Language and Cognition in Bilinguals and Multilinguals_ An Introduction

LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS



Language and Cognition in Bilinguals and Multilinguals An Introduction Annette M. B. de Groot

Published in 2011 by Psychology Press 270 Madison Avenue New York, NY 10016 www.psypress.com Published in Great Britain by Psychology Press 27 Church Road Hove, East Sussex BN3 2FA This edition published in the Taylor & Francis e-Library, 2011. To purchase your own copy of this or any of Taylor & Francis or Routledge’s collection of thousands of eBooks please go to www.eBookstore.tandf.co.uk Psychology Press is an imprint of the Taylor & Francis Group, an Informa business Copyright © 2011 by Taylor & Francis Cover design by Hybert Design All rights reserved. No part of this book may be reprinted or reproduced or utilized in any form or by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying and recording, or in any information storage or retrieval system, without permission in writing from the publishers. Library of Congress Cataloging-in-Publication Data A catalog record for this book is available from the Library of Congress ISBN 0-203-84122-0 Master e-book ISBN ISBN: 978–1–84872–901–8 ( Print Edition)

To the memory of my parents Johan de Groot (1917–2008) Cher de Groot-Fransen (1921–2001) To the memory of my twin sister Jeannette de Groot (1951–1994) For my son Jan Kraak



Contents About the author ix Phonological short- and long-term Preface and acknowledgements xi memory and foreign vocabulary learning 1 Introduction and preview 1 116 Psycholinguistics and the study of The role of form similarity between bilingualism 2 translation pairs in foreign 119 Defining and classifying bilinguals 4 vocabulary learning and teaching 123 Central themes, concepts, assumptions, 128 and questions 5 From focusing on form to focusing 144 Some naming conventions 7 on meaning 151 Preview of the chapters 7 The revised hierarchical model, its precursors, and other models Vocabulary acquisition in context Summary 2 Early bilingualism and age of 11 4 Comprehension processes: Word 155 acquisition effects on (first and) 11 recognition and sentence processing 155 second language learning 13 Introduction and preview 157 Introduction and preview 18 Methods and tasks Methods and tasks The processing of interlexical 163 Early bilingualism in development 47 homographs and homophones Age of acquisition effects and the 78 Models of language-nonselective 177 critical period hypothesis lexical access Summary Parallel phonological activation in 183 two languages 199 3 Late foreign vocabulary learning and The processing of cognates 206 Sentence processing 216 lexical representation 83 Summary Introduction and preview 83 Methods and tasks 85 The keyword method: Successes and limitations 93 Easy and difficult words: Evidence 5 Word production and speech accents from paired-associate learning 106 Introduction and preview 221 Methods and tasks 221 The role of prior knowledge in 222 foreign vocabulary learning 113 vii

viii CONTENTS Models of monolingual and Control in simultaneous interpreting 314 bilingual speech production 224 Summary 336 The core of speech production models: Lexicalization 229 7 Cognitive consequences of 236 Picture naming bilingualism and multilingualism 339 Color naming with distracters: The 254 259 Introduction and preview 339 Stroop task Word translation 265 Methods and tasks 342 Comprehension and production: 268 Cross-language influences on third Making a connection 276 Speech accents in non-native language use 343 speakers Loss of a former language 347 Summary The interacting languages of bilinguals and multilinguals: Effects on the first language 361 Bilingualism and linguistic relativity 371 6 Language control 279 The cognitive effects of bilingualism 385 Introduction and preview 279 Summary 401 Methods and tasks 281 Language control as switching 8 Bilingualism and the brain 405 between language subsets 283 Introduction and preview 405 Activation and deactivation of the Methods and nomenclature 406 language subsystems: The Language areas in the brain and bilingual’s language modes 288 their function 418 A neurolinguistic theory of bilingual Language areas in the bilingual brain 427 control 294 Language control in bilinguals 435 The emergence of language subsets 296 Language control in a language- Glossary 447 References 461 nonselective activation framework 302 Author index 497 Subject index 506 Language control through reactive suppression 306

About the Author Annette de Groot is Professor of Experimental Psycholinguistics at the University of Amsterdam. She moved to Amsterdam in 1987 after complet- ing a master’s degree in General Linguistics and a doctorate in Psycholinguistics at the University of Nijmegen, The Netherlands, and following a Research Fellowship awarded to her by the Netherlands Organization for Scientific Research. In Amsterdam she first held the position of Associate Professor of Cognitive Psychology and, since 1995, her current position. Her work has concentrated on word recognition and the struc- ture of the mental lexicon, the psychology of reading, and bilingualism and multilingualism. Her present research focuses on bilingualism, with a special interest in bilingual word process- ing, foreign language vocabulary learning, and simultaneous interpreting. She has been a mem- ber of the Editorial Boards of Memory and Cognition, and Psychonomic Bulletin & Review. Currently she is a member of the Editorial Board of Bilingualism: Language and Cognition, and of the Advisory Board of Interpreting. With Judith Kroll she co-edited Tutorials in Bilingualism: Psycholinguistic Perspectives and the Handbook of Bilingualism: Psycholinguistic Approaches. ix



Preface and Acknowledgements For about four decades following its emergence appeared since around 1995. But among those around 1950, psycholinguistics—the field of books, one is clearly missing: a comprehensive text science that examines the mental processes and that covers many of the topical sub-areas within knowledge structures involved in acquiring, the psycholinguistic study of bilingualism; that understanding, and producing language—had a integrates these sub-areas into a coherent whole; strong monolingual orientation. In a prototypical that is written at an introductory level; and that experiment native speakers of a language were reflects the fact that in the field, in addition to the asked to perform some task in that language more traditional behavioral methods, brain while it was implicitly assumed they did not speak research is becoming increasingly popular. Most any other language, or the possibility they might of the books published so far concern edited do so was simply taken for granted. Although this volumes with a relatively restricted scope (e.g., monolingual bias still characterizes mainstream foreign language vocabulary acquisition, or sen- psycholinguistics, from about 1990 a clear shift tence processing in bilinguals) and some are of focus toward the bilingual and multilingual monographs dealing with just one topic (e.g., language user can be witnessed. Two insights have bilingualism in development, or translation arguably caused this change of focus: the aware- and simultaneous interpreting). Instead, the ness that a large part of mankind speaks more Handbook of Bilingualism: Psycholinguistic than one language and that a person’s multiple Approaches, edited by Judith Kroll and myself languages interact with one another even when and published by Oxford University Press, has a one of them is selected for current use. Therefore, broad scope, but it does not qualify as an intro- in order to obtain a faithful account of man’s duction. The contributions to that volume were linguistic abilities, the study of human language all reviews of one specific sub-area of study writ- processing could not remain stubbornly ten by experts and aiming at an audience of monolingual. graduate students and researchers in the field. This same description applies to an earlier volume And it did not. Journals exclusively dedicated we edited together, Tutorials in Bilingualism: Psy- to the study of bilingualism have been founded, cholinguistic Perspectives, published by Erlbaum, workshops on bilingualism are being organized, except that it contained fewer contributions. and quite a few books on bilingualism have xi

xii PREFACE AND ACKNOWLEDGEMENTS The topics covered by the present text are Boroditsky, Laura Bosch, Anna Bosman, Mirjam largely the same as those dealt with in the above Broersma, Ingrid Christoffels, Dorothee Chwilla, two volumes, but this time they are presented at Vivian Cook, Albert Costa, Annick De Houwer, an introductory level, aiming at a less advanced Ton Dijkstra, Giuli Dussias, Wouter Duyck, Jim readership: far less prior knowledge is assumed, Flege, Wendy Francis, Cheryl Frenck-Mestre, methods and tasks are explained in relation to Daniel Gile, Jonathan Grainger, David Green, theory, and more illustration materials and a François Grosjean, Marianne Gullberg, Lynne glossary are included. Furthermore, I have Hansen, Jan Hulstijn, Sonja Kotz, Wido La Heij, worked with the same “model of the reader” in Batia Laufer, Renata Meuter, Barbara Moser- mind all through, trying to use the same balance Mercer, Michel Paradis, Aneta Pavlenko, Antonio between depth and scope when presenting each Rodriguez-Fornells, Ardi Roelofs, Rosa Sánchez- individual theme. Finally, I have used a uniform Casas, Núria Sebastián-Gallés, Peter Starreveld, structure for the majority of the chapters: They Janet van Hell, Walter van Heuven, and Jyotsna start with an introduction and preview of the Vaid. topics to be dealt with, continue with a Methods and Task section and, next, a number of sections One person is noticeably missing in this list, that each treats a different theme and the associ- the reason being that the fleeting inclusion of her ated evidence, and they conclude with a summary name among those of my other valued colleagues of the main findings. Chapter 1 does not follow would not do justice to the significant impact she this format because it is only meant to set the has had, and still has, on my academic career and stage for the rest of the book, explaining what beyond. Of course the person I am hinting at is psycholinguistics is about and providing some Judy Kroll, who I first met in 1989 and since built standard terminology and definitions and a pre- up a professionally productive and privately view of the chapters to come. Chapter 8 lacks a enriching relationship with. Even though my life summary because it introduces relatively few new in academia steered away in a different direction topics but focuses exclusively on the brain basis after I decided to broaden my perspective by join- of bilingualism and multilingualism. ing the University of Amsterdam administration, our close collaboration in the past has never The reader will notice that this book’s chapters stopped to inspire me and this book would not are generally quite long. The fact that little prior have been written without it. Thank you Judy for knowledge is assumed is partly responsible for all that! this. Further reasons are that I wanted to provide some history as well—relating the bilingual I am also indebted to a number of people I (and multilingual) studies to be presented to the first met in the more remote past. They mentored monolingual psycholinguistic studies they derive me during the initial stages of my academic career from—and be up to date at the same time. and some of them I became and stayed friends Especially meeting this last goal posed quite a with to this very day: Jan Elshout, Ken Forster, challenge because new articles are appearing at an Gerard Kempen, Remmert Kraak, Pim Levelt, accelerating pace. and Ar Thomassen. Many people have supported me, in many The following students have contributed very different ways, since I first started to work on tangibly to this book and I am deeply grateful to this book about four years ago and I am grateful them. Nick Naber always knew where I was get- to all of them. First of all there are my colleagues ting at when I gave him some general instructions in the field who contributed importantly by regarding the illustrations I wanted to include in providing information, sharing their ideas, and the book and then designed and produced all nourishing my views on bilingualism, through of them on the basis of these fuzzy indications. their published work and in personal contacts: Geoffrey Cramm embellished the figures of the Jubin Abutalebi, Panos Athanasopoulos, Teresa brain as designed by Nick. Arien op ’t Land Bajo, Ellen Bialystok, David Birdsong, Lera took upon her the rather unrewarding task of obtaining the publication rights of the figures that

PREFACE AND ACKNOWLEDGEMENTS xiii I borrowed from other publications and Hilde concert with Harm Pinkster, and a meal with Smedinga’s eagle eyes spotted many typing errors Lief Keteleer. Special thanks go to three of my and other infelicities in the list of references and friends: Ruud Bleijerveld, Kees Roza, and Marjan corrected them. Finally, Ingrid Singer helped me Freriks. In need of a place to retreat to for a compiling the chapters’ summaries. couple of weeks halfway into the project, Ruud and Kees generously offered me their home in the A next group of people to thank are a couple of German Eifel that, with some force of imagin- my colleagues, former and present, at the Uni- ation, can be seen to resemble Tuscany, and in the versity of Amsterdam. When my adventurous years to follow they bestowed their hospitality sojourn in the university administration ended, upon me in their Amsterdam home over grand Rector Paul van der Heijden and Chairman meals. On many Friday evenings I met with Mar- Sijbolt Noorda granted me a sabbatical that jan in her home, spending them according to an enabled me to start working on this book and the agreeable division of labor: she cooked, I ate (in administrators of the Department of Psychology, which she joined in), and we both discussed the Agneta Fischer, Gerard Kerkhof, and Klaas week’s noteworthy events. Visser, saw to it that I landed safely again in the Department at the conclusion of this sabbatical. I am also grateful to the staff at Psychology Johan van Benthem is probably unaware of the Press, and especially to Editors Cathleen Petree fact that he also had a role to play in me reaching and Paul Dukes for their pleasant and effective the finish; suffice to say that he did. I would like to guidance and to Production Manager Mandy thank Nol Verhagen for his role in the great col- Collison and Editorial Assistant Lee Transue, lection of journals and books the University of who provided advice and help during the final Amsterdam community has access to. I am also stages of the project. grateful to the Department’s secretarial staff, and especially to Hubert Eleonora and Anna Bogerd, And then, of course, there is my wonderful for supporting the logistics of the entire family, which provides attentive and loving com- enterprise. pany whenever I need it. But there is another reason why I am indebted to my family, each of I started my sabbatical in Tuscany, where its members: Lives are largely shaped by one’s Elda and Fernando Giannini guarded my health relatives of three generations, one’s own, the and general well-being and Italy’s unparalleled previous one, and the next. I dedicate this book cultural treasures took care of the rest. These to the main representatives of these three treasures were my reward after each week of generations: my parents, my twin sister Jeannette, unremitting labor. And then there were the and my son Jan. friends that boosted my spirits in times of loom- ing despair. I enjoyed an occasional lunch with Annette M. B. de Groot Walter Hoogland, a drink with Sander Bais, a Amsterdam, May 2010



