin, and Wilson (1979) and Eilers, Wilson, and Moore (1979) tests, not to be sensitive to this distinction may actually have been so, albeit at too low a level for their head-turning technique to show. In any case, it is implausible to attribute the discriminatory abilities of these infants to developing phonological skills. By definition, a phonological unit forms part of a linguistic system, and children of the relevant ages show no evidence of any ability to produce or understand sounds as tokens of linguistic items. Indeed, infants at this stage may well simply be cognitively incapable of distinguishing phonological classes (MacKain, 1982). The phenomena in question are rather examples of psychoacoustic or phonetic processing abilities (or possibly the development of the latter from the former; see Best, 1985; Watson 1985) at a stage when phonetic processing has not yet begun to serve linguistic needs. Unfortunately, experimental studies of bilingual-environment children in the prelinguistic stage are lacking. Longitudinal comparison between the development patterns of monolinguals and those exposed to two contrastive phonetic inputs at this prelinguistic stage might be very informative, although the apparent need for children in their early linguistic years to relearn these skills must leave a question mark over the exact relevance of early discrimination abilities. 31
Ian Watson First words There is no evidence that when the child's first words appear, they are stored, produced, and recognized as anything other than unanalyzed wholes, associated with specific situations (Barton, 1976; Ferguson & Farwell, 1975; Menyuk & Menn, 1979). An experiment by Barton (1976) showed that a number of children aged 20-24 months were unable to distinguish words on the basis of initial voicing differences without a considerable training period, suggesting that this was not a process they habitually used. Yet children are often able to produce words contrasting in this feature at this stage, the implication being that they are doing so without awareness of the phonetic/phonological units of which the words are composed. In producing these first words, they may simply be transferring to a meaningful use articulatory routines practiced during the babbling stage (Locke, 1986). As the child's vocabulary increases, it rapidly becomes uneconomic to store words as wholes. At this point words are analyzed into smaller units, probably first into syllables, then segments (Mann, 1986). However, the ability to produce appropriately sounds which, according to perceptual tests, are not categorized in an adult-like way (or, in the case of bilinguals, a monoiinguaMike way) is one which persists for many years (Carramazza et al., 1973; Simon & Fourcin, 1978). This is a paradoxical finding, both because it is also often the case that children can perceive distinctions they cannot accurately produce (see, e.g., Smith, 1973) and because, in general, children must learn their phonological systems and speech habits to a large extent by means of their perceptions of others' speech (see further discussion below). Although the early stages of phonological development in bilinguals depend on a multiplicity of factors, bilinguals are probably least different from monolinguals early in the one-word-utterance stage. The majority of the world's bilinguals are probably only initially exposed to one language, so that the first words produced will inevitably be in that language. Besides, even true primary bilinguals, that is those who have been exposed to two different languages almost equally since birth and have continued to be so exposed, are normally unconscious of being bilingual at this stage. The notion of different languages only develops around the age of two years. With the onset of true phonological awareness, two distinctions must be drawn. The first is between simultaneous and successive acquisition of the different languages. Although this is not truly a binary distinction - exposure to a second language can begin at any age - there may be significant differences in the strategies used for phonological acquisition by those who develop two systems absolutely simultaneously, and those who have at least part of one phonological system in place before the other is broached. This contrast is discussed below, in the light of what is known about monolingual acquisition at this stage. 32
Phonological processing in two languages The second contrast is that already alluded to, between production and perception. The presence of a particular sound in an appropriate place in a child's utterances is no guarantee that the child is able to discriminate that sound consistently. It may simply be an easy sound to produce, one developed at the babbling stage. Correct productions can occur without correct perception. However, the mastery of a sound system must be based on the child's correct perception of the production of others, and subsequent modification of its articulatory behavior in the light of those perceptions. In this sense, perception must normally precede production in development. There will often be sounds which children categorize correctly perceptually, but cannot produce properly. In judging a child's progress, it will be essential to monitor both aspects of speech behavior separately. It is inappropriate to assume that they are isomorphic. Monolinguals and bilinguals learning phonology Once past the one-word stage, how does phonological acquisition proceed? One view sees the process as largely one of differentiation. The child learns to recognize as separate sounds which had previously been confused. The most comprehensive attempt to explain acquisition in these terms came from Jakobson (1968), who saw the acquisition of phonological contrasts as being governed by the phonological complexity of the phonemes to be acquired, as expressed in his distinctive-feature system. There would thus be a universally applicable sequence of development, and the order in which all the phonemes used in a language are learnt would be totally determinable. A number of writers quote evidence which seems to support this view. Linell (1979) quotes the sudden differentiation of III and /k/ after a period of using only the former, and claims that when awareness of a particular opposition does come, it is applied to all the known words containing it even if these were previously confused with other words because the distinction was not known. Fourcin reports evidence of afixedorder of acquisition, typically followed by children learning English in England, and shows that the salient factor is not the frequency with which phonemes appear in the input to the young child (Fourcin, 1978). Despite this, the Jakobsonian view clearly provides an inadequate conceptual framework for investigating the processes of phonological acquisition (Elliot, 1981; Menyuk & Menn, 1979). Although phonological acquisition may be seen as being attained once the set of a language's oppositions has been mastered, the point has already been made that it is not clear what learning an individual opposition exactly means. During acquisition, perception and production are often not in synchrony, some allophones of a phoneme are learnt before others, and, despite Linell's 33
Ian Watson observation above, children often use the correct sound in one word, while substituting a different sound for it in others, especially at the stage where at least some words are still stored as gestalts (Locke, 1986; Menyuk & Menn, 1979). There is considerable variation in order and speed of acquisition across different children; individual children reach the same goal in a variety of ways. In modeling acquisition, all of these factors must be allowed their place. Most of the evidence concerning the effects of bilingualism between the ages of two and four years comes from case studies, usually based on parents' notes (there is a dearth of scientifically controlled studies of any children in this age group). The children may be divided into two groups; (i) those simultaneously acquiring two languages; (ii) those acquiring one language after another. Simultaneous bilinguals There is agreement in the literature that primary bilinguals can attain the facility to function in two different languages simultaneously, without taking twice as long - or, it seems, any significantly longer time - than a monolingual needs to acquire one. Initially, the main task is that of differentiation; with a plethora of different stimuli around him, the child must recognize first that these represent the combined output of two different linguistic systems, and then identify which elements - and in effect which categorical distinctions - belong to and constitute which system. Most bilingual children seem to be unaware that they are dealing with two different systems until the age of about two, by which time phonological development may be well under way. Leopold (1939-1949) reports that Hildegard treated all her linguistic knowledge in both English and German as forming a single system until shortly before this age. Soon after her second birthday she started to show awareness of the difference, even using the word German without having been taught it, then slowly developing a clear awareness of the two systems. Similarly, Imedadze's child, learning Russian and Georgian, first began to separate the two languages at about twenty months (Imedadze, 1967). Once the discovery is made, differentiation begins, and it immediately becomes very difficult to glean from the case studies what processes are being used. An ordinary monolingual child goes through several stages in which he simplifies the system he is trying to learn. Bilinguals simplify as well, but do so in each language and crosslinguistically. Any attempt to define patterns or rules in what they do is inevitably hampered by the problem of dominance - the condition of the child being more capable in one language than in the other, often due to greater exposure to one - which recurs, though not ostensibly, in many case studies. There is very 34
Phonological processing in two languages little documentation of balanced subjects, and those studies which do exist are uninformative about the area of interest here. Ronjat (1913) is more concerned with discussing the effect on Louie of the one language one person principle; neither Celce-Murcia (1978, an essentially synchronic study) nor Pavlovich (1920), while claiming that their subjects mastered their two systems simultaneously, go into detail. Amazingly, these three are almost alone in describing fairly balanced subjects. Leopold's Hildegard was exposed far more to one language than the other for long periods. Of course, perfectly balanced bilinguals are almost certainly the exception rather than the rule. Nevertheless, there may be different processes of differentiation corresponding to greater or lesser degrees of dominance, and the lack of clear information on this in the literature is frustrating. Leopold considered that Hildegard had perfected her English phonology between the ages of three and four, but her German system was not completely native-like until the age of five, when she spent a protracted period in Germany. The problems seem to have been both phonetic and phonological, in that she could neither accurately pronounce nor accurately distinguish the front rounded vowels which are a feature of German, but not of English. On her return from her visit to Germany, Hildegard had become dominant in German, and for a time exemplified the other major problem faced by bilinguals who are dominant in one language; the difficulty of preventing the incursion of one (already formed) language system on the domain of the other - of avoiding interference, as defined by Weinreich (1953). After four weeks in a totally German environment, Hildegard was reported as speaking English with a marked German accent; on her return to the U.S. A. this problem disappeared, and she began to have low-level phonetic problems in pronouncing German. It seems quite possible that Hildegard may have attained differentiation, at least in the latter stages, by a process of superimposing the unknown system on the known, that is, by using the first system as a base, and building onto it additions and alterations. In other words, she might to a large extent be using a procedure known to be used by secondary bilinguals (although this does not totally account for phonological learning in secondary bilinguals; see Anderson, 1981). Of course, balanced primary bilinguals might do the same to a certain degree, especially if their two languages are such that the phonological system of one is much more easily learnt than that of the other; in this case the learning of the former might in fact facilitate that of the latter. But it also seems possible that a more balanced subject might use a different method, starting with an averaged system which has contrastive patterns and perceptual category boundaries intermediate between the two target systems and slowly and more or less simultaneously moving away from this in either direction towards the differential phonologies. In all prob35
Ian Watson ability both balanced and non-balanced children may use both strategies, the only difference being one of emphasis. (Partly) consecutive bilingualism Fantini's Mario underwent a different pattern of exposure to his languages, resulting in a different pattern of phonological acquisition (Fantini, 1985). It is a difficult pattern to characterize. Fantini himself refers to Mario as "one whose mother tongue is Spanish, but whose initial contact with English was only slightly delayed . . . hence a 'simultaneous' bilingual" (Fantini, 1985, p.37), but indicates that at the age of two, when phonological acquisition was well under way, his son was only an incipient bilingual', indeed Mario's first productive use of English postdated that of Spanish by a year. The case serves to illustrate the arbitrariness of such labels, but Mario contrasts with the earlier cases cited in the slightness of his exposure to the second of his languages at this stage, and for our purpose he will be treated as a consecutive bilingual. Fantini's account of Mario's phonological development is based on Jakobsonian principles, although he does also give important details of allophonic variation. Differentiation was a protracted process for the child. His Spanish consonantal system was almost complete at the age of 2 years 6 months, at which stage he used virtually no English sounds other than those which had close equivalents in Spanish. The correct patterns for English began to appear only after the near-completion of the Spanish, in a quick burst from 2 years 6 months to three years - though aspiration of unvoiced consonants did not appear till three years and two months. Even then, the two systems were not properly differentiated in production; the /th/ and /dh/ sounds were not mastered until the age of six, and numerous Spanishinfluenced pronunciations (/x/ for /h/) remained until this stage. Nevertheless, by the age of five, a phoneme-perception test demonstrated that Mario was clearly able to distinguish correctly all the English phonemes. Mario also showed some signs of interference from English in his Spanish pronunciation. Once he had mastered the English aspirated consonants, he began inappropriately to aspirate certain Spanish consonants. Possibly the imperfect separation of his two sound systems facilitated this interference. It is far clearer in this case than in Hildegard's that one phonological system was developed by building out from the other. This is not to say that all of Mario's English system was developed in this way; there may also have been some independent developments. The /th/ phoneme may be one example of this. Its late development is, as Fantini points out, common among monolinguals. Two conclusions may be drawn from this survey of the development of the phonological system. The first is that the developing bilingual does 36
Phonological processing in two languages have to learn processing skills which are unnecessary for the monolingual. He or she has to recognize that a sound system is entirely arbitrary, in that it is possible to use more than one to communicate. He or she must therefore learn to assign similar physical events to different systems of oppositions according to the linguistic context. This can be done effectively, with certain provisos, noted in the next section. However, each phonological system is not necessarily acquired in a way analogous to monolingual acquisition (though it may be that procedure is possible). Commonly, one system, or at least aspects of it, will dominate the other, so that the child fails to make some oppositions in one language, or at least produces some sounds in a foreign way, due to interference. The second, more negative, conclusion is that remarkably little is known about this essential stage in bilinguals' development. We lack instrumental data, and have almost no data at all on completely balanced primary bilinguals. The absence of experimental data means that we cannot be certain how far claims of monolingual-like abilities in both languages are valid; there are examples from modern phonetic research of distinctions which are consistently made in speech, but which even trained phoneticians cannot detect. Fortunately, there are far more studies of children aged from four upwards; these are discussed in the next section. Older children With a few exceptions, by the age of four or soon afterwards, children have a fully developed phonological system. This means that they are capable of recognizing all the phonemic distinctions in their target language, and also of pronouncing the words of that language in a way judged acceptable by other native speakers. It does not guarantee that all the parameters involved in their speech perception and production correspond to those used by adult speakers. In this section, evidence is reviewed that in both of these areas bilinguals, while apparently highly proficient in both their languages, may actually exhibit different strategies from monolinguals. Some preliminary results from an ongoing project comparing French and English mono- and bilinguals are discussed which suggest that the latter sometimes choose phonetic options which reduce the difficulties created by their need to use two systems, without thereby sounding in any way abnormal in either. A favorite investigation area for child perception in general, and bilinguals in particular, is the voicing distinction between cognate word-initial stop consonants. This is particularly suitable because, although children apparently master the voicing contrast at an early stage, neither their production nor their perception of the various acoustic features associated with it is adult-like for a number of years. Furthermore, as already noted, the realization of this contrast is highly variable cross37
Ian Watson linguistically. Simon and Fourcin (1978) showed that one acoustic feature used distinctively, V.O.T., was not used categorically in perception until the age of four by a group of English-speaking monolinguals, and not until eight or nine by French monolinguals. A second feature, the presence of a voice-excited Fl transition, is only reliably distinctive in English, and was learnt by the English group by the age of eight, and not at all by the French group. It has also repeatedly been shown that children's productions of V.O.T. distinctions only slowly move towards adult norms. Similarly, Ohde (1985), in a production study, showed that variations in fundamental frequency consistently found in adult subjects were not found in six-year-olds. This combination of a protracted learning period and crosslinguistic differences gives an excellent tool for the comparison of monolingual and bilingual phonetic development. Bond, Eddey, and Bermejo (1980) and Konefal and Fokes (1981), both measured V.O.T. in the productions of Spanish-English bilingual children. Bond's subjects were two sisters, aged four and seven, the latter being severely language-disordered. Both were successive bilinguals, and indeed had only been learning English for six months; the older had reached only the two-word utterance stage in English, but the latter was "beginning to speak English in school" presumably with some fluency. Strikingly, the younger child had clearly differentiated V.O.T. patterns for her stops in the two languages, with figures corresponding well to monolingual norms. She prevoiced her voiced Spanish stops, aspirated her voiceless English stops, and produced Spanish voiceless and English voiced tokens in the short-lag region. The older child, in contrast, produced all her Spanish stops in the short-lag region - a pattern typical of younger children. Konefal and Fokes' subjects were three sisters, aged four, seven and ten, who also had Spanish as their mother tongue. The ten-year-old was language-disordered. All three had been exposed to English for three years at the time of the study. Again, the two normal children show clearly differentiated patterns in their two languages. The younger one contrasts with the four-year-old in the earlier study in that her Spanish voiced stops were concentrated in the short-lag region, with the voiceless set spanning the upper part of that region and the long-lag region. The seven-year-old, however, like Bond, Eddey, and Bermejo's four-year-old, consistently prevoiced Spanish /b, d, g/, and had short lags in /p, t, k/. Both children contrasted short- and long-lag stops in English although the younger of them did this less consistently, with a number of longer V.O.T.s in the voiced set. The oldest, disordered child showed considerable overlap between the two categories in both languages. The normal developmental pattern for V.O.T. production is for all stops to be produced in the short-lag range initially, with production extended, as appropriate for the language, to the long-lag or prevoiced ranges (Zlatin 38
Phonological processing in two languages & Koenigsknecht, 1976). For some children this process of extension may not be complete until well after the fourth year. We cannot, therefore, necessarily interpret the lack of prevoicing in Konefol and Fokes' four-yearold as being ascribable to English-Spanish interference. In fact, all of the normal subjects in these two studies seem to offer strong confirmation that, with respect to V.O.T., developing bilinguals can relatively easily produce different patterns in two different languages, even when those patterns are contradictory, in the sense that Spanish voiced stops and English voiceless ones are both in the short-lag region. The mastery of such duality of patterning seems to require very little time; Bond, Eddey, and Bermejo's four-year-old had achieved it in six months, although she had the advantage of coming to her second language at an age when V.O.T. production is rapidly developing in many children - and hence, perhaps, easily modified. But if in the overall pattern of V.O.T. development, bilinguals are similar to monolinguals, there is nonetheless evidence that their realizations of voicing contrasts do have some significant differences. Watson (in press) reports on a study comparing monolingual and bilingual development of a number of acoustic features associated with the voicing contrasts in English and French; two, V.O.T. and Fl onset frequency, will be discussed here. Bilinguals and monolinguals aged six and ten were compared, and there were also adult control groups. One of the striking results was the variability in productions, even by the monolingual adults. In French, phonologically voiced stops were most frequently prevoiced, but there were many instances where they were produced in the short-lag region, or were prevoiced, but with a break in the voicing during or immediately after closure (see Goudailler, 1983). French voiceless stops were sometimes significantly aspirated. Both of these tendencies were more pronounced in the child groups; indeed the monolinguals showed a strong tendency to aspirate velars. This tendency did not diminish between the two age groups. (See table 2.1.) The bilinguals showed a more consistent trend to reduce voice onset times with age, especially for the Frencn voiced tokens. However, they also produced the voiceless series with more aspiration than the monolinguals in both age groups and in both languages. This series thus comes to resemble the English series more than is typical for monolinguals (although their English voiceless stops are even more aspirated). This increase in aspiration results in a second disparity between the bilinguals and the monolinguals. In English there is a consistent, and perceptually salient, difference in the onset frequency of thefirstformant in vowels following voiced and voiceless stops; this follows inevitably from the presence of a period of aspiration in the latter, and its absence in the former. Because French does not systematically use aspiration, this difference is rarely present and has been shown to be perceptually unimportant 39
Ian Watson Table 2.1
Averaged V.O. T. production values
Subject group
Voiced
(a) V.O.T. velar- English (values in ms.) Bilingual 6-year-olds 37.5 Bilingual 10-year-olds 25.5 Monolingual 6-year-olds 30.05 Monolingual 10-year-olds 29.9 Monolingual adults 23.5 (b) V.O.T. velar-French (values in Bilingual 6-year-olds Bilingual 10-year-olds Monolingual 6-year-olds Monolingual 10-year-olds Monolingual adults
ms.) *29.9 *20.0 *33.1 *22.1 *10.5
Voiceless 105.4 103.2 84.1 94.0 62.0 68.7 65.2 58.2 60.9 28.5
Note: * before a figure indicates that the majority of tokens in this category had at least some prevoicing. However, there was also considerable aspiration in some tokens. The figure given here takes no account of prevoicing.
(Simon & Fourcin, 1978). The bilinguals in this study resembled the monolingual English subjects in their English speech, having a clear, statistically significant distinction for this feature. But they also showed such a distinction in their French productions, albeit a smaller one than in English. None of the French monolingual groups had any statistically significant difference in this feature, although there was a slight tendency in the same direction. For both of these parameters, the bilinguals' productions in French are systematically different from those of their monolingual peers, while in English their V.O.T.s are longer for the voiceless series. Nevertheless, given the intermittent tendency of the monolinguals to produce these same features, it is not the case that individual utterances by bilinguals would be distinguishable from monolinguals' utterances. This reflects the impressionistic judgments of both the children's own teachers and native speakers to whom the recordings have been played that there is nothing obvious which distinguishes the bilinguals' speech from that of monolinguals in either language. It is thus possible for bilinguals to use different production routines from monolinguals in their realization of the phonemes of their two languages, without this being perceptible to other native speakers. Just as hearers are not aware that children in both languages habitually produce longer V.O.T.s than adults, the differences between monolinguals and bilinguals go unnoticed. A compromise seems to be reached by the bilingual between the two needs to sound sufficiently like a native speaker to conform to two different language communities and to reduce the pro40
Phonological processing in two languages cessing load of having to master two different phonetic repertoires. Without matching exactly the norms of monolinguals of his or her age group, he or she finds a way of remaining within the limits of acceptability. Perceptual studies In the light of the infant studies, there is a surprising dearth of studies of perception by bilingual children. This may be a disguised blessing, as there is strong evidence that many of the results of perceptual tests carried out on monolinguals and on adults are misleading, because the stimuli used were insufficiently natural to evoke responses typical of normal speech behavior. Investigations of the voicing contrast are prime offenders (Watson, 1983; Williams, 1977). A wide range of acoustic cues can be identified with voicing perception (Lisker, 1978; Watson, 1983) and there is evidence that different languages, and even different subjects, attach different importance to different cues (see, e.g., Bailey, Summerfield, & Dorman, 1977); furthermore, there are interactions between different cues. If a set of test stimuli lack a feature which is highly salient for a particular group or subject, their responses may be difficult to assess. Nevertheless, such tests are potentially valuable as a source of knowledge about differences in monolingual and bilingual perceptual processing. Those studies which have been carried out indicate that there are more substantial differences in this aspect of speech behavior than in production. They also, therefore, point to considerable asymmetry in bilinguals' perceptual and production routines. As part of an ongoing study by the author, an experiment was carried out whose object was to compare the perceptual development with respect to English of a group of English-French bilinguals to that of the monolingual group tested by Simon and Fourcin (1978). The stimuli were identical to those used by Simon and Fourcin. Subjects were groups of primary balanced bilinguals who were living in England, but being educated in French, and those also experienced some French at home. There were four age groups; five, six, eight, and ten years. Simon and Fourcin's results had shown that by the age of five, English monolinguals already showed sharply categorical labeling of the V.O.T. continuum. The five-year-old bilinguals, however, showed quasi-random labeling, with a tendency to judge as voiced stimuli with very long V.O.T.s. By the age of six, they had almost caught up with the monolinguals, although their labeling was not quite as categorical as the latter group's. There was no disparity by the age of eight. A second aspect of the test compared sensitivity to a second acoustic feature, the Fl transition discussed above. The results for the five-year-olds were again chaotic, but the six-year-olds were clearly sensitive to the presence or absence of the cue. In its absence, the category boundary 41
Ian Watson
V.O.T.(ms.) Figure 2.1 English six-year-olds' percentage /g/ responses. The percentage of /g/ responses obtained from stimuli with (natural) and without (flat) a rising transition following the release burst
is lowered from about 26ms. to 20ms. (figure 2.1), a disparity which increases in the eight- and ten-year-old groups to 26ms. vs. 15 ms. There is thus a relative delay in the attainment of a stage of perceptual development by the bilinguals, but not one which has long-lasting consequences. There is no evidence that their perceptual processing strategies are any different from those of monolinguals. Obviously, an investigation of their perception in French, which is under way, may further elucidate this result. This finding runs counter to those of two studies of adult bilingual perception, which, however, pertain to bilinguals in quite different social situations. Carramazza et al. (1973) compared Canadian French-English bilingual to monolingual speakers of both languages. The bilinguals were all native speakers of French, but had learnt English from, at the latest, their seventh birthdays. The responses of the two monolingual groups showed the classic pattern found in such tests: quite distinct crossover points from voiced to voiceless categorization (figure 2.2), although the monolingual French group showed a less sharp crossover than the English. The bilinguals, however, showed virtually no cross-language differentiation in their response patterns. They were more categorical in their responses than the French monolinguals, but less than the English, and the point on the V.O.T. continuum where they crossed from predominantly voiced to voiceless judgments was intermediate between those shown by the two mono42
Phonological processing in two languages 120 100
c S
X X X X X X
H—h
Voiced
80H
Voiceless
60-
Discrimination
40Discrimination/ identification overlap
200
1 ) 1 ) 1
X K K
-20 0
20
40
60
80
V.O.T.(ms.)
Figure 2.2 Classic discrimination/identification function. The classic pattern of matched identification and discrimination typically found in V.O.T. categorization tests. The change from voiced judgments (line marked with vertical dashes) to voiceless (line marked with crosses) is very sharp and rapid, and corresponds to a peak in the discriminability of tokens which are adjacent on the continuum (unmarked line)
lingual groups. This in no way reflected their productions in either language, which corresponded closely to the monolingual norms. Similarly, Williams (1977) found that a group of eight Spanish-English secondary bilinguals, screened to be well balanced, kept their two languages quite distinct in producing voicing contrasts, but responded to a V.O.T. categorization (perception) task similarly in both language conditions. There was considerable intersubject variation in where the boundary was situated, with five subjects placing nearer the typical English boundary, and three near the Spanish boundary. Williams attributes these results to three phenomena. One is the composition of his stimuli, which included perceptually salient cues other than V.O.T. for English, but not for Spanish. The second is the hypothesis that the bilingual may make more use of these other cues than the monolingual, exposure to two different systems focusing attention on those features which are most distinctive in each language. Obviously, a failure to categorize a V.O.T. continuum differently in different conditions in the laboratory does not equate with a perceptual deficit in a real language situation. Finally, he indicated that there may genuinely be more variability in the perceptual use of acoustic cues by bilinguals. There is, then, ample evidence that in perception, the experience of 43
Ian Watson a bilingual upbringing can lead to the use of rather different strategies from those used by monolinguals, although the similarity in acquisition pattern between bilingual children and monolinguals in English demonstrates that this is not necessarily so. The disparity may be due to a difference in processing patterns between those who are already bilingual at the time of their first phonological development, and those who are consecutive bilinguals. However, far more detailed study of the responses of different types of bilinguals to different acoustic cues is needed before clear conclusions can be drawn. Considerable attention should obviously also be paid to individuals' results, in the light of Williams' identification of considerable variability in responses. In both production and perception, therefore, studies of older children (and adults) suggest that bilinguals behave in ways which are at once distinct from monolinguals and very similar to them. Bilinguals modify the same variables in the production of both their languages as monolinguals, within approximately the same limits, but within these limits the details of their use of these variables may differ. Their perceptual use of acoustic cues to phonemic identity may differ, although it may be less that they use different cues, than that they attribute different weightings to them. There does seem to be a disparity between the production and perception habits of bilinguals which is not found in monolinguals; this is an area where further research is urgently needed. Conclusions
Differentiation, avoidance of interference, learning to categorize acoustic input in two contrasting ways are all tasks which must be carried out by the bilingual and which the monolingual escapes. Obviously, all of these assume a degree of mental processing, involving strategies which will remain undeveloped in those exposed to only one language. The nature of these strategies may vary from individual to individual, and according to the type of bilingualism. However, having to adopt different strategies during phonological development does not necessarily entail the use of different processes once this development is complete. It would be quite possible, despite differences in the acquisition process, for a perfectly balanced bilingual to have two completely independent phonological and phonetic systems, each identical in all ways to those of monolinguals. On the other hand, there could be a degree of integration between the two systems. Or, as a third possibility, the bilingual may have two systems, but which differ in some way from those of monolinguals. The studies reviewed and described here support this third possibility, at least with respect to production. There is undoubtedly an early stage 44
Phonological processing in two languages when the two systems are mixed, but this is due to the undifferentiated state of the two languages as a whole. After differentiation, the children described in the literature clearly develop two sound systems, but there is variation in the correspondence of these systems to those of monolinguals, especially in their phonetic detail. At the phonemic level, even children who are dominant in one language seem to achieve a near-perfect match with monolinguals in both languages. In their productions they may, however, retain noticeable traces of interference in the realizations of particular phonemes, that is at the phonetic level. Such interference is not universal. All of the perceptual studies reported above had subjects who in their productions were, at least for the variables tested, impressionistically indistinguishable from monolinguals. Their productions were not, however, identical when measured instrumentally. In the case of the English-French subjects tested by Watson (in press) there was a tendency to produce voiceless stops in a similar way across the two languages. This might be indicative of an economy in the use of phonetic, articulatory routines, in a case where such an economy does not threaten either the satisfactory realization of the two different systems or the identification of the bilingual with either of the language communities. More widespread differences were found in perceptual processing. One (unlikely) interpretation of these would be that even subjects whose systems are separated with respect to production have a common system for perception. Such a situation would be likely to prejudice overt language behavior. It is more likely that these subjects do have two different sets of perceptual routines, but that these were not revealed in the experiments. Because the tests carried out have taken as their starting point features known to be salient for monolinguals, they have revealed when bilinguals do not use such features. They have, inevitably, been less effective at revealing what other features, less salient for monolinguals, they use instead. It is also possible that if more widespread experiments are performed, a greater degree of variation will be found in the perceptual processing of monolinguals than is currently recognized. As with production, we may find that bilinguals rarely use strategies monolinguals never use, but they do use them with different frequencies. The number of unanswered questions in the area of bilingual phonology is still enormously high. With the massive increase in bilingual studies over recent years, and the increasing availability of phonetic instrumentation, there is no excuse for these questions not to be tackled, especially at a time when interest in speech processing but almost exclusively monolingual speech processing - has never been higher. It is high time to invite Cinderella to the ball. 45
Ian Watson References Anderson, J. 1981. Considerations on phonological assessment. In J. G. Erickson & D. R. Omark (eds.), Communicative assessment of the bilingual bicultural child. Baltimore: University Park Press. Bailey, P. J., Q. Summerfield, & M. Dorman 1977. On the identification of sinewave analogues of certain speech sounds. Has kins Laboratory Status Report, 51152, 1-25.
Barton, D. P. 1976. The role of perception in the acquisition of speech. Unpublished doctoral thesis, University of London. Best, C. T. 1985. Discovering messages in the medium: speech perception and the prelinguistic infant. Haskins Laboratory Status Report, 81, 139-180. Bond, Z. S., J. E. Eddey, & J. J. Bermejo 1980. VOT del Espanol to English: comparison of a language-disordered and normal child. Journal of Phonetics, 8,287-291. Burnham, D. K. 1986. Developmental loss of speech perception: exposure to and experience with afirstlanguage. Applied Psycholinguistics, 7, 207-240. Carramazza, A., G. H. Yeni-Komshian, E. G. Zurif, & E. Carbone 1973. The acquisition of a new phonological contrast: the case of stop consonants in French-English bilinguals. Journal of the Acoustical Society of America, 54, 421-428. Celce-Murcia, M. 1978. The simultaneous acquisition of English and French in a two-year old child. In E. Hatch (ed.), Second language acquisition. Rowley, MA: Newbury House. Chomsky, N. & M. Halle 1968. The sound pattern of English. New York: Harper and Row. Eilers, R. E., W. Gavin, & W. R. Wilson 1979. Linguistic experience and phonemic perception in infancy: a cross linguistic study. Child Development, 50, 14-18. Eilers, R. E., W. R. Wilson, & J. M. Moore 1979. Speech discrimination in the language-innocent and the language-wise: a study in the perception of voice onset time. Journal of Child Language, 6, 1-18. Elliot, A. J. 1981. Child language. Cambridge: Cambridge University Press. Fantini, A. E. 1985. Language acquisition of a bilingual child: a sociolinguistic perspective (to age ten). Clevedon: Multilingual Matters. Ferguson, C. & C. B. Farwell 1975. Words and sounds in early language acquisition. Language, 57, 419-439. Fourcin, A. J. 1978. Acoustic patterns and speech acquisition. In N. Waterson & P. Snow (eds.), The development of communication. London: Wiley. Goudailler, J.-P. 1981. Exemple de traitement de Topposition de "sonorite" par les enfants du cours preparatoire - utilisation de la methode electroglottographique. Montreal: 12emes Journees d'Etudes sur la Parole. 1983. Diverses possibilites de materialisation du trait de voisement: etude electroglottographique des occlusives d'enfants ages de 7-8 ans d'un Cours Elementaire. lleme I.C.A., Paris, 267-270. Imedadze, N. V. 1967. On the psychological nature of child speech formation under conditions of exposure to two languages. International Journal of Psychology, 2, 129-132. 46
Phonological processing in two languages Jakobson, R. 1977. Child language: aphasia and phonological universals. The Hague: Mouton. Kohler, K. 1977. Einfuhrung in der Phonetik des Deutschen. Berlin: Erich Schmidt. Konefal, J. A. & J. Fokes 1981. Voice onset time: The development of Spanish/ English distinction in normal and language disordered children. Journal of Phonetics, 9, 443-444. Kuhl, P. K. & J. Miller, 1978. Speech perception by the chinchilla: identification functions for synthetic VOT stimuli. Journal of the Acoustical Society of America, 63, 905-917. Leopold, W. P. 1939-1949. Speech development of a bilingual child: a linguist's child (4 vols.). Evanston, IL: Northwestern University Press. Linell, P. 1979. Psychological reality in phonology: a theoretical study. Cambridge: Cambridge University Press. Lisker, L. 1978. Rabid vs. rapid: a catalogue of acoustic features that may cue the distinction. Haskins Laboratory Status Report, 54, 127-132. Locke, J. 1986. Speech perception and the emergent lexicon: an ethological approach. In P. Fletcher & M. Garman (eds.), Language acquisition. Cambridge: Cambridge University Press, MacKain, K. S. 1982. Assessing the role of experience on infants' speech discrimination. Journal of Child Language, 7, 527-542. Mann, V. A. 1985. Phonological awareness: nature and nurture both contribute. Annual Bulletin, Research Institute of Logopedics and Phoniatrics, Faculty of Medicine, University of Tokyo, 19, 95-110. Menyuk, P. & L. Menn 1979. Early strategies for the perception and production of words and sounds. In P. Fletcher & M. Garman (ed.), Language acquisition. Cambridge: Cambridge University Press. Nolan, F. J. 1982. The nature of phonetic representations. Cambridge Papers in Phonetics and Experimental Linguistics, 1. Cambridge University: Department of Linguistics. 1989. The descriptive role of segments: evidence from assimilation. Paper presented at the second conference on laboratory phonology, Edinburgh. Ohde, R. N. 1985. Fundamental frequency correlates of stop consonant voicing and vowel quality in the speech of preadolescent children. Journal of the Acoustical Society of America, 78, 1554-1561. Pavlovich, M. 1920. Le langage enfantin: acquisition du serbe et du frangais par un enfant serbe. Paris: Champion. Ronjat, J. 1913. Le developpement du langage observe chez un enfant bilingue. Paris: Champion. Simon, C. & A. J. Fourcin 1978. Cross-language study of speech pattern learning. Journal of the Acoustical Society of America, 63, 925-935. Smith, N. V. 1973. The acquisition of phonology: a case study. Cambridge: Cambridge University Press. Streeter, L. A. 1976. Language perception of 2 month old infants shows effects of both innate mechanisms and experience. Nature, 259, 39-41. Watson, I. M. C. 1983. Cues to the voicing contrast: a survey. Cambridge Papers 47
Ian Watson in Phonetics and Experimental Linguistics, 2. Cambridge University: Department of Linguistics. 1985. Direct perception of speech. Cambridge Papers in Phonetics and Experimental Linguistics, 4. Cambridge University: Department of Linguistics. In press. Acquiring the voicing contrast in French. A comparative study of monolingual and bilingual children. In W. Ayres-Bennett, W. Green, & J. N. Green (eds.), Variation and change in French. London: Routledge. Weinreich, U. 1953. Languages in contact. The Hague: Mouton. Williams, L. 1977. The perception of stop-consonant voicing by Spanish-English bilinguals. Perception and Psychophysics, 21, 287-297. Zlatin, M. & R. Koenigsknecht 1976. Development of the voicing contrast. A comparison of voice onset time values in stop perception and production. Journal of Speech and Hearing Research ,19, 92-111.
48
3
Second-language learning in children: a model of language learning in social context LILY WONG FILLMORE
In this chapter I discuss a view of language learning that has evolved from a decade and a half long period of research on the learning of English by young children for whom English is a second language. My goals in a series of longitudinal studies have been to understand the nature of the process, and to discover what causes the variation that exist among learners in respect to the ease with which they learn the new language. I have been particularly interested in the interplay of cognitive and social factors in the learners, with situational and linguistic factors in the settings in which learning takes place. In all, I have studied the development of English in well over 200 five- to eleven-year-old children, some of them for as long as three years. In this chapter, I discuss the model of language learning that emerged from that decade and a half long program of research. The word 'emerge' is used advisedly. The model was not an a priori view of language learning that led to the sequence of studies mentioned above: it evolved post hoc as I attempted to reconcile the sometimes contradictory findings from the investigations of individual factors in language learning. The first set of such factors examined were learner characteristics that earlier research (Gardner & Lambert, 1972; Schumann, 1978; Swain & Burnaby, 1976; Wong Fillmore, 1976, 1979) suggested were key contributors to variation in second-language learning. Among them were those aspects of personality and social style that can affect the learner's ability and inclination to interact with people who speak the target language and who can help with its learning: for example, sociability, communicative need, risk taking, and selfconfidence. Other learner characteristics examined were ones that were likely to affect the cognitive aspects of language learning: inductive reasoning, verbal memory, and pattern recognition - aspects of cognitive functioning that could affect the ease with which learners recalled and made sense of the language they heard people using, and detected the regularities that existed within such samples of speech. The main hypotheses tested in that research were that each of such social and cognitive characteristics contributed to 49
Lily Wong Fillmore variation in language learning, and that collectively they were predictive of speed and accuracy in language learning. The results from this line of research have been mixed, as they have been for other researchers (Beebe, 1983; Bialystok & Frohlich, 1977; Naiman et al., 1978; Rossier, 1976; Strong, 1982). The characteristics that are supposed to facilitate language learning have been found to be positively associated with rapid second-language development in some situations, but not in others (Wong Fillmore, 1983, 1989a). For example, children who score high in measures of sociability and communicative need are generally good learners, but only in social settings where they are outnumbered by speakers of the target language, and where speakers and learners can interact freely with one another. In such settings, the extent to which children interact with one another depends on two factors. The first - the learner's social style and communicative needs - affects the learner's ability to establish and maintain contact with speakers of the target language. The second relates to the availability of speakers of the target language, and their willingness and ability to interact with the learner in ways that promote language development. In such situations, then, the learner's social skills and styles proved to be important variables. Social characteristics such as sociability and communicative needs are far less useful as predictors of language-learning outcome in settings where there are many more learners than there are speakers of the target language, or where there is little freedom for the two groups to interact. In fact, it appears that if either condition obtains, social characteristics are not particularly useful as predictors. In one of the classrooms we studied, 40 per cent of the children who entered school at the beginning of the year had learned no English by the end of it (Wong Fillmore, 1982). Some of the children who did not learn English that year were ones we predicted would have difficulty. They were shy, reclusive, laconic, or socially inept. The others were quite the opposite. They were among the most socially skilled, outgoing and verbal children in the class. The children were free to interact with classmates, but because most of the children in the class were non-English speakers, they spent their time interacting with one another in their native languages or in learner varieties of English. The more sociable the child the greater the time spent interacting with peers, it appeared. The teacher and her assistant spoke English exclusively and the children might have learned English from them, but they did not because of the way the teachers structured their class. The class was one in which most of the instruction activities were "child-selected" and directed. The teacher and her assistant created a rich environment where there were many activities to select from, and freedom for the children to interact with one another while they worked. There were few instances of teacher-directed instructional activities, or 50
Second-language learning in children formal instruction over the course of the year. The teachers monitored the children's activities closely, and they discussed their work with them and instructed them largely on an individual basis. These interactions provided the children opportunities to hear and learn English, but they were not frequent enough for most of the learners to sustain their languagelearning efforts. In another classroom we studied, one that was structured quite differently but with essentially the same student composition, sociability and communicative need were less predictive of language-learning outcomes than were characteristics such as verbal memory and pattern recognition. This class was tightly structured by the teachers. Virtually every activity was teacher directed and controlled. The children had few opportunities to interact with one another in instructed activities, so their social inclinations and skills played relatively little part in getting access to speakers of English. They heard and practiced English during formal instructional activities, where learner characteristics such as attentiveness, verbal memory, and pattern recognition were more important than the social ones being discussed. Thus, it would appear that if learner characteristics are predictive of variation in second-language learning, they are not predictive in any simple or straightforward manner. This line of investigation did, however, reveal additional sources of variation to be examined, and ways to explain why the findings on learner variables have been as mixed as they are. The variation we find across learners in acquiring second languages simply cannot be accounted for by differences in learners alone. Other sources of variation can be found in the settings in which languages are learned as well as in the speakers who are providing access to the target language and assistance in learning it. Since then, I have investigated the extent to which variation in language learning stems from differences in target-language speakers and in the settings which bring learners and speakers together (Wong Fillmore et ai, 1985). The findings of these investigations dovetail nicely with studies that other second-language researchers have conducted in Canada and elsewhere (especially Cummins & Swain, 1986; Ellis, 1984; Harley etal., 1987; Lightbown, 1983; van Lier, 1988), and from this collective work, the following model of second-language learning has emerged. A model of language learning in social context The model is a complex social one, and is difficult to talk about in purely abstract terms. It is easier to show how its components fit together when they are discussed in relation to situations that are familiar to all of us. Typically, second-language learners in societies like the United States or Canada are members of immigrant families. The family speaks a language 51
Lily Wong Fillmore other than English, and continues to use it in the home and in the immigrant community. At the same time, however, the members of the family know they will have to learn English in order to get along in their adopted society. English is the key to getting acquainted with the people who live there, to getting an education, and most importantly, to getting jobs. They have a genuine need to learn the language, and are motivated to do so because it offers them access to the social and economic life of the community they are joining. Nonetheless, learning this new language is an enormously complex task. Each individual in the family will have to discover what the speakers of the target language know that allows them to speak and use it for social and communicative purposes. The language learner's task consists of figuring out and learning the full system of linguistic, social, and pragmatic rules that govern the language behavior of the speech community. Despite the complexity of the task most individuals can handle it. People, irrespective of background, do not come to the language learning task empty handed: in their efforts to figure out how the new language is used by its speakers, they are guided by prior social, linguistic, and general world knowledge. As members of one language community, they know what sorts of things people talk about, and they know in general what kinds of verbal behaviors are appropriate for a variety of social situations and settings. They have some understanding of what they will have to learn to say in English. In learning the new language, they will be looking for ways to express the thoughts and sentiments they need to express, and to achieve the communicative goals they customarily accomplish in their primary speech community. The immigrant family is in an ideal situation for language learning since it resides in a setting that provides generous exposure to the language in use. The members of the family can hear and learn the language in the workplace, classroom, neighborhood, and playground - wherever they come into contact with people who speak the target language well enough to help them learn it. Language learning requires the help and involvement of people who already speak it: their speech behavior allows the learners to figure out how the language works, provided they are mindful and considerate of the learners' linguistic limitations. The model's components The language-learning situation I have outlined contains the necessary ingredients for second-language learning. There are three major components: (1) learners who realize that they need to learn the target language (TL) and are motivated to do so; (2) speakers of the target language who know it well enough to provide the learners with access to the language and the help they need for learning it; and (3) a social setting which brings 52
Second-language learning in children learners and TL speakers into frequent enough contact to make language learning possible. All three components are necessary. If any of them is dysfunctional, language learning will be difficult, or even impossible. When all three are ideal, language learning is assured. Each of them can vary in a great many ways, however, and some of this variation can critically affect the processes by which language is learned. These are the three essential components of the model. Three types of processes come into play in language learning, each of them intricately connected with the others. These can be described as (1) social, (2) linguistic, and (3) cognitive. I will discuss each in turn. Social processes. By social processes, I refer to the steps taken by learners and TL speakers to create a social setting in which communication by means of the target language is possible and desired. Social processes figure in language learning in the following way. Learners ordinarily require social contact with people who speak the TL to learn their language. These contacts give the learners both the incentive and the opportunity to learn the new language. They allow the learners to observe the language as it is used by TL speakers in natural communication, such observations providing the learners with the linguistic and social data that eventually allow them to figure out how the language is structured and used. In the course of these contacts, learners have to make the speakers aware of their special linguistic needs, and get them to make whatever accommodations and adjustments are necessary for successful communication - a difficult task. Communication with learners is never easy because it takes special thought and effort to make oneself understood, and to figure out what the learner is trying to say. Many people find the effort unrewarding, and try to avoid such contacts. If the TL speakers happen to know the learners' LI (first language), they may find it easier to communicate by means of the shared language. When that happens, the contacts between learners and TL speakers obviously do not provide any exposure to the new language. In interactions in which the TL is used, the learners have to participate at some level, since the quality of their participation plays a crucial role in getting speakers to use the language in the special ways that make the speech samples from these contacts usable as language-learning data. When TL speakers and learners interact, both sides have to co-operate in order for communication to take place. The learners make use of their social knowledge to figure out what people might be saying, given the social situation. The learners assume that the speech used by the speakers is relevant to the immediate situation; if the TL speakers are being cooperative, this will indeed be true. This is possible when the social settings in which learning takes place provide meaningful contacts between learners 53
Lily Wong Fillmore and speakers of the language. Those situations that promote frequent contacts are the best, especially if the contacts last long enough to give learners ample opportunity to observe people using the language for a variety of communicative purposes. Those which also permit learners to engage in the frequent use of the language with speakers are even better. Linguistic processes. By linguistic processes I refer to the ways in which assumptions held by the speakers of the target language cause them to speak as they do in talking to learners - in other words, to select, modify, and support the linguistic data they produce for the sake of the learner. On the learners' side there are assumptions about the way language works that cause them to interpret the linguistic data they have available to them. Linguistic processes figure in language acquisition in several crucial ways. The first intersects with the social processes described above, and in a sense, involves linguistic processes principally from the perspective of the speakers of the target language as they interact with learners. In essence, what learners have to get out of these contacts is enough linguistic evidence to allow them to discover how the language works and how people use it. The end product of the acquisition process is linguistic knowledge the phonological, lexical, grammatical, pragmatic, and sociolinguistic knowledge that eventually allows learners to speak and comprehend the new language in a full range of social and communicative situations. What it takes to acquire this kind of knowledge is exposure to linguistic data in the form of situationally anchored speech produced by speakers of the language in the context of social interaction that involves the learner in one way or another. These linguistic data, together with the supporting social context in which the data are anchored, constitute what researchers studying first- and second-language acquisition refer to as "input" - the materials on which learners can base their acquisition of the language. Language produced by speakers in social contacts with learners can serve as input when it has been produced with the learners' special needs in mind. It is not ordinary language, but language which has been selected for content, and modified in form and presentation. It tends to be structurally simpler, more redundant and repetitive, and is characterized by greater structural regularity than is found in ordinary usage (Gaies, 1979; Hatch, 1983; Long, 1981, 1983). Krashen (1982) has argued that for language to serve as input, it must be "comprehensible" to the learner - well enough anchored situationally for its meaning to be completely recoverable from the context of use. Others who have examined the nature of the language used in native-non-native speaker discourse have argued that what is even more important is that modifications occur as learners negotiate with TL speakers on the meaning of the discourse that takes place between them (Gass & Varonis, 1985; Pica & Doughty, 1985; Tarone, 1981). 54
Second-language learning in children Linguistic processes figure in the making of these adjustments in that people generally operate with some sort of theory of what their language is like, and they have beliefs about the kind of adjustments they ought to make when talking to anyone who is not fully proficient in their language - as when talking to babies, or to foreigners. The modifications that speakers make in talking to language learners are based partly on notions they have about what people who do not know the language well would find difficult to understand, and what they would find easy. Studies of the phenomenon of "foreigner-talk" (Clyne, 1968,1982; Ferguson, 1977) suggest that modifications made by speakers on the basis of a priori beliefs about the relative difficulty of linguistic forms are not always helpful to learners, and can, in fact, mislead them as to what the target forms are like. More useful accommodations are based on actual feedback provided by the learners as to whether or not they understand what is being said to them. When learners appear to understand, speakers can assume that the adjustments they are making are appropriate or even unnecessary. When they appear not to understand or to be having difficulty following what is said, then the speakers make adjustments in the form of what they are saying, or they do something else, verbally or otherwise, to allow the learners to figure out what is being communicated. In an important sense, then, it can be seen that learners and speakers collaborate in producing the adjustments which benefit the learners. This leads to the second way in which linguistic processes appear to figure in language acquisition, a way that intersects with cognitive processes. Looked at from the learners' perspective, the initial problem in language learning is to make sense of what people are saying in the new language. In part, this is achieved by paying close attention to what is happening while people are talking, and by assuming a relationship between speech and the events in which it occurs. The learner must guess what people might be saying given the social situation at hand. This might seem like an impossible task, but second-language learners have special resources to help them in this effort. They have a prior language and thereby the means to make educated guesses about what people are likely to talk about or say in a variety of situations. Because they already have a language, they know about linguistic categories such as lexical item, clause, and phrase. This awareness of grammatical form and structure will predispose them to look for equivalent properties in the new language data they have available to them. Similarly, through the experiences they have had in their first language, learners are generally knowledgeable about the speech acts and functions that can be performed linguistically. They know about the uses of declarative and interrogative structures, about affirmation and negation, about expressions of certainty and uncertainty in speech, and the like. They have used and are therefore familiar with such 55
Lily Wong Fillmore forms for making requests, promises, denials, declarations, and questions. They know that one can ask questions, and that questions ordinarily require answers. They know that questions can serve as requests for information, as indirect requests for action, as greetings, and for a host of other communicative functions. This kind of prior linguistic knowledge and experience will lead secondlanguage learners to seek and to discover the means for accomplishing the same functions in the new language. In other words, they are guided in their language-learning efforts by what they know to be possible and useful from their knowledge of the first language. Thus, second-language learners start out with a fairly good idea of what to look for in the new language. The assumption that forms will be found in the L2 which are functionally equivalent to LI forms can lead learners to acquire them more efficiently than they might otherwise. At the same time, however, it can also interfere with learning, since this assumption sometimes leads learners to draw largely unwarranted conclusions that L2 forms are functionally and structurally identical to LI forms and usages. When that happens, wefindfirst-languageinterference or transfer. Overall, however, the result of this process is positive. By applying the knowledge they have of what people are likely to say in various social situations to what they know are possible forms, patterns, and functions in language, learners are more or less able to give meaningful interpretations to the language they hear, and thus, to discover eventually the principles that govern the structure and use of the language itself. Cognitive processes. The third type of process in language learning is that which can be described as cognitive. In a sense the cognitive processes in acquisition are the central ones since they involve the analytical procedures and operations that take place in the heads of learners and ultimately result in the acquisition of the language. Let us consider what the cognitive task involves. The primary linguistic data which learners have available to them as input for their analyses consist of speech samples produced by speakers of the target language during social contacts in which the learners are themselves participants, as I have argued. Hence what the learners have to work with are observations of the social situations in which the language was produced, and streams of vocal sounds produced by speakers according to complex and abstract systems of grammatical and social rules that systematically and symbolically link up sounds, meaning representations, and communicative intentions. What they must do with these data is discover the system of rules the speakers of the language are following, synthesize this knowledge into a grammar, and then make it their own by internalizing it. That in capsule form is what the cognitive task is for any language learner. 56
Second-language learning in children Figuring out how the speakers of the target language are using the noises they produce to represent meaning is the first step (Peters, 1983, 1985). This involves discovering the principles by which segments of the speech produced by TL speakers relate to events, ideas, experiences, objects, and the other things that people are known to talk about. Discovering how the speech serving as input segments, that is, figuring out where one thing ends and another begins, is critical to the procedure. Once the learners know what the units are, they can figure out how they are used to represent meanings. They eventually discover how such units can be assembled structurally to communicate more complex ideas and thoughts in the target language. Finally, the cognitive task involves figuring out the principles by which the speakers of the language use it to achieve their communicative goals and intentions: what do the speakers of the language talk about, and what can they do with the language they speak? Learners apply a host of cognitive strategies and skills to deal with the task at hand: they have to make use of associative skills, memory, social knowledge, and inferential skills in trying to figure out what people are talking about. They use whatever analytical skills they have to figure out relationships between forms, functions, and meanings. They have to make use of memory, pattern recognition, induction, categorization, generalization, inference, and the like to figure out the structural principles by which the forms of the language can be combined, and meanings modified by changes and deletions. Such cognitive tools can be described as general cognitive mechanisms. Some language-learning theorists would object to this claim. According to the prevailing linguistic theory, a special cognitive mechanism is involved in language acquisition (Chomsky, 1965, 1975). This mechanism, which is referred to as the "language-acquisition device," or L.A.D., is said to operate in a quite different way from ordinary cognitive processes. Its workings cannot be observed; they can only be inferred from the fact that all ordinary children learn a first language, and that they appear to do it in ways that cannot be explained by ordinary cognitive processes. One of the major arguments for the cognitive processes involved in acquisition being special ones is that many of the features of the grammar that learners eventually acquire cannot simply be induced from the linguistic data that are available to them. In fact, the argument goes, it would be impossible to explain how children can arrive at structure as complex and subtle as found in a competence grammar, based on the relatively meager structural evidence they are able to extract from the language spoken around them, unless we assume that a certain amount of that structure is already available to the child's innate language-learning mechanism. By this view of acquisition, the processes that have been described here as social and linguistic are regarded as incidental or peripheral phenomena. If they figure at all, 57
Lily Wong Fillmore they play only trivial roles; everything that is really important in language learning has to do with the working of L.A.D. Adherents to the L.A.D. proposal have argued that no matter what other kinds of information or help are available to learners, the primary data they have to work with are samples of speech consisting of phonological signals which are not cognitively penetrable - they are not tractable to ordinary cognitive manipulations or analytical procedures that are available to children. The only explanation to people who hold this view is that such rules are already 'known' to the acquisition device in some abstract sense, requiring only exposure to data in which such rules figure, to trigger their discovery. That may be the case in first-language learning. Nearly everyone does in fact end up learning a first language, despite considerable differences in general intellectual endowment and early language experiences, and no matter how difficult or complex the target language. It does not seem to be true in the case of languages after the first, where great differences can be found in language-learning ability across individuals. What I believe to be the case is this: language learning involves two kinds of cognitive processes, both those which are specialized for language learning (i.e., of the L.A.D. type), and those which are involved in more general intellectual functioning. In first-language learning, mechanisms of the first type figure more heavily than do the second. In fact, the cognitive skills of the second type are just developing while children are acquiring their first languages. By the time individuals are likely to find themselves learning a second language, however, such general cognitive skills are well developed. They figure in whatever cognitive tasks the individuals encounter, including the ones involved in learning a new language. The L.A.D.-type mechanisms that are said to figure so heavily in first-language learning may also figure in second-language learning. In fact, I would argue that both types of cognitive mechanisms are involved in language learning, whether the learner is dealing with a first or a second language. The degree of involvement of these two types of mechanisms is reversed, however, for second-language learning. While specialized language-learning mechanisms figure in an important way too, general cognitive mechanisms are more heavily involved. This, in fact, may be a crucial difference between first- and secondlanguage learning. There are two kinds of evidence for believing that general cognitive abilities and strategies figure more heavily in the acquisition of second languages than specialized abilities do. The first consists of observations of strategies that learners appear to follow when they tackle a second language (Bialystok, 1983,1984; Faerch & Kasper, 1980,1983; Klein, 1986; Klein & Dittmar, 1979; Meisel, 1983; Selinker, 1972; Wong Fillmore, 1976). The second relates to observations of individual variation in the learning 58
Second-language learning in children of second languages (Gardner & Lambert, 1972; Naiman etal., 1978; Wong Fillmore, 1979, 1983). The kind of cognitive strategies and skills that I have described in this chapter - namely, the ones we find learners applying in getting access to the language, in breaking it down into units, in figuring out its structural properties, and in extracting its principles of usage - make use of general cognitive mechanisms rather than specialized ones. The cognitive work learners engage in results in them figuring out and acquiring a lot of rules, principles, and patterns, etc. But such materials do not necessarily add up to a grammar. At some point, the knowledge which has been gained through the workings of general cognitive mechanisms must be consolidated and assembled, in a manner of speaking, into a competence grammar. This, I would argue, is where the language-specific cognitive mechanisms come into play; through these processes, what the learner has sorted out gets synthesized into a competence grammar, and perhaps a lot of the details of the grammar get refined here as well; that is the work of the L.A.D. or whatever one wishes to call the innate language-learning mechanism people seem to have. This last part, I admit, is speculative: there is no way of proving or disproving it. Variation in components These, then, are the processes that figure in language learning. We can now consider how they work, or do not work, in relation to variation in the components that have been identified: learners, speakers of the target language, and the social setting. Let us return to the example of the typical language learners in this society. The adults in the family typically come into contact with TL speakers in the workplace, in bureaucratic and service encounters, in informal social settings or in adult-education classrooms. Adult language classes are seldom ideal settings for language learning, however, since there is in most cases just one TL speaker (the teacher) to support the efforts of many language learners (the students). How well they work for language learning depends on the adequacy of the teacher's methods and materials, and the receptiveness of the students to formal language instruction. The kind of language the learners hear and practice in such settings can vary considerably in its richness and usefulness. Bureaucratic and service encounters can generally be dismissed as major opportunities for language learning. The interactions they invite between learners and TL speakers seldom last long enough, are repeated frequently enough, or provide enough incentive for the TL speakers to assist the learners to support language learning. Such encounters may motivate the learner to learn the language (in order to participate more successfully at future such 59
