South and Southeast Asian Psycholinguistics 9781107503540, 9781107017764

Citation preview

South and Southeast Asian Psycholinguistics

A large body of knowledge has accumulated in recent years on the cognitive processes underlying language, much of which comes from studies of Indo-European languages, in particular English. This ground-breaking volume explores the languages of South and Southeast Asia, which differ significantly from Indo-European languages in their grammar, lexicon and spoken forms. South and Southeast Asian Psycholinguistics raises new questions in psycholinguistics and enables readers to re-evaluate previous models in light of new research. With 36 chapters divided into three parts – Language acquisition, Language processing and Language and brain – it examines contemporary topics alongside new findings in areas such as first and second language acquisition, the development of literacy, the diagnosis of language and reading disorders, and the relationships between language, brain, culture and cognition. It will be invaluable to all those interested in the languages of South and Southeast Asia, as well as to psychologists, linguists, educationalists, speech therapists and neuroscientists. h e a t h e r w i n s k e l is a Senior Lecturer in Psychology at Southern Cross University, Australia. p r a k a s h p a d a k a n n a y a is Professor of Psychology at the University of Mysore, India.

South and Southeast Asian Psycholinguistics Edited by

Heather Winskel and Prakash Padakannaya

University Printing House, Cambridge CB2 8BS, United Kingdom Published in the United States of America by Cambridge University Press, New York Cambridge University Press is part of the University of Cambridge. It furthers the University’s mission by disseminating knowledge in the pursuit of education, learning and research at the highest international levels of excellence. www.cambridge.org Information on this title: www.cambridge.org/9781107017764 © Cambridge University Press 2014 This publication is in copyright. Subject to statutory exception and to the provisions of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published 2014 Printing in the United Kingdom by TJ International Ltd. Padstow Cornwall A catalogue record for this publication is available from the British Library Library of Congress Cataloguing in Publication data South and Southeast Asian psycholinguistics / Edited by Heather Winskel ; Prakash Padakannaya. pages cm Includes bibliographical references and index. ISBN 978-1-107-01776-4 (Hardback) 1. Psycholinguistics–South Asia. 2. Psycholinguistics–Southeastern Asia. 3. South Asia–Languages–Psycholinguistics. 4. Southeast Asia–Languages– Psycholinguistics. I. Winskel, Heather. II. Padakannaya, Prakash. P37.45.S68S68 2013 495.010 9–dc23 2013014906 ISBN 978-1-107-01776-4 Hardback Cambridge University Press has no responsibility for the persistence or accuracy of URLs for external or third-party internet websites referred to in this publication, and does not guarantee that any content on such websites is, or will remain, accurate or appropriate.

Contents

List of figures List of tables List of contributors Introduction h ea t he r w i ns k el I

page ix xiv xvii 1

Language acquisition

(i)

Spoken language

1

Studying language acquisition cross-linguistically s a bi n e s to l l a n d e le n a l i e ve n

19

2

Infant-directed speech: social and linguistic pathways in tonal and non-tonal languages christine kitamura

36

3

Pragmatic development of Mandarin-speaking young children: focus on communicative acts between children and their mothers j i n g z h ou

45

4

Referential forms in Thai children’s narratives t h ee r ap o rn ra t i ta m k u l

52

5

The acquisition of tense and aspect y a s u h i r o s hi r a i

60

6

The acquisition of Malay numeral classifiers k h a z r i y a t i s al e hu d d i n

71

7

The acquisition of Vietnamese numeral classifiers j en n i e t ra n

79

v

vi

Contents

8

An overview of the acquisition of Malay wh-questions no r h ai da a m an

89

9

Marking plurals: the acquisition of nominal number inflection in Marathi s h a l m a l e e pi t a l e a n d va i j a y a n t h i m . sa r m a

99

10

Issues in the acquisition of Tamil verb morphology va ij ay a n th i m . s ar m a

110

11

Fast mapping of novel words in bi/multilinguals vi s h n u k . k . n a i r , su n i l k u m a r r av i , s a p n a bh a t a n d s hy a m a l a k . c he n ga p p a

124

12

Studies on the acquisition of morphology and syntax among Malay children in Malaysia: issues, challenges and needs r o g a ya h a . r a z a k

133

Issues in developing grammatical assessment tools in Chinese and Malay for speech and language therapy l i x i a n j i n , r og a y a h a . r a za k , j a n e t w r i g ht an d jo h n s on g

145

13

(ii)

Written language

14

Reading and reading acquisition in European languages br i a n b y rn e , st e f a n s a m ue l s s o n an d ric h ard k. o ls on

159

15

Learning to read and write in Thai he a th e r w in s ke l

171

16

Learning to read and write in Malaysian/Indonesian: a transparent alphabetic orthography he a t h e r w i n s ke l a n d l a y w a h le e

179

17

Literacy in Kannada, an alphasyllabic orthography r. mal a te sh a josh i

184

18

Reading in Tamil: a more alphabetic and less syllabic akshara-based orthography b h u v an e s h w a r i b . a n d p r a k a s h pa d ak a n n ay a

192

Akshara–syllable mappings in Bengali: a language-specific skill for reading s h r ut i s i r ca r an d so n al i n ag

202

19

Contents

20

Diversity in bilingual children’s spelling skill development: the case of Singapore s u s a n r i c ka r d l i o w

vii

212

II

Language processing

21

Tones and voice registers arthur s. abramson

223

22

How to compare tones n a n x u ra t ta n a s o n e, vi r g i n i e a t ti n a , b en j a w a n k a s i s o p a an d de n i s b u rn h am

233

23

Studying sentence generation during scene-viewing in Hindi with eye-tracking ramesh mishra

247

Thai-specific and general reading processes in developing and skilled Thai readers j e e s u n kim a n d c h r i s d a v i s

256

Eye movement guidance in reading unspaced text in Thai and Chinese j i e- li t sai

265

24

25

26

27

Southeast Asian writing systems: a challenge to current models of visual information processing in reading r o n an re i l ly

272

Preferred Argument Structure and Thai varieties of English: evidence of cognitive processing limitations? t h o m h u eb n er

285

28

Cross-language perception of word-final stops k i m i k o t s uk a d a

297

29

Uncovering bilingual memory representations w i ns to n d . g oh , l i d i a su a´ re z a n d k e l l y ye o

305

30

Eye movements and reading in the alphasyllabic scripts of South and Southeast Asia h e a t h e r w i n s k e l , pr ak a s h pa d a k an n a ya a n d a p ar n a p a nd e y

315

viii

Contents

III

Language and brain

31

Aphasia to imaging: the neurolinguistic endeavor as it reflects on South and Southeast Asian languages l o r ai ne k. o b l e r a nd a va n t hi n i r an j a n p a p l i k a r

32

Neural bases of lexical tones j a c k s on t. g an d o ur a nd a na n t ha n a r a y a n kr is hn a n

33

Hemispheric asymmetry in word recognition for a right-to-left script: the case of Urdu c h a i t r a ra o , j y o t s n a v a i d a n d hs i n - c hi n c he n

34

35

36

The Declarative Procedural model of language: a new framework for studying the non-inflecting languages of Southeast Asia? tomasin a oh

329 339

350

362

Language-mixing in bilingual aphasia: an Indian perspective s a p n a bh a t a n d s hy a m a la ch e n ga p p a

372

The relationship between language and cognition he a t h e r w i n s ke l a n d s u da p or n lu k s a n e e y a na w i n

381

References Index

389 458

Figures

Figure 2.1 Model showing how infant-directed speech is modified according to the cultural background of the mother, and developmental forces in the infant. page 39 Figure 2.2 Mean F0 (top panel) and pitch range (bottom panel) for English and Thai at birth, 3, 6, 9 and 12 months, and in AD speech. 42 Figure 2.3 Mean F0 (top panel) and pitch range (bottom panel) in speech to Thai and English boys and girls at birth, 3, 6, 9 and 12 months, and in AD speech. 43 Figure 3.1 A comparison of three core types of social interchange in Chinese and American children. 47 Figure 6.1 Classification of Malay numeral classifiers (adapted from Salehuddin & Winskel, 2008). 73 Figure 6.2 The different types of responses made by children in the counting production task (Salehuddin & Winskel, 2009, p. 304). 74 Figure 10.1 Production across different person/number/gender in the corpus. 116 Figure 10.2 Default versus real agreement. 117 Figure 10.3 Overt NPs versus pro. 119 Figure 12.1 Verb phrase developmental patterns among young Malay children. 142 Figure 12.2 Development of level of utterances according to stages of LARSP. 143 Figure 13.1 The MCUS score sheet. 151 Figure 13.2 MCUS scores for 1 year 3 month to 1 year 5 months old Mandarin-speaking children in the Free Conversation task. 152 Figure 13.3 MCUS scores for 4 year 1 month to 4 year 11 months old Mandarin-speaking children in the Free Conversation task. 153 ix

x

List of figures

Figure 19.1 Sample of reading errors associated with akshara– syllable mismatch. Figure 19.2 Children’s success rate in syllable processing (Panel A) and phoneme processing (Panel B) on simple and complex syllables. Figure 19.3 Phoneme deletion errors in CCVC non-words. Figure 21.1 For the range from 50 to 500 Hz, the non-linear relation between Hertz and semitones. From Nolan (2003) with permission. Figure 21.2 Electroglottography of a Mon speaker. Upper graph: Two glottal pulses from /klan/ “naughty” in Breathy ¨ voice. Lower graph: Two glottal pulses from / klan/ “lick” in Clear (Modal) voice. From A.S. Abramson, G. Ramsay and L. Luangthongkum (unpublished data). Figure 21.3 The five tones of Standard Thai for much of the twentieth century. Adapted from Abramson (1962). Figure 22.1 Tone triangles plotted along F0 onset and offset dimensions in 3 months, 6 months and 9 months IDS and ADS for 3–6–9 cohort, and 6 months, 9 months and 12 months IDS and ADS for 6–9–12 cohort. Figure 22.2 Tone trajectories over time of Thai children’s tone productions. Figure 22.3 Mean tone triangle areas (and standard errors) of Thai children’s tone productions over six developmental age periods. Figure 22.4 Tone differentiation plots across age showing the distance between each Thai tone and all others combined. Figure 22.5 Tone ellipses for Mandarin, Thai, Cantonese for three tone space representations, F0 onset/F0 offset, F0 onset/F0 velocity, and F0 onset/F0 acceleration, with mean Bhattacharyya distances across all tone pairs in bold. Figure 23.1 Sample pictures used in the experiment in different conditions. Figure 23.2 Proportion of fixation to the different types of pictures for children and adults from the onset of the display till 8000 ms. Figure 23.3 Mean fixations to the verb and subject regions for the different picture types for children and adults. Figure 24.1 The top left panel shows Thai consonants grouped by similar form. Note that although Thai dictionaries list

204

208 210

225

226 229

239 241

242

242

245 251

252 253

List of figures

xi

44 consonant symbols, ฃ and ฅ are obsolete and have been replaced by ข and ค. The bottom panel shows 18 vowel symbols (for which 32 vowel forms can be constructed by combining the vowels with three consonants). Vowels and their components can be positioned before, after, below or above initial consonant(s) and these positions can also be combined. Note, here the consonant letter อ is used as a place-holder to indicate the vowel position; the consonants used to compose vowels are ย, ว and อ. The top right panel shows an example of how consonants and vowels can be written in various horizontal and vertical orthographic positions. 258 Figure 26.1 A schematic summary of the E-Z Reader model. Starting at the bottom, the current fixation feeds information to the lexical processing and saccade programming module, which operate in parallel. A saccade is initiated after a low-level assessment of a word’s familiarity. This can be overridden if the shift in attention also results in a successful ‘familiarity’ check, thus accounting for the phenomenon of word skipping. Note, however, that saccades are triggered by lexical processing. 279 Figure 26.2 A schematic summary of the SWIFT model. In the SWIFT model lexical processing occurs within a four-word attentional gradient. Saccadic programmes are initiated autonomously, by a timing mechanism, so as to maintain a mean rate of eye movements. The inhibitory link allows word identification to extend the duration of the current fixation (via increasing the duration of the time interval between saccades) if the word being fixated is difficult to process. 281 Figure 26.3 The Glenmore model has a connectionist architecture and comprises three main components: (1) an interactive activation network that is responsible for identifying words; (2) a saliency map that selects saccade targets; and (3) the saccade generator. Activation of the input units is propagated forward to the letter and saliency units so as to identity and localize the individual letters in the 30-unit input array. Letter activation is then spread to the word units (which provide top–down modulation of the letter units), the saliency units and a fixate centre unit.

xii

List of figures

Figure 28.1 Figure 28.2 Figure 28.3 Figure 29.1 Figure 29.2 Figure 29.3 Figure 32.1

Figure 32.2

Figure 32.3

A saccade is initiated to the target location that corresponds to the most active saliency unit whenever the activation of the fixate centre unit reaches a certain threshold. The top panel gives an overview of the architecture, the bottom panel a sample configuration for a specific fixation. Mean discrimination scores for Thai and English stimuli by six groups of listeners. The brackets enclose ± 1 SE. Mean discrimination scores for English stimuli by four groups of listeners. The brackets enclose ± 1 SE. Mean discrimination scores for Thai stimuli by four groups of listeners. The brackets enclose ± 1 SE. Proportion of foil intrusions across conditions in Sua´rez and Goh (2007). Pattern of spread of activation in the cross-language conditions. Proportion of foil intrusions across conditions in Yeo (2007). Positron emission tomography (PET) images show increased activity in the left anterior insular cortex when Chinese natives discriminate pitch patterns embedded in Mandarin words (top panel), but in the homologous area of the RH for those embedded in English words (bottom panel). In contrast, English speakers’ activity is circumscribed to the RH regardless of lexical function. (Adapted from Journal of Neuroscience, 24(41), 2004, 9157, with permission from Society of Neuroscience.) Activation in within-category deviant vs. across-category deviant contrasts elicited from a tonal continuum ranging from the Mandarin high rising to falling tone. Regions of activity are shown for within-category > across-category (panel A, STG; panel B, right STG) and across-category > within-category deviants (panel A, MTG; panel B, left MTG). STG, superior temporal gyrus; MTG, middle temporal gyrus. (Adapted from PLoS One, 6(6), 2011.) Peak amplitude and latency of MMN and P3a show that pitch contour and pitch height are important dimensions used in early processing of Cantonese tones. MMNs were larger in tonal pairs that differ greatly in initial pitch height (height–large, height–small).

282 301 302 303 310 313 313

342

343

List of figures

In contrast, pitch contour influenced the latency of P3a (contour–early, contour–late). FCZ, frontal–central electrode recording site. (Adapted from Neuroscience Letters, 487(3), 2011, 270, with permission from Elsevier Press.) Figure 32.4 Discriminant analysis of pitch strength indicates that moderate rising pitch is important for distinguishing tone language from non-tone language speakers. (Adapted from Journal of Neurolinguistics, 23(1), 2010, 89, with permission from Elsevier Press.) Figure 32.5 Comparisons of spectral f0 magnitudes reveal that pitch encoding is enhanced in musicians as compared to Chinese or non-musicians in the rapidly changing portion of Mandarin tone 2 (high rising) corresponding to the note B♭ of a discrete musical scale. (Adapted from Journal of Cognitive Neuroscience, 23(2), 2011, 431, with permission from MIT Press.) Figure 33.1 An illustration of the same word written in Hindi (Devanagari) and Urdu (Nastaliq) scripts.

xiii

346

347

348 353

Tables

Table 2.1 Percentage correct of lexical tone in ID speech at birth, 3, 6, 9, 12 months and in AD speech in the utterance-initial and utterance-final position page 37 Table 3.1 The development of communicative acts of Chinese young children at the age of 14 to 32 months 46 Table 4.1 Referential forms and discourse contexts 57 Table 5.1 Preference of verb semantics in the use of -te i-(ru) by children and caregivers 67 Table 5.2 Distribution of lexical aspect with -ko iss- by children 68 Table 5.3 Distribution of lexical aspect with -ko iss- by caregivers 68 Table 5.4 Use of kam0laŋ0 (progressive) and lε:w3 (perfective) markers by lexical aspect and age 69 Table 6.1 Malay shape-based numeral classifier acquisition based on the mean number of correct production responses across all children’s age groups (standard deviations are in parentheses) (Salehuddin & Winskel, 2009a) 75 Table 6.2 Malay shape-based numeral classifier acquisition based on the mean number of correct matching comprehension responses across all age groups (Salehuddin & Winskel, 2009b) 76 Table 7.1 The children in the longitudinal study 81 Table 7.2 The children in the cross-sectional study 82 Table 7.3 Proposed order of emergence of Vietnamese classifiers 84 Table 7.4 Errors 85 Table 8.1 Percent of correctly repeated responses 90 Table 8.2 Number of wh-questions in terms of wh-word and question structure by two Singapore Malay children and their mothers 91 Table 8.3 Proportion of in situ versus moved wh-questions across wh-words 92 xiv

List of tables

Table 8.4 Table Table Table Table Table Table Table Table Table Table Table

8.5 9.1 9.2 9.3 9.4 10.1 10.2 11.1 11.2 12.1 12.2

Table 13.1 Table 15.1 Table 15.2 Table 18.1 Table 19.1 Table 21.1 Table 22.1 Table 22.2 Table 22.3 Table 27.1 Table 27.2 Table 27.3 Table 27.4 Table 27.5 Table 27.6 Table 27.7

Proportion of in situ versus moved wh-questions across wh-words Means of embedded responses given by the children Noun classes in Marathi Results for testing the plural of real words Tokens for non-words Results for testing of plurals for non-words Match between predicate and subject case Imperative forms in Tamil Analysis of word errors in the L1 and L2 Analysis of word errors in the L1, L2 and L3 Pivot (P) and open class (O) structure in utterances Mean and standard deviation by age group for total MPLAT raw scores Available local SLT assessment tools in Malay and Mandarin Chinese Thai initial consonants (IPA symbols are in parentheses) Thai vowel expressions classified in terms of vowel combination types A comparison of the characteristics of a traditional alphabet and a syllabary Examples of children’s responses on non-word reading, syllable substitution and phoneme segmentation tasks List of Mon word pairs spoken four or five times each by four male native speakers Chao tone letters assignment for Cantonese syllable /fu/ (from Rose, 2000) Chao tone letters assignment for Thai syllable /kha/ Chao tone letters assignment for Mandarin syllable /ma/ (from Chao, 1948) Preferred Argument Structure constraints Non-proficient versus proficient speakers Number of transitive clauses with 0, 1, or 2 lexical core arguments – low- and high-proficiency speakers Distribution of lexical core arguments across A, S, and O roles for low- and high-proficiency informants Number and percentage of all arguments in A, S, and O roles that are lexical Number of verbs with 0, 1, or 2 new arguments – low- and high-proficiency speakers Number of transitive verbs with two new arguments

xv

94 97 101 106 106 108 117 120 130 130 136 141 148 172 173 193 207 232 234 234 234 286 290 291 292 292 293 293

xvi

List of tables

Table 27.8

Number and percentage of arguments in each role that are new Table 27.9 Number and percentage of all new arguments that appear in each grammatical role Table 27.10 Pronominal forms used by informants Table 29.1 Structure of the critical two-block trials in the short-term cued-recall paradigm Table 29.2 Manipulations of past studies using the short-term cued-recall paradigm Table 29.3 Manipulating phonological and visual similarity across foils and target fillers Table 29.4 Manipulating cross-language phonological similarity across targets and foils

294 294 295 306 307 309 312

Contributors

arthur s. abramson, Haskins Laboratories and University of Connecticut norhaida aman, English Language and Literature Academic Group, National Institute of Education, Nanyang Technological University virginie attina, Marcs Institute, University of Western Sydney sapna bhat, Guest Faculty, University of Southern Denmark bhuvaneshwari b., Department of Studies in Psychology, University of Mysore denis burnham, Marcs Institute, University of Western Sydney brian byrne, School of Behavioural, Cognitive and Social Sciences, University of New England hsin-chin chen, Department of Psychology, National Chung Cheng University shyamala k. chengappa, Department of Speech Language Pathology, All India Institute of Speech and Hearing chris davis, Marcs Institute, University of Western Sydney jackson t. gandour, Department of Speech Language Hearing Sciences, Purdue University winston d. goh, Department of Psychology, National University of Singapore thom huebner, Department of Linguistics and Language Development, San Jose State University lixian jin, Division of Speech and Language Therapy, De Montfort University, Leicester jing zhou, Faculty of Early Education, East China Normal University r. malatesha joshi, College of Education and Human Development, Texas A&M University xvii

xviii

List of contributors

benjawan kasisopa, Marcs Institute, University of Western Sydney jeesun kim, Marcs Institute, University of Western Sydney christine kitamura, School of Social Science and Psychology, University of Western Sydney ananthanarayan krishnan, Department of Speech Language Hearing Sciences, Purdue University lay wah lee, School of Educational Studies, Universiti Sains Malaysia elena lieven, Department of Developmental and Comparative Psychology, Max Planck Institute for Evolutionary Anthropology, and Max Planck Child Study Centre, School of Psychological Sciences, University of Manchester sudaporn luksaneeyanawin, Center for Research in Speech and Language Processing and Department of Linguistics, Faculty of Arts, Chulalongkorn University ramesh mishra, Centre for Neural and Cognitive Sciences, University of Hyderabad sonali nag, Centre for Reading and Language, Department of Psychology, University of York vishnu k. k. nair, ARC Centre of Excellence in Cognition and its Disorders, Department of Cognitive Science, Macquarie University loraine k. obler, Speech–Language–Hearing Sciences, Graduate School and University Center, City University of New York tomasina oh, Department of English Language and Literature, National University of Singapore richard k. olson, Department of Psychology, University of Colorado prakash padakannaya, Department of Psychology, University of Mysore aparna pandey, Department of Psychology, University of Mysore avanthi niranjan paplikar, Speech–Language–Hearing Sciences, Graduate School and University Center, City University of New York shalmalee pitale, Department of Humanities and Social Sciences, Indian Institute of Technology (ITTI) chaitra rao, National Brain Research Center, Manasar, New Delhi theeraporn ratitamkul, Department of Linguistics, Faculty of Arts, Chulalongkorn University

List of contributors

xix

nan xu rattanasone, Child Language Laboratory, Linguistics Department, Macquarie University sunil kumar ravi, Department of Speech Language Pathology, All India Institute of Speech and Hearing rogayah a. razak, Speech Sciences Program, School of Rehabilitational Sciences, Faculty of Health Sciences, National University of Malaysia ronan reilly, Department of Computer Science, National University of Ireland susan rickard liow, Division of Graduate Medical Studies, National University of Singapore khazriyati salehuddin, School of Language Studies and Linguistics, National University of Malaysia stefan samuelsson, Department of Behavioral Science, Linko¨ping University, Sweden vaijayanthi m. sarma, Department of Humanities and Social Sciences, Indian Institute of Technology (ITTI) yasuhiro shirai, Department of Linguistics, University of Pittsburgh shruti sircar, Department of Linguistics and Contemporary English, The English and Foreign Languages University, Hyderabad john song, Division of Speech and Language Therapy, De Montfort University, Leicester sabine stoll, Psycholinguistics Research Unit, University of Zurich lidia sua´rez, Department of Psychology, James Cook University, Singapore jennie tran, Department of Linguistics, University of Hawaii jie-li tsai, Department of Psychology, National Chengchi University kimiko tsukada, Department of International Studies, Macquarie University jyotsna vaid, Department of Psychology, Texas A&M University heather winskel, Department of Psychology, School of Health and Human Sciences, Southern Cross University janet wright, Division of Speech and Language Therapy, De Montfort University, Leicester kelly yeo, Department of Psychology, National University of Singapore

Introduction Heather Winskel

Traditionally, psycholinguistic research has predominantly focused on a small number of European languages. In more contemporary times, there has been a rapidly growing interest in investigating more diverse languages. By including a broader range of languages, we can build more comprehensive and representative universal models of psycholinguistic mechanisms and processes. The aim of the current volume is to contribute to this cross-linguistic endeavour and at the same time partially address this imbalance in research focus through reviewing some of the research conducted on the fascinating and varied languages of South and Southeast Asia. The people living in this large geographical region of South and Southeast Asia represent a large proportion of the world’s population and speak a large number of diverse languages. South Asia comprises Pakistan, India, Nepal, Bhutan, Bangladesh, Sri Lanka and the Maldives. These areas are home to several hundred languages belonging to the Indo-European, Dravidian, Sino-Tibetan and Austro-Asiatic families. The area of Southeast Asia encompasses Vietnam, Cambodia, Thailand, Laos, Burma, Malaysia, Indonesia and southern China with at least five language families (Tai-Kadai, Sino-Tibetan, Hmong-Mien, Austroasiatic, Austronesian) spoken in this area (Enfield, 2005, 2011). The rich diversity and significance of these languages is reflected in the number of languages spoken; for example, in India it is estimated as being 438, Indonesia 719, and Vietnam 106 (Ethnologue). Moreover, languages such as Hindi, Malay or Indonesian are spoken by millions of people. This volume on South and Southeast Asian psycholinguistics aims to highlight the rich diversity of languages and writing systems of this linguistically rich region. These languages are of growing interest to researchers as they differ significantly and in intriguing ways from the more commonly studied European languages in terms of grammar, lexicon, pragmatics or patterns in language usage, and written and spoken forms. Moreover, as a result of the historical genesis of psycholinguistics, most models and theories are based on a very limited number of languages and orthographies, predominantly European languages and Roman script; thus, it is essential to have data on a greater number of languages. As highlighted by Stoll and Lieven (Chapter 1), 1

2

Heather Winskel

‘we have only very rudimentary information, sometimes only focusing on one linguistic feature, from approximately 1 per cent of the languages spoken today and within this 1 per cent there is a major bias towards acquisition studies of a very few languages’. In order to delineate between what mechanisms and processes are universal or common across languages and what are shaped by the characteristics of the particular language, it is essential to investigate the acquisition and processing of a more representative and diverse range of languages. This volume was modelled on the very successful Handbook of East Asian Psycholinguistics series. However, in comparison to the languages covered by this series, namely Chinese, Japanese and Korean, much less research has been conducted and reached a wider international audience on the languages of South and Southeast Asia. The contributions in the current volume give a taste of the distinctive features of the languages of the South and Southeast Asian region. Even though this research represents only a tip of the psycholinguistic iceberg, it is important to acknowledge the significant contribution that the chapters in the current volume make to this enterprise. The volume covers topics and themes in first and second language acquisition, the development of reading and writing, diagnosis of language and literacy disorders in children and adults, and the relationship between language, brain, culture and cognition. It is hoped that this volume will form an invaluable resource to scholars and students interested in the languages of South and Southeast Asia as well as cognitive psychologists, linguists, educationalists, speech therapists and neuroscientists. An important goal of the current volume is to act as a catalyst to stimulate and encourage future research on these rich and diverse languages of South and Southeast Asia and their corresponding writing systems. First, we will provide a brief overview of some of the characteristics of the languages and writing systems of these two regions. Characteristics of the languages and writing systems of South and Southeast Asia The languages in the two geographic regions are very diverse in terms of both their linguistic characteristics and socio-cultural contexts. The Southeast Asian languages show striking convergence in terms of structure at all linguistic levels (Enfield, 2005, 2011; Matisoff, 1973, 2001). These languages are commonly analytic or isolating, thus, lack inflectional morphology (i.e. do not have agreement, case, gender/number/definiteness on noun phrases, tense-marking on verbs), have a tendency to be monosyllabic, have zero anaphora, rich inventories of sentence final particles, numeral classifiers and verb serialisation, and favour topic comment structure (Enfield, 2011). In addition, lexical tone is an important feature of many of these languages.