1 Introduction and Preview Bilingualism has become an omnipresent phe- In this introduction I will first describe what nomenon in our modern society of large-scale psycholinguistics is about and show how it has migration, international markets and finance, provided the basis for the study of bilingualism. backpacking youngsters, and a scientific com- Next I will present the colorful variety of munity in need of a lingua franca to disseminate language users who are all called bilingual. I will its achievements among its members. The aware- then introduce a number of central themes in the ness that bilingualism is not at all exceptional any study of bilingualism and, hence, in this book. more and may not have been so for a long time This will be followed by a brief section on con- has recently led to a steep growth in the number ventional nomenclature in the field and I will of studies on the implications of being bilingual conclude with a brief preview of the chapters to for language use and cognition in general. This come. book brings together the results of many of these studies. It presents the theories and views on But first some words are in order on the title of bilingualism that motivated these studies and this book: “Language and cognition in bilinguals emerged from them, but it also explains the and multilinguals: An introduction”. The vast research methods and tasks that were used to majority of the studies to be discussed tested address these theories and views in specific bilinguals; that is, people who know and use two experiments. Because of this latter feature this languages. Relatively few studies will be included book qualifies as an introduction to the study of that examined multilinguals—people who know bilingualism. However, it presupposes some basic and use three or more languages. This reflects knowledge of the research area of cognitive the fact that relatively few of these studies have psychology and, specifically, the psychology of been conducted and reported. Because of the language (or “psycholinguistics”). The issues that unbalance between bilingual and multilingual will be dealt with are largely based on those studies, I considered the alternative, smoother, addressed in the study of psycholinguistics, but title “Language and cognition in bilinguals: An they are approached from the perspective of introduction”, but for a couple reasons I decided bilingual language users. against it. The first is simply that it would not do justice to that fact that some multilingual studies 1

2 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS are in fact included. But a more principled reason lie these skills and what role they play in linguistic is that the theories, processes, and mechanisms behavior. In addition, psycholinguistics is con- that will be dealt with apply, I believe, to both cerned with how we acquire these skills. bilingualism and multilingualism. A multilingual language system is potentially noisier than a While exhibiting any of the above-mentioned bilingual language system, but the mental pro- four skills, the language user makes use of cesses and mechanisms that handle this increased knowledge units of various different types, each level of noise are presumably no different from of them relating to a separate domain of lin- those involved in dealing with the extra noise in a guistic study: phonology, morphology, syntax or bilingual system as compared with a monolingual grammar, vocabulary, orthography, and prag- system. To the extent that these claims are true, matics. Phonological units (“phonemes”) are the this book is as much about multilingualism as it smallest sound units of a language and the is about bilingualism, and in many places where corresponding field of study (phonology) I talk about bilinguals and bilingualism one attempts to discover the sound system of a might take these to imply multilinguals and language; that is, what sounds it contains multilingualism as well. Furthermore, as we shall and how these combine into larger sound units. see below, some people considered bilingual Morphological units (“morphemes”) are the according to one definition of bilingualism may smallest linguistic units that bear meaning and be considered multilingual by another (because the corresponding field of study (morphology) they also possess some knowledge of at least one investigates the way these meaning units can further language and can put it to effective combine into words. Syntactical knowledge is use, which counts as an additional language for knowledge about the way words can combine one theorist but not another). Finally, it is not into sentences and the corresponding domain unusual to call a particular study a bilingual one of study (syntax) tries to identify the rule system because two specific languages of the participants that underlies permissible word order and sen- are being examined while the fact is ignored that tence structure in a language. A language user’s the participants might also possess at least some vocabulary consists of all the words he or she basic knowledge of one or more other languages. knows. Knowing a word minimally involves knowing its spoken form and its meaning, but for PSYCHOLINGUISTICS AND THE STUDY OF literate language users it also implies knowing BILINGUALISM how it is spelt. A word’s spelling is called its “orthography”, but this name is also used for Psycholinguistics is an interdisciplinary field of spelling in general (especially in alphabetic study that connects the disciplines of psychology writing systems, in which the orthographic units, and linguistics. Briefly, linguistics is the field of the letters, represent phonemes). In addition it is science that describes the knowledge that under- used to refer to the field of study that investigates lies language, whereas psychology is the field of the rules of spelling of a language—how exactly science that explains behavior in terms of mental orthographic units map onto phonological units. processes. Psycholinguistics combines these two Finally, pragmatics concerns the study of how orientations by examining the mental processes people use language differently in different con- and types of knowledge involved in understand- texts, taking world knowledge and knowledge ing and producing language, in both its oral about the specific communicative circumstances and written forms. In other words, it deals with into account in choosing the exact wording. They the linguistic skills of listening, speaking, reading, might, for instance, use slang words when inter- and writing, trying to discover the cognitive acting with their peers in an informal setting but machinery and knowledge structures that under- choose more formal vocabulary when talking to their superiors, or they might choose to use some indirect form of language such as irony to maximize the communicative effect. Whereas

1. INTRODUCTION AND PREVIEW 3 linguistics is primarily concerned with describing psycholinguistics is that not all means have been these various sources of knowledge, psycho- exploited to become informed on the relation linguistics focuses on how we exploit them in between language and thought: If specific lan- using language. guages influence thought in specific ways, a per- son who masters more than one language may In addition to discovering how language is live in different worlds of thought depending on acquired and used, psycholinguistics tries to dis- the language currently used. Alternatively, this cover the relation between language and non- person’s way of thinking may be based on a linguistic cognition, specifically the relation merger of the worlds of thought associated with between language and thought. Across the cen- the separate language he or she speaks. turies, various views on the relation between language and thought have been advanced, such The insight that, as compared with mono- as the idea that thought is internal speech, that lingualism, bilingualism may alter language language is a tool to communicate thought, acquisition, representation, and processing as that language guides thought, and that language well as thought (and, as we shall see, other aspects influences thought (see Whitney, 1998, for an of non-linguistic cognition), has in recent years overview). The view that language influences led to a steep rise in studies on language use and thought also incorporates the idea that specific cognition in speakers of more than one language. languages influence thought in specific ways, with In this book I discuss many of these studies the effect that speakers of different languages and attempt to integrate the wide range of mis- might think and perceive the world differently. cellaneous findings that have emerged from the various sub-disciplines within this field of study Even now, psycholinguistics is characterized into a coherent whole. The study of bilingualism by a strong monolingual bias: The participants deals with largely the same topics and the same who are asked to perform some language task in linguistic domains and skills as covered by some experiment are typically native speakers of traditional psycholinguistics, and can therefore be the test language and it is implicitly assumed— regarded a branch of psycholinguistics. But possibly mistakenly—that they lack knowledge of because the study of bilingualism has only quite any other language(s). Alternatively, the investi- recently started to gain momentum it has not yet gators might be well aware that the participants grown as many branches as its source, traditional might speak one or more other languages in psycholinguistics, and some of the sub-areas addition to their native language, or that they that have started to emerge have not yet had are native speakers of two languages, but this the chance to develop into anything more than a possibility is simply ignored or taken for granted. tender twig. The monolingual orientation of psycholinguistics has arguably led to an incomplete conception, The study of bilingualism can be dissected into possibly even a false one, of human linguistic three main areas of study that map directly onto ability and language processing, because knowing the three main lines of study within traditional more than one language may have an impact on psycholinguistics. The first examines how lan- the way each individual one of them is mentally guage users understand (or comprehend) lan- represented and processed. If forced to single out guage input; the second how they produce the most salient result emerging from the study of language output. The main difference between bilingualism/multilingualism to date, I would the traditional psycholinguistic studies and the choose the ubiquitous effect of the language(s) bilingual studies is that only the latter address currently not in use on the one selected for cur- these topics from the bilingual perspective, testing rent use. In addition, acquiring a new language bilinguals and posing the question of how the is influenced by prior knowledge not only of the fact that they speak more than one language first language but of all further languages of influences the way they process language. The which the learner has at least some knowledge. A third main line of study is language acquisition further consequence of the monolingual bias in and can be divided into two sub-areas. One of

4 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS them deals with the simultaneous acquisition of allow for lower levels of fluency in one of the two languages from birth and how it compares languages or perhaps in both and, if, so, what with acquiring just one language. The second is minimal level would be required to be included? concerned with acquiring a second language after Is this minimum level of fluency required for all a first one is already partly or fully in place. of the four linguistic skills distinguished earlier or does it suffice, for instance, to be able to just speak As mentioned above, language comprehension in both languages or to just read in both lan- and production both occur in two modalities guages? In fact, in bilingualism research a wide depending on the nature of the input and out- range of more specific definitions of bilingualism put: speech or writing. Taking both dimensions can be encountered. These vary from only con- of classification (comprehension/production; sidering a person bilingual if he or she masters speech/writing) into account, four linguistic skills two languages at the same level of fluency and can be distinguished: listening, reading, speaking, with the same level of control as native speakers and writing. The first three of these have of the two languages—as if a bilingual person is established a clear presence on the agenda of bi- two monolinguals in one person—to regarding lingualism researchers, but studies on the writing people who only possess some minimal com- skills of bilinguals are still sparse. Another note- petence in one of the four linguistic skills as bi- worthy feature of the state of the art in bilingual- lingual. According to the first of these definitions ism research is that relatively many of both the the bilingual population is a very small one, if comprehension and the production studies only because of the fact that the language selected that have been performed have investigated the for current use is influenced by all of the other processing of words instead of larger linguistic languages known by this speaker (see above). units such as complete sentences or texts. For According to the second definition even people example, the majority of the bilingual speech who are in an initial stage of second language production studies conducted so far have learning count as bilingual. examined the production of single words out of context. Finally, looking at the current state of Bilinguals are classified according to a number the study of bilingualism from the viewpoint of dimensions (see e.g., Butler & Hakuta, 2004; of the various types of linguistic knowledge dis- Hamers & Blanc, 1989). Each of these dimen- tinguished above, it appears that (aspects of) sions will be explained briefly here and detailed vocabulary, phonology, and syntax have received further elsewhere in this book. One of them is a ample attention while studies on morphology and classification according to their relative com- pragmatics are still rare. The content of this book petence in both languages. So-called “balanced reflects this imbalance. bilinguals” possess similar degrees of proficiency in both languages, whereas “dominant bi- DEFINING AND CLASSIFYING BILINGUALS linguals” (or “unbalanced bilinguals”) are those with a higher level of proficiency in one language Which language users count as bilinguals? An than in the other. Balanced bilingualism does obvious and common answer to this question not necessarily imply a high competence in both is that any person who speaks more than one languages. For instance, a child who masters two language is a bilingual, but this answer does not languages to equal degrees is said to be a balanced take into account the wide range of differences bilingual even though in neither language has he between individuals who fit that definition. For or she reached full competence yet. Therefore a instance, it does not do justice to the fact that one further distinction to make is according to the can speak a language at different levels of fluency. level of fluency or proficiency in, especially, the Does the above definition imply that the level of second language. Bilinguals who have attained a fluency in both languages is nativelike or does it (near-) native level of proficiency in this language are called “proficient bilinguals”; those who have not are called “non-proficient bilinguals”.

1. INTRODUCTION AND PREVIEW 5 Three other dimensions concern the age at when there is a strict separation in the use of the which second language acquisition starts (the two languages. so called “age of acquisition” variable), the way words and their meanings are organized in A division that splits up bilinguals according bilingual memory, and the social status of each of to the status of their two languages is one the languages. The age of acquisition dimension between “additive bilinguals” and “subtractive splits up the bilingual population into “early” bilinguals”. Additive bilingualism is thought to and “late” bilinguals. Early bilinguals are those arise in circumstances wherein both languages are who acquired both languages in childhood, socially valued, whereas subtractive bilingualism whereas late bilinguals became bilingual beyond results from a situation in which one of them, childhood. In its turn, early bilingualism is usually the child’s native language, is devalued in divided into early “simultaneous” and early “con- his or her environment and there is social pressure secutive” or “sequential” bilingualism. Early not to use it. Additive bilingualism is considered simultaneous bilinguals have been exposed to to be beneficial for cognition and cognitive both languages from birth, whereas early con- development, whereas subtractive bilingualism is secutive/sequential bilinguals have first been thought to hamper them. The distinction between exclusively exposed to one language, their native additive and subtractive bilingualism has also language, and from some later point in time been referred to as one between “elite” and “folk” during early childhood have started to receive bilingualism, respectively. bilingual input. In some classifications late bilingualism is also split up into two sub- This enumeration of dimensions according to groups—adolescent and adult bilingualism. This which bilinguals can be categorized is by no subdivision relates to the view, held by some means exhaustive (see Butler & Hakuta, 2004, scientists in the field, that adolescence marks and Hamers & Blanc, 1989, for further divisions) a critical boundary in language-learning but will do to make the point that the bilingual ability. community are a colorful lot. One should there- fore think twice before generalizing a conclusion The classification that splits up bilingualism based on the results of a study testing bilinguals according to bilingual memory organization dis- of a specific type to another type, and it behoves tinguishes between “compound”, “coordinate”, the author of any bilingual study to provide and “subordinative” bilingualism. In bilinguals details about the type of bilinguals tested. of the compound type the two word forms of a translation-equivalent word pair map onto one CENTRAL THEMES, CONCEPTS, and the same meaning representation in memory, ASSUMPTIONS, AND QUESTIONS whereas in coordinate bilinguals each term in such a word pair maps onto a separate meaning Arguably the most central set of assumptions in representation. In subordinative bilingualism, as both general psycholinguistics and in the study in compound bilingualism, there is also just one of bilingualism is that linguistic knowledge units meaning representation for both elements of a like words, phonemes, and, perhaps, grammatical translation-equivalent word pair, but this time the rules each have a representation in long-term word form of the weaker language does not map memory, and that language processing involves directly onto this meaning representation but via the concerted activation and deactivation of the word form of the stronger language. These subsets of these memory representations. This three forms of bilingualism have been related to set of assumptions applies to both language different contexts of acquisition. For instance, it comprehension and language production. Take has been hypothesized that compound bilingual- for example word recognition, perhaps the most ism emerges when a child grows up in a home thoroughly examined language subskill in the where the two languages are spoken interchange- study of bilingualism. Each word representation ably, and that coordinate bilingualism emerges