Lily Wong Fillmore encounters), but they do not offer the learner much support for learning the language. Far better for language learning are informal social encounters with TL speakers, and encounters with them in the workplace. In most cases, however, the language limitations of the immigrant adult preclude jobs or friendships that require much TL communication. In the United States, the jobs that adult newcomers are likely to get are ones where most of their co-workers are also non-English speakers, or ones in which they do not need to interact much with English speakers. Thus, a non-English speaker seeking a job in a restaurant is more likely to get a job as a dishwasher or busboy than as a waiter or bartender. Such jobs work against language learning in two ways: they neither require the adult to use much English, nor do they offer much support for learning the language. From the language learner's perspective, the best jobs are ones that put the learner in close proximity to TL-speaking co-workers, and which require co-operation between workers to do what has to be done. The TL speakers are then motivated to communicate with the learners and help them learn the language, and the learners are motivated to do their part as well. None of the settings that the adult members of the immigrant family find themselves in, and few of the TL speakers they encounter can provide or match the support that is available to the children, however. The children of immigrant families are in the best position to learn a new language. They come into contact with speakers of the TL in school which, in many respects, is the ideal social setting for language learning (Ellis, 1984; van Lier, 1988). It was noted that language learning is possible when learners are in frequent contact with speakers of the target language, and the two groups are motivated and able to interact with each other in some fashion. Typically, immigrant children find themselves in classrooms surrounded by English speakers. In such a setting the social conditions for language learning outlined earlier are easily enough met. The learners are in constant social contact with speakers of the target language. The speakers (the teachers especially, but classmates as well) have ample reason to interact with the learners in this setting, and they are generally aware of the learners' limitations in the language. In their efforts to communicate with learners, they tend not to depend on speech alone but to accompany their speech with actions, or to give access to meaning in other ways. What the learners must do is observe carefully what is happening while others speak, listen to what they are saying, figure out what they are talking about, and how they do it. They apply their prior social and linguistic knowledge in making sense of their observations, and eventually, by exercise of general cognitive strategies they discover what speakers know that allows them to communicate in the target language. This is easier for some learners than for others. Let us consider the question of variation in second-language learning, 60
Second-language learning in children since this is what leads to the argument that general rather than specialized cognitive processes are involved in second-language learning, and that social and linguistic processes play the roles I have described. Variation in learners One of the most striking differences between first- and second-language learning is in the relative amounts of individual variation that can be found among the two types of learners. Clearly, there is variation to be found even among first-language learners (Nelson, 1981; Peters, 1977, 1983). The differences, however, are not regarded as great. While children may vary somewhat in their rate of primary language development, the variation among first-language learners is minor compared to that found even among relatively young second-language learners. Differences of up to five years can be found in the time children take to get a working command of a new language. Learners differ considerably both in how easily and how completely they master the grammatical details and intricacies of a second language. Some are able to learn it as completely and as well as they did their first language; others never completely master the forms or uses of the language. A substantial portion of this variation may be due precisely to the involvement of the kind of cognitive mechanisms that have been identified here as the ones that figure most heavily in second-language learning. Individuals apparently do not vary in having an innate capacity to learn language. If this mechanism were as heavily involved in the learning of second languages as it is said to be in the learning of first languages, we would not expect to find any more variation in second-language learning than in first-language learning. But if, as I have argued, the kind of cognitive processes that are involved in second-language learning are the ones which relate to general cognitive abilities, the variation makes sense. We know that there are considerable differences across individuals in such abilities. The differences in ability that figure in language learning (verbal memory, auditory perception, pattern recognition, and so on) are not ones that figure heavily in general intelligence, although some of them obviously are related (for example, categorization, generalization, and association). The point is that much of the variability found in second-language learning can be traced to differences found among learners in the application of these general mechanisms and abilities that figure in language learning. Learners who have poor auditory memory will have a difficult time remembering the things they hear in a new language. If they cannot remember what they hear, they will not find it easy to figure out what it means or how it is structured. Those who are poor in auditory perception will have difficulty discriminating the sounds of the new language, and hence will have 61
Lily Wong Fillmore difficulty making sense of what they hear and reproducing it. Learners who are poor in pattern recognition will have difficulty finding the patterns they must eventually acquire in the new language. Variation in language learning along the cognitive dimension is not just a matter of differences in learner endowments in cognitive abilities, however. It is also affected by other learner variables. Age is an obvious one. Research has shown that older learners have an advantage over younger learners with certain aspects of second languages (Ervin-Tripp, 1974; Krashen, Long, & Scarcella, 1979; Snow & Hoefnagel-Hohle, 1978). It has been argued that older learners acquire new structures and forms more rapidly and easily because they have better-developed learning strategies and cognitive abilities. But the cognitive advantages that come with age and experience do not always result in better language learning for older learners. Social and communicative needs, which become increasingly complex with age, can interfere with the process. In learning a new language, it is necessary initially to suppress both social and communicative needs to a degree. Learners must interact with TL speakers in order to get needed exposure to the language and assistance in learning it. They do not share a common language, however, so this kind of interaction is not easy. Learners cannot say all they need or want to say, nor can they establish themselves socially with the TL speakers as they might if they spoke their language. What they need are ways to interact and talk without much actual communication at an informational level. There are many more opportunities for young children to interact in this manner than there are for older ones, and for children in general than for adults. Children can play together without much real talk, but there are few ways for older learners to interact with one another without it. I have observed kindergarten-age language learners working and playing with classmates hour after hour, unhampered by the fact that they do not understand one another. By the third and fourth grade, children at eight or nine years of age not only have fewer opportunities for this kind of interaction, but they are much more hamstrung by the inability to communicate easily with one another. Thus, it appears that one way in which young learners have an advantage over older ones is that their social circumstances make it relatively easier for them to interact with TL speakers in ways that support language learning. Personality is another type of variable that can interact with age differences to affect both the social and cognitive processes involved in language learning. Personal characteristics which may not hinder language learning in young children may become major problems later in life. As individuals mature, there is an increasing investment in their views of self and in the traits they regard as their own. One personality characteristic in particular whose influence increases 62
Second-language learning in children with age is mental rigidity, a trait which varies considerably from person to person. Learning a new language requires an individual to consider multiple possibilities, and to avoid making premature decisions about the meanings and uses of the forms they are learning. Learners who are rigid in their thinking find it difficult to deal with multiple possibilities, or with things they cannot immediately understand. Unfortunately, there is much of that to deal with in the learning of a new language. The unwillingness or inability to accommodate new information or the unknown can make it extremely difficult for learners to handle many of the aspects of the task that I have outlined. Another personality or cognitive-style characteristic that can affect language learning relates to risk taking (Beebe, 1983). Language learning requires learners to apply inferential skills in figuring out what people are saying in the language, and in discovering how the language they hear relates to the social situations in which it is used. Some learners find it hard to take the risks involved in acting upon information gained through guessing, and in fact may be unwilling to risk much guessing at all. They find it hard to try out whatever knowledge they have acquired of the new language by their observations because they are afraid of being wrong, or of appearing foolish. Or they find it difficult to take the next step in language learning, which is to draw generalizations from the relationships they discover, and to test them out. Whatever the problem, the cognitive processes that should be operating do not function as they should for some learners, and as a result, they do not learn as quickly or well as they might. Variation in settings and TL speakers Studies of variation in second-language learning offer evidence of ways in which TL speakers and the social settings affect the social and linguistic processes that figure in acquisition, too. Let us first consider ways in which characteristics of social settings affect the social processes involved in language learning. Social settings vary in the opportunities they offer for learning language; these are partly a function of their composition (that is, the ratio of TL speakers to language learners), and partly a function of their structure and purpose. As noted earlier, social settings work best for language learning when TL speakers outnumber learners, and when they are structured in ways that maximize interaction between the two groups. Some settings provide learners with few opportunities to get close enough to speakers of the language to do them any good, or the kinds of contacts they provide are inadequate for language-learning purposes. The contacts may be too brief, or too infrequent, or too limiting in the kind of exchanges they allow. It was noted that classrooms are perhaps the ideal setting for language 63
Lily Wong Fillmore learning. But even classrooms can differ widely in the quality and quantity of input they provide learners (Allen etaL, in press; Lagaretta, 1979; Wong Fillmore, 1982, 1989b; Wong Fillmore et al., 1985). If learners outnumber TL speakers, the learners may not have enough contact with speakers. If the TL speakers are themselves imperfect speakers of the language, the kind of input they provide for learners may not be an adequate representation of what the learners ought to be aiming at. This has been well documented in the immersion classrooms that have been studied both in Canada and in the United States (Campbell, 1984; Cummins & Swain, 1986; Lambert, 1984; Lambert & Tucker, 1972; Snow, Galvan, & Campbell, 1980; Swain, 1984, 1988; Swain & Lapkin, 1982). Classrooms for language learning can also differ in the extent to which they require the learners themselves to play a role in getting the social contact needed for language learning, and in the role interaction itself plays in the process (Ellis, 1984; van Lier, 1988). It has long been assumed that direct interaction between learners and speakers is necessary in order for language learning to take place. But in our classroom research, we have found that some learners can in fact pick up a language pretty much by observing their teachers and peers, despite having little direct interaction with them (Wong Fillmore, 1983; Wong Fillmore et al., 1985). The ability to do so depends on the structure of the classroom settings, the kind of language used in them, and the characteristics of the individual learners. Children who lack the inclination or skills to interact socially with TL speakers can nevertheless learn the language, provided the setting offers them access to meaningful input, and opportunities to practice the language in the context of structured instructional activities. It appears that what is essential is that learners have access to language that is appropriately modified for them, and is used in ways that allow learners to discover its formal and pragmatic properties. This is a function not only of social settings, but of the TL speakers in them too. There were among the classrooms I have studied ones in which the English used by teachers for group instruction worked so well as input that all of the language learners in the class profited from the experience, irrespective of whether they were inclined to interact with speakers of the target language (Wong Fillmore, 1982). The children who make the best use of such opportunities are ones who score relatively high in the learner characteristics that appear to influence the cognitive processes involved in language, are highly motivated, and attentive enough to learn by observation. When there are regular activities that both invite and support the use of the target language in the context of learning about subject matter that is made relevant and interesting to the children, they learn the language, with or without much additional informal social contact with speakers. But classrooms, as noted earlier, can vary considerably as settings for language learning. Differences in structure and in class composition 64
Second-language learning in children are two important ways in which they can affect language learning. Differences in TL speakers in those classes can as well. TL-speaker variables figure in the process by affecting the quality and quantity of support available to language learners. Individuals who play the role of "TL speaker" have to know the language well enough to provide reliable samples of it for the learners to work on. In classrooms (for example, Canadian and American immersion programs) where nearly everyone is a language learner, the speech the children hear most often is produced by learners like themselves: it represents an interlanguage rather than a mature form of the target language (Selinker, Swain, & Dumas, 1975). Whether or not children in such situations can learn a standard variety of the language depends on their getting enough help from the TL speakers in the setting who do know the language well enough to support their efforts. If their teachers, who presumably are proficient speakers of the TL, provide them with enough opportunities to hear and use the language, and if they provide them with adequate corrective feedback when they use learner forms of the language, they do. But as Swain and her colleagues have found in their studies of language learning in Canadian immersion classrooms, little of the necessary feedback is provided in some cases (Swain, 1988). Attitudes and beliefs held by the TL speakers can affect the role they need to play in language learning. The ones that most critically affect their willingness to interact with the learners and the manner in which they do so are the following: how they feel about interacting with the learners; what they regard as their responsibility to the learners; what they believe about their own ability to communicate with the learners and about the learners' ability to understand and to speak; and what they think would be helpful to the learners. Individuals differ considerably in such matters. TL speakers who believe that their language-learning interlocutors will not comprehend authentic instances of the language may engage in the use of inappropriate adjustments, with the result that the input that is available to the learner is "foreigner talk" (Clyne, 1968, 1982; Ferguson, 1977). Or the speaker may bejieve that adjustments are unnecessary; in which case, the speech they use in interacting with learners may be incomprehensible to the learners. Or, if the TL speaker believes that the learners are not able to understand the language at all, they can abandon efforts to communicate by use of the target language. If they share a common language with the learners, they use it. If they do not, they try to communicate by non-verbal means, or they give up trying. These TL-speaker variables affect communication and learning in all settings - even in classrooms where the TL speaker is the teacher. In classrooms where the language used by the teacher does not work well as input, or in more unstructured situations, as in "open classrooms" or 65
Lily Wong Fillmore on the playground, where students cannot count on getting the free input provided by teachers in instructional activities as described above, the interaction between learners and TL speakers is all-important, and learners play a much greater role in getting the input and support they need for language learning. In such situations, learner variables such as personality and social skills can play a substantial role in language learning. Learners who find it easy or desirable to interact with TL speakers get many more of the social contacts needed for language learning than do learners who are not as interested, or motivated, or who are less able to manage the kind of social contacts needed. Variables such as personality, social style, social competence, motivation, and attitudes in both learners and speakers of the target language can affect language learning, in fact. Conclusion
What I have tried to show in this chapter is how the parts of the model that has been described here figure in the process of language learning, and how variation in the three components can crucially affect the outcome of the process. The attempt to find simple or straightforward explanations both for variation and for language learning has not been particularly fruitful. The processes that figure in language learning can be understood only when all of the components of the model are considered. Similarly, variation in language learning cannot be explained or predicted without a consideration of the complex interaction among the many factors that are involved in language learning. The model that has been discussed here is just a proposal, however. I continue to work on it, and to test it on the language-learning data my colleagues and I have collected over the years.
References Allen, P., M. Swain, B. Harley, & J. Cummins 1990. Aspects of classroom treatment: toward a more comprehensive view of second language education. In B. Harley, P. Allen, J. Cummins, & M. Swain (eds.), The development of second language proficiency. New York: Cambridge University Press. Beebe, L. M. 1983. Risk-taking and the language learner. In H. Seliger & M. Long (eds.), Classroom oriented research in second language acquisition. Rowley, MA: Newbury House. Bialystok, E. 1983. Some factors in the selection and implementation of communication strategies. In C. Faerch & G. Kasper (eds.), Strategies in interlanguage communication. London: Longman. 1984. Strategies in interlanguage learning and performance. In A. Davies & 66
Second-language
learning in children
C. Criper (eds.), Interlanguage: proceedings of the seminar in honour of Pit Corder. Edinburgh: Edinburgh University Press. Bialystok, E. & M. Frohlich 1977. Aspects of second language learning in classroom settings. Working Papers in Bilingualism, 13, 2-26. Campbell, R. N. 1984. The immersion approach of foreign language teaching. In Studies on immersion education: a collection for United States educators. Sacramento, CA: California State Department of Education. Chomsky, N. 1965. Aspects of the theory of syntax. Cambridge, MA: M.I.T. Press. 1975. Reflections on language. New York: Pantheon Books. Clyne, M. 1968. Zum Pidgin-Deutsch der Gastarbeiter. Zeitschrift fur Mundartforschung 35,130-139. (ed.) 1982. Foreigner talk. International Journal of the Sociology of Language, 28. Cummins, J. & M. Swain, 1986. Bilingualism and education. Aspects of theory, research and policy. London: Longman. Ellis, R. 1984. Classroom second language development. Oxford: Pergamon Press. Ervin-Tripp, S. 1974. Is second language learning like the first? TESOL Quarterly, 8, 111-127. Faerch, C. & G. Kasper 1980. Processes in foreign language learning and communication. Interlanguage Studies Bulletin, 5,47-118. 1983. On identifying communication strategies. In C. Faerch & G. Kasper (eds.), Strategies in interlanguage communication. London: Longman. Ferguson, C. A. 1977. Simplified registers, broken language, and Gastarbeiterdeutsch. In C. Molony, H. Zobl, & W. Stolting (eds.), Deutsch im Kontakt mitanderen Sprachen. Kronberg: Scriptor. Gaies, S. 1979. Linguistic input in first and second language learning. In F. Eckman & A. Hastings (eds.), Studies in first and second language acquisition. Rowley, MA: Newbury House. Gardner, R. C. & W. C. Lambert 1972. Attitudes and motivation in second language learning. Rowley, MA: Newbury House. Gass, S. & E. M. Varonis, 1985. Task variation in native/nonnative negotiation of meaning. In S. Gass & C. Madden (eds.), Input in second language acquisition. Rowley, MA: Newbury House. Harley, B., P. Allen, J. Cummins, & M. Swain 1987. Final report. The development of bilingual proficiency study. Toronto: Ontario Institute for Studies in Education. Hatch, E. 1983. Simplified input and second language acquisition. In R. Andersen (ed.), Pidginization and creolization as language acquisition. Rowley, MA: Newbury House. Klein, W. 1986. Second language acquisition. Cambridge: Cambridge University Press. Klein, W. & N. Dittmar 1979. Developing grammars. Berlin: Springer. Krashen, S. 1982. Principles and practice in second language acquisition. Oxford: Pergamon. Krashen, S., M. Long, & R. Scarcella 1979. Age, rate and eventual attainment in second language acquisition. TESOL Quarterly, 13, 573-582. 67
Lily Wong Fillmore Lagaretta, D. 1979. The effects of program models on language acquisition by Spanish-speaking children. TESOL Quarterly, 13, 521-534. Lambert, W. E. 1984. Overview of issues in immersion education. In Studies on immersion education: a collection for United States educators. Sacramento, CA: California State Department of Education. Lambert, W. E. & G. R. Tucker 1972. Bilingual education of children: the St. Lambert experiment. Rowley, MA: Newbury House. Lightbown, P. 1983. Exploring relationships between developmental and instructional sequences in L2 acquisition. In H. Seliger & M. Long (eds.), Classroom oriented research in second language acquisition. Rowley, MA: Newbury House. Long, M. 1981. Input, interaction, and second language acquisition. In H. Winitz (ed.), Native language and foreign language acquisition. Annals of the New York Academy of Sciences, 379, 259-278. 1983. Linguistic and conversational adjustments to non-native speakers. Studies in Second Language Acquisition, 5,177-193. Meisel, J. 1983. Strategies of second language acquisition: more than one kind of simplification. In R. Andersen (ed.), Pidginization and creolization as language acquisition. Rowley, MA: Newbury House. Naiman, N., M. Frohlich, H. Stern & A. Todesco 1978. The good language learner. Research in Education (no.7). Toronto: Ontario Institute for Studies in Education. Nelson, K. 1981. Individual differences in language development: implications for development and language. Developmental Psychology, 17, 225-287. Peters, A. M. 1977. Language learning strategies: does the whole equal the sum of the parts? Language, 53, 560-573. 1983. The units of language acquisition. Cambridge: Cambridge University Press. 1985. Language segmentation: operating principles for the perception and analysis of language. In D.S. Slobin (ed.), The crosslinguistic study of language acquisition. Vol. II: Theoretical issues. Hillsdale, NJ: Erlbaum. Pica, T. & C. Doughty 1985. Input and interaction in the communicative language classroom: a comparison of teacher-fronted and group activities. In S. Gass & C. Madden (eds.), Input in second language acquisition. Rowley, MA: Newbury House. Rossier, P. 1976. Extroversion-introversion as a significant variable in the learning of oral English as a second language. Unpublished doctoral dissertation. University of Southern California. Schumann, J. 1978. Social and psychological processes in second language acquisition. In J. C. Richards (ed.), Understanding second and foreign language learning: issues and approaches. New York: Academic Press. Selinker, L. 1972. Interlanguage. International Review of Applied Linguistics, 10, 209-230. Selinker, L., M. Swain, & G. Dumas 1975. The interlanguage hypothesis extended to children. Language Learning 25,139-152. Snow, C. & M. Hoefnagel-Hohle 1978. Age differences in second language acqui68
Second-language learning in children sition. In E. Hatch (ed.), Second language acquisition: a book of readings. Rowley, MA: Newbury House. Snow, M. A., J. Galvan, & R. Campbell, 1980. The pilot class of the Culver City Spanish immersion program: a follow-up report, or whatever happened to the immersion class of 78? In K. Bailey, M. Long, & S. Peck (eds.), Second language acquisition studies. Rowley, MA: Newbury House. Strong, M. 1982. Social styles and the second language acquisition of Spanishspeaking kindergartners. Unpublished doctoral dissertation. University of California, Berkeley. Swain, M. 1984. A review of immersion education in Canada: research and evaluation studies. In Studies on immersion education: a collection for United States educators. Sacramento, CA: California State Department of Education. 1988. Manipulating and complementing content teaching to maximize second language learning. TESOL Canada Journal, 6, 68-83. Swain, M. & B. Burnaby 1976. Personality characteristics and second language learning in young children: a pilot study. Working Papers on Bilingualismy 11,76-90. Swain, M. & S. Lapkin 1982. Evaluating bilingual education: a Canadian case study Clevedon, Avon, England: Multilingual Matters. Tarone, E. 1981. Some thoughts on the notion of communication strategy. TESOL Quarterly, 15, 285-295. van Lier, L. 1988. The classroom and the language learner. London: Longman. Wong Fillmore, L. 1976. The second time around: cognitive and social strategies in second language acquisition. Unpublished doctoral dissertation. Stanford University. 1979. Individual differences in second language acquisition. In C.J. Fillmore, W. S. Y. Wang & D. K. Kempler (eds.), Individual differences in language ability and language behavior. New York: Academic Press. 1982. Instructional language as linguistic input. In L. C. Wilkinson (ed.), Communicating in the classroom.New York: Academic Press. 1983. The language learner as an individual. In M. Clarke & J. Handscombe (eds.), In TESOL '82. Pacific perspectives on language learning and teaching. Washington, DC: Teachers of English to Speakers of Other Languages. 1989a. Language learning in social context: the view from research in second language learning. In R. Dietrich & C. Graumann (eds.), Language processing in social context. Amsterdam: Elsevier Science Publishers. 1989b. Teachability and second language acquisition. In R. Schiefelbusch & M. Rice (eds.), The teachability of language. Baltimore, MD: Paul Brookes. Wong Fillmore, L., P. Ammon, B. McLaughlin, & M. S. Ammon 1985. Learning English through bilingual instruction. Final Report to National Institute of Education. Berkeley, CA: University of California.
69
4
Interdependence of first- and second-language proficiency in bilingual children JIM CUMMINS
When children begin the acquisition of a second language (L2), whether in the home or at school, their cognitive resources clearly play a central role in the rapidity and ultimate success with which that language is acquired. Other individual factors, such as motivation, and contextual factors that determine amount and type of exposure to the second language, are also central to the acquisition process and in most situations are likely to interact with cognitive factors. I argue in this chapter that the process of second-language acquisition can be clarified by distinguishing between two dimensions of proficiency that relate in specific ways to determinants of the acquisition process, namely, attribute-based and input-based aspects of proficiency. Attribute-based aspects of proficiency refer to those dimensions of proficiency whose acquisition is strongly influenced by relatively stable attributes of the individual learner, for example, cognitive and personality variables. Input-based aspects of proficiency, on the other hand, are considerably less related to stable attributes of the individual than they are to the quality and quantity of L2 input received from the environment. In the initial stages of acquiring the L2 the distinction between attribute- and input-based dimensions of proficiency may not be apparent but, over time, the distinctive influences of attributes and input will result in differentiation between these dimensions. The chapter focuses in particular on the role of one aspect of the cognitive resources that children bring to L2 acquisition, namely, their first-language (LI) proficiency. The purpose is to elucidate the extent to which different aspects of LI and L2 proficiency are related in the L2 acquisition process. An initial step in this analysis is to clarify the dimensions of proficiency in one language. Dimensions of language proficiency
A number of investigators have pointed to a distinction between contextualized and decontextualized language as fundamental to understanding the nature of children's language and literacy development. The terms used 70
First- and second-language proficiency by different investigators have varied but the essential distinction refers to the extent to which the meaning being communicated is supported by contextual cues (such as the paralinguistic cues present in face-to-face interaction) or dependent largely on linguistic cues that are independent of the immediate communicative context. Among the distinctions that have been made are Bruner's (1975) distinction between communicative and analytic competence, Olson's (1977) utterance and text, Donaldson's (1978) embedded and disembedded thought and language, Bereiter and Scardamalia's (1981) conversation and composition, Cummins' (1981, 1984) contextembedded and context-reduced language proficiency (recently labeled simply as conversational- vs. academic-language proficiency; e.g., Cummins, 1989), and Snow et al.'s (this volume) contextualized and decontextualized use of language. Snow's terminology will be adopted in the present chapter. The distinction between contextualized and decontextualized language skills is consistent with the research of Biber (1986), who used psychometric analysis of an extremely large corpus of spoken and written textual material in order to uncover the basic dimensions underlying textual variation. Three major dimensions were labeled by Biber as interactive vs edited text, abstract vs situated content, and reported vs immediate style. The first dimension is described as follows: Thus, Factor 1 identifies a dimension which characterizes text produced under conditions of high personal involvement and real-time constraints (marked by low explicitness in the expression of meaning, high subordination and interactive features) - as opposed to texts produced under conditions permitting considerable editing and high explicitness of lexical content, but little interaction or personal involvement. (Biber, 1986, p. 385) The second factor reflects a ''detached formal style vs. a concrete colloquial one" (p. 396). Although this factor is correlated with the first factor, it can be empirically distinguished from it, as illustrated by professional letters, which, according to Biber's analysis, illustrate highly abstract texts with a high level of personal involvement. The third dimension "distinguishes texts with a primary narrative emphasis, marked by considerable reference to a removed situation, from those with non-narrative emphases (descriptive, expository, or other) marked by little reference to a removed situation but a high occurrence of present tense forms" (Biber, 1986, p. 396). Although Biber's three dimensions provide a more detailed analysis of the nature of language proficiency and use than the contextualized/decontextualized distinction (as would be expected in view of the very extensive range of spoken and written texts analyzed), it is clear that the distinctions highlighted in his dimensions are consistent with those distinguishing contextualized and decontextualized language. The narrower range of language 71
Jim Cummins used and/or required by developing children makes the contextualized/ decontextualized distinction adequate for purposes of this chapter. There is considerable empirical evidence for this distinction in the sphere of second-language acquisition. Several investigators (Collier, 1987; Cummins, 1981; Cummins & Nakajima, 1987) have reported that at least four years is required even for socioeconomically advantaged immigrant students to attain grade norms in English academic skills. Peer-appropriate conversational L2 skills are usually attained within a considerably shorter period (on average about two years of exposure to the L2 in the L2 context) (Cummins, 1984; Gonzalez, 1986; Snow & Hoefnagel-Hohle, 1978). Both Snow etal. (this volume) and Gonzalez (1986) have reported that contextualized and decontextualized language skills are relatively independent of each other among bilingual students. In summary, the distinction between contextualized and decontextualized language use has been drawn by a considerable number of investigators and is supported empirically in a substantial number of studies. To what extent are each of these dimensions related across languages? In other words, to what extent is there evidence that students who manifest high levels of contextualized and decontextualized skills in their LI will develop similarly high levels of these skills in their L2? The relation between LI and L2 proficiency
Investigations of the relationships between LI and L2 contextualized and decontextualized language skills among immigrant students will be reviewed according to the background of the students involved. The three major categories are Finnish students in Sweden, Hispanic students in the United States, and Asian students in the United States and Canada. Then some additional evidence from bilingual programs involving both majority and minority students will be reviewed and finally we will examine studies involving adult L2 learners. Studies of immigrant students Finnish minority students in Sweden Two sets of studies involving Finnishbackground students in Sweden are relevant; the first set consists of studies carried out in the early 1970s synthesized by Skutnabb-Kangas and Toukomaa (1976), while the second consists of more recent studies reported by Linde and Lofgren (1988). In reviewing these studies our purpose is to assess the extent to which students' acquisition of Swedish (L2) academic skills is related to their Finnish (LI) proficiency. In some studies, data are also reported on the relationships between contextualized (conver72
First- and second-language proficiency sational) and decontextualized (academic) language skills and these data will also be reviewed. The Skutnabb-Kangas and Toukomaa report The studies synthesized by Skutnabb-Kangas and Toukomaa (1976) were carried out primarily in Olofstrom and Gothenburg and were designed to determine the level of Finnish immigrant students' academic achievement in both Finnish and Swedish and to explore some of the determinants of their achievement. A variety of tests in both languages were used; most assessed cognitive and academic abilities such as vocabulary knowledge, synonyms, antonyms, etc. as well as academic achievement in reading, math, and other school subjects. Skutnabb-Kangas and Toukomaa summarize the Olofstrom data as follows: those who attended school in Finland (prior to immigration) approached the level of achievement of normal Swedish pupils . . . in the written comprehension test considerably more often than those who began school in Sweden. Those who attended school in Finland for at least three years did best. The explanation for this can perhaps be found in their better skills in their mother tongue, which laid the basis for understanding a test written in Swedish. Two years in a Finnish class in Sweden did not, on the other hand, make for as good a basis for learning Swedish as the corresponding time in Finland. (1976, pp. 65-66)
Significant correlations were observed for both the Olofstrom and Gothenburg samples between Finnish and Swedish verbal academic proficiency. For example, among the Olofstrom sample, the partial correlations (with length of residence held constant) between Finnish and Swedish for grades three to six students (N= 165) ranged from 0.20 (p<0.01) to 0.41 (p<0.001) (table 8, p. 60). For the grades seven to nine sample, five out of six partial correlations of Finnish with Swedish verbal academic skills were significant. A similar pattern of significant L1-L2 partial correlations was found for the Gothenburg sample (tables 11 and 12, pp. 64 and 65). The interdependence of LI and L2 verbal academic skills was further indicated by the fact that Finnish skills correlated about as highly with subject-matter achievement as did Swedish skills, despite the fact that all subjects were taught in Swedish (table 14, p. 68). The analyses presented by Skutnabb-Kangas and Toukomaa are limited by the absence of powerful statistical tools (such as path analysis) that would have allowed the interaction among predictor variables to have been assessed. From this perspective the studies reported by Linde and Lofgren present a more complete picture of factors involved in predicting L2 academic achievement. The Linde and Lofgren studies. Linde and Lofgren used path-analysis procedures to fit causal models to data on Finnish children's achievement 73
Jim Cummins in Swedish schools. The first study was longitudinal in nature and involved 32 grade-three children attending a bilingual instructional program. Children's proficiency in Finnish on entry to school was strongly related to achievement at the grade-three level (path coefficient 0.66) while initial Swedish proficiency was related to a lesser level (path coefficient 0.26). A positive relation between Finnish and Swedish language skills was also observed (path coefficient 0.36). The second and third studies involved 319 and 157 Finnish-background grade-six students respectively. In both studies a positive relation between Finnish and Swedish verbal academic proficiency was reported (path coefficients 0.30 and 0.33 respectively). In spite of these positive relations, a negative relation was observed between attendance at Finnish classes and Swedish proficiency. This may be partially a function of the fact that students who have been in Sweden for less time are more likely to be in Finnishmedium classes. Length of time in Sweden was not assessed in either of these two studies but was strongly related to Swedish proficiency in the first study. The fourth study involved 388 grade-eight students and again reported a positive relation between Finnish and Swedish proficiency (path coefficient 0.24). Unlike the grade-six studies, no negative relation was observed between amount of Finnish-medium instruction and Swedish proficiency. In other words, students instructed primarily through Finnish in the elementary school were performing as well in Swedish as those whose instruction had been largely through Swedish. As in the grade-six studies, however, a limitation of this study is that length of residence in Sweden was not included as a variable in the analysis. It should be noted that the relation between LI and L2 in these studies cannot be accounted for by specific transfer of linguistic elements across languages since Finnish and Swedish are not closely related to each other. In summary, in the Scandinavian studies a significant positive relation between Finnish and Swedish verbal academic proficiency is consistently obtained and transfer of academic skills across languages is evident in the fact that instruction through Finnish entails no long-term lag in the development of Swedish academic skills. Both of these sets of findings are consistent with the postulation of a moderate degree of interdependeace between LI and L2 verbal academic proficiency in the L2 acquisition process.
Hispanic students in the United States Two longitudinal studies provide strong support for the notion of linguistic interdependence. Ramirez (1985) followed seventy-five Hispanic elementary-school students in Newark, New Jersey, enroled in bilingual programs for three years. It was found that Spanish and English verbal academic language scores loaded on one single 74
First- and second-language proficiency factor over the three years of data collection. Only students with relatively high levels of Spanish academic proficiency (as assessed by the C.T.B.S. Espanol) on entry to the program developed high levels of English proficiency (as assessed by the Maculaitis English proficiency test). Hakuta and Diaz (1985) with a similar sample of Hispanic students found an increasing correlation between English and Spanish academic skills over time. Between kindergarten and third grade the correlation between English and Spanish went from 0 to 0.68. The low crosslingual relationship at the kindergarten level is likely to be due to the varied length of residence of the students and their parents in the United States, which would result in varying levels of English proficiency at the start of school. An evaluation study of five schools attempting to implement the Theoretical Framework for the Education of Language Minority Students developed by the California State Department of Education (1981) showed consistently higher correlations between English and Spanish reading skills (range r = 0.60-0.74) than between English reading and oral language skills (range r = 0.36-0.59) (California State Department of Education, 1985). Oral language skills were assessed by a detailed rating scale completed by teachers. In these analyses scores were broken down by months in the program (1-12 months through 73-84). It was found that the relation between LI and L2 reading became stronger as English oral communicative skills grew stronger (r = 0.71, N= 190 for students in the highest category of English oral skills). A well-designed study carried out by Gonzalez (1986) with Hispanic immigrant children similarly demonstrated a considerably stronger relationship between English and Spanish reading skills than between English reading skills and English oral communicative skills. Two groups of grade-six students attending a bilingual program were compared on English and Spanish measures: thirty-four students who were born and schooled for at least two years in Mexico prior to emigrating to the United States and thirty-eight students who were born in Mexico but emigrated to the United States before beginning school. Both groups were of low socioeconomic status. It was found that the Mexican-schooled group performed significantly better on both Spanish and English reading tasks (assessed by means of the C.T.B.S. Espanol and Stanford Reading Test respectively) than the group schooled entirely in the United States. The U.S.-schooled group outperformed the Mexican-schooled group on an English oral-language measure (the Bilingual Syntax Measure (B.S.M.) II) and on ratings of English communicative proficiency. Both groups showed a high level of competence in conversational aspects of Spanish proficiency, with the Mexican-schooled group scoring somewhat higher on the Spanish B.S.M. II and the U.S.-schooled group being rated more fluent in Spanish communicative skills. In the total sample, Spanish and English reading comprehen75
Jim Cummins sion scores showed a correlation of 0.55 (/?<0.01) with each other but neither reading score was significantly correlated with ratings of communicative competence. Gonzalez concluded that the academic foundation developed by the Mexican-schooled students transferred to the acquisition of English academic skills, giving them an advantage over their U.S.schooled peers, despite the fact that both groups attended the same bilingual program.1 An interesting aspect of Gonzalez' findings is that high crosslingual correlations were obtained between the ratings of Spanish and English communicative skills, suggesting transference of some aspects of oral-language skills in addition to academic-language skills. The studies outlined above used standardized reading tests as the major dependent variables in both English and Spanish. However, similar patterns of crosslingual transfer were observed in several other studies that used more authentic measures of verbal academic skills. Goldman (1985), for example, examined the retelling of narratives in both English and Spanish by bilingual children and compared these retellings with those of a monolingual group presented with narratives in English only. Children were in kindergarten through sixth grade. It was found that children used similar comprehension strategies whether they were exposed to the story in their LI or L2. The bilingual children performed as well in English as their monolingual peers in grades three, four, five, and six, although differences in favor of the monolingual children were observed in the lower grades. Carlisle (1986) examined the writing development of grades four and six Anglo students in regular programs and Hispanic students in either submersion or bilingual programs. An analysis of covariance revealed an advantage for the bilingual-program students in comparison to the submersion-program students in English rhetorical effectiveness, syntactic maturity, and productivity, although both groups performed less well than the Anglo students in regular programs on rhetorical effectiveness and overall quality of writing. A multiple regression analysis revealed a significant relationship between rhetorical effectiveness in Spanish and rhetorical effectiveness in English. A number of other studies have reported significant correlations between Spanish and English reading skills among Hispanic students in the United States. For example, a study of seventy sixth-grade Hispanic students in a bilingual program in Rochester, New York, showed a correlation of 0.59 between the Degrees of Reading Power test (a modified multiple-choice cloze procedure) in English and Spanish (Bull, 1986). Fradd (1983) similarly reported a significant correlation of 0.41 (p<0.01) between Spanish and English reading skills among forty-one Cuban-background high-school students in Florida. In summary, the findings of studies involving Hispanic students in the United States are consistent with those of Scandinavian researchers in point76
First- and second-language proficiency ing to a moderate degree of interdependence between minority students' LI and L2 academic skills. Asian students in Canada and the United States Genesee (1979) has reported that correlations between LI and L2 reading skills tend to be lower (although still significant) when the orthographies of the two languages are dissimilar. For example, his studies of bilingual and trilingual immersion programs found lower correlations between Hebrew and both English and French than between these latter two languages themselves. Languages such as Chinese and Japanese differ even more significantly from English in their writing systems. Consequently, investigations of L l L2 relationships involving these languages pose a stringent test for the interdependence hypothesis. Several studies involving Asian immigrant students to North America have been carried out that suggest that cognitive and personality attributes of individual learners, in particular their literate competence in LI, contribute significantly to the acquisition of certain aspects of L2, despite the dissimilarity of languages and writing systems. Cummins et al. (1984) set out explicitly to test the interdependence hypothesis in a study of ninety-one Japanese and forty-five Vietnamesebackground students in Toronto. The Japanese students were the children of temporary residents who were in Canada for business or professional reasons, whereas the Vietnamese sample consisted of refugee students. The Japanese students attended a Saturday Japanese school which aimed to help students keep up with the curriculum in Japan in order to ease scholastic reintegration when they returned (often after as much as five to six years of residence abroad). Students were selected from grades two to three and five to six (Canadian grades) in order to allow the effects of length of residence (L.O.R.) to be separated from age of arrival (A.O. A.). Thus, a grade-two student with two years L.O.R. has an A.O. A. of about five years whereas the A.O.A. for a grade-six student with two years L.O.R. is about nine years. All the Vietnamese sample were recent arrivals (L.O.R. 5-22 months) and ranged in age between nine and seventeen years. Thus, all the sample had received at least some education in Vietnamese prior to immigration to Canada. The dependent variables for the Japanese group consisted of five English decontextualized verbal academic measures (two reading measures from the grade-two Gates McGinitie reading tests and three oral-language tasks) and contextualized measures derived from ratings of student interviews (administered to a subsample, n = 59). Students were also interviewed in Japanese and were administered a Japanese standardized diagnostic reading measure. A factor analysis of English measures in the Japanese study revealed three factors: (1) a grammatical-competence factor; (2) an interactionalstyle factor related to the amount of elaboration and detail volunteered 77
Jim Cummins by students in the interview; and (3) a decontextualized or verbal academiccompetence factor. Regression analyses of these three factor scores on L2 exposure (i.e., L.O.R.) and personal attribute variables (e.g., parent ratings of student personality traits) showed a strong relationship between L.O.R. and all three dimensions. The proportion of explained variance ranged from 0.26 for the English grammatical factor to 0.17 for the verbal academic factor with the interactional-style factor in an intermediate position (R square = 0.21). A block of variables representing LI cognitive/ academic attributes of the students (e.g., Japanese reading T.-score, A.O.A. in Canada) was entered next into the equation. This block added only 3 percent and 6 percent to the explained variance for L2 grammatical competence and interactional style but 18 percent explained variance for the L2 verbal academic factor. Next, a block representing Japanese interactional style and parental ratings of their children's extraversion-introversion was added. This block accounted for only 4 percent and 2 percent increment to explained variance for L2 grammatical competence and verbal academic abilities but 17 percent increment for the English interactional-style factor. In other words, variables related to students' LI cognitive and literacy skills contributed significantly to the development of L2 cognitive and literacy skills while interactional-style dimensions in LI and L2 were closely related to personality attributes of the students. In contrast to the role individual attributes of the students played in the development of these aspects of L2 proficiency, grammatical proficiency in L2 was most significantly influenced by variables related to the amount of L2 input that students received (i.e., L.O.R.). We will return to this distinction between attribute-based and input-based predictors of L2 proficiency in a later section. The interdependence hypothesis was also supported in the Vietnamese study, where performance on a Vietnamese antonyms measure together with students' age accounted for 61 percent of the variance in an English antonyms measure. Probably due to the restricted range for L.O.R. among the Vietnamese sample, L.O.R. accounted for only 6 percent of the variance in the English antonyms measure when entered first into the regression equation. A more recent study of 273 grades two to eight Japanese students in Toronto (Cummins & Nakajima, 1987) reported findings consistent with those of the previous study. English and Japanese standardized measures of reading were administered to the sample together with assessments of writing skills in both languages. L.O.R. accounted for 35 percent of the variance in English reading scores with the verbal academic block (Japanese reading, A.O.A. and age) accounting for an additional 20 percent. Minimal variance was accounted for by these variables on measures of English writing (e.g. holistic ratings of writing quality and spelling errors). It was poss78
First- and second-language proficiency ible to examine the relationship between English and Japanese writing measures for a subsample (n = 70). A number of Japanese writing variables related significantly to overall quality of English writing and to English spelling. Among the strongest relationships was that between Japanese spelling (katakana) and English spelling. This relationship was independent of more general cognitive/academic variables such as Japanese reading proficiency and age. Consistent results are also reported by Iwasaki (1981) in a study of Japanese children in New York. She investigated the effect of initial age of intensive exposure to L2 and transferability of cognitive/academic skills across languages among grades seven and eight students in both full-time (N = 76) and part-time (N = 72) Japanese schools. The Gates McGinitie level E reading-comprehension test Normal Curve Equivalent scores were used as the dependent measure for English cognitive/academic proficiency. Japanese measures were developed by the investigator and combined into grade-normed scores so that the grades seven and eight data could be combined. Length of local schooling (an index of L.O.R.) was strongly related to English reading performance, accounting for 52 per cent of the variance in the regression equation (N=106). Performance on the Japanese measures added an additional 9 percent (p<0.001) to the explanation of variance. A chi-square analysis for thirty-nine students with more than twenty-seven months of local schooling showed a significant (p<0.01) relationship between Japanese and English proficiency. For example, of those who were in the high and medium ranges in Japanese proficiency eighteen were above the mean in English reading and only six below the mean. Of those low in Japanese proficiency only four were above the mean in English compared to eleven below the mean. Iwasaki also reported that grade-norm acquisition of English reading occurred at twenty-seven months of local schooling, suggesting an acquisition process that appears more rapid than the forty-eight months L.O.R. required by similar economically advantaged Japanese students in Toronto. The difference may be due, in part, to the non-equivalence of L.O.R. and length of local schooling indices. In short, moderately strong relationships are observed between reading performance in Japanese and English despite the differences in writing systems. Some crosslingual relationships are also evident in Japanese and English writing skills, although these are less clear-cut than those observed for reading. The final study to be examined in this section focuses on the bilingual development in English and Chinese of 112 grades four to six Chinese-background students in Seattle. This study, prepared by fhe Southwest Educational Development Laboratory (S.E.D.L.) (Hoover, 1983), included a large number of English oral language and literacy variables 79
Jim Cummins as well as a measure of LI (Cantonese) reading skills. Here we are primarily concerned with predictors of children's performance on the Interactive Reading Assessment System (IRAS2) developed by Calfee and Calfee (1979). Part of the sample had started schooling in the United States and had been in a Chinese-English bilingual program, while the remainder had immigrated to the United States after the start of formal schooling and had received varying amounts of LI literacy instruction abroad. Thus, it is possible to examine L1-L2 relationships for these two groups separately. For the U.S.-schooled students (n = 66) a significant positive correlation was found between IRAS performance and both amount of L2 instruction (r = 0.28, M = 112) and amount of LI instruction (r = 0.24, n = 112). The effect of bilingual instruction can be seen in the difference between subgroups who had minimal bilingual instruction (0-2 semesters, IRAS mean = 6.2) and those who had substantial amounts of bilingual instruction (8 semesters, IRAS mean = 7.5). For this group regression analyses showed length of residence in the U.S.A. accounting for 19 percent of the variance in IRAS scores, while an LI cognitive/academic block consisting of amount of bilingual instruction, Cantonese IRAS, and age of arrival in the U.S.A. accounted for an additional 14 percent. Of these latter variables only amount of bilingual instruction attains significance (p<0.05), while the other two variables entered subsequently approach significance (see Cummins, 1983; Hoover, 1983). For the initially foreign-schooled group, L.O.R. explained 29 percent of the variance in English IRAS performance while the LI cognitive/academic block explained almost 11 percent. Of the LI cognitive/academic block, only amount of LI instruction abroad attained significance (p = 0.05), accounting individually for 6 percent of the variance. The fact that Cantonese reading proficiency does not attain significance in this equation can be attributed to its overlap with LI instruction, since its partial correlation with English IRAS attains significance after L.O.R. is initially entered into the equation (r = 0.23, /?<0.02) but drops to insignificance after amount of LI instruction is entered. In summary, although the amount of variance accounted for by variables reflective of LI cognitive/academic proficiency is less than that accounted for by L.O.R. for both foreign-and U.S.-schooled groups (11 percent and 14 percent respectively), the effect does attain statistical significance and is consistent with the pattern observed for Japanese students in the Toronto studies. Thesefindingsappear consistent with Genesee's (1979) observation that the transfer of reading skills from LI to L2, while significant, is less when the languages are very dissimilar than is the case with similar languages. Sociocultural variables such as motivation to maintain and develop literacy skills in Chinese and Japanese are also likely to affect the relation between LI and L2 literacy skills. 80
First- and second-language proficiency Additional studies of bilingual students Several additional studies can be briefly noted. Falter (1988) reported correlations between English and French reading and writing tasks among 145 grade-five students in French immersion programs in Northern Ontario. The intercorrelations between English and French standardized reading, cloze, and writing measures were 0.59, 0.61, and 0.47 respectively. All were significant beyond the 0.001 level. Davidson, Kline, and Snow (1986) reported that decontextualized language skills (operationally defined as ability to give definitions) were highly correlated across languages in French-English bilingual children. Correlations across languages were stronger than correlations between decontextualized and contextualized skills within languages. This finding parallels that of the California State Department of Education (1985) reviewed above. Snow (this volume) also reports a strong relationship of definitional ability to reading achievement. Geva and Ryan (1987) investigated the interdependence hypothesis as part of a study of cognitive, memory, and linguistic-processing predictors of L2 reading development. Subjects were seventy-three grade five to seven children attending a bilingual Hebrew-English day school in Toronto. The test battery included measures of non-verbal intelligence, linguistic proficiency in LI (English reading, vocabulary, and clause-completion tasks3), linguistic proficiency in L2 (Hebrew oral-proficiency ratings and reading ability), and memory measures in LI and L2. The results pointed to an important role for memory processes in performing linguistic tasks in L2 as a result of the fact that L2 linguistic processing is less automatized than LI processing. A significant correlation (r = 0.37, /?<0.001) was found between the English clause-completion task and Hebrew reading, suggesting that those children who can more systematically employ executive-control functions in their LI are more likely to do so in their L2 as well. This correlation maintained significance (r = 0.26) even when grade and non-verbal intelligence were partialled out. Strong correlations were also observed between verbal memory-span tasks in Hebrew and English. However, the relationship between Hebrew and English reading was not significant. Geva and Ryan interpret these results as supporting the interdependence hypothesis, but caution that the "common underlying proficiency" involves more than just non-verbal intelligence; linguistic analytic skills or executive functions and memory span represent independent components of the common underlying proficiency." The lack of significant relationship between Hebrew and English reading in the Geva and Ryan study cannot be attributed to unique characteristics of the Hebrew language in view of the fact that Kemp (1984) reported that Hebrew (LI) cognitive/academic abilities accounted for 48 percent 81
Jim Cummins of the variance in English cognitive academic skills among 196 seventhgrade Israeli students. Malakoff (1988) reported crosslingual relationships of French-English bilingual students' performance on a verbal-analogies task. The students were sixth graders attending the International School of Geneva. Eighteen of the students were attending a program that carried out instruction primarily in English and twenty-four were in a predominantly French-language program. The verbal analogies were divided into easy and hard sets in both French and English. For the easy task the correlations between performance in French and English were r = 0.56 (/?<0.01) for the Englishlanguage program and r = 0.77 (p<0.001) for the French-language program group. For the hard analogies the correlations were also significant: r = 0.78 (p<0.001) for the English language program and r = 0.58 (p<0.01) for the French-language program. Canale, Frennette, and Belanger (1987) tested the extent to which the interdependence hypothesis applied to writing skills among a sample of grades nine and ten Franco-Ontarian students who were being educated predominantly through French. A sample of 128 essays (64 English, 64 French) written by thirty-two students equally divided between narrative and expository genres and randomly drawn from a larger data base of student writing formed the data for the study. It was found that crosslingual correlations between analytic scores were highly significant for the same mode of writing (r = 0.77 for narratives, r = 0.78 for expositions). Ho (1987) has reviewed three studies carried out in Singapore that examined crosslingual interdependence among English-Chinese (N = 296), English-Malay (7V = 91), and English-Tamil (N = 28) bilinguals. Significant correlations were obtained for verbal/academic performance among the Chinese and Tamil bilinguals (r = 0.21, p<0.01, and r = 0.56, p<0.0l respectively) but the correlation for the Malay group was not significant. This latter finding was attributed to the mixed nature of the Malay learners; when the group was separated into those of Malay ethnic origin and those of non-Malay origin the correlations were 0.28 (p<0.05) and —0.25 (n.s.) respectively for these two groups. The negative correlation in the latter group was attributed to the diversity of language background and abilities in this group. In summary, in these studies a moderately significant correlation was observed between languages that are quite dissimilar in orthography, syntax, and directionality. A European study described by McLaughlin (1986) also suggests that interdependence of academic skills operates across languages that are quite dissimilar, in this case Turkish and German. The study carried out by Rehbein (1984) found that 82
First- and second-language proficiency the ability of Turkish children to deal with complex texts in German was affected by their ability to understand these texts in their first language. Rehbein's investigations suggest that there is a strong developmental interrelationship between the bilingual child's two languages and that the conceptual information and discourse strategies acquired in the first language transfer to the second. (McLaughlin, 1986, pp. 34-35)
Finally, a study of 191 grade-seven Portuguese-background students in Toronto showed that measures of students' discourse proficiency (i.e., judgments regarding coherence and cohesiveness of text) in Portuguese and English were strongly interrelated (r = 0.54, p<0.001, n = 65). The English discourse measure loaded on a Portuguese proficiency dimension in a principal components analysis carried out on the data (Cummins, Lopes, & King, 1988). Studies of adult second-language learners Several studies of adult L2 learners support the interdependence hypothesis. Salgado (1988), for example, reported a correlation of 0.58 (p<0.001) between reading scores in Spanish and English among 201 Hispanic-community-college students in the New York area. In a sample of 182 Hispaniccommunity college students in Houston, Guerra (1984) studied the relationship between students' ability to judge and correct syntactic errors in English and their ability to judge and correct similar types of syntactic errors in Spanish. Other variables in the analysis included length of residence in the United States and amount of instruction in English as a second language (E.S.L.). Guerra reported that Spanish language skills and years of schooling in Spanish significantly predicted students' ability to recognize errors in English. In predicting ability to correct errors in English, E.S.L. instruction was equally significant to these variables. The ability to recognize and correct syntactic acceptability in Spanish was the highest predictor of the same ability in the second language (English). The interdependence notion is also supported in one of the few studies to focus on cognitive processes and text production (Cumming, 1987). The study assessed the relationship of writing expertise and second-language proficiency to adults' writing performance in E.S.L. Both writing expertise and second-language proficiency accounted for large proportions of variance in the qualities of E.S.L. texts and composing behaviors. These effects were independent of each other, however, suggesting that they are psychologically distinct abilities. The relation of these findings to the interdependence hypothesis is that writing expertise is common across languages but for effective writing performance in an L2 both expertise and specific knowledge of the L2 are required. As expressed by Cumming: the present research has identified the empirical existence of certain cognitive abilities entailed in writing expertise - problem solving strategies, attention to complex 83
Jim Cummins aspects of writing while making decisions, and the qualities of content and discourse organization in compositions - which are not related directly to second language proficiency but which appear integral to effective performance in second language writing. (1987, p. 175)
Writing expertise is viewed as a central cognitive ability with secondlanguage proficiency an additive factor that facilitates the operation of writing expertise in a new domain and possibly enhances writing expertise in subtle ways. In Cumming's study second-language proficiency was very closely related to length of residence in an English-speaking milieu, but both these variables were less fundamentally related to L2 composing behavior than was their writing expertise that manifested itself in both languages. Conclusion The data reviewed in this chapter suggest that both attributes of the individual learner and aspects of the input received by the learner contribute in important ways to the development of different aspects of L2 proficiency. The importance of quantity of input is clearly indicated by the consistently strong relationships observed between length of residence and L2 acquisition. However, L.O.R. was not equally related to all aspects of proficiency. For example, in the Cummins et al. (1984) study, acquisition of L2 conversational syntax was considerably more dependent on L.O.R. than either L2 academic proficiency or interactional style. Cognitive and personality attributes of the individual contributed as much to the explanation of variance in these dimensions as did L.O.R. These learner attributes, however, were unrelated to individual differences in L2 conversational syntax. In general, moderately strong crosslingual relationships are observed for attribute-based aspects of LI and L2 proficiency as a result of the fact that underlying attributes of the individual manifest themselves in the individual's performance in both languages. Within the sphere of cognitive attributes, the data show consistent moderate relationships between decontextualized aspects of LI and L2 proficiency in studies carried out in a wide variety of sociolinguistic situations and involving subjects ranging in age from early childhood to adult. Relationships tend to be somewhat smaller, albeit still statistically significant, in the case of languages that differ markedly in writing systems. Although the most consistent relationships are found in the sphere of decontextualized language proficiency, some crosslingual relationships were also observed in aspects of contextualized language proficiency that reflect particular learner attributes (e.g., interactional style). In short, the crosslingual relationships for academic or decontextualized aspects of proficiency are a function of the fact that both LI and L2 proficiencies are subsumed by cognitive attributes of the learner, while 84
First- and second-language proficiency
the learner's personality attributes are manifested in the cross-lingual relationship of interactional style. As Geva and Ryan (1987) suggest, it may be possible to develop a more refined model of the cognitive attributes that subsume LI and L2 decontextualized abilities; for example, by distinguishing the contribution of non-verbal ability, verbal analytic functions, memory span, etc. This direction is consistent with Cumming's (1987) conclusion that writing expertise represents a specific underlying competence that manifests itself in writing performance in both languages and is not reducible to other underlying cognitive attributes such as verbal or nonverbal I. Q. It should be emphasized that the distinction between attribute-based and input-based aspects of proficiency is a relative one in that individual learner attributes will be involved in most aspects of L2 learning to a greater or lesser extent and appropriate input is clearly essential for development of all aspects of proficiency. Thus, attributes and input are not totally independent of each other, as illustrated by the fact that highly motivated individuals are likely to seek out a greater amount of input than those less motivated. The point is, however, that the relative importance of attributes and input will vary for different aspects of the L2. The relatively weak relationship between contextualized and decontextualized (or conversational and academic) aspects of proficiency in a language can thus be understood as a function of the distinction between attribute-based and input-based aspects of proficiency. A weak relationship is observed either because different attributes are involved (e.g., cognition and personality) or because input characteristics are relatively more significant for acquisition (e.g., of L2 conversational syntax) than individual attributes. Clearly, the acquisition context must be considered also in that acquisition of L2 syntax may be considerably more dependent on cognitive attributes in formal classroom contexts than in naturalistic settings, where quantity and quality of input are primary determinants of acquisition. How does the attribute-based/input-based distinction relate to the common distinction made between contextualized and decontextualized aspects of proficiency and to Biber's three dimensions of language use? Biber's dimensions and the contextualized/decontextualized distinction refer primarily to characteristics of the language itself (i.e., to the type of input the learner receives) rather than to characteristics of the learner or user of the language. In these formulations, no relationship is posited between attributes of the learner and the development of proficiency in different dimensions of language use (e.g., contextualized/decontextualized). The present conceptualization goes beyond these distinctions to posit relationships that are predictable, at least in principle, between learner attributes and characteristics of the language. Thus, the strong crosslingual relationship between LI and L2 decontextualized language skills found 85
Jim Cummins in the research of Snow and her colleagues is interpreted as a function of the fact that decontextualized skills in LI and L2 reflect underlying cognitive attributes of the individual that manifest themselves in both languages. As the Cummins et al. (1984) study suggests, certain aspects of contextuaiized language skills are also reflective of underlying personality attributes, whereas for others (e.g., conversational syntax) the role of quality and quantity of input largely overshadows individual difference factors. Predictions that emerge from this conceptualization are (1) consistent crosslingual relationships between aspects of LI and L2 are reflective of underlying attributes of the individual in addition to characteristics of L2 input; and (2) aspects of LI and L2 proficiency that are unrelated across languages are also largely unrelated to underlying individual difference dimensions (e.g., cognition, personality) but strongly related to quality and/or quantity of L2 input received by the learner. Research directions suggested by this approach include refining the attribute dimensions that can be distinguished within the more general categories of cognition and personality and, in turn, the crosslingual aspects of language proficiency that are subsumed by these underlying attribute dimensions. Notes 1.