Introduction

3

In marked contrast, Dravidian and Indo-Aryan languages have synthetic morphology, that is, several pieces of grammatical information are found within a word. Thus, these languages inflect for tense, gender and number, whereas most Southeast Asian languages lack this information and do not inflect at all. In place of inflectional morphology, the Southeast Asian languages typically utilise separate functor words or lexemes. The rich scripts and writing systems of this region offer extremely fascinating opportunities for psycholinguistic research. Scripts of this region include Devanagari, Bengali, Urdu, Gujarati, Oriya, Kannada, Tamil, Telugu, Malayalam, Sinhala, Javanese, Thai, Lao, Khmer, Myanmar, Vietnamese and Tibetan. Many of the scripts of South and Southeast Asia share many features and characteristics as they are historically related and descendants of the ancient Brahmi script. They have been termed alphasyllabaries as they have hybrid characteristics of both alphabetic and syllabic scripts. Some of these scripts are used to write a diverse array of languages, for example Devanagari is used to write Hindi, Marathi and many other Indian languages. A common feature of many Indian scripts is akshara, which is basically an orthographic unit that represents sound at the level of the syllable, as well as marking constituent phonemes (see Joshi, Chapter 17; Bhuvaneshwari & Padakannaya, Chapter 18; Sircar & Nag, Chapter 19). Some of the Southeast Asian scripts do not have inter-word spaces (e.g. Thai, Lao, Khmer and Myanamar) and represent lexical tone orthographically (e.g. Thai, Lao, Vietnamese and Myanamar). Scope and structure of the book There are three sections to this volume. Part I: Language acquisition, which has two subsections – (i) Spoken language and (ii) Written language, Part II: Language processing and Part III: Language and brain. Some of the chapters are assigned somewhat arbitrarily to one of these sections, yet in reality, they cut across sections, disciplines and themes. A number of recurring themes occur in this volume. One important recurring theme is bilingualism and multilingualism. Many children in this region are growing up in multilingual environments, learning to speak two or more languages and also learning to read and write in more than one language. Prime examples of multilingual and multicultural environments are Singapore and India. This is reflected in the chapters of the current volume. In multilingual Singapore, Susan Rickard Liow (Chapter 20) focuses on spelling development in bilingual children and Winston Goh and colleagues (Chapter 29) home in on the representation of languages in the memory of bilinguals and biscriptals. Obler and Paplikar (Chapter 31) review case and group studies conducted on multilingual individuals with aphasia in South

4

Heather Winskel

and Southeast Asia. In India, Nair, Ravi, Bhat and Chengappa (Chapter 11) discuss the fast mapping of novel words in bilingual and multilingual children and Bhat and Chengappa (Chapter 35) review research conducted on bilingual aphasia in this multilingual milieu. An additional recurring theme is the importance of developing diagnostic language assessment instruments in the local language(s) spoken in the South and Southeast Asian region (Razak, Chapter 12; Jin, Razak, Wright & Song, Chapter 13; Nair et al., Chapter 11; Joshi, Chapter 17). The early identification of children with language, reading and communication problems, using appropriately designed assessment instruments in the child’s native language, is crucial so that timely intervention programmes can be implemented. Chinese has had a huge impact in the Southeast Asian region. It is a widely spoken and significant language in Southeast Asia. The important role that Chinese plays in the Southeast Asian region is reflected in the chapters by Jin et al. (Chapter 13) in Malaysia and Rickard Liow (Chapter 20), Goh, et al. (Chapter 29) and Oh (Chapter 34) in Singapore as well as the chapter on discourse pragmatics in Mandarin Chinese by Zhou (Chapter 3). Tone languages represent a large proportion of the spoken languages of the world (Yip, 2002), and yet lexical tone is an understudied feature in psycholinguistic research. In the current volume Kitamura (Chapter 2), Abramson (Chapter 21), Xu Rattanasone, Attina, Kasisopa and Burnham (Chapter 22) and Gandour and Krishnan (Chapter 32) all review research conducted on lexical tone and tonal languages. Psycholinguistic methods Psycholinguistic research, as represented by the current volume, includes a range of methodologies and techniques, naturalistic observations and case studies, as well as more experimental methodologies, which all play an invaluable role in increasing our understanding of psycholinguistic mechanisms and processes of these languages. There have been rapid advancements in the development of the new technologies available to psycholinguists. This volume includes chapters that highlight the use of brain imaging and eye-tracking technologies to gain greater insights and understanding of the psycholinguistic and neural correlates of behaviour. Part I: Language acquisition (i)

Spoken language

Even though the languages of the world pose a diverse set of challenges for young children in terms of phonological, morphological and syntactic structure, typically developing children throughout the world rapidly become

Introduction

5

competent and effective communicators in their particular language. There are striking commonalities in language development across diverse languages, but also intriguing language-specific effects on language acquisition. In the introductory chapter to this section, Sabine Stoll and Elena Lieven (Chapter 1) discuss the major theoretical approaches and methods used in language acquisition research. They also examine the role of variation in terms of language structure, contexts of child-rearing, and individual differences. Subsequently, they focus on three prominent areas of research: early vocabulary acquisition, spatial cognition and acquisition of argument structure, which are used to address directly the question of whether children are geared by innate predispositions or whether they adapt to languagespecific structures and categories from an early age. As stated by Stoll and Lieven, ‘one of the major challenges is to disentangle linguistic and cultural variables’. One way this can be achieved is by studying typologically distinct and diverse languages. The complex interplay between language-specific features, culture and input in language acquisition is illustrated by the chapters by Kitamura (Chapter 2), Zhou (Chapter 3), Ratitamkul (Chapter 4) and Shirai (Chapter 5). Kitamura and Zhou both discuss the important dual role of caregiver–child interactions in language acquisition and socialisation of the child. Kitamura discusses the role of infant-directed (ID) speech in tonal and non-tonal languages. A unique blend of features characterises ID speech in a diverse number of languages. However, there are also cross-cultural variations that reflect culture-specific beliefs and values. For example, Australian English and Thai mothers adapt their speech differently depending on gender of the child. Zhou (Chapter 3) focuses on pragmatics or patterns of language use between Mandarin-speaking mothers and their children. This chapter highlights the influence that culture has on interaction patterns between caregivers and their children. There are shared commonalities in interaction patterns across cultures but also culture-specific patterns shaped by the cultural beliefs and goals of the caregivers. Telling stories or telling others about things that have happened to us is an integral part of everyday conversation in all cultures. In order to create a coherent narrative, a speaker needs to sequence events in a comprehensible order and refer appropriately to characters occurring in the narrative, while concurrently observing the discourse norms and conventions of the particular culture. Ratitamkul (Chapter 4) examines the use of referential forms by Thai children when telling the famous Frog Story (Mayer, 1967). She found that Thai children’s use of referential forms to create coherence in discourse appears relatively late (around 9 years old), conforming to research findings on other languages. Shirai (Chapter 5) reviews research on the acquisition of tense-aspect markers in some Asian languages: Japanese, Korean, Mandarin Chinese,

6

Heather Winskel

Vietnamese and Thai. The pattern of tense-aspect acquisition has been of great interest to researchers in relation to the issue of nature versus nurture (Shirai, 2009). As an alternative to Bickerton’s innate bioprogram, Shirai and colleagues have proposed that the pattern of acquisition observed can be accounted for by the distribution of tense-aspect morphology occurring in the input to the child in conjunction with the cognitive processes involved in children’s early category formation. Based on a review of the data from research conducted on these Asian languages, he concludes that input frequency gives a more plausible account than innate predisposition (i.e. a bioprogram) for the patterns in acquisition observed. Individual and task differences are also important additional factors to consider. In order to gain greater insights into the nature of universals and particulars of language acquisition, future research needs to further investigate the acquisition of tense-aspect systems in Asian languages using multiple tasks and methods. Numeral classifier systems are a common feature of Southeast Asian languages. They are typically used in counting or referring to objects. In numeral classifier languages, objects are categorised into classifier categories based on their physical, conceptual and functional properties. The categories used by different numeral classifier languages have both shared and idiosyncratic characteristics. The complex semantic nature of numeral classifier systems poses significant challenges to young language learners. In the current volume, the numeral classifier systems of Malay (Salehuddin, Chapter 6) and Vietnamese (Tran, Chapter 7) form an interesting comparison. They categorise everyday objects and items in both common and quite distinct ways. The acquisition of Malay numeral classifiers is a relatively delayed and prolonged process in comparison to Vietnamese and other classifier languages. This appears to be due to the lesser degree of obligatoriness of this grammatical category in Malay. Aman (Chapter 8) provides an account of the acquisition of wh-questions of children who speak colloquial Singapore Malay. Malay, in contrast to European languages such as English, German, French or Spanish, allows questions with both in-situ and moved wh-words. European languages lack the latter in-situ question form. Thus, this study presents an alternative perspective on the acquisition of wh-question forms. Marathi, an Indo-Aryan language, is the official language of the state of Maharashtra and is spoken by over 70 million people. Moreover, Marathi is a richly inflected language. Pitale and Sarma (Chapter 9) investigate the acquisition of inflectional morphology of plural forms in Marathi using Berko’s famous wug task. In Marathi, there is a complex interplay between grammatical gender (whether a noun is masculine, feminine or neuter) and the final segment of the noun in determining the appropriate plural form to be

Introduction

7

used. This presents interesting challenges to the young language learner. Results are found to parallel those on the acquisition of English and German and indicate that morphological rules are induced by children. Morphological acquisition is governed by both universal principles of acquisition (e.g. developmental sequence, regularity of affixation and conditioning) and by specific features of the actual target language (e.g. number of nominal classes or genders, and phonotactics). Tamil is a Dravidian language, primarily spoken in the Indian subcontinent as well as in Sri Lanka, Singapore and Malaysia and by emigrant communities in many other countries. Similar to Marathi, Tamil is a morphologically rich language. The functions that are expressed by the verbal complex are structurally and functionally highly complex and consequently pose significant challenges to language learners. Sarma (Chapter 10) traces the acquisition of this complex verbal morphological system by a young child. Verbs are the key determinants of argument structure, and hence, particularly important to investigate in the field of language acquisition. Nair et al. (Chapter 11) discuss the fast mapping of novel words in bilingual and multilingual children. Fast mapping is a term coined to explain how children acquire vocabulary so rapidly during their early developmental years. The traditional perspective is that fast mapping is a process that enables children to create lexical representations for the unfamiliar words they encounter. However, very few studies have investigated fast mapping skills in children growing up in bilingual and multilingual contexts. This research can increase our understanding of children’s word-learning skills in each of the languages spoken and how lexical representations are created and stored in the bilingual child’s mental lexicon. Razak (Chapter 12) reviews studies conducted on the acquisition of Malay language. There is a great need for further research investigating the language development of Malay-speaking children. This research can be used to determine typical developmental patterns and milestones in language development in Malay. These milestones of language development can then act as indicators to identify developmentally delayed and at-risk children, so that early intervention programs can be implemented. Following on from this theme, Jin et al. (Chapter 13) discuss the issues involved in developing language assessment tools in two of the languages spoken in multilingual Malaysia, Malay and Chinese. Ascertaining the baseline norms and age-related language capabilities of normally developing Malay- and Mandarin-speaking children is the first step in developing language assessment tools. The overall aim is to provide a comprehensive, normative account of the acquisition of Malay and Chinese, so that children with communication problems can be identified at an early stage.

8

Heather Winskel

(ii)

Written language

As an introductory chapter to this section, Brian Byrne and colleagues (Chapter 14) give a comprehensive review of research on learning to read in European languages. English, in particular, is often used as a benchmark for literacy research even though it can be considered to be an ‘outlier’ orthography due to its high degree of inconsistency or irregularity (Share, 2008). As European languages have historically played a seminal role in literacy research, there is an expectation that research on other languages will make comparisons with these well-studied languages and orthographies. As Byrne, Samuelsson and Olson point out, our perspective on reading and writing development could have been quite different if the focus of research had evolved around Chinese or Hindi, for example. The chapters that follow review reading and writing development in a range of languages: Thai (Winskel, Chapter 15), Malay/Indonesian (Winskel & Lee, Chapter 16), Kannada (Joshi, Chapter 17), Tamil (Bhuvaneshwari & Padakannaya, Chapter 18), Bengali (Sircar & Nag, Chapter 19), and spelling development in bilingual children in Singapore (Rickard Liow, Chapter 20). This region offers rich opportunities for research on literacy acquisition in different scripts and children learning to read in more than one language and script. Thai, a tonal language, has its own distinctive alphabetic script that shares some common characteristics with Indic writing systems, due to common origins. The particular challenges this distinctive orthography poses to beginning readers and writers of Thai is discussed by Winskel (Chapter 15). Thai also has syllabic characteristics as it has implicit vowels for some consonants. Consonants are written in a linear order, but vowels can be written above, below or to either side of the consonant as full letters or diacritics, and commonly combine across the syllable to produce a single vowel or dipthong. There is a high level of consistency of mapping between phonemes and graphemes but there are multigrapheme to phoneme correspondences; consequently, spelling development lags behind reading in Thai. Thai is predominantly a monosyllabic language with words with very similar spellings, often varying in just one letter or the tone of the syllable, which adds to the challenges of learning this particular orthography. A variety of the Malay language is spoken in four Southeast Asian countries, namely Indonesia, Malaysia, Singapore and Brunei (Prentice, 1987). Winskel and Lee (Chapter 16) review research on learning to read and write in Malaysian/Indonesian. Malaysian/Indonesian language provides an interesting case study as it uses the same Roman script as English, but in contrast has a high degree of orthographic transparency. Furthermore, the syllable is a highly salient unit as it is predominantly bi- and multisyllabic and has a simple syllable structure with clear syllable boundaries.

Introduction

9

Joshi (Chapter 17) describes the characteristics of Kannada language and its orthography. Current research in reading development and reading breakdown is reviewed. Kannada is a polysyllabic agglutinative language with numerous inflections. The script of Kannada is an alphasyllabary and thus lends an interesting contrast between alphabetic languages like English and syllabic languages like Japanese Kana. Based on current evidence, Kannada seems to resemble English more than Japanese Kana. Kannada orthography has a fairly consistent relationship between the written unit, akshara, and the sound. Even though the orthography is transparent, literacy acquisition is still a difficult task for various reasons. The complex alphasyllabic nature of the akshara of Kannada is a contributing factor. In addition, poor teaching methods, large class sizes and lack of books in school and at home exacerbate the problem. It is important that standardised assessment materials exploiting the nature of Kannada aksharas are developed. Bhuvaneshwari and Padakannaya (Chapter 18) discuss the distinctive characteristics of Tamil language and its orthography and the challenges it poses to young readers as well as the interesting comparisons that it makes with other orthographies. Tamil script is a derivative of Brahmi script and its southern version, Grantha script. All elements of Tamil akshara are written as in line characters in contrast to other akshara-based scripts where characters can occur above or below the main text line. Tamil also has a relatively opaque or deep orthography in contrast to other Indic scripts, as it has one-tomany mapping of grapheme to phonemes. In addition, Tamil is diglossic with a marked difference between colloquial spoken language and more formal spoken and written forms. These distinctive characteristics pose significant challenges to the novice reader. A proficient reader needs to refer to sentence context and prior lexical knowledge to effectively read Tamil. Sircar and Nag (Chapter 19) discuss the nature of mapping between akshara and the phonological syllables in Bengali. They examine the relative difficulty in word reading where akshara–syllable mismatches occur, and the ways in which akshara–syllable mappings impact phonological processing. Children’s responses on word reading, syllable and phoneme processing tasks indicate that different types of akshara–syllable mappings are read and processed in different ways. These findings are one example of how specific properties of the spoken language and the writing system interact and influence the process of learning to read. In the final chapter of this section, Rickard Liow (Chapter 20) investigates the processes that contribute to spelling development in English for three groups of bilingual children (English–Mandarin, Mandarin–English and Malay–English) in Singapore. A series of empirical studies explore the processes that contribute to spelling in English for these three subtypes of bilingual children who all follow the same curriculum. Early spellings are

10

Heather Winskel

speech-based (Jalil & Rickard Liow, 2008), and the three main languages (English, Mandarin and Malay) are dissimilar in terms of phonology as well as orthography. These differences in the characteristics of the languages spoken and scripts learned by the different subtypes of bilinguals, influence both phonological representation and spelling development. Part II: Language processing Lexical tone is a common characteristic of the languages of Southeast Asia and yet it is psycholinguistically an understudied feature. The languages of Southeast Asia can be categorised into tonal languages or voice-register languages as well as languages that fall into neither category. Arthur Abramson (Chapter 21) presents an overview of acoustic and physiological research on tones and voice registers in the languages of Southeast Asia with particular attention to studies of both production and perception. These languages include not only official national languages with their regional and social dialects but also the languages of numerous ethnic minorities. Examples of tonal languages are Thai, Lao, Vietnamese and Burmese and examples of voice-register languages are Chong, Suai, Kh’mu and Mon. Languages such as Malay, Indonesian and Cambodian, at least in their standard dialects, belong to neither category. Xu Rattanasone and colleagues (Chapter 22) continue this theme by examining a number of methods by which tones can be described and compared, ahead of three examples of research using variations of a relatively recent method, that of tone space mapping. Mishra (Chapter 23) investigates sentence generation during scene-viewing in Hindi using eye-tracking technology. Eye movements offer real-time data related to cognitive processing and are particularly useful in studying language processing. Mishra presents and discusses the results from a recent study of sentence generation in children and adults in Hindi. The findings are discussed in relation to theories of language production and the multisensory interaction between language and vision. Kim and Davis (Chapter 24) review some of their recent experimental findings with both developing and skilled Thai readers and make comparisons with results from English readers. They review findings from tests that range from a short-term priming technique that probes print-to-sound mapping to examining the metalinguistic skills associated with the awareness of lexical tone. Patterns of results were found that are consistent with those for English (problems in processing print-to-sound connections appear to be a common feature of reading difficulties) but also results emerged that are most likely due to the specifics of Thai. Tsai (Chapter 25) and Reilly (Chapter 26) both use eye-tracking technology to investigate eye movements when reading Thai and Chinese. Both Thai and

Introduction

11

Chinese do not have inter-word spaces. Tsai reviews some of the empirical studies that have investigated the function of inter-word spaces when reading Chinese and Thai using eye-tracking technology. Eye movement research on languages without inter-word spaces raises the issue of the nature of parafoveal word segmentation. It suggests that, in place of inter-word spaces, language-specific information is used to indicate word boundaries when processing parafoveal words. Reilly specifically focuses on the visual information processing aspects of reading and computational modelling of these processes. Current modelling efforts on Roman-derived alphabetic writing systems have limited the generality of the resulting models. English plays an important role in this geographical region as it does globally. Due to the effects of globalisation, it has become of paramount importance in many regions of South and Southeast Asia to learn English as a second or foreign language. Thom Huebner (Chapter 27) reports on the distribution of referential devices among proficient and non-proficient Thai speakers of English. Research on Thai speakers of English as a foreign language has been particularly sparse and often inaccessible to readers outside of Thailand. The fact that the Thai pronominal system does not inflect for gender, number and case may place a heavier cognitive burden on the Thai speaker of English as a foreign language to find the appropriate morphological form once the pragmatic status of a given referent is established. Learners’ first language (L1) or their previous linguistic experience has a profound influence on subsequent language learning (Strange, 1995). This prior experience has a greater influence in adults than children, particularly in the area of phonetics/phonology (Flege, 1995, 2003; Toda, 2003). Tsukada (Chapter 28) reviews some recent studies on the perception of word–finalstop place contrasts in English and Thai by adult native listeners of four Asian languages: Cantonese, Korean, Japanese and Vietnamese. The findings indicate that these listeners discriminated final stop contrasts more accurately in English, a language they were familiar with in contrast to Thai, a language unknown to them. Results suggest that exposure to native input and familiarity with speech materials at both phonological and phonetic levels is needed for accurate discrimination of sounds in an unknown language. Goh et al. (Chapter 29) investigate the nature of memory representations in alphabetic versus logographic languages, and the activation of memory representations in bilingual speakers (and in particular, bilingual speakers of both alphabetic and logographic languages). The current state of research in this area seems to suggest that while alphabetic languages are predominantly represented phonologically in memory, logographic languages are represented using both phonemic and visual codes which can be activated to different degrees depending on the task performed. Evidence indicates that memory representations of both languages in a bilingual speaker are

12

Heather Winskel

activated, even when only one language is used in a task, which can be a source of interference in specific situations. Winskel, Padakannaya and Pandey (Chapter 30) discuss eye movements and reading research on the alphasyllabic scripts of South and Southeast Asia. Cross-scriptal similarities and differences within this family of scripts are discussed as comparative studies of these typologically related scripts are potentially very informative. Some potentially fruitful areas of research are discussed with reference to some recent studies that have been conducted on Thai, a member of this family of scripts. Part III: Language and brain In the first chapter of this section, Obler and Paplikar (Chapter 31) review the development of methods and techniques, current issues and findings in neurolinguistics. They have three main foci: language localisation, aphasia in bilinguals and multilinguals, and language-specific factors. Advances in neuroimaging techniques in conjunction with focused case and group studies has enabled us to substantially increase our understanding of the subregions and networks within each hemisphere involved in language processing. The authors emphasise the valuable contribution that research conducted on the languages of South and Southeast Asia has made in improving our understanding of the neurolinguistic organisation involved in language processing in the brain. The languages and orthographies of South and Southeast Asia both share common characteristics, but also provide interesting languagespecific differences in relation to the more studied Western languages. They thus provide us with the opportunity to confirm universals of language processing and to further investigate language-specific effects on language organisation in the brain. In relation to language-specific features, Gandour and Krishnan (Chapter 32) focus on the neural bases of lexical tone processing. Empirical evidence on the perception of lexical tones is examined from a variety of experimental methods including auditory electrophysiology and functional brain imaging techniques. Data are adduced from Mandarin Chinese, Cantonese and Thai to illuminate the effect of language experience, phonetic features, phonological categories, levels of representation and stages of processing in different cortical and subcortical structures of the brain. They conclude that subcortical sensory encoding interacts with cognitive processing in the cerebral cortex to shape the response of the perceptual system as required by the structural properties of a particular language. In essence, a fuller understanding of the neurobiology of language can only emerge from consideration of general sensorimotor and cognitive processes in addition to those associated with language.