6 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS in our mental lexicon (and each component part As we shall see, these questions have domin- of this representation) is assumed to be activated ated a substantial part of the bilingualism at some baseline level. If a particular word is research to date. The question of whether or not presented for recognition, it boosts the level of elements of the non-target language are co- activation of the corresponding lexical represen- activated with elements of the target language has tation. If the activation of this representation been extended from the vocabulary domain to surpasses some critical threshold level, the cor- other domains of language such as phonology responding word is recognized. Similar activation and grammar. The question of how bilinguals processes, but now incited by the intention to resolve the extra-fierce competition within their express a particular content of thought instead of mental language system has led to the insight that from an input presented to our ears or eyes, lead bilinguals possess a very effective control system to word production. which sees to it that linguistic performance meets their current goal; for instance, to now speak in A word presented for recognition increases the this specific language, not the other. Subsequent level of activation of not only “its own” lexical research has revealed that, presumably, the representation, but also that of representations control system in question is not exclusively of words similar in form (spelling or sound, dedicated to language control but is likely to be a depending on the modality of input) to the input general executive control system that is implicated word. Because the level of activation in one or in the control of behavior in general. This, in more of these simultaneously activated represen- turn, has led to the hypothesis that, as compared tations may also approach the critical threshold, with monolinguals, bilinguals might be experts in this leads to competition in the word recognition (some aspects of) executive control because they process. Similarly, the conceptualization of a par- exploit the underlying cognitive machinery inces- ticular content to put into a word activates not santly to always use the intended language. These, only the lexical representation of the targeted in a nutshell, are some of the most important word but also those of words with a similar mean- themes of this book. ing as the target word. This leads to competition in the word production process. A further central question in traditional psycholinguistics (and in linguistics) is whether Bilingualism researchers have wondered about humans have an innate capacity for learning the implications of these views on the activation language. Human languages, especially their dynamics of word recognition and word pro- grammars, are extremely complex systems and duction for bilingual speakers. Bilinguals possess yet children start to exhibit knowledge of the two stocks of lexical representations, one for each underlying rule system within a matter of about language. Does a spoken or written word (in 2 years. This feat has led a substantial part of the comprehension) or a specific conceptual content scientific community to assume that the human (in production) give rise to activation in both capacity for language learning is at least partly of these lexical subsets, or can bilinguals some- innate. Other researchers reject this idea and how “switch off” the contextually inappropriate believe that language learning is enabled by language? If the latter holds, what mental general learning mechanisms that are exploited mechanism is involved in switching off the non- for learning in other domains of cognition as intended language? If the former holds, this well. In this book an echo of this debate can be implies that bilinguals must experience more heard in, especially, the discussion on language competition during word recognition and word development in monolingual-to-be and bilingual- production than monolinguals do, and this to-be children, and in the review of studies on must have consequences for performance. A age of acquisition effects on (first and) second further question, then, is how bilinguals resolve language learning. An important conclusion to the extra-fierce competition and manage to draw from the studies on language development correctly recognize or produce the intended is that children have remarkably good general word.

1. INTRODUCTION AND PREVIEW 7 perceptual skills that bootstrap language acquisi- classes. In this book I will generally use the more tion, and that these skills on their own might neutral term L2, except in Chapter 3, which suffice to explain children’s rapid language largely deals with L2 vocabulary learning in development. The age of acquisition studies typical school settings. Of many of the studies clearly show that if one starts learning a language presented in this book I will explicate the two at a young age, ultimate linguistic performance is languages involved in the participants’ bilingual- generally better than when language learning ism. For example, I may talk about the Spanish– starts relatively late in life. This superior ultimate Catalan participants in a particular study or the performance of those who start young may French–English ones in another. Generally, in inform the present “nature–nurture” debate, these cases the language mentioned first is the because it might result from an innate language participants’ L1 and, at the same time, strongest faculty that is present early in life but is dis- language, and the one mentioned second their mantled at some point before adulthood (see L2 and weaker language. Exceptions to this rule e.g., Pinker, 1994). Other effects of acquisition will be noted. age that have been assumed are that language acquisition differs between early and late learners, In many situations bilinguals have to use one that early and late learners process language dif- specific language while trying to prevent intru- ferently, and that the network of brain structures sions from the other language. The language to that are recruited during language processing use in a specific setting is variously called the differs between them. These themes are also all “target” or “targeted” language, the “intended” dealt with in this book. language, the “selected” language, the “language- in-use”, the “chosen” language, or the “response” SOME NAMING CONVENTIONS language (and still further names might be used as well). Conversely, the language not to use in a For convenient reading, in this book the first and specific setting is referred to as the “non-target” native language is often referred to briefly as L1 or “non-targeted” language, the “non-intended” (for Language 1) and the second language is language, the “non-selected” language, the called the L2. The number conventionally refers “language-not-in-use”, the “non-chosen” lan- to the order of acquisition and any languages to guage, or the “non-response” language (and, be acquired later are numbered accordingly (L3, again, still further names occur). In this book L4, etc.). In the case of unbalanced bilingualism, I will use some of these terms, but most often I the L1 is often, but not always, the stronger one will talk about the target (or targeted) and non- of a bilingual’s two languages. I will use these target (or non-targeted) language. same conventions throughout. Whenever in a specific study the L1 and dominant language are PREVIEW OF THE CHAPTERS not one and the same (as might be the case for an immigrant who has resided in the new country for The first chapter to follow this introduction, quite some time), I will explicitly clarify which Chapter 2, consists of two parts. The first com- language is the stronger one. A distinction is pares language development in monolingual-to- sometimes made between a second language (L2) be and bilingual-to-be infants. It focuses on and a foreign language (FL). Where this is done, phonological and lexical development in the “foreign” means that the L2 in question is not first years of life as witnessed from infants’ per- an official language of communication in the formance on speech perception tasks. Due to the country where it is learned or used. For instance, availability of relatively new experimental tech- English learned in a school in Japan or Italy is a niques to assess infants’ perceptual skills, in this foreign language for the students in those English research area in particular substantial progress has been made in recent years. As a result, our

8 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS understanding of the development of speech evidence suggests that co-activation in the non- production has greatly increased as well (because targeted subsystem is ubiquitous if the word is the development of speech perception and speech presented out of context. Further questions to be production go hand in hand). The second part of dealt with are whether a linguistic context might Chapter 2 deals with the question of why L2 users constrain this nuisance co-activation and what who started to learn the L2 when they were young type of memory representations (phonological, generally achieve a higher level of proficiency orthographic, semantic) are simultaneously acti- than late learners. One popular explanation of vated in both language subsystems. In addition, this phenomenon will be highlighted in the a couple of bilingual word recognition models discussion: that during some bounded period will be discussed that account for evidence of early in life humans possess a heightened sensi- co-activation. The chapter concludes with a dis- tivity to language and that, for a language to be cussion of bilingual language comprehension acquired fully, the period of acquisition should beyond the level of the word, addressing the coincide with this period of heightened question of how bilinguals parse syntactically sensitivity. ambiguous sentences and process semantic and syntactic anomalies. Chapter 3 examines late foreign vocabulary acquisition and the ensuing types of representa- Chapter 5 first presents models of mono- tions in the bilingual mental lexicon. Late foreign lingual and bilingual speech production, zooming vocabulary learning differs substantially from in on the word production component con- vocabulary learning in L1 acquisition. One of the tained by the complete process. Subsequently, it differences is that in L1 vocabulary learning a focuses once again on the question of whether word’s form, its meaning, and the link between simultaneous activation occurs in a bilingual’s the two must all be learned. In foreign vocabulary two language subsystems, but this time from the learning the meanings of many of the words to be viewpoint of word production instead of word learned are already largely in place. After all, the recognition. The majority of the pertinent studies learner already knows the meaning of many L1 have used versions of two classes of tasks to words and these are largely similar to those of the examine this question: picture naming and corresponding foreign words. The only thing that color naming. The evidence from both classes needs to be done in the initial stage of foreign of tasks will be reviewed. In addition, a set of vocabulary learning is to connect the new foreign studies will be discussed that have investigated forms to the corresponding extant meanings. bilingual word production using word transla- The most popular methods that are used to tion. The chapter concludes with a discussion and make this connection, the keyword method and account of the speech accents that characterize paired-associate learning technique, are presented the speech of the vast majority of non-native and their efficacy will be discussed. In addition it speakers. will be explained why some types of words are easier to learn than other types, and attention will Bilinguals are generally quite good at avoiding be devoted to later stages of learning, in which the intrusions of the language they currently do not meanings of the foreign words are gradually want to use: Code switches into the other lan- refined and become more nativelike. Finally, guage are rare. Conversely, in situations where various views on the structure of bilingual lexical they need to switch between their two languages, representations and how they relate to levels of for instance in translation settings, they can do foreign language fluency will be presented. so. These skills, to maintain a language or to switch between languages, are together known as Chapter 4 largely deals with the question, bilingual “language control”. Chapter 6 presents introduced above, of whether a spoken or written various views on how bilinguals exert language word input causes activation in both of the lan- control and reviews the evidence for each of guage subsystems in bilingual memory or only them. A central assumption in most of these in the contextually appropriate subsystem. The views is that the activation level of the elements

1. INTRODUCTION AND PREVIEW 9 in a bilingual’s two language subsets can be Finally, Chapter 8 deals with the study of the regulated such that those of the target language bilingual brain. One of the questions examined in become more activated and those of the non- this chapter is what parts of the brain are target language deactivated. The mechanism recruited when bilinguals process language, and assumed to take care of these regulatory pro- whether these might differ from those involved cesses in the most topical of these views is the when monolinguals process language. It is well general executive control system introduced known that the two hemispheres of the human above. The chapter concludes with a discussion brain are not equally involved in language of the way control is exerted in what is arguably processing but that in the vast majority of the most difficult type of linguistic behavior: monolingual language users the left hemisphere simultaneous interpreting. dominates language processing. This is known as left “lateralization” of language. One of the One of the topics discussed in Chapter 7 is specific questions addressed in this chapter is language loss, of both the L1 and L2. Two further whether language might be lateralized differently ones concern aspects of the influence that each of in bilinguals, as some researchers have assumed to a bilingual’s or multilingual’s language exerts on be the case. Furthermore, this chapter reviews the other(s). One of these topics is which one studies that have tried to identify the brain of two earlier languages affects learning a third structures involved in bilingual language con- language most, and why it is that earlier lan- trol—the ability of bilinguals to maintain the guages have differential effects on learning a use of one language if such is demanded by the new one. The second concerns the influence of current circumstances and to switch between later languages on the first, in the domains the two languages under circumstances where of phonology, grammar, and semantics. The switching is required. As we shall see, this line of remaining parts of this chapter deal with effects research suggests that bilinguals exert language of bilingualism on non-linguistic cognition; control by exploiting the brain structures that specifically, on thought, general intelligence, and subserve general executive control. cognitive control.



2 Early Bilingualism and Age of Acquisition Effects on (First and) Second Language Learning INTRODUCTION AND PREVIEW and events rather than to individual entities and that they can do so even under circumstances On the face of it, acquiring a language is an easy in which the words’ referents are not actually thing to do. After all, it does not take long before present in the environment. newborn children start to utter the sounds of their native language and, some months later, One feature of the speech signal in particular their first words. Yet linguistic competence complicates language learning; namely, the fact requires the mastery of an extremely complex lin- that the primary meaning-bearing components guistic system which appeals to many sub-skills of language, its words, do not occur as discrete that all exploit a large database of knowledge. packages of information in the speech signal. Tens of speech sounds and thousands of words Visual records of speech fragments produced must be learned, as well as phonological and by means of a technical device called a spectro- grammatical constraints on combining and order- graph show that the breaks in a speech signal ing them in words and sentences and, ultimately, do not correspond straightforwardly to word ways of organizing sentences in coherent dis- boundaries. Often the pauses in the signal fall course. Word learning on its own is already a within words and, conversely, many boundaries multifaceted process, involving the learning of the between words are not marked by a clear speech words’ phonological forms (that is, their sound break. Pauses in the speech signal are thus un- patterns), the awareness that these forms carry reliable cues for developing a lexicon. A further meaning, the linking of the sound patterns to cumbersome characteristic of the speech input meaning, and the understanding that words is that one and the same phoneme (and, as a generally refer to whole classes of objects, people, consequence, one and the same word) is realized in an endless variety of ways in actual speech, depending on speaker characteristics (e.g., male 11

12 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS or female; child or adult), speech rate, and the native language speech perception bootstrap adjacent sounds. How then does the infant speech production, is another reason to highlight manage to unearth the relevant phonemes and the infant speech perception. first words from this noisy input so soon after birth, and what are the skills and mechanisms Whereas the first part of this chapter examines that get vocabulary acquisition going? These early bilingualism during development and, questions become all the more intriguing when we accordingly, the participants in the studies to be realize that newborn children growing up in a reviewed are always infants or toddlers, the bilingual environment do not take much longer to second part compares second language ability acquire the phoneme inventories and some in “early” and “late” adult bilinguals; that is, in vocabulary of both ambient languages than it adults who began to learn their second language takes infants growing up in a monolingual con- when young or relatively late in life. Innumerable text to acquire the corresponding knowledge studies have addressed the question of whether structures of their one native language. and how the age at which one starts to acquire a second language affects the level of proficiency Since Eimas and his associates introduced a that is ultimately reached in it. I will organize the sensitive new technique to assess the ability of discussion of this work around the popular “crit- infants to discriminate speech sounds (Eimas, ical period hypothesis”, which holds that during 1974, 1975; Eimas, Siqueland, Jusczyk, & some bounded period early in life humans possess Vigorito, 1971), and other researchers developed a special sensitivity for linguistic input and that clever new versions of this technique, our for a language to be acquired fully its acquisition knowledge about the way infants gradually come should coincide with this period of heightened to master the sound system(s) of their native sensitivity. The critical period hypothesis applies language(s) and, in parallel to this development, to the learning of any language, including the start to recognize and produce words, has tre- first, and studies that examined the effect of mendously increased. The technique in question acquisition age on first language learning will also will be explained next, in the Methods and Tasks be discussed in this section. Whereas the chapter’s section, together with other methods that are first part focuses on phonology and early lexical used to investigate this chapter’s themes: early skills, its second part primarily deals with bilingualism and age of acquisition (AoA) effects grammar. The question of how age of acquisition on (first and) second language learning and pro- affects second-language speech accent will be ficiency. In the first part of this chapter current dealt with in Chapter 5. insights into the development of language in monolingual- and bilingual-to-be infants will be In addition to examining infants or toddlers presented. The focus will be on phonological and versus adults, a further difference between the early lexical development as manifested in speech studies covered in this chapter’s two main sections perception tasks; that is, tasks that assess the concerns the type of early bilinguals that are infants’ ability to recognize phonemes and sound examined in both areas of study. The participants patterns that correspond to words. In addition in many of the developmental bilingual studies some attention will be devoted to the develop- are infants who are exposed to bilingual input ment of the ability to link word forms to from birth, a situation that has been coined meaning. The reason for dealing primarily with bilingual first language acquisition (BFLA) or speech perception, not speech production, is “2L1” by Meisel (1989, 2001) and that is also that choices had to be made and that especially referred to as simultaneous bilingualism. The the research area of infant speech perception development of bilingualism in these children, has recently witnessed substantial progress. The who can be said to have two native languages, insight that speech perception and speech pro- differs from early second language acquisition or duction do not develop independently from one sequential (or “successive”) bilingualism, where another, but that the infant’s gradual advances in the development of the native language is well under way the moment the child gets introduced