2. 3.
Findings reported by Baral (1979) appear at first sight inconsistent with the findings of Gonzalez' study. Baral reported that immigrant students who had at least two years of schooling in Mexico performed significantly lower in English academic skills than Mexican-American students born in the United States. However, the socioeconomic status of the immigrant students was significantly lower than that of the native-born students and their length of residence in the United States (two to five years) is likely to have been insufficient to catch up with the native-born students. IRAS scores can be read as approximations of grade equivalent scores. The clause-comple.tion task involved items such as "She cooked the potatoes and meat for him because (a) she wishes to be helpful; (b) he could fly airplanes; (c) he had eaten supper; (d) she wanted to help him." It was designed to assess children's level of analyzed linguistic knowledge in LI (see Bialystok & Ryan, 1985).
References Baral, D. P. 1979. Academic achievement of recent immigrants from Mexico. NABE Journal, 3, 1-13. Bereiter, C. & M. Scardamalia 1981. From conversation to composition: the role of instruction in a developmental process. In R. Glaser (ed.), Advances in instructional psychology, vol. 2. Hillsdale, NJ: Erlbaum. 86
First- and second-language proficiency Bialystok, E. & E. B. Ryan 1985. Towards a definition of metalinguistic skill. Merrill-Palmer Quarterly, 57, 229-251. Biber, D. 1986. Spoken and written textual dimensions in English: resolving the contradictory findings. Language, 62, 384-414. Bruner, J. S. 1975. Language as an instrument of thought. In A. Davies (ed.), Problems of language and learning, London: Heinemann. Bull, D. C. 1986. An evaluation of the modified cloze procedure as a measure of transferability of reading comprehension in bilingual children. Unpublished doctoral dissertation, State University of New York at Buffalo. Calfee, R. C. & K. H. Calfee 1979. Interactive reading assessment system. Stanford, CA: Stanford University. California State Department of Education 1981. Schooling and language minority students: a theoretical framework. California State University, Los Angeles. California State Department of Education 1985. Case studies in bilingual education: first year report. Federal Grant #G008303723. Sacramento: California State Department of Education. Canale, M., N. Frennette, & M. Belanger 1987. Evaluation of minority students writing in first and second languages. In J. Fine (ed.), Second language discourse: a textbook of current research. Norwood, NJ: Ablex. Carlisle, R. S. 1986. The writing of Anglo and Hispanic fourth and sixth graders in regular, submersion and bilingual programs. Unpublished doctoral dissertation. University of Illinois at Urbana-Champaign. Collier, V. P. 1987. Age and rate of acquisition of second language for academic purposes. TESOL Quarterly, 21, 617-641. Cumming, A. 1987. Writing expertise and second language proficiency in ESL writing performance. Unpublished doctoral dissertation, University of Toronto, Toronto. Cummins, J. 1981. The role of primary language development in promoting educational success for language minority students. In California State Department of Education (ed.), Schooling and language minority students: a theoretical framework. Evaluation, Dissemination and Assessment Center, California State University, Los Angeles. 1983. Policy report: language and literacy learning in bilingual instruction. A policy review of the Southwest Educational Laboratory Study submitted to the National Institute of Education. 1984. Bilingualism and special education: issues in assessment and pedagogy. Clevedon, Avon, England: Multilingual Matters. 1989. Empowering minority students. Sacramento, CA: California Association for Bilingual Education. Cummins, J. & K. Nakajima 1987. Age of arrival, length of residence, and interdependence of literacy skills among Japanese immigrant students. In B. Harley, P. Allen, J. Cummins, & M. Swain (eds.), The development of bilingual proficiency. Vol. Ill: Social context and age. Final report submitted to the Social Sciences and Humanities Research Council. Toronto: Ontario Institute for Studies in Education. Cummins, J., J. Lopes, & M. L. King 1988. The language use patterns, language 87
Jim Cummins attitudes, and bilingual proficiency of Portuguese Canadian children in Toronto. In B. Harley, P. Allen, J. Cummins, & M. Swain (eds.), The development ofbilingual proficiency. Vol. Ill: Social context and age. Final report submitted to the Social Sciences and Humanities Research Council. Toronto: Ontario Institute for Studies in Education. Cummins, J., M. Swain, K. Nakajima, J. Handscombe, D. Green, & C. Tran. 1984. Linguistic interdependence among Japanese and Vietnamese immigrant students. In C. Rivera (ed.), Communicative competence approaches to language proficiency assessment: research and application. Clevedon, Avon, England: Multilingual Matters. Davidson, R. G., S. P. Kline, & C. E. Snow 1986. Definitions and definite noun phrases: indicators of children's decontextualized language skills. Journal of Research in Childhood Education, 1, 37-48. Donaldson, M. 1978. Children's minds. Glasgow: Collins. Falter, P. 1988. The effect of metalinguistic instruction on the bilingual literacy skills of selected grade 5 French immersion children. Unpublished doctoral dissertation. University of Toronto, Toronto. Fradd, S. H. 1983. Language acquisition of 1980 Cuban immigrant junior high school students. Unpublished doctoral dissertation, University of Florida. Genesee, F. 1979. Acquisition of reading skills in immersion programs. Foreign Language Annals, 72,71-77. Geva, E. & E. Ryan 1987. Linguistic knowledge and cognitive demands for academic skills in first and second language. Paper presented at the meeting of the Canadian Psychological Association, Vancouver. Goldman, S. R. 1985. Utilization of knowledge acquired through the first language in comprehending a second language: Narrative comprehension by SpanishEnglish speakers. Report submitted to the Office of Bilingual Education and Minority Language Affairs, Washington, DC. Gonzalez, L. A. 1986. The effects of first language education on the second language and academic achievement of Mexican immigrant elementary school children in the United States. Unpublished doctoral dissertation, University of Illinois at Urbana-Champaign. Guerra, V. 1984. Predictors of second language learners' error judgements in written English. Unpublished doctoral dissertation, Corpus Christi State University. Hakuta, K. & R. M. Diaz 1985. The relationship between degree of bilingualism and cognitive ability: a critical discussion and some new longitudinal data. In K.E. Nelson (ed.), Children's language, vol.V. Hillsdale, NJ: Erlbaum. Ho, W. K. 1987. Proficiency across languages: an overview. In W. K. Leong, N. E. Guan, K. Abdullah, S. P. Thinnappan, & N. Govindasamy (eds.), The bilingual ability of a sample of primary three pupils: a collection of three papers. Singapore: Institute of Education. Hoover, W. 1983. Language and literacy learning in bilingual instruction. Cantonese Site Analytic study. Final report submitted to the National Institute of Education. Iwasaki, M. 1981. A study of sociological and developmental factors in first language 88
First- and second-language proficiency maintenance and second language acquisition of kaigaishijo in New York. Unpublished master's thesis, University of Hawaii. Kemp, J. 1984. Native language knowledge as a predictor of success in learning a foreign language with special reference to a disadvantaged population. Unpublished master's thesis, Tel-Aviv University. Linde, S. G. & H. Lofgren 1988. The relationship between medium of instruction and school achievement for Finnish-speaking students in Sweden. Language, Culture and Cognition, / , 131-146. McLaughlin, B. 1986. Multilingual education: theory east and west. In B. Spolsky (ed.), Language and education in multilingual settings. Clevedon, Avon, England: Multilingual Matters. Malakoff, M. 1988. The effect of language of instruction on reasoning in bilingual children. Applied Psycholinguistics, 9, 17-38. Olson, D. R. 1977. From utterance to text: the bias of language in speech and writing. Harvard Educational Review, 47, 257-281. Ramirez, C. M. 1985. Bilingual education and language interdependence: Cummins and beyond. Unpublished doctoral dissertation, Yeshiva University. Rehbein, J. 1984. Diskurs und Verstehen: Zur Role der Muttersprache bei der Textarbeitung in der Zweitsprache. University of Hamburg. Salgado, L. J. 1988. Language proficiency and retention among Hispanic students in community colleges. Unpublished doctoral dissertation, Seton Hall University. Skutnabb-Kangas, T. & P. Toukomaa 1976. Teaching migrant children's mother tongue and learning the language of the host country in the context of the sociocultural situation of the migrant family. Helsinki: The Finnish National Commission for UNESCO. Snow, D. E. & M. Hoefnagel-Hohle 1978. The critical period for language acquisition: evidence from second language learning. Child Development, 49, 1114-1128.
89
5
Giving formal definitions: a linguistic or metalinguistic skill?1 CATHERINE E. SNOW, HERLINDA CANCINO, JEANNE DE TEMPLE, and SARA SCHLEY
As children go farther and farther in school, they are required to devote relatively more and more attention to the form as compared to the content of their language productions. Tasks which might seem straightforwardly communicative and linguistic, such as writing paragraphs, telling stories, answering comprehension questions, and explaining how things work, must increasingly also be dealt with as metalinguistic problems as form becomes a codeterminant of success along with content. One such task, which we will argue shifts from being linguistic to metalinguistic in the course of the elementary grades, is defining words. Young children very reasonably respond to a question like "What's a hat?" with "You wear it," and such a response is tolerated if the child is young enough. 2 Older children, on the other hand, are expected to respond to such questions by giving "formal definitions," which conform to particular standards for form as well as for content, for example, "A hat is an article of clothing worn on the head." Definitions, though a rather specialized speech genre, are of both theoretical and practical interest to students of language development. In school settings, definitions are often requested of children, and giving definitions (or having children look them up in dictionaries and copy them) is a standard and frequent technique for vocabulary training. Theoretically, definitions are interesting because they constitute one example of what has been referred to as decontextualized language use - language used in ways that eschew reliance on shared social and physical context in favor of reliance on a context created through the language itself. A good defini tion avoids exploiting the availability of information that happens to be shared between speaker and hearer, restricting itself instead to information shared by the entire speech community. Furthermore, giving good definitions requires analyzing one's own knowledge of word meaning to distinguish "definitional" from incidental information about the target concept, as well as control of the conventional form for giving definitions (a copula, a superordinate, a restrictive complement). As such, giving 90
Giving formal definitions formal definitions constitutes a metalinguistic task in which both analysis of knowledge and control of processing, Bialystok and Ryan's (1985) terms, are crucial to success. Evidence that success in giving definitions has a metalinguistic component derives from an analysis of what constitutes a "good definition" in comparison to what constitutes a communicatively effective one. Perhaps the most communicatively transparent definitions are ostensive ("What's a clock?" - "That thing on the wall"), but such highly contextualized definitions do not resemble the formal, "Aristotelian" definitions typically required of children at school ("A clock is a machine that tells time"). Formal definitions typically identify a superordinate class to which the definiendum belongs, and add relevant information about restrictions on membership in that class, while avoiding specific or idiosyncratic information that may be true but not central (e.g., can be made of wood or metal; can be wind-up, electric, or battery-operated; some have alarms). In order to give a good formal definition, one must analyze all one knows about a concept to separate the crucial from the irrelevant information. Growth in the ability to perform this sort of analysis of one's own knowledge may, thus, lie at the heart of the development of the ability to give good definitions. We know that children have considerable difficulty either in giving or in understanding formal definitions before the middle-school years. Young children typically include in their definitions information that is incidental, idiosyncratic, and highly personal (Hagtvet, 1986; Litowitz, 1977; Snow, 1990; Watson, 1985), though they can sometimes be induced to improve their definitions by adding a superordinate if pushed (Watson & Olson, 1987). Older children who demonstrate with some words that they know what form a definition should take fail to adhere to the constraints that only "definitional" information be given for other words (Snow, 1990; Snow, Cancino, Gonzalez, & Shriberg, 1989). Even adults often fail fully to respect the constraints on the definitional genre for abstract nouns, and when giving definitions for verbs and adjectives (Markowitz & Franz, in press). Understanding formal definitions is also quite difficult for children as old as ten or twelve (Miller & Gildea, 1985), perhaps because the conventions that govern the form are opaque to children who have not yet developed a reliable ability to analyze their own word and concept knowledge. The suggestion that skill at giving formal definitions relates to the ability to analyze one's own knowledge about a word's meaning is supported by the finding that definitional skill relates in rural populations to number of years of schooling (Tapia-Uribe, 1988) and to school achievement, particularly in literacy (Snow, Cancino, Gonzalez, & Shriberg, 1989). Literacy tasks, like some oral-language tasks, require analysis of knowledge; this is particularly true for those literacy tasks in which new information must 91
Catherine E. Snow et al. be comprehended and/or remembered by virtue of being related to old information. Thus, it makes good sense that performance on such tasks relates to performance on the formal-definitions task. In addition to analysis of knowledge about word meanings, performing well on formal definitions requires analysis of the form itself, and acquisition of the linguistic structures required to replicate it. Formal definitions are, despite their saliency in classrooms, a rather infrequent form, and one which may require the availability of relatively rare vocabulary items for identifying the superordinate class. Accordingly, one might expect that children perform worse in giving definitions in their second language, in which they have had less opportunity for exposure to the relevant form and vocabulary. Alternatively, however, if the second language is also the curricular language, and thus the language in which definitions have mostly been heard, one might expect better performance. Malakoff (1988) and Johnson (this volume) both found that performance on school-like tasks was better in the school language even for learners whose overall proficiency in that language was limited. Finally, one might expect that performance in giving formal definitions would relate to bilingualism. Considerable evidence suggests that bilingual children have higher levels of metalinguistic skill than monolingual children. Bilingual children have had an opportunity to develop control of processing to a greater extent than monolingual children (Bialystok, 1986), particularly children who have acquired the second language for use in academic circumstances. One might expect, then, that performance on a task like giving formal definitions would be better for children who have had the opportunity to achieve higher degrees of bilingualism. Alternatively, one might expect that children who are good at the formal-definitions task will be more skilled at foreign-language learning, a task which is heavily dependent on metalinguistic skills. Data relevant to all these issues will be presented in this chapter. The study Setting Data were collected from children attending the United Nations International School (UNIS), an independent school in Manhattan that serves the U.N. and international community of the New York City area, as well as American children whose parents are interested in a high-quality, culturally diverse educational setting. English is the curricular language of UNIS, and French is taught as a foreign language to all the children in the elementary grades who are fully integrated into the English-language curriculum. Excellent English as a Second Language (E.S.L.) services are 92
Giving formal definitions available to the approximately 50 per cent of the children at UNIS who are not native speakers of English; after exiting from E.S.L., children spend approximately one year in the "transition" program, during which they get special help from the classroom teacher during the time their classmates attend foreign-language classes.
Subjects One hundred and fifty second through fifth graders participated in the larger study of which the definitions task was one part. Of these, 137 provided usable data on the definitions task in English and seventy-one provided usable definitions in French. The children were distributed approximately equally over second, third, fourth, and fifth grades. They represented the full range of children attending UNIS, in terms of their family backgrounds (children from monolingual English-speaking families, children born in the U.S. A. into families where one or more other languages are usually spoken, children who had started school abroad and received E.S.L. training after transferring to UNIS), though children whose English was still very poor could not complete the test battery. An attempt was made to test all the children in French as well as in English, but this was only possible for sixty-four of them; the others had either started French classes much more recently, or were simply poorer students and thus had not learned enough French to be able to perform this task. Thirty-two of the sixty-four subjects had English as a major or exclusive home language and were learning French as a second language in the classroom. Twelve had French as their home language, were fully integrated in the English curriculum, and thus were enrolled in French as a second language. The remaining twenty children used a language other than English or French at home, attended school in English, and were learning French as a curricular subject. Seven children were tested in French but not English.
Procedure The definitions test was administered as part of a longer test battery. The adult experimenter and the child went to a quiet, private corner within the school for the individual test session. The instructions for the definitions task were simply "Tell me what xxx is." The words used were the first ten nouns from the Wechsler Intelligence Scale for Children (WISC-R). These words were selected because we wished to use items for which all the children would be likely to know the meaning. Translation equivalents were used in French, with occasional substitutions when the translation was ambiguous. 93
Catherine E. Snow et al. Coding The coding procedure is described here in full for the reader's ease, though basic results from some of the scores have been reported elsewhere (Snow, 1990). The coding was designed to reflect the degree to which children approached the norm of adult-like, Aristotelian definitions, with a superordinate, a restrictive complement, and the inclusion of criterial information about the word's meaning. In outline, coding proceeded as follows: the transcribed definition was first sorted into one of two broad classes. It was called a formal definition if it included anywhere in the stretch of talk a copular construction and some superordinate predicated of the definiendum. Definitions with vague superordinates {something, a thing) and quasisuperordinates {a kind of) were scored as formal. Each child received a score indicating the percent of words defined that were given a formal definition (%F.D.). Formal definitions were then scored on a variety of categories: 1.
2.
3.
4.
Syntax: the degree to which the definition given approached the ideal definitional syntax "x is a y [restrictive complement]" was scored on a four-point scale, in which no complement was scored as 1 and a full relative clause was scored as 4. Superordinate: the quality of the superordinate was rated on a four-point scale, with thing and a kind of scored as 1, something or someone scored as 2, something or someone with added specific qualifications was scored as 3, and the adult-like superordinate (e.g., animal, vehicle, utensil) was scored as 4. Complement: the semantic content of the complement was rated on a four-point scale with no complement scored 0, an incorrect or misleading complement scored 1, a correct but insufficiently limiting complement scored 2, and a properly restrictive complement scored 3. Definitional features: the number of correct definitional features mentioned was counted. Definitional features were decided on in advance for each word, based on elicited adult and dictionary definitions of the words.
A summed score based on the above four categories was calculated per word; this score is referred to as the formal definitional quality (F.D.Q.)If non-definitional but correct information about the meaning of a word was provided in addition to a formal definition (e.g., examples of its use, descriptions, comparisons) instances were scored under the category formal definitional supplement (F.D.S.). F.D.Q. and F.D.S. scores were averaged over the number of words which received formal definitions to calculate the child's score. If a definition did not qualify as formal, it was then scored to reflect how much information was provided about the definiendum. Points were given for correct information about definitional features, descriptive features, functions, examples, etc. These points were summed per word 94
Giving formal definitions to provide a score of informal definitional quality (I.D.Q.). The I.D.Q. word scores were averaged over the number of words which received informal definitions to calculate the child's score. Thus, the I.D.Q. score was independent of the F.D.Q. and F.D.S. scores, since it was based on a different set of words. The entire definition, whether formal or informal, was rated on a fourpoint scale for communicative adequacy (C.A.). If it was possible to know what word was being defined just from the definition, a score of 4 was assigned; if one would probably select the correct word, but there remained some ambiguity about exact identity, a score of 3; etc. The C.A. score was averaged over all the words for which definitions were given. Finally, the incidence of use of conversational features (C.F.) in either formal or informal definitions was counted. Conversational features included all gestures or verbal behaviors that exploited the conversational interactive context, for example, asking direct questions of the adult, requesting help with words, using gestures, soliciting acknowledgement, language switches, and linguistic fillers. Two coders rated all the English definitions from the children, and served as trainers for the single French-definitions coder. All coders had to reach perfect agreement on five definitions protocols before coding independently.
Results
Nature of development Development of skill at giving formal definitions has two components: recognizing that formal definitions are called for, and being able to give formal definitions of adequate precision and conventionality. Development in both these domains occurred between second and fifth grade in the UNIS sample (Snow, 1990, presents details of data on development of definitional form). Percent of English definitions that were formal increased from 49 percent at second grade to 79 percent at fifth (Snow, 1990). To demonstrate the kind of development in quality that occurred, we present here definitions that were scored at approximately the average score achieved by second, third, fourth, and fifth graders giving formal definitions (scores achieved are in parentheses). A donkey is . . . a baby horse, I think, and you can ride on it, but it's not as big as a horse (6.5). A donkey's something you can ride on, and it sounds like this, hee-a, hee-a(7.5). Donkey? Well, it's a sort of horse, a very small horse, which . . . is very 95
Catherine E. Snow et al. lazy, and ... if not really ... pulled ... can just sit there and ... not do anything (8). An animal ... that looks like a horse; it's the same family as the horse, and it's ... hard to ... uh ... move around (9). To extend the contrast let us consider the lowest scoring and the highest scoring formal definitions of donkey offered by any of the children: It's a dumb animal (4). A donkey is an animal that most people use to wor ... to make them work for them, or ... to ... use to ride (11.5). These definitions do not differ very greatly in semantic content. Though the superordinates vary from something to sort of horse to animal, the lowest-scoring definition here actually included animal, the most sophisticated superordinate used by anyone in the sample for donkey. The notion that donkeys are used as beasts of burden is at least hinted at in all but the poorest definition, and indications of stubbornness are also quite general. These definitions are not distinguished from one another by semantic content as much as by form. The examples given here, which are typical of the range of formal definitions offered by children in this study, suggest that the most sophisticated definitions differ from the less sophisticated in the degree to which the information given is organized into a single, well-planned structure. It was characteristic of the best definitions, and of those given by older children, that they contained many false starts, hesitations, and restructurings. The poorer definitions, and those given by younger children, tended, on the other hand, to be much more fluent. The dysfluencies in the definitions given here clearly reflect the complexities of syntacticizing into a single sentence information which is simple enough to convey in several separate sentences: donkeys are animals, people cause donkeys to perform work, people ride donkeys, donkeys do not work willingly, donkeys resemble and are related to horses. The dysfluencies associated with going beyond a simple listing of donkey attributes to a fully developed definition suggest that success at giving formal definitions shares with other distanced, decontextualized language tasks the need for selfmonitoring and careful planning. Children exhibit this on-line planning and their struggles with the constraints of the form through their hesitations and reformulations. While the formal definitions showed the development exemplified above, the quality of informal definitions (I.D.Q.) showed no age-related changes (Snow, 1990). Since the scores on the informal definitions reflect the amount of information they contain, this suggests that development in giving definitions is not controlled primarily by the fact that children know more about the word's meaning or about the target concept. Rather, they know better 96
Giving formal definitions Table 5.1 Means of English and French communicative adequacy (C. A.) and conversational features (C.F.) scores Grade CA English*
French** CF English* French**
2
3
4
5
F
P
0.99 1.78 0.06 0.25
1.33 1.35 0.05 0.18
1.58 1.22 0.03 0.96
1.64 1.59 0.08 0.83
11.99 1.33 0.66 2.12
0.001 n.s. n.s. n.s.
Notes: *df= 3/129; **df= 3/62.
what information is relevant to the task, and they know better how to package that information into a linguistic structure. If the growth in percent and quality of formal definitions reflects a growing ability to assess the adequacy of one's own speech, and to adjust to the needs of the listener, then we would expect that the communicativeadequacy scores would increase with age as well. As can be seen from table 5.1, C A . scores in English do show a significant increase with grade, though even the fifth graders' scores are not yet close to the highest possible rating of 4. On the other hand, there are no grade-related changes in conversational features (gestures, linguistic fillers, etc.), the incidence of which remains quite low across the entire age range observed (see table 5.1). The low level of conversational features suggests that even the youngest children understood that this task was meant as a formal, non-conversational task. Nonetheless, as we shall see below, the presence of conversational features related to other aspects of the children's performance on this and other tasks. Home versus school sources of definitional skill Since definitions are a school task, it makes sense that skill in giving definitions should relate to one's school experience more strongly than to one's home experience. Indeed, this is precisely what previous analyses have revealed (Snow, 1990). A multiple regression analysis showed that exposure to English at school explained considerably more of the variance on English formal definition scores than did exposure to English at home. Furthermore, exposure to French at home explained none of the variance on formal definition scores in French (Snow, 1990). These results recall the findings of Malakoff (1988) and Johnson (this volume), who also show that school exposure to a language enables children to perform well on school-like tasks in that language, even for children whose overall proficiency in the language is relatively low. One wonders, then, whether it is ever possible to achieve high levels 97
Catherine E. Snow et al. of performance on a task like definitions in a language where one has had relatively little exposure to them. Many notions about second-language acquisition suggest this should not be difficult - that knowledge from an earlier learned language about the need for analysis of content and of form can be transferred quite easily to a second language (Cummins et al., 1984; Mace-Matluck, 1979; Snow, 1987). The hypothesis that such knowledge can be easily transferred from a stronger to a weaker language is not supported by the relatively low and sometimes negative correlations obtained between definitions scores in English and French for this sample (Snow, 1990), though very high cross-language correlations were obtained for a sample of good bilinguals who had received schooling in both French and English (Davidson, Kline, & Snow, 1986) and considerably higher correlations have been found in Puerto Rican Spanish-English bilinguals attending school in both Spanish and English (Velasco & Snow, in preparation). These results exemplify the impact of language of instruction on the development of decontextualized language skills. A more careful breakdown of the seventy-one subjects who provided definitions in French in this study confirmed the earlier finding that exposure to French in the home is not necessarily related to higher scores on French definitions. Furthermore, of the forty-nine children who had minimal or no home exposure to French, a small group (n = 9) were identified by their French teachers as excellent school learners of French. These children achieved very high French definitions scores, almost equivalent to their own scores in English (see table 5.2). For this subgroup, then, the transfer from the stronger to the weaker language seems to have been successful. These findings suggest that transfer of the analysis skills that underlie giving good definitions may simply not be possible before a certain, evidently fairly high, level of L2 proficiency is achieved. Furthermore, the children who are relatively quite advanced in the analysis skills that are needed to give good definitions may be those children who are best able to cope with the academic demands of foreign-language study (a notoriously difficult task for elementary-school children). Of particular interest, six of the nine children identified as excellent school learners of French used a language other than French or English at home (e.g., Hungarian, Arabic, Portuguese, German); thus French was at least the third language they were learning. This suggests that their previous successful experience in learning a second language (English) may have been beneficial to them in their study of yet another language (French), perhaps by providing them with additional sources of support for the development of metalinguistic abilities. College students' success in foreign language study has been shown to relate to their level of metalinguistic skill in their first language (Webster, 1986). The French definitions scores of the "good school learners" may reflect the same relationship. 98
Giving formal definitions Table 5.2 Scores in English and French for children with home vs school exposure to French School
Home good learner
average learner
n = 16* n = U*
/i = 7
n = 99
AI = 9
H = 44
72% 69% 6.97 7.83 1.04 0.94 1.92 2.49 0.14 0.06 1.27 1.55
81% 70% 8.85 7.45 0.88 0.49 1.57 1.50 0.00 0.63 1.91 1.55
69% 71%
Measure Eng.: Fr.: % F.D. Eng. Fr. F.D.Q. Eng. Fr. F.D.S. Eng. Fr. I.D.Q. Eng. Fr. C.F. Eng. Fr. C.A. Eng. Fr.
7.86 5.52 1.06 0.60 2.47 1.18 0.04 1.05 1.42 1.35
P
n.s. n.s. n.s. 0.006 n.s. n.s. n.s. 0.02 0.0004 0.001 0.05 n.s.
Note: *n's for English and French means differ because a few of the children could not perform the task in English, and many could not perform it in French.
Bilingualism and definitional skill If giving definitions is a task which relates to metalinguistic skill in general, then we might expect it would also relate to degree of bilingualism, since there is ample evidence that bilingual children show precocious metalinguistic skills (Ben-Zeev, 1977; Bialystok, 1986; Cummins, 1978; Galambos & Hakuta, 1988; Ianco-Worrall, 1972; Leopold, 1939-1949; Slobin, 1978). We assessed this possibility by dividing the children into three categories: those who were minimally bilingual, having had exposure to a second language only through foreign-language classes; those who had had some exposure to and need for the use of a second language at home or in other schools, but who were not fully bilingual; those who were fully bilingual, that is, used some language in addition to English regularly and well. The fully bilingual group scored slightly below the other two groups on English F.D.Q., but also lowest in communicative adequacy, on which the partially bilingual group scored highest (see table 5.3). These differences were only marginally significant (p = 0.09 and 0.06 respectively). The minimally bilingual group used significantly more conversational 99
Catherine E. Snow et al. Table 5.3
Effects of status as a bilingual on scores in English and French
Measure Eng.: Fr.: % F.D. Eng.
Fr. F.D.Q. Eng.
Fr. F.D.S. Eng.
Fr. I.D.Q. Eng.
Fr. C.F. Eng.
Fr. C.A. Eng.
Fr.
Minimally bilingual
Partially bilingual
Fully bilingual
n = 49* n = 20*
n = 59 n = 34
Al = 1 6
71% 66%
69% 71%
68% 71%
7.99 5.42 1.04 0.49 2.32 1.23 0.04 1.40 1.38 1.44
7.82 6.43 0.99 0.77 2.36 1.67 0.04 0.54 1.52 1.48
6.98 6.75 1.03 0.68 1.80 1.78 0.10 0.48 1.23 1.43
P
* = 29 n.s. n.s. 0.09 n.s. n.s. n.s. n.s. n.s. n.s. 0.0001 0.06 n.s.
Note: * n's differ somewhat because many children tested in English could not be tested in French, and a few tested in French were not tested in English.
features in French (mean of 1.40, compared to 0.54 and 0.48 for the partially and fully bilingual groups; F(2,67) = 10.77,p<0.001); the fact that conversational features were more frequent in French but not in English suggests the effect is due to this group's lower proficiency in French rather than any direct effect of bilingualism. Thus, we have no strong support in these data for any proposed positive (or negative) consequence of bilingualism per se. It should be pointed out, though, that being bilingual is to some extent compounded in this sample (as it would be in almost any sample drawn in the U.S.) with being a non-native speaker of English. We have already noted above that English performance on formal definitions relates to school exposure to English - of which the native English speakers had more. To explore further the possibly subtle effects of multilingualism on performance in the definitions task, we also looked at the number of languages the children spoke, and at the amount of exposure to the most dominant language other than English or French. Neither of these classifications related in any significant way to scores on the definitions task.
Relations across measures One way of viewing language skill is to think of it as a global, undifferentiated capacity; such a view would predict that children who do well at 100
Giving formal definitions Table 5.4 scores
Within-language correlations for English and French definitions
% F.D. Eng. Fr. F.D.Q. Eng. Fr. F.D.S. Eng. Fr. I.D.Q. Eng. Fr. C.F. Eng. Fr. C.A. Eng. Fr.
% F.D.
F.D.Q.
0.55*** 0.32** 0.16* -0.11 -0.31*** -0.38*** -0.16* 0.10 0.40*** 0.03
0.15* 0.45*** -0.04 0.36** -0.10 -0.32** 0.56*** 0.33**
F.D.S.
I.D.Q.
0.35*** 0.49*** 0.14* 0.29*** -0.23* -0.23* 0.26** 0.06 0.18 0.06
C.F.
C.A.
0.03 0.07
Notes: *O.OOKp <0.05;**0.0009
Catherine E. Snow et al. Table 5.5 Between-language correlations for English and French definitions scores English % F.D. F.D.Q. F.D.S. I.D.Q. C.F. C.A.
French % F.D.
F.D.Q.
0.27* 0.13 -0.15 -0.07 -0.01 0.09
0.10 -0.03 -0.19 -0.05 0.03 0.11
F.D.S.
I.D.Q.
-0.01 -0.09 0.20* 0.16 0.08 -0.10
-0.02 -0.19 -0.07 -0.27* 0.05 -0.10
C.F.
C.A.
0.05 -0.08 0.04 0.07 0.12 0.40*** 0.06 0.28* -0.07 -0.03 0.02 0.28*
Notes: *0.001 ?<0.05; **0.0009
Relations across languages The correlations between scores in French and English definitions (table 5.5) were not very high, a result we attribute to the relatively low average level of proficiency in French of children in this sample. As a result, individual differences in French proficiency swamped the results, making cross language relationships difficult to identify. However, a more qualitative analysis of individual students' responses in French and English might reveal whether they were using the same strategies to define specific words and their translation equivalents. We present here data from a few individual children, who represent something of the range available in the entire sample. Christophe was a native German-speaker who was in the fifth grade at UNIS. He had been taking French for three years, and was an average French student, though his performance in English made clear his abilities to learn a second language. Christophe provided the following definition in English for bicycle. Uh-it's . . . something with two wheels, and you can steer the front wheel, and there're many kinds, there're racing bikes . . . and you can travel from one place to another, you have to pedal.
In French, he was hindered by lack of relevant vocabulary, but nonetheless achieved an equivalent level of sophistication in terms of superordinate selected and the crucial bits of information (two wheels and to ride): C'est quelque chose pour monter et umm ... c'est quelque chose avec deux . . . umm, roues? et il y a un de ... tu fais ride. A harder word, alphabet, was treated almost identically in both languages: 102
Giving formal definitions It's all the letters in the [giggle] all the letters ... that... in the language. C'est des lettres . . . c'est toutes les lettres. Similarly, the attempts at thief are distinguished only by the presence of a bit more detail in English: It's a person [giggle] that comes in your house mostly through the window and steals things. C'est une personne qui entre dans la maison et vole quelque chose. The extra difficulties introduced by being asked to define an adjective, favorite, become equally evident in Christophers stronger language, English, and in French: Favorite is when you have a favorite person or someone you like the best or something you like the best. Favori? Ton livre favori? Tu aimes beaucoup. In English, Christophe is reduced for favorite to the more immature "x is when you . . . " form, and in French he seems to associate the word favori primarily with one referent - favorite books. These responses clearly remain much more tied to frequently encountered uses of the term, not influenced by analysis of those uses to establish a core meaning. Beatrice, a native Portuguese speaker in fifth grade at UNIS, was one of the group of excellent school learners of French. Like Christophe, she tended to use similar strategies in giving definitions across the two languages. A knife is something that you cut with ... if you pi- if small children play with a knife, they might cut their finger. Un couteau c'est quelque chose qui coupe, f a peut couper de la viande, le poulet - quelque chose comme 9a. Interestingly, she gave a higher proportion of informal than formal definitions even in English (her stronger second language), though with hesitations and dysfluencies characteristic of self-monitoring. Despite the fact that she gave a more sophisticated definition for favorite in French than in English, her definitions showed the expected relationship - more detail and more self-monitoring in English: Favorite means something ... like, I-I have - my favorite animal is a dog, it's a-one you like the best. Favori c'est une chose que tu aimes beaucoup, c'est mon favori animal. Beatrice had achieved a remarkably high level of fluency in French, but still in general showed more attention to the form of her speech in English, as one would expect. Nicole was a third grader who spoke French at home with her parents. As one would expect, then, the amount of detail in her French definitions 103
Catherine E. Snow et al. was greater than in her English definitions, but the crucial, definitional information tended to be identical in both. Her definitional style was in general rather terse: Knife is something that you cut with. Cest quelque chose que tu coupes avec, avec . . . comme tu coupes decoupes du viande, et tout 9a. Umbrella's what when it's rains you put over your head. Cest quand il pleut tu . . . il faut que tu le prends. A nail is something that you could screw in the wall and put a painting on, put a picture on it, Cest quelque chose quand tu vas . . . euh . . . veux mettre quelque chose sur une mur, tu vas mettre ce dessus, et apres tu vas . . . euh . . . mets ce que tu veux mettre sur le mur.
These comparisons show that, although the children in both the tested languages tended to reveal a similar degree of analysis of their own knowledge, their actual performance is greatly affected by ease of retrieval and processing in the target language. Although the children who speak French at home might have an advantage retrieving certain kinds of words and structures in French, nonetheless the forms specifically required for definitions - superordinates, generic constructions - seem not to be more easily accessible to them in French than in English. As a result, their performance in French was superior only for informal definitions, not for the quality of formal definitions.
Conclusions
The data we have reported here and elsewhere (Davidson, Kline, & Snow, 1986; Snow, 1990; Snow, Cancino, Gonzalez, & Shriberg, 1989) about the development of definitions suggest the range of accomplishments represented by the good formal definition. To provide a formal definition, children must first analyze the task situation in which they find themselves, decide that a formal definition is required, and recall the form peculiar to this particular genre; then they must reflect on the meaning of a target word, analyze what they know about that word in order to decide what is central to its meaning, and to organize that crucial information into the standard form for the definitional genre. At every stage of this process analysis is crucial - analysis of the situation first, of the genre-specific demands second, and then of one's own lexical knowledge. The centrality of this process of analyzing knowledge, rather than simply communicating or in other ways acting upon it, is the reason we feel definitions reflect a good deal of metalinguistic as well as linguistic ability. 104
Giving formal definitions Relationships between definitions and other language skills It is perhaps illuminating to go a bit beyond the data on the UNIS children's production of definitions, to consider how they performed on other language tasks in English and in French. In addition to the definitions task, we asked the children to provide picture descriptions under varying conditions (Ricard & Snow, 1990), and to interview the adult for four minutes, as if for a television talk-show (Snow & Dolbear, 1988; Schley & Snow, 1990). This latter task was designed to elicit information about the children's conversational abilities. The talk produced during each of these other tasks was extensively analyzed for a variety of linguistic characteristics. Very few of the coded characteristics showed any relation to performance on the definitions task. However, the production of topic starters in the talk-show task was highly correlated with the use of conversational features during the definitions. A sophisticated use of topic starters was predictive of quality of performance in the talk-show, specifically the ability to introduce new topics, and yet maintain the flow of conversation by not overusing this skill. But in the definitions task, a high proportion of conversational features detracted from the child's formal definitional skill. Thus we see that the specific problemsolving strategies a child adopts may be highly successful or not, depending on the context of interaction (e.g., formal classroom language vs. casual exchange) and the decision made by the adult concerning how the task is to be defined. For example, Beatrice scored low for use of conversational features in the definitions task, and introduced new topics judiciously in the talk-show task after following up on her partner's responses. Beatrice: Qu'est-ce que tu aimes faire? Experimenter: Oh - toutes sortes des choses ... j'aime lire, ecrire des histoires, j'aime . . . suivre des cours de danse, jouer au squash, marcher dans la ville . . . Beatrice: Qu'est-ce que c'est "squash"? Experimenter: Squash c'est un peu comme le tennis, mais ga se joue dans une petite salle, ... et vous frappez, chacun a son tour, le bal contre le mur. Beatrice: C'est difficile? Experimenter: Oui - qa se joue rapidement - enfin - si on joue bien. Beatrice: D'accord. Si tu £tais dans une bibliotheque, quelle partie de livres est-ce que tu choisis? Experimenter: C'est une tres bonne question,... Beatrice followed up on the experimenter's responses to her questions several times before initiating the new topic about books. By contrast, Eleanor, a native English-speaking fifth grader, overuses topic starters in her French interview, and her definitions show correspond105
Catherine E. Snow et al. ingly low formal skill: Eleanor: Experimenter: Eleanor: Experimenter: Eleanor: Experimenter: Eleanor: Experimenter:
Est-ce que tu aimes lire? Oui, j'aime beaucoup lire. Est-ce que tu regardes beaucoup la television? Euh - je la regarde de temps en temps. Est-ce que tu fais la gymnastique? Non - je ne fais pas de gymnastique. Je ne pourrais pas! Oh, Est-ce que tu aimes la cuisine chinoise? Oui, j'aime bien la cuisine chinoise.
She provides very informal definitions of chapeau and alphabet. Chapeau: Alphabet:
C'est pour mettre sur la tete. Ben - c'est les lettres.
However, in English, her interview involves more follow-up to her partner's comments, and her definitions are more formal. Eleanor: Experimenter: Eleanor: Experimenter:
Do you like animals? Yes I like animals. Do you have any as a pet? Yes, well not right now, but we've always had pets, dogs and cats. Eleanor: Okay, um . . . what's your favorite animal? Experimenter: Oh - for the house? Or favorite animal? Eleanor: Any animal. Experimenter: Well, as far as an animal to be around, I do like dogs. Eleanor: Okay. Um . . . do you like to play . . . with other people?
And her definitions for hat and alphabet in English are strikingly more formal than were her definitions for the same words in French: Hat: Alphabet:
It's something you can wear, in the cold, or even when it rains, and it usually has round edges. The alphabet means . . . um um . . . it's a group of letters that you - that means . . . well, it makes up words, if you put them - some together . . . and . . . it's usually for kids that start in kindergarten.
Christophe also has high definitional skills, and makes some attempt in both English and French to maintain a topic for several exchanges before introducing a new one, despite his evident struggle with French. Christophe: Experimenter: Christophe: Experimenter:
106
What did you do before you did . . . the movie? Well, I used to interview children. And you like to do that, too? Oh, I like to do that very much. In fact I was really thinking of when I sort - when this . . . movie business stops, I'll probably go back to do that again. 'Cause I enjoy that . . . I did that a long long time.
Giving formal definitions Christophe: And you think it'll stop soon, the movie business? Experimenter: Maybe I'll - another year, I'll do it, and then that'll be it. Yes. Mm . . . and . . . Interviewing children, you do the higher Christophe: grades, the lower grades? Experimenter: Well, usually the lower - usually elementary school, so that is from first grade to the sixth grade, usually? Yes, uh-hm. Christophe: And you do each child, or only a few? Experimenter: Well, usually I do a few. I will do . . . maybe twenty . . . in a school. Christophe: You pick out special ones, or just by random? Experimenter: Usually by random. Usually by random. Yes, it's important, when you do work like this, to pick out things by random. Christophe: And you . . . what d'you do with them? Experimenter: Well, we're interested in seeing how children use their language, and how well they can talk to people, how they can converse, say, like we're doing! And uh . . . all sorts of things about language. Uh. And you do this f - if - you do this just f - to know Christophe: how much a - how they use the language? Experimenter: Yes, because as you know, people use language differently. Christophe: Uh-hm. Experimenter: Yes. Christophe: And you did that long, too? Experimenter: .Yes, I did that a very long time. I've done that for - oh my goodness, I would hate to tell you how long! The strongest relation between the picture-description task and performance on definitions occurred for incidence of hesitations in describing pictures, which correlated positively with the quality of formal definitions. This relationship is not surprising if we think back to the very hesitant quality of the best definitions cited above. Children who engage in selfmonitoring reduce the fluency of their output, but they indicate that they are adjusting their speech on-line to match an internal model. Thus, these children produced better formal definitions, and also fewer irrecoverable references in the picture descriptions. Eleanor, for example, produced a high proportion of formal definitions in English (see examples above), and exhibited quite a bit of disfluency in her decontextualized picture descriptions: Eleanor: It's like the other picture. There's a boy with a lamb, except he's holding it up in the air and playing with it. And the other boy's getting mad at the goat, cause it's trying to run away. And there's still farm houses in the background. And even farth - urn behind 107
Catherine E. Snow et al. the farmhouses, there's water and a boat, with . . . a small boat, and the looks very - one of - the boy . . . with the lamb running away from he looks very . . . angry, but the other boy is having fun with the lamb. However in French, where Eleanor did not produce such formal definitions, she is not as disfluent in her decontextualized picture description: Eleanor: II y a les - des garcons, qui pechent du lac, et un gargon attrape un poisson, et l'autre reste sur un rocher. II y a des - how do you say mountains? Des montagnes um a cote des gargons, il y a d'her d'her - grass d'herbe, et est-ce que je peux dire qu'est-ce qu'ils portent? Um un gargon porte des chaussures noires avec des blue jeans et um, une chemise et um, l'autre gargon porte un chapeau avec un jacket et un blue jeans, et il a les chaussures bruns. Christophe, formal in both his French and English definitions, produces picture descriptions that are fairly disfluent: English: In the front, there's a grass hill, and there're two boys, one boy has a . . . yellow coat . . . and a a hat . . . with black stripes, blue - bluejeans and a d - a white dog just ran away from him down the hill. And the other boy . . . ha - is lying on the grass, he has ye - uh yellow hair, has um a white sweater with blue stripes, with black stripes, and nn bluejeans, yellow shoes, and he's playing with a dog. And next to him is a bo - a bottle, a drinking bottle, and nn and in bef - front of the two boys there are two houses, one with a green roof and . . . chim - chimney on top, and is yellow, the house that's yellow has two windows on one side, and you can see only two windows on the other side. And the other house has a red roof, wh - is white, and has two windows on one side and you can see three windows on the other side; and then there're two more houses that you can only see a part of. One of them has a red roof and is yellow, and the other one you can only see in the shadow. And there're people running . . . in front of the houses, to the boat, I think, and . . . uh . . . let's see . . . you can't see what things they're wearing, and . . . they . . . and the boy with the yellow coat h a s . . . has black hair, and there's a pole which the dog is attached to that's running away. French: II y a deux enfants, il va - et une gargon a une poisson et il tombe dans l'eau, et il a des pantalons bleus, l'autre gargon a aussi des pantalons bleus, mmm, et il y a une chapeau, et il - il y a des il y a un gargon qui tombe dans l'eau, il a une une petite poisson. L'autre gargon est content, et il y a un tricot jaune et il y a des [unintelligible] et le gargon qui tombe dans l'eau a un tricot blanche, avec des avec un peu de - de noir - . II a aussi des cheveux jaune, mmm, et l'autre gargon a des cheveux noirs, et et des - non, sur le tricot jaune il y a un autre tricot rouge, et il n'a pas de poissons. 108
Giving formal definitions Je ne sais pas le mot ... Mountains. II y a beaucoup de montagnes, etc'esttout. The children's performances in English and French showed a relationship in the degree of decontextualization and disfluency in both the definitions and picture-description tasks. A tendency to use conversational features, which is negatively related to skill at formal definitions and at decontextualized picture descriptions, does in fact assist a child in the talkshow task. Contexts for acquisition The lack of global relationships between levels of performance on quite different tasks emphasizes again the need to analyze contexts for acquisition if we are to understand the kinds of skills children acquire in a second language. Although skills already acquired in afirstlanguage can ultimately be transferred to a second language, this transfer evidently does not occur until a relatively high level of proficiency in the second language is acquired. It is promoted, though, if the contexts to use any specific skill are provided in the second-language environment; indeed, if the second-language context requires it, language skills can be acquired in L2, even at an early stage in the acquisition of L2, that surpass similar skills in LI. Most children at UNIS were achieving proficiency in French via exposure in second-language classrooms, and showing transfer to French of skills acquired in the curricular language, English. But similar transfer of skills is found when looking at children tested in a second language that is also a curricular language. Malakoff (1988) exemplifies this with French-English bilinguals (in either French or English programs) at the International School of Geneva, Switzerland. In solving difficult analogies, the children's performance was faster when the language of presentation matched their language of school instruction. For easy analogies, students in a French curriculum performed the task faster when the analogy was presented in English (their native language); however, with difficult analogies, they performed faster with the analogy presented in French (their curricular language). Where there is greater cognitive demand, the language imbalance is more visible. The language of instruction thus enhances decontextualized skills in solving a challenging language task, when being tested in a curricular (but not native) language. Using a metaphor task, Johnson (1989, this volume) similarly studied bilingual children whose second language was a curricular language rather than a language learned in a foreign-language context. Comparing three groups of subjects living in Toronto, Canada (monolingual English subjects, Spanish speakers with less than three years of residence, and Spanish speakers with at least five years of residence), Johnson found that per109
Catherine E. Snow et al. formance in interpreting metaphors was related to conceptual knowledge rather than specific language capacity. Her bilingual subjects were already quite proficient in their second language, with Spanish as their main home language, and English as their curricular language. Children were tested both in Spanish and English in a metaphor-interpretation task. The performance of the older children was higher than that of the younger children in both languages. However, the increase was less extreme in Spanish. The older children were more proficient in English, having studied in an English curriculum for longer, and thus, the difference between the English and Spanish scores was greater in the sample of older children than it was in the sample of younger children. A full understanding of oral-language proficiency and of literacy skills in a second language depends on an analysis of the context in which the language is acquired. The foreign-language classroom promotes transfer between a native language and a second language, but only once a solid level of proficiency has been attained. The native language (used in the home) certainly enhances conversational skills, but does not appear to directly enhance those language skills that are called upon in the classroom context. It is these latter classroom skills (e.g., giving formal definitions, decontextualized picture descriptions, etc.) that are most predictive of successful literacy and school achievement. More information about the characteristics of the language used in the home and school environments will provide us with a greater understanding of the sources of decontextualized language skills, and the reasons for discrepancies between skills in languages learned exclusively at home, and languages learned in a curricular or foreign-language classroom environment. Notes 1.