Introduction

13

Rao, Vaid and Chen (Chapter 33) examine hemispheric asymmetry in word recognition in Urdu, a right-to-left directional script. Urdu, an Indo-European language, is widely spoken in the north of India and is a national language of Pakistan. Urdu makes particularly interesting comparisons with Hindi, as the languages are closely related and share phonology, lexicon and grammar. However, interestingly, the two languages use quite distinct scripts. Urdu is written using an alphabetic Perso-Arabic script, whereas Hindi is written using an alphasyllabic script, Devanagari. In addition, Urdu is relatively opaque or non-transparent in terms of sound/symbol correspondences, whereas Hindi is highly transparent and written from left to right. By comparing readers of Urdu with readers of Hindi, the effects of these script differences on word recognition can be examined. The fact that Urdu (like Hebrew, Arabic and Persian) is read and written from right to left makes it an interesting test case for claims about the nature of left/right cerebral hemisphere differences in word recognition. Results indicate that while the left hemisphere is dominant in overall word recognition in Urdu, the right hemisphere of Urdu readers is also capable of additional processing of aspects of text. This right hemisphere processing of text contrasts with results found for English readers. Oh (Chapter 34) examines the Declarative Procedural (DP) model of language as a potential framework for studying the analytic, non-inflecting languages of Southeast Asia. This model has previously been applied to synthetic languages such as English, Italian (Walenski et al., 2009), Spanish (Eddington, 2009) and German (Smolka, Zwitserlood & Rosler, 2007) but not to analytic languages. In this chapter, the DP model is utilised to examine whether there is one or more than one default numeral classifier in Mandarin Chinese. Bhat and Chengappa (Chapter 35) review research conducted on language mixing in individuals with bilingual aphasia in the Indian context. In the Indian context, multilingualism is the norm and language mixing is highly prevalent. The authors highlight the differences in language-mixing patterns across societies. Differences in bilingualism between Western countries and India are discussed in order to understand the implications for languagemixing research in clinical bilingual populations in the Indian context. Finally, there has been much debate about the relationship between language and cognition. An intriguing question is whether the language we speak shapes or influences the way that we think or perceive of the world around, termed the Whorfian hypothesis or linguistic relativity debate. Winskel and Luksaneeyanawin (Chapter 36) point out that the rich and diverse array of languages spoken in the South and Southeast Asian region with their distinctive linguistic features offer an ideal opportunity to investigate this age-old question further.

Part I

Language acquisition

(i) Spoken language

1

Studying language acquisition cross-linguistically Sabine Stoll and Elena Lieven

Introduction One of the fundamental questions in the enterprise of cognitive science is what enables children to learn any of the approximately 7000 languages spoken today and how they cope with the extreme variation exhibited in the structures of these languages? On the one hand, there must be human-specific universals that allow a child to learn any language. On the other, we know from research on linguistic relativity that language influences the way we characterize and categorize the world. Thus the way children learn language sheds light on the question of how language and cognition interact. Prelinguistic children show a similar development of the abilities relevant to producing and understanding language (Callaghan et al., 2011). Some of these abilities are joint attention, pattern recognition, imitation and role reversal. They also show an approximately similar timetable for major landmarks such as babbling and first words. However, as soon as children start producing language, many aspects of development across languages and also, within languages, diversify radically due to the different demands on the learner exhibited by the structures of the language a child is learning and the cultural and social environments a child grows up in. All acquisition theories assume that language learning involves the interplay of biological and environmental factors but the nature of this interaction is hotly debated. To answer these theoretical issues, empirical acquisition research on a wide range of languages is mandatory. One of the main challenges is to explain how children are able to cope with such a wide variation both of structures and of cultural traditions and beliefs. There have been two somewhat different approaches to the question of variation. One started with Slobin’s cross-linguistic endeavor (Slobin, 1985a, 1992, 1997), where the main focus was on the different challenges posed by the grammars of typological diverse languages. A second approach is more anthropologically oriented and focuses on the role of cross-cultural differences and their consequences for language acquisition (Ochs & Schieffelin, 1984; Keller, 2007). These two major branches of comparative research run somewhat in 19

20

Sabine Stoll and Elena Lieven

parallel. As noted by Budwig and Chaudhary (1996, p. 461) a major problem is that ‘both approaches simultaneously covary typology and cultural setting’. Thus, one of the major challenges is to disentangle linguistic and cultural variables. The main difficulty in doing this is the lack of available data. We have only very rudimentary information, sometimes only focusing on one linguistic feature, from approximately 1 per cent of the languages spoken today and within this 1 per cent there is a major bias towards acquisition studies of a very few languages (Stoll, 2009; Lieven & Stoll, 2010). Most language families except for some European languages of the Indo-European family and some languages of East Asia (Chinese, Japanese, Korean) have been neglected and we know close to nothing about how most languages of the world are acquired and what challenges they pose for acquisition. Thus an important task for future research is to acquire information about languages that are typologically very different from those previously studied and to disentangle the social context of language learning from typology. In the following, we first briefly discuss the two major theoretical approaches in language acquisition research focusing on the way in which the roles of nature and nurture are characterized. Second, we present the methods used in acquisition studies and discuss their advantages and disadvantages. Subsequently, in the third section, we discuss the role of variation at several levels: language structure, contexts of child-rearing and between individuals. There are relatively few, but important, contributions from South and Southeast Asian languages concerning these issues and some of them will be presented below. In the fourth section, we focus in somewhat more detail on three major research areas, which directly address the question of whether children are geared by innate predispositions or whether they rather from early on adapt to language-specific structures and categories. The three areas we discuss are: (i) early vocabulary acquisition, (ii) spatial cognition, and (iii) acquisition of argument structure. Theories: the balance between nature and nurture All modern language-acquisition theories assume interplay between innate features that are specific to humans and learning. The main difference between the two major types of theories in the field lies in what they assume to be innate and how specific to language this is. Linguistic-nativist approaches to language acquisition (e.g., Chomsky, 1981a, 1986; Pinker, 1984, 1989) assume innate grammatical principles that enable children to learn any human language, and see themselves broadly as within the Chomskian tradition of generative grammar. However, researchers vary considerably as to which stage in the development of Chomskian theory they use as their theoretical base. Despite the fact that the newest version of the theory,

Studying language acquisition cross-linguistically

21

‘Minimalism’, contains only two principles (‘move alpha’ and ‘merge’) and, in even more recent accounts, recursion is claimed as the only uniquely human capacity relevant to language (Hauser, Chomsky & Fitch, 2002), the theory of principles and parameters (Chomsky, 1981b) is still probably the theoretical framework adopted by most child language researchers working within this tradition (Hyams, 1986; Valian, 1994; Wexler, 1998; Legate & Yang, 2007). In this theory, universal grammar consists of a set of principles which govern anaphora resolution, recursion, etc. and parameters which are set by the input language, such as head direction and pro-drop. These parameters are meant to account for variation between languages. The number of parameters proposed, their initial setting and the number of possible settings all vary between theorists. While this is, in principle, an elegant solution to the problem of language variation, the fact that the number of parameters proposed and their settings tends to increase with each new language studied has undermined their impact on researchers not a priori committed to this approach. The other major type of acquisition theory is the so-called usagebased or constructivist theory (Tomasello, 2003; Goldberg, 2006). In this type of theory, the assumption of innate components necessary to learn a language is restricted to general cognitive features, such as pattern finding, distributional learning, imitation, joint attention and symbolic representation, not all of which are necessarily species-specific. Other species are equipped with one or several of these abilities, but only in humans are all of them present and this probably constitutes a unique interplay in evolution and development. The focus of usage-based approaches is on identifying the strategies a child employs in extracting patterns of form–meaning mappings from what they hear. Unlike classic Chomskian approaches where the input is thought to play a relatively minimal role in setting the parameters, in usage-based theory input, in interaction with the child’s cognitive and social developmental stage, is seen as being critical. Correlations between the frequencies of particular forms and structures in the input and children’s language development are ubiquitous (for instance vocabulary composition, verb argument structures, complex sentences) but as we shall see below, factors other than input frequency also play an important role. Methods There are three main methods in acquisition research: (i) diary studies, (ii) recordings of natural speech, both cross-sectional and longitudinal, and (iii) experiments. The earliest studies of acquisition were diary studies and these are still a very useful research method (Leopold, 1949; Stern & Stern, 1965[1928]; Dromi, 1987; Mervis et al., 1992; Tomasello, 1992). Diary studies are only

22

Sabine Stoll and Elena Lieven

suitable for very young children who do not yet talk very much or for focusing on a narrowly defined issue (such as children’s first verbs: Naigles, Hoff & Vear, 2009; or the development of wh-questions: Rowland, 2007). In diary studies the goal is to record every instance of a feature in its specific context of occurrence. The advantage of this method is that it potentially provides a complete overview of all the situations and functions for which a child uses a given construction. In most cases such a study works only if the researcher is also the child’s primary caregiver, which might complicate the situation in terms of recording bias. Thus, caregivers might inadvertently focus on these constructions in their own speech thus influencing their children to use the construction earlier and more often. The other, more frequently used, method is the recording of natural speech. Since the middle of the last century with the advancement of technology more, and larger, corpora of spontaneous speech have been collected. This can be done cross-sectionally, i.e. several children are recorded for an hour or more at specific ages (e.g. age 2;0, 2;5, 3;0, etc.). If the samples at each age are large enough it is possible to extrapolate a description of developmental progress over a range of time, even though the same children are not being studied at each age. Longitudinal studies involve recordings of the same child at specific intervals over a period of several months or years. They allow monitoring the development of a specific child in its natural context. So far, there are studies on 34 languages freely available through the CHILDES program (http://childes.psy.cmu.edu) initiated by MacWhinney and Snow (1985). For South and Southeast Asian languages, two corpora are openly available. One corpus is of 18 Thai children of an age period between approximately 0;6 month to 2 years contributed by Chulalongkorn University (2004) to CHILDES and the other is a corpus of ten Indonesian learning children collected and contributed by Gil (2004). The Thai corpus, however, is not translated and glossed and the Indonesian corpus is only rudimentarily glossed. Further, there are large corpora on Tamil and Hindi collected by Narasimhan and one on an endangered South Asian language called Chintang (Sino-Tibetan, Kiranti, Eastern Nepal) collected by Stoll, Lieven and Bickel (DoBes Archive and www.spw.uzh.ch/clrp/acquisition.html). Thus, only a handful of the approximately 1500 South and Southeast Asian languages have acquisition corpora. The third type of method for studying language acquisition is experimental. There are a number of complicating factors to the use of experiments for cross-linguistic comparisons. Experiments introduce a cultural context in which the behaviour of a child is tested. To make sure it is the test variable that influences the child and not the cultural context, one needs to control for these cultural variables. Thus, for instance, in a culture that is not familiar with watching television or films, an experiment with such

Studying language acquisition cross-linguistically

23

stimuli might get results that are caused by this unfamiliarity rather than by an inability to use a given form in a specific context. Thus, to be able to compare results across different cultures one needs to be very careful in the choice of stimuli and experimental set-up. A famous cross-cultural experiment that has been applied to a wide variety of languages is the Frog Story experiment (Berman & Slobin, 1994; Stromqvist & Verhoeven, 2004) where children of various ages were presented with a picture story and their task was to tell the story to an interlocutor not familiar with the story. The main focus was on the expression of motion events, a feature which divides languages typologically. The experiment has also been conducted in Thai (Zlatev & Yangklang, 2003). However, as Berman and Slobin (1994) note, it is debatable whether a story in which one of the main protagonists is a pet frog is appropriate to all cultures. Variation in language acquisition A major challenge in researching how children learn languages is to take three important types of variation into account. Structural variation There are major differences in the structures of the different languages and language families. To make generalizations about language acquisition, we need to know how children learn as wide a range of different structures as possible. Children are able to learn any of these structures provided they grow up in the respective linguistic and cultural context. Both in the areas of South and Southeast Asia, there is huge variation in the features exhibited. Both areas are assumed to constitute a so-called Sprachbund, that is, the languages of a specific area belong to a variety of language families but due to contact they have assimilated strongly on various levels. Geographically, the area South Asia comprises Pakistan, India, Nepal, Bhutan and Bangladesh, Sri Lanka and the Maldives. These areas are home to several hundred languages belonging to the Indo-European, Dravidian, Sino-Tibetan and Austro-Asiatic families. In India alone 438 languages are spoken (Ethnologue). Languages vary in their number of speakers and their geographical extension. Some languages have only a few speakers left and are on the verge of extinction, others are spoken by millions of people distributed over large areas as, for instance, Hindi or Bhojpuri. On the one hand the linguistic features of the different language families vary strongly. On the other hand, South Asia has also been claimed to be a Sprachbund especially the Indo-Aryan and Dravidian languages. The languages of Nepal and the Tibeto-Burman languages of Northern India are

24

Sabine Stoll and Elena Lieven

left out in these discussions (Ebert, 2001, p. 1529). Ebert cites the following features that are commonly attributed to the South Asian Sprachbund: retroflex consonants, morphological causatives, OV word order, converbs compound verbs, dative subjects, oblique stems, reportative particles, etc. These features would lend themselves to a comparative study across language families in the Sprachbund, but hardly anything is known about their acquisition. A similar situation concerning common features is found in Mainland Southeast Asia where ‘language “genealogies”, have been dramatically permeated by socio-historical contact’ (Enfield, 2005, p. 181). Thus, languages that belong to very different families show similar traits due to contact. The area encompasses Vietnam, Cambodia, Thailand, Laos, Burma, parts of peninsula Malaysia and southern China with at least five language families (Mon-Khmer, Tai-Kadai, Hmong-Mien, Sino-Tibetan and Austronesian) spoken in this area (Enfield, 2005). Phonological structure in these languages usually comprises large vowel systems, they exhibit a monosyllabic structure and most of non-Khmer languages employ tone and/or phonation distinctions such as creaky or breathy voice. As Enfield summarizes, these languages are morphologically mainly analytic or isolating. Argument structure is not marked morphologically and inflectional categories such as tense, number and gender are absent (Enfield, 2005). Except for the Sino-Tibetan languages, which are predominantly verb-final, Mainland Southeast Asian languages exhibit VO order. Further, topic comment structure is favoured, which is in contrast to the structures we know from common European languages (see Enfield, 2005). The different language families of South and Southeast Asia nicely illustrate the wide variety of linguistic features a child has to be able to deal with in learning language. Phonological inventories vary a lot. In Dravidian, children have to learn a distinction between, dental, alveolar and retroflex, whereas most children growing up in Southeast Asia are confronted with tone and a phonological inventory in which the voice is used (e.g. creaky voice). Concerning morphology, Dravidian has synthetic morphology, that is, several pieces of grammatical information are found within a word. In most Southeast Asian languages grammatical information is not accumulated within a single word but rather distributed across words, that is, one word per information. Further, Dravidian languages and also Indo-Aryan languages inflect for tense, gender and number, whereas most Southeast Asian languages lack this information and do not inflect at all. Most South Asian languages show OV word order while very many Southeast Asian languages exhibit SVO. There are many more features in which these two linguistic areas vary and every child could in principle learn any of these languages.

Studying language acquisition cross-linguistically

25

Variation in the cultural context of child-rearing The way children are treated and raised is another important variable that differs widely across cultures. The majority of longitudinal studies have been conducted in the culturally rather homogeneous situation of urban, middleclass families in industrialized societies. The recordings consisted mainly in one-on-one interactions between mother and child irrespective of whether this is the typical situation in which the child spends her day. Many children in these societies do indeed spend the first years with their mother at home, though there are many others that go to daycare from early on. To learn how children behave in their typical contexts, an inclusion of more varied situations in these longitudinal studies even within these cultures would be necessary. Even if a child spends all her time with her mother in the early years, one cannot assume that the mother plays all day with the child, which is the context in most recordings. Thus, the contexts recorded and also the speech gathered may be rather context-specific. A further issue is that large numbers of children in the world grow up in very different social contexts. Many are raised in extended families or in villages where there is no single caregiver who looks after the child exclusively, at least not after around 6 months of age. These cultural differences, including the numbers of other people surrounding the child and the ways in which they interact, are factors of variation that are important to recognize (Keller, 2007). Another factor that influences the daily life of children and therefore the context of recording is the climate. In a cold climate, the movements of children are much more restricted than in a warm climate where children can roam around more easily and freely. In Chintang, children spend most of their days outside in peer groups. Adults in the vicinity look after the children and often talk to them but do not interact in the same way as one would for instance expect of an American mother, i.e. concentrated interaction and play (Stoll et al., 2012). Older children are held responsible for the well-being of younger children and children mostly interact with peers. This is less the case in industrialized countries even when children spend their days away from home, because childcare tends to be more formally organized and controlled by adults. A further variable that affects the contexts in which children grow up is the belief about what babies or young children can understand and how one should treat them (Gaskins, 2006). For instance, Ochs and Schieffelin (1984), in a study comparing children of three very different cultures (American, the Kaluli of Papua New Guinea, and Samoan) showed how important beliefs about child-rearing are and how different the input and social environment can be due to such different cultural practices. In the US American context, babies were treated as full conversational partners from early on. Greetings, questions, interpretations of the child’s sounds and

26

Sabine Stoll and Elena Lieven

reactions were a constant part of the interactions even with babies. Both the language and the environment in this cultural setting are most often adapted to the child and there are often rounds of turn taking even in these early proto-conversations. Every utterance or sound of the child is interpreted as intentional and usually is taken as part of a ‘conversation’. The environment is very much child-centred and adults constantly address the child. Ochs and Schieffelin isolate two main strategies in this environment: (i) a self-lowering strategy, meaning that the adult simplifies his or her language in talking to someone lower in social hierarchy, and (ii) a child-raising strategy, whereby the child is treated as being further developed than s/he is. This situation is radically different for both the Kaluli (Polynesian, Papua New Guinea) and the Samoans. Despite very different social structures in these two societies, what is common to both of them is that the child has to adapt to interlocutors rather than the other way around. Children are not recognized as conversational partners until certain hallmarks are reached and not much speech is addressed to them until they start to speak themselves. Thus, there are no protoconversations or overt interpretations of the child’s intention as is common in the Anglo-American and European contexts most typically studied and, in the case of Samoan, adults mostly do not address very small children at all. Recently, these issues have again become the subject of intensive research around the question of the influence of such child-rearing practices on the development of communication. On the one hand, extensive work by Keller and her associates has shown clear correlations between the ideologies and practices of child-rearing in different cultures (Keller, 2007). On the other, a series of cross-cultural experiments on the foundations of intention-reading and communication (e.g. pointing, imitation, helping and cooperation) suggest that, in the early stages, these develop somewhat independently of the cultural setting and that the effects of varying cultural practices and ideologies start to have a greater influence once the child reaches the toddler stage and beyond (Callaghan et al., 2011). If this is correct, the implication is that, within wide limits, young infants are on their own developmental timetable for emerging socio-cognitive and communicative abilities. This would explain both why there is a broad consistency in the timing of their emergence and why, within this, there is a range of individual differences. Individual variation Individual variation adds a substantial complication to the study of child language. We know from a large number of studies that children differ in their linguistic and also general cognitive and motor development to a significant degree and that, within quite wide limits, biological age is not a good predictor for development. Some children start to say their first words

Studying language acquisition cross-linguistically

27

between 7 to 9 months and are classified as ‘early talkers’, whereas ‘late talkers’ only start out with their first recognizable words at around 18 months or later. Norms can, of course, be established by collecting data from large numbers of children of the same age. Fenson et al. (1994) in a study with 659 American-English babies and 1130 preschoolers found a wide range of individual differences in both word comprehension and production, especially in the very early phases of acquisition. Eight-month-old babies in the 90th percentile understood around 75 words whereas the children in the lowest percentile understood none. Production is later but variation is similar. Eleven-month-old children in the highest percentile produced around 20 words, whereas the children in the lowest percentile did not produce any and this did not change much for children in the lowest percentile up to the age of 16 months (Fenson et al., 1994). These findings raise major issues for the interpretability of longitudinal studies based on a small number of children. Since longitudinal studies need a lot of resources and are extremely time-consuming, rarely more than a handful of children are recorded. This means that the children may be at very different stages of language development despite being of similar age, which reinforces the necessity of treating each child as its own control in terms of mapping the trajectory of language development. Given the fact there is so much individual variation, the first step before pooling children into groups thus needs to be an assessment of their individual development. Traditionally, mean length of utterance in either words or morphemes (MLU) has been used instead of age, to chart individual progress and as a measure of comparison between children. Leaving aside the problem of comparing MLUs across languages, which differ widely in their degree of analyticity and their richness of morphology, even within a particular language MLU is a very crude measure, which can vary widely for an individual child between sessions. It is strongly affected by the context in which the recording is taking place (for instance, many questions from the caregiver requiring short answers from the child will substantially lower the MLU). This variation between sessions will also be true of almost any phenomenon investigated. To determine developmental stages that are consistent and take account of variation between recording sessions Gries and Stoll (2009) have suggested a statistical clustering method. The individual development of either a particular phenomenon or the stages obtained by the different children can then be compared. The question that arises is why there are such extreme individual differences. Research suggests a number of contributing factors including the child’s socio-cognitive development, processing skills and environmental input. In addition, there are almost certainly personality differences that might, for instance, lead some children to be more outgoing and talkative and others to be shyer or more reflective. Although cognitive differences in

28

Sabine Stoll and Elena Lieven

learning capacity (such as memory skills, speed of pattern recognition and willingness of imitation) are to be expected and may well influence language development, they are hard to measure in very young children. But there is clear evidence of correlations between aspects of children’s language development and their non-verbal behaviour with objects (McCune-Nicolich, 1981; Gopnik & Meltzoff, 1986) and their socio-cognitive development (Carpenter, Nagell & Tomasello, 1998; Markus et al., 2000). We further know that the role of the environment plays a crucial role in the linguistic development of a child. Studies by Huttenlocher, Haight, Selzer and Lyons (1991) and Hart and Risley (1995) and many others on the input to Anglo-American children show that differences in the amount of speech children hear in a day are very great. For instance, Hart and Risley followed 42 families of children of about 13 months of age for 18 months on a monthly schedule. The results show that some children get an overall input of about 30 million words in their first three years, whereas children of lower socio-economic status only get about 10 million. There was a clear correlation between socio-economic status and the amount of speech that caregivers addressed to the children. Hart and Risley also showed that the amount of speech addressed to children has a direct influence on how much children talk, and on the development of vocabulary and the learning of specific grammatical structures. It is possible that these differences may be greater for children growing up in advanced industrial societies than in non-technological, rural societies where they are in large groups interacting with peers nearly all the time, but there is no research to date on this issue. After having looked at these different types of diversity in the way children are brought up and in their input, the question arises, how come all children learn language though at different speeds and maybe to different degrees of sophistication? Given all these differences, there may still be common traits in the input children receive (Lieven, 1994). We hypothesize that the following features are good candidates for universal features of child directed speech, whether exhibited by adults or other children. First, that the content of the conversation is framed in a way that the speaker gets across the message, i.e. the speaker frames the utterance to make sure the child understands. This often involves focusing on what the child is attending to; using short, explicit utterances with repetitions and expansions of what the child has said; and relatively simple semantic content. One of the key components is repetition and reframing to get the message across. Second, frequency in the input is known to be an important factor for acquisition and hence these repetitions have a double impact: they help the child to parse and understand the utterance and at the same time they have an impact on the memory of the child, which then enhances learning. Third, another probably very frequent feature of child-directed speech is the use of leading questions and

Studying language acquisition cross-linguistically

29

explanations of the context (see also Ochs & Schieffelin, 1984, who discuss these features for the Anglo-American, Kaluli and Samoan contexts). However, we should not expect that all cultures use the same combination of features and, given the difficulty of collecting data from a wide range of cultures, it is not surprising that there is almost no research on the causes of individual differences between children within more than a very few cultures. What gears language acquisition: universal cognitive preconditions or language-specific features? One of the most hotly debated questions in acquisition research is the question of whether language development is geared by innate linguistic principles (Chomsky, 1980; Pinker, 1989) or cognitive preconditions (Slobin, 1985a) or whether from early on children adapt to the categories, features and distributions expressed in their ambient language (e.g. Bowerman & Choi, 2001) as assumed by usage-based approaches to language acquisition. Of course these positions are not mutually exclusive: no theorist denies that humans are biologically adapted for language and no theorist denies that the child has to hear and analyse the ambient language to learn it. But there are major differences of emphasis and the major focus of argumentation has tended to be around the issue of how specifically linguistic such biological precursors are and to which aspects of language they might apply. In the following, we explore these issues by discussing studies of South and Southeast Asian languages in three core areas of acquisition research: (i) early vocabulary acquisition with a focus on nouns and verbs, (ii) spatial cognition, and (iii) the acquisition of argument structure. These topics are especially relevant to the above hypotheses because it has been widely assumed that in these three areas there are universal cognitive and/or linguistic universals that lead children to show a similar path of acquisition independently of the language they are learning. Early vocabulary The first words children say give us a glimpse on how they structure and see the world. This is one of the reasons why early vocabulary acquisition has been such a hot topic in acquisition research. A main focus in this research has been on whether nouns are learned before verbs. Universalist approaches to early vocabulary acquisition have claimed that there is an innate noun bias (Gentner, 1982; Macnamara, 1982; Gillette et al., 1999; Gentner & Boroditsky, 2001). In a cross-linguistic study on a number of typologically unrelated languages (English, German, Turkish, Japanese, Kaluli, Mandarin Chinese and Japanese), Gentner (1982) found

30

Sabine Stoll and Elena Lieven

that nouns outnumbered verbs in early acquisition in all of these languages. She argued that nouns are generally easier to learn because they refer to more concrete, naturally individuated referents (Gentner, 2006, p. 544), whereas verbs are more complex in their structure and are intrinsically relational which makes them more difficult to learn. This study resulted in her Natural Partition Hypothesis, which claims that there is a universal preference for nouns in early acquisition due to the fact that nouns are conceptually easier than other parts of speech. Subsequent research, however, has shown that there are languages in which children do not show such a preference and in which nouns are neither necessarily the first parts of speech, nor are they the most frequent parts of speech used in early vocabulary, even in English (Nelson, 1973; Gopnik, 1981, 1988; Bates, Bretherton & Snyder, 1988). As argued by Pine, Lieven and Rowland (1996) assessments of the distribution of early vocabulary depend very much on the method chosen. One very frequent method used is the Bates–MacArthur Communicative Development Inventory (MCDI) parental questionnaire. Nearly all these studies, with the exception of Childers, Vaughan and Burquest (2007) for Ngas (Chadic, AfroAsiatic, spoken in Nigeria), show a preference for nouns. However, the MCDI has far more nouns in the inventory than verbs, which might well bias the results (Stoll et al., 2012). A second methodological issue relates to how nouns are defined. In a comparative MCDI study comparing the use of nouns and verbs in English, Mandarin Chinese and Cantonese, Tardif et al. (2008) divided nouns into nouns denoting objects and nouns denoting persons and found differences between the three languages. They found that the first ten nouns used by children of all three languages mostly referred to people. However, English children used more object nouns than verbs, Mandarin Chinese children used twice as many verbs and Cantonese children showed an equal distribution of object nouns and verbs among the first ten words. Thus, there might be a universal but pragmatically determined preference for nouns referring to persons in early acquisition, but this needs to be explored for more languages in addition to noting that for some cultures, kinship terms are used in place of proper names. It seems clear that there are a number of factors which influence the relative proportions of nouns and verbs in children’s early vocabularies. For instance, a study by Tardif, Shatz and Naigles (1997) comparing the acquisition of nouns vs. verbs in English, Mandarin Chinese and Italian showed that children did not directly mirror the distributions found in the corresponding child directed speech. Both the positions of nouns and verbs in an utterance and their relative morphological complexity were factors, potentially competing, that could explain the behaviour of the children in these three languages.