2. EARLY BILINGUALISM AND AGE EFFECTS 13 to a new, the second, language. The early habituation phase and eventually be back at some bilingual adults partaking in studies that look at low baseline level. A test phase then follows the role of age of acquisition on ultimate attain- in which a new speech sound is presented which, ment are typically of this kind. In sequential if the difference between the new and old stimulus bilingualism, by the time the child is first exposed is detected at all, will arouse the child’s attention to the second language, phonological, lexical, again. A control condition may also be included and grammatical knowledge about the native in which one and the same sound is presented language will partly be in place already, the more during habituation and test. In this condition of it the later the child starts receiving bilingual attention should remain at the same low level as input. This situation is known to give rise to reached at the end of habituation. transfer between the two languages and, espe- cially, from the native onto the second. On the To apply this paradigm in the study of speech other hand, in BFLA the two languages appear perception in babies a technique is required that is to develop relatively independently from one sensitive enough to detect fluctuations in atten- another (e.g., De Houwer, 2005; Genesee, tion in children this young. One such technique Nicoladis, & Paradis, 1995; Meisel, 2001). A measures the rate at which infants suck on an further difference between BFLA and early artificial pacifier, a high rate indexing attention sequential bilingualism relates to the fact that, in and a low rate indexing that they have lost inter- tandem with increasing brain weight, general est. Eimas et al. (1971) first used this technique cognitive skills improve during maturation. The in an experimental procedure (called the high- implication is that also as a consequence of amplitude sucking paradigm, HAS) that they improved general cognition, second language developed to find out whether young infants, just development in sequential child bilingualism pro- as adults, perceive speech sounds in phonemic ceeds differently from language acquisition in categories. This phenomenon, called categorical BFLA. Age-related differences in cognitive perception, implies that listeners can hear the dif- skills—both the improvement thereof during ference between two speech sounds that represent maturation and cognitive decline beyond a certain two different phonemes but fail to discriminate a age—should also be taken into account when pair of speech sounds that are variants of one and evaluating the findings from studies on the role the same phoneme. This even holds if the acoustic of age of acquisition on ultimate L2 attainment. differences between the two phonemes on the one hand and the two variants of one phoneme on the METHODS AND TASKS other hand are equally large. Figure 2.1 illustrates the HAS paradigm, with the infants’ sucking rate The primary research method used to assess the on the y-axis and the time from the start of a speech perception abilities of newborn children trial’s habituation phase on the x-axis. In the makes use of the fact that babies pay more atten- example the trial’s total time lasts 9 minutes and tion to novel stimuli, visual and aural, than to consists of 5 minutes of habituation followed by stimuli they have perceived before (e.g., Melson & 4 minutes of testing. The dashed line represents McCall, 1970). This phenomenon was exploited the moment at which, following habituation, a in the development of the habituation paradigm new speech sound is presented to an experimental or familiarization paradigm in which, in the group of infants but not to a control group. habituation phase, young infants are repeatedly Figure 2.1 shows a situation in which the experi- presented with one and the same stimulus, say a mental group noticed the difference between the speech sound. When this sound is first presented, speech sounds presented during habituation and it will arouse the infant’s attention. Attention testing, as evidenced by the increased sucking will subsequently gradually lessen during the rate, indexing renewed attention, upon the pre- sentation of the new sound. Employing this procedure, Eimas and his associates (1971) demonstrated categorical

14 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS Mean number of sucking responses as a function of time since trial onset and experimental condition. Total trial time is 9 minutes. The dashed line indicates the moment the experimental group is presented with a new speech sound. The baseline sucking rate is also indicated. Adapted from Eimas et al. (1971) with permission from AAAS. Spectrograms of two synthetic plosives, one with a VOT of plosive is uttered. It is illustrated in Figure 2.2, +10 ms, the other with a VOT of +100 ms. The formants are which shows spectrograms of two synthetic bands of relatively intense acoustic energy in the speech plosives, one with a VOT of +10 ms and the spectrum. From Eimas et al. (1971). Reprinted with second with a VOT of +100 ms. A spectrogram is permission from AAAS. a visual record of a speech fragment produced by a spectrograph and shows an analysis of perception of (synthetic, machine-made) plosives the speech fragment in the constituent sound that only differed from one another in voice onset waves and how they develop over time. The hori- time (VOT) in 1- and 4-month-old infants. VOT is zontal and vertical axes of a spectrogram repre- the time between the release of the air and the sent time and frequency of the constituent sound moment the vocal cords start to vibrate when a waves, respectively. The intensity of the speech sound at any moment in time is indicated by the degree of blackness of the marks on the paper— the higher the intensity, the blacker the markings. The black bands in Figure 2.2, called formants, thus represent frequency bands of relatively intense acoustic energy in the speech spectrum. The investigators found that the sucking rate increased at test when the two plosives presented during habituation and at test, respectively, had VOTs that are contrastive in English (which means that two plosives that only differ in this one respect represent different phonemes, in the present case the phonemes /p/ and /b/). The increased sucking rate at test indicates that the infants could discriminate between these two plosives. In contrast, the sucking rate remained low at test when, again, two different plosives were presented during habituation and at test, however the two now had VOT values that are not contrastive in English but are manifestations of one and the same phoneme. This suggests that the

2. EARLY BILINGUALISM AND AGE EFFECTS 15 infants failed to discriminate between this pair of decrease when the infant gets used to the con- plosives. Importantly, this result occurred despite current speech sound. (Because looking time the fact that the VOT difference between the non- serves as a measure of listening time and it is the contrastive pair of plosives was equally large as latter that the researcher is interested in, the the difference between the plosives of the con- technique is also called the preferential listening trastive pair. In subsequent studies Eimas general- technique). At the onset of the familiarization ized these findings to other so-called “minimally phase the looking time per trial will be relatively contrastive” pairs of consonants, such as a con- long because of the novelty of the presented trast between plosives in their place of articulation speech sound. The habituation phase is ended the (Eimas, 1974; e.g., /p/ versus /t/) and the contrast moment looking time drops below some preset between /r/ and /l/ (Eimas, 1975). Because con- minimum duration. At that point either one of sonants cannot be uttered separately from a two new speech stimuli is presented; for instance, vowel, in all experiments of this type the unit of a new variant of the phoneme presented during presentation is a syllable, consisting of the critical habituation or an exemplar of a different phon- consonant and a vowel, the latter being identical eme. An increase in looking time indicates dis- for the two consonants in a contrasting pair. For crimination between the sounds presented during instance, the contrast between /p/ and /b/ is tested habituation and test. If looking time does not with the synthesized syllables /pa/ and /ba/. recover at test one may conclude that the test stimulus is not perceived as different from the Infants’ increased attention to novel stimuli stimulus presented during habituation. can also be detected with other experimental pro- cedures. One is the heart-rate paradigm, which This general procedure has not only been exploits the fact that an increase in attention is used to chart the speech sounds that infants can accompanied by an increase in heartbeat (e.g., and cannot discriminate at various ages, but also Lasky, Syrdal-Lasky, & Klein, 1975). Changes in to discover whether and when infants can dis- a cardiogram that occur the moment a new criminate between languages. It has also been speech sound is presented thus suggest that the used to find out how word learning in infants infant has noticed the change. A technique that comes about, including the prerequisite ability to has more recently become popular is a visual segment the speech stream into words (see Werker fixation habituation procedure sometimes called & Byers-Heinlein, 2008, for a review). Some of the preferential looking technique. It makes use of the studies that employed the preferential looking the fact that infants look at the place where an paradigm will be presented in more detail in the interesting new sound comes from and stop sections to come. As you will see, the exact pro- looking when the novelty of the sound wears off. cedure used differs between studies. For instance, As in the high-amplitude sucking and heart- the stimuli presented during familiarization, if rate paradigms, the infants are generally first uttered by real speakers and not generated by a familiarized with one type of speech stimulus, speech synthesizer, may be delivered by one and for instance a phoneme (embedded in a syllable, the same speaker or by different speakers. The e.g., /ba/). The speech stimulus is accompanied latter procedure mimics the fact that in natural by some visual stimulus, typically presented on a speech different instantiations of one and the TV screen, to attract the infant’s attention. The same linguistic unit typically differ somewhat source of the speech stimulus is located near the from one another due to speaker variability and TV screen. The infant’s looking time at the screen co-articulation. A further difference between is registered for each trial—a trial consisting of studies using this paradigm concerns the duration the repeated presentation of the speech stimulus. of the familiarization phase. In the habitation The underlying assumption is that looking time procedure as described above the infant controls is a reflection of listening time which, in turn, is the duration of this phase: Familiarization stops a reflection of attention time, and that as a con- the moment looking (= listening) time drops below sequence of habituation the looking time will some predetermined level. In another version of

16 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS the procedure the duration of familiarization is 30 degrees toward this light it continues flashing fixed for all participants within a study. Depend- until the infant looks away from it again for a ing on the duration of this fixed period and, thus, certain duration. At this moment the trial ends. the number of exposures to the familiarization After familiarization a test phase follows wherein stimulus, the infant may show a listening stimuli are played from only one loudspeaker at preference for the novel or for the familiar the time and this time stimulus presentation is stimulus. Specifically, it has been noticed that contingent on the infant’s behavior: When the after a relatively short familiarization period the infant turns his or her head toward the flashing infant may show a familiarity preference at test, side light, a stimulus is presented from the loud- listening longer to the familiar stimulus than to speaker on that side. The presentation of the the novel stimulus. In contrast, the infant may stimulus and the flashing of the light both stop show a novelty preference after a relatively long when the infant fails to maintain the head turn familiarization procedure, listening longer to the for a predetermined amount of time. The stimuli novel stimulus (Saffran, 2001). But the crucial presented at test are either the ones presented finding in these studies is the occurrence of a during familiarization (or related to them in some difference in listening time for familiar and novel way or other) or unfamiliar ones. If the test phase stimuli, not its direction. A final difference shows a difference in looking (= listening) between studies to mention here is that the time between the stimuli presented duration preferential looking procedure (or any other of familiarization on the one hand and the new the above habituation procedures) does not stimuli on the other hand it can be concluded that always require an initial familiarization phase as the infant notices the difference between them. part of the experiment, because prior linguistic This procedure has also successfully been used experience in a natural language environment with stimuli of various kinds: phonemes, words, may also serve as the familiarization phase. For and complete sentences. instance, to examine whether, say, 10-month-old infants recognize particular frequently occurring To examine the development of a particular words, one may simply compare listening times, linguistic ability a cross-sectional design is as assessed by means of the preferential looking typically employed in which groups of infants technique, for these words and a set of words are selected that differ in age but are matched unlikely to be known at that age. with respect to relevant background variables such as the parents’ socioeconomic status. The A related technique is the head-turn procedure, studies that focus on language development in which also comes in various versions and was bilingual (more precisely, bilingual-to-be) originally developed by Kemler-Nelson et al. children typically include monolingual (to-be) (1995). This time, speech sounds are emitted control groups so that differences in the language from two loudspeakers positioned left-front and development of monolingual and bilingual right-front of the infant and, as in the above children can be assessed. Incidentally, “mono- paradigms, a test phase is preceded by a familiar- lingual” in these studies does not always imply ization phase. In the familiarization phase of the that the children in question are exclusively original version the infant learns to turn his or her exposed to their native language, nor does head towards a light that flashes near the one “bilingual” imply that the children concerned or the other loudspeaker while a stimulus set is receive balanced bilingual input. Many of the emitted continuously from both loudspeakers. A bilingual studies are carried out in bilingual familiarization trial starts with the flashing of a communities where language-pure exposure does light that is centered between the two speakers not occur. For instance, in Wales some exposure and extinguishes when the infant looks in the to English cannot be prevented so that children direction of this light. At that moment the side from Welsh monolingual families are exposed light near one of the speakers begins to flash. to some English as well. Similarly, in Catalonia After the infant has made a criterion head turn of exposure to Spanish cannot be completely

2. EARLY BILINGUALISM AND AGE EFFECTS 17 prevented. What “monolingual” means in these of some type. This technique is described more studies is that exposure to the one and the other completely in later chapters (Chapters 4 and 8). language is extremely unbalanced, say 95% versus Here it suffices to say that the presentation of 5%. Instead, “bilingual” means that the dual lan- a particular stimulus induces small voltage guage exposure is more balanced, say minimally changes in the electroencephalogram (EEG). 40% for each language. These voltage changes, which can be positive (indicated by P) or negative (indicated by N), are A requirement of many of the studies that the ERPs. One and the same stimulus can give focus on vocabulary development is that infor- rise to several ERPs that take different amounts mation is available on how many and what words of time from stimulus onset to develop. An ERP are likely to be known by the participating is often named after its polarity (that is, whether infants or toddlers. This information is needed to it is positive or negative) and its “latency”, the select the experimental materials and to properly time it takes to develop. For instance, the P200 match bilingual infants to monolingual controls. is an ERP with a positive polarity that takes Instruments that are often used for these pur- about 200 ms following stimulus onset to develop. poses are the infant and toddler versions of the ERPs are very sensitive measures of cognitive American-English MacArthur Communicative activity, sometimes pointing up some ongoing Development Inventory (CDI; Fenson et al., cognitive process that behavioral measures fail to 1993, 2000) adapted to the language under study detect. (see e.g., De Houwer, Bornstein, & De Coster, 2006). A CDI concerns a list of words arranged in Obviously, the central manipulation in studies various categories that the participants’ parents examining the relation between the onset age of or other caregivers check off, indicating which L2 acquisition and L2 language proficiency is ones they believe the child understands, or under- the age of acquisition manipulation: Groups of stands and produces. In bilingual studies CDIs participants are tested on a specific linguistic in both languages are used and dominance ability, say L2 grammar, the different groups differences in lexical knowledge between the consisting of participants that first started to participants’ two languages can be established. learn the L2 within different age ranges (e.g., Furthermore, Total Conceptual Vocabulary size 3–6, 13–16, and 23–26 years). Other studies have (TCV) in bilingual infants can be determined by not compared the performance of discrete age first calculating the sum of the words reported to of acquisition groups but have correlated age of be known in each of the languages separately and acquisition and performance scores of a large then subtracting from this summed score the number of individuals that together cover a large number of translation-equivalent pairs among part of the whole age of acquisition spectrum. this total set (e.g., Conboy & Mills, 2006; Pearson Ideally, all participants in these studies have used & Fernández, 1994). A procedural decision that the language under study a large number of years must be made prior to running an experiment is so that it can safely be assumed they have reached whether the stimulus materials will be delivered in their ultimate level of proficiency in it and do not child directed speech (CDS or motherese) or in still qualify as learners. Furthermore, wherever speech with which adults are typically addressed. possible potentially confounding variables, such CDS is characterized by exaggerated intonation as the chronological age of the test takers, their and clear articulation, is spoken slowly and, if the length of residence in the L2-speaking country, stimuli concern sentences at all, the sentences are their educational level, and their amount of daily typically very short. L2 use, should be controlled for. To examine age of acquisition effects on first language learning, In addition to using the above behavioral the linguistic skills of deaf people can be assessed methods, the linguistic skills of infants and while varying the age at which they first started toddlers can, and have, also been studied by to have full access to a language, typically sign registering “event-related potentials” (ERPs) language. A second method that has been used while they are presented with language materials