2.
We would like to express our appreciation to the Spencer Foundation, which funded the collection of the data reported on here; to the Office for Educational Research and Improvement, which funded the analyses and writing up through the Center for Language Education and Research; to Susan Kline and Paulina Gonzalez, who tested the children in French; to Loriana Novoa, who carried out statistical analysis; and to the staff, faculty, parents, and children of the United Nations International School, who co-operated willingly in the research. Herlinda Cancino's current address is Stanford Graduate School of Education, Stanford, CA 94305. An alternative way of categorizing young children's definitions is that they incorporate scriptal knowledge, as in the classic "A hole is to dig." All the definitions in A hole is to dig (Krauss, 1952) reflect scripted uses of the definienda, e.g., a mustache is to wear on Halloween, a castle is to build in the sand, dishes are to do, and pertinently, a hat is to wear on a train. 110
Giving formal definitions References Ben-Zeev, S. 1977. The influence of bilingualism on cognitive strategy and cognitive development. Child Development, 48,1009-1018 Bialystok, E. 1986. Factors in the growth of linguistic awareness. Child Development, 57,498-510. Bialystok, E. & E. Ryan 1985. Toward a definition of metalinguistic skill. MerrillPalmer Quarterly, 31, 229-251 Cummins, J. 1978. Metalinguistic development of children in bilingual education programs: data from Irish and Canadian Ukrainian-English programs. In M. Paradis (ed), The fourth Lacus Forum. Columbia, SC: Hornbeam. Cummins, J., M. Swain, K. Nakajima, J. Handscombe, D. Green, & C. Tran 1984. Linguistic interdependence among Japanese and Vietnamese immigrant students. In C. Rivera (ed.), Communicative competence approaches to language proficiency assessment: research and application. Clevedon, Avon, England: Multilingual Matters. Davidson, R., S. Kline, & C. E. Snow 1986. Definitions and definite noun phrases: indicators of children's decontextualized language skills. Journal of Research in Childhood Education, 1, 37-48. Galambos, S. & K. Hakuta 1988. Subject-specific and task-specific characteristics of metalinguistic awareness in bilingual children. Applied Psycholinguistics, 9,141-163. Hagtvet, B. E. 1986. " 'Dress' is birthday" and " 'Sleep' is night." Young children's definitional strategies in reacting to word definition tasks. Unpublished manuscript. Norwegian Institute of Special Education, Granasen 4, 1347 Hosle, Oslo, Norway. Ianco-Worrall, A. D. 1972. Bilingualism and cognitive development. Child Development, 42,1390-1400. Johnson, J. 1989. Factors related to cross-language transfer and metaphor interpretation in bilingual children. Applied Psycholinguistics, 10,157-177. Krauss, R. 1952. A hole is to dig: a first book of first definitions. New York: Harper and Row. Leopold, W. F. 1939-1949. Speech development of a bilingual child: a linguist's record, 4 vols. Evanston, IL: Northwestern University Press. Litowitz, B. 1977. Learning to make definitions. Journal of Child Language, 4, 289-304. Mace-Matluck, B. J. 1979. Order of acquisition: same or different in first- and second-language learning? The Reading Teacher, 32, 696-703. Malakoff, M. 1988. The effect of language of instruction on reasoning in bilingual children. Applied Psycholinguistics, 9,17-38. Markowitz, J. & S. Franz in press. The development of defining style. International Journal of Lexicography, 1. Miller, G. & P. Gildea 1985. How to misread a dictionary. AILA Bulletin, Pisa. Ricard, R. J., & C. E. Snow 1990. Language use in and out of context: evidence from children's picture descriptions. Journal of Applied Developmental Psychology, 11,251-266. Ill
Catherine E. Snow et al. Schley, S., & C. E. Snow 1990. The conversational skills of school-aged children. Unpublished manuscript, Harvard University, Cambridge, MA. Slobin, D. I. 1978. A case study of early language awareness. In A. Sinclair, R. J. Jarvella, & W. J. M. Levelt (eds.), The child's conception of language. Berlin: Springer. Snow, C. E. 1983. Literacy and language: relationships during the preschool years. Harvard Educational Review, 53,165-189. 1987. Beyond conversation: second language learners' acquisition of description and explanation. In J. Lantolf & A. Labarca (eds.), Research in second language learning: focus on the classroom. Norwood, NJ: Ablex. 1990. The development of definitional skill. Journal of Child Language, 17, 697-710. Snow, C. E., H. Cancino, P. Gonzalez, & E. Shriberg 1989. Giving formal definitions: an oral language correlate of school literacy. In D. Bloome (ed.), Literacy and classrooms, Norwood, NJ: Ablex. Snow, C. E. & M. Dolbear 1988. The relations of conversational skill to language proficiency in second language learners. Unpublished manuscript. Harvard University, Graduate School of Education, Cambridge, MA. Tapia-Uribe, F. M. 1988. Women's schooling, fertility, and child survival in a Mexican village. Unpublished doctoral dissertation, Harvard Graduate School of Education, Cambridge, MA. Velasco, P. & C. E. Snow in preparation. Bilingual children's oral decontextualized language skills and reading comprehension. Unpublished manuscript. Harvard University, Cambridge, MA. Watson, R. 1985. Towards a theory of definition. Journal of Child Language, 12, 181-197. Watson R. & D. Olson 1987. From meaning to definition: a literate bias on the structure of word meaning. In R. Horowitz & J. Samuels (eds.), Comprehending oral and written language. Orlando: Academic Press. Webster, M. 1986. Relationship between metalinguistic ability and second language learning. Unpublished doctoral dissertation, Harvard Graduate School of Education, Cambridge, MA.
112
6
Metalinguistic dimensions of bilingual language proficiency ELLEN BIALYSTOK
The [bilingual] child learns to see his language as one particular system among many, to view its phenomena under more general categories, and this leads to awareness of his linguistic operations. (Vygotsky, 1962, p. 110)
Why would we expect bilingual children to display different metalinguistic abilities from monolingual children? If language proficiency, and, in particular, oral-language proficiency, is construed as a universal achievement of all members of the species, governed by innate mechanisms that assure its propitious evolution from the first words heard and spoken, then there is little basis for expecting significant variation in speakers' knowledge of language as a function of simply how many linguistic systems they have mastered. But language proficiency is not simply a skill or procedure which is mastered independently of other forms of knowledge and other types of abilities. Language is also a logical symbolic system, capable of being known itself and capable of guiding and shaping other aspects of cognition. Entry into these uses of language that involve its own properties and structure is more or less what researchers mean by metalinguistic awareness. It is because metalinguistic aspects of language are not necessarily specific to particular languages that their discovery may be influenced by the mastery of two languages, and it is because metalinguistic awareness is consequential for other aspects of cognition, both linguistic and non-linguistic, that its study is important. This chapter will begin by defining what metalinguistic might mean operationally and reviewing some of the constraints that have been proposed to set metalinguistic ability apart from linguistic ability. The conception of metalinguistic ability adopted in our own research will be outlined. The conclusion of this section will be that metalinguistic is best defined in terms of the operations necessary to solve a set of tasks. In the second section, language use will be considered in terms of three broad task domains oral, literate, and metalinguistic. Within each, the processes necessary to operate in a variety of tasks are discussed. Learners can perform the tasks 113
Ellen Bialystok in the different domains if they have the required level of processing skill for those operations. In this way, there is no particular ability that determines that the child will be able to solve metalinguistic problems, as the processing skills required for these are the same as the processing skills required for some of the problems in the oral and literate domains. If monolingual and bilingual children develop these processing skills in different ways or at different rates, then there will be differences in all three domains of language use and not simply in metalinguistic tasks. The third section will further consider some of the ways in which these differences in underlying processing skills may lead bilingual children to learn and process language differently from monolingual children.
Metalinguistic aspects of language
The term metalinguistic is used broadly in scope and application. The literature is replete with arguments concerning the proper domain of language skills that are metalinguistic as opposed to purely linguistic, the nature of the evidence that is considered decisive in attributing these skills to language learners, and the age at which children can potentially begin to develop such skills. Many of these arguments are vitiated by the fact that in some cases researchers arguing different positions are simply not talking about the same thing: what is metalinguistic for one is part of ordinary language use for another. One example of such a contentious behavior is the ability to self-correct or repair oral speech. On some accounts this is metalinguistic because it is an operation on language forms, considered on some views to be the determining feature of metalinguistic performance. On other accounts it is simply part of the normal performance of conversation, not distinguished by any particular awareness by the speaker nor any explicit consideration of language forms. Since very young children repair speech, the attribution of metalinguistic abilities to children depends on whether or not such behaviors are to be considered metalinguistic. A further confusion in the application of the term metalinguistic is that it is used interchangeably to apply to tasks, skills and levels of awareness. When applied to tasks, certain uses of language, such as making repairs or judgments about well-formedness, are classified as being metalinguistic or not. A learner performing a task classified as metalinguistic is demonstrating metalinguistic ability (see Clark, 1978). When the term is applied to skills, learners are considered to have developed metalinguistic awareness when they have demonstrated an ability to focus on language forms, irrespective of any particular task for assessing this skill. The term metalinguistic in this case is describing a mental state of the learner and not an aspect of language performance (see Lundberg, 1978). Applied to levels 114
Metalinguistic dimensions of awareness, virtually any performance can have metalinguistic properties if it is carried out with the deliberate control and awareness of the language learner. On this view, it is not sufficient to merely perform a metalinguistic task (such as repair) nor have demonstrated an independent ability to focus on language forms, but performance must unequivocally indicate that the learner is aware of the forms and functions of the language being manipulated (see Bowey, 1988). These three applications are sometimes signaled by the different terms metalinguistic task, metalinguistic ability, and metalinguistic awareness. The problem is that these terms are not used systematically to distinguish among the three applications. Nor has there been any attempt to consider the relations among them. Rather, they are applied loosely to the variety of behaviors and abilities described above. Some definitions combine them. In Vygotsky's statement, the language learner was led ultimately to an awareness of linguistic operations, invoking both the ability and awareness interpretations. With so much latitude in its interpretation, what is the essence of metalinguistic? Does it make sense to distinguish metalinguistic from non-metalinguistic uses of language? Without a clear consensus for the proper application of the term, the issue of possible differences among children in their metalinguistic skills as a function of bilingualism becomes unexaminable. In order to consider metalinguistic abilities of bilingual children, therefore, a more precise definition for what the metalinguistic function entails must be explicated. The remainder of this section will outline a conception of metalinguistic that involves all three aspects identified above: characteristics of the task, abilities of the learner, and properties of awareness. In the next section, this conception of metalinguistic will be applied to the language development of bilingual children. The basis of the explanation for metalinguistic abilities is an informationprocessing description of cognition. On these views, systematic processes are applied to organized mental representations. This is the central notion that distinguishes information-processing explanations from other approaches. A current alternative to this paradigm is parallel distributed processing, or connectionism (e.g., McClelland & Rumelhart, 1986; Rumelhart & McClelland, 1986). These models assume a fundamentally different architecture for the mind and offer radically different accounts of learning and cognition (see Fodor & Pylyshyn, 1988, for a comparison of information processing and connectionist accounts of cognition). On these views, processing is explained by patterns of activation that spread rapidly through the mental networks. No higher-order organization is involved in learning and using language. Without such an organizing structure, it is difficult to imagine how constructs like metalinguistic would be defined. 115
Ellen Bialystok Using the perspective of information processing, the assumption is that language is assigned a mental representation, and using language involves the application of identifiable processes to those representations. Descriptions of both the form and structure of the representation and aspects of the processes are necessary to characterize a cognitive or linguistic ability. Following information-processing conceptions, then, metalinguistic performance will be defined by reference to both the underlying mental representations involved in these problems and the cognitive operations that are carried out in the solution to tasks. In addition to these general properties shared by information-processing models, the present model has one further constraint. The model is explicitly addressed to issues of development. Hence the central function of the model is to account for change. How do the representations of linguistic knowledge and the operations applied to those representations evolve? Language proficiency will be described as a function of the level of development of the linguistic representation and the particular cognitive processes applied to those representations. The central feature of the model is that language processing is described in terms of two components. Each component bears responsibility for one aspect of language processing and its development. The first responsibility is the nature of the learner's mental representation of language. Mental representations evolve by becoming more structured, more explicit, and more interconnected. The processing component by which these changes occur is called analysis of linguistic knowledge. The second responsibility is the attention function, which selects information from a representation and directs attention to that information for the purpose of performing in a specific context. Selective attention improves with advances in proficiency, resulting in more fluent and more accurate performance. The processing component by which this selective attention is guided is called control of linguistic processing. The two processing components are considered to be the mechanisms by which language proficiency improves through age, experience, and instruction. They are also the mechanisms which are responsible for a language learner's ability to carry out various language functions. There are, of course, a large number of processes implicated in a task as complex as learning and using language. Some of these are the "fast processes" (Jackendoff, 1987) or "automatic processes" (Schneider & Shiffrin, 1977) that are involved in low-level aspects of language processing, functions such as letter and sound recognition and some aspects of semantic interpretation. These processes are different in nature from those examined here and will not be considered in the present model. The fast processes are essentially inaccessible to study and impervious to development. The study of these fast processes also raises different issues from the ones considered 116
Metalinguistic dimensions here, issues such as the role of modular and central processes on performance. The present account is neutral with respect to these issues. Representation of linguistic knowledge The first processing component is the progressive analysis or restructuring of the mental representations of language. In an excellent review of the literature on representation and its role in children's cognitive development, Mandler (1983) distinguishes between two meanings for the term representation in the psychological literature. The first meaning refers to knowledge and the way in which it is organized. This is the problem of knowledge and its structure. The second meaning refers to the use of symbols. This is the sense in which representations refer to words, artifacts, or concepts. This latter meaning is more complex in that representation involves a relation between a symbol and its referent. She also points out that these two meanings correspond more or less to Piaget's distinction between "representation in the broad sense" (the first meaning) and "representation in the narrow sense" (the second meaning). The difference between these two meanings stands at the core of what is involved in metalinguistic awareness of language and, crucially, provides a means of distinguishing it from language proficiency in its more usual sense. Representations of language in the broad sense are organized around meanings and uses. In this way, language is a set of semantic relations that organize our knowledge of the world. This is the structure of language that was explored in early psycholinguistic studies of the representations of language. In these studies, reaction-time paradigms were used to determine the nature of semantic connections between various linguistic terms, such as bird and robin, robin and sparrow or bird and animal. Various proposals for semantic structure followed from these studies, such as network theories (Collins & Quillian, 1969; Smith, Shoben, & Rips, 1974), relational theories (Rumelhart, Lindsay, & Norman, 1972), and prototype theories (Rosch, 1975). Linguistic representations described in this way are sufficient for most uses of language. The learner may or may not be aware of these representations and their organization for them to support activities such as conversation, some interpretation of text, and a variety of semantic judgments. Metalinguistic uses of language are based on representations of the second kind. To solve metalinguistic problems, the underlying mental representation must be organized around forms and structures and specifically indicate the nature of the symbolic relation, or how it is that the forms refer to meanings. These representations, then, differ in two respects from the representations that are sufficient for most language functions: first, they are organized around formal symbolic categories and not semantic 117
Ellen Bialystok or empirical categories; and second, the structure of those categories is explicit. Representations of this type are called analyzed representations of linguistic knowledge. Language learners constantly reorganize their knowledge of the linguistic system to increasingly explicate its formal structure. The result of this process of analysis is that new analyzed symbolic representations emerge out of the non-analyzed or semantic ones. An example of how a domain of linguistic knowledge can become restructured from being organized around semantic notions to being organized around more formal ones is provided by Karmiloff-Smith (1986). She uses the example of the acquisition of the indefinite article in French (un). At first, children produce this form in contexts requiring non-specific reference, the numeral "one," and the appellative function, all appropriate uses for the form. Her claim, however, is that at this point the form un is simply represented in each of these semantic contexts with no means of connecting them to each other. At some point in development, these separate representations are reorganized so that the individual forms that are structurally the same are linked together under the formal category indefinite article. This is the process of analysis. The semantically based representation that links the form to contexts via meaning relations is not lost; rather, the learner has added to the knowledge of language a new level of organization based on formal structure. The sequence of stages from unanalyzed semantic representations through to analyzed symbolic representations actually includes several intermediate positions. There are at least three identifiable stages in this process. Between the unanalyzed representations of meanings and the symbolic representations of language forms is a stage of explicit knowledge of forms that lacks specification for the way in which those forms refer to objects, meanings, or other symbols. These representations, then, are explicit, and even metalinguistic in the sense that they are about language, but they are not symbolic. As learners gain proficiency in a language, whether it is the first language being acquired by a child or a subsequent language being mastered by an adult, the impression of that learner's progress with analysis is that of a continuum, not discrete stages. Learners build up their representations for language bit by bit, and different aspects of the linguistic system can be represented at different levels of analysis. Thus, young children may have symbolic representations for the letter-to-sound correspondences that constitute our writing system, explicit formal representations for the form class of pronouns (without understanding the class of meanings to which they refer), and unanalyzed semantic representations for word-order constraints. In time, each of these will become analyzed to a higher level. But because different aspects of the system are at different stages in this process, the general classification of that learner's level of analysis is in 118
Metalinguistic dimensions some way an average of all these stages of representation. In this way, it appears that learners move smoothly and progressively through to higher levels of analysis. For the present discussion, analysis will be considered in this way as a continuum, and the simplifying assumption will be that learners can be located along that continuum as a general reference to their level of analysis. Thus, reference will be made to relative degrees of analysis without specifying absolute levels. The important consequence of representational change through the process of analysis is that some applications of language require a more formal or a more analyzed representation than others. The development of language proficiency, then, includes as one of its essential components the continual analysis of linguistic knowledge to create representations based on formal or symbolic structure.
Attention to representations The second processing component that is claimed to be central to language proficiency is control over attention. This aspect of processing is the selective attention to different representations, or different aspects of a representation, carried out within real time. Processing is always selective, and control of linguistic processing is the process of selecting, with or without awareness, the information that will be attended to in the solution to a problem. In most information-processing models, attention is usually handled by some executive component that has a controlling function over other aspects of processing (e.g., Case, 1985; Sternberg & Powell, 1983). The present model attempts to escape the hierarchical structure of these executive systems by positing control over selective attention as a process at the same level of functioning as other processes, such as the analysis of representations. In this way, the process of analysis and the process of control over attention are intimately connected. The distinction, that is, is not between representation and process, but between analysis and control at low (meaning-related) levels of representation and analysis and control at higher (symbolic) levels of representation. The reason for rejecting a distinction between representation and process as category boundaries is that such a distinction functionally translates into a distinction between knowledge and skill. Discussion of the representation quickly slips into discussion of the knowledge, and consideration of process becomes synonymous with skill. A knowledge-skill distinction, however, is not the intention of analysis and control. The processing components of analysis and control both reflect aspects of both knowledge and skill. The process of analysis depends heavily on knowledge-based represen119
Ellen Bialystok tations but demands as well the analytic skill to restructure those representations at a higher level. The process of control depends heavily on the skill of directing attention, but such attention would be vacuous without an adequate knowledge-base, including a set of procedures or strategies for selecting an appropriate focus of attention. Hence the processing components are considered to exist at equivalent levels in some hierarchy of processes. Attention, then, is fundamental to language processing but is not superordinate to other essential processes. This conception of attention follows closely from the model of cognition proposed by Jackendoff (1987). He identifies two functions for attention: the selecting function and the directing function. The selecting function controls the level of detail that is activated in the representation for a given problem. This selecting function corresponds to (or depends on) the process of analysis having made the necessary structures and relations available. The directing function controls the attention to the specific information that is necessary for the solution. Different uses of language differ in the types of attentional strategies necessary for their execution. The simplest form of attention to linguistic representations is to pay attention to meanings. Indeed, Stroop (1935) showed that meaning was an irresistible aspect of language processing. When subjects were presented with ccrtor names written in different colors of ink and asked to pay attention only to the color of ink in which the word was written, the color named by the word interfered and made this problem difficult. More complex uses of language involve more complex or more demanding strategies for controlling attention, and the increasing ability to attend to linguistic representations in these ways is the developing process of control. As with analysis, learners' progress with control appears to proceed continuously. Learners obtain higher levels of control when they have mastered the appropriate attentional strategies for more difficult problems. They learn with experience what information is relevant to processing in certain situations. Reading, for example, requires an intricate balance of attention divided between forms and meanings. As these attentional strategies are learned, they allow the learner to engage in language uses that are more demanding in this way. Again, learners' relative progress with the mastery of such attentional strategies can be indicated by a position on a continuum.
The processes of language proficiency The implication of viewing language proficiency as the development of the two component processes corresponding to analysis of linguistic rep120
Metalinguistic dimensions resentations and control of processing is that language learners will advance through different language uses as they increasingly master these two processing components. There are two hypotheses that follow from this view regarding the development of metalinguistic skills of bilingual children. First, the ability to solve metalinguistic problems is not distinct from the ability to solve certain other problems in the oral and literate domains. Higher levels of analysis and control provide learners with access to more difficult problems, including metalinguistic ones. But those same abilities are the ones necessary for advanced uses of language in the oral and literate domains as well. Hence, if there are differences between monolingual and bilingual children with respect to their development of these two processing components, then this difference will be reflected not only in their solutions to metalinguistic problems but also in their solutions to certain oral and literacy problems. Second, if there are differences in the development of these two skill components for monolingual and bilingual children, there may be different relations evident for the development of each of them. Bilingual children, that is, may be advantaged relative to monolingual children in one of these processes but not the other. This arrangement would mean that differences between monolingual and bilingual children in language processing would need to be described in more detail in terms of the types of processes required by specific problems. Domains of language use
By positing (at least some degree of) independence for the two processing components, they can be represented as two orthogonal axes (see figure 6.1). The space created by this matrix can be used to quantify the demands of a variety of language uses on each of the two components. Each axis marks increments in the demands placed upon each processing component. Plotting a variety of tasks within this space allows a direct comparison of the processing difficulty presented by each task. Two tasks that may appear to present overall similar levels of complexity, or reside together in the same language-use domain, may be different from each other in their processing demands. Some tasks may rely heavily on high levels of analysis of knowledge while others make more severe demands on control of processing. Which tasks we expect learners to master depends on the learners' level of mastery of both analysis and control. The issue of bilingual differences must be examined from the point of view of both the processing demands of specific tasks under investigation, and the specific processing differences between bilingual and monolingual children. Overall assessments of proficiency would not supply this information. 121
Ellen Bialystok High control
metalinguistic
- • High analysis
Low analysis « literate
oral
Low control
Figure 6.1 Domains of language use
An alternative way of presenting the argument would have been to use the axes to plot levels of learner proficiency for each processing component rather than the demands made by each task. That organization, however, is less conducive to displaying development changes. Learners at a particular point in time would occupy a particular position in the matrix. When their skill with one of the processing components changes, a new position for that learner would need to be defined. In this way, a particular matrix could indicate the configuration of a learner's proficiency at only one point in time. Learners would move inevitably to higher levels of analysis and control, absorbing tasks on the way, but the relations among those tasks would not be clear. A matrix organized around learners' levels of proficiency would also fail to capture the point that learners of high ability who are solving tasks that require only low levels of a particular processing component perform only up to the level necessary for that task. Some language learners, for example, may have very high levels of both analysis and control and so a profile of their language proficiency would involve positions in the upper right quadrant. But when these same people were performing tasks that required only low levels of proficiency, such as a friendly conversation, there would be no way to indicate the difference, namely that processing in that case invokes only minimal levels of these components. The purpose of the chart is to represent task differences in order to
122
Metalinguistic dimensions account for the different processes they entail. There are even reversals when the diagrams are changed to plot positions of learner ability instead of levels of task demands. Second-language problems are depicted as demanding higher levels of skill than comparable problems in a first language because the demands placed upon performance for speakers in a second language are more taxing than those imposed for speakers carrying out the same function in a first language. Learners, however, actually have less skill in performing in the second language, so a chart based on learner ability would place the second-language tasks at a lower position. Positing levels of analysis along the x-axis and levels of control along the y-axis, the Cartesian space created by these axes indicates positions jointly described by the two skill components. These co-ordinate positions provide a means of representing developing language proficiency. Children's language acquisition generally progresses from oral-conversational uses of language, to the literate uses of reading and writing, and finally to the more metalinguistic uses required by tests and special types of problem solving. In most accounts of language proficiency, this progression is expressed as a unitary development. In the present framework, the progression is only apparently unitary; the child is in fact developing competence with two separate processing components. If the analysis and control demands for specific applications within each domain are averaged, then these averages occupy increasingly high positions in the Cartesian space. On average, that is, each domain places successively higher demands on each of the skill components, making each domain more difficult than the previous one. This ordering is illustrated in figure 6.1. No boundaries demarcate the division between domains, as these fade gradually one into the other, but the median position for each domain follows the pattern shown in the figure. Thus, the usual order of achievement for children learning their first language, moving from conversational through literate to metalinguistic uses of language, is attributed to the increasing and correlated demands of each of these domains on the components for analysis and control. Within each of the domains, a large range of tasks can be identified, and the full spectrum of tasks within a domain covers the majority of the matrix. Thus, in spite of the ordering of the general domains shown in figure 6.1, each domain can also be shown to present a range of problems to learners that differ in these demands. Mastery within a domain, then, can be conceptualized in the same way as mastery across domains: learners will initially be able to solve problems characterized by relatively lowdemand values for the processing components and will advance to more difficult problems with advances in their mastery of analysis and control. These relations are plotted in figures 6.2-6.4 below by identifying the relative matrix position of the skills involved in performing specific tasks. Speci123
Ellen Bialystok fie domains, such as metalinguistic, are not distinguished solely by virtue of the level of demand they place on the two processing components. The overlap across domains makes such distinctions facile. There is, rather, a dynamic interplay between the levels of processing incorporated by the two components, the nature of the task, and the domain in which it resides. Oral domain Although the oral domain of language use is usually the first to which learners have access, the domain is in no sense "primitive." High levels of oral use make demands on analysis and control that are as challenging as those found in the other two domains. There is a tendency for the most common oral uses of language to cluster around those activities which indeed make only minimal demands on the learners' levels of analysis and control. But these levels of processing skill, as repeatedly argued, cannot be defining for the domain. In addition to specific values for these processing skills, the oral domain is defined by the criterion of attention to meanings conveyed through spoken language. Usually there is interaction involved, but this too is not essential. Competence in oral uses of language involves the ability to express and interpret meanings as they develop through spoken discourse. Some examples of specific oral uses of language are plotted infigure6.2. Mastery of oral uses of language is often gauged by the criterion of fluency. Operationally, fluency needs to be defined in terms of processes. This approach to a definition of fluency substantiates the role of the two processing components in oral uses of language. Fillmore (1979) considers the problem of what constitutes fluency in oral uses of language and argues that the concept cannot be reduced to a single achievement. Rather, he claims, fluency takes different forms, and each is a different expression of a well-developed oral skill. One type of fluency is the ability to speak continuously, without pause, but without necessarily relying on any depth of form or meaning. This, he claims is the special achievement of disk jockeys. Game-show hosts may be another example. No extraordinary level of analysis motivates the system, as the chatter is generally banal, structurally simple, and repetitive, but the control processes that access the system by means of extremely efficient attention procedures to integrate the formulaic expressions into a continual stream of speech are impressive. A second expression of fluency is the person who can speak reasonably well about a wide range of topics, displaying just enough informative knowledge to participate in a conversation. These speakers also rely on formulaic expressions, but must show greater analysis of those expressions and the 124
Metalinguistic dimensions High control
lecturing
disk jockey
simultaneous translation
talk-show host
creative talk
- • High analysis
Low analysis <
definitions L2 conversation
LI conversation
children's conversation
Low control
Figure 6.2 Oral uses of language corresponding linguistic forms since they must be adapted to a range of purposes. Thus, these speakers may have less control over the forms than do disk jockeys, but they need nonetheless a more highly analyzed conception of language to apply the system to the different purposes. Fillmore offers certain of the "talk-show hosts" as individuals endowed with this ability. A third type of fluency is the master of speech. Such speakers not only produce the fluency associated with high levels of control, but also produce speech which is eloquent, thoughtful, and perfectly formed. Although lecturing in its supreme form demands such skill and lecturers may aspire
125
Ellen Bialystok to achieve it, its ultimate attainment is probably quite rare. Classical societies placed a high value on such skills. In the absence of widespread literacy, the orator carried the responsibility for dissemination of information and formation of attitudes and opinions. This type of fluency exacts high levels of both analysis and control. Finally, Fillmore identifies a fourth fluency which is associated with creative uses of language. These are the speakers who can invent puns and jokes in the course of conversation, seeing the possibilities in both the linguistic forms and the meanings within the time constraints of the interaction. This form of fluency demands attention to multiple aspects of the conversation, but also sufficient analytic ability to creatively manipulate the forms and concepts. Fillmore claims that all types of fluency, these being four examples of highly achieved skill, depend on both knowledge of language, including grammar, vocabulary, and forms of expressions, and control of a number of processes for creating expressions. Speakers who display different types of high levels of fluency differ, according to Fillmore, in the extent to which they have mastered the knowledge and control of language. These knowledge and control bases described by Fillmore correspond in a general way to the analysis and control components of language processing described in the present system. Fillmore's examples, then, serve to illustrate the variety of linguistic functions that share membership in the oral domain and the utility of processes such as analysis and control to distinguish among them. The distribution of oral-language tasks according to their demands for analysis and control permits an examination of possible differences between monolingual and bilingual proficiency in the oral domain. As a reference point, begin by considering conversation in afirstlanguage, which is plotted at a moderately low position on the graph. Speakers need to have some analytic knowledge of the system, both linguistic and conceptual, and sufficient attentional control over processing so that the real-time constraints of oral interaction are not violated. This situation can be contrasted with two other types of conversations - those of children and those of second-language learners. Children's conversations reduce the necessity for both analysis and control. Diminished reliance on analysis is the consequence of the highly contextualized nature of children's conversations. Children talk about things present, the "here and now," in the presence of the objects and events being referenced (Snow & Ferguson, 1977). In this situation, conveying and interpreting meanings can be achieved with only casual attention to the language itself (Macnamara, 1972). Thus the pressure for analyzed conceptions of language is reduced. Similarly, the problem of attention is reduced, in that attention to overall meanings without the need for considering details of the linguistic 126
Metalinguistic dimensions code or monitoring the complexities of anaphoric reference is a simple and sufficient attentional strategy. The conversations of second-language learners differ from conversation in afirstlanguage primarily in control. To the extent that adult conversations normally impose greater control demands than do children's conversations in that more integration of multiple sources of information must be achieved within severe time constraints, these attentional demands are even greater for an imperfectly known language. It is the decreased levels of control possessed by the second-language learner that mark the unique character of second-language conversation. Second-language learners frequently manage to reduce the demands for analysis by the use of highly practiced and conventionalized ''chunks" or patterns of language. These patterns require little analysis, as they allow whole phrases to function as single items. Hakuta (1976) describes the use of these chunks as follows: "the learner will employ a strategy which 'tunes in' on regular patterned segments of speech, and employs them without knowledge of their underlying structure, but with the knowledge as to which particular situations call for what patterns" (p. 331). These chunks, then, are treated as meaning units, part of the unanalyzed representations of language that are the basis of the system of words and meanings. As these chunks become analyzed, they will be used in more creative contexts and be separated into their components. The demands of carrying out conversations in a second language and achieving fluency require higher levels of control of processing than does the more usual experience of speaking in a first language. But many other oral uses of language also increase the demand for higher levels of control. Bilingual children have the experience of conducting a conversation in a second language, an experience quite obviously not encountered by monolingual children. And while this experience places increased demands on the child's level of control of processing, it may be as well that the child benefits from this experience in that levels of control are enhanced through the challenge. Bilingual children who are perfectly balanced in their knowledge of the two languages may find the levels of control necessary for oral-language uses equivalent in the two languages. But it is rare for children to have such equal proficiency in two languages. It is more likely the case that one of the languages puts bilingual children in the position of stretching their level of control of processing available in order to sustain the interaction. Other aspects of oral-language use contribute as well to increased levels of control. Bilingual children routinely attend to the form in which a language is spoken to decide which of their linguistic systems is relevant. They attend to differences between language systems to make choices when addressing different speakers. Functioning in the oral domain, then, may 127
Ellen Bialystok High control
skimming
shifting style
writing poetry studying/writing Low analysis <<-
- • High analysis LI reading
L2 reading
early reading
reading readiness
Low control
Figure 6.3 Literate uses of language serve to increase the control abilities of bilingual children by placing additional demands on that processing component. Literacy domain Tasks in the literacy domain are also concerned with meanings, but the construction of those meanings proceeds from interactions with text. In addition, literacy tasks in general require higher levels of analysis and control than do oral uses of language (see figure 6.3). Reading is a decontextualized use of language, that is, the language is presented outside of the empirical context to which it refers. The intended meaning of the text, therefore, can be obtained only through correct interpretation of linguistic
128
Metalinguistic dimensions forms, thereby increasing the need to understand the structure of these forms. No contextual cue will lessen the burden for accurate and precise interpretation as examination of the language forms is the only route to constructing the meanings. Moreover, because graphemic forms are initially parasitic upon phonological forms, that is, young children attach the letters to sounds and not directly to meanings, greater analysis of the linguistic system is necessary to accommodate this two-step path to meaning. Reading also demands more sophisticated control over attention than do oral uses of language. Information from the graphemic forms must be carefully sampled and integrated with information about emerging meanings. Conversation, by contrast, may proceed adequately on the basis of chunks of meaning that depend less critically on each word or constituent. Reading, then, requires shifting away some of the attention that naturally falls to meaning in conversational uses of language and redirecting it to more careful examination of the forms. Some children who experience difficulty in learning to read can be shown to be deficient in each of these processing components. In a study of some children with severe reading difficulties but normal intelligence (including verbal I.Q.), two groups of children were identified (Bialystok & Mitterer, 1987). One group of children could sound out words correctly and appeared to be reading fluently, but they had poor comprehension of what they read. These children were shown empirically to lack control of processing, although their levels of analysis of linguistic knowledge were no different from those of normal readers. They read, that is, by paying excessive attention to the forms and inadequate attention to the developing meanings. Another group of children did not have the necessary analyzed conceptions of language to even develop adequate decoding skills, and they read by focusing all their attention on the meanings, disregarding the forms. Most of the errors these children made were to substitute words that were roughly appropriate to the context but bore no formal similarity to the word written in the text. Reading, like speaking, is not a single activity. Several types of reading are plotted on figure 6.3. One feature of highly skilled reading is the ability to shift styles to accommodate specific goals. One reads for information differently from the way in which one reads for gist; one reads novels differently from the way in which one reads textbooks. These different styles of reading entail different strategies for scanning and integrating the information in the text. During the first several years in which children learn to read, one of the important skills they develop is this ability to adjust their style of reading to the purpose. Children who are considered to be good readers show greater facility with this adjustment than do children who are weaker in reading (Brown, 1980). This ability to shift styles is a function of control, and tasks which require children to adjust style 129
Ellen Bialystok to read in specific ways demand high levels of control. The main feature distinguishing among different literacy tasks, however, is the level of analysis of knowledge implicated in the problem. Writing places greater demands on the analysis of the system than does reading. Manipulation of language forms, especially for creative writing, demands a highly organized representation of the linguistic system that makes structure and possibilities of meaning transparent. The supreme achievement in this respect is writing poetry. Increased demand for analysis of linguistic knowledge is also the prime characteristic of reading in a second language. Because reading itself is so dependent upon the reader's knowledge of language structure, the reader must have analyzed conceptions of a second language that are more organized and more explicit than are the conceptions of language needed for conversation. Becoming literate in a second language, therefore, forces the language learner to examine the structure of the second language through the process of analysis so that the language is represented as a formal system. This means that bilingual children who are also biliterate have had the experience of analyzing two linguistic systems, the result of which must translate into a more powerful and more analytic conception of language in general. Metalinguistic domain The first section of this chapter presented some of the issues that are involved in the confusion about the definition for metalinguistic. Following from the processing framework that is being used here, the definition of metalinguistic becomes fairly straightforward. First, metalinguistic cannot be used to apply to a unique set of abilities possessed by some language learners. This is because the processing components invoked in the solution to metalinguistic tasks are exactly the same (but to different degrees) as the processing components invoked in the solution to oral and literate tasks. Second, the criterion of awareness does not serve to clearly demarcate the domain, since various problems in oral and literate uses of language were characterized by different degrees of awareness, even though the problem was not metalinguistic. Accordingly, the best application of the term metalinguistic appears to be to a group of tasks, or language uses. These would be those uses of language characterized by three criteria: relatively high demand for analysis of linguistic knowledge; relatively high demand for control of processing; and the topic is language or structure. Metalinguistic tasks differ from each other in their primary reliance on either analysis or control, and empirical differences in children's metalinguistic abilities can be reconciled by attention to the task demands. Some examples of metalinguistic tasks are shown in figure 6.4. Levels of learners' 130
Metalinguistic dimensions High control
judge anomaly
symbol substitution
counting words in sentence
sun/moon problem
segment text
- • H i g h analysis
Low analysis *^~ produce rhyme, synonymy, paraphrase
detecting errors
correcting sentences
judge correct sentences
Low control
Figure 6.4 Metalinguistic uses of language analysis of knowledge and control of processing should determine which types of metalinguistic tasks they can solve. If bilingual children approach metalinguistic tasks with different levels of ability in these two processing components, then bilingual children will solve metalinguistic problems differently from monolingual children. The discussion of the oral and literate domains suggests two ways in which the processing skills of bilingual children may be different from their monolingual peers. First, experience in oral uses of language should enhance
131
Ellen Bialystok the levels of control of processing for bilingual children to monolingual children. Since (virtually) all bilingual children have some experience with oral language, all bilingual children should perform better than monolingual children on metalinguistic tasks that depend heavily on control of processing. Second, experience in using literacy skills in a second language should enhance the levels of analysis of knowledge for bilingual children compared to monolingual children. Not all bilingual children are biliterate. Those who are should perform better than monolingual children and bilingual children who are not biliterate on metalinguistic tasks that depend heavily on analysis of knowledge. A common task used to assess metalinguistic ability is a grammaticalityjudgment problem. Different versions of the problem and different presentation conditions, however, change the underlying demands of analysis and control (Bialystok, 1986a). Manipulating these versions provides access to children's levels of analysis and control. The instructions to children are to respond only to the grammaticality of sentences presented orally. A number of training trials are presented to ensure that children understand that the task requires judgments only of syntax. One condition consists of meaningful grammatical sentences (Why is the dog barking so loudly?). These sentences are very easy to judge, requiring little analysis of structure and little control over attention. Ungrammatical sentences (Why the dog is barking so loudly?) have consistently been shown to be more difficult to judge, and this difference is attributable to the increased need for analysis in detecting an error. Correcting errors in sentences places even greater burden on levels of analyzed knowledge. Sentences can also be made anomalous (Why is the cat barking so loudly?), and these sentences are hard because attention must be diverted from the compelling anomaly in meaning and remain fixed only on the forms. Judgments of this type require control. In a study of monolingual and bilingual children between the ages of five and nine years, children's ability to solve these various versions of the problem depended on both their level of literacy and whether or not they were bilingual. In general, children who were older and more literate scored higher on the versions demanding higher levels of analysis than younger children. Children who were bilingual, irrespective of age or literacy, scored higher than monolingual children on the items demanding higher levels of control of processing. Children who were bilingual and biliterate also demonstrated some advantage on items demanding high levels of analysis. Some metalinguistic problems are specialized for their demands on control, even in the absence of much need for analysis of the linguistic system. The symbol-substitution task, created by Ben-Zeev (1977), is an example 132
Metalinguistic dimensions of a problem that imposes high demands on control of attention. Her problem is as follows: in this game, every time I say the word they, you say the word spaghetti. The first sentence is "They are good children." The child must respond, "Spaghetti are good children," and this is very difficult to do. The sun/moon problem created by Piaget (1929) is another example of this type of problem. The problem is: suppose everyone got together and decided to call the sun the moon and the moon the sun. What would be in the sky when we go to bed at night? (Answer: the sun.) What would the sky look like? (Answer: dark.) In a study comparing monolingual, partially bilingual, and fully bilingual (and biliterate) children, both groups of bilingual children scored significantly higher than the monolingual group in their solutions to the sun/moon problem. The two bilingual groups were not different from each other in level of performance. In both these problems, the solution depends on being able to focus attention only on the forms without being distracted by meanings that are either misleading or irrelevant. Metalinguistic problems of somewhat median demands for analysis and control are the range of tasks that involve producing or judging word pairs on criteria such as rhyme, synonymy, or paraphrase. These types of problems are often used in conjunction with assessments of children's ability to cope with literacy demands, and according to the placement of these tasks on the framework compared to fluent reading, the correspondence between the two activities could be justified by their similar demands for analysis and control. Finally, there is a group of problems that are highly integrative in that they make rigorous demands on both analysis and control. One example is the problem in which subjects (usually children or illiterate adults) are asked to count the number of words in a sentence. Young children cannot effectively solve this problem until they are about six or seven years old. Regarding analysis, the problem is difficult because the subject needs a fairly well-constructed notion of word boundary and an understanding of how this unit relates to other units of speech, such as phrases, articlenoun combinations, and the like. Regarding control, the problem is difficult because the subject's attention is inevitably drawn to the meaning of the sentence, but correctly counting the words, especially the functors, requires paying no attention to that meaning and looking only at the formal units. Again, by constructing different versions of the problem that differentially rely on analysis and control, it is possible to separate children's ability with respect to each of these components. Consistent with our previous research, bilingual children were shown to be more advanced than monolingual children in the versions of the problem dependent upon control of processing, while biliterate children demonstrated further advantages 133
Ellen Bialystok in versions based on higher levels of analysis of knowledge (Bialystok, 1986b). These counting problems are part of a large group of similar tasks involved with segmentation of linguistic passages. The various metalinguistic problems sampled in this discussion, as well as the many not included, differ from each other in the extent to which each of the two processing components is involved in the solution. Various levels of involvement for each of the two processing skills produces an array of problems that are considerably different from each other in overall level of difficulty, and in some cases, superficially different from each other in the kinds of abilities being investigated. What makes them all metalinguistic, however, is that they each depend on relatively high levels of at least one of the two processing skills and the topic, or purpose of the task is language and its structure. The abilities children need to solve the problem are not metalinguistic; they are the same abilities that children needed to solve oral and literacy problems. They generally need them, however, at a higher level of mastery than that which was sufficient for these other two domains. Bilingual children solve metalinguistic problems the same way as do monolingual children; but bilingual children approach these problems with different initial levels of mastery of analysis and control than do their monolingual peers. Bilingual children have higher levels of control than monolingual children, so metalinguistic problems based primarily on control will be more easily solved by bilingual children. Some bilingual children, notably those who are biliterate, also have advanced levels of analysis, and these children will have an advantage in solving those metalinguistic problems that are expressly dependent upon that processing skill. Implications for bilingual processing The characterization of language use in terms of different domains leads to a conception of language proficiency that is both dynamic and contextdependent. What we mean by proficiency, that is, is determined by the general cognitive-linguistic level of the learner and by the demands imposed by specific forms of language use. Learners can develop proficiency in certain aspects of language use while their ability to perform in other contexts remains limited. Thus, language learners, particularly secondlanguage learners, may "specialize" for the oral conversational uses of the language or for the technical written aspects. These two contexts demand different combinations of the underlying components required to process language, and learners can deliberately elaborate their resources necessary for one of these contexts. Developing language proficiency in two languages is a special case of moving through to increasingly high levels of analysis and control so that increasingly demanding problems come within reach. The very experience 134
Metalinguistic dimensions of learning to read in a second language, to switch languages when addressing different speakers, to have completely different systems for expressing such notions as past tense, bestow upon the bilingual child the opportunity to make rapid advances in their mastery of the two processing components that underlie all language use across the domains. These advantages show up when the child is then asked to solve a problem that happens to need just that skill component. Language learning and use
The chapter began with an attempt to define what metalinguistic might mean in an operational sense of language processing. The conclusion of the first section was that language processing depends on the two components of analysis and control, and that these components are at a premium in metalinguistic uses of language. Their continuity through other uses of language, however, demonstrates that these components are not different for metalinguistic uses of language, but that they seem to be demanded at higher levels. In this way, analysis and control can be considered to be the metalinguistic dimensions of language proficiency. They are the processes that define performance across tasks and that determine entry into the metalinguistic domain. In the second section, it was argued that bilingual children (and adults) have different abilities to solve problems in three language domains than do monolingual peers because of their different levels of mastery of analysis and control. In this way, language processing of various types was traced to underlying demands for analysis and control. In this section, some further issues in language processing will be considered. These are other aspects of performance that can be traced to the two processing components. The intention is to demonstrate how mastery of analysis and control by bilingual language learners determines how language is processed, how it is learned, and how it is used. Three issues will be briefly discussed, pedagogy, variability, and strategies. Language learning and language instruction When proficiency is redefined in terms of components, there are direct implications for language instruction. Different methods of language learning differentially nurture the development of the two processing components. This is true for different forms of instruction and for different kinds of learners. Different programs of language teaching aim to develop different aspects of proficiency. Some programs, for example, place great emphasis on teaching rules of grammar and structure. In these programs there is a stress 135
Ellen Bialystok on written work and formal accuracy is highly valued. These programs are addressed to developing the learners' analysis of linguistic knowledge. Other programs are focused on the development of oral-language skills. Evaluation is based on fluency and the learners' ability to express meanings, even in the absence of formal accuracy. These programs aim to develop the learners' control of processing. Few language-instruction programs, of course, are clear and simple cases of these two extremes. Most programs are eclectic to varying degrees and develop a broader-based competency than these simplistic alternatives would allow. Yet programs differ from each other in emphasis. Learners might want to consider language instruction in these terms in order to select a program that will develop the sort of competency that most closely matches their purpose in studying the language. Learners also differ in which of the two processing components is most easily developed in a new language. Individuals vary in their style of learning and in their style of approaching language. Some learners may find it easier to learn a language by systematically applying a set of rules, while others find it easier to discover broad patterns of meaning. Adults and children may also differ in this regard. In general, adults would be more likely to profit from a rule-oriented approach to language instruction than would young children. Variability in language use It is commonly observed that second-language learners demonstrate considerable variability in their apparent control over the forms of the new language. Correct forms seem to slip in and out of the learners' speech, defying any accurate measure of their progress with the second language. Tarone (1988), following the work of Labov (1972), has offered one explanation for this. She posits that learners have variable rules of the form: use form X in Y percent of obligatory contexts. Such a rule would both account for the learners' command of the form when it appears correctly and predict its absence on certain occasions. Moreover, the rules are refined so that variability is systematically related to variation in context, making it possible to determine those contexts in which the correct form will emerge. In formal situations where learners pay careful attention to speech, the correct form will more likely be used. The rule in this case may be: use form X in 100 percent of obligatory contexts. In informal, casual situations, the appearance of the standard form is less likely. Here the rule may be: use form X in 50 percent of obligatory contexts. Although this seems correct in essence, there may be a different sort of explanation for why these correct forms seem to have such a fragile existence in the speech of second-language learners. For these learners, 136
Metalinguistic dimensions variability arises from two distinct sources, and an explanation of variability in speech must account for both. The first type of variability is cognitive and reflects changes in the mental representation. This type of variability is expressed diachronically over periods of time; it occurs because the mental representation changes, through the process of analysis. As learners come to master new forms, there is a period of learning and uncertainty where all the constraints are not yet in place, and sometimes the forms are used correctly but because the conditions of application are fuzzy, errors occur. This variability is objective and can be measured through criterion-referenced evaluation, namely, comparison to the target language. Young children learning their first language exhibit the same phenomenon. It is often difficult to determine the point at which a child has learned a new form. In the seminal study of children's first-language acquisition, Brown and his colleagues (Brown, 1973) documented the acquisition of fourteen morphemes and decided that the child had mastered a form when it was used correctly in 90 percent of the obligatory contexts. The second type of variability is performative and reflects variations in situation. This variability is synchronic: at one point in time a learner can use the same form correctly and incorrectly, depending on the context of use. This variability is determined by the processing constraints imposed by the situation and reflects changes in control. Thus, when the context for language use leads the learner to direct attention to language forms, then the forms are more likely to be produced correctly than when the situation has placed a premium on attention to meanings. Native-speaker speech shows this synchronic variability. In certain situations, it seems more propitious to pay attention to how something is said than to what is said. Measures of this type of variability must be normreferenced so that language learners are judged according to how native speakers would perform in a similar context. The explanation developed by Tarone is primarily addressed to this aspect of variability. The difference between the two types of variability is their dependence upon either the analysis or control aspects of language processing. Diachronic variability is a reflection of changes in the learners' analysis of the linguistic system. Synchronic variability is a reflection of changes in the learners' (or native speakers') selective attention to aspects of the language. All language users are subject to synchronic variability. Only children and second-language learners are subject to diachronic variability. Strategies for communicating One of the perils of speaking in a second language is having an inadequate vocabulary or grammar to convey intended meanings. While the problems 137
Ellen Bialystok endemic to this shortage are suffered most acutely by bilingual speakers, there are benefits for these speakers as well. The solutions to the problem of communicating in the absence of adequate linguistic resources have been referred to in the literature as communicating strategies. The types of solutions learners devise were first studied seriously by Tarone (1977) and elaborated by others. The range of solutions, however, can be distilled to two options, each based on one of the two processing components. This explanation is detailed elsewhere (Bialystok, 1990). In brief, the options are either to express the concept some other way but keeping within the same linguistic system, or to attend instead to a different means of expression, such as a different linguistic system or non-linguistic means of communicating. Following the first option depends on the extent to which the linguistic system has been analyzed, permitting substitutions for synonyms, equivalent structures, and the like. Following the second option depends on the extent to which attentional resources can be redirected to alternative sources of information that are less salient. All speakers, both monolingual and bilingual, have the processing resources to employ both these alternatives. Bilingual speakers have the additional advantage of being more frequently required to test these alternatives. The effect of this experience may well be to enhance the levels of analysis and control for bilingual speakers.