Studying language acquisition cross-linguistically

31

The role of morphological complexity was investigated in more detail for children learning Chintang (Sino-Tibetan) by Stoll et al. (2012). Chintang is a polysynthetic language exhibiting extremely complex verb morphology with over 1840 verb forms per verb. The analysis of longitudinal data of four children (two children were aged 2 at the first recording and two children aged 3) showed (i) children do not mirror the low noun-to-verb ratios found in the surrounding adult language, (ii) as soon as they become more proficient with verb morphology children start to match the distributions of the adults. On the other hand, for a language like Mandarin, which hardly expresses morphology at all and exhibits almost no difference between the complexity of noun and verb morphology, this might be the reason why children match the adult noun and verb distributions earlier, that is, verbs are much more transparently accessible than in a language like Chintang. In a language like English the frequency of nouns in the input, the saliency of their position and maybe also the somewhat more complex morphology on verbs than on nouns could account for the higher proportion of nouns in early vocabulary. Spatial cognition Spatial cognition is an aspect of cognition which has been hypothesized to be universal. All humans not only have the same visual apparatus but move around in the world the same way. Further, there is cross-cultural evidence for similar development in object permanence and movement perception in infancy (Dasen, 1977). However, recent cross-linguistic research has shown that there is a range of possibilities for how humans structure and express space (Levinson, 2003). Most cultures in Europe use a left–right system together with an intrinsic system for the orientation of objects (e.g. ‘in front’, ‘behind’, ‘under’). However, this is by no means the only way in which space can be conceptualized. For instance, in Belhare (Sino-Tibetan, Nepal) (Bickel, 1997) spatial orientation is absolute: the points of orientation are fixed landmarks of the environment, that is, the hills of the region. So how do children learn the system used by their language and culture? In principle there are two possibilities: (i) all children start out with the same hypotheses as to how space is structured and expressed or (ii) they orient toward language-specific categories right from early on. Cross-linguistic research by Bowerman and Choi (2001) on Korean, Dutch and English, where containment includes very different features, for instance a contrast between loose and tight containment in Korean which is not marked in English or Dutch, suggests that, as soon as children start to comprehend and produce language, they use language-specific categories. Similar results were found in a comparative study by Narasimhan and Brown (2008) who examined whether Hindi (Indo-European, India) and Tzeltal (Mayan, Mexico) showed

32

Sabine Stoll and Elena Lieven

a preference for semantically specific expressions of containment (in/inside). Children of both languages were found to use the most frequent semantically abstract form without restrictions to specific contexts and irrespective of syntactic categories. In Tzeltal children also used semantically specific forms appropriately and they did not overextend these forms even in the very early phases of acquisition. This suggests that the children in both cultures are not simply repeating the forms at a lexically specific level but making an abstraction based on the distinctions made in their language. However, matching the ambient language from early on does not happen in all instances. Hindi children underextended the use of these forms and did not use them for temporal expressions, which is typical for adult usage. Similar to the results on vocabulary acquisition discussed above, comparative research on Tamil and Dutch (Narasimhan & Gullberg, 2006) has shown that typological characteristics of the input language can have an impact on how easily a linguistic feature is acquired. Narasimhan and Gullberg compared the role of frequency and semantic transparency in Tamil and Dutch placement verbs. Tamil posture verbs are semantically very transparent and ‘consist of morphemes that individually label the causal and result sub events’ (Narasimhan & Gullberg, 2006, p. 504) while Dutch posture verbs are expressed by whole words and are therefore opaque as well as conflating across different types of events. Tamil-speaking children aged 4 to 5 were tested with infrequent placement verbs and it turned out that they used them correctly. Dutch children, by contrast, were tested with frequent placement verbs and even at age 5 children used them inappropriately, despite their frequency. Thus, learning to express space in a language-adequate way is conditioned not only by the frequency of expression in a language but other factors such as morphology and semantic transparency. These results are similar to those on early vocabulary acquisition, which show that frequency alone is not responsible for the behaviour of children. Acquisition of argument structure Children have to learn how argument structure is marked in the language they are learning together with the valencies or argument frames of particular verbs. Case-marking, word order and agreement-marking are potential cues depending on the language and there are also semantic and discourse biases around the marking of agency, animacy, etc. In a language like English where there is no overt case-marking on either subject or object (except for pronouns), word order is an important cue. The question is how children know or learn that the position before the verb marks subject. There are two hypotheses which both rely on the idea of innate knowledge ‘bootstrapping’ the child into the system: one through semantics and the other through syntax.

Studying language acquisition cross-linguistically

33

In the semantic bootstrapping hypothesis (Pinker, 1984), children have an innate linking between the semantics of agency and the syntactic subject, which allows them to work out how agents are marked in their language and to bootstrap from this into the more abstract syntax of argument structure. By contrast, the syntactic bootstrapping hypothesis argues that children learn verb meanings from the syntactic contexts in which they occur and there is innate knowledge about the relationship between particular syntactic contexts and the broad semantics they express. For instance, a sentence with three semantic roles is likely to reflect some sort of giving (Gleitman, 1990). One major difficulty with both approaches is that, in most languages of the world, the input is notoriously unreliable with respect to the expression of arguments because most languages ubiquitously omit arguments. Thus children learning a language that shows pervasive argument ellipsis encounter a very different learning environment from that which has been taken for granted by studies of Western European languages such as English, German and French, where arguments are usually realized. A study by Narasimhan, Budwig and Murty (2005) on the acquisition of Hindi argument realization examined whether children mirror the pervasive argument ellipsis they encounter in childdirected speech. One expectation might be that they make errors with transitive verbs due to this frequent ellipsis. The study was based on the discourse of 12 child–parent dyads (2;10–4;3) recordings of approximately an hour. Narasimhan et al. found that even though there is massive argument ellipsis in child-directed speech, children did not make mistakes in their use of arguments and from early on they seemed to ‘exhibit sensitivity to discoursepragmatic influences’, grammatical role and referential form (Narasimhan et al., 2005). While this study shows that children do seem to observe the valency frames of the verbs they use, it is not yet clear exactly how they work this out. Additionally, there is, as always, the issue of sampling to be considered since the study depends on relatively limited amounts of discourse and this might make it difficult to find errors. To answer the question what children can do and would do if pushed hard testing in an experimental setting is needed. This was done in a comparative study by Chan, Lieven and Tomasello (2009) on English, German and Cantonese with children aged 2;6, 3;6 and 4;6. Transitive utterances in English and German usually have both arguments expressed but around 20 per cent of transitives in German child-directed speech have OVS word order. Cantonese child-directed speech, on the other hand, shows a very high degree of ellipsis of both subject and object arguments. Children were tested with novel verbs in three types of active transitive constructions: sentences in which the subject was animate and the patient inanimate (A–I, the most frequent type in the child-directed speech of all three languages); sentences in which both agent and patient were animate (A–A); and, finally, sentences

34

Sabine Stoll and Elena Lieven

in which there was an inanimate agent and an animate patient (I–A). Even the 2½-year-olds in all three languages chose the first noun as agent when hearing the A–I sentences, but all groups were at chance in the ‘counter-prototypical’ condition in which animacy and word order conflicted (I–A). Across sentence types, children made significantly fewer first-noun-as-agent choices when the animacy contrast was neutralized (AVI vs. AVA) or conflicted with word order (AVA vs. IVA). This was true for Cantonese children at all three ages; for German children at ages 2;6 and 3;6; and for English children at 2;6. The suggestion is that the results are related to the frequency of SVO active transitives and the expression of their arguments. As noted above, 20 per cent of active transitives in German have OVS word order. This means that German-speaking children are presented with a cue conflict in word order which takes them longer to resolve than for the children learning English in which this conflict does not exist. And the degree of argument ellipsis in Cantonese results in children having much less evidence for SVO word order, again explaining their slower progress. Researchers from different theoretical backgrounds have suggested that children might have an innate agentivity bias (Braine, 1976; Pinker, 1984; Slobin, 1985b). In Pinker’s case this was termed the ‘semantic bootstrapping hypothesis’ (Pinker, 1984, p. 461). This account predicts that children form a semantic concept of an agent and then identify agentive arguments as being similar and mark them in a similar way, that is, as subjects in both transitive and intransitive sentences. This would actually lead to erroneous marking in languages with ergative case in the early phases of acquisition because in ergative languages the agent of a transitive sentence is marked differently than the subject of an intransitive sentence. Thus, if children have an agent bias, one would expect them to extend the case-marker of the most prototypical agent, that is, the A of a transitive sentence to the S of an intransitive sentence. However, so far none of the studies on a variety of languages with case marking has found such an overextension of ergative marking from A to S. Even in languages with an additional complication of a split ergative system like Hindi, children did not show such overextension. Narasimhan et al. (2005) studying the longitudinal development of three children found no overextension errors even though the split system is rather intricate in that only perfective verbs in the past tense mark their A with the ergative and in all other cases there is accusative marking. This suggests that children are sensitive to the distributional patterns of the input. The three examples briefly outlined above strongly suggest that as soon as children start extracting form–meaning mappings from what they hear, they are remarkably sensitive to the distributional patterns in the input, rather than adhering to innate, universal categories.

Studying language acquisition cross-linguistically

35

Summary/conclusion We have briefly summarized some of the major theoretical and methodological issues involved in the study of children’s language acquisition. It is clear that almost none of these questions can be properly addressed in the absence of detailed information about the acquisition of a wide range of typologically very different languages and that this poses a stark challenge for researchers in terms of commitment of both time and other resources. However, we have indicated a number of areas of theoretical interest where a clearly focused typological question with the appropriate methodology could contribute a great deal to unravelling this fascinating area. Although the collection of a well-sampled corpus of language development in languages that contrast typologically with those already studied is the ideal and a most urgent necessity, the use of targeted diaries or elicitation situations to focus on a particular structure would also be extremely useful. We should also emphasize the importance of interdisciplinary teamwork, particularly if the language in question is previously relatively undocumented and/or spoken by communities in non-technologically advanced settings. The insights and analyses of typologists, developmental psychologists and anthropologists are all necessary for understanding the relationship between children’s language development and the contexts in which it occurs, including the fact that the majority of the world’s children grow up in multilingual rather than monolingual environments.

2

Infant-directed speech: social and linguistic pathways in tonal and non-tonal languages Christine Kitamura

Social interaction is integral to language acquisition (Kitamura & Burnham, 2003; Snow, 1989). A key attribute of early social interaction is the caregivers’ use of infant-directed (ID) speech with its greater affective salience, higher fundamental frequency (F0; pitch), wider pitch range (Kitamura & Burnham, 2003; Kitamura et al., 2002), and larger vowel space than adultdirected (AD) speech (Burnham, Kitamura & Vollmer-Conna, 2002; Kuhl et al., 1997). Evidence shows the emotion in its exaggerated prosody provides simultaneous attentional, and social-affective benefits (Kitamura et al., 2002; Snow, 1989), as well as assists infants in many of the challenges of early speech perception, e.g. to segment the speech stream (Theissen, Hill & Saffran, 2005), find stressed words in sentences (Fernald & Mazzie, 1991), discriminate multisyllabic words (Karzon, 1985), and parse clauses (Kemler Nelson et al., 1989). Overall, ID speech is ideal as a didactic tool due to its slowed speech rate, its simpler syntax and exaggerated prosodic and affective features. In this chapter, studies that include ID speech in tonal and non-tonal languages are reviewed to elucidate the universal and language-specific components as they relate to social, cultural and linguistic imperatives. Pitch, intonation and tones The concern in tonal and pitch-accented languages is that the heightened pitch and expanded pitch range in ID speech might thwart the phonological function of tonal or pitch accent information. In non-tonal languages, pitch is used in stress-timed languages such as English at the syllable level as a rhythmic timing device; at the word level to emphasise key words in a sentence; and at the sentence level, pitch is used to convey the emotions, attitudes, and intentions of the speaker. However, in pitch-accented languages such as Japanese, and tonal language such as Thai, pitch also acts independently at the syllable level to distinguish lexical meaning. For instance, in standard Thai there are five lexical tones (low, mid, high, rising and falling). Each syllable in the Thai language carries one of these tones, and tone can differentiate lexical meaning of the same sequence of phonetic segments: for 36

Infant-directed speech: social and linguistic pathways

37

example, /kha/ mid ‘to be stuck’, /kha/ low ‘galangal, a rhizome’, /kha/ high ‘to engage in trade’, /kha/ rising ‘leg’ and /kha/ falling ‘to kill’. The challenge for speakers of languages in which pitch is used to distinguish semantic meaning is to maintain lexical tone in the context of sentence intonation. With adults, despite finding some Thai tones may become almost indistinguishable when superimposed on the intonation of a question or unfinished statement, all other tones remain phonologically distinct in the context of emotional and attitudinal prosody (Luksaneeyanawin, 1998). There is evidence that speakers of Thai and other tonal languages restrict their use of pitch for affective signalling in order to protect the phonetic integrity of the tonal information in their language (Ross, Edmondson & Seibert, 1986). However, this does not reduce their ability to signal emotional and other information effectively because Thai speakers compensate by using final particles to express mood, social status and the relationship between the speaker and the listener. For example, male and female speakers use different final particles, ‘khrap’ by males and ‘kha’ by females to signal their gender, and for the purposes of politeness. The use of final particles provides a complement to any pitch restrictions that are necessary to keep lexical tone intact, while still conveying social and emotional information. It would be expected that the integrity of lexical tones in high pitch ID speech would be particularly susceptible to corruption. This is not the case. Utterance-initial and utterance-final tones were compared in Thai ID speech at birth, 3, 6, 9 and 12 months of age and in AD speech (Kitamura et al., 2002). As shown in Table 2.1, lexical tones in the utterance-initial position remained identifiable at all ages, but were significantly less identifiable in the utterance-final position in speech to newborns, 3-, 6- and 9-month-olds than AD speech. Importantly, there was no difference in the number of identifiable utterance-final tones in speech to 12-month-olds and adults. Thus, it appears that tones in Thai ID speech are not corrupted in the syllable-initial position but are, to some extent, in the syllable-final positions. Overall, it seems mothers ensure reasonably accurate production of tones, and importantly by 12 months, tonal information is similar in ID and AD speech (Kitamura et al., 2002). Table 2.1 Percentage correct of lexical tone in ID speech at birth, 3, 6, 9, 12 months and in AD speech in the utterance-initial and utterance-final position Position of tone

Birth

3 mths

6 mths

9 mths

12 mths

ADS

Utterance-initial Utterance-final

99.3 72

99.3 71.5

100 77.3

99 73.3

100 78.3

100 85.5

38

Christine Kitamura

Cross-cultural universals, interaction and ID speech It is well documented that infants prefer ID over AD speech in English (Fernald, 1985; Panneton Cooper & Aslin, 1990; Werker & McLeod, 1989) and Japanese (Hayashi, Tamekawa & Kiritani, 2001), and that infants’ ID preferences are based on its heightened vocal affect (Kitamura & Burnham, 1998; Singh, Morgan & Best, 2002) conveyed by its pitch patterns (Fernald & Kuhl, 1987). Affective messages in ID speech have distinctive pitch contours designed to directly influence the infant’s affective state; and these pitch contours are more likely to be used in particular contexts, for example, falling contours are used to calm the infant, bell-shaped contours to reward and rising contours to gain attention (Fernald & Simon, 1984; Papousek, Papousek & Symmes, 1991; Stern, Spieker & MacKain, 1982). Furthermore, ID affective messages, including attention-bid, approval and comfort, are much easier to identify by adults in ID than AD speech (Fernald, 1989), and this occurs across language and cultural barriers (Bryant & Barrett, 2007). Even Shuar adults from an Ecuadorian indigenous tribe recognise categories of ID affective intent produced by North American mothers (Bryant & Barrett, 2007). The production and perception of affective intent in ID speech is intrinsically tied to its prosodic patterns, and many studies speak strongly to the universality of its affective expression. Affective messages of approval, comfort and prohibition are expressed using the same intonation contours whether the language is tonal Mandarin or non-tonal American English (Papousek et al., 1991). Infant preferences for ID speech are also not language-specific: Cantonese Chinese and English infants are as attentive and affectively responsive to ID speech whether it is presented in their own or non-native language (Werker, Pegg & McLeod, 1994). In addition, English-learning 5-month-olds show the appropriate affective responses to approving and disapproving ID contours whether presented in English, German, Japanese or Italian (Fernald, 1993). These studies indicate that the perception of affect in ID speech is very similar across cultures, and consistent with Kitamura and Burnham’s (1998) contention that a distinctive and definitive characteristic of ID speech is its emotive nature. It is this emotive quality which acts to lubricate social interaction with a cognitively immature infant, whatever the language background of the infant. It is agreed that this style of interaction teaches the infant the protocols of social interaction, and facilitates language acquisition (Snow, 1977, 1989). The infant’s early introduction to language may be through the attentional and affective qualities of ID speech and then, as the infant gets older, its features are modified to more directly facilitate language acquisition (Fernald, 1992; Kitamura & Burnham, 2003; Kitamura et al., 2002). Thus,

Infant-directed speech: social and linguistic pathways

Developmental forces shape infant feedback

INFANT predisposed to listen to ID speech

Feedback from infant modifies mother’s input

MOTHER predisposed to use ID speech

39

Cultural forces shape mother’s speech and response to feedback

Figure 2.1. Model showing how infant-directed speech is modified according to the cultural background of the mother, and developmental forces in the infant.

the emotional qualities of ID speech provide the foundation for language learning. Figure 2.1 illustrates the relationship between the factors in mother– infant interaction. Underlying the interactive behaviour is the generally accepted notion that young infants have a predisposition to listen to ID over AD speech (Fernald, 1985; Panneton Cooper & Aslin, 1990; Werker & McLeod, 1989) and that caregivers are predisposed to use this speech style in the presence of their infant. For the caregiver, ID speech is modified by the caregiver’s cultural background, and reflects culturally important beliefs and values (Fernald & Morikawa, 1993; Grieser & Kuhl, 1988; Kitamura et al., 2002), and attitudes to child-rearing (Schieffelin & Ochs, 1983). For the infants’ part, feedback to the caregiver changes as developmental forces act on the growing infant, such that infants’ developmental needs are specified via this feedback mechanism. Thus, it is within this loop that mothers’ input is modified by the infants’ developmental stage and the social customs the mothers wishes to inculcate in her infant. Culture-specific variations in ID speech Early studies pointed to a unique blend of features in ID speech in diverse languages, for example Arabic, Comanchi and Gilyak (Ferguson, 1964), Spanish (Blount & Padgug, 1977), Marathi (Kelkar, 1964) and Latvian (Ruke-Dravina, 1977). More sophisticated analyses of ID prosody have also been conducted in tonal languages such as Thai (Kitamura et al., 2002) and Mandarin Chinese (Grieser & Kuhl, 1988; Papousek & Hwang, 1991), and in pitch-accented languages such as Japanese (Fernald et al., 1989). More often, analyses are conducted in non-tonal languages such as American English (Stern et al., 1983), Australian English (Kitamura & Burnham, 2003; Kitamura et al., 2002), British English (Shute & Wheldall, 1989) and German (Fernald & Simon, 1984; Fernald et al., 1989). In short, the primary difference between ID speech in tonal and non-tonal languages is that pitch range in tonal languages is much narrower. Mean pitch, on the other hand,

40

Christine Kitamura

tends to be lower but not decreased to the same degree as pitch range (Fernald et al., 1989; Grieser & Kuhl, 1988; Papousek et al., 1991). In addition to subdued ID intonation, Asian mothers use other culturespecific features in their semantic content; for example, American mothers use more information-oriented utterances whereas Japanese mothers use more affect-oriented utterances – nonsense, onomatopoeic sounds, baby talk and baby names (Bornstein et al., 1992; Toda, Fogel & Kawai, 1990). Japanese mothers also talk more to their infants, and engage infants in social routines twice as much as American mothers (Fernald & Morikawa, 1993). These cross-cultural variations appear to reflect the relative importance of culturespecific beliefs and values: Japanese mothers are more interested in teaching their infants the norms of polite social interaction, while American mothers may be more intent on encouraging the development of language and cognitive skills. Despite the cross-cultural variation in affective expression, these studies indicate that ID speech provides a means by which the young infant is socialised into the appropriate behaviours of the culture (Fernald & Morikawa, 1993; Toda et al., 1990). For some cultures (see Stoll & Lieven, Chapter 1), cultural attitudes to child-rearing impact on the amount of speech addressed directly to the child. For instance, children in Samoan and Papua New Guinean cultures, such as the Kaluli, typically do not directly address children until they begin to talk (Ochs & Schieffelin, 1984) leaving infants to gain linguistic knowledge from listening to adults talking to other adults. Infants in such non-industrialised cultures, of course, would be surrounded by many more adults in their daily life than infants in industrialised cultures. While, it is shown that the quantity and the quality of speech directed to infants does effect aspects of language development in industrialised societies (Hart & Risley, 1995), in other nonindustrialised societies, less is known about how differences in means of input influence the timing of specific milestones in the development of language. Age- and sex-related adjustments to ID speech In tonal and non-tonal languages mothers tend to adjust pitch and affective intent in response to signs of development in the infant, and in keeping with the social mores of the culture (see Figure 2.1). In the first year of development, infant socialisation is most intense around 6 months of age (see Kitamura & Burnham, 2003). By 9 months, there is perceptual reorganisation in linguistic and cognitive realms and with it attention to native language over non-native vowels, consonants and tones (Mattock & Burnham, 2006; Polka & Werker, 1994; Werker & Tees, 1992). Age-related features in ID speech have the potential to provide insights into the attentional, social and/or linguistic significance of ID speech and how it advantages social and

Infant-directed speech: social and linguistic pathways

41

linguistic development. Sex-related adjustments, on the other hand, more clearly show how cultural expectations impinge on pitch adjustments in speech to sons and daughters. A longitudinal study examining adjustments to mean pitch and pitch range at birth, 3, 6, 9 and 12 months shows mean F0 and pitch range is more constrained in Thai than Australian English ID speech (Kitamura et al., 2002). Thai mothers’ restrictions on pitch levels serve at least one purpose to advantage linguistic development as they maintain the integrity of tonal information. Age trends at 3-monthly intervals from birth in the Kitamura and colleagues study show how mothers from tonal and non-tonal languages accommodate infants’ developmental level, depending on whether they are in a social (0–6 months) or linguistic (9–12 months) phase (Kitamura et al., 2002). As shown in the bottom panel of Figure 2.2, pitch range follows a similar path in Thai and English, but mean F0 (in the top panel of Figure 2.2) takes a different path across age. The age trends for mean F0 from birth to 6 months are similar during socialisation, but during the emerging linguistic period at 9 months, Australian English and Thai mothers converge at a similar level of mean pitch. Mothers from the two language environments appear to engage in different strategies to achieve a similar language advantage at a critical age for native speech perception: Australian English mothers decrease mean F0 to dampen the high level of positive vocal affect evident in speech to infants at younger ages (Kitamura & Burnham, 2003) whereas Thai mothers become more emphatic to achieve the same effect. When infants are uttering quasi-words at 12 months, the age trends diverge, and English mothers increase average pitch, while Thai mothers reduce pitch to almost adult levels. This may allow Thai mothers to approximate the tonal information found in adult Thai speech at a time when there is a shift in infants’ attention to the language-specific components of speech. Moreover, these language-specific differences may be directed at exposing infants to specific characteristics of the ambient language. Mothers in both cultures adapt their speech differently depending on whether they have a girl or boy baby. As shown in Figure 2.3, Thai and Australian mothers always use higher pitch range in speech to girls. Mean pitch, on the other hand, while it is elevated more for girls than boys in Australian English, in Thai, mothers are more restrictive in their use of mean F0 with girls, especially during the intense socialisation phase around 6 months (Kitamura et al., 2002). This configuration in Thai ID speech seems designed to expose infants to sex-appropriate behaviours, which may be more highly valued for girls in Asian countries. In Australian English, mothers adjust pitch similarly in speech to male and female infants until about 6 months, but after this age, there is an emerging differentiation in pitch level, most likely based on infant social responsiveness, and the emergence of