18 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS is to study the rare cases of people deprived of 0 ms, or +40 ms and +60 ms). A third subgroup language during childhood because of physical was presented with exactly the same sound during isolation from other human beings. habituation and test. The important finding was that only the first of these subgroups of infants, in EARLY BILINGUALISM IN DEVELOPMENT both age groups, showed release from habituation at test, as evidenced by an increase in sucking Categorical speech perception in infants rate the moment they were presented with the test stimulus. This finding suggests first, that, just The introduction of the high-amplitude sucking as adult listeners, infants this young perceive paradigm and the related techniques described speech sounds in categories, only detecting above has boosted research into infants’ per- differences between speech sounds that are lin- ceptual abilities and these efforts have been guistically meaningful. Second, it suggests that rewarded with a greatly increased understanding the boundary between the two phoneme cate- of how speech perception develops during the gories that are perceived as different from one first year of life. An important first finding was another coincides with the adult boundary. already briefly described above—that, just as adults, infants exhibit categorical perception of The study of Eimas et al. (1971) inspired speech sounds. Eimas and his collaborators many new ones testing other minimal contrasts (1971) presented their 1- and 4-month-old infant between speech sounds, including the two already participants, born of English-speaking parents, mentioned above: between plosives that only dif- with a number of machine-synthesized instances fer in their place of articulation (e.g., /p/ versus of the bilabial plosives /p/ and /b/ (with a vowel /t/, both voiceless; Eimas, 1974) and between added onto the critical plosive), manipulating the the /l/ and /r/ (Eimas, 1975). These studies have plosives’ VOT value. The VOT values included in revealed that infants can discriminate many their study were −20 ms, 0 ms, +20 ms, +40 ms, speech sound categories, including these two +60 ms, and +80 ms. The minus and plus signs further distinctions tested by Eimas and his indicate that the vocal cords start to vibrate collaborators and differences between vowel before and after the moment the air is released, categories (e.g., Trehub, 1973). Other discrimin- respectively. A VOT of 0 ms indicates that the ations appear more difficult for infants this release of the air and the onset of the vocal cords’ young, such as the minimal contrast between vibration coincide. Earlier studies had shown pairs of fricatives such as /v/ and /f/ or /z/ and /s/. that adult speakers of English identify bilabial plosives with VOT values below +25 ms as /b/ and Because of the tender age of the participants those with VOT values above +25 ms as /p/. In in these studies it has been suggested that the other words, the first three of the above six speech ability to perceive speech sounds in phonemic sounds are perceived as voiced /b/, the last three categories is innate. If correct, all infants should as voiceless /p/. initially be sensitive to the same phoneme boundaries irrespective of the language that The infants within each age group were surrounds them. Interestingly, the number of the divided in three subgroups. One subgroup was critical boundaries on a given speech dimension, familiarized with the +20 ms stimulus and, after and hence the number of phonemic categories, habituation, tested with the +40 ms stimulus. differs between languages. In a cross-linguistic The stimuli presented during habituation and test study of 11 natural languages, Lisker and thus fell on opposite sides of the adult phonemic Abramson (1964) showed that some languages boundary. The habituation and test stimuli cut up the VOT continuum into three categories, presented to a second subgroup both fell on the distinguishing between speech sounds with a same side of the phonemic boundary (−20 ms and “short voicing lag” (where the vocal cords start to vibrate very soon after the air is released), those with a “long voicing lag” (where they start to vibrate relatively late after air release), and

2. EARLY BILINGUALISM AND AGE EFFECTS 19 those with a “voicing lead” (where the vocal cords native Spanish contrast between VOT −20 ms and start to vibrate relatively long before air release). VOT +20 ms should also have been perceived. The The authors showed that some of the examined fact that it was not indicates that in acquiring languages exploit all three of these VOT values, the relevant phoneme contrasts of the native whereas others only exploit two of them. In the language there is a role for learning through latter type of language the intermediate short language exposure as well. A third conclusion to voicing lag always occurred, accompanied by draw from the above results is that 6.5 months of either the long voicing lag or the voicing lead (the exposure to the ambient language (here Spanish) latter occurring, for instance, in Tai, where a does not suffice for (all) of the native language’s boundary exists somewhere between VOT values contrasts to be acquired. of −60 ms and −20 ms, a contrast that is not dis- tinctive in English). In addition to a difference in The precise nature of the apparently innate the number of phoneme boundaries, the exact ability to perceive a number of phonemic con- locus of the boundary or boundaries differs trasts is a source of dispute among scientists. between languages. For instance, whereas the Some hold the view that the phenomenon of boundary between voiceless and voiced con- categorical perception only applies to language sonants lies near a VOT value of +25 ms in and not to other forms of perception and that it English, in Spanish it lies somewhere between the is unique to humans, and does not apply to other values of −20 ms and +20 ms. In other words, a species (e.g., Eimas & Miller, 1991). This view voiceless plosive is spoken with a longer VOT in instantiates the more general view that the English than in Spanish. linguistic ability of Homo sapiens is enabled by a special innate language faculty that is inde- Using the heart-rate paradigm, Lasky et al. pendent of general cognition. Other scholars (1975) addressed the question of which contrasts assume that categorical perception is not unique on a VOT continuum ranging from −60 ms to +60 to language and the human race but results from ms Spanish infants, varying in age between 4 and general perceptual abilities of the auditory system 6.5 months, could detect. The results showed they of mammals. These scientists’ reason for taking could detect a difference between a VOT of −60 this stance is that categorical perception in ms and one of −20 ms, suggesting that they were humans has also been observed for non-speech sensitive to the Tai boundary. In addition, they sounds (Jusczyk, Pisoni, Walley, & Murray, could detect a difference between VOT values of 1980) and that non-human mammals such as +20 ms and +60 ms, demonstrating sensitivity chinchillas and macaques have been shown to to the English boundary between voiceless and perceive the same speech contrasts as humans voiced speech sounds. Interestingly, when the do (e.g., Kuhl & Miller, 1978). As argued by Kuhl habituation and test stimuli had VOT values of (1986, 2004), the match between the boundaries −20 ms and +20 ms, respectively, no increase in between phonetic categories in natural languages heartbeat was observed at test, suggesting that the on the one hand, and the way humans and difference, contrastive in the participants’ native non-humans categorize speech sounds on the language Spanish, could not be detected yet. other, is not accidental. She assumed that during This combination of findings suggests, first, that their development the world’s languages have the ability to discriminate between contrastive primarily exploited the auditory discontinuities speech sounds is indeed innate, because how to which the mammalian auditory system is espe- else can it be explained that contrasts that are cially sensitive, showing a tendency to place the linguistically relevant in some languages (here, phoneme boundaries at the points of highest Tai and English) but that do not occur in the auditory sensitivity. In agreement with this view, native language (here Spanish) can be perceived at and getting back to the studies of Eimas et al. all? Second, it suggests that this predisposition (1971) and Lasky et al. (1975) presented above, does not extend to all of the contrasts occurring Gerken (1994) hypothesized that on the voicing in natural languages, because if it would, the continuum English has exploited the peak of

20 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS highest auditory sensitivity of the auditory process in children growing up bilingual may, in system in creating the phoneme boundary addition to delaying the learning process, lead to between voiced and voiceless plosives (e.g., /b/ two systems that differ from the corresponding versus /p/) and that this is why infants in both an systems of monolingual speakers. Alternatively, English and a Spanish environment are sensitive the two emerging systems may resemble the to the English boundary. For one reason or other, corresponding monolingual speakers’ systems during the development of Spanish this peak of but interact with one another during speech per- highest sensitivity was not used to create the ception and production (see pp. 363–365 for evi- boundary between voiced and voiceless plosives dence from production studies). Early (and late) but, instead, a point of lesser sensitivity. This is sequential bilingualism confronts the special the reason why 4- and 6.5-month-old Spanish problem that non-native phonetic contrasts that infants do not yet perceive the Spanish contrast. were initially perceived but subsequently lost may Further exposure to Spanish is required for them have to be relearned because they are meaningful to discover the critical boundary. in the second language. The above discussion leads to two questions A substantial amount of research has that have engaged researchers in this area of study addressed the first two of the above questions. In after the discovery of the phenomenon of the next section I will first outline the major categorical perception and its constraints: First, results of this work and briefly discuss some of do humans continue to perceive phonetic dis- the mechanisms that have been proposed to tinctions that are contrastive in some languages account for the perceptual changes involved. but not in the ambient language, and if not, at Next, I will address the third question, discussing what age does the ability to perceive these con- the little bilingual work there is in more detail. trasts decline? For example, are English adults still sensitive to the Tai boundary on the VOT The unlearning of non-native phoneme continuum despite the fact that it is not distinctive contrasts and the learning of native in English? Second, how long does it take humans phoneme contrasts to develop sensitivity for contrasts that are lin- guistically meaningful in the ambient language The collected studies show that the unlearning of but that are not innate? For example, when do non-native phoneme contrasts and the learning Spanish infants become sensitive to the boundary of native phoneme contrasts that are not per- between VOTs of −20 ms and +20 ms, which they ceived from the outset go hand in hand, and that do not yet perceive at 6.5 months of age but will many of the perceptual changes involved occur have to learn at some point in order to come to during the first year of life. Table 2.1 presents perceive and produce Spanish properly? the various developmental steps encompassed in the “universal language timeline” of speech A third question is particularly relevant in perception development as envisioned by Kuhl view of our special interest in bilingualism: Does (2004). According to this timeline infants under the learning of contrasts that are linguistically about 6 months of age can perceive phoneme meaningful in the ambient language(s) but not contrasts of all languages (but, as we have seen, innate, and the unlearning of innate contrasts not necessarily all contrasts of all languages). that are not functional in the ambient lan- After 6 months of age the perception of non- guage(s), differ between children growing up native contrasts declines and the ability to per- monolingual and those growing up bilingual? ceive contrasts specific to the native language Evidently, children growing up in a bilingual increases (e.g., Werker, Gilbert, Humphrey, & environment must come to acquire the sound Tees, 1981; Werker & Tees, 1984). The latter systems of both their native languages, a state of happens sooner for vowels than for consonants. affairs that may delay the acquisition process and The increase in native language perceptual ability especially so if the two systems exploit different not only involves the learning of contrasts not phoneme boundaries. The double acquisition

2. EARLY BILINGUALISM AND AGE EFFECTS 21 TABLE 2.1 mental pattern has been observed for infants growing up in different language environments Universal language timeline of speech perception and has been tested with a varied set of phoneme and speech production development contrasts (see Werker, 1989, for a review). Speech perception Speech production In addition to the points in time at which native and foreign phoneme contrasts start to 0–5½ months: 0–3 months: appear and disappear, respectively, Table 2.1 Infants discriminate Infants produce non- shows the times at which other perceptual abilities phonetic contrasts of speech sounds emerge—for instance, the sensitivity to the all languages ambient language’s phonotactics; that is, the 3 months: specific sequences of sounds that do and do not Infants produce vowel- occur in the language. Although the develop- like sounds mental change in infants’ speech perception is not yet completely understood, it is generally believed From 6 months: that the developing perceptual skills bootstrap Statistical learning language acquisition in general. The fact that (distributional landmarks in the development of speech pro- frequencies) and duction during the first year of life closely follow language-specific specific perceptual landmarks (see Table 2.1) perception for vowels supports this idea. A final point stressed by Kuhl (2004) and other authors (see e.g., Kuhl, Tsao, & 7 months: Liu, 2003; Sebastián-Gallés & Bosch, 2005) is Canonical babbling that the decline in infants’ sensitivity to non- native contrasts and the increase in sensitivity From 8 months: to native contrasts both result from perceptual Statistical learning reorganization processes caused by linguistic (transitional probabilities) experience. In other words, the decreasing sensi- and detection of typical tivity of infants to foreign phonetic contrasts also stress patterns in words indexes growth in phonetic development. From 9 months: Still, linguistic experience does not result in the Recognition of language- loss of discrimination ability for all non-native specific sound contrasts. The earliest study to demonstrate a combinations lasting ability to discriminate between particular pairs of non-native speech sounds examined the From 10 months: perception of dental versus lateral click conson- Language-specific ants in isiZulu, the native language spoken by speech production the Zulu people (Best, McRoberts, & Sithole, 1988). Both English-speaking adults and infants From 11 months: up to 14 months of age (the oldest infants tested) Decline in foreign detected the difference between these consonants. language consonant Attempts to explain the developmental change in perception and increase in perceiving non-native speech contrasts must take native language this variability into account. Best and McRoberts consonant perception (2003) contrasted three such accounts. One of them distinguishes between “fragile” and 12 months: “robust” contrasts (Burnham, 1986). Robust con- First words produced trasts involve distinctions that are acoustically salient and common across languages, whereas Adapted from Kuhl (2004). present initially but also the sharpening of the boundaries between the native phonetic cate- gories that are present from the outset (see e.g., Kuhl, Williams, Lacerda, Stevens, & Lindblom, 1992). By the time infants are about 11 months of age they have largely lost the ability to dis- criminate between pairs of contrastive foreign speech sounds, especially if they concern con- sonants, behaving like the adult speech listeners in their language community. Discriminatory ability for foreign language vowel contrasts lasts longer (Polka & Werker, 1994). This develop-