Bilingual processing
The argument advanced in this chapter has been that the metalinguistic dimensions of language processing are the processing components of analysis and control. As claimed throughout, these components are not unique to metalinguistic tasks but are the essence of language processing in the broad sense. Vygotsky's statement, cited at the beginning of this chapter, was that bilingual children would have an advanced awareness of language processing because of their rich and unique experience of interacting with the world through two linguistic systems. Translated into the processing components of the present account, the claim is that bilingual children have enhanced awareness of the analysis and control components of processing. Each of the sections attempted to document the ways in which the experience of using two languages made the levels of analysis and control different for bilingual children than for monolingual children. There are no universal advantages, nor are there universal liabilities in being bilingual. But processing systems developed to serve two linguistic systems are necessarily different from the same processing systems that operate in the service of only 138
Metalinguistic dimensions one. Bilingual children, then, ultimately and inevitably process language differently from monolingual children. References Ben-Zeev, S. 1977. The influence of bilingualism on cognitive strategies and cognitive development. Child Development, 48,1009-1018. Bialystok, E. 1986a. Factors in the growth of linguistic awareness. Child Development, 57, 498-510. 1986b. Children's concept of word. Journal of Psycholinguistic Research, 15,
13-32. 1990. Communication strategies: process and problem in using a second language. Oxford: Basil Blackwell. Bialystok, E. & J. Mitterer 1987. Metalinguistic differences among three kinds of readers. Journal of Educational Psychology, 79,147-153. Bowey, J. A. 1988. Metalinguistic functioning in children. Geelong, Australia: Deakin University Press. Brown, A.L. 1980. Metacognitive development and reading. In R. J. Spiro, B. C. Bruce & W. F. Brewer (eds.), Theoretical issues in reading comprehension. Hillsdale,NJ:Erlbaum. Brown, R. 1973. A first language: the early stages. Cambridge, MA: Harvard University Press. Case, R. 1985. Intellectual development: birth to adulthood. New York: Academic Press. Clark, E. V. 1978. Awareness of language: some evidence from what children say and do. In A* Sinclair, R. J. Jarvella, & W. J. M. Levelt (eds.), The child's conception of language. Berlin: Springer. Collins, A. M. & M. R. Quillian, 1969. Retrieval time from semantic memory. Journal of Verbal Learning and Verbal Behavior, 8, 240-247. Fillmore, C. J. 1979. On fluency. In C. J. Fillmore, D. Kemper, & W. S-Y: Wang (eds.), Individual differences in language ability and language behavior. New York: Academic. Fodor, J. & Z. W. Pylyshyn, 1988. Connectionism and cognitive architecture: a critical analysis. In S. Pinker & J. Mehler (eds.), Connections and symbols. Cambridge, MA: M.I.T. Press. Hakuta, K. 1976. A case of a Japanese child learning English as a second language. Language Learning, 26, 284-297. Jackendoff, R. 1987. Consciousness and the computational mind. Cambridge, MA: M.I.T. Press. Karmiloff-Smith, A. 1986. From metaprocess to conscious access: evidence from children's metalinguistic and repair data. Cognition, 28, 95-147. Labov, W. 1972. Sociolinguistic patterns. Philadelphia: University of Pennsylvania Press. Lundberg, I. 1978. Aspects of linguistic awareness related to reading. In A. Sinclair, R. J. Jarvella, & W. J. M. Levelt (eds.), The child's conception of language. New York: Springer. 139
Ellen Bialystok McClelland, J. L. & D. E. Rumelhart, 1986. Parallel distributed processing. Vol. II: Psychological and biological models. Cambridge, MA: M.I.T. Press. Macnamara, J. 1972. The cognitive basis of language in infants. Psychological Review, 79, 1-13. Mandler, J. M. 1983. Representation. In J. H. Flavell & E. M. Markman (eds.), Handbook of child psychology. Vol. Ill: Cognitive development. New York: Wiley. Piaget, J. 1929. The child's conception of the world. London: Routledge and Kegan Paul. Rosch, E. 1975. Cognitive representations of semantic categories. Journal of Experimental Psychology: General, 104, 192-223. Rumelhart, D. E. & J. L. McClelland 1986. Parallel distributed processing. Vol. I: Foundations. Cambridge, MA: M.I.T. Press. Rumelhart, D. E., P. H. Lindsay & D. A. Norman, 1972. A process model for long-term memory. In E. Tulving & W. Donaldson (eds.), Organization of memory. New York: Academic. Schneider, W. & R. M. Shiffrin 1977. Controlled and automatic human information processing. I: Detection, search, and attention. Psychological Review, 84,1-66. Smith, E. E., E. J. Shoben, & L. J. Rips 1974. Structure and process in semantic memory: a featural model for semantic decisions. Psychological Review, 81, 214-241. Snow, C. E. & C. A. Ferguson (eds.) 1977. Talking to children: language input and acquisition. Cambridge: Cambridge University Press. Sternberg, R. J. & J. S. Powell 1983. The development of intelligence. In J. H. Flavell & E. M. Markman (eds.), Handbook of child psychology. Vol. Ill: Cognitive development. New York: Wiley. Stroop, J. R. 1935. Studies of interference in serial verbal reactions. Journal of Experimental Psychology, 18, 643-662. Tarone, E. 1977. Conscious communication strategies in interlanguage. In H. D. Brown, C. A. Yorio & R. C. Crymes (eds.), On TESOL '77. Washington, DC: TESOL. 1988. Variation in interlanguage. London: Edward Arnold. Vygotsky, L. S. 1962. Thought and language. Cambridge, MA: M.I.T. Press.
140
7
Translation skill and metalinguistic awareness in bilinguals1 M A R G U E R I T E M A L A K O F F and KENJI
HAKUTA
Scholars and educators have studied the effects of bilingualism on cognitive and linguistic abilities for almost a century. The question has generally focused on how a child with more than one language mentally organizes language, and on the repercussions of bilingualism on cognitive and linguistic development. Both these questions grew out of what might be called the monolingual-norm assumption: the belief that monolingualism is the cognitive-linguistic norm and that the child's cognitive system is fragile and designed to cope with only one language. The monolingual-norm assumption gave rise to the negative myths surrounding bilingualism: bilingualism has been blamed for cognitive, social, and emotional damage to children (see Hakuta, 1986, for review). Recent research, however, does not support the view that simply speaking two languages taxes either the cognitive or the linguistic system. Studies of balanced bilinguals (bilinguals who have roughly equivalent abilities in the two languages) have suggested that bilingualism has a positive effect on cognitive development, especially under certain conditions of additive bilingualism where both languages are supported academically and emotionally by both the community and the society at large. The influence of the environment plays a large role in determining whether the bilingual situation will be additive or subtractive (Lambert, 1975). Subtractive bilingualism occurs when the mother tongue is a low-status minority language that is rapidly being replaced by the high-status majority (and second) language. There is evidence that suggests that minority-language Hispanic students in the United States show positive cognitive gains from bilingualism while they are becoming bilingual; that is, while they are receiving academic support from Spanish and learning English (Hakuta, 1988). The majority of studies on bilingualism to date have focused on comparing bilinguals to monolinguals, and most measures used have been derived from and for monolingual samples. Bilingual performance is thus directly compared to monolingual performance - bilinguals may be handicapped or cognitively enhanced, depending on how they measure up to their monolingual counterparts. Such a design assumes that the cognitive-linguistic experience of 141
Marguerite Malakoff and Kenji Hakuta the two populations is comparable. Bilinguals, however, differ from monolinguals in a very major way: the bilingual child experiences the world through two languages - two languages which are used in alternation. For the bilingual, linguistic experience is spread over two languages: experience is encoded in either of two languages and can be expressed in both languages, and information representation can be switched between the languages. The most explicit process in which this occurs is in translation, the topic of this chapter. Although the professional translation literature distinguishes between translation and interpretation, where translation refers to the written modality and interpretation generally refers to the oral modality, we use the term translation to refer to all modes of reformulating a message from one language (the source language) into another language (the target language). For many bilingual children throughout the world, translation is an everyday activity, a part of their lives as bilinguals (Grosjean, 1982). Yet translation is a poorly understood phenomenon. In this chapter, we will first provide a review of literature and concepts relevant to the study of translation ability. We will then report two studies that investigate translation ability among language-minority Puerto Rican children in the United States. Theoretical perspectives on translation
The empirical literature on translation is sparse; for translation by children it is barren. The majority of the literature on child translation comes either from anecdotal evidence by linguists (for example, Leopold, 1939-1949) and other scholars, or from indirect evidence from studies in which translation was observed or used as a research technique, but not directly studied (see Harley, 1986; Paivio, Clark, & Lambert, 1979; Swain, 1972; Swain, Naiman, & Dumas, 1974). Although the linguistic nature of translation is often discussed in the more recent translation literature, we have found little empirical study of translation as a linguistic skill, from either a linguistic, a psycholinguistic, or a sociolinguistic perspective. Several reasons for this paucity of research may be cited. One reason is the lack of theoretical and methodological coherence in the still young science of translation (Harris, 1977, 1980; Nida, 1976; Seleskovitch, 1976; Wilss, 1976, 1982). Wilss (1982) notes that the applied science of translation is younger yet, while Vinay (1975) noted that the heading translation theory is more often than not an indication of a discussion of translation problems, and not one of theoretical formulation. Another reason is that until recently, child bilingualism has been viewed with a wary eye especially when the children are from minority language and lower socioeconomic status (S.E.S.) families. The greater issue here has been on keeping the languages separate and reaching proficiency in the majority 142
Translation skill and metalinguistic awareness language, rather than on the benefits and advantages of knowing two languages. Whether the focus has been on minority-language students or middle-class children, the emphasis both in education and in research has been on the process of second-language acquisition. Further, translation - used since Roman times to teach a second language - fell out of fashion in the postwar era. Together, this blacklisting of translation in foreign-language education and the focus on second-language acquisition contributed to the lack of interest in translation as a bilingual skill. Yet, clearly, all bilinguals are able to translate at some level. To what extent the nature of their translation resembles that of trained interpreters is an open question - mainly because so little is known about the natural translation ability of bilinguals. Recent theories of translation (Nida, 1976; Pergnier, 1984; Seleskovitch, 1976; Seleskovitch & Lederer, 1984; Wilss, 1982) have emphasized the communicative aspect of translation. The fact that it is possible to communicate meaning in the absence of correct sentence structure has been overlooked, probably because these theories have focused on professional adult translators and translation. The oldest, and, until recently, the most prevalent model of translation is a two-stage model. Ljudskanov's (1969) twostage model of translation is typical of this approach: the first stage is the analysis of the source-language text, and the second stage is the synthesis of the information into a target-language text. However, Seleskovitch (1976,1978) argues that this binary model does not capture the communicative dimension of the translation, and adds a third intervening step: the comprehension of the meaning. At the level of the reformulation, however, she makes no distinction between the communicative demands and the metalinguistic demands of the task. It is possible that Seleskovitch does not distinguish between these two demands because her work is both inspired by and focused on professional translation, a domain where linguistic sophistication is a given. A professional translator is more likely to miss the intent of the speaker if not familiar with the subject matter than to use the wrong syntax (Seleskovitch, 1976). Bilingualism and natural translation Brian Harris has argued that the empirical study of translation should include, even begin with, the study of translation by naive translators and, in particular, naive child translators - bilingual children without any special training in translation (see Harris, 1977, 1980; Harris & Sherwood, 1978). Harris adopted the term natural translation to refer to this type of translation, a type which he contrasted with professional translation as carried out by highly trained and sophisticated translators. Natural translation refers to the cognitive skills involved, not to the translation situation. That 143
Marguerite Malakoffand Kenji Hakuta is, natural translation is produced by a child (or adult) who has received no formal training in translation and is relying on a set of natural linguistic skills. Harris and Sherwood (1978) propose that all children can translate in all cultures, in all languages and registers, throughout history and from the time the individual starts to acquire a second language. To support this claim, they cite a number of findings, including the early age at which natural translation is found; the prevalence of spontaneous translation in young children; the small exposure to language that older children need before starting to translate, and the lack of correlation, in children, between the ability to translate and instruction in translation. Natural translation is thus a necessary concomitant to bilingualism, just as the ability to communicate comes with being the speaker of any language. Translation is not a learned skill, such as learning a foreign language in school, but, rather, it is a skill which is developed from a natural and existing base, similar to the development that occurs in mother-tongue language abilities. Although Harris (1977, 1980) remarks that natural translation can be improved under guidance just like any other natural skill, he does not take a clear stand on the issue of individual differences. Translation is typically viewed as a valuable skill that is available only to the highly trained and linguistically sophisticated bilinguals who come out of interpreter and translator training school. It is not a skill that is generally considered to be within the repertoire of just any bilingual, much less children, much less minority-language children. Yet studies have found that children can both interpret and translate materials that are within their comprehension and vocabulary (Hakuta et al., 1988; Hakuta & Malakoff, 1987; Harris, 1980). Shannon (1987) has documented children interpreting for adults in medical, legal, and administrative situations. These findings generally support Harris' claim about natural translation ability.
Translation and context More recently, the importance of contextual meaning in the translation process and the link between the comprehension of meaning and the translation has been stressed in the translation literature. In an early linguistic formulation of translation, Catford (1965) argued that translation theory must necessarily draw upon a theory of meaning. Processing the text for translation requires taking into account the full context of the text (Ballard, 1984; Mininni, 1981; Nida, 1976; Seleskovitch, 1976; Seleskovitch & Lederer, 1984). A number of authors have further suggested that the processes involved in the comprehension of text can be better understood through translation and interpretation (Mininni, 1981; Nida, 1976; 144
Translation skill and metalinguistic awareness Seleskovitch, 1976). It is assumed within this literature that a deeper comprehension of the text as a whole should improve the quality of the translation (Mininni, 1981; Pergnier, 1978; Seleskovitch, 1978). There is still little empirical basis on which to base such a claim (Mininni, 1981). The theoretical link between translation and comprehension finds its roots, probably, in Catford's theory of meaning (Catford, 1965). Catford argues that meaning in a text can be analyzed at many different levels or units. There are the meanings of the individual words, of the phrases, and of clauses that constitute the sentence; there are the meanings of the individual sentences that constitute a passage, and there is the meaning of the passage as a whole. Catford (1965) argues that below the level of the sentence, equivalence of meaning between two languages cannot be established at the same level: a word in one language may require a phrase in the other, a clause in one language may require only a phrase in the other. It is only at the sentence level that the meaning of a source-language unit (the sentence) may be entirely captured in an equivalent targetlanguage unit (another sentence). Translation "implies the substitution or replacement of textual material in one language by equivalent textual material in another language" (Catford, 1965, p. 20). These levels of meanings may be thought of as windows through which the source-language text is processed: the size of the window determines how much of the text is used to process the meaning for translation. In word-for-word translations, the focus is primarily on the meaning of the individual words that constitute the sentence. If the focus is exclusively on the words, the meaning of the target-language text may be quite different from that of the source-language text: for example, Me gusta el perro pequeho in Spanish produces Me pleases the dog little instead of / like the little dog. If the window size is larger, individual phrases or clauses may be translated sequentially, each with an appropriate syntax, but with the whole lacking a coherent sentence structure or meaning. This may be particularly true in the case of idiomatic expressions, which take their meaning, in part, from their use within the entire sentence. Written translation is said to offer the particular temptation of "translating and then understanding" (Seleskovitch, 1976). Two languages are rarely so similar that a translation equivalent is a word-for-word or phrase-for-phrase transposition from one language to the other. Between even closely related languages there are more or less subtle differences in syntax and idiomatic expressions. Hence, there is generally a certain amount of syntactic and lexical restructuring that must be done in reformulating the original source-language meaning in the targetlanguage sentence structure. Sensitivity to specific differences between the two language systems should result in fewer literal translation errors when moving between the two languages. It is sensitivity to specific language 145
Marguerite Malakoff and Kenji Hakuta differences that permits the use of translation strategies: certain structures are automatically given a red flag that warns the translator to beware. Code-switching and translation Code-switching is the use of two (or more) languages in alternation within a single discourse, sentence, or constituent (Poplack, 1979). It is a bilingual mode of communication that is frequently and extensively used in amongst members of bilingual communities (Gumperz, 1982; Pedraza, 1978; Poplack, 1979; Zentella, 1981). Code-switching, like translation, should be considered a bilingual language skill. However, the goals, uses, and demands of code-switching differ from those of translation. Translation typically involves replacing an utterance in the source language with an equivalent utterance in the target language to enhance communication to monolingual speakers of the target language. Translation aims to reproduce as closely as possible in the target language the meaning of an utterance (or text) in the source language. Code-switching, on the other hand, is used to enhance or complement communication to bilingual speakers. It does not seek to reproduce what has already been said, but to enhance what is being said. Code-switching takes advantage of a larger bilingual vocabulary, playing on subtle differences between the two languages in connotative, denotative, or sociolinguistic meaning. Thus, while translation takes advantage of similarities across two languages, code-switching takes advantage of the differences. As a sociolinguistic strategy, code-switching is used for signaling group boundaries, conveying emphasis, role playing, and establishing sociocultural identity. It is also used to redefine an interaction (Scotton & Ury, 1977), to signal the level of intimacy (Asuncio-Kibdem, 1981) or emotional charge (Marcos & Alpert, 1976; Marcos & Urcuyo, 1979). Code-switching may also be used when a particular word or phrase has a more specific denotative or connotative meaning in the other language. Finally, a code switch may be used when a word is more salient in the other language or unknown in the current language. Poplack (1979) found bilingual ability to be a good predictor of the type of code switch produced: speakers with the greatest bilingual ability were more likely to produce intrasentential switches, a more complex type of code switch. Code-switching and translation have been viewed in a very different light, particularly when used by children: the former is treated as evidence of an absence of linguistic differentiation (Bergman, 1976; Lindholm & Padilla, 1978) while translation is seen as evidence of linguistic separation (Bergman, 1976; Lindholm & Padilla, 1978; Swain & Werch, 1975). This argument assumes that code-switching is not a conscious activity, that the switch is not intentional, and that the child is not aware of the linguistic switch. Weinreich (1953) regarded code-switching in adults as evidence 146
Translation skill and metalinguistic awareness of a lack of control in maintaining linguistic separation. Within U.S. circles, the phenomenon of code-switching has also been considered evidence of the disintegration of Hispanic language and culture, and among educators, as a deviation from some bilingual "norm" (see Poplack, 1979, for discussion). Studies of code-switching, however, suggest quite the opposite: codeswitching is both rule-governed and function-specific, and not evidence of linguistic interference (Chan, Chau, & Hoosain, 1983; Grosjean, 1982; Poplack, 1979; Sridhar & Sridhar, 1980). The point where the language switch occurs, which may be at the word, phrase, or sentence level, is specified by a set of rules; in particular, the word at the switch point must be an appropriate syntactic structure in both languages (Lindholm & Padilla, 1978; Poplack, 1979). Poplack (1979) found that there were virtually no ungrammatical combinations of the two languages in close to two thousand switches produced by bilingual adults; this was true regardless of the bilingual ability of the speaker. Studies of child code-switching suggest that children's usage does not differ greatly from that of adults (Genesee, 1987; Lindholm & Padilla, 1978). Children appear to make more switches that are unacceptable to bilingual speakers than do adult code-switchers; however, monolingual children also make more grammatical errors than adult monolinguals. In a study of eighteen Spanish-English bilingual children between the ages of two and six, Lindholm and Padilla (1978) found that the switches served both sociolinguistic and communicative strategies. The children used lexical switches more often than phrasal switches; the most common switch occurred for a single noun. The switches conformed to rules dictating number agreement and the use of functors; most errors occurred for article agreement. That children make use of rule-governed code-switching is evidence not only of linguistic separation, but also of children's knowledge of the differences and similarities across languages. It is suggestive of a sophisticated metalinguistic awareness.
Metalinguistic awareness
At the most general level, metalinguistic awareness may be defined as an awareness of the underlying linguistic nature of language use. Metalinguistic awareness allows the individual to step back from the comprehension or production of an utterance in order to consider the linguistic form and structure underlying the meaning of the utterance. Cazden (1976) defines the construct as "the ability to make language forms opaque and attend to them in and for themselves" (p. 603). A metalinguistic task, in the most general sense, is one which requires the individual to think 147
Marguerite Malakoff and Kenji Hakuta about the linguistic nature of the message: to attend to and reflect on the structural features of language. Tunmer and Herriman (1984) note that: to be metalinguistically aware is to begin to appreciate that the stream of speech, beginning with the acoustic signal and ending with the speaker's intended meaning, can be looked at with the mind's eye and taken apart. The extent to which this appreciation must be conscious is one of the many tangles in the web. (1984, p. 12)
Bialystok and Ryan (1985) point out that the traditional conceptualization of metalinguistic skill as a unique linguistic ability has proved less than useful. They argue that the term metalinguistic should be applied "not to a specific mental accomplishment but rather to a set of problems which share certain features. The theoretical issue, then, is to determine what cognitive skills underlie the solutions to this set of problems" (1985, pp. 230-231). Metalinguistic awareness is treated not as a unique ability, but as the ability to successfully approach and solve certain types of problems. In this sense, it is both an awareness and a skill: the problem is metalinguistic and the skill is recognizing the nature and demands of the problem. Vygotsky (1962) suggested that bilingualism facilitates certain types of language awareness, a finding that has since been supported by a number of researchers (e.g. Ben-Zeev, 1977; Bialystok, 1988; Galambos & Hakuta, 1988; Ianco-Worrall, 1972; Peal & Lambert, 1962). Studies of middle-class children suggested that bilingualism leads to increased levels of metalinguistic awareness at an earlier age: for example, bilingual children were aware of the arbitrary relationship of names and objects at a younger age than monolingual children (Ianco-Worrall, 1972). A similar bilingual advantage, however, has generally not been found in children who are non-proficient in their second language or children who are from minority-language groups.
Metalinguistic awareness and translation skill Metalinguistic awareness and bilingual proficiency are separate but related linguistic skills: for a given metalinguistic level, there can be a range of bilingual proficiencies, and for a given bilingual repertoire there can be a range in metalinguistic awareness. In the case of elementary- and middleschool bilingual children, the two skills are likely to be correlated. The correlation between these skills is in part the result of the influence of academic experience on both skills: language skills are an important part of elementary- and middle-school curricula. Children who have a more developed sense of metalinguistic awareness are likely also to have more developed language skills in general; this appears to be particularly true 148
Translation skill and metalinguistic awareness for written-language skills. Cummins (this volume) argues that these metalinguistic and written-language skills are also shared across both languages. Translation proficiency might be thought of as the product of an interplay between metalinguistic maturity and bilingual proficiency. A third factor, translation strategies, would enhance performance, but not beyond an optimal-level limit determined by the two linguistic factors. The concept of an optimal-level limit was proposed by Fischer and Pipp (1984). Their model characterizes the acquisition of skills by a series of stage-defining optimal-level limits - the upper limit of performance under optimal conditions of support, experience, and feedback. Within the stage defined by each limit, there is room for individual variation in performance when children are left to their own devices. A translation strategy is a learned strategy that helps improve performance; for example, when going from Spanish to English, a simple strategy might be "always start by reversing the order of the adjective and noun." However, since the adjective can also precede the noun in Spanish, this strategy will induce the translator into error unless it is accompanied by an understanding of the semantic differences for word order in Spanish (bilingual proficiency) and a sensitivity to and monitoring of the resulting English word order and meaning (metalinguistic skill). Thus, the translation strategy cannot raise performance beyond an upper limit of performance determined by the interplay of metalinguistic skill and bilingual proficiency. However, within this limit, it can enhance performance. Adult bilinguals, because of their more developed linguistic abilities and metalinguistic awareness, have a greater range of strategies available to them - use of a dictionary, paraphrase, knowledge of morphological rules, reliance on cognates are a few examples.
A two-level model of the translation process Translation requires the manipulating of language at two levels: it must apprehend and convey the meaning of the source language text; and it must formulate an appropriate target-language sentence structure in which to embed this meaning. Natural translation involves four processes: comprehension of the vocabulary of the original source-language text; comprehension of the meaning of the original source-language message; reformulation of the message in the target-language; and judgment of the adequacy of the target-language text. In the complete translation process, reformulation and judgment must operate at the levels of both meaning and structure. The translator must both reformulate the meaning into the target language and insert that meaning into an appropriate target-language sentence structure. That is, the translator must evaluate both the equivalence of the 149
Marguerite Malakoffand Kenji Hakuta meaning of the source-language and target-language texts and evaluate the appropriateness of the sentence structure used to convey the meaning. From this two-level perspective, translation is a composite of communicative and metalinguistic skills - skills that are "translinguistic," in the sense that they are not particular to any one language. The ability to grasp the meaning of an utterance and to convey that meaning to another person is certainly a communicative skill; this is true whether the conveying of meaning is within a language or across languages (see Brumfit, 1984; Catford, 1965; Mininni, 1981; Steiner, 1975; Wilss, 1982). The evaluation of the target-language sentence, both in terms of the meaning it conveys and the sentence structure in which that meaning is embedded, requires the ability to recognize language as a tool and as a rule-governed system. The translator must evaluate his or her use of the tool, that is, whether he or she has successfully conveyed the message, and his or her abidance by the rules of the target-language system, that is, whether he or she has embedded the meaning in a correct sentence structure. It is this necessity to reflect on language and language use across two languages that makes translation a metalinguistic skill, par excellence (Carroll, 1978; Flesch, 1972; Fuchs, 1982). In natural translation, linguistic sophistication and explicit knowledge of contrastive linguistics is generally not the norm, especially among gradeschool children. Although children of this age can speak two (or more) languages correctly, they do not yet have a conscious awareness of the specific differences between language systems. Despite the absence of such linguistic knowledge, children are able to communicate meaning; although the meaning may be embedded in poor sentence structure. For example, as we will show in our studies, we found that elementary-school students were generally able to communicate the meaning of the source-language text despite errors in target-language syntax and some literal translation errors.
Translation and paraphrase Paraphrase is particularly relevant to the study of translation because of potential parallels with translation. As is the case with translation, paraphrase has been used as a testing tool, but there has been little direct study of the process. A number of authors have argued that it is a part of language competence. For some, it is a measure of the speaker's semantic mastery of the language (Fuchs, 1982; Katz & Fodor, 1963). Gleitman and Gleitman (1970) based their study of language competence on the assumption that paraphrase reflects grammatical competence better than did the traditional classification procedure. 150
Translation skill and metalinguistic awareness Translation has been called "interlanguage paraphrase," or paraphrase "intralanguage translation." In both, the objective is to take a piece of information and recode the meaning in a different linguistic form - in one case the form is a different language and in the other the form is within the same language (Fuchs, 1982). A number of authors argue that paraphrase is a metalinguistic skill (Flesch, 1972; Fuchs, 1982). Paraphrasing consists in finding the meaning of two compared sequences and showing its equivalence, and this identification constitutes a judgment on the sequences (Fuchs, 1982; Kintsch & Van Dijk, 1978). This argument resembles that made for the link between translation and metalinguistic awareness. It is thus intuitively plausible that paraphrase ability is related to that of translation. The degree of this relationship is, however, a topic of debate: one school of thought argues that translation and paraphrase are the same process distinguished only by language mode (Fuchs, 1982; Wilss, 1982); the other position argues that the language distinction between intra- and interlingual communication makes a fundamental difference in the process (e.g., Kade, 1968). Kade argues that in translation, L2 units are matched with LI units, and translation takes place at the level of content. In paraphrasing, on the other hand, a new code in L2 is created; the process thus involves creating a new code within the same language, not simply recoding L2 into LI. Kade, however, ends by blurring the distinction between the two processes, stating that they "rarely appear in their pure form but rather, as a rule, overlap" (p. 17). Mininni (1981) appears to make a similar claim: he offers the "general rule" that the more the contents are assimilated by the translator through an activity of paraphrase, the more adequate the translation will be. Translation and paraphrase are both metalinguistic tasks that depend on the ability to extract meaning from an utterance and capture the equivalent meaning in another utterance. The vocabulary demands of the two tasks differ, however. Typically, paraphrase requires a large vocabulary within one language, while translation requires only a basic vocabulary in each of two languages. Nonetheless, one would expect that if the vocabulary demands of a paraphrase task were made equivalent to the vocabulary demands within any one language in translation, that performance across both tasks would be correlated. That is, a child who performed well on a translation task would also perform well on a paraphrase task, // the within-language vocabulary demands were the same. This claim, however, requires empirical support. Translation, code-switching, and paraphrase are all tasks that require a knowledge of similarities and differences within the speaker's active vocabulary. All three also demand that the speaker anticipate the audience's knowledge of language. In the case of translation and code-switching, the 151
Marguerite Malakoff and Kenji Hakuta vocabulary spans two languages; in the case of paraphrase, only one vocabulary is involved. Two studies of translation
As the above discussion of translation shows, it is an excitingly rich area of investigation that brings together a number of complex psycholinguistic and sociolinguistic problems. Two recent studies of translation ability we conducted provide empirical support that late elementary-school students are able to produce good written and oral translations, and give glimpses into the properties of this ability. The first study looked at children who were considered to have had some experience in translating, usually for their family. The goal of the study was to investigate the psycholinguistic properties of translation among these "experts." The second study addressed the question of the generality of translation ability among a less selected group of bilingual students. Both of these studies were conducted with Puerto Rican children in New Haven, Connecticut, who are from extremely low socioeconomic backgrounds. The community bilingualism can be characterized as generally subtractive in nature, and has been described in some detail elsewhere (Hakuta, 1988). Study 1: Properties of translation ability A broad sweep of translation skills was made in this study, ranging from written translation tasks to on-line measures of word and sentence translation to assess ability in terms of speed. Subjects Sixteen translators (8 girls, 8 boys) were recruited through a local education advocacy and service agency that offered various adulteducation and after-school and summer programs for Hispanic children. The mean age of the subjects was 10.7 years, ranging from nine to twelve years old. Six of the subjects had just completed fourth grade, ten had completed fifth grade. Nine were born in New Haven, seven in Puerto Rico. All but one subject had been in a bilingual-education program at some point. Ten were in the bilingual program in the year preceding the study, but were about to be placed in English-only classes due to their high English proficiency. Five subjects had been in a bilingual program at some previous point in their educational history, for a mean of 1.8 years. Mothers were interviewed in Spanish by a native Venezuelan to obtain information on the children's translation activities as well as their general opinions about bilingualism. All mothers reported that their children did some translating; eleven of the children translated regularly for one or both parents. The children who did not translate for their parents translated 152
Translation skill and metalinguistic awareness for at least one adult (uncle, grandparent, or other adults in the home), as well as for their peers. Many of the children would interpret spontaneously, as well as when asked to do so. In general, mothers thought that their children had been interpreting since third or fourth grade. They reported that the children's services were most often required for visits to the hospital and social services; however, most mothers also said that they needed someone to interpret television programs and telephone conversations. The mothers overwhelmingly viewed bilingualism as a useful skill and not a source of confusion to children. All the mothers believed that interpreting ability would improve employability, whether on the mainland or in Puerto Rico. Several mothers remarked that a good knowledge of English was becoming indispensable to obtain a good job in Puerto Rico. In general, they saw English as improving employability and Spanish as maintaining contacts with the Hispanic community. When asked to estimate their children's use of each language in situations at home, in school, and outside the home, the majority of mothers reported that the children used English more than Spanish. Although all but three mothers reported encouraging the use of Spanish over English at home, the only situations in which the children used primarily Spanish were with their mother or another adult family member. None of the children used primarily Spanish with their siblings, although in no case did they use exclusively English. Measures Language proficiency in both languages was assessed by the Woodcock Language Proficiency Battery (W.L.P.B.), an individually administered standardized proficiency measure with a variety of components, including vocabulary, analogies, reading. In addition, an on-line analogies task was created and administered on a Macintosh computer in each language. Translation tasks involved words, sentences, and stories as translation stimuli. The tasks were conducted in both source-target directions, that is, going from Spanish to English, and from English to Spanish. Stimuli were chosen to minimize vocabulary complexity, so as not to confound translation ability with vocabulary knowledge. Sentences and stories were constructed in order to provide ample opportunities for grammatical pitfalls that would cause intrusion errors (i.e., errors in which the source-language structure intrudes into the translation). In addition, sentences were administered in both a straight and an imagery condition. In the imagery condition, after the subject read the source sentence, a "thought" balloon would appear on the screen during which s/he was instructed to make a mental picture of the sentence. Then the source sentence would reappear on the screen, and the subject could pro153
Marguerite Malakoff and Kenji Hakuta ceed with the translation. It was thought that the imagery condition would expand the "window space" for translation, whereby the meaning of the sentence would be processed to a greater extent and grammatical pitfalls through literal translation would be avoided. The vocabulary and sentence stimuli were administered on the Macintosh computer to enable measurement of translation time. In these on-line tasks, the experimenter controlled the timing of the stimulus presentation with the mouse. When the subject was ready, the source appeared on the screen accompanied by a tone. The subject was instructed first to read the source, and then to provide the translation. The stimulus remained on the screen throughout the trial. The sessions were tape-recorded to enable assessment of the accuracy of the translation. The story translations were administered in written format. In addition to the translations, we attempted to assess translation proficiency through a word identification task, in which subjects were simply asked to say whether words appearing on the computer screen were English or Spanish. The words contained no obvious cues as to language, such as accent marks or obvious spelling patterns. We thought of this as a first approximation of measuring access time to the two lexicons. (A number of people have pointed out to us that this task can be successfully executed with knowledge of only one language, i.e., a monolingual English speaker could make decisions in this task based on whether it is an English word or not an English word. Observation of the subjects in the course of the task suggests that this was not the basis for their decisions, but we have since modified the task to include nonsense words to eliminate this possibility.) Results Several major conclusions emerged from this study. They can be stated in terms of propositions: Conclusion 1: The subjects were extremely good translators and made few errors in both source-target directions. Support for this conclusion comes from an assessment of the quality of sentence translations. Each response was coded for whether it was correct (as an adult balanced bilingual would translate it), incorrect (where a word or key concept was omitted in the translation), or wrong (where the translation had more flaws than the omission of a single concept). In coding these responses, morphological errors were ignored, as they did not affect the major meanings of the translations. Percentages of each coding category were computed across the straight translation and imagery conditions of the sentence translations. In the Spanish-English translations, only 2 percent were wrong and 10 percent were incorrect; in the English-Spanish translations, only 1 percent were wrong and 6 percent were incorrect. 154
Translation skill and metalinguistic awareness Another indicator of the quality of translation is the proportion of two error types in which we took particular interest. A source-word intrusion error was defined as one where a word from the source language worked its way into the translation. For example, one subject translated laproxima semana as the proximate week. A second error type was the source-word order error in which the word order of the source language worked its way into the translation. For example, one subject translated iDonde estara mi hermano? as Where should be my brother? Error rates for these two types of errors were extremely low. Going from Spanish to English, only 0.033 source-word intrusion errors and 0.036 source-word order errors occurred per sentence. From English to Spanish, the respective rates were 0.069 and 0.028. The written translations of stories reveal some intriguing contrasts, although this should not be taken as a direct contrast of modalities between spoken and written responses. Going from Spanish to English, there were 0.048 source-word intrusion errors per sentence, and 0.151 source worderrors per sentence. Going from English to Spanish, the respective proportions were 0.22 and 0.137. The notably higher rates of source errors in this written task suggest that the written modality is considerably more demanding, thereby reducing the window size of processing and causing more source-language-based word-order errors that result from word-forword translations. Conclusion 2: The imagery manipulation did not have an effect on whether literal or non-literal translations were made. Translations were defined as non-literal when the literal order of words and ideas in the source sentence was not preserved, such as through changes in the placement of adverbs or prepositional phrases. It was assumed that forcing subjects to form a mental image of the sentence would lead them to provide more non-literal translations because these superficial features of the target sentence would be diluted with greater focus on the meaning. This manipulation was thought to have an effect on the "window space" of translation. However, the data did not bear out the hypothesis, as can be seen in table 7.1. The proportion of non-literal translations did not vary as a function of the manipulation. This lack of effect, however, may be due to the ineffectiveness of the imagery manipulaticfn, since we observed that subjects found it tedious and tiring to form images continually across a large number of sentences. Conclusion 3: Translation was more efficient translating into English than translating into Spanish, reflecting English dominance. Comparison of means for word- and sentence-translation times revealed significantly more efficient translation going from Spanish to English than 155
Marguerite Malakoffand Kenji Hakuta Table 7.1 Percentage of responses that were non-literal or literal translations in different translation directions and conditions Non-literal (%)
Literal (%)
English-Spanish, straight English-Spanish, imagery
16 13
76 80
Spanish-English, straight Spanish-English, imagery
15 17
75 68
English to Spanish. For words, there was a mean difference of 0.73 seconds (r(15) = 1.96, p < 0 . 0 5 ) , and for sentences, the mean difference was 0.48 seconds (f(15) = 2.19, p<0.05). This asymmetry is attributable to greater English dominance. The dominance is also supported by the results from the time for solving the analogies problems, in which solution times were significantly faster for the English problems than the Spanish problems (f(15) = 3.23, p<0.01). The asymmetry is not surprising, considering the subtractive bilingualism that characterizes the community, as well as the emphasis on English language skills in the bilingual programs that have a transitional rather than maintenance policy. Conclusion 4: For words, translation speed is better predicted by proficiency in the target language than in the source language; the pattern is less clear for the sentence translations.
Regressions were calculated to estimate the power of translation-speed language proficiency in English and Spanish (as measured by the W.L.P.B.) in predicting translation speed for the word and sentence tasks in both directions. As seen in table 7.2, the results are very clear for the wordtranslation task, where Spanish proficiency predicts 0.19 of the variance in the English-Spanish direction, and English proficiency predicts 0.40 of the variance in the Spanish-English task. The pattern is less clear for the sentence-translation tasks. It is possible that the source-language proficiency comes to play a greater role as the unit of language that needs to be processed gets larger, as it does in going from individual words to sentences. Conclusion 5: In addition to proficiency in the two languages, there appears to be a translation proficiency, as measured by performance on the word-identification task, that predicts translation speed.
The data support the hypothesis that in addition to proficiency in the two languages, translation skill requires an additional component of accessibility of the two lexicons. As the hierarchical regression results from table 7.2 156
Translation skill and metalinguistic awareness Table 7.2 R2 values obtained through regression predicting translation times for words and sentences on English and Spanish proficiency, and the word-identification (W.I.) task
Predictors English proficiency Spanish proficiency English + Spanish proficiency English + Spanish proficiency + W.I.
Word translation S->E E-»S 0.01 0.19 0.25 0.57
0.40 0.00 0.44 0.51
Sentence translation E—S S->E 0.21 0.13 0.27 0.55
0.12 0.08 0.16 0.40
indicate, there is a significant increase in R2 values when the word-identification measure is added to the equation with both English and Spanish proficiencies. Discussion The results of this first study show that the subjects were very good translators, and reveal various properties of their translation skill. The generally low incidence of source-language intrusion errors provides striking evidence of the separation of the two languages. Evidently, the source-language proficiency matters little in word translation, although it may matter more when it comes to sentence translation. Target-language proficiency seems to be an important factor in translation efficiency. Finally, some evidence was found for translation ability that goes beyond the sum of the two language proficiencies. Indeed, it may well be that translation ability is related to metalinguistic skills, a hypothesis that must be explored in future research. Study 2: Distribution of translation ability These initial findings encouraged us to ask whether translation skills are distributed across a less selected group of bilingual children. Subjects Fifty-two subjects (27 girls, 25 boys) from fourth- and fifth-grade bilingual-education classes were selected on a random basis, the only constraint being that they had sufficient proficiency in the two languages to complete the written story-translation task in both directions, as judged by the teacher. Approximately two-thirds of the students in the bilingual classes met this criterion. There were 24 subjects in fourth grade, 27 in fifth grade, and 1 in sixth grade. Of these, 18 were assigned to mainstream 157
Marguerite Malakoff and Kenji Hakuta Table 7.3. Examples of errors and codes in translation tasks Spanish to English (S. E. W. 1) articles and quantifiers en la oscura casa sus grandes pies
to that dark house the big feets
(S.E.W.2) nouns, verbs, adjectives, adverbs Los ninos jugaban the boy played en la noche in the dark (S.E.W.3) source Rapidamente el malcriado la proxima semana
Fastly the malobedient the proximate week
(S. E. W. 4) prepositions en el parque a la casa
on the park in the house
(S. E. 1.1A) ungrammatical addition El martes el pajaro se murio
The other Tuesday the bird it died
(5. E.I.W.B.) semantic addition bebia aqua con su almuerzo las cuatro sillas he drank water in lunch with her son azules blue the seats are round and they are (S.E.I.2A) ungrammatical deletion Los ninos jugaron El vecino ganara el juego
The boy (was) playing The neighbor (will) win the game
(S.E.I.2B) semantic deletion en el bano temprano No encontro papel en la caja
in the bathroom (early) I didn't find paper (in the box)
(S.E.M.I) case Por la puerta entro Tiene ella un viejo vestido
Him through the door Has her one old dress
(S.E.M.2) agreement Empiezan las clases Pero entiendo poco
The classes starts but he understands a little bit
(S. E. M.3) part of speech oramos por la comida porque dice que yo soy tonta
we prayer for the food because I'm too dummy
(S.E.T.) tense esta fria Luego, se vieron
is gonna be cold Then they were seeing
158
Translation skill and metalinguistic awareness Table 7.3. Cont. (S.E.O.) order ^Donde estara mi hermano? El bebe triste
Where should be my brother? That baby sad
(5. E. P.) paradigmatic sus grandes pies crecen rapidamente
his big feets grow fastly
(S.E.S.) subject El termina temprano Llam6 mi madre
I finished early / called my mother
English to Spanish (E.S. W.I) articles and quantifiers To a game tomorrow Saw the boy
/1/juego mariana Vio al nino
(E.S.W.2) nouns, verbs, adjectives, adverbs Dejar unpapel Leave a message El telefono casi suena The telephone often rings (E.S.W.3) source We ate ice cream The chicken is ready to eat
Comimos ice cren El polio esta redi para comer
(E. S. W. 4) prepositions toward the mouse wait for me at the table
sobre el raton esperame en la mesa
(E.S.I. 1 A) addition ungrammatical You can tell my teacher I can't study
Tu puedes decir/e a mi maestra No puedo a estaudiar
(E.S.I.IB) addition semantic She saw the word Maria will arrive tomorrow
Ella vio las palabras Maria va a llegar aqui manana
(E.S.1.2A) deletion ungrammatical To visit her friends The door . . . opened
A visitar (a) sus amigos La puerta . . . (se) abrio
(E.S. 1.2B) deletion semantic My neighbor goes to the store to buy fish The problem was too hard to understand
159
Mi vecino va (a la tienda) a comprar pescado El problema era muy duro (para entender)
Marguerite Malakoff and Kenji Hakuta
Table 7.3. Cont. (E. S.M.I) case None (E.S.M.2) agreement Every day My friends
Todos los dia Mi amigos
(E. S. M.3) part of speech knew that lesson perfectly mother entered quickly
sabia esa Iecci6n muy perfecto mi mama entro el cuarto rdpido
(E.S.T.) tense The teacher told him to read The letter is in the envelope
La maestra le dijo que lea La carta estaba en el sobre
(E.S.O.) order The new desk The famous actor who we saw
La nueva mesa El famoso actor nosotros vimos que
(E.S.S.) subject Make room for me With her friends she played
Cogemos un cuarto Con su amigoyugaron penlota
(E. S. P.) paradigmatic My sister knew that lesson To stop the drinking
Mi hermana sabio esa leccion Que parada de beber
classrooms for the next year, 29 were scheduled to continue in the bilingual program the following year (information was missing for five subjects). Measures The story-translation task from study 1 was chosen because it can be group-administered. In order to compare performances on this task in studies 1 and 2, a detailed error analysis was conducted following a coding scheme found in table 7.3. Two independent judges conducted the coding, and when there was disagreement, it was solved by reaching a consensus. Results A comparison of errors between studies 1 and 2 is presented in table 7.4, along with a test of significance in the difference between the
160
Translation skill and metalinguistic awareness means on each type of error. In the Spanish-English translation direction, there are only two significant differences, going in opposite directions. Study 1 made fewer semantic-addition errors than study 2, but more part-ofspeech errors. The lack of any consistent pattern suggests to us that these significant effects are spurious. In the English-Spanish translation direction, four significant differences emerge, all showing fewer errors by study 2 subjects. Thus, the less selected subjects provided better English-Spanish translations. This somewhat surprising finding is explained by the fact that all of the study 2 subjects are in bilingual-education classes, where instruction in Spanish is provided, while this was not the case for study 1 subjects. Since the bilingualism of the community is generally subtractive in nature, it is not surprising that those students who are not in the bilingual classes reach a plateau in their Spanish proficiency. A comparison of the error patterns between the two groups of subjects showed good consistency. The correlation between the two groups of their respective percentage of each error type showed r = 0.87 for the SpanishEnglish translations, and r = 0.91 for the English-Spanish translations. Discussion The results suggest that translation ability is a widespread skill among students in this population. Unfortunately, due to limitations in resources, we could not administer any measures of response latency. However, the error patterns in the written task revealed similarities between the two groups, and the more selected group in study 1 did not enjoy any general advantages in translation ability.
Conclusions The findings from these studies demonstrate that translation skill is widely found in bilingual children by late elementary school. Bilingual children are able to translate, albeit with flaws, and their translations reflect their understanding of the communicative importance of translation. That is, when the quality of the translation suffers, the errors are usually in sentence structure and not in meaning. Furthermore, translation efficiency and quality appear to vary according to a number of dimensions, including targetlanguage proficiency, processing-window size (as suggested by the difference between the written and oral tasks), and translation proficiency. Although we have only begun to scratch the surface of the empirical relationships between the different abilities involved in translation, there are already some exciting implications of our research beginning to emerge. The fact that natural translation is an ability to be expected of bilingual children suggests that its use as a tool for both research and language161
Marguerite Malakoff and Kenji Hakuta Table 7.4 Comparison offrequencies of error types in the story-translation task across subjects from study 1 and study 2 %
articles, etc. nouns, etc. source prepositions addition ungrammatical addition semantic deletion ungrammatical deletion semantic agreement case part of speech possessive order subject paradigmatic tense
articles, etc. nouns, etc. source prepositions addition ungrammatical addition semantic deletion ungrammatical deletion semantic agreement case part of speech possessive order subject paradigmatic tense
M.
S.D.
Spanish to English Study 2 N = 52 14 2.17 3.77 17 2.62 2.10 0.63 2 0.27 9 1.39 1.17 4 0.67 1.26 4 0.58 0.87 2 0.33 0.56 10 1.52 1.58 3 0.46 0.78 0 1 <1 13 2 3 16
0
0
0.08 0.21
0.27 0.41
1.9
1.9
0.35 0.46 2.42
0.48 0.73 2.29
English to Spanish Study 2 N = 52 8 1.08 0.84 34 4.71 2.48 1.01 6 0.87 1 0.19 0.40 6 0.87 1.21 6 0.81 1.34 0.63 3 0.39 13 1.85 1.66 0.94 6 0.77 < 1 0.02 0.14 0 0 9 1 1 6
0 0
0 0
1.31 0.10 0.15 0.81
0.98 0.30 0.36 0.91
Notes: * P< 0.05; ** P< 0.01; ***/>< 0.001.
162
%
M.
S.D.
t
Study 1 ^J = 14 7 0.93 22 3 3 0.43 10 1.43 1 0.14 3 0.36 1 0.21 6 0.86 3 0.36 0 0 5 0.64 < 1 0.21 10 1.36 2 0.29 4 0.57 22 3
0.62 1.23 2.83 -0.56 1.34 -0.66 1.28 -0.11 0.36 1.55 0.63 2.75** 0.43 0.75 1.43 1.35 0.43 0.75 0 0 0.84 -4.31*** 0 0.42 0.95 1.99 0.43 0.47 0.85 -1.42 2.11 -0.85
Study 1 N = 16 5 25 10 3 11 5
1
5.15 2
0.62 2.31 1.0
5 0.92 19 3.77 3 0.54 0 <1 0 6 1 3 4
0.82 1.28 1.78 0.87 1.60 1.26 1.04 1.83 0.78
0
0
0.08
0.28
0
0
1.23 0.15 0.69 0.85
0.60 0.38 1.12 0.80
0.33 -0.69 -1.03 -2.87** -3.89*** -0.51 -2.52* -4.00*** 0.88 0 0 0 0.31 -0.56 -3.18** 0.16
Translation skill and metalinguistic awareness proficiency assessment is viable, as suggested fifteen years ago by Swain, Dumas, and Naiman (1974), and should be aggressively pursued. For example, we have begun using a task in which subjects provide judgments of the goodness of translations, which we believe taps deeply into metalinguistic skill. Another exciting extension is in the area of pedagogy. We have worked in a number of classroom settings (e.g., see Shannon, in press) to use translation ability as a psycholinguistic and sociolinguistic hook into amplifying the bilingual skills of students. Translation provides an easy avenue to enhance linguistic awareness and pride in bilingualism, particularly for minority bilingual children whose home language is not valued by the majority culture. In sum, the study of translation provides a superb research "preparation," in which basic psycholinguistic and sociolinguistic issues can be integrated with educational practice. Note 1.
Support for this paper was provided in part by a subcontract from the Center for Language Education and Research (CLEAR) to Kenji Hakuta, and in part by the Bilingual Research Group at the University of California, Santa Cruz. We are grateful to the following individuals: Marcus Rivera and Margarita Rodriguez-Lansberg, who helped in data collection; Daniel d'Andrea and Julia Kushner, who helped code the data; Hortensia Calvo, who helped develop the translation stimuli; Laurie Gould, who helped in the development of the error coding; and Jose Capuras, who wrote the computer program to collect translation-time data. Correspondence should be addressed to Kenji Hakuta, School of Education, Stanford University, Stanford, California 94305.
References Asuncio-Kibdem, N. C. 1981. Intercultural communication. In N. C. AsuncioKibdem & E. M. Pacasio (eds.), Building bridges across cultures. Perspectives on intercultural communication: theory and practice. Manila, Philippines: Solidad. Ballard, M. 1984. La Traduction reteve-t-elle d'une p^dagogie? In M. Ballard (ed.), La traduction: de la theorie a la didactique. Lille: University of Lille III. Ben-Zeev, S. 1977. The influence of bilingualism on cognitive strategy and cognitive development. Child Development, 48,1009-1018. Bergman, C. D. 1976. Interferences vs. independent development in infant bilingualism. In G. Keller, R. Teschner, & S. Viera (eds.), Bilingualism in the bicentennial and beyond. New York: Bilingual Press/Editorial Bilingue. 163
Marguerite Malakoff and Kenji Hakuta Bialystok, E. 1988. Levels of bilingualism and levels of linguistic awareness. Developmental Psychology, 24, 560-567. Bialystok, E. & E. B. Ryan 1985. Toward a definition of metalinguistic skill. MerrillPalmer Quarterly, 31,229-251. Brumfit, C. 1984. Communicative methodology in language teaching: the roles of fluency and accuracy. Cambridge: Cambridge University Press. Carroll, J. B. 1978. Linguistic abilities in translators and interpreters. In D. Gerver & H. W. Sinaiko (eds.), Language interpretation and communication. New York: Plenum Press. Catford, J. C. 1965. A linguistic theory of translation. Oxford: Oxford University Press. Cazden, C. B. 1976. Play with language and metalinguistic awareness: one dimension of language experience. In J. S. Bruner, A. Jolly, & K. Sylva (eds.), Play. New York: Penguin. Chan, Mun-Chee, H. L. H. Chau, & R. Hoosain 1983. Input/output switch in bilingual code switching. Journal of Psycholinguistic Research, 4, 407-415. Fischer, K. & S. Pipp 1984. Processes of cognitive development: optimal level and skill acquisition. In R. Stern berg (ed.), Mechanisms of cognitive development. New York: W. H. Freeman. Flesch, R. 1972. The pursuit of translation. In R. Flesch (ed.), The art of clear thinking. New York: Macmillan. Fuchs, C. 1982. La Paraphrase. Paris: Presse Universitaire de France. Galambos, S. & K. Hakuta 1988. Subject-specific and task-specific characteristics of metalinguistic awareness in bilingual children. Applied Psycholinguistics, 9,141-162. Genesee, F. 1987. Bilingual development: a critical view. Unpublished manuscript. Gleitman, L. & H. Gleitman 1970. Phrase and paraphrase. New York: Norton and Norton. Grosjean, F. 1982. Life with two languages. Cambridge, MA: Harvard University Press. Gumperz, J. 1982. Discourse strategies. New York: Cambridge University Press. Hakuta, K. 1986. The mirror of language. New York: Basic Books. 1988. Language and cognition in bilingual children. CLEAR Progress Report. Hakuta, K. & M. Malakoff 1987. Translation skills in bilingual children. Paper presented at the Society for Research in Child Development, Baltimore, MD, April. Hakuta, K., L. Gould, M. Malakoff, M. Rivera, & M. Rodriguez-Landsberg 1988. Translation and interpretation in bilingual Puerto-Rican students. Paper presented at AERA, New Orleans, April. Harley, B. 1986. Age in second language acquisition. Clevedon, Avon, England: Multilingual Matters. Harris, B. 1977. The importance of natural translation. Working Papers in Bilingualism, 72,96-114. 1980. How a three-year-old translates. In Patterns of bilingualism. RELC anthology series 8. Singapore: National University of Singapore Press. Harris, B. & B. Sherwood 1978. Translating as an innate skill. In D. Gerver & 164
Translation skill and metalinguistic awareness H. W. Sinaiko (eds.), Language interpretation and communication. New York: Plenum Press. Ianco-Worrall, A. 1972. Bilingualism and cognitive development. Child Development, 43, 1390-1400. Kade, O. 1968. Zufallund Gesetzmassigkeit in der Ubersetzung. Berlin, Schonberg: Langenscheidt. Katz, J. J. & J. A. Fodor 1963. The structure of semantic theory. Language, 39, 170-210. Kintsch, W. & T. A. Van Dijk 1978. Toward a model of text comprehension and production. Psychological Review, 85, 363-394. Lambert, W. 1975. Culture and language as factors in learning and education. In A. Wolfgang (ed.), Education of immigrant students. Toronto: Ontario Institute for Studies in Education. Leopold, W. 1939-1949. Speech development of a bilingual child: a linguist's record (4 vols.). Evanston, IL: Northwestern University Press. Lindholm, K. & A. M. Padilla 1978. Child bilingualism: report on language mixing, switching, and translation. Linguistics, 211,23-44. Ljudskanov, A. 1969. Traduction humaine et traduction mecanique. Paris: Dunod. Marcos, L. R. & M. Alpert 1976. Strategies and risks in psychotherapy with bilingual patients: the phenomenon of language independence. American Journal of Psychiatry, 133,1275-1278. Marcos, L. R. & L. Urcuyo 1979. Dynamic psychotherapy with bilingual patients. American Journal of Psychotherapy, 33, 331-338. Mininni, G. 1981. On the need of psycholinguistic research on translation. International Journal of Psycholinguistics, 8, 79-101. Nida, E. A. 1976. A framework for the analysis and evaluation of theories of translation. In R. W. Brislin (ed.), Translation: applications and research. New York: Gardner Press. Paivio, Clark & W. E. Lambert (1979). Bilingual coding theory and semantic repetition effects on recall. Unpublished manuscript, University of Western Ontario. Peal, E. & W. Lambert 1962. The relation of bilingualism to intelligence. Psychological Monographs, 76 (whole no. 546). Pergnier, M. 1978. Language meaning and message meaning: towards a sociolinguistic approach to translation. In D. Gerver & H. W. Sinaiko (eds.), Language interpretation and communication. New York: Plenum Press. 1984. La traduction, les structures linguistiques et le sens. In M. Ballard (ed.), La traduction: de la theorie a la didactique. Lille: University of Lille III. Poplack, S. 1979. Sometimes I'll start a sentence in Spanish y termino en Espanol: Towards a typology of code-switching. Working paper No. 4. Centro de Estudios Puertorriquenos, New York. Scotton, C. M. & W. Ury 1977. Bilingual strategies: the social function of codeswitching. International Journal of the Sociology of Language, 13, 5-20. Seleskovitch, D. 1976. Interpretation, a psychological approach to translation. In R. W. Brislin (ed.), Translation: applications and research. New York: Gardner Press. 165
Marguerite Malakoffand Kenji Hakuta 1978. Language and cognition. In D. Gerver & H. W. Sinaiko (eds.), Language interpretation and communication. New York: Plenum Press. Seleskovitch, D. & M. Lederer 1984. Interpreter pour traduire. Paris: Publications de la Sorbonne. Shannon, S. M. 1987. English in the Barrio: a sociolinguistic study of second language contact. Unpublished doctoral dissertation, Stanford University, Stanford, in press. Spanish for Spanish speakers: a curriculum involving translation skills. In A. Padilla, C. Valdez, & H. Fairchild (eds.), Advances in language education. New York: Sage. Sridhar, S. N. & K. K. Sridhar 1980. The syntax and psycholinguistics of bilingual code-mixing. Canadian Journal of Psychology, 34, 407-416. Steiner, G. 1975. After Babel: aspects of language and translation. London: Oxford University Press. Swain, M. 1972. Bilingualism as a first language. Unpublished doctoral dissertation, University of California, Irvine. Swain, M. & M. Werch 1975. Linguistic interaction: case study of a bilingual child. Language Sciences, 37,17-22. Swain, M., G. Dumas, & N. Naiman 1974. Alternatives to spontaneous speech: elicited translation and imitation as indicators of second language competence. Working Papers on Bilingualism, 3, 68-79. Tunmer, W. E. & M. L. Herriman 1984. The development of metalinguistic awareness: a conceptual overview. In W. E. Tunmer, C. Pratt, & M. L. Herriman (eds.), Metalinguistic awareness in children. New York: Springer. Vinay, J.-P. 1975. The theory of translation: myth or reality. In M. S. Batts (ed.), Translation and interpretation: the multicultural context. A symposium. Vancouver: CAUTG. Vygotsky, R. L. 1962. Thought and language. Cambridge, MA: M.I.T. Press. Weinreich, U. 1953. Languages in contact. The Hague: Mouton. Wilss, W. 1976. Perspectives and limitations of a didactic framework for the teaching of translation. In R. W. Brislin (ed.), Translation: applications and research. New York: Gardner Press. 1982. The science of translation. Tubingen: Gunter Narr. Zentella, A. 1981. Language variety among Puerto Ricans. In C. A. Ferguson & S. B. Heath (eds.), Language in the USA. New York: Cambridge University Press.