42

Christine Kitamura

Mean F0

350

F0 (Hertz)

300 250

200 English

Thai

150 Birth

3 mths

6 mths

9 mths 12 mths

Adult

Pitch range

16

Semitones

14 12 10 8 6

English

Thai

4 Birth

3 mths

6 mths

9 mths 12 mths

Adult

Figure 2.2. Mean F0 (top panel) and pitch range (bottom panel) for English and Thai at birth, 3, 6, 9 and 12 months, and in AD speech.

more sex differentiated behaviours. Australian English and Thai mothers want children who fit well in their society. Australian mothers may allow their infants more scope for individual expression, but Thai mothers appear more intent on raising polite children, with the expectation that politeness is more important for girls than boys. The commonalties in ID speech in tonal and non-tonal languages, together with the cultural differences, provide a rich source of information with regard to the functions of ID speech. Both Australian English and Thai mothers use higher pitch characteristics when speaking to their infants. Cross-language

Infant-directed speech: social and linguistic pathways Mean F0 (Australian English)

1.6

1.6

1.4

1.4

1.2 1.0

Girls Boys ADS

0.8

Mean IDS/ADS

Mean IDS/ADS

Mean F0 (Thai)

0.6

1.2 1.0

Girls Boys ADS

0.8 0.6

Birth

3 mths

6 mths

9 mths 12 mths

Birth

Pitch range (Thai)

1.6

1.4 1.2 1.0

Girls Boys ADS

0.8 0.6

Birth

3 mths

6 mths

9 mths 12 mths

Mean IDS/ADS

1.6 Mean IDS/ADS

43

3 mths

6 mths

9 mths 12 mths

Pitch range (Australian English)

1.4 1.2 1.0

Girls Boys ADS

0.8 0.6

Birth

3 mths

6 mths

9 mths 12 mths

Figure 2.3. Mean F0 (top panel) and pitch range (bottom panel) in speech to Thai and English boys and girls at birth, 3, 6, 9 and 12 months, and in AD speech.

evidence almost universally shows that mothers from diverse linguistic and cultural backgrounds increase their mean F0 and pitch range in the presence of an infant. Similarly, pitch characteristics increase from birth to 6 months in Thai and English ID speech to girls and boys (Kitamura et al., 2002). The one caveat to this is that mothers from tonal or pitch-accented language backgrounds produce a narrower pitch range. These findings are consistent with the position that mothers across cultures exaggerate the pitch characteristics in ID speech to their infants’ advantage and in tune with the cultural mores of their language background. Maternal communicative practices appear to be driven by cultural social factors at younger ages, but take more account of linguistic development by 9 months. If we assume a reciprocal relationship exists between mother and infant, that is, each is able to influence the other’s behaviour, then the degree to which mothers allow infants to drive the relationship may depend on the expectations of the society. The way Thai mothers interact with their infants, especially in the early months, is more clearly culturally determined while its use at older ages is more likely determined by a need to maintain the structure of the Thai tonal system. Australian English mothers are not constrained in this way, and may be using higher pitch to emphasise key words in an

44

Christine Kitamura

utterance, and do this more so with girls than boys. However, despite the overriding commonalties in Australian English and Thai ID speech and their implications for infant language and social development, the results show that sex-specific modifications to F0 in ID speech may be more attuned to developing culturally appropriate communicative practices.

3

Pragmatic development of Mandarin-speaking young children: focus on communicative acts between children and their mothers Jing Zhou

The development of communicative competence is a critical component of early language acquisition. In order to interact and socialise effectively in a particular cultural context, it is essential to learn not only the grammar and vocabulary of a language but also the communicative or pragmatic skills associated with that particular language and culture. This chapter discusses the gradual development of communicative and socio-interactional skills in Mandarin-speaking children aged from 1 to 6 years old as they interact with their mothers. First, the development of communicative acts in infancy will be discussed, followed by the development of pragmatic skill acquisition in older children. Finally, Chinese mother–child interactions will be discussed in relation to the socio-cultural context. Pragmatic development in infancy and early development Typically developing prelinguistic children develop socio-interactionalrelated skills including joint attention, imitation, vocalisations, gestures, symbolic representation and role reversal, which form the foundation for subsequent language development (Stoll & Lieven, Chapter 1). As children’s language begins to emerge, the socio-cultural context plays an increasingly influential role. Many aspects of language development begin to diversify due to the influences of both the structure of the language as well as the socio-cultural environment. Pragmatic skills develop and are shaped through interactions with members of the child’s particular cultural group. In order to investigate the acquisition of verbal and communicative acts in prelinguistic and linguistic children, Ninio and Snow (1996) developed the Inventory of Communicative Acts-Abridged (INCA-A) (Ninio et al., 1990). It forms a system for coding non-verbal and verbal communicative acts and intentions. INCA-A distinguishes between three levels of communicative acts: (i) the level of social interchange in which speakers encode their communicative intentions, (ii) the speech act level that expresses the speaker’s communicative meaning, and (iii) the social interchange and speech 45

46

Jing Zhou

Table 3.1 The development of communicative acts of Chinese young children at the age of 14 to 32 months Age

Proportion of attempts interpretable at social interchange level (%) Proportion of attempts interpretable at speech act level (%) Number of interchange types Number of speech act types Pragmatic flexibility (types of combinations of social interchanges and linguistic expressions)

14 months (N ¼ 20)

20 months (N ¼ 20)

26 months (N ¼ 20)

32 months (N ¼ 20)

67

90

98

98

41

80

94

97

3.9 (0.56) 4.0 (1.63) 5.5

5.2 (1.13) 8.6 (2.36) 12.1

5.3 (1.41) 12.2 (4.18) 15.4

6.7 (0.94) 13.4 (2.67) 18.5

Source: Adapted from Zhou (2002).

act combination level, which is also called the level of pragmatic flexibility (Ninio et al., 1994; Ninio & Snow, 1996). Snow and her colleagues have used this framework with typically developing English-speaking children and found that they rapidly expand their communicative repertoire over a period from 14 to 32 months (Snow et al., 1996). In a parallel study, the communicative acts of young Mandarin-speaking children aged from 14 to 32 months were investigated (Zhou, 2002; Zhou & Fletcher, 2006). The communicative repertoire increased from 14 months to 32 months in terms of the number of different types of communicative acts used at all three levels with a corresponding decline in the proportion of uninterpretable communicative attempts made. The data in Table 3.1 reflects the changes in communicative acts of this group of Mandarin-speaking young children. The longitudinal observations of a single child at 14, 20, 26 and 32 months confirmed this pattern of development. The results from both the cross-sectional as well as longitudinal studies showed a similar pattern in pragmatic development. The results found in this study on Chinese children corresponded with Snow and colleagues’ study on American children. Even before the children have the ability to express clearly their communicative intentions, the 14-month-old Chinese children used three different forms of communication: (i) communicative attempts with expressive intentions; (ii) communicative attempts with non-verbal expressions of intentions; and (iii) communicative attempts with no specific expressions of intentions. During the period

Pragmatic development of Mandarin-speaking young children

47

45% 40% 35% 30% 25% NANJING

20%

HARVARD

15% 10% 5% 0% DHA

NIA 14 m

DJF DHA

NIA 20 m

DJF DHA

NIA

DJF

32 m

Figure 3.1. A comparison of three core types of social interchange in Chinese and American children.

under study, the interpretable communicative attempts significantly increased (see Table 3.1). The intelligibility of the children’s communicative intentions gradually improved, which indicates that they are acquiring the ability to become more intentional communicators (Zhou & Fletcher, 2006). Another noticeable change was the increase in the types of communicative acts produced at all three levels. At the social interchange level the number increased from 3.9 to 6.7, at the speech act level from 4.0 to 13.4, and at the combination level from 5.5 to 18.5. Young Chinese children were able to use three core types of social interchanges: DHA (Directing hearer’s attention), NIA (Negotiating immediate activity) and DJF (Discussing a joint focus). They were also able to use a variety of different speech act types including DCC (Discussing clarification), DFW (Discussing fantasy world) and DNP (Discussing non-present) to communicate with their mothers. At the social interchange level, the Chinese children were similar in the number of types of interchange and in the core set of interchange types with the American children in the Harvard study. However, as illustrated in Figure 3.1, there were some interesting differences in the acquisition of communicative acts emerging between the two cultural groups. The Chinese children tended to use more DJF (Discussing a joint focus) at all ages of 14 months, 20 months and 32 months in comparison with the American children. In particular, from the age of 20 months DJF became the most frequent speech act used by the Chinese children while the English-speaking

48

Jing Zhou

children used more NIA (Negotiating immediate activity) than any other type. Thus, we can see in these young Chinese and American children that their style of communicating is already diversifying. Pragmatic development in preschool children The results of a national Chinese project (Zhou, 2006) have illustrated that there is continued growth in the communicative repertoire of Chinese children in the preschool period. The study found that the communicative repertoire of children from 3 to 6 years old increased in terms of the number of communicative acts used at the three different levels of the INCA framework. The children’s number of interchange types increased from 7.9 to 9.2 and speech act types from 14.1 to 16.5. The pragmatic flexibility score, a measure of the combined interchange and speech act levels, increased over the study period from 18.5 to 22.5 (Zhou, 2006). Several types of speech acts emerged prior to 3 years of age such as SI (State intent), DC (Create a new state of affairs by declaration), QN (Ask a product wh-question), AN (Answer in the negative to yes/no question) and RD (Refuse to carry out act requested by other). Some relatively late-emerging speech act types indicate a growing trend for increased engagement, as illustrated by the use of FP (Ask for permission to carry out act), GR (Give reason) and CS (Counter-suggestion; an indirect refusal). The proportion of occurrences of these speech act types increased in frequency after the children’s third year (Zhou, 2006). These results indicate that preschool children are gradually developing a wide range of different strategies for expressing their communicative intentions. The relationship between the development of communicative acts and children’s syntactic development was also investigated. The syntactic development of the Chinese children was measured using mean length of utterance (MLU) and mean length of the five longest utterances (MLU 5). Over the study period, the Chinese children showed a gradual increase in MLU and a clearly increasing trend for MLU 5. MLU and MLU 5 correlated positively with the children’s speech act repertoire. This suggests a complex crossdomain relationship between aspects of pragmatic ability and syntactic knowledge at particular periods of development. Two other studies have also revealed a growing repertoire of speech act types in 3- to 6-year-old Mandarin-speaking children. Li (2008) reported that Chinese children produced an increasing number of different types of speech acts during peer play time. According to Li, linguistic acquisition and the development of speech acts contributed to children’s metacommunication skills in the play context. Yang (2009) analysed data of 3- to 6-year-old Chinese children and showed the growing capacity of these children in terms of the different speech acts produced as well as their linguistic abilities in

Pragmatic development of Mandarin-speaking young children

49

conversations. These studies (Li, 2008; Yang, 2009; Zhou, 2006) suggest that these two aspects of language, syntax and pragmatics, are intricately interconnected and illustrate the role that different social contexts and experiences play in children’s language development. The relationship between children’s pragmatic and syntactic development is the subject of much debate, particularly in relation to the proposition that the development of children’s communicative intentions is the source of children’s linguistic growth (e.g. Bates, 1979; Ninio & Snow, 1996). Children gradually develop language skills through interaction with others in sociocultural contexts. Initially, children start to communicate with very limited language abilities. Varied social exchanges enable children to gradually expand and master new forms of language. Thus, it can be envisaged that pragmatic skills and syntax concurrently develop through these varied sociocultural contexts (Zhou, 2002, 2006). These studies demonstrate the complex relationship between syntactic and pragmatic development. Understanding how these two language domains interact and co-drive language development may facilitate a richer and more integrated approach to early language teaching and education. Chinese mother–child interactions Communicative interactions and exchanges bear important socio-cultural information. The particular skills and orientation that children develop are rooted in the specific historical and cultural contexts of the community in which children and their companions interact (Rogoff, 1990, 2003). Embedded in the interactions that occur between children and their caregivers are everyday socio-cultural ‘lessons’. In the study of pragmatic development of Mandarin-speaking young children from 14 months to 32 months, it was found that Chinese mothers in comparison to American mothers had a smaller repertoire of communicative acts but showed their intentions more clearly by their frequent use of this limited repertoire. At all stages, the Chinese mother’s focus in their interactions with children were on DJF (Discussing a joint focus), DRP (Discussing the related to present), DNP (Discussing the non-present) and DRE (Discussing a recent event). Strong correlations were found between the frequencies of the communicative acts of mothers and their children (Zhou, 2002). These findings illustrate how Chinese mothers have their own characteristic ways of communicating and transmitting information when interacting with their children, and in turn the mother’s social behaviour influences the child’s linguistic development. In a more recent study (Zhou & Li, 2010), striking commonalities in the language characteristics of Mandarin-speaking mothers from diverse

50

Jing Zhou

educational backgrounds when interacting with their 3- to 6-year-old children, were observed. The mean number of types of speech acts at all three levels, the intention of using discussion types of interchanges, and the intention of asking questions were remarkably similar. This illustrates a common culture-specific style of interacting by Chinese mothers with their young children regardless of educational background. Chinese mothers typically use discussion-based speech acts when interacting with their children. This characteristic way of interacting also applies to mothers of autistic children. In a study by Li (2008), it was found that young autistic Chinese children outperformed their typically developing peers in communicative acts (types of social interchanges and speech acts). The 18-month longitudinal study showed that Chinese autistic children had a relatively high frequency of communicative acts, but had not developed initiative functional communication (Li, 2008). It is believed that the communicative behaviour of Chinese autistic children may be influenced by the cultural learning environment and/or by the expectations and perceptions of parents. Chinese mothers tended to offer a question–answer framework and use more frequent communicative interchange types when interacting with their autistic children. The following speech acts – DJF (Discussing a joint focus), DRP (Discussing the related to present), DNP (Discussing the non-present) and DRE (Discussing a recent event) – frequently occurred in the mother’s speech. This illustrates the differences in these Chinese mothers’s communication style in comparison with the mothers of typically developing children (Li, 2008; Zhou et al., 2008). Chinese mothers attach much importance to information exchange with their children. Le et al. (1997) propose that the role played by Chinese mothers is that of a teacher. Chinese mothers emphasise the transmission of knowledge, through discussion of joint focus, related to present, non-present and recent events. In mother-child interactions, they seldom used MRK (Marking) to express socially expected sentiments on specific occasions such as thanking, apologising, or to mark some event. At the same time there were very few questions used to converse about the hearer’s and speaker’s feelings. Fifty-two per cent of the language used by the mothers was related to different kinds of questions. The implication is that Chinese mothers, whether consciously or subconsciously, have a tendency to play the role of the teacher. As mothers instigate their children’s participation in discussions on various topics around the age of 3, children have plenty of opportunity to learn how to discuss a topic with a joint focus. In interaction with their mother, Chinese children spend a large proportion of their time answering their mother’s wh-questions or in stating their views. They also asked few questions and made few objections. This illustrates how mothers’ guidance shapes Chinese children’s behaviour and how children acquire the social role of the student.

Pragmatic development of Mandarin-speaking young children

51

This kind of information-exchange-centred interactional process predominantly used by the Chinese mothers is different from the commonly known child-centred model or adult-centred model of learning. In the child-centred interactive model Western middle-class parents often adopt an ‘adult lowering’ interaction strategy (Ochs & Schieffelin, 1984; Ozcaliskan & Goldin-Meadow, 2005). In their interaction with children, adults promote interactions by focusing on the topics and activities of interest to the child (Pan et al., 1996). Traditional Chinese parents, in contrast, typically adopt a more authoritarian model of interaction. However, Chen et al. (2000) found that Chinese mothers with a relatively higher level of education seldom use the authoritarian model in interactions with their children. Le et al. (1997) also mention that, in interactions between middle-class Chinese mothers and their children, children are both positive and proactive. We have found from our observations that contemporary Chinese mothers neither completely adhere to the traditional adult-centred model or to the Western child-centred model. The way children are treated and raised is an important variable that differs widely across cultures (Stoll & Lieven, Chapter 1) and influences language development. Thus, it is essential that language development is investigated in diverse cultural contexts. The South and Southeast Asian contexts offer many rich opportunities for further research on language development. It is imperative that the development of communicative or pragmatic skills as well as syntax and grammar is investigated.

4

Referential forms in Thai children’s narratives Theeraporn Ratitamkul

Overview Among many linguistic skills to be acquired by a competent speaker of any language is the ability to produce a coherent discourse. In storytelling, a narrator needs to sequence events in a comprehensible order, make appropriate reference to story characters, both new and given, observe the norms and conventions in the language spoken, and at the same time take into account the listener’s attention state. One of the important linguistic devices used to establish coherence in narratives is referential forms. As Halliday and Hasan (1976) point out, cohesion in discourse is strongly related to reference. While telling a story, the narrator thus constantly makes decisions about referential choice to be used at different points in the story. It is interesting to examine referential forms in children’s storytelling as this reflects their narrative skills. Researchers have used various methods to elicit narrative data from children. One common methodology is a picture-based task where children tell stories stimulated by a series of pictures of connected events. A notable example is the use of Mercer Mayer’s (1967) wordless picture book, Frog, Where Are You? Data have been collected from children speaking different languages, and various aspects of narratives have been investigated, for instance the description of motion events, the expression of spatial and temporal relations, and the use of referential terms to refer to entities in the story (see Berman & Slobin, 1994; Stro¨mqvist & Verhoeven, 2004). This trend of research also allows for a developmental study of narratives across age groups and languages. Other methods of narrative data elicitation include, for example, the use of a silent film as a stimulus and a recall or retell task in

Part of this work was presented in a poster entitled ‘A developmental study of referential strategies in the Thai Frog Stories’ presented at IASCL 2011 in Montreal, Canada. This work was partly supported by the Chulalongkorn University Centenary Academic Development Project and the Higher Education Research Promotion and National Research University Project of Thailand, Office of the Higher Education Commission (HS1153A). I would also like to thank Chutichol Aemdit, Sumintra Maklai, Teeranoot Siriwittayakorn and Junyawan Suwannarat for their help in data coding.

52

Referential forms in Thai children’s narratives

53

which children tell familiar stories. Work on child narratives can therefore vary with regard to modes of stimulus presentation, length and complexity of the stimulus as well as shared knowledge between the child and the listener. It is clear that methodological discrepancies can lead to different results and conclusions pertaining to referential forms (see Hickmann, 2003 for details). Referential forms in child narratives have received a great deal of attention over the past four decades. Research has been conducted in several languages, covering a wide range of language families and socio-cultural backgrounds. The list includes, but is not limited to, Cantonese, English, French, German, Japanese, Mandarin Chinese, Spanish, Turkish and Warlpiri (Bamberg, 1987; Bavin, 2000; Clancy, 1992; Gutierrez-Clellen & Heinrichs-Ramos, 1993; Hickmann & Hendriks, 1999; Jisa, 2000; Karmiloff-Smith, 1981; Ku¨ntay & Koc¸bas¸, 2009; Wigglesworth, 1990; Wong & Johnston, 2004). Little is still known about how children use referential forms in South and Southeast Asian languages. This chapter will focus on referential forms Thai children choose in narratives and the development of their referential strategies with age. Cross-linguistic studies of children’s referential forms in narratives All languages have some means to distinguish between new and given information. According to Chafe (1976), new information is described as ‘what the speaker assumes he is introducing into the addressee’s consciousness by what he says’ while given information is ‘that knowledge which the speaker assumes to be in the consciousness of the addressee at the time of the utterance’ (p. 30). While new information tends to be explicitly expressed, given information is likely to appear in attenuated forms such as pronouns or ellipses. While this seems a general principle that governs forms referring to new and given expressions universally, it is widely recognized that crosslinguistic variability exists regarding referential forms available along with aspects to be encoded. To illustrate this, English and French usually make use of nominal and pronominal forms; null forms are rare. These languages also require that a distinction concerning definiteness be grammatically expressed through definite/indefinite articles. On the other hand, in many Asian languages like Chinese, Japanese, Korean, Lao, Thai and Vietnamese, ellipses commonly take place, though not to the same extent. Moreover, definiteness is only optionally specified in these languages. This implicates that children acquiring different languages have to learn what acceptable and conventionalized forms in their languages are and which aspects of the grammar to which they need to pay attention. Adults’ production of referential forms is influenced by multiple factors, one of which is the discourse context (e.g. Clancy, 1980; Ratitamkul, 2007).

54

Theeraporn Ratitamkul

Different forms are used when the speaker introduces new referents, continues to talk about the same referent, or switches to another referent. Likewise, children have also been found to differentiate referential forms occurring in disparate contexts. Clancy (1992) examined narratives of Japanese children aged 3 years 8 months (3;8) to 7 years 4 months (7;4) and adults, which came from narration of a series of pictures and retellings of a short animation. She found that discourse context played an important role in children’s referential choice. At all ages, nominal forms were used more often when new characters were introduced than when there was a switched subject and used least often when the subjects remained the same in two consecutive clauses. The youngest children (3;8 to 4;0) used nominal forms to introduce new characters and to talk about switched subjects less frequently than children over 5 and adults. No age differences were found for same subject reference; elliptical forms were preferred by all age groups. Clancy concluded that the context of new introductions was the most difficult for young children, followed by the context of switched subjects and the context of continuing subjects, respectively. In another study by Wong and Johnston (2004), Cantonese-speaking children (3;0, 5;0, 7;0 and 12;0) told a story based on picture sequences. Their referential forms were later judged for informational adequacy. It was revealed that at 3 years old, children were still unable to produce informationally adequate forms. The 5-year-old group fared better than the youngest group, but it was the two older groups who were able to use appropriate forms according to the context. Given the development in referential adequacy, the most difficult context for the children was the reintroduction context in which a main character was reintroduced, followed by the introduction context in which a second new character was introduced, and the easiest context was the one in which children maintained reference to a known character. It can be observed that the order of referential difficulty in this study was not the same as that in Clancy (1992). This could have resulted from differences in criteria and methodology applied in the two studies. On the whole, previously conducted research on children’s referential selection in narrative discourse has reported that the ability to use referential forms at the discourse level appears to be a rather late development (e.g. Hickmann, 2003; Karmiloff-Smith, 1981). Based on an extensive review, Hickmann (2003) claims that children do not use referential forms for discourse-internal purposes until the age of 7 or 10, although different forms are available to children from a very young age. In Karmiloff-Smith’s (1981) study of French and English children aged between 4 and 9 years, it was found that children under the age of 6 predominantly used pronouns deictically. That is, their pronominalization heavily depended on paralinguistic gestures and on other extralinguistic information. The next step seen in

Referential forms in Thai children’s narratives

55

children’s pronominal use was that pronouns were reserved for the thematic subject. After an initial introduction of the main protagonist, only this character would be pronominalized and appear in a subject position. Children at this stage started to use pronouns anaphorically at the discourse level, indicating some sensitivity to narrative discourse structures. KarmiloffSmith’s study thus shows children’s acquisition of pronominalization, progressing from deictic usage to a discourse-internal device. Although it is clear that children as young as 2 years old can produce felicitous forms to convey new and given information in other types of discourse such as in conversations (e.g. Allen, 2000), children’s relatively late development of referential use in storytelling seems to stem from the complex nature of narrative discourse. Thai children’s narratives The present study analyses narratives of Thai children with the goal of answering two questions. First, what are the referential forms children use in different discourse contexts and, second, how does the ability to use referential forms to form a coherent discourse develop with age? Data came from the Thai Frog Stories (Zlatev & Yangklang, 2004), which can be found in the CHILDES database (http://childes.psy.cmu.edu/). Narrators consisted of ten participants from each of the following age groups: 4, 6, 9 and 11 years old and a comparison group of Thai adults. They told a story elicited through Mercer Mayer (1967)’s picture-book, Frog, Where Are You? The book is a story about a boy and his dog, who went on an adventure to look for his missing frog. Animate referents in the subject position in adults and children’s narratives were coded for referential forms (Proper name, Lexical, Pronominal or Null). Discourse contexts in which those forms occurred were also identified. Adapted from Jisa (2000), the contexts were Introduction (INT), Reintroduction (REIN), Promotion (PROM) and Maintain (MA). In the INT context, a new referent was introduced into the discourse. The other three contexts, on the other hand, contained old referents. In the REIN context, a referent had been mentioned earlier but not in an immediately preceding clause. The PROM context was the one in which a non-subject referent or a referent that was part of the subject of the previous clause was promoted to a subject position in the current clause. In the MA context, a referent was the subject of the previous clause and continued as a subject in the current clause. Examples of a referent in each context are given in (1) to (4), with a focused referent underlined. Excluded from the analysis were utterances that were irrelevant to the story, quotes, incomplete sentences and special constructions such as questions and fixed phrases.