22 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS fragile contrasts are non-salient and occur only lated to one and the same native category and, at rarely in the languages of the world. According to the same time, they are considered to be equally this view, the ability to discriminate robust con- good representatives of this category, the two trasts remains intact irrespective of whether they elements in the non-native contrastive pair are occur in the ambient language. In contrast, fragile perceived as similar to one another; in other contrasts can only be maintained if they occur in words, they cannot be distinguished from one the language environment. As noted by Best and another. If instead one element of the contrasting McRoberts, a problem with this account is that it pair is assimilated to one native phonemic is circular, because the notions of fragility and category whereas the other is assimilated to a robustness are hard to define independent of the different native category, discrimination is easy development of the behavioral responses to the and performance near ceiling. An intermediate contrasts in question (that is, independent of situation arises if the two elements in the non- whether or not the contrast continues to be native pair are assimilated to one and the same perceived, as suggested by the experimental native category but at the same time are not con- evidence). sidered as equally good exemplars of the native category. In this case discrimination ability lies in A second account that takes variability in the between that in the above two cases: Sometimes decline of foreign language phonetic contrasts the two phones in a non-native contrastive pair into consideration (Tees & Werker, 1984) holds are regarded as similar to one another; at other that non-native contrasts continue to be perceived times they are judged to be dissimilar. A special if they occur as allophones of one and the same feature of PAM is that it defines perceptual phoneme in the language environment, but are similarity in terms of the way movements of the lost if they do not. Best and McRoberts (2003) articulatory apparatus shape the speech signal. queried this account on the basis of counter- Specifically, it is assumed that sounds that are evidence. Recall that in Spanish the boundary produced by similar movements of the articula- between voiced and voiceless plosives (for tory organs are perceived as similar. instance, between /b/ and /p/) lies between VOTs of −20 ms and +20 ms whereas in English it lies Employing the visual fixation habituation around a VOT of +25 ms. Yet Spanish voiced procedure described earlier (p. 15), Best and and voiceless plosives do occur in English, but McRoberts (2003) examined the predictions of as allophones of one and the same (voiced) PAM regarding the perception of a set of non- phoneme. Despite their occurrence as allophones, native consonant contrasts (occurring in isiZulu English listeners have been shown to experience and in Tigrinya, a language spoken in Ethiopia) problems in discriminating between pairs of by 6–8- and 10–12-month-old English-learning Spanish voiced and voiceless plosives. infants. The specific hypothesis tested was that the ability to discriminate non-native contrasts The third account of why the sensitivity to remains intact if the two contrastive speech non-native contrasts is variable across such con- sounds are produced by different articulatory trasts is framed within the perceptual assimilation organs (e.g., lips, larynx, and velum) but that it model (PAM) developed by Best (e.g., Best, 1994; declines if one and the same articulatory organ Best et al., 1988). The central premise of the is involved in producing them. The results con- model, from which it derives its name, is that firmed this hypothesis by showing that infants “mature listeners have a strong tendency to per- in both age groups could detect the difference ceptually assimilate non-native phones to the between contrastive non-native sounds produced native phonemes they perceive as most similar” by different articulatory organs and that dis- (Best & McRoberts, 2003, p. 186). The model crimination of these contrasts was equally good predicts that the discrimination of non-native in both groups of infants. In contrast, the contrasts depends on the way the two elements in younger group outperformed the older group on a contrasting pair are assimilated to the extant the contrasts that involved a single articulatory native phonemic categories. If both are assimi-

2. EARLY BILINGUALISM AND AGE EFFECTS 23 organ, suggesting a decline in performance with couple of studies testing the perceptual abilities aging. of adults have shown evidence of non- discrimination of non-native contrasts when Apart from once again demonstrating that behavioral testing methods were used, whereas the decline in discriminating between non-native the same participants demonstrated the ability to contrastive speech sounds is not a uniform still perceive these contrasts when ERPs served as phenomenon affecting all non-native contrasts the measurement technique. Conversely, studies equally, these results also support PAM’s tenet in which the perception of new speech sounds is that speech perception and speech learning being trained show evidence of speech sound exploits articulatory information in addition to learning in the ERPs earlier on during training acoustic information. As assumed by the authors, than behavioral measures do. In other words, attending to the articulatory gestures produced ERP measurements suggest that the insensitivity by native speakers provides infants (with intact to non-native contrasts sets in later, and the sensi- eyesight) with an additional, visual, source of tivity to new sounds earlier, than previously information next to the acoustic information thought on the basis of just the behavioral during speech learning. This hypothesis is sup- data. Accordingly, the authors hypothesized that ported by the well-known finding that infants are the above infant studies, which all employed sensitive to the congruence or incongruence behavioral measures, might have underestimated between talking faces on the one hand and the longevity of the ability to perceive non-native audible speech signals on the other (e.g., Kuhl contrasts and overestimated the time it takes to & Meltzoff, 1982; Rosenblum, Schmuckler, & learn not-innate native contrasts. Johnson, 1997). It is also consistent with the observation that 4- and 6-month-old infants from To test this possibility, Rivera-Gaxiola et al. monolingual homes can discriminate between (2005) examined infants’ discrimination of native their native language and a non-native language and non-native speech contrasts by means of the just by looking at silent video clips of speakers ERP technique. The participants in this study reciting sentences in two different languages were American infants tested twice, at 7 months (Weikum et al., 2007). At 8 months they have lost and 11 months of age. At both moments in time this ability, while bilingual-to-be infants tested on they were presented with (many instances of) their two native languages exhibit it at both 6 and three consonant-vowel syllables differing in VOT 8 months (they were not tested at 4 months). Of and perceived as either /da/ or /ta/, the syllables’ course, lip reading in the deaf also corroborates VOT values being −24 ms, +12 ms, and +46 ms. the hypothesis. As we have seen before, the boundary between voiced and voiceless consonants differs between Best and McRoberts (2003) thus demonstrated Spanish and English: It falls in between VOTs of that the developmental decline in perceiving −24 ms and +12 ms in Spanish but between VOTs foreign language contrasts is not a unitary of +12 ms and +46 ms in English. In other words, phenomenon but depends on specific acoustic/ the syllable with a VOT of +12 ms is voiceless in articulatory characteristics of the foreign sounds Spanish but voiced in English. The experimental involved. A study by Rivera-Gaxiola, Silva- procedure used concerned a “double-oddball” Pereyra, and Kuhl (2005) goes one step further paradigm in which the ambiguous +12 ms VOT by showing that the discrimination of specific syllable was presented as the “standard” stimulus non-native contrasts, which at first sight seems to 80% of the time and each of the remaining two be lost, still appears to be intact at some level. “deviant” stimuli were presented 10% of the Reviewing the literature, these authors observed time, the presentation order being randomized. that event-related potentials (ERPs) provide a The critical question was whether and at what more sensitive marker of speech perception age the infants could distinguish between the abilities than do the common behavioral standard on the one hand and one or both measures such as looking time, head turning, or deviants (the “oddballs”) on the other hand, as sucking rate. This is suggested by the fact that a

24 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS evidenced by differences between the brain nificant for both groups. It thus seems that responses to the standard and deviants. A dif- 11-month-olds have not yet lost the ability to ference between the ERPs to the +12 ms standard perceive the non-native contrast between Spanish and the +46 ms deviant would suggest sensitivity voiced and voiceless plosives. Exactly why the to the native (English) contrast. A difference brain responses of 11-month-olds may exhibit between the ERPs to the +12 ms standard and the this sensitivity in different ways across individual −24 ms deviant would indicate sensitivity to the participants is a question that needs further non-native (Spanish) contrast. research. In agreement with the behavioral data To summarize, loss of the ability to discrimin- reported in the literature, the analysis of the ate non-native contrasts is variable, depending on neural responses of all participants combined the acoustic and articulatory characteristics of suggested that at 7 months of age the infants dis- the contrasting speech sounds. In other words, criminated both the native and the non-native not all non-native phonetic contrasts are lost. contrast. This was suggested by a different ERP Furthermore, discrimination of non-native con- response to the standard on the one hand and trasts that appears to be lost when relatively both oddballs on the other hand in a time window insensitive behavioral measures are employed from 250 to 550 ms after stimulus onset. Further- may turn out to still be intact when ERP patterns more, a larger amplitude of the ERP to the are used to index discrimination ability. Both deviant contrastive in English (+46 ms) at 11 findings suggest good news for the sequential months of age than at 7 months suggested bilingual, who after first being immersed in an that between 7 and 11 months of age infants’ sen- exclusively or predominantly monolingual sitivity to native language contrasts increases. In environment for some time starts to receive input contrast, at 11 months of age nowhere over the in a second language and must come to learn the entire time window tested did the ERP signals to sound system of this new language. The residual the standard (+12 ms) and the non-contrastive sensitivity to non-native contrasts as detected by deviant (−24 ms) differ significantly from one means of the ERP methodology suggests that another, suggesting, again in agreement with the regaining sensitivity to contrasts apparently lost behavioral data, that by this age infants have lost does not have to start from scratch but may the ability to perceive the non-native contrast. build on dormant memory traces. A final finding However, further data analysis proved this con- emerging from the above studies is that learning clusion premature. During an examination of the the sound system of the ambient language and ERP signals of the individual participants it the unlearning of innate non-native contrasts became clear that the group as a whole was made develop in parallel, suggesting that the language up of two subgroups, showing different brain learners’ increasing insensitivity to non-native responses to the non-contrastive deviant. In phonetic contrasts is the direct consequence of about half of the infants this deviant aroused an their exposure to the native language and the early positive component in the ERP signal as phonetic knowledge that emerges from this compared with the standard, whereas the exposure. According to Rivera-Gaxiola et al. remainder of the infants showed a slightly later (2005), the process involved in both increased negative component for this deviant. Plausibly, sensitivity to the native language sound system the opposite brain responses to the non- and decreased sensitivity to foreign sounds is a contrastive deviant in these two subgroups gradually increasing “neural commitment” of cancelled one another out in the overall analysis. brain tissue to the speech patterns of the ambient An analysis of the data for the separate language. The corollary is a gradual decrease in subgroups of 11-month-olds confirmed this the availability of neural tissue to subserve non- hypothesis by showing that the difference between native speech. As we will see toward the end of the brain response to the standard and to the this chapter, the ubiquitous age of acquisition non-contrastive deviant was statistically sig- effects that are observed in adult second language

2. EARLY BILINGUALISM AND AGE EFFECTS 25 speakers and in all domains of language—not is subsequently exposed to L2 English will have only in phonology but also in morphology and to learn to discriminate between /l/ and /r/ again syntax—have also been attributed to this in order not to run into comprehension and pro- phenomenon of neural commitment to native duction problems in English (where, e.g., lip and language structures. rip or loom and room mean different things). Plausibly, the later the onset of second language Early bilingualism: Reversing exposure, and therefore the further advanced decreased phonetic sensitivity and native language phonetic development and neural doubling phonetic development commitment, the more effort it will take to restore the perceptual abilities required for proper per- As described above, exposure to the native ception and production of the new language. language and the native language phonetic structuring that comes with it lead to a decrease Kuhl et al. (2003) examined the conditions in discriminatory ability for non-native phonetic that enable a reversal of the decline in foreign categories. What are the consequences of this language phonetic discrimination during the first typical phonetic development for the child in year of life. In 12 laboratory sessions of 25 min- a sequential bilingual setting where it first utes spread out over a period of 4 weeks, they experiences pure monolingual input to which exposed a group of 9- to 10-month-old American some time later a second language is added? And infants, from families where only English was how does the phonetic development of children spoken, to Mandarin Chinese. During these growing up under circumstances of simultaneous sessions four different native Mandarin Chinese bilingualism compare to that of children growing speakers read children’s books to the infants up in a monolingual environment? Now that and played games with them. The number of research has revealed the general pattern of Mandarin syllables heard by each infant over monolingual phonetic development, the first these sessions amounted to over 25,000. This studies addressing these questions are beginning experimental set-up was designed to closely to appear. I will discuss these in the present mimic natural adult–infant interaction, the section. naturalness of the situation being enhanced further by the use of child directed speech. A Sequential bilingualism control group of American infants of the same age was exposed to 12 similarly designed sessions The consequences for early sequential bilinguals but held in English and with native speakers of of the reduced ability to discriminate non-native English interacting with the infants. speech sounds from about 11 months of age are obvious: By the time the child is introduced to Following training, Mandarin speech percep- the second language his or her perceptual ability tion ability of the two infant groups was tested, has started to shift from a language-general using a version of the head-turn procedure to a language-specific pattern of phonetic dis- described earlier (p. 16). As experimental stimuli crimination and the child has generally become two computer-synthesized speech sounds were less sensitive to non-native phonetic contrasts, chosen that are contrastive in Mandarin but not including contrasts that are distinctive in the in English. One of these sounds was an affricate, second language to which the child is now the second a fricative, and both were articulated exposed. To come to master the second language, in the alveolar-palatal region of the vocal tract. sensitivity to these contrasts will have to be Prior studies had shown that adult native restored. To illustrate, a Japanese-born child who, speakers of Mandarin show near-perfect dis- at 12 months of age, can no longer perceive the crimination of these two sounds whereas adult difference between /l/ and /r/ (which in Japanese native speakers of English perform poorly on are instances of the same phonetic category) and them. As usual, the head-turn procedure con- sisted of a habituation phase followed by a test phase. In the habituation phase the infants were

26 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS trained to turn their heads when they heard a The results clearly indicate that the foreign change from a repeated background sound (the language intervention to the infants in the fricative) to the target sound (the affricate). A test experimental group had reversed their relative phase then followed in which the number of head insensitivity to this Mandarin contrast: The turns upon a change in speech sound served as the infants in the Mandarin training group responded dependent variable, each head turn indicating with more head turns to a change of input than that the change in input was perceived. The those in the control group exposed to English question of interest was whether the infants who during training, and in fact were as sensitive to had been exposed to native Mandarin in the prior this contrast as the infants of comparable age training sessions would exhibit heightened dis- who had never been exposed to any other crimination of the Mandarin contrastive pair as language than Mandarin. compared to the control group of infants trained in English. The results are shown in Figure 2.3, A second experiment used the same head-turn which also shows the results of a group of native testing procedure and tested the same two speech Mandarin Chinese infants of the same age tested sounds but this time the Mandarin training in their home country (Taiwan). sessions did not involve natural live adult– child interaction but the presentation of either Mandarin Chinese speech discrimination in American 9- to 10-month-olds after exposure to Mandarin Chinese or American English in 12 sessions of natural live adult–child interaction (the two left bars) or to Mandarin Chinese in 12 non-interactive audio-visual or audio-only sessions (the middle two bars). The performance scores indicate the percentage of head turns upon a change from a Mandarin affricate to a Mandarin fricative or vice versa. The performance of a control group of native Mandarin Chinese infants is also shown (the right bar). Adapted from Kuhl et al. (2003). Copyright 2003 National Academy of Sciences, USA.