166
8
Towards an explanatory model of the interaction between bilingualism and cognitive development R A F A E L M. D I A Z and C Y N T H I A
KLINGLER
During the past twenty years research has shown, with some degree of consistency, that learning a second language in childhood, either by simultaneous acquisition or in the context of bilingual education, is associated with positive cognitive gains. In both bilingual-monolingual comparisons and in studies using "within-bilingual" designs, children's bilingualism is positively related to concept formation, classification, creativity, analogical reasoning, and visual-spatial skills, to name a few (Diaz, 1983; Hakuta, Ferdman, & Diaz, 1987). In addition, as is evident in several chapters of the present volume, bilingual children have demonstrated a particularly refined awareness and control of the objective properties of language, commonly referred to as metalinguistic skills. Ben-Zeev (1977), for example, found that bilingual children approached linguistic tasks with a special sensitivity to language structure and detail. More recently, Bialystok (1986) has shown that children's bilingualism positively affects their increasing ability to solve problems involving high levels of control of linguistic processing. Even though we have substantial documentation of the cognitive and metalinguistic advantages of childhood bilingualism, an important issue remains unresolved; namely, researchers have not yet developed and tested the validity of an explanatory model of how or why bilingualism has such positive effects. To date, it is not clear, for example, how bilinguals' metalinguistic skills are related to advantages in cognitive abilities not directly related to language, such as classification or visual spatial skills. The present chapter proposes an explanatory model of the relation between bilingualism and cognitive abilities that specifies the role of language awareness in the development of non-linguistic cognitive skills. Several researchers in the past have formulated different explanatory hypotheses when discussing their positive findings (see Diaz, 1985a, for a review). Perhaps the most widely known is Peal and Lambert's (1962) code-switching hypothesis. These investigators believed that the possibility of switching linguistic codes while performing cognitive tasks gave bilingual 167
Rafael M. Diaz and Cynthia Klingler children an added flexibility that monolingual children did not enjoy. In their own words: The second hypothesis is that bilinguals may have developed more flexibility in thinking . . . bilinguals typically acquire experience in switching from one language to another, possibly trying to solve a problem while thinking in one language and then, when blocked, switching to the other. This habit, if it were developed, could help them in their performance on tests requiring symbolic reorganization since they demand a readiness to drop one hypothesis or concept and try another. (Peal & Lambert, 1962, p. 14)
The code-switching hypothesis has been quite influential in the field, giving birth to one of the most popular concepts regarding bilinguals' cognitive advantages, namely, bilinguals' cognitive flexibility. However, as pointed out by Diaz (1983), the code-switching hypothesis is limited by the fact that there is no empirical evidence (beyond anecdotal accounts of questionable validity) that bilinguals indeed switch languages while performing cognitive tasks. Moreover, the concepts of individual code-switching during problem-solving activity (in contrast to code-switching in communicative social situations) and its resulting cognitive flexibility have not been formulated in the context of current theoretical accounts on the relation between thought and language. Vygotsky's (1962) concept of verbal thinking or Luria's (1976) accounts of the'role of speech in the formation of higher psychological functions would both appear to be relevant to this problem. It is our belief that any successful explanation of the interaction between bilingualism and cognitive development must fulfil two basic requirements: first, the model should be formulated, developed, and tested within a solid theoretical framework regarding the relation between language and thought in development; second, the model should be constrained by the available data. In other words, the model should be developed in order to explain the reliable findings to date on bilingual cognitive development. In order to fulfil the first requirement, we have selected Vygotsky's (1962, 1978) theory of thought and language, especially his emphasis on the selfregulation of cognitive functions through the use of private speech. There are at least four reasons why Vygotsky's theory seems especially suited for the task at hand, namely: 1.
2.
Among current theories on the relation between thought and language (see Hakuta, 1986, for a brief review) Vygotsky's theory uniquely postulates that the use of language promotes a radical transformation of the course of cognitive development. In the context of his theory, the diversity of linguistic experiences has differential effects on cognitive abilities (Luria, 1976). In Vygotsky's theory, the relation between thought and language changes at different times in development. The theory is in essence a developmental 168
Towards an explanatory model one and, therefore, uniquely suited to understanding the effects of childhood bilingualism on cognitive development. 3. Vygotsky's theory places special emphasis on the increasing executive control and self-regulation of cognitive processes through the uses of language in private speech. As Bialystok (this volume) suggests, the metalinguistic skills observed in bilingual children are based on control of different aspects of language processing and this may be related to an executive. 4. Finally, Vygotsky's theory provides a specific account of the relation between social and psychological processes in development (Wertsch & Stone, 1985). The theory is uniquely qualified to guide our understanding regarding the relation between bilingual situations and bilingual minds. In order to fulfil our second requirement for the development of an explanatory model, we review the literature in search of findings that must be explained. In what follows, we discuss six different sets of findings regarding the relation between bilingualism and cognitive development. These findings are presented as data points that both constrain and direct the model-building effort.
Findings to be explained
Cognitive advantages Research on the relation between bilingualism and intelligence has been plagued by a number of methodological shortcomings (see review by Diaz, 1983; Hakuta, 1986). However, in our opinion, there are two research paradigms that provide reliable findings on the efforts of bilingualism on cognitive development. The first acceptable paradigm is modeled after Peal and Lambert's (1962) seminal study. This paradigm involves a comparison between monolingual and balanced bilingual children. Two major requirements are involved. The first requirement is that children in the bilingual sample must demonstrate some degree of "balance" or comparable proficiency in the two languages. Even though there is no general consensus of what constitutes "balanced bilingualism" in absolute terms, a useful working definition is that balanced bilingual children demonstrate age-appropriate abilities in both languages. The second requirement is that researchers using this paradigm should attempt to make the two groups (bilingual-monolingual) comparable on relevant variables, such as socioeconomic status, parental education, years of schooling, or any other possible confounding variables that could make the two groups different for reasons other than bilingualism. The second acceptable paradigm has been labeled the "within-bilingual" design (see, e.g., Hakuta & Diaz, 1985). The aim of this paradigm is to 169
Rafael M. Diaz and Cynthia Klingler examine the effects of bilingualism by studying children who vary in their second-language proficiency. In these studies, children's relative proficiency in the second language or their "degree of bilingualism" constitutes the independent variable; different cognitive measures constitute the dependent variable. An example of this kind of study is to assess the cognitive abilities of three different groups of children: (a) proficient bilinguais, (b) partial bilinguais, and (c) limited bilinguais. The main requirement for this approach is that children in all three groups must be equivalent in their first-language ability but widely different in their second-language proficiency. Group differences in cognitive ability, therefore, could be attributed to children's varied second-language ability or relative bilingual proficiency. Another example of the within-bilingual design is to use a multiple-regression approach, where cognitive ability is the criterion variable and children's proficiency in their first and second language the predictor variables. The effects of bilingualism can be assessed by examining the relation between second-language proficiency and cognitive ability once the effects of the first-language ability have been partialed out from the equation (see Hakuta & Diaz, 1985). Since Peal and Lambert's study (1962), a number of studies have shown that bilingualism can have a positive influence on the acquisition and development of different cognitive skills. Using the two research paradigms discussed above, investigators have reported bilingual advantages on concept formation (Liedtke & Nelson, 1968), creativity (Torrance etai, 1970), Piagetian conservation tasks (Duncan & De Avila, 1979), visual-spatial abilities (Hakuta & Diaz, 1985), analogical reasoning (Diaz, 1985b), and classification skills (Diaz & Padilla, 1985). Even though each of these studies could be criticized on several methodological and conceptual grounds, the convergence of positive findings across different measures and designs is impressive. Since the findings of these and other studies have been recently reviewed elsewhere (see, e.g., Hakuta, Ferdman, & Diaz, 1987), we will briefly review the last three studies reported in 1985. The three studies are reviewed below not only as they apply to our model-building endeavor, but also to synthesize for the reader the types of cognitive effects that have been observed in bilingual children. Hakuta and Diaz (1985) tested the relation between degree of bilingualism and cognitive abilities in a sample of 123 Puerto Rican children. At the time of testing, children in the sample were enroled in kindergarten and first-grade bilingual-education programs. Degree of bilingualism was operationalized as variation in second-language ability (English), controlling for variation in the first language (Spanish). The study showed a significant positive relation between degree of bilingualism and performance on the Raven Progressive Matrices, a well-known non-verbal test of cognitive ability. The positive relation between bilingualism and Raven performance 170
Towards an explanatory model remained significant even when potentially confounding variables, such as age and socioeconomic factors, were controlled in the analyses. Also, since first-language ability was measured with the Spanish Peabody Picture Vocabulary Test (P.P.V.T.), individual differences in basic ability were also controlled in the analyses. Hakuta and Diaz' data are important on several counts. First, the study makes a significant contribution to the operationalization of degree of bilingualism. The authors' methodological suggestions make it possible to assess the amount of variance explained in cognitive abilities by children's degree of bilingualism, while controlling for potentially confounding variables. Second, using repeated observations of variables across time, the study examined the direction of causality between bilingualism and cognitive ability, testing two possible causal models through multipleregression techniques. The data supported the model where children's degree of bilingualism is the possible causal factor. Finally, the study documented positive effects of bilingualism in a sample of immigrant children of low socioeconomic status. The study showed that the potentially disruptive effects associated with folk or subtractive bilingualism (Fishman, 1977) could be prevented through bilingual-education programs, where both languages are supported and develop in an additive fashion. In further analyses of the same data base, Diaz (1985b) reported on the effects of bilingualism on children's analogical reasoning. In this report, 100 children from the original sample were divided into extreme groups of low and high second-language (English) proficiency; the two groups were equivalent on first-language (Spanish) proficiency. Analogical reasoning was measured using a modified version of the Stanford-Binet Intelligence Scale subtest of opposite analogies. Children were instructed to verbally fill in the blank immediately after the experimenter had read aloud an incomplete analogy, such as "The princess is beautiful, the monster is " In order to obtain a passing score for that item, children had to respond saying the word "ugly." All the items were administered in Spanish (i.e., Laprincesa es bonita, elmonstruo esfeo). The following results were obtained: (1) At time 1, children with higher second-language (English) proficiency scored significantly higher on the analogical-reasoning test given in Spanish; when socioeconomic variables were taken into account, however, the differences were only marginally significant. (2) Degree of bilingualism explained a significant portion of analogical-reasoning variance only for children in the low second-language proficiency group (a puzzling, and seemingly contradictory, finding that will be discussed later in the section (see pages 177-180)). (3) At time 2, initial differences in analogical reasoning between the two proficiency groups disappeared, showing that children of low second-language proficiency caught up in their analogical-reasoning skills. Since second171
Rafael M. Diaz and Cynthia Klingler language proficiency predicted significant portions of the analogical-reasoning variance within the low-proficiency group, the catch-up at the end of the year could be attributed to their increased bilingualism. Even though the findings reported in Diaz (1985b) are complex (e.g., varying for children at different stages of second-language learning), the multiple-regression equations showed a strong relation between degree of bilingualism and analogical reasoning. Specifically, for the low Englishproficiency group, the second-language variable explained 12 percent (at time 1) and 6 percent (at time 2) of the variance in the analogies measure. Similar results were obtained for the Raven Progressive Matrices (4 percent and 24 percent of the variance) and for two tests of visual spatial abilities (9 percent and 11 percent of the variance). In a different sample of Mexican-American bilingual preschoolers, Diaz and Padilla (1985) reported positive effects of bilingualism on several ageappropriate tasks of cognitive ability. A total of thirty-four preschoolers who attended Spanish-English bilingual preschool programs in El Paso, Texas, were videotaped while performing three different tasks: block designs, classification, and story-sequencing tasks. After brief instructions, children were asked to work on their own for a period of five minutes for each task. The effects of bilingualism on performance on the three tasks were analyzed in multiple-regression equations controlling for ability in the first language and months of preschool education. Children's degree of bilingualism, operationalized as suggested by Hakuta and Diaz (1985), predicted significant portions of performance variance in both classification and story-sequencing tasks. The effects, however, did not meet the significance criterion for the block design task. Some interesting effects were also found on children's private speech during the tasks; these effects will be discussed later in the chapter. In sum, since Peal and Lambert's (1962) study, research has frequently documented advantages of bilingualism on several cognitive abilities. The advantages have been shown on different tasks, including psychometric and experimenter-made tests of both verbal and non-verbal abilities. The positive results have appeared in studies comparing balanced bilingual children and in studies that have used within-bilingual designs. Cognitive advantages have appeared for children who are simultaneous learners of both languages and for preschool and school-age children in the context of bilingual-education programs. Let us now turn to a brief discussion of the effects of bilingualism on metalinguistic skills. Metalinguistic abilities Guided by theoretical statements from the works of Vygotsky (1962) and Leopold (1939-1949), researchers have explored the effects of bilingualism 172
Towards an explanatory model on children's metalinguistic awareness. Even though workers in the field recognize that all children, bilingual and monolingual, increasingly become aware of language as an objective and arbitrary sign system, they hypothesize a bilingual advantage in metalinguistic development. Vygotsky believed that, seeing their language as one system among many, bilingual children enjoyed a special awareness and control over their linguistic operations. Following closely the bilingual development of his daughter Hildegard, Leopold observed an early separation of word and referent as a result of the bilingual experience. Both Vygotsky and Leopold hypothesized that bilinguals' refined awareness of language, involving a precocious understanding and manipulation of symbols, played an important role in the mental development of young bilingual children. In current research parlance, metalinguistic awareness refers to a set of abilities involving an objective awareness and control of linguistic variables, such as understanding the arbitrariness of word-referent relations and the capacity to detect and correct syntactical violations. Moreover, metalinguistic awareness is seen as a crucial component of cognitive development because of its documented relation to language ability (Smith & TagerFlusberg, 1982), symbolic development (Van Kleek, 1982) and literacy skills (Bowey, 1986). Research has shown positive effects of bilingualism on the following metalinguistic abilities: early word-referent distinction (Ianco-Worrall, 1972); sensitivity to language structure and detail (BenZeev, 1977); detection of ambiguities and analysis of tautological sentences (Cummins & Mulcahy, 1978); syntactic orientation in sentence processing (Galambos, 1985); correction of ungrammatical sentences and detection of language mixing (Diaz, 1985b); control of language processing (Bialystok, 1986); understanding of referential and non-syncretic arbitrariness of language (Edwards & Christophersen, 1988). Even though some researchers have suggested caution in the interpretation of results on account of methodological shortcomings (see, e.g., Bowey, 1986), the consistency of positive findings across different samples, measures, and research designs provides substantial support to Vygotsky's (1962) and Leopold's (1939-1949) claims. Recent analyses of different metalinguistic tasks have yielded an important distinction between tasks that demand analysis of language knowledge on one hand, and tasks that demand control of linguistic processing on the other (Bialystok, 1986). Some metalinguistic tasks such as grammatically judgments, for example, require an objective awareness and analysis of implicit language knowledge. Other tasks, however, require not only explicit knowledge but also a certain degree of control of linguistic processing, such as "the ability to switch back and forth between form and meaning" (p. 499). Bialystok has demonstrated bilinguals' metalinguistic advantage, especially for items that demand a high degree of control. In 173
Rafael M. Diaz and Cynthia Klingler her two studies, bilingual children performed better than monolinguals in grammaticality judgments of sentences that were grammatically correct but had anomalous meanings. Judging the grammaticality of correct but anomalous sentences demands a great deal of control, since the meaning of the sentence must be suppressed or overlooked in order to focus exclusively on aspects of form. Similar abilities in the control of linguistic processing have been found in studies of bilingual children by Ben-Zeev (1977), and Galambos and Hakuta (1988). Ben-Zeev compared the performance of bilingual and monolingual children on a symbol-substitution task. During the task, administered in a game-like fashion, children had to substitute words in conversation at the experimenter's request. For example, the experimenter would ask the child to substitute the word "I" for the word "Macaroni." In response to the question, "How do you say 4I am warm'?" children had to answer: "Macaroni am warm" in order to obtain a correct score. In this task, children were asked to violate the rules of grammar and inhibit a more automatic, spontaneous response such as "Macaroni is warm," which, though grammatically correct, violated the rules of the game. Bilingual children outperformed their monolingual counterparts by a large margin in the symbol-substitution task. The capacity to produce ungrammatical sentences upon request requires a high degree of executive control of language processing, considering the automaticity involved in the usual production of grammatically correct utterances. Different results are obtained for metalinguistic tasks involving different kinds of processing. In an examination of paradigmatic word associations, a task that reflects children's knowledge and objective analysis of language, no bilingual-monolingual differences were found. In a more recent study by Galambos and Hakuta (1988), the investigators assessed children's detection of three types of ambiguities: homophonous (pearslpairs), polysemous (bark/bark), and phonetically ambiguous (engineer/engine ear). The results suggested first that the three types of ambiguities differed in their level of difficulty, where phonetically ambiguous items were hardest to detect and polysemous items the easiest. Galambos and Hakuta interpreted the difference between item types in terms of their differential demands on control of processing: A possible explanation for the difference in difficulty of the items observed in our experiment can also be found in the information-processing literature. From this perspective, a prediction could be made that, to detect phonetic ambiguity, it would be necessary to encode the construction quite exhaustively in order to be able to restructure the information. A fair bit of cognitive control would be required to accomplish this. Detecting polysemy, on the other hand, would be much easier. Only a small amount of cognitive control would be required to reinterpret the meaning of the word, as automatized procedures would probably 174
Towards an explanatory model already have been developed to access familiar meanings of a word. Finally, the link between the meanings of homophonous items would probably be less strong and less automatized than that for polysemous items. (Galambos & Hakuta, 1988, p. 158)
Using a "within bilingual" design, Galambos and Hakuta examined the effects of bilingualism on the detection of different types of ambiguity. The study included Spanish-dominant children who were learning English in the context of Spanish-English bilingual-education programs, and involved longitudinal observations at two points in time. The findings favored children of a higher degree of bilingualism, for phonetically ambiguous and polysemous items. However, the effects did not reach the specified level of statistical significance. The investigators found the largest significant improvement over time for children of a higher degree of bilingualism on the phonetically ambiguous items. Taking together the results of Ben-Zeev (1977), Bialystok (1986), and Galambos and Hakuta (1988), an interesting picture begins to emerge regarding the metalinguistic skills of bilingual children. Namely, that bilingualism positively affects children's executive control of language processing. It is possible that the systematic separation of form and meaning that is experienced in an early bilingual experience gives children an added control of language processing, as the works of Vygotsky (1962) and Leopold (1939-1949) had suggested. Above all, the findings just reviewed suggest that any model of the relation between bilingualism and intelligence must pay special attention to the effects of bilingualism on cognitive executive functions. Additive and subtractive situations A third finding that must be taken into account is the fact that the positive effects of bilingualism are closely connected to additive bilingual situations (Cummins, 1976). Additive bilingualism refers to situations where both languages are supported and develop in parallel. Subtractive situations, in contrast, are characterized by a gradual loss of the first language as a result of increasing mastery and use of the second language. Within subtractive situations, at a given point in the process of one language replacing the other, children might appear as semi-linguals, that is, as seriously limited in both languages. Even though the construct of "semi-lingualism" has been heavily criticized (Martin-Jones & Romaine, 1986), there is indeed some evidence that subtractive situations are related to academic difficulties and cognitive disadvantages in immigrant children (Skutnabb-Kangas, 1984). It is not clear, however, whether these findings reflect some permanent negative effects on cognitive development as a result of the subtractive experience. An alternative explanation is that, in the process of language 175
Rafael M. Diaz and Cynthia Klingler shifting, children experience verbal limitations in both languages that negatively affect their test performance temporarily. It is no surprise, therefore, that the first positive findings reported in the psychological literature involved "balanced" bilingual children, or children who had achieved similar age-appropriate abilities in the two languages. Clearly, balanced bilingualism is the result of additive situations, while "pseudo-bilingualism," a term used by Peal and Lambert (1962) to describe children in the bilingual samples of early studies in the field, is most commonly associated with subtractive situations. "The pseudobilingual knows one language much better than the other, and does not use his second language in communication" (p. 6). In fact, the majority of studies that have reported positive effects of bilingualism have involved samples in additive-bilingual situations such as simultaneous learners of the two languages or children in bilingual-education programs that involve some kind of systematic use and support of the two languages. One possible explanation for why the positive effects of bilingualism occur only in additive situations was formulated by Cummins (1976) in his threshold hypothesis. According to Cummins, the positive effects of bilingualism should occur only after the child has achieved a certain threshold of competence in the second language. Moreover, he suggested that the bilingual experience could have negative effects for those children whose proficiency in the two languages remains below a certain threshold of acceptable competence. Low levels of proficiency in the two languages tend to occur in subtractive situations, where there is low proficiency in the language of schooling and the first language is not promoted through literacy or instruction. According to the threshold hypothesis, in order for bilingualism to have a positive effect on intellectual development, a relatively high degree of proficiency in the two languages must be attained. Needless to say, such a high degree of bilingual competence, while being the stated goal of additive situations, could not be expected in subtractive situations. The importance of additive situations for the positive effects of bilingualism has been questioned by two recent findings in the research literature, adding complexity to the issues. One source of complexity is that positive effects of bilingualism have been found for children in transitional bilingualeducation programs within communities that are experiencing subtractive effects (Hakuta, 1987). U.S. bilingual-education programs are described as "transitional" because their stated goal and actual implementation involve the rapid transition to monolingual English education with minimal secondary support of the first language. As Hakuta (1986, 1987) has convincingly argued and demonstrated, transitional models of bilingual education foster subtractive features, such as the gradual replacement of languages and eventual loss of the native language. An important corollary 176
Towards an explanatory model is that children in transitional bilingual programs do not usually attain balanced bilingualism. Nonetheless, positive findings of bilingualism have been found for Puerto Rican children in transitional bilingual-education programs, even though these children were not balanced bilinguals (and probably never would be) and lived in communities experiencing a language shift in a subtractive fashion (Diaz, 1985b; Hakuta, 1987; Hakuta & Diaz, 1985). These findings suggest that transitional bilingual programs could provide, at least temporarily, some additive features (an "additive oasis,'* in the words of Kenji Hakuta) so that, even within a larger subtractive situation, the bilingual experience could produce positive cognitive effects. One immediate consequence of these findings for our present purpose is that any explanation of positive effects should describe additive bilingual situations in terms of additive features in the child's immediate language-learning and educational context, rather than in terms of ultimate consequences regarding the future maintenance of the two languages. A second source of complexity regarding the mediating effects of additive situations in the relation between bilingualism and cognitive development is that, unlike what is expected in Cummins' threshold hypothesis, the positive effects of bilingualism seem to occur more strongly for children at early stages of bilingual development and attenuate as children become more proficient bilinguals. We now turn to this set of findings.
Timing of positive effects In a recent longitudinal study of the effects of bilingualism on cognitive ability, Hakuta (1987) made repeated observations of children's language and cognitive abilities over a three-year period. The sample consisted of approximately 200 Puerto Rican children enroled in bilingual-education programs, where both the first (Spanish) and second (English) languages were used substantially as media of instruction. At the beginning of the study, the youngest subjects attended kindergarten and at the end of the study, the oldest children attended sixth grade. The study, therefore, provides valuable data about the effects of bilingual education throughout the elementary-school years. The most striking finding of Hakuta's study is that bilingualism (defined as bilinguals' ability in the second language controlling for relative ability in the first language) predicts substantial portions of the variance in cognitive ability for younger children in kindergarten and first grade while the effects attenuate for the older children in grades four through six. In fact, no significant effects of bilingualism on cognitive abilities were found for children in grades five and six. For example, in first grade, degree of 177
Rafael M. Diaz and Cynthia Klingler bilingualism explained 22 percent of the variance in Raven's test scores, while in fourth grade it predicted only 4 percent of the variance for the same test. Two possible explanations can account for the attenuation of effects. One explanation is that the effects of bilingualism are mediated by age, and only in the early grades should we expect an interaction between the bilingual experience and the development of cognitive abilities. We will call this explanation the age hypothesis. Because in Hakuta's study older children had a higher degree of bilingualism than younger children (e.g., English vocabulary scores increased linearly with age from a mean of 21.88 in kindergarten to 72.77 in sixth grade), there is an alternative explanation for the attenuation of positive effects: namely, that the effects of bilingualism occur at the beginning stages of second-language learning and, therefore, after a certain threshold level of second-language proficiency, no additional cognitive variance is explained by increasing levels of bilingual proficiency. We shall label this second explanation the level of bilingualism hypothesis. In a study of kindergarten and first-grade children from the same sample, Diaz (1985b) examined the effects of bilingualism separately for children of high and low levels of second-language proficiency, within kindergarten and first-grade age groups. The reported findings give substantial support to the level of bilingualism hypothesis. In both kindergarten and first grade, degree of bilingualism predicted significant portions of the cognitive variance for children of low second-language (English) proficiency level, while the effects attenuated or virtually disappeared for children of high secondlanguage proficiency of the same age. Interestingly enough, within the first grade, bilingualism predicted 24 percent of the variance in Raven's test scores for children of low English proficiency, while it predicted only 4 percent of the Raven variance for children of high English proficiency. This finding replicates, within an age group, the attenuation found by Hakuta across grades. Taking into account the findings from both studies it seems unlikely that the attenuation of effects is simply due to age changes. In fact, Diaz' (1985b) findings give substantial credibility to the level of bilingualism hypothesis. The positive effects of bilingualism reported in Diaz (1985b) were so clearly connected to low levels of second-language proficiency that a new threshold hypothesis was formulated. In contrast to Cummins' threshold hypothesis, Diaz suggested that only before a certain (as yet unspecified) threshold of second-language ability, would bilingualism have a strong impact on cognitive ability. This new threshold hypothesis is somewhat confusing and counterintuitive, so let us explain, using a trite example from academia. As we all know, the number of publications that an assistant professor lists on his or her C.V. carries a lot of weight for a tenure and 178
Towards an explanatory model promotion decision. For example, the difference between listing two versus twelve publications can be the decisive factor in denying or granting tenure. The variability established within a range of ten - between two and twelve publications - therefore, can be of crucial importance and can have a significant impact on the tenure decision. On the other hand, the same variability established within a range of ten at a higher level of productivity, let us say between twenty and thirty publications, is most likely irrelevant for a decision to grant or deny tenure. We would like to make a similar argument for the relation between bilingualism and cognitive abilities. It is very possible that the variability at lower ranges of second-language proficiency (for example, the difference between speaking just a couple of words and producing an original grammatical sentence in the second language) might make a significant difference for children's cognitive abilities, for reasons not yet specified. On the other hand, the same variability established at higher ranges (for example, the difference between producing grammatical sentences and understanding word jokes in a second language), might have importance for social communication or for finding a job at the U.N., but this variability might be completely irrelevant in affecting cognitive abilities. Cummins' and Diaz' threshold hypotheses are difficult to compare and test as alternative explanations with our current data, for several different reasons. First, neither hypothesis has specified the "threshold" in terms of an objective level of language proficiency. Second, Cummins' hypothesis was formulated as a meta-analysis of seemingly contradictory findings from early and more recent studies that compared bilingual and monolingual children. Diaz' hypothesis, on the other hand, was formulated to reflect the differential explanatory power of second-language proficiency in predicting cognitive variance within groups of different levels of bilingual proficiency. Finally, Diaz' hypothesis refers to effects of additive situations, while Cummins' describes the possible effects of both additive and subtractive situations. Above all, we believe that, considering the current state of our knowledge, the two threshold hypotheses provide fertile ground for further research in the field. It is important to note, however, that Diaz' threshold hypothesis does not suggest that the effects of bilingualism disappear or "wash out" with increasing bilingual proficiency. On the contrary, as witnessed by the cognitive advantages of balanced bilingual children, the positive effects are clearly maintained. However, the data do suggest that the effects of bilingualism on cognitive development are most likely mediated through the processes and experiences associated with early phases of second-language learning in an additive context. In addition, even though more research is needed to answer the question of age effects in the relation between bilingualism and intelligence, the findings to date underscore the potential 179
Rafael M. Diaz and Cynthia Klingler benefits of an early bilingual experience. We will now turn to an examination of the effects of bilingualism in a sample of preschoolers. Bilingual private speech A major problem in understanding and explaining the relation between bilingualism and cognitive development is that research has focused mostly on outcome rather than process variables. The exclusive reliance on test scores and outcome data has hindered our understanding of bilinguals' different styles and strategies in cognitive performance. In other words, although we have an abundance of data on the outcome of bilingual children's performance on psychometric and experimental tasks, it is not clear whether bilingual children simply do better (more of the same) than monolinguals or whether they perform differently on the tasks. Needless to say, a clarification of this issue is essential for the development of an explanatory model. One hypothesis concerning bilinguals' different approach to task performance is the code-switching hypothesis (Peal & Lambert, 1962), described in some detail at the beginning of this chapter. However, in a critical review of Peal & Lambert's code-switching hypothesis and its resulting cognitive flexibility, Diaz (1983) noted several assumptions regarding bilinguals' use of language in the solution of non-verbal tasks. Peal & Lambert's codeswitching hypothesis was developed in order to explain bilinguals' cognitive advantages on non-verbal tasks, such as the Raven Progressive Matrices. As Diaz noticed, the code-switching hypothesis involved three underlying and, at the time, untested hypotheses, namely, that (1) bilingual children are thinking verbally while performing the non-verbal tasks; (2) bilinguals switch from one language to the other while performing these tasks; and (3) that bilinguals' habit of switching languages while performing these tasks results in improved task performance. Even though there is substantial evidence that with increasing age children rely on language to solve cognitive tasks, most likely this reliance on language during cognitive tasks or verbal mediation is a variable that reflects individual differences in strategy use, rather than a universal given of task performance. In fact, the first assumption of Peal and Lambert's code-switching hypothesis suggests an alternative, somewhat independent, hypothesis, that is, a verbal-mediation hypothesis. In the words of Hakuta and Diaz (1985, p. 341): "It is possible . . . that childhood bilingualism fosters a rather precocious use of verbal mediation in the processing of information and this, in turn, explains bilinguals' improved performance on nonverbal tasks." Because of the scarcity of process data, as mentioned above, little empirical support could be given at the time to either the code-switching or verbal-mediation hypothesis. Fortunately, some recent 180
Towards an explanatory model findings on bilingual private speech do shed some light on the validity of both hypotheses. In order to test the code-switching and verbal-mediation hypotheses, Diaz, Padilla, and Weathersby (in press) examined the private speech of thirty-four Mexican-American preschoolers. Private speech refers to children's use of overt language during cognitive tasks for the purpose of planning, guiding, and monitoring their own activity. Self-regulatory private speech can be seen as the overt and observable precursor of covert verbal mediation (Diaz, 1986). In the authors' opinion, the private speech of bilingual children presents a unique opportunity to study not only a process variable in bilingual task performance, but also an excellent data base to explore the validity of the code-switching and verbal-mediation hypotheses. At the time of the study, children in the sample, aged three to six, were attending Spanish-English bilingual preschool programs in El Paso, Texas. The preschool programs were designed for children to achieve a high degree of bilingual proficiency by the time they enter first grade. During the day, children participated in different activities; some activities were conducted mostly in English while others mostly in Spanish. At all times, however, teachers were encouraged to switch languages as much as possible and give instructions or announcements in both languages. The investigators were able to observe the preschool programs in action, and we observed many instances of social code-switching in administrators, teachers, staff, and the children. After frequent observations and interactions with the administration, teachers, and staff, it was clear that the adults involved in the program valued bilingual proficiency as a true asset for academic, social, and financial success in their society. In short, the preschool programs had multiple features of additive bilingualism in terms of both language use and status. In order to elicit children's spontaneous use of self-regulatory language, children were video-taped individually while performing the following three cognitive tasks: block designs, classification, and story-sequencing tasks. The block-design task involved the creation of three-dimensional block structures according to two-dimensional models. For the classification task, children were given several sets of randomly ordered cards portraying different familiar objects. Children were asked to match pairs that go together; typical matches were hammer and nails, toothbrush and toothpaste, lock and keys, etc. The story-sequencing task consisted of different sets of cards, randomly ordered, where each set portrays a sequenced event (e.g., making a cake: mixing the ingredients, baking them in the oven, decorating cake with frosting). Children were asked to arrange the cards in the correct temporal order. Before each of the three tasks, children were given brief instructions with two examples. 181
Rafael M. Diaz and Cynthia Klingler Instructions were given bilingually by the first author, repeating every line of instruction in both languages, as well as answering children's questions in both languages, regardless of the language in which the question was asked. Since we were interested, among other things, in observing the process of language-switching in private speech, it was important to create through instructions a climate where code-switching was done frequently and naturally in the context of the testing session. For each of the tasks, after completing the two examples, children were asked to do as many sets of cards/designs as they could in five minutes. The experimenter remained in the room, but told the child that he could not help and that they had to work on their own. These instructions were intended to elicit subjects' self-regulatory speech and self-regulated behavior. At a later time, children's spontaneous speech was transcribed from the video-tapes and coded into social and private speech categories (see Diaz, Padilla, & Weathersby, in press, for details). Prior to the video-taped sessions, children's language abilities in English and Spanish were tested using both receptive and productive measures. Following Hakuta and Diaz' (1985) recommendations for a within-bilingual design, the sample was divided into low and high bilingualism based on their second-language scores, controlling for ability in their first language. In order to test the code-switching hypothesis, the following three subhypotheses were formulated: (1) bilingual children will switch languages in their private speech while performing cognitive tasks; (2) bilingual children will switch languages more often in tasks of increasing difficulty; and (3) the frequency of language-switching will be meaningfully related to task performance. As it turned out, the researchers found only twenty instances of code-switching representing a language switch in less than 2 percent of all coded utterances. In addition, no meaningful relations were observed between the few instances of code-switching and task difficulty, task performance, or degree of bilingualism. As Diaz, Padilla, and Weathersby (in press, p. 28) conclude: "The most sensible conclusion that can be drawn from these data is that, even though bilinguals engage in frequent code-switching in social communication, there is no evidence that bilingual children spontaneously switch languages in their problem-solving or cognitive performance." On the basis of the present findings, the authors rejected Peal and Lambert's code-switching hypothesis. In order to investigate the validity of the verbal-mediation hypothesis, Diaz, Padilla, and Weathersby examined the differences between highand low-bilingualism groups on two measures of verbal mediation. One measure examined the overall frequency of task-relevant private-speech utterances. A second measure assessed the quality of verbal mediation, operationalized as the number of different task-relevant functions used by the child in his or her private speech. For example, children use private 182
Towards an explanatory model speech for different functions, such as labeling relevant aspects of the task, planning, making inter-item transitions, pacing motor movements, etc. We reasoned that children who use their private speech for a larger number of functions, independent of frequency of utterances, can be considered to have a higher quality of verbal mediation than children who only use one or two functions. If language can be used for different cognitive mediating functions, children who use a higher number of functions can be said to use their regulatory language more efficiently than those children who use language for only one or two functions. In this scheme, the private speech of a very talkative child who only uses the labeling function, for example, is of a lesser quality than the less talkative child who uses language for labeling, describing own activity, and planning. The results of the study suggested a positive effect of bilingualism for both frequency and quality of private speech. Even though more bilingual children used a much higher number (almost double!) of task-relevant private-speech utterances (M = 13.53 vs. M — 26.20), this group difference failed to reach statistical significance. The lack of statistical significance is a common problem in most studies that compare the amount of private speech between groups. As it turns out, there are large individual differences in the use of private speech, including a large number of children who do not talk at all during the tasks. The resulting large standard deviations within groups make it difficult to find significant between-group comparisons (see Diaz, 1986, and Frauenglass & Diaz, 1985, for a more detailed discussion of these issues). Group differences in the quality measure revealed also a bilingual advantage in the number of private-speech functions that children used during the tasks (M= 1.53 vs. A/= 2.20); even though the effect was substantial (F(l,30) = 3.36, p<0.07), the difference failed to meet the prespecified level (/?<0.05) of significance.
Summary of findings Regarding the interaction between an early bilingual experience and cognitive development, we can summarize the reliable findings to date in the following manner: 1. 2. 3.
4.
Bilingual children show consistent advantages in tasks of both verbal and non-verbal abilities. Bilingual children show advanced metalinguistic abilities, especially manifested in their control of language processing. Cognitive and metalinguistic advantages appear in bilingual situations that involve systematic uses of the two languages, such as simultaneous acquisition or bilingual education. The cognitive effects of bilingualism appear relatively early in the process 183
Rafael M. Diaz and Cynthia Klingler
5.
6.
of becoming bilingual and do not require high levels of bilingual proficiency nor the achievement of balanced bilingualism. Bilingual children show advantages in the use of language for verbal mediation, as evidenced by their higher frequency of private-speech utterances and their larger number of private-speech functions. Even though bilinguals switch languages easily and frequently in the context of communicative social speech, there is no evidence of bilinguals' spontaneous code-switching in their overt private speech or covert verbal mediation.
For the purpose of outlining a model of the interaction between bilingualism and cognitive development, it is important to note some important consequences, points of convergence, and relationships among the findings. First, because the positive cognitive effects appear relatively early in the process of becoming bilingual, the code-switching hypothesis, as formulated by Peal and Lambert (1962) can be ruled out as a plausible explanatory hypothesis. Code-switching demands a relatively high command of the two languages (at the level of production), and it cannot explain the advantages that we have observed early in the process. Ruling out the code-switching hypothesis is also a prudent route to follow, considering the lack of data about bilingual code-switching while verbally solving cognitive tasks. Second, it is important to note the convergence between the metalinguisticfindingsand thefindingsabout bilinguals' private speech. Recent findings on bilinguals' metalinguistic awareness suggest that the advantage is related to issues of language control rather than language knowledge. Specifically, the findings underscore bilinguals' capacity to exercise executive control over the automaticity of language comprehension and production. By the same token, private speech refers to the uses of language for the purpose of planning, guiding, and monitoring one's behavior. Private speech, therefore, signals the activity of the self-regulatory executive functions. In fact, private speech increases with task difficulty when processing demands a high degree of executive control and virtually disappears when a given task or skill can be performed automatically. In short, both the metalinguistic and private-speech findings suggest a beneficial effect of bilingualism on self-regulatory executive functioning. We turn now to Vygotsky's theory of thought and language in search of a theoretical framework to guide our model-building effort. Vygotsky's theory of thought and language According to Vygotsky (1962), the human child is endowed at birth with two separate systems: thought and language. Thought refers to the system of biologically endowed elementary functions and processes such as perception, attention, and memory that constitute the child's native intellectual 184
Towards an explanatory model endowment. The thought system can be understood as practical or preverbal intelligence and is roughly equivalent to Piaget's notion of sensorimotor intelligence. The language system, on the other hand, refers to the system of communication that is present at birth in the form of cries and smiles. The language system gradually develops through babbling and the use of words into the complex system that constitutes the adult's linguistic competence. Vygotsky's major claim is that, in the human infant, as well as in many other species, these two systems of thought and language are separate and functionally independent. The separation of the language and thought systems, however, is shortlived. Very early in development, children begin to use language not only for communication but also as a tool to plan, guide, and monitor their activity in a self-regulatory fashion. The use of language as a tool of thought, private speech, transforms the course of intellectual development and gives birth to a new form of intellectual activity, verbal thought. In Vygotsky's own words: The most significant moment in the course of intellectual development, which gives birth to the purely human forms of practical and abstract intelligence, occurs when speech and practical activity, two previously completely independent lines of development, converge ... as soon as speech and the use of signs are incorporated into any action, the action becomes transformed and organized along entirely new lines. (1978, p. 24) Our readings of Vygotsky suggest that the proposed "transformation" consists of an increasing self-regulation of processes that were originally controlled and regulated by the external physical and social environment. For example, at birth, young infants attend to whatever is perceptually salient; also, early on, caregivers manipulate the infant's attention by the use of sound, movement, and gesture. At these early stages, the infant's attention is, physically and socially, externally regulated; attentional capacities are, so to speak, at the mercy of the concrete, external environment. At one point in development, however, children begin to use speech in the form of labeling and descriptions in order to organize and restructure their perceptual field. By saying the words the red one, for example, children focus their attention voluntarily on a specific feature (color) of the stimuli and manipulate at will figure-ground relations in their immediate perceptual field. By saying the word red, children make color the salient perceptual figure while other stimulus features, such as size, shape, or texture, remain as ground. In other words, attention develops, through the use of language, from a set of externally controlled processes to an internally self-regulated voluntary function. According to Vygotsky (1978), the use of language as a tool of thought has three major consequences for the child's intellectual development. First, 185
Rafael M. Diaz and Cynthia Klingler the child's cognitive operations gain greater flexibility, freedom, and independence from the concrete stimulus field. For example, with the use of speech, children can bring to their problem solving elements that are not immediately present. Also, as mentioned above, through the use of language children organize and restructure their perceptions in terms of their own goals and intentions. Second, through the use of speech, the child's operations and actions become less impulsive. Speech breaks down the immediate spontaneous connection between the stimuli and the child's responses, allowing the child to act reflectively according to a plan rather than responding impulsively to the objective properties of the stimuli at hand. Finally, through the use of speech as a tool of thought, children gain mastery and control over their own cognitive processes. The rigidity of innate perceptual preferences is broken down under the control of voluntary attention; arousal-habituation responses are mastered through the capacity for sustained attention and inhibition of distracting stimuli; the limitations of eidetic memory processes are overcome by the use of mnemonic strategies; inefficient trial and error is gradually replaced by problem solving through verbalized planning and self-monitoring. Needless to say, Vygotsky's theory provides a most fascinating account of how language transforms the course of cognitive development, proposing that the use of language as a tool of thought gives children increasing executive control over cognitive operations. While Vygotsky proposed his theory of thought and language as a universal process of development, it is a fact that children vary enormously in terms of their use of private speech and in terms of self-regulatory capacities (see, e.g., Diaz & Lowe, 1987; Frauenglass & Diaz, 1985). A consideration of individual differences in the development of self-regulation is important also in order to address the relation between bilingualism and intelligence and explain bilinguals' advantages on private-speech and executive functions. We must look, therefore, for sources of individual variability in the use of private speech and self-regulation and examine how the bilingual experience might influence the progression of self-regulatory development. The developmental path towards self-regulation can be described in terms of children's increasing takeover of the adult caregiver regulatory role. In fact, self-regulation begins when the same sign system that adults use to regulate children is used by children to regulate themselves. In other words, in order to regulate their own activity, children begin to use the same signs that adults were using to regulate them. In order to understand the nature of individual differences in self-regulatory development, it is important to understand the reason why children begin to use the language of caregivers in this self-regulatory fashion. Vygotsky has given a very interesting answer to this question. He suggests that it is children's awareness of the regulating power of signs that leads them to rely more frequently 186
Towards an explanatory model on speech for cognitive functions. In his essay on the genesis of higher psychological functions, Vygotsky (1960/1981) describes the development of the use of signs for self-regulation as occurring in four different stages. Stage 2 describes the regulating role of adult-given auxiliary signs, while stage 3 describes the child as actively manipulating signs and organizing their stimulus field in order to achieve a desired response. The main developmental question is, therefore, what accounts for the transition from stage 2 to stage 3, that is, what accounts for the movement from other-regulation to self-regulation? Vygotsky believes that the transition is motivated by children's increasing knowledge and awareness of language (metalinguistic awareness?) as a system of signs that have regulatory power. Commenting on children's use of adult-given signs for mnemonic functions during stage 2, and their subsequent transition to stage 3, Vygotsky states: In mastering naively what this memory operation consists of, children already move on to the next stage. If we give them pictures in random order, they arrange them in a certain order and establish a certain connection. They no longer operate superficially with signs; rather they know that the presence of such signs helps them carry out the operation, i.e., helps them remember through use of the given signs. (1960/ 1981, p. 182)
In the present authors' opinion, Vygotsky's proposed connection between metalinguistic awareness and the use of signs for self-regulation provides an excellent framework to integrate the findings to date on bilingual cognitive development. Specifically, Vygotsky's theory allows for an integration of metalinguistic findings regarding awareness and control. It is to this task that we now turn. Outline for an explanatory model
At the beginning of the chapter we suggested two requirements for the development of an explanatory model on the interaction between bilingualism and cognitive development. First, we suggested that the model should take into account the findings to date in toto, rather than just a set of findings from one particular study. Second, we argued that the model should be grounded on a solid account regarding the interaction between language and thought in development. Having reviewed, interpreted, and integrated the findings to date, we are now ready to provide an outline for an explanatory model with Vygotsky's theoretical framework. The model will be outlined in the form of three propositions. Proposition 1. At an early age, the exposure to two languages in a systematic additive fashion leads to an objective awareness of language. Bilinguals' 187
Rafael M. Diaz and Cynthia Klingler metalinguistic awareness includes not only awareness and knowledge about grammatical rules, but also awareness of the non-communicative uses and functions of language. We would like to argue that an additive bilingual experience, for reasons as yet unspecified, allows children to step outside and function outside of one particular language system. This experience leads to an objectification not only of language rules and structures, but also of language function. Proposition 2. Increased awareness and understanding of the cognitive functions of language leads to an increased use of language as a tool of thought. In the same way that bilinguals' awareness of grammaticality yields greater control over automatic grammatical productions, the awareness of functional properties of language should lead to an increased self-regulatory control over different language functions. Proposition 3. The increased and more efficient use of language for cognitive functions will give children an advantage in cognitive performance across verbal and non-verbal tasks. Bilinguals' increased reliance on private speech and verbal mediation will give children a head start on the development of cognitive executive functions. Those advantages will be more evident in tasks that demand non-automatic processing. Based on previous research, the model outlined suggests that the systematic exposure to two languages found in bilingual additive situations gives children a unique advantage in the objectification of language. At first sight, this first proposition reveals nothing new beyond the objectification hypothesis as stated by Leopold and Vygotsky and espoused by several researchers in the field. Proposition 1, however, adds two new features to the popular objectification hypothesis. First, the outline specifies exposure to rather than use of two languages, in keeping with findings that the effects of bilingualism occur at the beginning of the second-language learning process. Second, the outline underscores bilinguals' awareness of language beyond knowledge of grammatical rules and structures, such as the understanding of cognitive self-regulatory functions of language. The model outlined proposes further that the increased objectification of language at an early age will foster a more efficient use of language as a tool of thought. Research on children's private speech has supported Vygotsky's (1962) notion that children use language not only for communication but also as a tool to plan, guide, and monitor their cognitive activity (Berk, 1986; Diaz, 1986). However, even though children vary in the amount, quality, and effectiveness of their private speech, the sources of such individual variation are not presently known. In line with Vygotsky's theorizing, proposition 2 hypothesizes that the early objectification of 188
Towards an explanatory model language function will increase the actual use and effectiveness of private speech in the performance of cognitive tasks. Finally, proposition 3 logically suggests that the increased frequency and quality of private speech will manifest in improved cognitive performance across tasks that demand executive control of processing. To complete this section, we would like to note that our outline for a model on the interaction between bilingualism and intelligence not only fulfils the two initial requirements we imposed on the model-building effort, but also addresses four important gaps in our understanding of the phenomena at hand. First, the outlined model incorporates both the findings about bilinguals' cognitive advantages and thefindingsabout bilinguals' metalinguistic advantages, meaningfully relating the constructs of language awareness, language control, and cognitive benefits. Second, by postulating effects of bilingualism on executive cognitive functions rather than on specific cognitive tasks, the outlined model explains the seemingly puzzling finding that a linguistic experience shows positive effects on a wide range of tasks, verbal and non-verbal. Third, the outline provides a conceptual integration of both aspects of metalinguistic ability, awareness and control, that have been recently observed in bilingual children. Finally, the outline suggests that the effects of bilingualism are not simply an acceleration of cognitive development, but rather an enriched approach to the use of language as a tool of thought. We believe that an early bilingual experience offers children a certain cognitiveflexibilityin their task performance; however, unlike the code-switching hypothesis, we believe thisflexibilitystems from a greater use of and reliance on the self-regulatory functions of language. Why only an outline? It comes as no surprise that the development of an explanatory model on the relation between bilingualism and cognitive development is limited by a serious scarcity of empirical observations. The outline, however, is useful in pointing out the path for future research. In order to move from a simple three-proposition outline to a full explanatory model, further research is needed in the following areas: 1. First, we need a specification of the specific features of a bilingual experience that characterize it as an additive situation. 2. Once the additive features are specified, how do they foster an increasing awareness of language? 3. We need further research on the awareness of language functions. The truth is that we have virtually no data, from bilingual or monolingual children, on their increasing awareness of the self-regulatory functions of language. 4. More work, conceptual and empirical, is needed to specify the relation 189
Rafael M. Diaz and Cynthia Klingler
5.
between language awareness and control. Is Vygotsky right in his suggestion that the use of self-regulatory speech is fostered by an awareness and knowledge of the regulatory power of signs in human activity? The proposition seems like a very ambitious claim to explain the origins of private speech in preschoolers. Finally, further work is needed to specify the functional relations between private speech and cognitive activity. The literature on private speech is admittedly small, and the observed relations between private speech and performance are unclear and problematic, to say the least.
We offer the above list of research gaps not as a discouraging statement underscoring the limitations of our knowledge, but rather as an invitation for further research in this fascinating and important line of scientific investigation.