56

Theeraporn Ratitamkul

(1)

INT context mii1 mǎa have dog ‘There was a dog.’

(2)

REIN context

(3)

PROM context

(4)

MA context

(a 6-year-old)

The four contexts in fact varied with respect to cognitive salience of the referents. According to Arnold (2003), a referent that appears as the subject of the previous clause is the most salient in the mind of the interlocutors. The second most salient is a referent that occurs in the previous clause but is not a subject. It could be an object, subpart of the subject or an object, or both a subject and an object of the previous clause. Next is a referent that has been mentioned earlier but not in the previous clause. And the least salient is a 1 2

An argument of the existential verb mii ‘to have’ was considered a subject in this study. In this study, the abbreviation CONN stands for connective, and the abbreviation PART stands for particle.

Referential forms in Thai children’s narratives

57

Table 4.1 Referential forms and discourse contexts Proper name INT 4-year-olds 6-year-olds 9-year-olds 11-year-olds Adults REIN 4-year-olds 6-year-olds 9-year-olds 11-year-olds Adults PROM 4-year-olds 6-year-olds 9-year-olds 11-year-olds Adults MA 4-year-olds 6-year-olds 9-year-olds 11-year-olds Adults

— — — 2 (5%) 2 (4%)

Lexical

38 (100%) 47 (100%) 40 (100%) 41 (95%) 43 (96%)

Pronominal

Null

— — — — —

— — — — —

— — 60 (26%) 70 (33%) 165 (54%)

146 (83%) 245 (93%) 136 (58%) 114 (54%) 77 (25%)

9 (5%) 9 (3%) 21 (9%) 12 (6%) 26 (8%)

20 (11%) 9 (3%) 17 (7%) 14 (7%) 39 (13%)

— — 14 (34%) 11 (21%) 25 (36%)

22 (81%) 28 (70%) 19 (46%) 22 (42%) 13 (19%)

— 6 (15%) 3 (7%) 7 (13%) 14 (20%)

5 (19%) 6 (15%) 5 (12%) 13 (25%) 17 (25%)

— — 20 (8%) 36 (16%) 34 (12%)

40 (37%) 75 (47%) 33 (13%) 22 (10%) 13 (5%)

15 (14%) 27 (17%) 41 (17%) 31 (14%) 48 (17%)

54 (50%) 56 (35%) 153 (62%) 131 (60%) 188 (66%)

referent that is newly introduced into the discourse. Hence, the order of reference saliency regarding context of occurrence from the highest to lowest is: MA, PROM, REIN, INT. As shown in Table 4.1, it was found that both children and adults used different referential forms in different discourse contexts. For each age group, there was a tendency for implicit forms (null and pronominal forms) to increase with the degree of salience. By contrast, increased use of explicit forms (lexical forms and proper names) was observed in contexts of decreased saliency. In other words, explicitness of the forms negatively correlated with the degree of cognitive salience; the more salient a referent, the less likely it was to be expressed with explicit forms. Results suggested that even the youngest group in this study showed some sensitivity towards differences in cognitive status of referents and adjusted their choice of referential forms accordingly. Comparing across age groups, all children and adults almost always used lexical forms when they introduced new referents in the INT context. In the

58

Theeraporn Ratitamkul

REIN context, where a subject was an old referent that had been mentioned before in an immediately preceding clause, children chose lexical forms most frequently while adults preferred proper names. Only 9- and 11-year-olds showed some use of proper names. Referents in the PROM context, where a referent in the previous clause was promoted into a subject position, also exhibited the same pattern. As for the context of reference maintenance (the MA context), adults and children, except the 6-year-olds, used null forms most often. A closer examination revealed that there were small discrepancies in referential selection between the two older groups and adults, while the 9-year-olds and 11-year-olds were not different from each other. In contrast, differences between the two younger groups and adults were striking. As an illustration, to maintain reference in two consecutive clauses, both 4- and 6-year-olds used a large number of lexical forms in comparison to adults (see Ratitamkul, 2010 for details about 4-year-old children). Unlike the younger groups, both adults and older children relied on null forms in this context. Furthermore, stories told by children in the two younger groups, by the 4-year-olds in particular, were more like picture descriptions than a coherent story. This is consistent with Clark’s (2003) observation, that young Englishspeaking children’s narratives are incoherent without the pictures. She also notes that 3- and 4-year-olds tend to use full noun phrases to refer to given referents, cases in which older children would use pronouns. Around age 5, children then start to add referential cohesion with pronouns. Interestingly, proper names could be observed only from 9 years old onwards. Four children in the 9-year-old group, five in the 11-year-old group and all ten adults used proper names. We can then see that the 9-year-olds in this study started to adapt their narrative styles to be compatible with the genre of storytelling by giving names to the main characters (the boy, the dog, and sometimes the frog). Moreover, adults’ narratives showed a repetition of proper names across sentences, which may appear rather strange in English but is not unacceptable in Thai. Likewise, narratives of both 9- and 11-yearolds also shared this pattern of use. Below are examples from a 9-year-old (Example 5) and an adult (Example 6). (5)

thɔm kà p fé nfaay cʉŋ cɔ̂ ŋmɔɔŋ thɔ̂ nmá ay ná n dû ay khwaamsǒ ncay . . . lέʔ thɔm kà p fé nfaay kɔ̂ dəən khâ w pay klâ y klâ y . . . thɔm bɔ̀ ɔk fé nfaay wâ a . . . ŋîap ŋîap sı`ʔ câ w fé nfaay . . . dι̌ aw kò p cà ʔ tʉ̀ ʉn mò t . . . lέɛw thɔm kà p fé nfaay kɔ̂ chaŋô ok nâ a loŋ pay duu lǎ ŋ thɔ̂ nsuŋ thɔ̂ n ná n . . . thɔm kà p fé nfaay dâ ay phó p kò p sɔ̌ ɔŋ tua yù u lǎ ŋ thɔ̂ nsuŋ ‘Tom and French Fries then looked at that log with interest . . . And Tom and French Fries then walked closer . . . Tom told French Fries . . . “Be quiet, French Fries . . . or the frog will wake up.” And Tom and French Fries then leaned forward to look behind that log . . . Tom and French Fries saw two frogs behind the log.’

Referential forms in Thai children’s narratives (6)

59

thɔm mâ y than rú utua . . . thɔm piin khʉ̂ n pay bon tô nmá ay . . . Ø hě n phrooŋ. . . thɔm kɔ̂ thı`aw hǎ a câ w kò p nɔ́ ɔy taam phrooŋmá ay ‘Tom was not aware . . . Tom climbed up a tree . . . (Ø) saw a hole . . . Tom then searched for the little frog in the hole of the tree.’

In general, Thai children and adults made referential decisions depending on context, which reflected their attention to the status of referents in the discourse. However, the two younger age groups, the 4- and 6-year-olds, differed noticeably from the adults. Apparently, those children were still learning to manipulate referential forms to produce a coherent discourse. They seemed to be able to use appropriate forms to introduce new referents earlier than to maintain reference. This is in contrast to the findings of Clancy (1992) and Wong and Johnston (2004) and can be explained by young Thai children’s overall preference for lexical forms. As for older children, the 9and 11-year-olds showed more similarities in referential use to adults than younger children. Instead of being just picture descriptions, their narration formed a narrative discourse, revealing appropriate referential forms for the genre of storytelling. By 9 years of age, Thai children are beginning to show the signs of becoming a skilled narrator. Conclusion An analysis of the Thai Frog Stories shows the development of children’s ability to use referential forms appropriately in narrative discourse. Thai children’s use of referential forms to create coherence in discourse appears relatively late (around 9 years old), conforming to research findings on child narratives in other languages. In fact, at 9 years of age, children’s overall narrative performance reaches a turning point (Berman, 2009). As Berman claims, children at this stage are capable of producing a story that shows both general narrative structure and detailed individual events, as well as adding their own personal comments. The development of Thai children’s narrative skills also corresponds to their syntactic growth. Piyapasuntra (2010) discovered that the Frog Stories of 9-year-olds contained greater syntactic complexity than those of the younger children. Moreover, the present study reveals that Thai children gradually acquire the referential strategies of their language. It should be noted that this study concentrates only on animate referents that appear in subject position. Further studies on referential acquisition in Thai need to include referents in other grammatical positions and constructions in which referents occur. Cross-linguistic comparisons would also prove fruitful in achieving a better understanding of the different referential strategies that children use when producing narratives.

5

The acquisition of tense and aspect Yasuhiro Shirai

The acquisition of tense-aspect markers has been an important topic in psycholinguistics, especially in relation to the roles of innate predisposition and cognitive development in language acquisition (Bickerton, 1981; Slobin, 1985). This debate has largely centred on the acquisition of European languages, such as French (Bronckart & Sinclair, 1973), Italian (Antinucci & Miller, 1976), Polish (Weist et al., 1984) and English (Bloom, Lifter & Hafitz, 1980; Shirai & Andersen, 1995). This chapter reviews more recent studies that have investigated the acquisition of Asian languages, and examines how these typologically distinct languages contribute to this issue. They include Japanese, Korean, Mandarin Chinese, Vietnamese and Thai. Terminology I will first review some terminology of tense and aspect, which forms the basis for understanding psycholinguistic issues in this area of inquiry. Tense locates a situation in relation to another reference point in time, and therefore is deictic. For example, English simple past tense locates a situation in relation to the speech time as the reference point. Aspect, on the other hand, concerns the temporal contour of the situation to be described by the verb (phrase). There are two levels of aspect that need to be considered: lexical aspect and grammatical aspect. Lexical aspect (also known as situation aspect or aktionsart) refers to the temporal semantic characteristics of the predicate, which Smith (1991) called ‘verb constellation’. Vendler’s (1957) four classes are the standard classification of lexical aspect: State: love, contain, know, think that..; Activity: walk, think about.., play the guitar; Accomplishment: make a chair, walk to the store; Achievement: reach the summit, die, notice; In terms of temporal semantic features, state terms refer to stative situations that do not change unless some other force changes them. States are stative, while the other three classes are dynamic. Among the dynamic terms, 60

The acquisition of tense and aspect

61

activities involve duration, but they do not have an inherent endpoint, that is, one can stop the action at any time point, but the fact remains that the action has been done. Activities are therefore atelic (non-telic), not involving an inherent endpoint. Accomplishments, on the other hand, are telic, involving a necessary endpoint. Finally, achievements refer to situations that are linguistically conceived as punctual or instantaneous. Achievements, therefore, are [þpunctual] and [þtelic]. In addition to Vendler’s four-way categories, Smith (1991) proposed the fifth category Semelfactives, which are atelic and punctual in that they do not result in change of states (e.g., jump, knock). Grammatical aspect, on the other hand, is typically realized by inflections and auxiliaries, which normally consist of lexical items that have taken on the grammatical meaning as a result of linguistic change (Bybee, Perkins & Pagliuca, 1994). It denotes what Smith (1983, 1991) calls ‘viewpoint aspect’, such as imperfective, perfective, progressive, which specifies how the speaker views the situation to be described. Perfective aspect presents a situation as a totality (external view), while imperfective aspect focuses on the internal structure of a situation (internal view) (Comrie, 1976). Progressive is a type of imperfective aspect, with an added component of dynamicity (i.e., dynamic imperfective).

Cognitive constraints on past tense-marking The acquisition of tense and aspect marking was first treated as a window into children’s conceptual development. Several studies published in the 1970s and the early 1980s showed that children’s past tense marking in various languages (e.g., English: Bloom et al., 1980; French: Bronckart & Sinclair, 1973; Italian: Antinucci & Miller, 1976; Turkish: Aksu, 1978; Greek: Stephany, 1981) is primarily restricted to change-of-state verbs, and some researchers (e.g. Bronckart & Sinclair, 1973; Antinucci & Miller, 1976) suggested that this is because of cognitive constraints. It was argued that children cannot code remote past using past tense marking because they do not have the notion of pastness (i.e. cognitive deficit); instead they use past tense marking to denote ‘change-of-state’, the resultant state which is cognitively more salient to children. This strong version of the ‘cognitive constraints’ account of children’s initial limited use of verbs was invalidated by Weist et al. (1984), who investigated the acquisition of Polish tense-aspect marking. They showed that Polish children can use past tense marking very early, and can use it to refer to not just result states, but also to remote past, marking truly deictic past. However, the fact remains that children’s past marking is primarily restricted to change-of-state verbs, which needs to be explained.

62

Yasuhiro Shirai

The Aspect Hypothesis Based on the observation in L1 acquisition of tense-aspect morphology, the following generalizations were proposed as a universal pattern of development (Shirai, 1991; Shirai & Andersen, 1995; Andersen & Shirai, 1996), and are often referred to as the Aspect Hypothesis: (1) Learners use (perfective) past marking on achievement or accomplishment verbs, eventually extending use to activity and state verbs. (2) In languages that encode the perfective–imperfective distinction morphologically, imperfective past appears later than perfective past, and imperfective past marking begins with stative and activity (i.e. atelic) verbs, and then extends to accomplishment and achievement (i.e. telic) verbs. (3) In languages that have progressive marking, it begins with activity verbs, and then extends to accomplishment/achievement verbs. (4) Progressive marking is rarely incorrectly overextended to stative verbs. (1) and (2) are based on the studies discussed above (e.g. French, Italian, Greek, English); (3) and (4) are mainly based on English (e.g., Brown, 1973; Kuczaj, 1978; Bloom et al., 1980). These generalizations are mostly agreed upon as descriptive statements regarding the L1 acquisition of tense-aspect markers (Shirai, Slobin & Weist, 1998), and are referred to using various labels (e.g. the aspect before tense hypothesis, the aspect first hypothesis, the lexical aspect hypothesis, the primacy of aspect hypothesis, the defective tense hypothesis). However, at the level of explanation – that is, why these patterns are observed – there is still no agreement in the field. One explanation proposed was the primacy of aspect over tense (Bloom et al., 1980). Bloom et al. noted that grammatical marking of aspect is more common than that of tense across languages of the world, and that aspect marking is physically closer to the verb stem than tense marking. This suggests, they argued, that aspectual notions are more salient than tense. Bickerton (1981, 1984) further argued that the restricted pattern is innate. He proposed the Language Bioprogram Hypothesis, based on the observation that genetically unrelated creole languages share common linguistic features in domains as varied as the causative construction, nominal specificity and tenseaspect modality. Bickerton hypothesized that these commonalties must come from a bioprogram – that is, knowledge that is innately specified for all human children. In the area of tense-aspect, he proposed that PNPD (the punctual–nonpunctual distinction) and SPD (the state–process distinction) were universally marked in creole languages, and therefore part of the bioprogram. Bickerton then argued that the data from tense-aspect acquisition support the bioprogram hypothesis. He interpreted (1) and (2) as evidence for PNPD, claiming that children are marking telicity (punctuality in his terminology)

The acquisition of tense and aspect

63

rather than pastness when they use past tense morphology, disregarding adult grammar. Regarding (3) and (4), Bickerton claimed that, because children rarely (if ever) make the error of incorrectly attaching progressive marking to stative verbs, they must already know the distinction between states and dynamic processes (i.e. SPD). Input distribution as explanation Shirai (1991, 1994; Shirai & Andersen, 1995; Li & Shirai, 2000) proposed an alternative account for the acquisitional patterns of tense-aspect morphology, which Bickerton had earlier attributed to the bioprogram. The proposal is that the pattern of acquisition is primarily driven by the distribution of the tenseaspect morphology in the input, and by children’s initial process of category formation. In reviewing the acquisition of Italian (Antinucci & Miller, 1976), Bickerton (1981, p. 174) suggested that in child Italian, perfective past and imperfects are in complementary distribution, the first being used for telic verbs, the second for atelic ones, which disregards the adult model. But while it may be possible to attach perfective or imperfective past forms to any verb types in adult grammar, not all verb classes are evenly attached to perfective and imperfective past forms in discourse. There are natural relationships between perfective past and telic verbs on the one hand, and between imperfective past and atelic verbs on the other. Leone (1990) reported that in an adult–adult native speech sample (an interview), 97% of perfective past forms are attached to telic verbs. If such a distributional bias is present in the input, it is not surprising that children follow this trend in acquiring tense-aspect morphology. Shirai tested this hypothesis by analysing the use of verbal morphology by three children acquiring English, and their mothers, from the CHILDES database (MacWhinney, 2000). Shirai and Andersen (1995) report that the earliest past inflections are mostly attached to achievement verbs (i.e. [þpunctual], [þtelic]). At the same time, however, they found that approximately 60% of past tense markers in the caregivers’ input occurred with achievement verbs which they argued to be the source for the near absolute tendency observed in the children’s output. Regarding (3), the restriction of progressive marking to activity verbs, Bickerton (1981) claimed that children acquiring English first mark nonpunctuality (i.e. lack of telicity) using -ing, and then later mark telicity by using irregular past, according to the distinction prescribed by the bioprogram. This, however, can be attributed to input: approximately 60% of the mothers’ uses of progressive are with activity verbs (i.e. [telic], [þdynamic]) (Shirai & Andersen, 1995).

64

Yasuhiro Shirai

Regarding (4), the lack of overextension of progressive onto stative verbs, Shirai (1994) showed that children do incorrectly attach progressive marking to stative verbs. It was found that one of the three children studied used progressive marking with state verbs frequently, and some of the uses were obviously ungrammatical from the adult’s point of view (e.g. *seeing light, *needing). The study further shows that the pattern of error is guided by parental input: the child who made frequent errors of this type was exposed to maternal speech in which the ratio of progressive used with stative verbs (a usage that is not necessarily ungrammatical; see Smith, 1983) is higher (3.8%) than observed in the speech of the caregivers of the other two children. In fact, the other two mothers never used progressive with stative verbs in the samples analysed (0%). In sum, the evidence from language acquisition that Bickerton discussed to support his bioprogram hypothesis can be accounted for by input-based learning. The strong initial correlation between punctual/telic verbs and past marking, and that between progressive/imperfective marking and atelic verbs, which Bickerton interpreted as children’s bioprogram-based marking of the punctual–non-punctual distinction (PNPD), can instead be attributed to the skewed distribution in the input. The lack of overextension of stative verbs, which supports Bickerton’s state–process distinction (SPD), was not observed in Shirai (1994), and incorrect use of stative progressive was attributed to input. Distributional learning and prototype-based initial representations The mechanism behind this input-driven early restriction is, it is argued, that of distributional analysis and prototype formation (see Shirai & Andersen, 1995; Andersen & Shirai, 1996; Li & Shirai, 2000). Children are not just passively imitating what caregivers say. They actively reorganize their linguistic representations based on the distributional information in the input, and create an initial prototype. In the case of tense-aspect morphology, children appear to be guided by temporal semantic features such as telicity and punctuality in creating the initial prototype. If we assume this type of learning mechanism on the part of children, we do not need to posit an innate bioprogram to explain the pattern of past tense acquisition in French and Italian, which Bickerton used as supporting evidence for his bioprogram hypothesis. Japanese Although the skewed distribution in the input suggests the strong possibility that input is the driving force rather than a bioprogram, the above studies cannot resolve this question. In particular, both the bioprogam hypothesis and

The acquisition of tense and aspect

65

the input-distribution hypothesis make the same prediction for English and other European languages. Both predict early association of past tense marking with telic verbs, and progressive marking with activity verbs. Japanese has an interesting organization of tense-aspect marking that can potentially resolve this issue because the two theories make contrasting predictions. The Japanese past tense marker -ta is similar to English in its meaning, and both theories predict that it will be associated with telic verbs. In contrast, the Japanese imperfective marker -te i- can denote not only progressive meaning (when combined with durative verbs), but also resultative meaning when combined with achievement verbs (Kindaichi, 1950; Shirai, 2000). Thus, Bickerton’s bioprogram would predict for Japanese that children first mark the punctual side of PNPD with past tense form -ta, and then mark the non-punctual side of PNPD with imperfective aspect marker -te i-. In contrast, the input scenario does not make the same prediction for -te i-, it only predicts that it depends on the input. In fact, in adult speech Japanese -te i- is most frequently used with achievement verbs and denotes resultative state (e.g. Shirai & Nishi, 2005). We first review research on the acquisition of Japanese to examine which theory fits the data better. Horiguchi’s (1981) longitudinal study showed that a child used -te i- to denote progressive earlier (at 1;10) than resultative (at 2;0). The boy also used the form to denote progressive more frequently at the earlier period (1;10 to 2;5, 68.9% and 2;6 to 2;11, 72.9%) than at the later period, when this preponderance was lost (3;0 to 3;5, 52.2%). She also reported on a girl studied by Okubo (1967), whose -te i- form appeared at 1;8, and who used it to refer to both progressive and resultative situations. She analysed all 60 tokens of -te i- used by the girl (1;8 to 3;5) and found that 61.7% were used for progressive meaning. Rispoli and Bloom (1985) argued that Japanese children have no problem acquiring -te i- regardless of whether it is used for progressive or resultative. They stated: ‘In Japanese, the feature continuative . . . can be coupled with the feature punctual . . . This arrangement causes Japanese children absolutely no problem. From as early as 22 months the same inflection is used to signify both continuing incomplete aspects and continuing complete aspects of events’ (p. 473). Cziko and Koda (1987) analysed the data from a published longitudinal corpus of a Japanese boy (Sumihare) from age 1;0 to 4;11. They found that punctuality did not necessarily correlate with the past inflection -ta (only about 50% of -ta were used with punctual verbs). This clearly contradicts the claim that the early past inflections are used with punctual/telic verbs. However, Shirai (1993) analysed the same published corpus, and showed that after a period of sporadic uses of -ta (only 10 tokens between 0;11 and 1;5), the correlation became high (73% at 1;6), and gradually declined

66

Yasuhiro Shirai

(66% at 1;7 and 59% at 1;8). Since Cziko and Koda’s ‘early use’ analysis included 1;9, 1;10, 1;11 and 2;0, during which time correlation is expected to become weak, it appears their analysis was not able to identify the earliest correlation. Shirai (1993) also found that the boy exhibited a period in which the use of -te i- was more frequent with activity verbs to denote action in progress than with achievement verbs to denote resultative state. Shirai (1998) analysed the longitudinal data of three boys (Aki, Yocchan, Taachan), and found that past tense -ta had a strong correlation with achievement verbs for all three children. On the other hand, individual differences were found for the emergence of -te i-. Yocchan’s data did not include the emergence of -te i-, and therefore I focus here on Aki and Taachan. First, Aki’s early use of -te i- was predominantly with activity verbs to denote action in progress. However, Taachan used -te i- with achievement verbs very frequently (84%) when his use of -te i- began to emerge at age 2;2. Shirai and Suzuki (in press) analysed longitudinal data from four children (Taro, Jun, Ryo and Tai) from CHILDES. None of these children showed particular preference for activity verbs used with -te i- and they slightly preferred achievements, about 60% of -te i- being used with achievements. The data from Japanese acquisition are not conclusive regarding the explanation of early semantic bias in the acquisition of tense-aspect morphology. On the one hand, there seems to be some preference for activity and progressive meaning, which is congruent with the bioprogram. On the other hand, the bioprogram does not predict individual variation to this degree. One possible source of the preference for activity/progressive is input. Shirai (1993, 1998) analysed input addressed to Sumihare and Aki. Both children showed preference for activity verbs with -te i-, and indeed input showed some indication of high frequency of activities even though in adult–adult speech achievement/ resultative is much more frequent with -te i-. This is rather preliminary because the input data recorded for Aki and Sumihare are not complete, but it is possible that because of discourse context (primarily ‘here and now’), there may be more activity/progressive used in child-directed speech than in adult-adult speech. To further test this possibility, Shirai and Suzuki constructed Table 5.1 to illustrate the pattern of input–output relationships. The studies conducted on Hiroki (Rispoli, 1981), H (Horiguchi, 1981) and Taachan (Shirai, 1998) did not look at input, so they cannot be informative. Among the studies that analysed input, Aki (Shirai, 1998) and Sumihare (Shirai, 1993) seem to indicate some degree of influence of input frequency, in that both the children and the caregiver showed some preference for activity verbs, as noted above. All the other cases show preference for achievements, both in child-directed speech and children’s use. This seems to indicate that the input frequency account explains the available acquisition data in Japanese better than the bioprogram account.