2. EARLY BILINGUALISM AND AGE EFFECTS 27 audio-visual Mandarin materials (to one infant same stimuli, meaning that stimuli from the cen- group) or of audio-only Mandarin materials (to ter of the VOT continuum occurred relatively a second group). Interestingly, the performance often. The authors predicted that the infants of both these Mandarin training groups now familiarized by means of the bimodal distribution equaled that of the live English training group in pattern would develop two phonetic categories, the first experiment (see Figure 2.3). The investi- while those familiarized with the unimodal pat- gators concluded that a reversal of the typical tern would develop a single phonetic category. developmental decline in foreign language speech The results obtained in the test phase supported perception in infancy requires true-to-life social this prediction by showing that only infants in interaction. (See Bijeljac-Babic, Nassurally, Havy, the bimodal condition, in both age groups, dis- & Nazzi, 2009, for another demonstration of the criminated tokens from the endpoints of the efficacy of social interaction in early sequential continuum. foreign language learning.) It remains to be seen whether true-to-life social interaction is a neces- Simultaneous bilingual infants are exposed to sary requirement for learning all perceptual two speech systems that differ from one another aspects of foreign speech and for all sequential in a number of ways, among others with respect bilinguals. Plausibly the generally better cognitive to the distribution of the phonetic values of skills of older second language learners can the systems’ phonetic elements on the various compensate for the suboptimal learning context acoustic dimensions (such as the VOT in that an audio-visual or audio-only learning plosives). Maye et al.’s (2002) discovery of sensi- environment provides. tivity to the statistical distribution of speech sounds in the speech input as a mechanism for Simultaneous bilingualism learning phonetic categories and contrasts thus gives rise to the question of how infants growing A couple of recent studies have examined the up in a bilingual setting from birth come to development of native phonetic contrasts in master the speech systems of their two native infants growing up in a bilingual language languages, and how this development compares environment from birth. These studies focused on to that of infants exposed to just one language. the role of statistical distributional information Plausibly, because of the relatively complex lin- in the ambient languages in the development of guistic input, it takes longer for simultaneous phonetic discrimination. In a study testing 6- and bilingual-to-be infants to develop the two 8-month-old infants from monolingual English targeted phonetic systems than it takes mono- homes, Maye, Werker, and Gerken (2002) were lingual infants to develop their single phonetic among the first to show that infants can exploit system, and the developmental trajectories may statistical distributional information of speech differ between these two groups of learners. These sounds in the language input to build phonetic questions have been addressed in a small set of categories. In this study the researchers used the recent studies that together have examined the preferential looking procedure described earlier perception of a couple of vowel and consonant to first familiarize infants with eight stimuli on a pairs in Catalan–Spanish (Bosch & Sebastián- continuum from the voiced to voiceless stop Gallés, 2003a, 2003b) and French–English consonants /da/ and /ta/. The infants were (Sundara, Polka, & Molnar, 2008) bilingual divided over two familiarization conditions. One infants and matched monolingual control groups. group was presented with a bimodal distribution of all eight stimuli, which meant that during Using a version of the familiar head-turn familiarization stimuli near the endpoints of the procedure, Bosch and Sebastián-Gallés (2003b) continuum occurred more often than stimuli from examined the development in Catalan–Spanish the center of the continuum. The other group was infants of a vowel contrast that exists in Catalan presented with a unimodal distribution of the but not in Spanish. Two age groups were tested, 4-month-olds and 8-month-olds. This choice of age groups was motivated by the fact that a shift

28 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS from language-general to language-specific vowel vowels. The experimental stimuli were different perception has been shown to occur at around 6 instances of /e/ and /ε/ (embedded in the disyl- months of age in monolingual infants (see Table labic nonwords /deði/ and /dεði/), thus imitating 2.1). Comparable groups of 4- and 6-month-old the natural variation in the way phonemes are infants from Spanish and Catalan monolingual realized in actual speech as a consequence of environments were also tested so that the devel- variables such as individual speaker character- opmental pace and trajectory of the infants istics and speech rate. All /e/ and /ε/ tokens exposed to both languages from birth could be used (18 of each) are shown in Figure 2.4 in a evaluated against a monolingual standard. In all phonetic vowel space formed by the vowels’ first then, six infant groups participated, three per age and second formants (recall that formants are level. The vowel contrast under study concerned concentrations of acoustic energy in particular the vowels /e/ (as in bait) and /ε/ (as in bet). This frequency ranges in speech sounds). As can be vowel pair is contrastive in Catalan although seen, the first-formant frequencies of all tokens acoustically no clear demarcation exists between formed a continuum; that is, there was no clear manifestations of /e/ on the one hand and of acoustic boundary between the /e/ tokens on the /ε/ on the other hand. The Spanish phonetic one hand and the /ε/ tokens on the other hand. system only contains /e/, which acoustically over- During familiarization half of the participants laps with both Catalan /e/ and /ε/ and whose were presented with /e/ exemplars and the other prototypical exemplar is intermediate between half with /ε/ exemplars. In the subsequent test the prototypical exemplars of the two Catalan phase all were presented with “same” and First- and second-formant mean values (in Hz) for all /e/ and /e/ stimuli. Adapted from Bosch and Sebastián-Gallés (2003b). Copyright © 2003 SAGE Publications. Reprinted by permission of SAGE Publications.

2. EARLY BILINGUALISM AND AGE EFFECTS 29 “different” trials. On a same test trial the stimulus ference between same and different trials in the was a token of the same category (/e/ or /ε/) as opposite direction in the Spanish group was not presented during familiarization. On a different significant.) This suggests a lost sensitivity to a test trial the stimulus was a token from the other contrast that does not exist in the ambient category. language or, in other words, that all exemplars of both /e/ and /ε/ are perceived as realizations of a The authors predicted that all three groups of single phonetic category. Of special interest are 4-month-olds, including the Spanish monolin- the results observed for the bilingual 8-month- guals for which /e/ and /ε/ are not contrastive, olds. These show that, despite having been would perceive the contrast, because at this age exposed to Catalan at least part of the time, at 8 language-specific phonetic sensitivity has not yet months these infants have lost the ability to per- emerged. Because language-specific perception ceive the difference between exemplars of the one should have developed by 8 months, the Catalan or the other category, behaving like Spanish monolingual infants were predicted to have monolingual infants of the same age. This result remained sensitive to the contrast, still showing gives rise to the question of how long it takes discrimination at 8 months. In contrast, at 8 before sensitivity to the Catalan contrast is months of age the Spanish monolingual infants restored again. To answer this question the should have lost the ability to perceive the con- researchers tested a new group of Catalan–Span- trast because it is absent in the ambient language. ish bilingual-to-be infants, now at 12 months of age. The data for this group clearly suggest that Regarding the bilingual-to-be infants the discrimination ability has been restored at this age authors pondered that two factors might play a (see the upper part of Figure 2.5). role in discrimination ability: continued versus discontinued exposure to the Catalan /e/–/ε/ The authors concluded that due to the over- contrast, and the distributional overlap between lapping distributions of instances of Spanish Spanish /e/ and Catalan /e/ and /ε/. If only con- /e/ and Catalan /e/ and /ε/, bilinguals first develop tinued exposure matters, the bilingual-to-be a single extended phonetic category that includes 8-month-olds should have developed two separate all three vowels and therefore fail to discriminate phonetic categories and, thus, perceive the con- between them. They suggested that this one trast. After all, they had been exposed to Catalan category is ultimately separated into three from birth, albeit only part of the time. In (Catalan /e/ and /ε/ and Spanish /e/) as a result of contrast, the distributional overlap between either extended exposure to Spanish and Catalan Catalan /e/ and /ε/ and the acoustically inter- or of progressed lexical development. In two mediate Spanish /e/ might foster the emergence further Catalan–Spanish studies Bosch and of a single phonetic category in the bilingual Sebastián-Gallés (2003a, 2005) similarly showed infants and the ensuing inability to perceive the a differential developmental pattern for mono- contrast. The upper part of Figure 2.5 shows the lingual and bilingual infants, now testing the results of this experiment. Specifically, it shows participants’ ability to discriminate between the the average listening time for “same” and “differ- fricatives /s/ and /z/ and between the vowels /o/ ent” trials in the experiment’s test phase for all and /u/ (as in boat and boot), respectively. As with infant groups. the vowels /e/ and /ε/, the fricatives /s/ and /z/ are contrastive in Catalan but not in Spanish and As shown, in agreement with the predictions, instances of these phonemes show a distri- all three groups of 4-month-olds perceived the butional overlap. Instead, the vowels /o/ and /u/ contrast. This suggests that, for infants this age, are contrastive in both Spanish and Catalan, speech perceptual abilities are not yet influenced with their prototypical acoustic values differing by the specific language(s) they have been exposed somewhat between the two languages and, again, to. Also in agreement with expectation, the instances of the two phonemes exhibiting dis- Catalan 8-month-old monolinguals but not the tributional overlap. The fricative contrast was Spanish monolinguals this age discriminated between instances of /e/ and /ε/. (The small dif-

30 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS (a) Mean listening time to same and different (= switch) trials of Catalan and Spanish monolingual infants and Catalan– Spanish bilingual infants of different ages. Adapted from Bosch and Sebastián-Gallés (2003b). (b) Mean listening time to same and different (= switch) trials of French and English monolingual infants and French–English bilinguals of different ages. Adapted from Sundara et al. (2008) with permission from Elsevier. tested on 4½-month-old and 12-month-old age group was tested to find out when dis- monolingual Catalan, monolingual Spanish and crimination ability of this contrast is restored Catalan–Spanish bilingual infants. At 4½ months again in bilinguals. The /o/ and /u/ contrast both the Catalan and Spanish monolingual (Bosch & Sebastián-Gallés, 2005) was tested on infants and the bilingual infants could dis- 4- and 8-month-old monolingual and bilingual criminate /s/ and /z/ (if the contrast occurred in infants (the authors do not explicitly mention word-initial position). At 12 months, however, the whether the ambient language of the mono- Catalan monolinguals still could, whereas the lingual-to-be infants was Catalan or Spanish, monolingual Spanish and the bilingual infants so presumably both Spanish and Catalan mono- failed to do so. Unfortunately, no older bilingual linguals were included among the monolingual

2. EARLY BILINGUALISM AND AGE EFFECTS 31 group). At 4 months both the monolingual and acoustic overlap. Still, English adults, but not bilingual infants could discriminate /o/ and /u/, French adults, can distinguish between the but at 8 months only the monolinguals could do French dental /d/ and the English alveolar /d/. so. The authors mention that preliminary data (Sundara and her colleagues suggested that a rea- from a group of 12-month-old Catalan–Spanish son that English adults are sensitive to the distinc- infants show evidence of discrimination again, tion might be that French dental /d/ is per- so the same developmental pattern emerged as ceptually similar to English /ð/.) Importantly, with the /e/–/ε/ contrast: initial sensitivity to French dental /d/ and English alveolar /d/ occur the contrast, followed by loss of sensitivity (pre- very frequently in their respective languages. As sumably because a single phonemic category is hypothesized by the authors, thanks to the high temporarily formed), followed by restored frequency of occurrence of these speech sounds sensitivity. English–French bilingual-to-be infants exposed to both languages from birth might show the It thus appears that cross-language distri- same developmental path as monolingual English butional overlap of speech sounds delays the controls, despite the fact that the distributions of building of language-specific contrastive cate- French and English /d/ overlap. gories in bilinguals. Sundara et al. (2008) hypoth- esized that infants may be sensitive not only Employing the visual fixation habituation to distributional characteristics of the speech procedure, three groups of 6–8-month-olds and input but also to the frequency of occurrence of three groups of 10–12-month-olds were tested: phonetic elements. If so, a high frequency of monolingual French, monolingual English, and occurrence of particular phonetic elements might simultaneous English–French bilinguals. During offset the adverse effect of cross-language dis- habituation half of the infants of each of the tributional overlap on discrimination ability. In six groups heard repetitions of four different other words, bilingual and monolingual infants French /d/ tokens excised from words spoken might show the same developmental trajectory on by three different monolingual French speakers. contrastive pairs consisting of speech sounds that The other half heard repetitions of four English both occur frequently in speech. /d/ tokens excised from words spoken by three different monolingual English speakers. At test, The researchers tested this hypothesis by com- each participant heard some new /d/ tokens paring the ability of monolingual French, mono- from the language of familiarization and some lingual English, and bilingual French–English tokens from the other language. The question infants to discriminate between exemplars of of interest was whether dental French /d/ and French /d/ and English /d/ (presented as the onset alveolar English /d/ were discriminated at test, as in syllables that always contained the same evidenced by relatively long listening times for vowel). The average French and English /d/ differ trials involving a language switch as compared in their place of articulation. In French, /d/ is with same-language trials. Because language- produced “dentally”, that is, by putting the specific consonant perception has not yet tongue to the back of the upper teeth, whereas in developed at 6–8 months of age, the infants in all English it is realized by putting it somewhat three groups this age were expected to discrimin- further back in the mouth, on the so-called ate French and English /d/. The data, shown in “alveolar ridge”. These differences in place of the lower part of Figure 2.5, supported this pre- articulation are reflected in acoustic differences diction. Furthermore, monolingual English and between French and English /d/. The difference French 10–12-month-olds were expected to between dental /d/ and alveolar /d/ is phonemic behave in the same way as monolingual English in some languages but not in French or English: and French adults, respectively—the English All realizations of dental and alveolar /d/ in monolingual infants noticing the difference both French and English instantiate one and between French and English /d/ but the French the same phoneme /d/. Furthermore, exemplars monolinguals failing to do so. Again, the data of French /d/ and English /d/ show considerable