References Ben-Zeev, S. 1977. The influence of bilingualism on cognitive strategy and cognitive development. Child Development, 48, 1009-1018. Berk, L. E. 1986. Relationship of elementary school children's private speech to behavioral accompaniment to task, attention, and task performance. Developmental Psychology, 22, 671-680. Bialystok, E. 1986. Factors in the growth of linguistic awareness. Child Development, 57, 498-510. Bowey, J. A. 1986. Childhood bilingualism and metalinguistic development. Paper presented at the Fourth National Developmental Conference, Sydney, Australia, August. Cummins, J. 1976. The influence of bilingualism on cognitive growth: a synthesis of research findings and explanatory hypotheses. Working Papers on Bilingualism, 9,1-43. Cummins, J. & R. Mulcahy 1978. Orientation to language in Ukrainian-English bilingual children. Child Development, 49,1239-1242. Diaz, R. M. 1983. Thought and two languages: the impact of bilingualism on cognitive development. In E. W. Gordon (ed.), Review of research in education, vol. X. Washington, DC: AERA. 1985a. The intellectual power of bilingualism. The Quarterly Newsletter of the Laboratory of Comparative Human Cognition, 7,16-22. 1985b. Bilingual cognitive development: addressing three gaps in current research. Child Development, 56,1376-1388. 1986. The union of thought and language in children's private speech. The Quarterly Newsletter of the Laboratory of Comparative Human Cognition, 8, 90-97. Diaz, R. M. & J. R. Lowe 1987. The private speech of young children at risk: a test of three deficit hypotheses. Early Childhood Research Quarterly, 2,
181-194. 190
Towards an explanatory model Diaz, R. M. & K. A. Padilla 1985. The self-regulatory speech of bilingual preschoolers. Paper presented at the April Meeting of the Society for Research in Child Development. Toronto, Canada. Diaz, R. M., K. A. Padilla, & E. K. Weathersby in press. The impact of bilingualism on the self-regulatory private speech of Mexican-American preschoolers. In E. Garcia (ed.), The Mexican-American child: developments in language, cognitive, and social development. Tempe, AZ: University of Arizona Press. Duncan, S. & E. De Avila 1979. Bilingualism and cognition: some recent findings.
NABE Journal, 4, 15-50. Edwards, D. & H. Christophersen 1988. Bilingualism, literacy and meta-linguistic awareness in preschool children. British Journal of Developmental Psychology, 6, 235-244. Fishman, J. A. 1977. The social science perspective. Bilingual education: current perspectives, vol. I. Arlington, VA: Center for Applied Linguistics. Frauenglass, M. & R. M. Diaz 1985. Self-regulatory functions of children's private speech: a critical analysis of recent challenges to Vygotsky's theory. Developmental Psychology, 21, 357-364. Galambos, S. 1985. The development of metalinguistic awareness in bilingual and monolingual children. Paper presented at the April Meeting of the Society for Research in Child Development. Toronto, Canada. Galambos, S. J. & K. Hakuta 1988. Subject-specific and task-specific characteristics of metalinguistic awareness in bilingual children. Applied Psycholinguistics, 9,141-162. Hakuta, K. 1986. The mirror of language: the debate on bilingualism. New York: Basic Books. 1987. Degree of bilingualism and cognitive ability in mainland Puerto Rican children. Child Development, 58, 1372-1388. Hakuta, K. & R. M. Diaz 1985. The relationship between degree of bilingualism and cognitive ability: a critical discussion and some new longitudinal data. In K. E. Nelson (ed.), Children's language, vol. V. Hillsdale, NJ: Erlbaum. Hakuta, K., B. M. Ferdman, & R. M. Diaz 1987. Bilingualism and cognitive development: three perspectives. In S. Rosenberg (ed.), Advances in applied psycholinguistics, vol. II Reading, writing and language learning. New York: Cambridge University Press. Ianco-Worrall, A. D. 1972. Bilingualism and cognitive development. Child Development, 43, 1390-1400. Leopold, W. F. 1939-1949. Speech development of a bilingual child: a linguist's record (4 vols). Evanston, IL: Northwestern University Press. Liedtke, W. W. & L. D. Nelson 1968. Concept formation and bilingualism. Alberta Journal of Educational Research, 14, 225-232. Luria, A. R. 1976. Cognitive development: its cultural and social foundations. Cambridge, MA: Harvard University Press. Martin-Jones, M. & S. Romaine 1986. Semilingualism: a half-baked theory of communicative competence. Applied Linguistics, 7, 26-38. Peal, E. & W. Lambert 1962. The relation of bilingualism to intelligence. Psychological Monographs, 76, 1-23. 191
Rafael M. Diaz and Cynthia Klingler Skutnabb-Kangas, T. 1984. Bilingualism or not: the education of minorities. Clevedon, Avon, England: Multilingual Matters. Smith, C. L. & H. Tager-Flusberg 1982. Metalinguistic awareness and language development. Journal of Experimental Child Psychology, 34,449-468. Torrance, E. P., J. J. Wu, J. C. Gowan, & N. C. Aliotti 1970. Creative functioning of monolingual and bilingual children in Singapore. Journal of Educational Psychology, 61, 72-75. Van Kleek, A. 1982. The emergence of linguistic awareness: a cognitive framework. Merrill-Palmer Quarterly, 28, 237-265. Vygotsky, L. S. 1960/1981. The genesis of higher mental functions. In J. V. Wertson (ed. and trans.), The concept of activity in Soviet psychology. New York: M. E. Sharpe. (Reprinted from Razvitie vysshikhpsikhicheskikhfunktsii, 1960, 182-22.) 1962. Thought and language. Cambridge, MA: M.I.T. Press. 1978. Mind in society: the development of higher psychological processes. Cambridge, MA: Harvard University Press. Wertsch, J. V. & C. A. Stone 1985. The concept of internalization in Vygotsky's account of the genesis of higher mental functions. In J. V. Wertsch (ed.), Culture, communication and cognition. New York: Cambridge University Press.
192
9
Constructive processes in bilingualism and their cognitive growth effects1 JANICE JOHNSON
The research literature on childhood bilingualism provides a rich data base for examining relations between language and cognition. Classic developmental theory in psychology addresses the interplay between language and cognition in the process of development, but tends to place the emphasis on one versus the other. Piaget (Piaget & Inhelder, 1969), for example, collapses language into logic (or conceptualization); by contrast, Vygotsky and his followers (Luria, 1961, 1976; Vygotsky, 1962, 1978) collapse logic into language. In the current chapter, I use constructs from dialecticalconstructivist theory - a neo-Piagetian theory (Pascual-Leone, 1976a, 1984, 1987; Pascual-Leone & Goodman, 1979) - to attempt to define the nature of linguistic and cognitive structure and the processes through which they may interact in the bilingual child. I use these constructs to organize findings in the literature on cognitive effects of bilingualism, and I present new data on metaphor interpretation in bilingual and monolingual children. Let us consider first the sorts of knowledge structures the child may construct in the course of experience - sorts that often are organized into knowledge layers or levels of representation. Three sorts are of particular interest here (Pascual-Leone, 1976a, 1984). First are structures that, after Piaget (Piaget & Inhelder, 1967), I will call infralogical.2 These are particular-experiential structures; Pascual-Leone (1976a, 1984) called them mereological and Husserl (1970) referred to them as the structures of the life world. Infralogical structures are the substance of experience; they represent actual objects or things in the environment. Our internal representations for distal objects, whether organized as prototypes (Rosch, 1975), scripts (Nelson, 1986; Schank & Abelson, 1977), or schemas (Anderson, 1985), are infralogical structures. Logological structures embody invariances across types of objects or situations.3 Piaget called them logical structures. They encode into kinds (classes, relations, propositions) the diversity of infralogical (i.e., particular-experiential) structures. Logological structures are of a higher order than their infralogical referents (i.e., they can represent infralogical 193
Janice Johnson structures but are distinct from them). Logological structures are generic kinds that often represent common characteristics of the infralogical token structures from which they are constructively abstracted. Linguistic structures are the structures that mediate the function of communication among humans. They encode invariances in the linguistic environment (e.g., lexical terms, grammatical relations). Language is a symbolic system received from society and geared to communication, and a structure is linguistic only if communication is necessary for its learning. Linguistic structures often denote or symbolize logological structures (e.g., classes, relations, propositions), which in turn describe infralogical structures. There are also, however, linguistic structures that derive directly from infralogical structures. This is the case of signals (Luria, 1961; Vygotsky, 1962) and of many conventional symbols - much of early language acquisition might proceed directly from being "heard" infralogical structures to becoming linguistic structures the child uses for communication (Pascual-Leone, Johnson, & Benson, 1989). Linguistic structures take their meaning from other structures (either logological or infralogical) for which they stand. Infralogical structures are constructed in the context of concrete experience (e.g., Piaget's "physical experience"). Logological structures can be constructed "bottom-up" by internalizing invariances across infralogical structures (e.g., Piaget's "logico-mathematical experience") or "top-down" through instruction. In the-latter case human mediation via language plays an important role (Sigel, 1981,1982; Vygotsky, 1978). Once acquired, however, logological knowledge can be language-free; it is not reducible to linguistic structures. Consider the example of a professor trying to convey to students the meaning of an abstract concept, that is, trying to bring the students to construct internally a logological structure as referent for a new technical (lexical) term. If the students are sophisticated and knowledgeable in the relevant content area then a definition may suffice. The definition points to the essential invariants that underlie the concept, and if the students already have these invariants encoded as logological structures they may be able to co-ordinate them into the new concept. If our definition is met by puzzlement we may assume that the students do not have the appropriate logological structures, and we then attempt to lead them to construct the new concept in a more bottom-up way. We use a series of concrete examples (whose referents are infralogical structures in the students), and we choose the examples so that what varies across them is irrelevant to the concept in question and what is invariant is the essence of the concept. Our hope is that the students will co-ordinate into a new structure not our words nor the particular situations, but what is common across the situations. The students do acquire a new linguistic structure - the label for the concept 194
Constructive processes in bilingualism - but through a process of internal reorganization they also construct the referent of the label, which is not linguistic (Pascual-Leone, 1987). The literature on bilingualism provides much evidence for interdependence between languages (e.g., Cummins, 1979; Cummins et al., 1984; Davidson, Kline, & Snow, 1986; Galambos & Hakuta, 1988; Geva & Ryan, 1987; Hakuta, 1987; Johnson, 1989; Malakoff, 1988). An obvious basis for such interdependence is a repertoire of infralogical (i.e., particularexperiential) and logological (i.e., generic-conceptual) structures that can be accessed through either language. A number of researchers have been concerned with transfer of decontextualized language skills or of cognitiveacademic proficiency (e.g., Cumminsetal., 1984; Davidson, Kline, & Snow, 1986; Malakoff, 1988). Here, the basis for transfer is most likely logological structures. Transfer of contextualized language skills (e.g., Davidson, Kline, & Snow, 1986) is probably mediated by infralogical structures, if by contextualized we mean performance produced by structures that are easily cued by concrete features of the situation. Following the lead of Leopold (1961) and Vygotsky (1962) a number of researchers have proposed that experience with two linguistic systems leads the bilingual child to an early awareness of the arbitrary connection between linguistic forms and meanings (e.g., Bialystok, 1987; Cummins, 1978a; Diaz, 1983; Ianco-Worrall, 1972). As we have already suggested, structural learning (i.e., learning that reflects relational patterns across lower-level schemes or structures) involves recording patterns of invariance. When we use two languages to express the same meaning (i.e., the same proposition or actual state of affairs), we hold constant the underlying proposition while varying the linguistic means. This holding constant the meaning while varying the language facilitates separation of the meaning invariant from the linguistic form; that is, it facilitates detachment of logological and infralogical structures from linguistic structures. Malakoff (1988) provides a striking example of separation of linguistic and logological structures. She found that bilingual children performed more accurately on an analogies task when the task required use of two languages (i.e., analogy stem in one language, response in another) than when a single language was used. Malakoff suggests that the mixed-language condition encouraged the child to rely less on linguistic association and more on reasoning based on the conceptual relationship between the terms in the analogy. That is, the mixed-language condition was more likely to elicit processing at the logological level. Once logological and linguistic structures have been detached, we can use our logological structures to examine language from an intellective vantage-point. This is the stance that some researchers have called metalinguistic and that Russian psychologists, and others, have labeled "objectification" of language (Cummins, 1976). Russian psychologists have also 195
Janice Johnson emphasized that language can serve to control cognition (Luria, 1961; Vygotsky, 1978; Wertsch, 1985). That is, the processes through which one gains voluntary control over one's cognitions and actions are internalized symbolic-mediational structures. These structures are internal reconstructions of external, socially mediated activity - especially, for the Russians, linguistic activity. Today many would call such cognitive-control processes executives (Case, 1985; Pascual-Leone, 1984; Pascual-Leone etal., 1978), strategies (Brown etal., 1983), or metacomponents (Sternberg, 1985). Executives are structures engaged in mental planning and the temporal structuring of mental processing and action across all sorts of tasks. One can define two sorts of executive structure (Pascual-Leone et al., 1978). Task executives (or plans) are addressed to the external situation and monitor the subject's interaction with the environment (i.e., his or her performance). Control executives are addressed to the monitoring of organismic internal resources-e.g., the allocation of cognitive capacities such as mental attention. As we will see in more detail later, research accrued over the past two decades points to cognitive benefits of bilingualism across several types of task situation (Diaz, 1983; Hakuta, 1986). One might attempt to explain such data in terms of a multiplicity of content (i.e., specificknowledge) structures - each structure appropriate for a different type of task. A more parsimonious hypothesis, however, is that learning two languages facilitates construction of certain kinds of executive structure (task and/or control), and that these executives guide performance across certain kinds of task. To explore this hypothesis I examine proposals for how executive structures are constructed. Note that executives may be logological or infralogical structures. Infralogical executives are developed within particular contexts and tend to be situation-bound; logological executives are generic in that they apply across contexts (Pascual-Leone, 1984).
The Vygotskian viewpoint from a dialectical-constructivist perspective
As mentioned above, Vygotsky (1978) and Luria (1961) were concerned with development of higher mental processes that allow the child to control his or her own action and cognition; that is, the development of executive structures. Their proposal was that these internal mediational structures are developed by means of internalization (i.e., reconstruction on the intrapsychological plane) of the structure of social-mediational situations. Speech is seen to play a privileged role in this process. For example, through the use of language the mother may guide the child to successful performance on a task the child could not solve on his or her own. The mother's language directs the child's attention and the sequencing of the child's action - that is, she externally induces in the child the executive structures 196
Constructive processes in bilingualism needed to produce the performance. With repetition, the child internally reconstructs this executive sequence and can use it to guide his or her own performance in the future. For Vygotsky and Luria this executive control is exercised initially through the child's external speech and then through inner speech. What role might a bilingual social environment play in such development? Luria (1961) speaks of speech as initially serving a primary signaling function vis-a-vis action, that is, it initially serves to impel or initiate action (e.g., "Go! Go!"). As such it is tied to the immediate context, including the action the child is performing, and is not effective in inhibiting ongoing action. For speech to regulate and inhibit action, it must take on a symbolic (or significative) function (e.g., "Press for the red light. Don't press for the green light"), that is, it must take on a meaning independent of the immediate context. Likewise, Vygotsky emphasized a developmental principle that Wertsch (1985) labels "decontextualization of mediation means"; this is "the process whereby the meaning of signs becomes less and less dependent on the unique spatiotemporal context in which they are used" (Wertsch, 1985, p. 33). Using the dialectical-constructivist terminology introduced above, we can say that development of higher mental functions involves a transition from low infralogical (i.e., signalic) to high infralogical (i.e., experiential-symbolic) structures, and from there to logological (i.e., generic-symbolic) structures as referents for communicative signs (i.e., linguistic structures). Note that I understand by a symbol a signifier that in mental processing stands for a referent, but is functionally detached from this referent in the sense that it evokes the possibility, but not the actuality, of the referent. This is the sense of symbol found in Cassirer (1944), Langer (1957), Peirce (1955), and Piaget (Piaget & Inhelder, 1969; Piaget distinguishes between "signs" and "symbols" - 1 use "symbol" to refer to both of these). Contrary to the logical-empiricist tradition (e.g., Morris, 1946), I do not equate symbol with language: non-linguistic (e.g., perceptual, imaginal, motor) symbols exist that can be infralogical (i.e., standing for particular experiences) as well as logological (i.e., standing for generic representations Pascual-Leone, Johnson, & Benson, 1989). As suggested above, varying the linguistic means while maintaining the same context and propositional content may facilitate decontextualization/ detachment of mediational structures from specific linguistic contexts or utterance events. In a bilingual learning environment, the child is likely to encounter similar social mediational situations (social contexts) conducted through different languages. The invariant across these situations will be not the specific linguistic means, but the structure of the socially constructed (task or control) executive (e.g., control of attention, sequencing of action). This sort of mediated learning could lead to earlier internal 197
Janice Johnson reconstructions of executive structures in bilingual children. Although for Vygotsky and Luria speech plays a central role in the development of internal mediational structures, the specifics of the language used are rather irrelevant. Since the bilingual situation naturally provides variation in this linguistic aspect, it should hasten elimination of the linguistic aspect from the structure of interest (thus hastening decontextualization). We have been addressing the development of general task and control executives. As stated above, Vygotskian theory also predicts that bilingualism could facilitate the objectification of language (Vygotsky, 1962). That is, as language becomes detached from its referents, it becomes an object whose structure can be examined in its own right. As Vygotsky states, structural phenomena common to the two languages may be seen to fall "under more general categories" (1962, p. 100) - that is, language structure itself may come to be represented by means of decontextualized, logological structures. This notion will prove to be important in our later analyses of bilinguals' cognitive performances, so let us examine it more closely using dialectical-constructivist constructs. We make sense of experience by means of our infralogical (particularexperiential) structures (i.e., we are always in a context). Thus language as communication has its reference (is validated) at the infralogical level; it is difficult to understand communication that does not make contact with our experience (Husserl, 1970; Pascual-Leone, 1976a, 1984). Signals are contextual features that directly cue infralogical structures. For example, perceptual features of a context serve to cue our structures for (allow us to recognize) objects. Developmentally, language starts out having a signalic function, and some features of language retain this function (e.g., "Go! Go!" or "Fire!"; Luria, 1961; Pascual-Leone, Johnson, & Benson, 1989). Language also gradually takes on a symbolic function, that is, it allows us to entertain the idea of the referent, without making the referent present (Cassirer, 1944). Logological structures (the logic of thought) are also symbolic in this sense. Let us now characterize linguistic and linguistic-logological structures as lexical and grammatical structures that are specific to a language. Linguistic-logological structures are generic structures within a given language. An example would be "Because SJC,S)>" - a generic structure for generating because sentences in English, where Sx and Sy are slots that can be filled by sentential clauses (Johnson, Fabian, & Pascual-Leone, 1989). The presence of variable slots makes the structure generic. In a concrete sentence the slots would be filled with sentential clauses referring to the actual events (i.e., infralogical structures) x and y. A judgment of grammaticality of the sentence would be carried out at the linguistic-logological (or the logological, see below) level; a judgment of truth or meaning at the infralogical 198
Constructive processes in bilingualism level (i.e., experientially, are events x and y in the temporal and causal relation described by the sentence?). Purely linguistic structures are language-specific and non-generic, at least from the perspective of the user. For instance, although any name (e.g., Peter, Janice) might be regarded as generic from an observer's view-point, the intended referent for the user is definitely non-generic: it is a given, particular person. Similarly, early in development the observer might note that a given noun (e.g., slug, car, horse) refers to a quasi-generic infralogical collection (an actual set of slugs, cars, or horses); however, the user (young child), whose logological structures are weak, does not in fact distinguish between the instances or tokens and the type, as Piaget (1962) and others have shown. In contrast, linguistic-logological structures are generic, and seem to correspond, in part, to what Bialystok (Bialystok & Ryan, 1985) has called analyzed knowledge. In addition to within-language structures, there are logical categories that, as categories, are common across languages - for example, the notions of word, noun, verb, adjective; reference to the past, the future; causality; etc. We experience language in the context of communication, and for this reason the connection between linguistic structure and infralogical structure is strong (we most readily connect language to sense). Across linguistic/communicative events, however, our cognitive system may begin to notice invariances in pragmatic functions, and these invariances may be encoded as new categorical structures that are not properly linguistic, but rather are logological. The linguistic means for indicating these functions may change from language to language, but the functional invariant remains the same and can come to be detached as a logological structure. An example would be the concept of word or of causality.4 Again, a bilingual learning environment provides increased variance in the linguistic means for expressing these functions and should facilitate detachment of the functions as logological (as distinct from purely linguistic) structures (Ben-Zeev, 1977b). Note that the more similar the bilingual's two languages are in terms of shared functional categories, the more likely these categories are to become detached as logological structures. The bulk of the research on cognition and bilingualism concerns subjects whose languages are similar in this respect (i.e., Spanish-English, English-French). These theoretical claims find some empirical support in research by Davidson, Kline, and Snow (1986), who measured separately decontextualized language skills and contextualized language skills in children attending a bilingual French-English school. They found that correlation was higher within these two dimensions of language skill across languages than it was across the dimensions within languages. That is, more important than the specific language was whether performance was guided by a decontextualized executive or was cued directly by features of the context. 199
Janice Johnson A limitation on these claims is suggested in research by Malakoff (1988). She found that bilingual children being schooled in one language (i.e., English or French) had faster reaction times when solving difficult analogies in the language of instruction than in the other language (which they received only as a language arts subject). (Note that the sample being schooled in English was heterogeneous in terms of home language; the majority of those being schooled in French had English as their home language.) This result suggests that for bilingual children, when certain kinds of mediational situations are consistently presented in only one language (e.g., in the language of schooling), logological structures may become more detached from linguistic structures in the instructional, as compared with the non-instructional, language. The result would be that although logological structures would also be accessible through the other language, they would be analogically evoked (i.e., activated by purely relational cues) more readily in the context of the instructional language. As Hakuta (1986) has pointed out, the importance of socialization in Vygotskian theory compels us to consider the historical circumstances of language use in the child. Davidson, Kline, and Snow's subjects were being schooled in two languages, Malakoff s in one. The Piagetian viewpoint from a dialectical-constructivist perspective
In his analysis of the sensorimotor period Piaget describes the development of action performance driven by features of the immediate situation - such performance is motivationally immediate, not mediated by decontextualized structures (Pascual-Leone et al., 1978; Pascual-Leone, Johnson, & Benson, 1989). Beyond the sensorimotor period, Piaget describes development of mediated performance, performance that is increasingly freed from the immediate context as the child develops higher levels of operativity. Piaget is concerned with the development of structures that guide performance across kinds of tasks, that is, he is concerned with the child's construction of logological structures. Piaget differs from Vygotsky in that he sees these structures as emergent in the child, rather than emergent in social interaction. These structures result from co-ordination of relations among one's own actions, not from internalization of the structure of externally guided performance. Mental co-ordination of action requires representational structures (i.e., the symbolic function) to relate past with present action, to anticipate future results. The driving force in development is equilibration - the drive to maximize adaptation (i.e., the number of kinds of situations with which one can successfully interact), while minimizing internal structural complexity (Case, 1985; Pascual-Leone et aL, 1978). Failure of adaptation leads to disequilibration, followed by reorganization aimed at restoring a new (i.e., higher-level) equilibrium. 200
Constructive processes in bilingualism Piaget's emphasis was on the child's interaction with the physical, not the social environment. Peal and Lambert (1962) proposed that the bilingual child may be exposed to a wider range of experiences, because his or her experiences stem from two different cultures. If this is the case, one might conclude that a by-product of bilingualism could be structural enrichment, within strict Piagetian terms. Such enrichment is not likely to take us very far, however, in explaining bilingual benefits across a range of types of task. Thus we must attempt to integrate the language environment within the Piagetian learning situation. Such an integration has been attempted bySigel(1981). Like Piaget, Sigel is concerned with representational thinking. He sees representational thought as having the same roots in development as did Piaget. He focuses, however, on mental "distancing experiences" as determiners of the level of representational thinking an individual will attain. Mental-distancing experiences are "a class of events and interactions which 'demand' the child to separate himself/herself mentally (via representation) in space or time from the ongoing observable field" (Sigel, 1981, p. 206). Distancing strategies are social experiences during which mental-distancing demands are made on the child. Such strategies appear to be mainly verbal and may take the form of a statement or a question to the child. Sigel (1982) describes a range of strategies that vary in the mental-distancing demand made on the child. For example, a demand to observe ("Look at the book" or "Do you see a butterfly?") involves minimal distancing. Intermediate levels of distancing are carried, for example, by demands to interpret ("What does it mean to . . . ? " ) , infer similarities, and sequence ("What do we do next?"). High levels of distancing are carried by demands to resolve conflict ("If there were only one apple left, we could still both have some") and to predict the result of transformation ("What do you need to do to the water to change it into ice?"). Sigel (1981) proposes that asking (i.e., inquiry at the various distancing levels) plays a particularly important role in the development of representational thought in the preschool child. An inquiry can create a mental discrepancy, which may result in disequilibrium. The child will try to resolve the disequilibrium through some mental action. Although the child's cognitive-developmental level will constrain to some extent the resolution he or she can achieve, creation of discrepancy provides the opportunity for structural reorganization as well as practice in moving mentally from the present situation to the past or the future. Mental-distancing strategies clearly are aimed at development of decontextualized structures that enable mental movement beyond the present to reconstruction of the past or anticipation of the future. Such "representational competence" (Sigel, 1981) is essential for development of executive structures. Sigel also highlights the role that language may play in Piagetian 201
Janice Johnson development. We cannot assume that children in a bilingual learning environment are exposed to more mental-distancing strategies, as described by Sigel, than are monolinguals. Might there be some advantage, however, in encountering distancing experiences in two languages? The advantage, again, would seem to be in adding linguistic variance to the contexts in which mentaldistancing experiences are encountered. This should facilitate the detachment of linguistic structures from the (logological and infralogical) representational structures needed for mental distancing. Also, discrepancies may be produced by differences between the two languages themselves (e.g., grammatical differences or lack of exact translation equivalents). This may result in spontaneous use of certain distancing strategies (e.g., infer similarities, transform) and construction of higher-level representational structures for operating on language itself. The dialectical-constructivist perspective
The important role of executive structures in guiding performance is well recognized by neo-Piagetian theory (Case, 1985; Pascual-Leone, 1976a, 1984; Pascual-Leone et al., 1978) and by current theories of learning (Brown etal., 1983) and of intelligence (Sternberg, 1985). As stated above, PascualLeone (Pascual-Leone et al., 1978) distinguishes two kinds of executive: task (or plan) executives serve to co-ordinate relevant action schemes into a performance and to direct the sequence of mental or behavioral steps that generate this performance; control executives monitor the use and regulation of cognitive resources. Executives are acquired by way of relational learning, which, according to Pascual-Leone (Pascual-Leone & Goodman, 1979), comes about when mental or behavioral schemes (or structures) that are cofunctional (i.e., directed toward the same situational praxis) are activated simultaneously or in close sequence. This pattern of activated schemes eventually is "rewritten" as a new mental structure, which can subsequently cue and be cued by the component schemes. Performance patterns in which the new structure participates can be written recursively into yet higher-level relational structures. Infralogical (experiential-symbolic) executive structures are developed within particular contexts and serve to guide performance within restricted domains - they are often developed through repeated practice (e.g., learning to knit or to solve crossword puzzles). Because the context is a constant in the learning, its representation will become one of the component schemes in the infralogical executive; that is, the executive will not be decontextualized. Logological (generic-symbolic) executive structures reflect performance invariances across contexts that the subject construes as functionally related 202
Constructive processes in bilingualism to some praxis (i.e., pragmatic achievement). According to Pascual-Leone, logological executives are never developed through simple repeated practice, but rather through practice and the application of mental effort that is, co-ordination of a new structure through application of mental attention to the generic, relevant invariant aspects (Pascual-Leone, 1976a, 1984). Human mediation (see discussion of Vygotsky above) is often needed for learning logological executives. Recall that in a mediational situation the tutor often highlights (i.e., brings to the subject's attention) executive components as distinct from the particular context or action schemes needed to carry out the plan. In the process the subject may learn executives that might be applicable to other different tasks, such as memory retrieval, mentally moving backward and forward in time, mentally moving from the perceptual to the intellective level of representation, temporal organization of performance, etc. These are generic executives that can apply across many types of context. The mental-distancing strategies described by Sigel are aimed at developing such logological executives along with generalized infralogical (i.e., experiential-symbolic and quasi-generic) ones. An important function of executives is to mobilize mental-attentional energy and allocate it to structures that are not highly activated, but are relevant for performance. That is, the executive can guide application of mental effort in pursuit of the executive goal. Executive-guided application of mental attention will be needed in the following kinds of situation (Johnson, Fabian, & Pascual-Leone, 1989; Pascual-Leone et al., 1978): when the structures needed for performance are not strongly activated by the situation (i.e., when there is not strong contextual support for performance); when correct performance is not highly determined by past learning (e.g., in novel task situations); and especially when contextual or learning factors cue structures that, in their application, are incompatible with correct performance. This last case constitutes a misleading situation (Johnson, Fabian, & Pascual-Leone, 1989; Pascual-Leone et al., 1978). As we shall see later, tasks where bilinguals exhibit superior performance are often of the misleading type. The sorts of language tasks that Bialystok (Bialystok & Ryan, 1985) considers "high-control" tasks are in fact misleading in the sense defined here. Claims made in the discussions of Vygotsky and Piaget, regarding possible relations between bilingualism and the acquisition of executives and logological structures, are consistent with dialectical-constructivist theory. Within this theory Pascual-Leone has refined the notion of control executives (or controls). Using the term "controls" in the narrow sense defined above, we might ask: what role might a bilingual learning situation play in the development of controls? Unique to the bilingual situation are the tasks of flexibly switching between languages (as a speaker and as a listener) and of translating between languages. Consider the task of translating and 203
Janice Johnson its requirements for movement among levels of mental representation. Translating from language A to language B requires the translator to mentally move from the linguistic-representational level of A, to the logologicai level of word reference (i.e., "meaning" as defined by Vygotsky, 1962), to the infralogical level for the precise sense in the current context; processing must then move back to the logologicai level of meaning and finally to the linguistic-structural level of language B. Continuing to translate requires continued mental movement among these levels of representation. In the process the subject receives practice in decentration - in executiveguided change of mental attention among different levels of representation of the same content (Pascual-Leone, 1984). The development of decentration executives in the language context may facilitate acquisition of generic decentration controls that can apply as needed in other contexts. This may be an explication of the "cognitive flexibility" that has been attributed to bilinguals (Ben-Zeev, 1977a, 1977b; Cummins, 1976; Diaz, 1983; Peal & Lambert, 1962). Note that with much practice at translation, the mental movement among levels in the translation context may become automatized resulting in increased speed and fluency. Research on bilingualism and cognition
Let us use the ideas developed above to explore the literature on bilingualism and performance on language and cognitive tasks. I will organize this literature into four kinds of task, each kind to be discussed in order. The kinds are: (1) verbal tasks in which situationally cued infralogical processing conflicts with necessary logologicai processing; (2) perceptual-cognitive tasks in which the subject's initial organization of the stimulus situation conflicts with necessary reorganization; (3) tasks requiring linguistic control of overt action, in situations in which verbal action conflicts with motor action; and (4) verbal tasks in situations that do not engender cognitive conflict. Note that the first three kinds of task all involve situations that are misleading in the sense defined above: situations in which contextual or learning factors cue structures that, in their application, are incompatible with correct performance. The fourth kind of task situation is not misleading. This distinction will prove to be important for explaining the data on bilingualism and cognition. The discussion below will address effects that bilingualism could have in principle. The literature contains much discussion of moderating effects of the context of bilingualism (e.g., additive vs. subtractive bilingualism, threshold levels of bilingualism, etc.; Cummins, 1976, 1979; De Avila & Duncan, 1980; Diaz, 1985; Hakuta, 1986). These are important issues that will not be dealt with in detail here. It is important to note, however, that bilingualism is only one possible means to the kind of structural enrich204
Constructive processes in bilingualism ment I have described. There are other means (environmental and human mediational), as described for example by Vygotsky, Sigel, Piaget, and neo-Piagetians (Case, 1985; Pascual-Leone etal., 1978; Pascual-Leone & Ijaz, 1989). These circumstances make essential proper matching of bilingual and monolingual samples, so that valid comparisons can be made. Verbal tasks in misleading situations There is a class of verbal task in which bilingual children are often found to excel their monolingual peers. These are tasks that are misleading in that they create in the subject a conflict between solutions by means of two kinds of structure. Successful performance requires use of logological structures, but the task situation strongly cues infralogical structures. There is cognitive conflict because the two kinds of structure can apply on the same content, but they lead to different performance (Johnson, Fabian, & Pascual-Leone, 1989; Pascual-Leone & Goodman, 1979; Pascual-Leone et al.9 1978). In these situations, bilinguals appear more likely than monolinguals to resolve the conflict in the logological direction. Characteristic tasks of this class are those that Bialystok (Bialystok & Ryan, 1985) claims require high cognitive control. Take, for example, her grammaticality-judgment task (Bialystok, 1986). In this task the child must judge whether sentences are grammatical or not. Four types of sentence are presented: those in which grammar and semantics are congruent (either correct or incorrect) and those in which they are incongruent (one correct and the other incorrect). Bialystok predicts that the type requiring highest cognitive control is the incongruent case, where the grammar is correct but the semantics anomalous. The child here must judge the sentence correct despite its inappropriate semantics. Consistent with Bialystok's prediction, bilingual children perform better than monolinguals on this type of item, but not on other types. What does this task require of the child? We have said that the communicative function of language makes strong the connection between linguistic and infralogical structures. In other words, as a result of past learning the linguistic situation is highly likely to cue infralogical processing. Anomalous semantics implies that there will be a mismatch between linguistic and infralogical structure (i.e., the sentence will not "make sense"). Infralogical processing will therefore lead to rejection of the sentence. The task instructions (which are carried in the task executive), however, direct the child to attend to the grammar. Presumably these sentences have correct organization in terms of functional categories, but the instances of the categories violate semantic selection restrictions. The sentence therefore must be processed at the categorical - the logological - level. Infralogical structures will be activated by the content of the sentence. Logological 205
Janice Johnson structures can be activated only by relational patterns in the sentence; executive-guided application of mental attention is likely to be needed for logological structures to dominate processing. Correct performance is thus more likely to come about if the child has acquired the appropriate logological structures and decentration executives; I argued above that such acquisition may be facilitated by a bilingual learning environment. Note that the claim is not that bilingual children necessarily know more of the particular, concrete grammar of a given language (linguistic structure): Bialystok (1986) did not find bilinguals better able to correct grammar; Ben-Zeev (1977b) found no bilingual advantage in ordinary grammarrule usage; and Hakuta (1987) found that bilingualism related more to ability to note grammatical errors than ability to correct them (see Bialystok, 1988, for an exception). The claim is rather that bilinguals have better structures for the logic of grammar (logological structures). Thus they may be better able to focus on logical structure, as opposed to meaning, when making decisions about grammaticality in misleading situations. Logological structures are not sufficient for correcting grammar, however; here linguistic or linguistic-logological structures within the target language are required. In sum, a bilingual advantage is likely to occur when the task situation pits logological processing against other forms of processing that are cued by the task situation. Further examples from Bialystok's work are bilingual children's superior ability to (1) avoid correcting meaning when correcting grammar (Bialystok, 1986); (2) not be disrupted in word counting in the context of meaningful sentences (vs. scrambled strings) and words of more than one syllable (vs. monosyllables; Bialystok, 1987); and (3) not be disrupted by a misleading context when choosing rhymes or synonyms, as directed, for a stimulus word (Bialystok, 1987). When the misleading aspect is removed, monolinguals do as well as or better than bilinguals in all these tasks. Another characteristic task of this class (conflict between use of a logological functional-category vs. an overlearned linguistic-infralogical connection) are verbal symbol-substitution tasks. For example, in tasks that require the child to switch common labels for objects, the overlearned connection of the label to the infralogical referent becomes misleading. Although there are exceptions (Ben-Zeev, 1977b; Cummins, 1978b, study 2), the literature indicates a strong tendency for bilingual children to outperform their monolingual peers in this type of task (Ben-Zeev, 1977a; Bialystok, 1988; Cummins, 1978a; Feldman & Shen, 1971; Ianco-Worrall, 1972; Edwards & Christophersen, 1988, report a correlation between degree of bilingualism and performance on verbal-symbol substitution). Cummins (1978a) provides additional examples of this general class of verbal task. He found bilingual children better able to assert the (logologi206
Constructive processes in bilingualism cal) meaning of words in the face of hypothetical destruction of the words' (infralogical) referents. A second task required the subject to judge and justify the truth, falsity, or indeterminacy of "non-empirical" statements (i.e., contradictions or tautologies) about poker chips that were either in the view of the child or hidden. Bilingual children were better able than their monolingual peers to respond based on the logic of the sentence (logological) rather than their actual knowledge concerning the chip (infralogical). Cummins (1978b) found bilingual children better able to recognize and justify grammatically based ambiguities in sentences. In a task such as this, the first meaning seen is still highly activated at the point when one seeks an alternative meaning; that is, the first response becomes misleading vis-d-vis the second (Goodman, 1971). One needs good decentration executives to allow one to deactivate the first meaning and apply mental attention to an alternative logical structure for the sentence. The verbal-transformations task (Ben-Zeev, 1977a, 1977b) presents a similar case at the perceptual level. Let us consider one last verbal task in this section. Bialystok (1988) found bilingual children to be better than monolinguals at giving formal definitions in response to the questions "What is a word?", "How can you tell if something is a word?" Properly, this is not a misleading situation. What is required is to respond by describing the functional category "word" rather than giving concrete referents of "word." If, as has been suggested, a bilingual learning environment facilitates construction of such a category (logological structure), then we would expect superior performance by young bilinguals. Perceptual-cognitive tasks in misleading situations A second class of task is constituted by many of the non-verbal tasks on which bilingual children have been found to excel their monolingual peers. These tasks are not purely perceptual; rather, they require some inference or mental reorganization - thus I label them "perceptual-cognitive." More specifically, these are tasks in which an initial organization of the stimulus situation becomes misleading vis-a-vis subsequent necessary reorganization. Correct performance on this sort of task requires executive mobility - that is, ability to flexibly change one's structuring of the situation. There are two forms of executive mobility: figurative and operative (Pascual-Leone, 1984,1989). Figurative structures stand for cognitive states - for objects and for properties therein (including verbal/linguistic objects). Operative structures are transformations that apply to change physical objects (and/or the mental objects that represent them). Verbs (other than the copula) are examples of linguistic operatives.5 207
Janice Johnson A task that requires figurative mobility requires the subject to generate more than one figurative construal of the stimulus situation (e.g., mentally reorganize his or her initial perception). Tasks that require operative mobility require the subject to apply more than one operative strategy to the stimulus situation (i.e., revise tactics and apply a new approach to the task). In both these cases, the initial construal of the situation becomes misleading vis-a-vis subsequent reorganization, and often these tasks are constructed so as to bias an initial incorrect construal (e.g., due to gestaltist features or a misleading learning set). Whether the mobility required is in the figurative or the operative domain, flexible decentration executives are needed so that a new assignment of (figurative or operative) structures can be generated. It was suggested above that, other things being equal, bilingualism might foster acquisition of such executives. It is difficult to say for the general case whether bilingualism would foster more executives for figurative or operative mobility. The need to switch grammatical structure would involve operative transformation; switching labels would take place in the figurative domain (Goodman, 1971). The context of learning and language use (e.g., how much language-switching is done) would play a role here. Insofar as a bilingual learning environment could be one way to foster growth of decentration executives (other ways are described, for example, by some of Sigel's mental-distancing strategies), we might expect some bilingual advantage on tasks that require executive mobility. The claim for a relation between language processing and figurative and operative mobility is supported by Goodman's (1971) research (done in collaboration with Pascual-Leone). Goodman found that non-verbal measures of field dependence-independence (related to figurative mobility) and adaptive flexibility (related to operative mobility) predicted the ability of adult monolinguals to produce multiple meanings for ambiguous sentences. High adaptive flexibility was related in particular to generation of multiple syntactical forms for an input sentence. Relevant to the issue of executive mobility are results reported by Peal and Lambert (1962). These researchers found that bilingual children excelled over their monolingual peers on non-verbal tasks that required "mental reorganization," but not on tasks with simple spatial-perceptual requirements. As suggested above figurative and/or operative mobility may be required for solution of "mental-reorganization" tasks. One of the tasks for which Peal and Lambert (1962) found a bilingual advantage - Ravens Progressive Matrices - requires both kinds of mobility: figurative, to imagine each alternative as filling the missing space in the matrix; operative, to switch (i.e., to scan and select) among alternatives. Ben-Zeev (1977a) also employed the Ravens. Her bilingual subjects (who were younger than Peal and Lambert's) did not exhibit superior perform208
Constructive processes in bilingualism ance, but they did show fewer "failure to scan" errors, as compared with monolinguals; and this suggests the possibility of greater operative mobility. Hakuta (1987), working with young Spanish-English children, reports some relation between degree of bilingualism and performance on Ravens. BenZeev (1977a) found no bilingual-monolingual difference in solution of a matrix transposition task; but she did find bilinguals to outperform monolinguals on the descriptive part of the task. That is, under prompting by the experimenter to say in what way pairs of cylinders were "the same" and "different" from each other, bilingual children were better able to attend to both relevant dimensions (i.e., apply two figurative schemes height and weight); this again requires decentration in the figurative domain. Embedded-figures tests - which require the subject to find a simple figure within a more complex drawing - also fall within this class. Figurative mobility is required in embedded figures in order to mentally break the complex pattern (whose perception is fostered by gestaltist features) in order to mentally search for features of the simple figure (Pascual-Leone, 1989; Pascual-Leone & Goodman, 1979). Balkan (1970, cited in Diaz, 1983) and De Avila and Duncan (1980) have found bilingual children to perform better than their monolingual peers on embedded-figures tasks. Several researchers have found young bilingual children to perform better than their monolingual peers on conservation-like tasks: Liedtke and Nelson (1968 - conservation of length), Feldman and Shen (1971 - object constancy, which is similar to identity conservation), De Avila and Duncan (1980, paper-and-pencil measures of perspective-taking and of conservation of substance). For each kind of conservation, group differences were found in the age-ranges when the conservation should just have been emerging. Pascual-Leone (1976b; Pascual-Leone et al., 1978) has task analyzed the various conservations, specifying the cognitive requirements for their spontaneous solution (i.e., solution without prior acquisition of a logical structure for conservation). All the conservations present a misleading perceptual aspect; this creates in the subject a bias towards a perceptual strategy that leads to incorrect solution. Correct solution requires activation of an alternate strategy (e.g., an "identity-reversibility" strategy in the case of substance conservation), and executive-guided application of mental attention to task-relevant non-perceptual schemes. Spontaneous solution will be facilitated if the subject has a decentration executive to monitor the shift from the perceptual to the non-perceptual intellective strategy (Pascual-Leone, 1976b; Pascual-Leone et al., 1978). If the bilingual situation fosters the practice of decentration, this could give young bilinguals an initial advantage in spontaneous solution of conservation problems. Ben-Zeev (1977b) found that bilingual children performed better than monolinguals on a classification task. Subjects had to classify a set of items 209
Janice Johnson into two groups, and then switch to a second and a third type of classification of the same items. The situation is facilitating for the first sorting - the subject simply picks one distinctive feature of the items and sorts on that. This first sorting then becomes misleading vis-a-vis the second: in the face of the perceptual configuration produced by the initial sorting, the subject must activate a strategy for a different sorting - must switch strategies. Operative mobility will aid performance on second and third classifications. Note that when a subject persisted with the same classification twice, instead of switching, the experimenter added on the next trial items ("hints") that would make difficult continuing with the same strategy (e.g., if the subject was classifying on the basis of two item-shapes, the experimenter would introduce a third shape). Ben-Zeev (1977b) reports that the "bilinguals were significantly better able to use these hints as cues to successful restructuring" (p. 40). This suggests that the bilinguals might have acquired an executive strategy to decentrate when faced with contradiction - perhaps based on experience with contradiction between their two language systems.
Tasks requiring linguistic control of overt action in misleading situations The data of most relevance here come from an exemplary study by Bain and Yu (1980). The study examined the possible relation between bilingualism (specifically, language acquisition in a one-parent, one-language home environment) and growth of linguistic control of action in the child. Working within a Vygotskian framework (Luria, 1961; Vygotsky, 1962), Bain and Yu proposed that a one-parent, one-language bilingual upbringing might foster the child's early mastery of language and consequently early voluntary control of cognitive processes. Subjects came from France, Canada, and Hong Kong; at each location there were one bilingual and two monolingual samples. The children were tested twice, once at 22-24 months of age and once at 46-48 months, on tasks that had been devised by Luria (1961); bilingual children received alternate trials in each of their languages. Testing at the younger age involved hiding, outside the child's view, a marble under a tumbler or a cup, and then verbally instructing the child where to find the marble. The simplest version of the task involved simple direct instruction ("The marble is under the tumbler . . . Find the marble"). The next level of the task introduced a 10 sec. delay between the information regarding location and the instruction to find the object. The third version (like Piaget's object-permanence tasks) introduced a misleading learning set: the child was instructed on three consecutive trials to find the object in one location ("The marble is under the cup . . . Find the marble" for three trials), then on the fourth trial was told to find it in the other location ("The marble is under the tumbler . . . Find the marble."). The fourth 210
Constructive processes in bilingualism version of the task was like the third, but with introduction of a 10 sec. delay (as in version 2). This set of tasks concerns the ability of language to initiate action in the child; this is basically a signalic function. The child must be able to hold mentally active the linguistic term (or its infralogical referent) during the delay period. As well in trial 4 of versions 3 and 4, linguistic control of action (in this case, the ability of the linguistic term to cue its infralogical referent in the child) must be strong enough to overcome the misleading action scheme acquired in the three prior trials. The bilingual samples performed consistently better than the monolingual samples (the differential increasing with the difficulty of the task version), but this difference was not statistically significant. Statistically significant results obtained in the second testing of the children (see below) suggest that the small performance differential at two years might be a valid, early symptom of a bilingual-monolingual difference. So let us consider how a bilingual environment might relate to performance on the first set of tasks. Cups and tumblers share numerous properties and from the viewpoint of a two-year-old are likely to serve much the same function. Linguistic labeling thus may play a significant role in drawing the child's attention to distinctive features of these two objects; that is, in the child's construction of distinct infralogical referents for the objects. This task concerns the ability of the linguistic label to impel action toward the appropriate concrete referent. It is possible that some consistent linguistic variation (i.e., dual labels for each object) aids in construction of the infralogical referent. In addition, the difficulty of sorting out the two languages at this age may cause increased attention to language in the child; such attention is needed in this set of tasks. The children were tested again at about four years of age. The task now concerned the ability of language to regulate and inhibit action. This is the symbolic function of language, which requires that meaning be detached from the immediate context. The child had to follow verbal instructions regarding whether or not to squeeze a ball in response to onset of a red or a green light. The first, and easiest, version of the task involved simply following the adult's instructions which varied from trial to trial: "When the red light goes on squeeze the ball"; "When the green light goes on squeeze the ball"; "When the red light goes on squeeze the ball. When the green light goes on don't squeeze the ball"; etc. The second version of the task required the child to instruct him or herself overtly: "When the green light goes on say 'squeeze' and squeeze the ball. Say it and do it." More importantly, in this version some trials introduced a conflict between what the child had to say and to do (i.e., overt verbal action was made misleading vis-a-vis overt motor action): "When the red light goes on say 'squeeze' but don't squeeze the ball. Say it but don't 211
Janice Johnson do it"; "When the red light goes on say 'squeeze' and squeeze the ball. Say it and do it. When the green light goes on say 'don't squeeze' and don't squeeze the ball. Say it but don't do it." The third version of the task was similar to the second, but required the child to instruct him or herself covertly (silently); the trial was scored incorrect if the child made an overt verbal response. Bain and Yu found no group differences in performance on version 1, but found a consistent and statistically significant bilingual advantage on versions 2 and 3. Note that, particularly in versions 2 and 3, if language is playing an impelling (or signalic) function for the child (i.e., if the linguistic term serves to make real the infralogical referent), the child is bound to fail. Verbal action will impel motor action: both "squeeze" and "don't squeeze" will initiate squeezing, as Luria (1961) found. For language to serve the required symbolic or significative function here (i.e., for it to mediate rather than impel action) the linguistic structure must be functionally detached from its action referent. I argued above that a bilingual learning environment might facilitate this detachment process. Note that in the current case a main problem is for the child to use the task instructions to construct an infralogical executive to guide performance (infralogical because it is geared to action processing in the immediate task context). An executive is an internal mediational structure, and necessary for its construction is that the key terms in the instructions have symbolic rather than signalic value for the child. One of the sources of difficulty in version 3 (covert self-instruction) is probably the need to suppress overt verbal action; the ability to do this is a further symptom of language taking on a symbolic rather than a signalic function. The claim of Diaz and Klingler (this volume) that bilingual children show advantages in the use of self-regulatory language seems a logical extension of the findings of Bain and Yu. Self-regulatory language provides overt evidence for the child's use of task executives.
Verbal tasks in non-misleading situations Included in this section are verbal tasks that show no particular bilingual advantage. This class of task generally presents no misleading aspect, and is concerned with semantics (i.e., meaning or sense) rather than grammatical structure. Peabody Picture Vocabulary - a vocabulary measure often used in studies of bilingualism - is an example of such a task. In the Peabody, the subject is given a lexical term and must select its proper referent from among four pictures; because of the format, the lexical terms usually have concrete referents. Such a task is facilitating (Johnson, Fabian, & PascualLeone 1989; Pascual-Leone et aL, 1978) in that all task aspects (including 212
Constructive processes in bilingualism the subject's past learning) are relevant to correct solution; there is no need for sophisticated executives here. The subject must have previously acquired the lexical term and must be able to recognize a depiction of its referent. Several studies have found monolinguals to have an advantage over bilinguals on picture vocabulary (Ben-Zeev, 1977a, 1977b; Bialystok, 1988). Ben-Zeev (1977b) claims that this is to be expected, because "having to share their language experience between two languages, the bilinguals have less opportunity for experience with the vocabulary of either" (p. 35). This is consistent with my claim above that there is no reason to expect a bilingual advantage at the level of purely linguistic structure. In fact, as Ben-Zeev states, on average, one would expect a monolingual advantage here. Ben-Zeev (1977a, 1977b) predicted a bilingual advantage in paradigmatic word association (i.e., giving a response that is of the same part of speech as the stimulus word), because of an expectation that bilinguals have more developed conceptual categories than do monolinguals. Note that under the operational definition, paradigmatic associates need not fall under the same conceptual category (logological structure); they might instead both relate to the same experiential situation (i.e., be components of an infralogical structure; e.g., eat - chew). In two samples of bilingual children BenZeev found no group differences in paradigmatic word association; she attributed this to the lower vocabulary of the bilinguals. Word association is a facilitating task - one simply mijst "open one's mind" to allow the stimulus word to cue other structures. One of these structures will come to be more highly activated than the others (due to learning factors or "semantic-atmosphere" effects) and will dominate performance. The researcher scores the response as being based on either a semantic or linguistic association. Although Ben-Zeev's hypothesis is not unreasonable and her explanation may be correct, given the lack of misleading factors there seems no strong reason to expect a bilingual advantage for paradigmatic associates. Similarly, there is little reason to predict bilingual-monolingual differences on verbal-recall tasks, unless less well-developed linguistic structures (e.g., vocabulary) might lead to bilingual disadvantage (Cummins & Gulutsan, 1974; Goldman, Reyes, & Varnhagen, 1984). Snow etal. (this volume) report that bilingual children show no advantage over monolinguals in giving formal definitions, and that bilinguals are only slightly more likely to give formal and complete definitions in their stronger as compared with their weaker language. Note that constructing a definition is also a task without misleading aspects. In defining a term, the subject must first go to his or her infralogical structures - that is, to representations of experiential situations that serve as referents for the term. If the subject describes one of these structures the result will be an example, considered 213
Janice Johnson by Snow et al. to be an "informal definition/' A formal definition requires the subject to look for commonalities across relevant infralogical structures in order to abstract the essential invariant (the logological structure) that is the meaning of the term. Linguistic structure is needed to initiate cuing of the infralogical structures (i.e., the subject must be familiar with - have referents for- the lexical term) and it is needed to communicate the conceptual invariant. The subject may also have previously learned the conventional form for giving definitions. Construction of the definition requires, however, relevant infralogical structures and sufficient mental-attentional capacity to enable the subject to notice and hold in mind essential invariances across the infralogical structures. Young children probably fail to give formal definitions due to a combined lack of relevant experiential structures and sufficient mental-attentional capacity.