The acquisition of tense and aspect

67

Table 5.1 Preference of verb semantics in the use of -te i-(ru) by children and caregivers Children’s speech

Input

Hiroki (Rispoli, 1981) H (Horiguchi, 1981) Sumihare (Noji, 1976) Taachan (Okubo, 1982) Aki (Miyata, 1995) Taro (Hamasaki, 2002) Jun (Ishii, 2004)

No preference Progressive Progressive–Activity Resultative–Achievement Progressive–Activity Resultative–Achievement Slight preference for Achievement

Ryo (Miyata, 2004) Tai (Miyata, 2000)

Resultative–Achievement Resultative–Achievement

NA NA Progressive–Activity? NA Progressive–Activity? Resultative–Achievement Slight preference for Achievement Resultative–Achievement Resultative–Achievement

Chinese The acquisition of Chinese has also been investigated in relation to the issues relevant to the semantic bias in the acquisition of aspectual markers. Li and Bowerman (1998), in three cross-sectional experiments (comprehension, production and elicited imitation), investigated children’s (age 2;9 to 6;4) knowledge of Mandarin aspect markers with verbs of different lexical aspect. Their findings support the Aspect Hypothesis: in all tasks children associated the perfective aspect marker -le more frequently with telic verbs (achievement and accomplishment verbs) than with atelic verbs (activity and semelfactive verbs), and they associated the progressive aspect marker zai more often with atelic verbs than with telic verbs. Chen and Shirai (2010), in analysing longitudinal data from four children acquiring Mandarin Chinese, investigated the associations between lexical aspect and grammatical aspect markers. Although the findings are generally consistent with the Aspect Hypothesis, they also observed unexpectedly higher association between stative predicates and perfective aspect -le. They attributed this to the high frequency of this combination in the input the children received. Korean Go¨kmen and Lee (2002) investigated the acquisition of past tense marker -o ss- by analysing the longitudinal data from three children (1;2–2;4). The children used it almost exclusively (roughly 95%) with telic (accomplishment and achievement) verbs, which is consistent with the Aspect Hypothesis.

68

Yasuhiro Shirai

Table 5.2 Distribution of lexical aspect with -ko iss- by children

-ko iss-

Activity Accomplishment Achievement

JONG

JOO

YUN

13 (38.2%) 5 (14.7%) 16 (47.1%)

12 (66.7%) 1 (5.5%) 5 (27.8%)

16 (43.2%) 10 (27.0%) 11 (29.7%)

Table 5.3 Distribution of lexical aspect with -ko iss- by caregivers

-ko iss-

Activity Accomplishment Achievement

JONG

JOO

YUN

222 (67.5%) 37 (11.2%) 70 (21.3%)

489 (76.5%) 40 (6.3%) 110 (17.2%)

275 (65.8%) 36 (8.6%) 107 (25.6%)

Ryu and Shirai (in press) analysed longitudinal data from three Korean children (1;3–3;10), focusing on the imperfective aspect markers -ko iss- and -o iss- (dynamic durative and static durative, respectively: Lee, 1991). Since the status of -o iss- is not clear in relation to the issue of input versus bioprogram, here I focus on the results concerning -ko iss-. Although it primarily marks progressive aspect, it denotes resultative state when combined with transitive achievement verbs of wearing, carrying and body posture (Shirai, 1998; Lee & Kim, 2007). The distribution of the children’s use of -ko iss- is shown in Table 5.2 (calculated based on Table 7 of Ryu and Shirai, in press). As can be seen, the results are not necessarily consistent with the Aspect Hypothesis. First, achievements are quite frequent, even more frequent than activities for JONG. JOO’s results are the most consistent with the Aspect Hypothesis with the highest ratio of activity verbs. This distributional pattern is also found when the first ten tokens produced by the children were analysed. What then is adults’ use of -ko iss- like? Table 5.3 shows the distribution for the caregivers (calculated from Ryu and Shirai’s Table 10). The caregivers’ use of -ko iss- appears to be more consistent in that all three children are exposed to -ko iss- in the frequency order of activities (66–77%) ! achievements (17–25%) ! accomplishments (6–11%). This distribution, however, is not always followed by the children. JONG seems to have ignored input frequency while JOO and YUN exhibited use more similar to the adult model. Therefore, acquisition of Korean aspect seems to be characterized more by individual variations than Japanese or Chinese, which appears to be more dependent on input frequency.

The acquisition of tense and aspect

69

Table 5.4 Use of kam0laŋ0 (progressive) and lε:w3 (perfective) markers by lexical aspect and age State

Activity

Accomplishment

Achievement

4 years kam0laŋ0 lε:w3

0 3

8 1

4 2

3 3

6 years kam0laŋ0 lε:w3

0 2

9 0

1 0

2 10

9 years kam0laŋ0 lε:w3

1 1

9 1

2 0

2 8

Adults kam0laŋ0 lε:w3

0 2

4 3

2 0

2 13

Vietnamese Tran and Ud Deen (2003) analysed natural interaction data of a Vietnamese girl at 1;9 for their use of aspect markers. Although their focus was different, their results can be interpreted from the Aspect Hypothesis perspective. They coded all the predicates that appeared with perfective aspect marker roi for their lexical aspect values. Of the 23 predicates, 4 were states, two were activities, 7 were accomplishments, and 10 were achievements, that is, 73.9% of perfective aspect markers were used with telic verbs, consistent with the Aspect Hypothesis.

Thai Winskel (2007) analysed Frog Story narratives (Berman & Slobin, 1994) from 4-, 6- and 9-year-olds and adults from CHILDES to investigate the acquisition of temporal expressions by Thai children. Although not her main focus, Winskel reported quantitative analysis of lexical aspect used with different aspectual morphemes. She found that the association between lexical aspect and aspectual morphemes are stronger in adults than in younger children, which goes against the prediction of the Aspect Hypothesis. Regarding the progressive marker kam0laŋ0, all groups used it most frequently with activity verbs, However, the concentration is not the highest among 4-year-olds (53.3%), but among 6- (75.0%) and 9-year-olds (69.2%) (refer to Table 5.4). Perfective lε:w3 also shows the same tendency in that

70

Yasuhiro Shirai

its concentration with the prototype (i.e. achievements) is higher for 6- and 9-year-olds (83.3% and 80%), rather than the youngest group (33.3%). This is not consistent with the Aspect Hypothesis in that it predicts the association is strongest at the earliest stage of aspectual development. This may be due to the fact that this study analysed elicited picture description narratives. It has been pointed out that younger children tend to give static description of the pictures rather than give a sequenced narrative in telling a Frog Story narrative (Berman & Slobin, 1994). Further research with naturalistic, longitudinal studies on Thai will help clarify this issue. Summary and conclusion In this chapter, I reviewed the acquisition of Asian languages with regard to the relationship between lexical aspect and grammatical aspect/tense. Although the bulk of research so far has been on Japanese, recently more research has been conducted on other Asian languages (Chinese, Korean, Thai and Vietnamese). We can conclude the following based on the review: (1) The Aspect Hypothesis is generally supported in that past tense (Japanese and Korean) or perfective aspect (Chinese, Vietnamese and Thai) are strongly associated with telic verbs (achievements and accomplishments), and that progressive marking (Korean, Chinese, Vietnamese and Thai) are associated with activity verbs. (2) Regarding the explanation for the semantic bias, input frequency is more plausible than innate predisposition (i.e. a bioprogram) because of the strong association between caregiver speech and children’s speech for Japanese -te i- and Chinese -le. (3) Korean progressive marking exhibits clear individual variations which cannot be explained either by a bioprogram or input frequency. (4) The effect of task appears to explain the lack of support for the developmental prediction of the Aspect Hypothesis (Thai). The issue of semantic bias in tense-aspect acquisition has been extensively investigated in the past 40 years, and hotly debated in relation to the issue of nature versus nurture (Shirai, 2009). Although the debate in first-language acquisition mostly centred around European languages in the first 20 years, the last 20 years have seen more and more research on Asian languages, which has revealed that input is even more important than was thought, and that individual and task variations are more conspicuous than we had been assuming. Further research in Asian languages with multiple tasks would help reveal the true nature of universals and particulars of language acquisition.

6

The acquisition of Malay numeral classifiers Khazriyati Salehuddin

Most Sino-Tibetan and Austronesian languages have numeral classifiers (Adams & Conklin, 1973; Aikhenvald, 2003; Allan, 1977; Craig, 1986). Numeral classifiers are a syntactic–semantic category that is typically used in counting or referring to objects. Syntactically numeral classifiers occur predominantly as independent lexemes that occur contiguously with numerals and expressions of quantity in a noun phrase (Craig, 1986). Semantically, numeral classifiers explicitly classify nouns by denoting ‘some salient perceived or inputted characteristic’ of the entity in question (Allan, 1977, p. 285). This semantic classification includes categorising objects based on their physical, conceptual, and functional properties. Categorisation and labelling are essential processes in human cognition and language development. This process is influenced by the knowledge and information the speaker has about a particular object (Keil, 2003; Rosch et al., 1976). In numeral classifier languages both cognition-based universal features (e.g. Adams & Conklin, 1973; Allan, 1977) and culturebased idiosyncratic features (e.g. Yamamoto & Keil, 2000) influence the categorisation of objects into the classifier categories. Studying numeral classifier acquisition enables researchers to examine how children learn to categorise and label objects in their environment using a constrained framework and how this ability develops and becomes more refined over time. The present chapter gives an overview of the research on Malay numeral classifier acquisition. First, a brief introduction to the Malay numeral classifier system is given. Then the research that has so far been conducted on the acquisition of the Malay numeral classifier system is reviewed. Finally, further possible studies investigating numeral classifier acquisition are discussed. Introduction Human beings consciously and unconsciously assign everyday objects and items into particular linguistic and non-linguistic categories. When categorising and labelling a particular object, stored information is retrieved and 71

72

Khazriyati Salehuddin

evaluated, and only when there is a satisfactory resemblance between new and stored information is an object accepted into the respective category (Barsalou, Huttenlocher & Lamberts, 1998). Concurrently, those objects that are considered dissimilar are categorised into different categories (Hampton, 1998). This categorisation process occurs in relation to numeral classifier languages, but the way objects are categorised into these categories can be somewhat arbitrary and specific to the particular classifier language. The categories used by numeral classifier languages have both shared and idiosyncratic characteristics. The Malay numeral classifier system Malay numeral classifiers occur in noun phrases, with a numeral preceding and a noun following the numeral classifier. Num Tiga Three ‘three children’

NumCl orang NumCl

N kanak-kanak child

Malay numeral classifiers are divided into two major categories, that is, animate and inanimate (Figure 6.1). For animate nouns such as children (anak) and cats (kucing), their classification is relatively straightforward and semantically transparent as all humans are classified with orang (which literally means ‘human’), and all animals are classified with ekor (which literally means ‘tail’). However, for nouns such as ‘cars’ (kereta) and ‘cups’ (cawan) classification is not straightforward as ‘cars’ are classified with buah, which literally means ‘fruit’, and ‘cups’ with biji, which literally means ‘seed’. Moreover, nouns that are homonymous with the numeral classifiers never collocate with their respective homonyms (e.g., *tiga buah buah and *tiga biji biji); instead, they are classified with other numeral classifiers, for example tiga biji buah (three seed fruits) and tiga butir biji (three grain seeds). While the classification of animate items is straightforward, the classification for inanimate items is more complex. The classification of inanimate items in Malay in terms of shape involves mixed semantic criteria. The assignment of long (1D), flat (2D) and rounded (3D) objects into their respective numeral classifier category is also based on rigidity (for 1D and 2D objects) and size (for 3D objects) criteria. In Malay there is not a clearly designated general classifier, as occurs in other numeral classifier languages such as Chinese, Japanese and Thai (Salehuddin, 2010).

The acquisition of Malay numeral classifiers

73

NOUN COUNT

MASS

ANIMATE

INANIMATE

HUMAN

ANIMAL

LITTLE

MUCH





SPECIFIC

SHAPE SIZE

DIMENSIONALITY

RIGIDITY 1-D

2-D

3-D

+ RIGID





– RIGID



FINE SMALL MEDIUM BIG ABSTRACT



FLOWER FIREARM KNIFE

Figure 6.1. Classification of Malay numeral classifiers (adapted from Salehuddin & Winskel, 2008).

Acquisition of Malay numeral classifiers Previous research on the acquisition of numeral classifiers has been conducted on Japanese (e.g. Matsumoto, 1985, 1987; Naka, 1999; Sanches, 1977; Uchida & Imai, 1999; Yamamoto, 2009), Chinese (e.g. Chien, Lust & Chiang, 2003; Erbaugh, 1986; Loke & Harrison, 1986; Ng, 1989; Uchida & Imai, 1999; Wei & Lee, 2001) and Thai (e.g. Carpenter, 1991; Gandour et al., 1984; Tuaycharoen, 1984). Prior to studies by Salehuddin and Winskel (2007), Malay numeral classifier acquisition has not been previously studied. The first Malay numeral classifiers acquired by children are the animate numeral classifiers followed by the inanimate numeral classifiers (Salehuddin & Winskel, 2007). Within the animate numeral classifier category, children produce the numeral classifier for humans (orang) prior to the numeral classifier for animals (ekor). This trend is similar to results found in Japanese (Hansen & Chen, 2001) and Chinese (Erbaugh, 1984). In an elicited production task with 140 6- to 9-year-old Malay children, Salehuddin and Winskel (2009a) found that the youngest children studied (6- and 7-year-olds) tended to initially omit numeral classifiers when they counted the number of items, whereas older children (8- and 9-year-olds)

74

Khazriyati Salehuddin

35 30 25

Correct

20

Alternative classifier No classifier

15

Others 10 5 ts ul Ad

ds 9ye a

r-o l

ds r-o l 8ye a

r-o l 7ye a

r-o l 6ye a

ds

0 ds

Mean number of correct responses

40

Age group

Figure 6.2. The different types of responses made by children in the counting production task (Salehuddin & Winskel, 2009, p. 304).

tended to substitute an alternative numeral classifier in place of the correct numeral classifier (see Figure 6.2). It was found that most of the alternative numeral classifiers that children used in place of the correct numeral classifier shared the same dimensionality or rigidity features with the correct numeral classifier. For example, biji [3D: þrigid] (a 3D numeral classifier) was predominantly used (77%) in place of other 3D numeral classifiers (i.e. buah, ketul and butir) in preference to 1D (8%) and 2D (15%) numeral classifiers. Similarly, batang [1D: þrigid] (a numeral classifier for rigid items) was predominantly used (61%) in place of keping [2D: þrigid]. These results suggest that children are having difficulty in selecting the appropriate numeral classifier due to difficulty differentiating between finer distinctions within the same dimensionality category. The production of Malay numeral classifiers appears to be a relatively delayed and prolonged process (Salehuddin & Winskel, 2009a) in comparison to other numeral classifier systems (Chinese: Erbaugh, 1984, 1986; Ng, 1989; Japanese: Matusumoto, 1985; Thai: Carpenter, 1991; and Vietnamese: Tran, Chapter 7). At the age of 6, Malay children produced only 5% correct numeral classifier responses. Furthermore, the 9-year-old Malay children (51%) lagged behind the adults (88%) (Salehuddin & Winskel, 2009a), which indicates that Malay numeral classifier acquisition is a relatively delayed and prolonged process and that acquisition continues to develop into late childhood and adolescence.

The acquisition of Malay numeral classifiers

75

Table 6.1 Malay shape-based numeral classifier acquisition based on the mean number of correct production responses across all children’s age groups (standard deviations are in parentheses) Numeral classifier

Mean number of correct responses

helai [2D: rigid] batang [1D: þrigid] buah [3D: big] biji [3D: small] utas [1D: rigid] keping [2D: þrigid] ketul [3D: medium] butir [3D: fine]

0.53 0.43 0.36 0.35 0.24 0.18 0.12 0.05

(0.40) (0.27) (0.32) (0.27) (0.32) (0.24) (0.22) (0.17)

Source: Salehuddin & Winskel (2009a).

The production order of Malay shape-based numeral classifiers began with a 1D and a 2D numeral classifier (i.e., batang [1D: þrigid] and helai [2D: rigid]), followed by the production of two 3D numeral classifiers (i.e. buah [3D: big] and biji [3D: small]) (Table 6.1). There are two forms for 1D and 2D numeral classifiers (i.e. þrigid and –rigid) and four forms for 3D numeral classifiers (i.e. big, medium, small, fine). The production of a 1D and a 2D numeral classifier prior to that of 3D numeral classifiers is consistent with semantic complexity theory (Clark, 1973; Matsumoto, 1987), whereby objects with a higher cumulative number of semantic features are categorised later than those with a lower cumulative number of semantic features (Clark, 1973). These results are similar to those found for Chinese and Japanese numeral classifier acquisition, where numeral classifiers involving more complex features are acquired later than those involving less complex features (Uchida & Imai, 1999). However, the production order of Malay shape-based numeral classifier is not entirely consistent with semantic complexity theory (Clark, 1973). The other 1D and 2D numeral classifiers (i.e. utas and keping) were produced only after the production of buah and biji. This can be explained in terms of the children’s linguistic environment and input. Buah and biji were found to be the most frequently occurring numeral classifiers in a Malay children’s corpus in comparison to other shape-based numeral classifiers (Salehuddin & Winskel, 2009a). Hence, this could be a contributing factor to the relatively early emergence of buah and biji in comparison to utas and keping. Children comprehended Malay shape-based numeral classifiers (Salehuddin & Winskel, 2009b) earlier than they produced them (Salehuddin & Winskel, 2009a), which is consistent with the general claim that children tend to comprehend a particular linguistic construction before they

76

Khazriyati Salehuddin

Table 6.2 Malay shape-based numeral classifier acquisition based on the mean number of correct matching comprehension responses across all age groups (standard deviations are in parentheses) Numeral classifier

Mean number of correct responses

helai [2D: rigid] batang [1D: þrigid] keping [2D: þrigid] utas [1D: rigid] ketul [3D: medium] buah [3D: big] biji [3D: small] butir [3D: fine]

0.66 0.65 0.61 0.51 0.47 0.44 0.37 0.36

(0.32) (0.23) (0.26) (0.36) (0.28) (0.35) (0.30) (0.34)

Source: Salehuddin & Winskel (2009b).

appropriately produce it (Clark, 1973, 2003; Conroy & Lidz, 2007; Gennari & MacDonald, 2009; Ingram, 1974) and with previous research on numeral classifier acquisition (Hu, 1993; Uchida & Imai, 1999). However, the comprehension order (Salehuddin & Winskel, 2009b) (Table 6.2) does not directly correspond to the production order (Salehuddin & Winskel, 2009a). In the comprehension task, children had difficulty matching the two 3D numeral classifiers (buah and biji) with the appropriate object exemplars. However, in the production task, biji and buah were frequently used as default numeral classifiers probably as a result of their availability in the children’s linguistic environment and their broader application or usage in comparison with other Malay shape-based numeral classifiers. This disparity in production and comprehension order obtained could be due to the broader usage of these classifiers. According to Clark (2003), the usage of a default numeral classifier is similar to children’s early overextensions of meaning of lexical terms (e.g., ‘dog’ is initially used to refer to all four-legged animals). In relation to the acquisition of Malay shape-based numeral classifiers, usage of buah and biji (the frequently occurring numeral classifiers in the Malay children’s linguistic environment) as default numeral classifiers helps reduce the semantic load involved in learning numeral classifiers (see Gandour et al., 1984). Children’s act of overextending the usage of buah and biji decreases when appropriate numeral classifiers are acquired, and when children begin to restrict and refine their use of the semantic rules for the numeral classifiers buah and biji. The degree of typicality of objects is an additional factor that influences the production and comprehension of Malay numeral classifiers. Children comprehended and produced correct numeral classifiers and with faster response times (Salehuddin & Winskel, 2011) for typical rather than atypical

The acquisition of Malay numeral classifiers

77

exemplars of classifier categories (Salehuddin & Winskel, 2009a, 2009b). These findings concur with previous research on numeral classifier acquisition (e.g. Carpenter, 1991; Matsumoto, 1985; Uchida & Imai, 1996, 1999) and previous categorisation studies (Gelman & Coley, 1990; Hampton, 1998). According to Mervis and colleagues (Mervis & Pani, 1980; Rosch & Mervis, 1975), the redundancy of semantic features associated with typical exemplars of a particular numeral classifier category facilitates categorisation. The more the stored and the succeeding objects overlap with each other, the easier and faster objects are categorised (Mervis & Pani, 1980; Rips, Shoben & Smith, 1973; Rogers & McClelland, 2004; Rosch, 1973). This is in line with the view that category formation occurs through learned experience from categorisation of exemplars and prototypes (Barsalou et al., 1998). The disparity between typical exemplars and members of different categories also facilitates categorisation in young children (Mervis & Pani, 1980; Rips, Shoben & Smith, 1973; Rogers & McClelland, 2004; Rosch, 1973). In a ‘select the odd-one-out’ task that required children to select one exemplar that did not belong to a particular numeral classifier category, young Malay children initially had difficulty in identifying atypical exemplars as being members of a particular Malay numeral classifier category. As a result, they selected the atypical exemplars as the ‘odd-one-out’ and tended to categorise perceptually similar non-members into the numeral classifier categories (Salehuddin & Winskel, 2010). For example, younger children tended to select the atypical exemplar of biji [3D: small] (‘plate’) as a non-member of the biji numeral classifier category and yet included ‘planet’ in the category. Since the ‘redundancy structure of the category as a whole’ (Rosch & Mervis, 1975, p. 602) is often not reflected in atypical members, atypical exemplars are more likely to be excluded by younger children. In categorising numeral classifier exemplars, it appears that younger children do not have a complete understanding of the semantic rules of numeral classifier categories and hence rely on perceptual similarities of exemplars per se to determine numeral classifier membership. A greater awareness of the rules associated with category membership gradually develops. The acquisition of Malay numeral classifiers is a relatively delayed and prolonged process in comparison to other numeral classifier languages. One plausible explanation for this is the usage and the degree of obligatoriness of Malay numeral classifier usage. Malay numeral classifiers predominantly occur when counting or in referring to an already mentioned item (Hopper, 1986; Salehuddin, Winskel & Maros, 2011). An important consideration is that when numeral classifiers are not used, speakers can still be understood. A study on caregiver–child interactions in a counting game showed that interaction between caregivers and their children took place with no difficulty even though numeral classifiers were largely absent (Salehuddin & Winskel,

78

Khazriyati Salehuddin

2011, 2012). Malay numeral classifiers appear to be less obligatory in usage than in other numeral classifier languages (e.g. Thai: Carpenter, 1986; Japanese: Matsumoto, 1985; Chinese: Erbaugh, 1986; and Vietnamese: Tran, Chapter 7). This is a contributing factor to the relatively late acquisition of Malay numeral classifiers. Another consideration is that the categorisation of some objects into a particular Malay numeral classifier category is not always clear-cut as there is a varying degree of flexibility or choice in how some objects are classified. Some objects are more difficult to categorise as they can be placed into several different categories based on sometimes quite fine perceptual distinctions. For example, coklat (‘chocolate’) can be classified with ketul [3D: medium] if it is perceived as chunky, or with biji [3D: small] if it is perceived as small, or with batang [1D: þrigid] if it is perceived as long (finger-like). Hence, the manner some items are sorted into different categories by different speakers can vary to quite an extent. In conclusion, the categorisation of objects into the Malay numeral classifier system is a complex process that involves building up a comprehensive knowledge of complex inherent semantic characteristics of the different numeral classifier categories. The acquisition of Malay numeral classifiers is a relatively delayed and prolonged process. Future research can be conducted to investigate if the same developmental pattern of Malay numeral classifiers is identified in the acquisition of Malay numeral classifiers among non-native speakers of Malay who are learning or acquiring Malay as a second or third language, in particular those whose mother tongue is a numeral classifier language such as Chinese.

7

The acquisition of Vietnamese numeral classifiers Jennie Tran

Overview From a semantic perspective, numeral classifiers are unbound function words that categorize the head noun based on inherent or salient features of the noun’s referent, such as animacy, shape, function or arrangement. The classifier system is not easy for children to acquire. The complex semantic nature of a classifier system makes the learning of the full set of numeral classifiers difficult and slow. There are many different classifiers to learn, and the semantic relation between classifier and noun is often quite intricate. So far, there have been studies on numeral classifier development in six languages – Cantonese, Japanese, Korean, Mandarin, Malay and Thai. East Asian languages have been the most studied; the only Southeast Asian languages hitherto investigated are Thai and Malay (see Salehuddin, Chapter 6). This chapter presents the acquisition of the numeral classifier system of a third Southeast Asian language, namely Vietnamese. To date, there has been no other study of the acquisition of Vietnamese classifiers. Since Vietnamese has a rich and complex classifier system with many classifier categories, an investigation of its developmental pattern adds cross-linguistic depth to the classifier acquisition literature. The current chapter examines which Vietnamese classifiers are acquired first by young Vietnamese-speaking children, in what order they learn them, and what errors they make.