32 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS confirmed these predictions (see Figure 2.5, Lexical development lower part). Introduction Of special interest was the pattern to be observed for the bilingual 10–12-month-olds. If So far the discussion has primarily focused on cross-language distributional overlap on its own the way infants growing up in a monolingual determines the time course of phonetic develop- environment gradually attune to the phonetic ment, the bilingual 10–12-month-olds should fail system of their native language and, in parallel to to perceive the contrast, grouping the instances of this development, become less sensitive to foreign French and English /d/ into a single category. speech sounds and foreign phonetic contrasts. If, however, a high frequency of occurrence of In addition, a few studies were reviewed that particular speech sounds counteracts the effect addressed the question of how phonetic devel- of distributional overlap on discrimination, the opment in infants exposed to just one language bilingual 10–12-month-olds should behave like compares to that of infants that receive dual- their English monolingual peers and like English language input from birth or soon thereafter. This adults. This is in fact the pattern that emerged joint research has revealed important insights (Figure 2.5, lower right part). into the emergence of language-specific phonetic categories and contrasts in infants, monolingual In conclusion, the joint results of the above and bilingual. Given the fact that phonetic studies indicate that simultaneous bilingual categories are the building blocks of words, this infants develop some phonetic contrasts at the work pertains to lexical development as well. Yet same pace and following the same trajectory as it does so only indirectly, exploiting techniques their monolingual peers, but that the develop- and using stimulus materials especially developed ment of other contrasts is delayed and follows a to learn about the emergence of phonetic cate- different route. They furthermore suggest that gories and contrasts, not words. both cross-language distributional overlap of representatives of different phonetic categories Other researchers have addressed the question and their frequency of occurrence in actual of how infants acquire vocabulary more directly. speech determine whether or not bilinguals show The goal of some of them was to discover when a different development from matched mono- and how infants start to detect sound patterns linguals for a specific phonetic contrast. More that correspond to words in fluent speech, an generally, these data show that infants are sensi- ability that is a prerequisite for acquiring vocabu- tive to statistical distributional information lary. As mentioned in the introduction to this regarding phonetic units in the speech signal, chapter, word boundaries are typically not and exploit this source of information in learning marked by breaks in the speech signal and, con- the phonetic categories of their language(s). As versely, the acoustic information conveying a such, these results add to the growing body of single word may be interrupted by one or more evidence suggesting that statistical distributional pauses within the word. How then do listeners learning plays a pivotal role in language learning segment the speech stream into words, the in general, and challenging theories of language primary linguistic units to convey meaning? acquisition that assume language acquisition to For adults, the fact that they have a vocabulary be driven largely by innate linguistic universals presumably contributes greatly to solving the (see Saffran, 2003, for a review). So far statistical segmentation problem because they can monitor learning has not only been shown to play a role the speech input for familiar sound patterns; in phonological development but also to con- that is, for sound sequences that match the tribute to the development of the lexicon (e.g., phonological representations of words stored in Saffran, Aslin, & Newport, 1996a) and grammar their mental lexicon. This is not an option for (Gomez & Gerken, 1999; Saffran & Wilson, infants because they have no lexicon yet, or only a 2003). It is to some of the relevant evidence that I very rudimentary one. The evidence suggests that will now turn.

2. EARLY BILINGUALISM AND AGE EFFECTS 33 caregivers do not generally facilitate children’s of the first words around 12 months (see vocabulary acquisition by providing them with Table 2.1), the vocabulary of infants initially words clearly separated from their linguistic con- grows very slowly, with just one or two newly text: Woodward and Aslin (1990; in Jusczyk & produced words a week. Around 18 months a Aslin, 1995) have shown that even in situations in sudden steep growth in vocabulary, the so-called which mothers are explicitly asked to teach their “vocabulary spurt” or “lexical spurt”, is observed infants new vocabulary the vast majority of the (e.g., Benedict, 1979; Nelson, 1973). Nazzi and targeted words are not presented in isolation but Bertoncini (2003) hypothesized that this vocabu- in continuous speech. lary spurt reflects a qualitative shift from the associative to the referential mode of word What, then, are the cues to segmentation of acquisition, the latter allowing an increase in the speech input, and at what age and through cognitive economy and, thereby, a reduction what mechanisms can infants exploit these cues in cognitive load: If a particular sound pattern and start recognizing the sound patterns of words is connected to a category of objects, far fewer in connected speech? Obviously, recognizing a word–meaning pairings have to be learnt (namely, word’s sound pattern does not yet equal knowing one for the whole category of objects) than the word in question, nor does it imply that the when words are paired with separate objects. In infant has some awareness that the sound addition, the reduced cognitive load would allow patterns that he or she recognizes are linguistic- the infants to attend to the words’ phonological ally meaningful. A word can only be said to be forms more carefully than before, thus gradually known if a connection has been established replacing the phonetically underspecified word between its sound pattern—that is, its phono- forms by more precisely specified ones. logical form—and its meaning. The ability to link the form and meaning of words develops In the next sections I will discuss a number of later than the ability to recognize the sound pat- studies that examined when and how infants start terns of words in continuous speech. Further- to recognize the phonological forms of words in more, there is some evidence to suggest that the the speech stream and subsequently connect these pairing of words to their meanings consists of to meaning. In so doing I will move from a couple two successive stages during which different of monolingual and cross-language investigations learning mechanisms are operative and that result to the specific case of infants exposed to two in different types of word–meaning linkages languages. (Nazzi & Bertoncini, 2003). In the first stage an associative lexical acquisition mechanism Segmenting the speech stream links phonetically underspecified sound patterns to specific objects through the repeated co- Statistical bootstrapping. Employing the head- occurrence of sound pattern and object. In the turn procedure, Jusczyk and Aslin (1995) tried to second stage a referential lexical acquisition find out at what age infants start recognizing the mechanism pairs phonetically specified sound sound patterns of words in connected speech. patterns with categories of objects rather than The investigators first familiarized a group of single objects. Nazzi and Bertoncini regard the 7½-month-olds from American-English homes word–meaning connections resulting from the with two monosyllabic words by presenting both first, associative, stage as “proto-words” and of them repeatedly until some predetermined those resulting from the second, referential, stage familiarization criterion was attained. At test, as genuine words. Only after a word has been four passages, each consisting of six sentences, learned at the referential level is there an under- were presented. Two of these passages each con- standing that words refer to things and events and tained six repetitions of one of the two words that, to do so, these referents need not be present presented during familiarization. The remaining in the environment. two passages each contained six repetitions of two novel words, one novel word per passage. The It is well known that, following the emergence

34 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS authors reasoned that if the infants noticed the a statistical learning mechanism similar to the similarity between the words presented during one involved in the development of phonetic familiarization and testing, this should show in categories and contrasts (pp. 27–32), plausibly different listening times at test for the passages even the very same mechanism, plays a pivotal containing the familiar words and those contain- role in bootstrapping this particular aspect of ing the novel words. Such a result was indeed lexical development. They tested this idea by obtained: The average listening time for the pas- familiarizing a group of 8-month-olds from an sages containing familiar words was more than a American-English language environment with second longer than for the passages containing 2 minutes of fluent synthesized nonsensical unfamiliar words. Accordingly the authors con- speech. The speech stream consisted of four cluded that at 7½ months infants can isolate three-syllabic nonsense words (pabiku, tibudo, from fluent speech words they have before been golatu, and daropi) repeated in a random order familiarized with by just listening to them being (e.g., tibudodaropipabikutibudogolatudaropigolatu repeated in isolation. A replication of this . . .). It did not contain any pauses, stress differ- experiment with 6-month-olds showed equally ences between the syllables, nor any other pros- long listening times to passages containing odic cue signaling the boundaries between the familiar and unfamiliar words, suggesting that “words”. The only cue that pointed at the word the ability to detect familiar sound patterns in boundaries was the sequential (“transitional”) fluent speech develops some time between 6 and probabilities of the neighboring syllables, these 7½ months. always being 1.0 within words but much lower between words (.33). For instance, ti is always fol- As mentioned earlier, child directed speech lowed by bu, bu is always followed by do, but do does not generally consist of words presented in may be followed by go, da, or pa. At test the isolation, so the ability to match words first pre- infants were presented with repetitions of two of sented in isolation to these same words presented the four “words” presented during training (e.g., in fluent speech later may not help infants greatly pabiku and tibudo) and with repetitions of two in learning to identify words in continuous syllable triads that were created by joining the speech. Jusczyk and Aslin (1995) therefore per- final syllable of one of the words to the first two formed a further experiment that more closely syllables of a second word (e.g., tudaro and pigola, mimics the segmentation problem faced by called “part-words” by the authors). Inspecting infants. A new group of 7½-month-olds served as the above speech stream the reader will discover participants. This time the order in which the that the infant has also encountered these part- infants encountered the passages and the isolated words while listening to the input, but much less words was reversed so that they were now often than the syllable triads imitating words. If at first familiarized with two words presented in test the infants responded differently to the words passages. At test they listened to four different than to the part-words, this would mean they had lists of isolated words, two lists consisting of extracted from the input the relative frequencies repetitions of either one of the two words that of the transitions between syllables (high within had occurred repeatedly in the passages and the words; low between words). A difference in listen- remaining two consisting of repetitions of either ing times for the words and part-words was one of the two novel words. The infants listened indeed obtained, supporting the idea that infants significantly longer to the former two lists than exploit a learning mechanism that detects sequen- to the latter. As concluded by the authors, this tial probabilities as a tool in developing a lexicon. finding constitutes strong evidence that 7½- month-olds have developed the ability to In a later study Saffran (2001) strengthened recognize the sound pattern of words in fluent this conclusion by showing that infants treat the speech. output of statistical speech learning—that is, the sequences of syllables they have become familiar But what exactly is the mechanism underlying with—as potential words and not simply as sound this ability? Saffran et al. (1996a) suggested that

2. EARLY BILINGUALISM AND AGE EFFECTS 35 patterns that have no special linguistic status (in transitional probabilities of syllables because the same way as they might come to recognize syllables are more salient perceptual units than other recurrent sound sequences such as tonal phonemes. Using the head-turn procedure, patterns in a musical composition). Only if Chambers and her colleagues first familiarized a infants assign a special linguistic status to the group of 16½-month-old infants from American- familiar syllabic patterns is it legitimate to con- English speaking homes with 25 CVC syllables clude that this statistical learning mechanism (C = consonant; V = vowel). These syllables bootstraps vocabulary acquisition. In this study instantiated a particular (artificial) set of phono- Saffran used the earlier procedure (Saffran et al., tactic constraints: Five consonants only occurred 1996a) to familiarize a group of American in syllable-initial position (e.g., /b/ and /m/) and 8-month-olds with four three-syllabic nonsense five different consonants only occurred in final words. During testing the “words” and “part- position (e.g., /p/ and /g/). At test the infants words” were not presented in isolation (as in heard novel CVC syllables that either respected Saffran et al.) but in either an English context the phonotactic constraints of the syllables pre- (I like my pabiku or What a nice tudaro) or in a sented during training (that is, their initial and nonsense context (Zy fike ny pabiku or Foo dray final consonants were members of the syllable- miff tudaro). Saffran argued that if the syllable initial and syllable-final set, respectively; e.g., sequences encountered during training were /bεg/) or that violated them (e.g., /pim/). Listening treated as mere syllable sequences instead of times were longer for the illegal syllables than potential words, the type of context in which they for the legal ones, suggesting that the participants are embedded at test, English or nonsense, should discriminated between the two types of stimuli. not affect the infants’ listening times during test- This discrimination ability was not based on ing. Conversely, if they were treated as potential familiarity or unfamiliarity at the level of the words, type of context should affect listening syllables as wholes because all of the syllables time. The results indicated that type of context presented at test were new. It therefore seemed did indeed matter: A significant listening time legitimate to conclude that during training the difference between words like pabiku versus participants had discovered the phonotactic part-words like tudaro was obtained when they system underlying the training set. were embedded in English contexts but not when they occurred in nonsensical contexts. Other studies have shown that this sensitivity Accordingly, the author concluded that “. . . to sublexical phonotactic constraints is in fact infants are not just detecting the statistical already in place in infants much younger than properties of sound sequences. Instead, they are those tested by Chambers et al. (2003) and that using the statistical properties of sound sequences it may coincide with infants’ sensitivity, at 8 to acquire linguistic knowledge: possible words in months, to sequential probabilities of syllable- their native language” (Saffran, 2001, p. 165). sized speech segments (Saffran et al., 1996a; see above) and with their ability to isolate familiar Whereas the Saffran studies showed that monosyllabic word forms from fluent speech infants are sensitive to the sequential probabilities (Jusczyk & Aslin, 1995). Jusczyk, Friederici, of speech segments the size of syllables, Wessels, Svenkerud, and Jusczyk (1993) presented Chambers, Onishi, and Fisher (2003) gathered American and Dutch infants of about 9 months evidence that infants are sensitive to the sequen- of age with a series of Dutch and English word tial probability of separate phonemes. This lists. Each list contained words in one language sensitivity is required for language learners to only and all of the presented words were abstract come to master their language’s phonotactics; and had a low frequency of occurrence, so that it that is, the system of permissible phoneme was unlikely any of them was familiar to any of sequences and their positional constraints. the participants (examples of the English words Acquiring this type of sequential constraints pre- used are kudos, aglow, butane, and scutcheon). sumably poses a greater challenge than learning The majority of the English words violated the