One final task I will examine is that of metaphor interpretation. A metaphor relates terms from two disparate, conceptual or experiential, domains. These domains are often labeled the topic and the vehicle of the metaphor. The topic is what the metaphor is about, and the vehicle an expression used to say something about the topic (e.g., in "My sister was a mirror," the topic sister is described in terms of the vehicle mirror). Metaphor interpretation is a process of semantic transformation, in which semantic aspects of the vehicle are transformed so that they can apply in the semantic domain of the topic. Developmental change in metaphor interpretation can be characterized in terms of degree of transformation underlying construction of the interpretation: with age, increasingly more complex transformations are applied (Johnson, Fabian, & Pascual-Leone, 1989; Johnson & Pascual-Leone, 1989). My research has employed context-free (i.e., decontextualized) metaphoric sentences, which the subject is asked to interpret. The sentences have similar syntax ("topic was a vehicle") and use topic and vehicle terms with familiar referents (topics: sister, shirt; vehicles: rock, mirror, butterfly). Like the other tasks of this final class, the metaphor task is facilitating (there are no strong misleading aspects, and all the subject's knowledge about the topics and vehicles is potentially relevant), and it is concerned with meaning (Johnson, 1990; Johnson, Fabian, & Pascual-Leone, 1989). Thus, there is no strong reason to predict a bilingual advantage in metaphor interpretation. I have reported elsewhere (Johnson, 1990) results of research aimed at examining the role of linguistic versus logological structure in metaphor interpretation. In this research a bicultural developmental design was used, 214
Constructive processes in bilingualism with children of English and of Spanish home background. The goal was to have samples differing in level of English proficiency. By properly selecting subsamples, however, it is possible also to examine effects of bilingualism, and I do this here. Children (ages 7-8, 9-10, and 11-12 years) constituting the subsamples came from primary schools in working-class communities of Toronto; all were being instructed in English. The English-speaking subjects were monolingual. English was the second language of their Spanish-speaking classmates, who were selected on the basis of a parental questionnaire; for selected children, parents spoke Spanish at home often or always, and children spoke Spanish at home at least sometimes. The children were tested with a non-verbal measure of mental-attentional capacity (i.e., developmental level; the Figural Intersections Test, PascualLeone & Ijaz, 1989) and with measures of oral-language proficiency (i.e., English and Spanish versions of the oral-language subscale from the Woodcock Language Proficiency Battery - Woodcock, 1980, 1981 - and a story retelling task - De Avila & Duncan, 1983). Within each age group of the Spanish-speaking sample, six "bilingual" subjects were selected, that is, subjects showing evidence of good proficiency in both languages. The Woodcock standard-score scale is based upon a mean of 100 and a standard deviation of 15. Subjects were selected into the bilingual group if they had an English standard score of 83 or higher combined with a Spanish standard score of 82 or higher or a Spanish standard score in the 70s accompanied by a Spanish story-retelling score at the "proficient-speaker" level. These were the most stringent criteria that could be applied and still yield a minimally acceptable sample size. A group of "marginally bilingual" subjects was constituted by six Spanish-speaking subjects per age group who were matched with the "bilinguals" on English standard score, but who were low in Spanish. "Monolingual" subjects were English-speaking children matched with bilinguals on English standard score. The bilingual group performed higher than the marginal bilingual group on the Spanish Woodcock. A series of analyses of variance yielded no statistically significant effects of language group (bilingual vs. marginal bilingual vs. monolingual) for non-verbal mental capacity, English Woodcock, or English metaphor score. There were no interactions with language group. A definitive test would require larger sample sizes and subjects who are more clearly proficient in both languages than were the current bilingual subjects. Results from the present study suggest, however, that bilingualism yields neither benefits nor deficits in level of metaphor interpretation. See Johnson (1989,1990) for discussion of group differences in metaphor interpretation when proficiency in the target language is allowed to 215
Janice Johnson vary, as well as discussion of structural dynamics in language processing (bilingual and monolingual) as suggested by correlational data. Conclusion
The chapter has described constructive processes through which bilingualism may foster cognitive growth. A central distinction was made among linguistic, logological, and infralogical structures; and cognitive-developmental theory (i.e., an integration of classic - Piaget, Vygotsky - and contemporary - Pascual-Leone - theories) was used to explain processes by which these structures are acquired. I have placed emphasis on how the bilingual experience might be accommodated within these theories of cognitive growth and what explicit predictions the theories could make regarding cognitive effects of bilingualism. The major conclusion is that the impact of bilingualism is on the construction of logological and executive structures, not on purely linguistic structures. These predictions were used to organize the literature on bilingualism and cognition into four classes of task, varying in the kinds of processes needed for task performance. Task analysis was used to explain the patterns of positive and negative findings in the literature, regarding cognitive advantages of bilingualism. Bilingual advantages tend to occur on the following: (1) verbal tasks in which situationally cued infralogical processing conflicts with necessary logological processing; (2) perceptual-cognitive tasks in which initial subjective organization of the stimulus situation conflicts with necessary reorganization; and (3) for young children, tasks requiring linguistic control of overt action in situations where verbal action conflicts with motor action. Misleadingness - a cognitive conflict between structures needed for correct performance and other structures cued by the situation - is a common psychological aspect across these three classes of task. It is proposed that specific kinds of logological and executive structures the construction of which may be facilitated by a bilingual learning environment - may aid the bilingual in resolving such cognitive conflicts in the correct direction. In the fourth class of task language serves as a cue to processing at the infralogical and logological levels, in the absence of misleading situational factors. Here no bilingual advantages are found. New data on bilingualism and metaphor interpretation were presented. Consistent with predictions, bilingualism had no effect on the level of metaphor interpretation. The chapter has focused on kinds of cognitive effects that bilingualism could have in principle. Moderating effects of the context of bilingualism have not been examined in detail. I believe, however, that such moderating effects could be accommodated within the model proposed here. What 216
Constructive processes in bilingualism is required is to examine the context in terms of the opportunities it provides for relational learning. Limits that developmental factors may place on task performance and relational learning are discussed elsewhere (Johnson, Fabian, & Pascual-Leone, 1989; Pascual-Leone, 1984,1987; Pascual-Leone & Goodman, 1979; Pascual-Leone etal., 1978). The literature provides strong evidence that bilingualism can foster cognitive growth. The present chapter has used cognitive-developmental theory to explicate the relevant research data and the mechanisms by which bilingual experience may affect cognitive functioning.
Notes 1. Preparation of this chapter was facilitated by a grant from the Social Sciences and Humanities Research Council of Canada. I am grateful to Dr. J. Pascual-Leone for valuable comments and to C. Arnold for assistance with preparation of the manuscript. 2. Elsewhere I have called these structures "experiential" or "concreteexperientiar (Johnson, 1989; Johnson, Fabian, & Pascual-Leone, 1989; Johnson & Pascual-Leone, 1989). 3. Logo means "speech" and "clear idea" - these are the logical structures subtending abstract thought and language. Elsewhere I have called these structures "conceptual" or "generic-conceptual" (Johnson, 1989; Johnson, Fabian, & Pascual-Leone, 1989; Johnson & Pascual-Leone, 1989). 4. I am not suggesting here that the concept of causality is necessarily acquired through language. It may well be acquired by means of non-linguistic experience. More complex concepts such as the relations expressed by the terms moreover or nevertheless may well be constructed with the aid of language. The not uncommon misuse of these terms by university students suggests that the students have not constructed the appropriate logological structures, rather these terms function more as purely linguistic formulas. 5. Note that the figurative-operative distinction cuts across the infralogicallogological-linguistic dimension; figurative and operative structures occur at each of the infralogical, logological, and linguistic levels (Pascual-Leone, 1984).
References Anderson, J. 1985. Cognitive psychology and its implications. New York: W. H. Freeman. Bain, B. & A. Yu 1980. Cognitive consequences of raising children bilingually: 'one parent, one language/ Canadian Journal of Psychology, 34, 304-313. 217
Janice Johnson Ben-Zeev, S. 1977a. The influence of bilingualism on cognitive strategy and cognitive development. Child Development, 48,1009-1018. 1977b. Mechanisms by which childhood bilingualism affects understanding of language and cognitive structures. In P. Hornby (ed.), Bilingualism: psychological, social, and educational implications. New York: Academic Press. Bialystok, E. 1986. Factors in the growth of linguistic awareness. Child Development, 57, 498-510. 1987. Words as things: development of word concept by bilingual children. Studies in Second Language Acquisition, 9,133-140. 1988. Levels of bilingualism and levels of linguistic awareness. Developmental Psychology, 24, 560-567. Bialystok, E. & E. Ryan 1985. A metacognitive framework for the development of first and second language skills. In D. Forrest-Pressley, G. MacKinnon, & T. Waller (eds.), Metacognition, cognition, and human performance. New York: Academic Press. Brown, A., J. Bransford, R. Ferrara, & J. Campione 1983. Learning, remembering, and understanding. In P. Mussen (series ed.), J. Flavell, & E. Markman (vol. eds.), Handbook of child psychology. Vol. Ill: Cognitive development. New York: Wiley. Case, R. 1985. Intellectual development: birth to adulthood. New York: Academic Press. Cassirer, E. 1944. An essay on man. New Haven: Yale University Press. Cummins, J. 1976. The influence of bilingualism on cognitive growth: a synthesis of research findings and explanatory hypotheses. Working Papers on Bilingualism, 9,1-43. 1978a. Bilingualism and the development of metalinguistic awareness. Journal of Cross-Cultural Psychology, 9,131-149. 1978b. Metalinguistic awareness of children in bilingual education programs: data from Irish and Canadian Ukrainian-English programs. In M. Paradis (ed.), Aspects of bilingualism. Columbia, SC: Hornbeam Press. 1979. Linguistic interdependence and the educational development of bilingual children. Review of Educational Research, 49, 222-251. Cummins, J. & M. Gulutsan 1974. Some effects of bilingualism on cognitive functioning. In S. Carey (ed.), Bilingualism, biculturalism and education. Edmonton: University of Alberta. Cummins, J., M. Swain, K. Nakajima, J. Handscombe, D. Green, & C. Tran 1984. Linguistic interdependence among Japanese and Vietnamese immigrant children. In C. Rivera (ed.), Communicative competence approaches to language assessment. Clevedon, Avon, England: Multilingual Matters. Davidson, R., S. Kline, & C. Snow 1986. Definitions and definite noun phrases: indicators of children's decontextualized language skills. Journal of Research in Childhood Education, 1, 37-48. De Avila, E. & S. Duncan 1980. The language minority child: a psychological, linguistic, and social analysis. In J. Alatis (ed.), Current issues in bilingual education. Washington, DC: Georgetown University Press. 1983. Language assessment scales (LAS). San Rafael, CA: Linguametrics Group. 218
Constructive processes in bilingualism Diaz, R. 1983. Thought in two languages: the impact of bilingualism on cognitive development. Review of Research in Education, 10, 23-54. 1985. Bilingual cognitive development: addressing three gaps in current research. Child Development, 56,1376-1388. Edwards, D. & H. Christophersen 1988. Bilingualism, literacy and meta-linguistic awareness in preschool children. British Journal of Developmental Psychology, 6, 235-244. Feldman, C. & M. Shen 1971. Some language-related cognitive advantages of bilingual five-year-olds. Journal of Genetic Psychology, 118, 235-244. Galambos, S. & K. Hakuta 1988. Subject-specific and task-specific characteristics of metalinguistic awareness in bilingual children. Applied Psycholinguistics, 9,141-162. Geva, E. & E. Ryan 1987. Linguistic knowledge and cognitive demands for academic skills in first and second language. Paper presented at the meeting of the Canadian Psychological Association, Vancouver, June. Goldman, S., M. Reyes, & C. Varnhagen 1984. Understanding fables in first and second languages. NABE Journal, 8, 35-66. Goodman, D. 1971. Cognitive style factors in linguistic performance with ambiguous sentences. Unpublished master's thesis, York University, Toronto. Hakuta, K. 1986. Mirror of language: the debate on bilingualism. New York: Basic Books. 1987. Degree of bilingualism and cognitive ability in mainland Puerto Rican children. Child Development, 58,1372-1388. Husserl, E. 1970. The crisis of European sciences and transcendental phenomenology. Evanston, IL: Northwestern University Press. Ianco-Worrall, A. 1972. Bilingualism and cognitive development. Child Development, 43, 1390-1400. Johnson, J. 1989. Factors related to cross-language transfer and metaphor interpretation in bilingual children. Applied Psycholinguistics, 10, 157-177. 1990. Developmental versus language-based factors in metaphor interpretation. Manuscript submitted for publication. Johnson, J. & J. Pascual-Leone 1989. Developmental levels of processing in metaphor interpretation. Journal of Experimental Child Psychology, 48, 1-31. Johnson, J., V. Fabian, & J. Pascual-Leone 1989. Quantitative hardware stages that constrain language development. Human Development, 32, 245-271. Langer, S. 1957. Philosophy in a new key. Cambridge, MA: Harvard University Press. Leopold, W. 1961. Patterning in children's language learning. In S. Saporta (ed.), Psycholinguistics. New York: Holt, Rinehart, and Winston. Liedtke, W. & L. Nelson 1968. Concept formation and bilingualism. Alberta Journal of Educational Research, 14, 225-232. Luria, A. 1961. The role of speech in the regulation of normal and abnormal behavior (J. Tizard, ed.). New York: Liveright. 1976. Cognitive development: its cultural and social foundations (M. Lopez-Morillas & L. Solotaroff, trans.; M. Cole, ed.). Cambridge, MA: Harvard University Press. 219
Janice Johnson Malakoff, M. 1988. The effect of language of instruction on reasoning in bilingual children. Applied Psycholinguistics, 9,17-38. Morris, C. 1946. Signs, language and behavior. New York: Prentice-Hall. Nelson, K. (ed.). 1986. Children's event knowledge: structure and function in development. Hillsdale, NJ: Erlbaum. Pascual-Leone, J. 1976a. Metasubjective problems of constructive cognition: forms of knowing and their psychological mechanism. Canadian Psychological Review J7, 110-125. 1976b. On learning and development Piagetian style, I: a reply to LefebvrePinard. Canadian Psychological Review, 17, 270-288. 1984. Attention, dialectic, and mental effort: toward an organismic theory of life stages. In M. Commons, F. Richards, & C. Armon (eds.), Beyond formal operations: late adolescent and adult cognitive development. New York: Praeger. 1987. Organismic processes for neo-Piagetian theories: a dialectical causal account of cognitive development. International Journal of Psychology, 22, 531-570. 1989. An organismic process model of Witkin's field-dependence-independence. In T. Globerson & T. Zelniker (eds.), Cognitive style and cognitive development. Norwood, NJ: Ablex. Pascual-Leone, J. & D. Goodman 1979. Intelligence and experience: a neoPiagetian approach. Instructional Science, 8, 301-367. Pascual-Leone, J. & H. Ijaz 1989. Mental capacity testing as a form of intellectualdevelopmental assessment. In R. Samuda, S. Kong, J. Cummins, J. Lewis, & J. Pascual-Leone (eds.), Assessment and placement of minority students. Toronto: Hogrefe International. Pascual-Leone, J., J. Johnson, & N. Benson 1989. Mental capacity constraints on symbolic processing. In J. Johnson (chair), Mental capacity constraints on cognition: studies from the theory of constructive operators. Symposium conducted at the meeting of the International Society for the Study of Behavioural Development, Jyvaskyla, Finland. Pascual-Leone, J., D. Goodman, P. Ammon, & I. Subelman 1978. Piagetian theory and neo-Piagetian analysis as psychological guides in education. In J. Gallagher & J. Easley (eds.), Knowledge and development, vol. II. New York: Plenum. Peal, E. & W. Lambert 1962. The relation of bilingualism to intelligence. Psychological Monographs: General and Applied, 76 (27, whole no. 546). Peirce, C. S. 1955. Philosophical writings of Peirce (J. Buchler, ed.). New York: Dover. Piaget, J. 1962. Play, dreams and imitation in childhood. New York: Norton. Piaget, J. & B. Inhelder 1967. The child's conception of space. New York: Norton. 1969. The psychology of the child. New York: Basic Books. Rosch, E. 1975. Cognitive representations of semantic categories. Journal of Experimental Psychology: General, 104, 192-223. Schank, R. & R. Abelson 1977. Scripts, plans, goals, and understanding. Hillsdale, NJ: Erlbaum. Sigel, I. 1981. Social experience in the development of representational thought: distancing theory. In I. Sigel, D. Brodzinsky, & R. Golinkoff (eds.), New directions in Piagetian theory and practice. Hillsdale, NJ: Erlbaum. 220
Constructive processes in bilingualism 1982. The relationship between parental distancing strategies and the child's cognitive behavior. In L. Laosa & I. Sigel (eds.), Families as learning environments for children. New York: Plenum. Sternberg, R. 1985. Beyond IQ: a triarchic theory of human intelligence. New York: Cambridge University Press. Vygotsky, L. S. 1962. Thought and language (E. Hanfmann & G. Vakar, trans.). Cambridge, MA: M.I.T. Press. 1978. Mind in society (M. Cole, V. John-Steiner, S. Scribner, & E. Souberman, eds.). Cambridge, MA: Harvard University Press. Wertsch, J. 1985. Vygotsky and the social formation of mind. Cambridge, MA: Harvard University Press. Woodcock, R. 1980. Woodcock Language Proficiency Battery, English form. Hingham, MA: Teaching Resources Corporation. 1981. Woodcock Language Proficiency Battery, Spanish form. Hingham, MA: Teaching Resources Corporation.
221
10 Language, cognition, and education of bilingual children ELLEN BIALYSTOK and JIM CUMMINS
The theory and research reported in the preceding chapters of this volume describe a variety of aspects of cognitive and linguistic processing by bilingual children. Bilingual children of different ages and having had different experiences with bilingualism were studied for their ability to learn the second language, perform in school tasks, use language for special purposes, and solve a variety of both linguistic and non-linguistic problems. The purpose of this chapter is to evaluate the evidence reported in this volume with respect to its implications for the study of the cognitive development, linguistic proficiency, and educational prognosis for bilingual children. At the same time, however, more general insights into these areas of cognition, language ability, and education will inevitably emerge. The implications of the study of bilingual children, that is, extend beyond a description of bilingual competence into broader domains of inquiry. Our discussion is divided into two parts. First is an evaluation of the place and prognosis of bilingual children in educational systems. This discussion will also address the implications of these findings for the various programs available to bilingual children in schools. Second is an evaluation of the consequences of bilingualism for cognitive processing. This discussion will focus on the relation between linguistic and other cognitive skills and attempt to set the parameters for a description of the cognitive and linguistic development of bilingual children. Although the findings presented in this volume appear to be at times inconclusive and at times contradictory, our assumption is that they in fact form a coherent body of data that holds an important key to our understanding of the cognitive abilities of both bilingual and monolingual children. These are the findings that have emerged from investigations that vary enormously in methodology, subject populations, and theoretical orientation. Our interpretation of them must attempt to uncover the factors that make them all true descriptions. These are the findings that need to be explained if we are to understand the cognitive and linguistic function222
Language, cognition, and education ing of bilingual children and to appreciate the type of experience they have in schools. The theoretical and practical consequences of this issue are immense. More precise descriptions of language learning and language use for both monolingual and bilingual children can form the basis for educational practice and innovation. Current research studies of bilingual children, including those reported in the present volume but extending as well to the general style of research undertaken in the last decade, differ from the early research of several decades ago in important ways. Accordingly, the nature of the evidence accumulated about the abilities of bilingual children has changed with the change in paradigms. These changes have invited revaluations of our conceptions regarding bilingual children. Early research into the effects of bilingualism on children's cognition had traditionally followed from categorical predictions of possible outcomes. The motivating question was to determine if bilingualism had a general debilitating or enhancing role in language and cognitive development. Some of the research was biased by the political and educational agendas described in the next section, and outcomes were almost predetermined (Hakuta, 1986). Research undertaken in that spirit of making categorical decisions about bilingual children was predictably indecisive. One of the primary contributions of the papers in this volume is that each begins by setting out distinctions that must be followed before such categorical questions can even be addressed. Rather than assuming that bilingualism and monolingualism are the relevant categories, each chapter proposes some further distinction within which systematic differences can be discussed. These distinctions pertain to categories within the type of bilingualism, the type of task being used to assess performance, and the type of competence invoked by the task. Regarding type of bilingualism, for example, important differences are attributed to the distinction between simultaneous and successive acquisition of the second language (Sharwood Smith, Watson). Diaz and Klingler further suggest that within successive bilingualism, the level of balance between the two languages needs careful attention as outcomes on cognition vary as a function of this balance*. Regarding the type of task, a recurring distinction is drawn between contextualized (e.g. conversational) and decontextualized (e.g. school) tasks (Cummins, Snow et al.). Regarding the types of processing involved, distinctions made refer not only to broad domains of processing, such as cognitive versus linguistic (Wong Fillmore) but also to types of linguistic (Bialystok) and types of cognitive (Johnson) processing needed to solve specific problems. All of these distinctions turn out to be significant in describing performance and all have bearing on the course of development for specific achievements. The question to be addressed in this chapter is to assess the significance of these findings, 223
Ellen Bialystok and Jim Cummins and, more particularly, these distinctions, for the education and cognitive development of bilingual children. The education of bilingual children
The upsurge of academic interest in the phenomenon of bilingualism during the past twenty years has paralleled the increasing frequency and social significance of the phenomenon in many countries. Rapid population shifts, ease of travel and telecommunications, and increasing global economic interdependence have resulted in considerably greater crossnational and crosslinguistic contact. Language planning, as an area of scientific inquiry, has also emerged and gained legitimacy from these developments. Language planning intersects with the phenomena of individual and societal bilingualism insofar as planning the status and functions of different languages in society often involves provision of greater (or fewer) opportunities for the acquisition of different languages in the educational sphere. Similarly, educational provision designed to address equity issues (e.g. bilingual education of minority students in the United States) will simultaneously affect the status and functions of different languages and consequently the impact on language planning decisions. While demographic changes affect all spheres of society, the most immediate impact is experienced in education. The extent of population shift and its implications for education are readily apparent in recent statistics from countries of the Organization for Economic Co-operation and Development (O.E.C.D.). For example, O.E.C.D.'s Centre for Educational Research and Innovation (CERI) (1987) reported that in seven European countries surveyed, the average foreign enrolment ranged from 4.8 percent in the Netherlands to 18 percent in Switzerland and 38 percent in Luxembourg. The proportions in Belgium (13.5 percent), France (10.1 percent), Germany (11.9 percent), and Sweden (8.7 percent) were also substantial. These figures can be compared to the virtually 0 percent of foreign enrolments that were recorded in the late 1960s. With large population bases, these proportions represent a very large number of children. The profound changes that are taking place in the schools of these countries can be illustrated in the fact that between 1974/1975 and 1981/1982, the number of students in German schools (excluding the preprimary level) fell by 700,000, the net result of a loss of 1.1 million German children and a gain of 400,000 foreign children. The rate of demographic change has also escalated in the United States. In California, for example, the proportion of limited-English-proficient students rose by 14 percent between 1988 and 1989 (to 740,000 from 652,439 students). In the five years prior to that the annual average increase was 7 percent. Faced with massive demographic change, policy makers in many coun224
Language, cognition, and education tries have attempted to devise strategies for promoting educational equity for minority groups while minimizing intercultural tensions. One common, but controversial, strategy has been the teaching of minority children's mother tongue or its use as a medium of instruction within the public school system. For example, in countries such as the United States and Sweden various types of bilingual programs have been implemented involving the use of the minority child's mother tongue as a medium of instruction in order to bridge the gap between home and school languages. Most frequently these are "transitional programs" that use the minority language only as a temporary instructional medium until the student is judged to have sufficient competence in the majority language to follow instruction in that language. Such programs usually do not aspire to promote bilingualism. In the United States, widespread implementation of transitional bilingual programs resulted from the 1974 Supreme Court decision in the Lau vs. Nichols case that required school systems to take some affirmative action to help students from language-minority backgrounds acquire proficiency in the language of school instruction. At the time the U.S. Office of Civil Rights interpreted this decision as virtually requiring transitional bilingual programs. A major problem in the usual rationale for transitional bilingual programs is the fact that mandating one programmatic response for all languageminority students ignores the enormous diversity among students, both within and across minority groups. For example, not all minority students perform poorly under conditions of a home-school language switch; the success of some (but by no means all) groups of Asian students has been well documented and publicized in the United States and Canada. Black students in the United States perform as poorly as Hispanic students despite the fact that there is usually not a linguistic mismatch between home and school, although there often is a mismatch in cultural and dialectal patterns. Thus, linguistic mismatch in itself cannot account for the failure of minority students (Cummins, 1989). Even within minority groups there is considerable variation in academic achievement related to socioeconomic and other contextual factors. For example, Hispanic students who arrive in the United States after several years schooling in Latin America appear to have better academic prospects than Hispanic students born in the United States (see Cummins, this volume). In short, given the heterogeneity of the population of "minority" students, it is not surprising that considerable variation is observed in patterns of achievement. Occasionally, "enrichment" bilingual programs are implemented for minority students that do include full bilingualism and biliteracy as goals. In the United States context, these programs are termed "developmental" or "two-way bilingual" programs. They usually involve both minority and 225
Ellen Bialystok and Jim Cummins majority students in the same class with each group acting as linguistic models for the other. Instruction is modeled after French immersion programs in Canada in which most of the instruction in the early grades is through the medium of the minority language with a gradual shift to about half the time through each language by the end of elementary school. Evaluation data from these programs are very positive in that students from both majority and minority groups develop bilingual and biliteracy skills with no apparent loss with respect to progress in the majority language (English). However, enrichment bilingual programs for minority students are still relatively scarce; "enrichment" goals are considerably more likely to be found in bilingual programs for children from the majority group (e.g. French immersion programs in Canada). Parents of children from the majority group have increasingly pressured school systems for better language-learning opportunities for their children. Progress toward European integration and mobility of labor has made biand multilingualism an important economic asset in the European job market. In Canada, more than 250,000 students from mainly English backgrounds are enroled in French immersion programs as a means of developing bilingualism. The initial results of these programs tended to be highly positive, with students developing close to native-like receptive skills in French and relatively fluent (but by no means native-like) expressive skills (Lambert & Tucker, 1972). However, as students entering immersion programs have gradually diversified with respect to socioeconomic and linguistic background, results have become more mixed; for example, not all students succeed equally well and a small but significant proportion drop out of the program during the early grades. A number of investigators (e.g. Cummins, 1987) have suggested that pedagogy in immersion programs requires greater individualization to address the learning needs of an increasingly diverse student population. It is clear that the development of bilingualism is characterized by an enormous array of acquisition patterns both within the school context and outside the school. The policy situation is complicated further by the fact that social and educational policies with respect to language and bilingualism are by no means neutral with respect to intergroup power relations. The degree to which particular languages are used in the institutions of society reflects the status and power of particular groups, and societal rewards (e.g. jobs) result from the degree to which ethnic languages are institutionalized. For example, the adoption of Spanish and other minority languages as languages of instruction within the United States after the Lau vs. Nichols Supreme Court decision in 1974 gave rise to increased job opportunities within the educational sphere for Hispanic and other minority groups as teachers, administrators, and instructional assistants. An increase of status and power by a minority group is frequently resisted 226
Language, cognition, and education by the majority or dominant group and thus it is not surprising that policies relating to bilingualism in education have been highly controversial in many countries. The most extreme example has been in the United States, where the English-only movement has campaigned to make English the official language of the nation and has strongly resisted attempts to expand bilingual-education programs for minority students. Their argument is that such institutionalization of minority languages threatens the unity of the nation and is counterproductive in view of minority students' need to acquire English. The assumption is that dilution of exposure to English instruction through bilingual education will impede children's acquisition of English. In Canada, the teaching of minority or "heritage" languages has been similarly controversial in cities like Toronto, where more than half the school population come from non-English-speaking backgrounds. In view of the volatile context of many political and educational situations relating to bilingualism, it is not surprising that the research data regarding bilingual children's cognitive and academic development have frequently been distorted and misinterpreted. For example, there is no evidence that bilingual instruction impedes children's acquisition of literacy in the majority language (see Cummins, this volume; Hakuta, 1986). In fact, there tends to be a strong relationship between literacy levels in first and second languages. Similarly, the data reviewed in this volume suggest that bilingual children experience more varied possibilities for language processing and that, under some conditions, this may result in cognitive and academic advantages. These data tend to be either ignored or denied by researchers and policy makers who fear that bilingual instruction will promote societal division. Each of these types of bilingual-education programs differs importantly in the kinds of initial abilities it assumes children to have with regard to proficiency in the two languages and the kinds of skills that it attempts to develop in children. The programs, that is, are not alternative solutions to the same problem but specific solutions to different problems. Children whose initial level of language proficiency in a single language is highly developed may be in a position to benefit from enrichment bilingual programs while other children have the need for various types of transitional and other bilingual education programs. The social and educational context of the child in the school, that is, must be considered in the evaluation of specific bilingual programs. Fundamental to this consideration is the cognitive and linguistic functioning of the child. Cognition and the bilingual child
The usual methodology for research into cognitive and linguistic aspects of functioning in bilingual children is cross-sectional research (although 227
Ellen Bialystok and Jim Cummins a few longitudinal studies mercifully appear, for example Hakuta and Diaz, 1985). The logical form of such research is a comparison between two groups in which evidence for differences between the groups is attributed to differences in bilingualism. This type of research is obviously limited in its power, as the problem of individual differences in functioning on these incredibly complex tasks is rarely addressed and, in any case, is only poorly understood at best. But the evidence needed to address the fundamental questions in cognitive and linguistic proficiency of bilingual children is subtle and elusive. The rough and somewhat contaminated data that are yielded by between group comparisons can only provide hints of importantfindingsand become the basis for descriptions that are nebulous approximations to cognitive models. The best assurance of reliable findings is the convergence of results across studies. Remarkably, the convergence of many studies, in particular those reported in this volume, is very high for a few general principles of comparison. There are two relevant forms of comparison in the evaluation of bilingual children. The first form is a judgment of the similarity between the way in which bilingual children perform in certain situations or solve certain problems and the performance of monolingual peers for the same problems. The second form of comparison is an evaluation of the way in which bilingual children approach a particular task in their first or second language. Cumulatively, these two comparisons set constraints on the description of processing by bilingual children by determining those aspects that are unique to speakers of two languages and those that appear not to be influenced by linguistic competence. Some types of achievement, such as second-language acquisition and translation, are exclusively bilingual activities, but even here, relevant comparisons between performances in the first and second language point to the aspects of these achievements that are influenced by the child's general linguistic competencies and those that are not. Most important is the examination of development. By studying the ways in which bilingual children proceed in their cognitive and linguistic growth and by comparing that development to both the development of monolingual children and the development of these competencies within the two languages of the bilingual child, it is possible to extract general principles of cognitive development. Issues such as interdependencies between language and thought, as well as interrelations among various aspects of language proficiency are essentially inaccessible to study if one considers only the progress of monolingual children. The central finding in all the studies reported in this volume is the need for a more differentiated conception of the competencies of bilingual children. Such a conclusion leads one inextricably to the debate concerning the modularity of mental processes. To what extent is processing influenced by central cognitive operations as opposed to isolated structures specific 228
Language, cognition, and education to a domain? It is on these problems that our results of the linguistic processes of bilingual children can be brought to bear. The modularity view of cognitive organization assigns processing responsibility for different functions to different, non-interacting "modules." In strong versions of the theory, such as that of Fodor (1983), only a limited set of high-level processes are carried out by the central processor, and these events are largely inscrutable under empirical investigation. Most cognitive processing is carried out in the modules, and these systems are autonomous. Structuralist accounts of cognition, in contrast, place greater priority on central processes that evolve from and influence cognition across domains (Case, 1985; Piaget, 1970). The evidence accrued from the studies of bilingual children gives clear support to neither general position. Considerfirstthe predictions that follow from a completely modular organization for cognition. In Fodor's (1983) conceptualization, modules are defined by input systems that are self-contained units of fast, encapsulated processing. Language, then, constitutes a modular input system: listeners have no control over the (initial stages of) processing of the linguistic input as it is channeled through this dedicated system. But what about processing of the input of a second language? If it is part of the same language module as the one designed to process data in the first language, then the shared resources in the early stages of processing should produce considerable convergence between the two languages in the ways in which they are understood, used, and the kinds of cognitive problems to which they can be applied. Conversely, if the second language is processed through its own unique module, then little interference or overlap would be expected. Each language, on this view, may become specialized for the solution to quite different problems. The evidence supports neither of these predictions. There are indeed interactions between first- and second-language skills (Cummins, this volume) but at the same time there is evidence for specialization of the languages (Malakoff & Hakuta, this volume). Other problems, notably school tasks, seem impervious to the language in which the problem is presented and are solved equally by both languages (Snow et al., this volume; Johnson, this volume), although sometimes this too becomes an important feature as solutions to problems are more easily carried out in the language in which those problems were presented (Malakoff, 1988). Modular input systems do not fit readily with these findings. A central-processing theory of cognition is somewhat more compatible with the performance of bilingual children, but here too, there is not clear support for the model. Although the findings of Snow et al. (this volume) and Johnson (this volume) suggest the centrality of cognitive operations in certain logical and manipulative linguistic tasks, research by Malakoff (1988) shows language-specific effects on logical reasoning. In particular, 229
Ellen Bialystok and Jim Cummins the language of instruction was more likely to lead to faster performance in logical-reasoning tasks, even if it was the child's weaker language. In this case, central processors were strongly influenced by language-specific factors. Johnson (this volume) suggests that this separation of cognitive and linguistic operations may be a function of certain logical operations becoming specialized in one language because of repeated experience or exposure to those operations in that language. Johnson's analysis of the logical demands associated with different types of task situations adds precision to a central-processing description in that it allows the same finite domain of operations to function differently across a variety of problems without recourse to segmentation of those operations. Thus differentiation is achieved with a common set of cognitive operators. Another solution to the problem of characterizing the relation between linguistic and cognitive operations is to reconceptualize modules along different lines. Jackendoff (1987) provides a detailed review of Fodor's account of modularity and assesses it against his own description of cognitive functioning in a computational mind (the same metaphor for mind as that used by Fodor). Jackendoff concludes that processing modules form an essential structure of mind, but that the identification of modules as input systems is incorrect. He proposes a system of modules that is in fact more finely differentiated than that described by Fodor, but, more importantly, determined not by input systems but by representational systems. These representation-based modules cut across functional domains and enclose instead operations according to their level of representation. Hence, the processing of such information as "shape" is shared across the linguistic function of reading and the cognitive function of object recognition or pattern identification. Aside from the economy of representation and processing, there are several other advantages to modules defined in these terms. The system, for example, is able to engage in bidirectional processing, a function precluded with input-based modules, and mechanisms for transmitting information from the modules to the central processors can be specified in greater detail. Jackendoff takes the reader through a detailed account of the operation of the computational system he describes. The evidence from bilingual children suggests that a complex description such as that discussed by Jackendoff within a modular approach or Johnson within a constructivist approach is necessary to describe the organization of linguistic and cognitive processes. Descriptions of children's performance on linguistic tasks minimally demand that the task be characterized as being either conversational, contextualized, informal on the one hand, or academic, decontextualized, formal on the other. Different processes are involved in the performance of each, and each bears a different relation to central cognitive processors. The primary implication of these highly differentiated descriptions of 230
Language, cognition, and education cognition is that there are cognitive and linguistic operations that can be discretely identified and that have the potential for functioning autonomously. But much more importantly, they have the potential for developing autonomously. Hence there is no reason to expect a uniform pattern of development across all operations for bilingual children, or across all bilingual children. Bilingual children may differ from monolingual children in some but not all of the constituents of thought. Moreover, some bilingual children may differ from other bilingual children in these patterns of development. Some of the social, contextual, and educational factors that conspire to produce these differences have been discussed in some of the chapters of this volume. If we dispense with universality, is there any general truth that remains to describe the language or cognitive development of bilingual children? The most prudent course at this time appears to be to continue to seek out the relevant structures that differentiate performance in children and follow through to the development of specialized educational programs that are sensitive to such differences. Paradoxically, it will most likely be the study of the diversity that exists among bilingual children in their social, cognitive, and linguistic proficiency that will lead to the most universal understanding of the ways in which language and cognitive processing is carried out by bilingual children. The challenge for researchers who investigate the phenomenon of bilingualism is to weave the increasing amount of data that has emerged into coherent models of the cognitive processes that characterize different types of bilinguals and multilinguals. We believe that the chapters in this volume have taken some important steps in this direction.
References Case, R. 1985. Intellectual development: birth to adulthood. New York: Academic
Press. Centre for Educational Research and Innovation 1987. Immigrants' children at school. Paris: O.E.C.D. Cummins, J. 1987. Immersion programs: current issues and future directions. In L. L. Stewin & S. J. McCann (eds.), Contemporary educational issues: the
Canadian mosaic. Toronto: Copp Clark. 1989. Empowering minority students. Sacramento: California Association for Bilingual Education. Fodor, J. 1983. The modularity of mind. Cambridge, MA: M.I.T. Press. Hakuta, K. 1986. Mirror of language: the debate on bilingualism. New York: Basic
Books. Hakuta, K. & R. M. Diaz 1985. The relationship between degree of bilingualism and cognitive ability: a critical discussion and some new longitudinal data. 231
Ellen Bialystok and Jim Cummins In K. E. Nelson (ed.), Children's language, vol. V. Hillsdale, NJ: Erlbaum. Jackendoff, R. 1987. Consciousness and the computational mind. Cambridge, MA: M.I.T. Press. Lambert, W. E. & G. R. Tucker 1972. Bilingual education of children: the St. Lambert experiment. Rowley, MA: Newbury House. Malakoff, M. 1988. The effect of language of instruction on reasoning in bilingual children. Applied Psycholinguistics, 9, 17-38. Piaget, J. 1970. Structuralism. New York: Harper and Row.
232
Index
adult data, 4, 20, 38, 39, 42, 59, 60, 72, 83-84,91,147,149,187 Albert, 17 Aliotti, 170 Allen, 51,64 Alpert, 146 Ammon, M. S.,51,64 Ammon, P., 51, 64,196, 200, 202, 203, 205, 209,212,217 analysis of linguistic knowledge: see processing components Anderson, 35,193 Asuncio-Kibdem, 146 attention, 84,116,119-120,127,185,196, 197, 203 babbling, 29, 30, 32, 33,185 Bailey, 41 Bain, 210, 212 Balkan, 209 Baral, 86 Barmejo, 38, 39 Barton, 32 Bates, 5 Beebe,50,63 Belanger, 82 Benson, 194,197,198,200 Ben-Zeev, 20,21, 99,132,167,173,174, 175,204, 206, 207, 208, 209, 210,213 Bereiter, 71 Bergman, 146 Berk,188 Best, 28,31 Bialystok, 8,14,20,50,58, 86,91,99,129, 132,134,138,148,167,169,173,175, 195,199, 203, 205, 206,213, 223 Biber,71,85 bilingual advantage, 17,20-22, 92,98, 99^100,141,146,148,161,163,167, 168,177,180-183,184,186,189 academic achievement, 73, 74, 75,109, 110 cognitive achievement, 62,126-128,130,
233
131-135,136,138-139,169,170-172, 177-178,195, 204-217, 223-224, 227-231 difficulties, 8,19, 63,141,175, 213 model for, 187-191 bilingual proficiency, 143,146,148,149 bilingual programs, 72, 74, 75, 76, 80, 81, 92, 152,156,161,170,181 bilingualism, additive, 175-177,179,181,187,188,189, 204 balanced, 35, 37,141,169,176 consecutive: see sequential private speech, 180-184,185,186,188,190 sequential, 10,17, 36-37, 38 simultaneous, 10,17,18,19, 34-36 subtractive, 152,156,161,171,175-177, 179, 204 Bond, 38, 39 Bowey, 115,173 Bransford, 196, 202 Brown, A., 129,196,202 Brown, R., 3,137 Brumfit, 150 Burner, 71 Bull, 76 Burnaby, 49 Burnham, 31 Burt, 3 California State Department of Education, 75,81 Campbell, 64 Campione, 196,202 Canale,82 Cancino, 8, 71, 72, 91,104, 223, 229 Carbone,32,42 Carlisle, 76 Carramazza, 32, 42 Carroll, 150 Case, 5,119,196, 200, 202,205, 206, 229 Cassirer, 197,198 Catford, 144,145,150
Index Cazden, 147 Celce-Murcia, 35 Chan, 147 Chau, 147 Chomsky, 3,11,26,57 Christopherson, 173 Clark, E., 114 Clark, H . H . , 142 Clifton, 15 Clyne,55,65 code-switching, 17, 22,146-147,151,167, 168,180,182 r 184,189 cognition, 49, 85, 86,115,141, 204-215, 227-231 cognitive flexibility, 168,189,203, 204 cognitive skills, 49,53,58,59,60,61,78,81, 83, 84, 85,143,148,167,168, 170,187 Collier, 72 Collins, 117 communication strategies, 52,53,55,56,57, 60,71,137-138,147,150 communicational adequacy, 95,97,99,100, 101,102 connectionism, 115 consciousness, 146 content, 90,96 contextualized language use, 71,72,77,81, 84,85,223 control of linguistic processing: see processing components conversation, 71, 72,75; see also communication strategies conversational features, 84, 95, 97, 100,101, 102,105 crosslinguistic influence, 16,17, 19, 28, 38, 75,76,79,82,84,85,86 Cumming, 83, 84, 85 Cummins, 8, 20, 21, 51,64, 71, 72, 77, 78, 80, 83, 84, 86,98, 99,173, 176,178, 179, 194, 204, 206,207, 223, 225, 226, 227, 229 Davidson, 81,98,104,194,199, 200 DeAvila, 170,204,209,215 decentration, 204,209,210 Dechert, 16 decontextualized language use, 70, 71,72, 73, 77, 81, 84, 85, 90,96, 98, 107, 108, 109,195,197,198,199,201,223 definitions, 90,92, 213, 214 categories, 94 formal, 90,91-92, 94-95, 96-100, 101, 103-109,214 informal, 95, 96,101,103-104, 214 task, 103-109 De Temple, 8, 71, 72, 213, 223, 229 dialectical-constructivist perspective,
234
196-200, 200-202, 202-204 Diaz, 8,21, 75, 167,168,169,170,171, 172, 173,177,178, 179,180, 182, 183, 186, 188, 195, 196, 204, 212, 223, 228 disequilibration, 200, 201 Dittmar, 58 Dolbear, 105 Donaldson, 71 Dorman, 41 Doughty, 54 Dulay, 3 Dumas, 65, 142, 163 Duncan, 170, 204, 209, 215 Eddey, 38, 39 education, 2, 222,224-227 foreign language classrooms, 64, 92, 93, 99,109,110,143 immersion classrooms, 1,2, 51-52, 65, 72, 73,81,226 schooling, 73, 79, 83, 91, 97-99 Edwards, 173 Eilers, 30, 31 Elliot, 33 Ellis, 51, 60, 64 encyclopedic knowledge, 11,12-13,17, 20 Ervin-Tripp, 62 Fabian, 198, 203, 205, 212, 214, 217 Faerch, 58 Falter, 81 Fantini, 36 Farwell, 29, 32 Feldman, 206, 209 Ferdman, 170 Ferguson, 29, 32, 55, 65,126 Ferrara, 196, 202 Ferreira, 15 Fillmore, 124, 125, 126 first language (LI), 16,17, 18, 20, 53, 55, 56, 58, 61, 70, 72, 73, 74, 76, 77, 78, 80, 81,84,85,86,109,151,170 Fischer, 149 Fishman, 171 Flesch, 150,151 Fodor,10, 11, 12,13, 20, 115, 150, 229 Fokes, 38, 39 form, 54, 90, 92, 96,132-133, 173, 174, 195 Forster, 14,15 Foster, 13 Fourcin, 26, 30, 32, 33, 38, 40, 41 Fradd, 76 Franz, 91 Frauenglass, 183, 186 Frazier, 14,15,18,22 Frennett, 82 Frohlich,4,50,59
Index Fuchs, 150,151 Gaies, 54 Galambos, 99,148,173,174,175,195,196 Galvan, 64 Gardner, 49,59 Gass, 16,154 Gavin,30,31 Genesee, 77,80,147 Geva,81,85,195 Gildea, 91 Gleitman, H., 150 Gleitman, L., 150 global processes, 100,101 Goldman, S., 213 Goldman, S. R., 76 Gonzalez, 72,75,76,91,104 Goodman, 193,196,200,202,203,205,207, 208,209,212,217 Goudallier, 27,39 Gould, 144 Gowan, 170 grammatical competence, 11-12,13,14,15, 17,22,55,57, 59, 77,78,132,150,188, 206,207 Green,77,84,86,98,195 Grosjean, 142,147 Guerra, 83 Gulutsan, 213 Gumperz, 146 Hagtvet, 91 Hakuta, 6,8,75,99,127,141,144,148,152, 167,168,169,170,171,172,174,175, 176,177,180,182,195,196, 200,204, 206,209,223,227,228,229 Halle, 26 Handscombe, 77,84,86,98,195 Harley,51,64,142 Harris, 142,143,144 Hatch, 54 Herriman, 148 Ho, 82 Hoefnagel-Hohle, 4,62,72 Hoosain, 147 Hoover, 79,80 Huserl, 193,198 Hyltenstam, 16 Ianco-Worral, 99,148,173,195,206 Ijaz, 205,215 Imedadze, 34 immigrant children, 1,7, 51,60,72,73,77, 86,171,175 information processing, 115-116,119 Inhelder, 193,197 instruction, 12,50,51,59,64,65,66,74,80,
235
82, 83,109,135-136,177 interlanguage, 65,150 Iwasaki, 79 Jackendoff, 11,12,116,120 Jakobson, 33 Johnson, 2,4,8,21,92,97,109,110,194, 195,197,198,200,203,205,212,214, 215,217,223 Kade, 151 Karmiloff-Smith, 20,118 Kasper, 58 Katz, 150 Kellerman, 16 Kemp, 81 Kessler, 21 King, 83 Kintsch, 151 Klein, 58 Kline, 81,98,195,199,200 Klingler, 8,21,212,223 Koenigsknecht, 39 Kohler, 26 Konefal, 38,39 Krashen,54,62 Krauss, 111 Kuhl, 30 Labov, 136 Lado,3 Lagarretta, 64 Lambert, W. C , 49,59,64,141,142,148, 167,168,169,170,172,176,180,184, 201, 208, 226 Lange,197 language acquisition, 12,13,14,16,22,25, 28,54-56,70,72,74,79,109-110,143, 210 language acquisition device, 57,58,59 language domain, 121-124 language learners, 61-66 language learning cognitive, 49,56-59,60,62,63,64,70, 135-137 instruction, 50 linguistic, 54-56,60 model, 51-53 social aspects, 49, 50,51-52,53-54,56,59. 60,63,64,70 language proficiency, 8,70-86,110,116,118 120,121,142,153,171,176 attribute-based, 70,78,84,85,86 input-based, 70,78,84,85,86 Lapkin, 64 Lasnik, 12 Lederer, 143,144
Index Leopold, 2,21,34,99,142,173,175,188, 195 Licdtke, 170 Lightbown, 51 Linde, 72,73,74 Lindholm, 146,147 Linell,33 linguistic interdependence (between LI and L2), 72-83,195 linguistic transfer, 74,76,109,195 Lisker, 41 literacy domain, 128-130 literacy skills, 78,80,91,110,114,130,227 Litowitz, 91 Ljudskano, 143 Locke,29,32,33 Lofgren, 72,73,74 Long, 54,62 Lopes, 83 Lowe, 186 Lundberg, 114 Lundsay, 117 Luria, 168,193,194,1%, 197,198,210,212 Mace-Matluck, 98 MacKain, 31 Macnamara, 18,126 MacWhinney, 5 Malakoff, 8,82,92,97,109,144,195,200, 229 Mandler, 117 Mann, 32 Markowitz, 91 Martin-Jones, 175 McClelland, 115 McLaughlin,51,64,82,83 meaning, 91,96,132-133,149,150,173,174, 195 comprehension, 73,144-146,147,149 theories of meaning, 144-146,149,195 mediated learning, 1%, 197,200,203 Mehler, 18,19 Meisel, 17,58 Menn,32,33,34 mental distancing, 201,202,203 mental representation, 13,16,20,116, 117-118,119^120,141,193,195,201 Menyuk, 32,33,34 metalinguistic ability, 6,13,43,92,98,104,114,115, 117-118,121,132,149,157,167, 172-175,184 awareness, 20,21,22,113,147,148-149, 151,187 definition, 90,113,130,147,148,194 domain, 130-134 metamode, 20-22
236
Miller, 30,91 Minninni, 144,145,150,151 misleading situation, 203,204,205-207, 207-210,210-212 Mitterer, 129 mobility figurative, 207,208-209 operative, 207,208,209 modularity, 10,11-16,17,20,22,229 monolingual children, 21,25,28,29,32,37, 39,40,41,42,44,76,92,93,109,114, 132,141,142,148,169,174,189, 205, 206,207,209,210,211,222,228 Moore, 31 Morris, 197 Mulcahy, 173 multilingualism, 100 Naiman, 4,50,59,142,163 Nakajima, 72,77,78,84,86,98,98,195 Nelson, K., 61,193,209 Nelson, L. D., 170 Newport, 2,4 Nida, 142,143,144 Nolan, 26 Norman, 117 Norris, 18,19 Obler, 16,18 Ohde, 38 Olson, 71 oral domain, 84,114,124-128,141 Padilla, 146,147,170,172,182 Paivio, 142 paraphrase, 149,150-152 Pascual-Leone, 193,194,195,1%, 197,198, 200,202,203,204,205,207,208,209, 212,214,215,216,217 Pavlovich, 35 Peal, 148,167,168,169,170,172,176,180, 184,201,204,208,226 Pedraza, 146 Peirce, 197 Perecman, 17,22 Pergnier, 143,145 Peters, 57,61 phonology acquisition, 29-33,33-34,36,37,38,44, 58 bilingual, 25,27,28,29,31,32,33-34, 34-41,43,44 monolingual, 25,28,32,33-34,37,38,39, 40,41,42,43,44 perceptual studies, 41-44 representation, 26,27
Index Piaget, 133,185,193,194,197,199,200,201, 203,205,210,216,229 Pica, 54 Pipp,149 Poplack,146,147 Powell, 119 processing developmental, 10 knowledge, 10,11-13 on-line, 11-16 processing components analysis of linguistic knowledge, 90,91-92, 98,116,117-119,122-138,173,199 control of linguistic processing, 17,20, 116,119-120,122-138,167,173,174, 175,203,205 Pylyshyn, 115 Quillian, 117 Quinn, 21 Ramirez, 74 Raupach, 16 reading, 5,75,76,78, 79,80,81,128-130 Rehbein, 82,83 Reyes, 213 Ricard, 105 Rips, 117 Rivera, 144 Rodriguez-Landsberg, 144 Romaine, 175 Ronjat, 35 Rosen, 117,193 Rossier, 50 Rumelhart, 115 Ryan, 81,85,86,91,194,199,203,205 Sajavaara, 18 Scarcella, 62 Scardamalia, 71 Schley, 8, 71, 72,105, 213, 223,229 Schneider, 116 Schumann, 4,49 second language(L2),10,16,17,18,21,50, 51, 55,56, 58,60, 61,70, 72, 73, 74, 76, 77,78, 80, 81, 83, 84, 85, 86,98,99,109, 126,127,136,152,170,171 second-language acquisition, 3,7,223,229 Segul, 18,19 selective attention: see attention Seleskovitch, 142,143,144,145 Selgado, 83 Seliger, 16 Selinker, 16,58,65 sensorimotor period, 200,209 Shannon,144,163 Sharwood-Smith, 7,14,16,17,223
237
Shen,206,209 Sherwood,143,144 Shiffrin, 116 Shoben,117 Shriberg,91,104 Siegler, 5 Sigel,194,201,205 signals, 198,211 signs, 173,185,186,187,189,197 Simon, 26,30,32,38,40,41 Skutnabb-Kangas, 72,73,175 Slobin, 99 Steiner, 150 Smith, C. L., 173 Smith, E . E . , 117 Smith, N . V . , 32 Snow, 4 , 8 , 6 2 , 6 4 , 7 1 , 7 2 , 8 1 , 8 6 , 9 1 , 9 4 , 95,97,98,101,104,105,126,195,199, 200,213,223,229 spatial abilities, 170,172 speech, 12,17,18,19,25,26,28,32,37,40, 41,49,55,58,65,136,168,172, 180-183,185,186,188,190,196-197 Sridhar,K. K.,147 Sridhar,S. N.,147 Stern, 4,50,59 Sternberg,5,119,196,202 Stone,169 Streeter, 31 Strong, 50 Stroop, 120 structures executive structures, 175,188,196,197, 198,201, 202,203, 204, 208,216 infralogical, 193,194,195,196,197-206, 213,216 logological, 193,194,195-206,216 mereological, 193,194 Subelman, 196,200,202,203,205,209,212, 217 Summerfield, 41 Swain, 49,51,64,65,77, 84,86,98,142, 146,163,194 symbols, 56,118,173,194,197,198,206, 211
Tager-Flusberg, 173 Tapia-Uribe, 91 target language, 29,49,50,52,53,54,57, 59,60,62,63,64, 65,66,137,141,143, 145,146,149,150,157,161 Tarone,54,136,138,137 tests Embedded Figures, 209 Gates-McGinitie, 77,79 Grammatically Judgments, 83,132,173, 174,205-206
Index tests (cont.) Interactive Reading Assessment System, 80 Peabody Picture Vocabulary Tests, 171, 212,213 Ravens Standard Progressive Matrices, 170,171,172,177,178,180,20&-209 Stanford Reading Tests, 75 WISC, 93 Woodcock Language Proficiency Battery, 152,155,215 Todesco, 4,50,59,84,86 Torrance, 170 Toukomaa, 72,73 Tran, 8,86,98,195 translation and bilinguals, 93,143-144,203,204 natural, 143-144 and paraphrase, 150-152 skills, 14&-149 study,152-161 theory, 142-147,149-150 Tucker, 64,226 Tunmer, 148
Velasco, 98 verbal-mediation hypothesis, 180-182 Vinay,142 Vygotsky,113,115,138,148,168,169,172, 173,175,184,185,187,190,193,194, 195,196,197,198, 200, 203, 204, 205, 210
Vago, 16 Van Dijk, 151 Van Kleek, 173 vanLier,51,60,64 Varnhagen,213,216 Varonis, 54
Yu, 210, 212 Yeni-Konshian,32,42
238
Wu, 170 Watson, I., 7,28, 31, 39, 41, 45, 223 Watson, R., 91 Weathersby, 182 Webster, 99 Weinreich, 16,35,146 Werch, 146 Wertsch, 169,196,197 White, 12 Williams, 41,43,44 Wilson, 30, 31 Wilss, 142,143,150,151 WongFillmore, 7, 8, 49, 50, 51, 58, 59, 64, 223 writing, 5, 71, 76,77, 78, 79, 82, 83, 84, 85, 118,130,134,141,149,152
Zentella, 145 Zurif, 32,42 Zlatin, 38