Previous research on the semantic development of classifiers Evidence from previous cross-linguistic studies (e.g. Carpenter, 1987; Erbaugh, 1986; Uchida & Imai, 1996; Wong, 1998) shows that children employ certain strategies in the process of acquiring the semantic specialization of classifiers. A predominant strategy up to age 4 is to use a general classifier. Children use it to replace classifiers that they have not yet acquired. Its use as a place-holder and its overuse for many referents continue even up to age 7, as reported by Carpenter (1987) and Salehuddin and Winskel (2009). Specific classifiers are acquired relatively late. 79

80

Jennie Tran

Among the specific classifiers, there are conceptual preferences. The animal classifier is reported by most studies as the one most appropriately used and earliest acquired. Such findings are consistent with Clark’s (1977) proposal that animacy is one of the first and most frequent distinctions in children’s early noun overextensions. Other classifiers that are acquired relatively early are shape and then function classifiers. Within the shapebased classifier assignments, features most frequently generalized are vertical extensions and small size (Erbaugh, 1986). Within the function-based classifier assignments, transportation seems to be the most salient. Both in Cantonese (Wong, 1998) and Mandarin (Loke & Harrison, 1986; Mak, 1991) the classifier used for vehicles is the first one acquired among function-based classifiers. This developmental pattern is indeed in line with the theory of noun acquisition that holds that when young children define noun meanings, shape is more important than function (Bowerman, 1975; Clark, 1973). Carpenter (1987) suggests that shape and function are the two most important criteria for innovative class extensions and erroneous generalization of classifier categories. However, unlike the general trend, in Thai, shape-based classifiers are acquired later than classifiers referring to arrangement and quanta (Gandour et al., 1984). This is possibly due to the fact that Thai has many semantically complex shape classifiers that have a large number of contrasts encoded in them. It has six classifiers for one-dimensional, four for two-dimensional and four for three-dimensional objects. In contrast, the Japanese one-dimensional classifier hon and two-dimensional classifier mai are acquired relatively early, according to Matsumoto’s (1985) report, because they are semantically less complex. In general, as reported in Fang (1985), classifiers with a limited semantic field whose referents are limited – to books, for example, or to paired items – are acquired before those with a large semantic field, such as two- and three-dimensional objects. Wong (1998), however, made an interesting contrary observation in this respect: classifiers that children acquire first are those that refer to more than a single type of noun. For example, a classifier that refers to long and slender objects only is acquired later, but a classifier like the animal classifier is acquired earlier because it refers to three different types of noun (animals, utensils and some body parts). Thus, what seems to facilitate acquisition is wider distributional potential and a one-to-many mapping of forms (disjunctive) rather than a one-to-one mapping (non-disjunctive). Most of the previous classifier acquisition studies were elicited production studies. Erbaugh’s (1982, 1986) and Wong’s (1998) studies are the only two longitudinal studies examining naturalistic data from very young children at regular recording intervals. Their participants were in the age ranges of 1;10–3;10 and of 1;09–2;09, respectively. These studies provide valuable information on the first appearance of classifiers. All the other studies are

The acquisition of Vietnamese numeral classifiers

81

experimental and cross-sectional. The only studies that examined the syntactic properties of classifier development are those by Erbaugh (1982) and Hu (1993) on Mandarin, Carpenter (1987) on Thai and Wong (1998) on Cantonese. All the others focus on semantic aspects of classifiers. Research on the acquisition of Vietnamese classifiers Two different types of data collection methods – longitudinal naturalistic and cross-sectional experimental elicited-production – have been utilized to collect data on the acquisition of Vietnamese classifiers in young children. The focus was solely on children’s production and not on their comprehension of classifiers. The longitudinal naturalistic data was collected from four children from monolingual families living in Vietnam, ages 1;9, 1;11, 2;4 and 2;5, twice a month over a period of 6 to 9 months. Table 7.1 provides information about these children. Each recording session lasted about 1 hour of interaction between the child and his/her parents or caregiver in the child’s home. All sessions were both audio-taped and video-recorded. The recording scenarios were playtime, mealtime, bedtime, bath time and television time. Various types of toys, props, puppets, pictures and picture-books were utilized to encourage parents/ caregivers and children to use various types of nouns, since the targeted syntactic constructions were classifier phrases involving nouns. For uncontrolled elicitation, picture-books and toys were used. The cross-sectional experimental elicited-production data was collected every day for a period of 6 weeks from 38 children between the ages 2;10 and 5;7 in a daycare centre in Ho Chi Minh City. The 38 children were divided into three groups (Table 7.2). The aim of the production task was to investigate the obligatoriness of classifiers by eliciting constructions with numerals. The targeted syntactic structure is numeral þ classifier þ noun. The children’s responses were audiorecorded. There were 114 stimuli, including pictures, toys and real objects. The child was shown pictures of multiple objects and entities on picture cards Table 7.1 The children in the longitudinal study Name

Gender

Age range

Duration of recording

Minh Ha` Mi Lieˆm Giang

Boy Girl Boy Girl

1;9–2;3 1;11–2;5 2;4–2;10 2;5–3;2

6 6 6 9

months months months months

82

Jennie Tran

Table 7.2 The children in the cross-sectional study Group

Gender

Age range

I (youngest) II (mid) III (oldest)

6 girls, 5 boys 7 girls, 5 boys 8 girls, 7 boys

2;10–3;7 3;8–4;4 4;7–5;7

and asked to say how many of the same objects s/he sees. The question itself did not contain a classifier. The quantity of the objects ranged from two to ten. There were 58 different pictures and objects, which were used to elicit 17 different classifiers. For data collection, both familiar and unfamiliar objects and entities were equally used. The longitudinal data provide evidence for the very first Vietnamese classifiers that young Vietnamese-speaking children acquire, the order they acquire them, and the errors they make. The later acquisition of classifiers (ages 3–5) is explored using the cross-sectional data as evidence.

Classifiers in Vietnamese The following five classifiers are disjunctive and were elicited in the production task. General classifier: (1) ca´i (a) general classifier (b) body parts (c) inanimate objects (e.g., furniture, kitchenware, utensils, appliances, tools) (d) some food items (e) abstract entities Specific classifiers: (2) con (a) animals (b) human beings (e.g., person, boy, daughter) (c) animate entities (e.g., ghost, devil, eye, heart, number, street) (d) objects with animate connotations (e.g., doll, knife, boat) (3) quả or tra´i (a) fruits and vegetables (b) roundish, big, spherical objects/entities (e.g., ball, egg, bomb, grenade, heart, kidney, globe, earth, mountain, hill, pendulum)

The acquisition of Vietnamese numeral classifiers

83

(4) chie´ˆ c (a) vehicles (b) one of a pair (c) secondary general classifier for certain sophisticated objects (5) caˆy (a) plants (b) long-straight-rigid, plant-like objects Classifiers used solely for indicating specific shapes are referred to as nondisjunctive or as displaying one-to-one mapping. Eight such classifiers were elicited in the production task. (6)

sợi /cọng long, thin, small, flexible, string-like (e.g. hair, thread, string, cord, rope, necklace, noodle)

(7)

tờ flat, thin, sheet-like objects (e.g. paper, newspaper, calendar sheet)

(8)

taˆ́ m flat, thin, hard objects (e.g. picture, photograph, mirror, carpet, cloth, board)

(9)

lá small, light, flat–thin, leaf-like objects/entities (e.g. card, flag, letter, sail, liver, lung)

(10)

cục small objects/entities with no defined shape (e.g. battery, bone, eraser, block, meatball, coal)

(11)

vieˆn small roundish objects of same material and having same shape (e.g. pill, tablet, candy, bullet, brick, tile)

(12)

hoˆṭ /hat tiny, oval, flat, seed-like objects/entities ˙ (e.g. grain, seed, salt, rice, egg, button, raindrop)

(13)

cuoˆṇ roll-like objects/entities (e.g. toilet paper, cloth, wool, film)

Two non-disjunctive classifiers not indicating shape were elicited in the production task. (14) (15)

củ root vegetables (e.g. onion, garlic, carrot, potato, manioc, turnip, cabbage) quyển/cuoˆ´n

book, volume, notebook, calendar, diary, novel

Emergence order of the first classifiers Based on the results and the analysis obtained from both the longitudinal data of four children and the elicited production task with 38 children, the

84

Jennie Tran

Table 7.3 Proposed order of emergence of Vietnamese classifiers Classifier

Meaning

Time of emergence

con ca´i quả/tra´i caˆy cuc ˙ củ chieˆ´c

animal, animate (only animal category) general, inanimate fruit/3D, big–round 1D, long–straight-rigid 3D, small–roundish, undefined root vegetable vehicle, one of a pair, secondary general (only vehicle category) 1D, long–thin-flexible 2D, flat–thin, sheet-like volume-like 2D, flat–thin, hard 3D, small–round 3D, small–roundish, seed-like roll-like 2D, flat–thin, leaf-like

before 1;9 before 1;9 before 1;9 1;10–2;2 1;10–3;7 from 2;0 from 2;2

sợi tờ quyển ta´ˆ m vieˆn hoˆṭ cuoˆṇ la´

from 3;8 " " from 4;7 " " " " or later

following preliminary order of emergence was proposed for Vietnamese classifiers (Table 7.3). Both the longitudinal and cross-sectional data confirm that the five earliestacquired classifiers are con, ca´i, quả/tra´i, caˆy and cuc. Interestingly, the ˙ general classifier ca´i was not the classifier with the highest percentage of correct use; it was instead second to the animal–animate classifier. The animal classifier was the earliest-acquired classifier, and also the one with which children made the fewest errors. Củ and chieˆ´c were acquired at different times by different children; they may be the sixth and seventh classifiers acquired, or may be acquired a bit later, depending on each child’s noun repertoire, needs and immediate environment. Except for cuc, all seven of these are disjunctive classifiers. When children first acquire ˙disjunctive classifiers they were generally only able to apply them to one of the relevant semantic categories. This shows that the disjunctive classifiers that have one-to-many mappings are acquired first, before the shape classifiers that have one-to-one mappings (non-disjunctive). This finding is consistent with Wong’s (1998) results. Regarding the remaining classifiers (non-disjunctive ones), the youngest group knew them with only one noun in a fixed amalgam. A closer analysis of how the youngest group used the classifiers sợi, tờ and quyển with each test item showed that they used the expected classifier only with specific nouns. There was no sign of productive use; therefore, it cannot be concluded that children between 2;10 and 3;8 have acquired these classifiers. During the age

The acquisition of Vietnamese numeral classifiers

85

range of the middle group (3;8–4;4), a break away from amalgams occurs as more classifiers emerge. After about age 4, children display a more consistent, appropriate and productive use of shape classifiers. The remaining classifiers, ta´ˆ m, hoˆṭ , vieˆn and cuoˆṇ , are acquired later, during the age range encompassed by the oldest group (4;7–5;7). Vietnamese shape-based classifiers are acquired later than non-shape ones. Among the shape classifiers, the first acquired are caˆy and cuc. This is ˙ an early consistent with previous studies that have found that children have preference for one-dimensionality classifiers, and that the less specific shapes seem to be acquired before those used for more specific shapes.

Errors The errors that children made were also analysed (Table 7.4). It was found that children made six different types of errors. Errors make up only a very small percentage of the total number of nominals. The first, third and sixth error types were structural errors, syntactically impermissible. The other three were syntactically permissible, but semantically inappropriate. Classifier omission errors involved the ungrammatical omission of an obligatory classifier in a classifier structure containing a numeral (Num-CL-N), as shown in (1). (1)

*CHI: hai meo luoˆn. two cat EMP ‘There are two cats!’

(Ha` Mi, 2;3)

The most common overgeneralizations were to use the general classifier (ca´i) to replace the specific classifiers. The children made two types of errors with the general classifiers: (1) dispreferred though not ungrammatical phrases such as those denoting vehicles, one of a pair, or long–rigid items, and (2) phrases that were completely unacceptable in adult Vietnamese. For example, children used the general classifier to refer to animals, fruits/vegetables, root Table 7.4 Errors Type of error

Tokens

Percentage

Classifier omission Overgeneralizations Double classifiers Classifier þ verb Classifier þ other Word order errors

101 108 82 78 20 6

4–17.5 4.2–6.7 general, 2.1–6.2 specific 1–2 0.5–2

86

Jennie Tran

vegetables, and most nouns classified by shape-based classifiers, mass nouns that required a mensural classifier (e.g. water, soup, meat, powder, grass), non-classified nouns that did not need a classifier (e.g. baby), and novel nouns (e.g. Superman). After they progressed beyond this general-as-default classifier stage, Vietnamese-speaking children begin to employ a ‘juggling’ strategy, that is, they used both the general and the specific classifier (the target) concurrently, until the specific classifier was firmly incorporated into the child’s classifier repertoire. The exact time-frame during which the specific classifier completely replaces the general one varies depending on individual differences and the child’s preferences. This is supported by the finding in the crosssectional study that the strategy of employing the general classifier as a default still prevailed in the speech of some children in the middle age group (3;8–4;4) and even in some children in the oldest group (4;7–5;7). In their responses, there was no uniformity; some children responded with the target classifier, and some with the general one. Furthermore, the use of the specific classifier did not prevent children using the general classifier as many children still used the general classifier correctly with nouns. Overgeneralizations with the specific classifier involve using a specific classifier in place of another specific classifier (e.g. using the animal/animacy classifier for ‘banana’). Out of 2012 elicited tokens, only 55 (2.7%) represented such errors. The children between 3;8 and 4;4 made the most overgeneralizations. Semantic overextensions included three main types of errors. First, shape bias errors (e.g. using the fruit/big-round classifier for ‘bell’ and ‘sun’) were made. Second, across-shape errors, that is, using a shape classifier in place of another similar shape classifier, for instance a classifier designating big items was used in place of a classifier designating objects of a similar shape but of a small size (e.g. quả with ‘candy’ and ‘marble’, items that are normally classified with cuc and are rather small to be classified with quả). Another example is using ˙cuc for nouns that require hoˆṭ . The children made ˙ and ‘peanut’. There was also a tendency to such errors with ‘rice grain’ overextend big–round to small–round objects, but not vice versa. The third type of semantic extension was cross-taxonomic, which involved using, for example, quả to refer to root vegetables (carrot, garlic, onion), or using củ to refer to fruits and beans. Presumably, children in such instances are not distinguishing between root vegetables and fruits. Using củ for ‘bean’, which requires hoˆṭ , may be considered a taxonomy-to-shape overextension. There were no overgeneralizations with caˆy, chieˆ´c and cuc. ˙ and (3). Classifierþverb errors are shown in Examples (2) (2)

*CHI: caˆy ve˜ CL draw

(Minh, 1;9)

The acquisition of Vietnamese numeral classifiers (3)

*CHI: ca´i mu´c canh CL ladle soup

87

(Ha` Mi, 2;2)

The CL-V sequence is permissible in Vietnamese (verbal classifiers), but those that the children produced were semantically inappropriate. This error is a circumlocution strategy that the children used when they did not know a particular noun. They describe the expected noun with a verb phrase following the classifier. They mostly used the general classifier ca´i before the verb. Double classifier errors involved the ungrammatical use of two different classifiers consecutively, with the first classifier being redundant. The most frequent double classifier pattern took the form [general classifierþspecific classifierþnoun], when the child added the general classifier to a well-formed, semantically appropriate classifierþnoun construction, as in Example (4). (4)

ca´i

tra´i

banh

CL(general)

CL(fruit/big-round)

ball

(Ha` Mi 2;0, Lieˆm 2;9, Giang 2;9, 2;11)

There were a few errors in the form of [specific classifierþspecific/general classifierþnoun] where the second [classifierþnoun] was also well formed and semantically appropriate, as in Example (5). (5)

*CHI: con CL(animal)

cuc ˙ CL(small-roundish)

đa´ stone

(Ha` Mi, 2;2)

Final discussion and conclusion Vietnamese classifier acquisition has similarities with classifier acquisition in other numeral classifier languages, but it is also different in important ways as well. It confirms the following eight cross-linguistic findings and strategies of classifier acquisition: (1) The first step is the acquisition of a set of five to six basic core classifiers. (2) Children produce many classifiers, but use only a few of them productively. This is consistent with the findings of Erbaugh (1982), Hu (1993), Wong (1998) and Matsumoto (1987). (3) Children make very few errors when constructing classifier phrases. Such errors involve classifier omissions, double classifiers and wrong word order, as reported by Wong (1998), Erbaugh (1986, 2002), Hu (1993), Carpenter (1987) and Tse et al. (2007). (4) Children use the general classifier as a default substitute when they lack knowledge of more semantically specific classifiers, as has been reported in most of the previous studies. (5) Specific classifiers seem to be acquired relatively slowly and late; between 2;6 and 3;0, according to Erbaugh (1986) and Hu (1993). (6) For disjunctive categories, young children use the classifier first for only one of the branches, in line with the findings of

88

Jennie Tran

Carpenter (1987), Hu (1993) and Matsumoto (1985). (7) Among the specific classifiers, the animal classifier is the one that is most often used appropriately, as reported by Erbaugh (2002), Tse et al. (2007) and Wong (1998). (8) Young children rely on shape rather than function in selecting a classifier. Mostly, shape-based classifiers are acquired before function-based classifiers, as found by Poon (1980) and Hu (1993). In addition, Vietnamese-speaking children showed (i) a higher rate of correct specific classifier use than that found in other previously studied classifier languages and (ii) a lower rate of general classifier use in comparison to that found in previous studies.

8

An overview of the acquisition of Malay wh-questions Norhaida Aman

Introduction This chapter aims to provide a brief account of the acquisition of wh-questions of colloquial Singapore Malay, the language to which Malay children in Singapore are exposed at home and in the speech community before they learn the standard, formal language in school. Many studies on wh-questions have been conducted on English and other languages. However, Malay, unlike English, is a language that allows questions with both in situ and moved wh-words. In (1), the wh-word remains in situ (or in its thematic base-generated position), while in (2), the wh-word has been fronted to the beginning of the question. English has only the moved option similar to (2). (1)

Liyana jumpa apa semalam? Liyana find what yesterday ‘Liyana found what yesterday?’

(2)

Apai Liyana jumpa ti semalam? ‘What did Liyana find yesterday?’

(3)

Apai [yang Liyana jumpa semalam] ti ? ‘What (was it) that Liyana found yesterday?’

The third type of wh-question seen in Example (3) is known as a focused question. A focused question consists of a copula-like construction with a headless relative clause (preceded by the relativizer yang ‘that’) in the subject position (Aman, Cole & Hermon, 2009). It should be noted that this question structure is rather formal and is only occasionally found in naturalistic data. The following issues will be addressed in this chapter: (i) children’s knowledge and ability to produce the different options for asking simple questions (wh-in situ, moved questions, and focused questions), (ii) their knowledge of the three question structures in long-distance questions and (iii) the role of island constraints. Data from spontaneous production and quasi-experimental studies like elicited imitation and picture story comprehension will be examined to illustrate the acquisition of wh-questions by Malay-speaking children. 89

90

Norhaida Aman

The acquisition of simple wh-questions Results from production and comprehension tasks reveal that children between 3 and 4 years are able to produce and respond to both simple in situ and moved questions (Aman, 1999). A production task known as Elicited Imitation (EI) (Cohen & Lust, 1993; Lust, Flynn & Foley, 1996; Santelmann et al., 2002) was carried out to assess Malay children’s knowledge of the three question structures: in situ, moved and focused questions. The EI task elicits imitation of particular sentences by young children; children are asked to repeat sentences. The assumption is that if s/he can correctly repeat a sentence, then it can be inferred either that the child has the grammatical competence for generating the sentence structure or that the child can simply repeat what s/he hears. Structures that are not part of the child’s competence will be repeated less accurately than structures that are. The types of errors committed provide interesting information on his/her grammatical knowledge. The percentage of correct responses for this experiment is shown in Table 8.1. The data revealed two main findings: (i) children (between 3;4 and 4;6, mean age 4;2) were equally competent in both in situ and moved questions with the percentage of correct responses at 76.7% and 78.3% respectively; (ii) their poor performance on focused questions suggested a lack of mastery of the focused question construction. These results are similar to results found in spontaneously produced data. Aman (1999) collected naturalistic data from two Malay-speaking children about age 3 over a period of 3 months and found that both mothers and Table 8.1 Percentage of correctly repeated responses Question structure

Wh-word

% Correct responses

In situ

apa ‘what’ sapa ‘who’ kenapa ‘why’ kat mana ‘where’

78.3% 73.3% 88.3% 66.7% 76.7%

Moved

apa ‘what’ sapa ‘who’ kenapa ‘why’ kat mana ‘where’

83.3% 66.7% 90.0% 73.3% 78.3%

Focused

apa ‘what’ sapa ‘who’ kenapa ‘why’ kat mana ‘where’

43.3% 33.3% 16.7% 13.3% 26.7%

The acquisition of Malay wh-questions

91

Table 8.2 Number of wh-questions in terms of wh-word and question structure by two Singapore Malay children and their mothers Question structure

Wh-word

WH in isolation

In situ

Child

Mother

Child

Mother

96 50 3 11 7 8 4 1 2 0

42 5

184 3 31 22 1 8 child: 7 mother: 12a 14 12 21 12 2 16 0 9 19 3 0 8 1 0 1 4 0 1 0 5 0

apa ‘what’ 25 s(i)apa ‘who’ (object) 8 s(i)apa ‘who’ (subject) 1 (kat) mana ‘where’ 48 (ma)cam mana ‘how’ 0 kenapa ‘why’ 11 apasal/pasal apa ‘because (of) what’ 1 apa benda/barang ‘what thing’ 0 berapa ‘how many’ 1 bila ‘when’ 2 a

12 0 0 0 0 1 0

Moved Child

Mother

The percentage of embedded responses was calculated as a proportion of the total number of relevant responses.

children asked moved and in situ questions (Table 8.2).1 However, as shown in Table 8.3, the proportion of in situ versus moved questions differed significantly depending on whether the wh-word was an argument or an adjunct. With argument questions (involving who and what questions), both mothers and children showed a strong tendency to leave the wh-word in situ. On the other hand, with adjunct questions (how and why questions), they demonstrated a tendency to use the moved question structure. Both groups showed similar tendencies towards producing in situ argument questions and moved adjunct questions. These strong preferences were not evident in the EI study (Table 8.1). The absence of an argument–adjunct asymmetry in the 4-year-olds’ responses indicates equal competence in both the in situ and moved structures. What these two different sets of data reveal is that while the input and the children’s output in spontaneous speech show strong preferences to leave the argument wh-word in situ and to move the adjunct wh-word, Malay-speaking children are aware that both wh-question options are acceptable in their language. Consider now the focused questions where the children in the EI experiment performed poorly. As discussed earlier, the data in Table 8.1 reveal a 1

In their first sessions, the two children were 3;0 and 2;6 years. Weekly visits were made over a period of 3 months. A year later, a follow-up session was carried out with the older child who was then 3;8.

92

Norhaida Aman

Table 8.3 Proportion of in situ versus moved wh-questions across wh-words Child

Mother

Wh-word

In situ

Moved

In situ

Moved

apa ‘what’ s(i)apa ‘who’ (object) (kat) mana ‘where’ (ma)cam mana ‘how’ kenapa/apasal/pasal apa ‘why/ because of what’a berapa ‘how many’ apa benda/barang ‘what thing’ bila ‘when’

42 (93.3%) 5 (83.3%) 12 (50%) 0 (0%) 0 (0%)

3 (6.7%) 1 (16.7%) 12 (50%) 2 (100%) 9 (100%)

184 (85.6%) 22 (73.3%) 14 (40%) 12 (42.9%) 3 (10%)

31 (14.4%) 8 (26.7%) 21 (60%) 16 (57.1%) 27 (90%)

a

1 (100%) — 0 (0%)

0 (0%) 5 (100%)

4 (80%) 1 (50%) —

1 (20%) 1 (50%)

An adult control group included in the study did not show this strong preference. They were equally likely to give a matrix/embedded response (Aman, 1999).

lack of mastery in focused question structures. An analysis of errors further revealed that one of the main errors committed was the deletion of the relativiser yang, resulting in moved wh-question structure like that in Example (2) instead of a focused question with the headless relative clause structure shown in Example (3). Naturalistic data revealed that Malay children around 3 years old do in fact produce yang, albeit very rarely.2 Their yang constructions, however, do not have full-fledged relative clause structures. Children used yang mainly in headless relative clauses in which the relative clause does not appear to display the full range of elements, which typically appear in normal IPs. The relative clauses were restricted to examples such as yang þ adjective, as in yang besar punya (that big possessive) ‘the big one’. There was also a single example of a focused (clefted) sentence: Witch dia yang mati (witch 3sg that die) ‘It was the witch who died.’ The mothers, on the other hand, used yang as a relative clause pronoun (relativizer) in both headless and headed relative clauses. These data indicate that the children were aware of the use of yang as a relativizer at a young age. Aman (1999) argued that this poor performance was due to processing difficulties. A focused question is a copula-type construction with a headless relative clause in the subject position and a moved question word. The complexity of the structure appears to be too complex for the children, resulting in poor performance. In other words, at this age, they have not yet acquired the focused question structure in Example (3). The deletion of yang 2

No analysis was made for the subject questions – whether the subject remains in situ in [Spec,IP] position, or if vacuous movement to [Spec,CP] has taken place. The subject questions were coded as being neither in situ nor moved.

The acquisition of Malay wh-questions

93

in the EI task suggests the children resorted to ‘simplifying’ the structure – from a focused (headless relative structure) to a fronted wh-question structure that they are more familiar with. Long-distance questions In the de Villiers, Roeper and Vainikka (1990) study, 25 English-speaking children from 3;7 to 6;11 years were read stories which involved temporal or manner information, and were shown accompanying contextual pictures. At the end of each story, a question was asked. All the stories were ambiguous as in each story there were two possible answers to the question asked. For example: Story This boy loved to climb trees in the forest. One afternoon he slipped and fell to the ground. He picked himself up and went home. That night when he had a bath, he found a big bruise on his arm. He said to his dad, ‘I must have hurt myself when I fell this afternoon!’

(4)

Wheni did the boy say _(at night)_ he hurt himself _(this afternoon)_ ? 50% 44%

If the child associated the trace with the main verb and answered ‘at night’, this response was classified as the ‘short-distance’ movement (or the matrix response) option. However, if the child believed the wh-word originated from within the embedded clause and answered ‘that afternoon’, then the response was classified as the ‘long-distance’ interpretation (or the embedded response). De Villiers et al. found that 44% of the time, children gave the long-distance interpretation, i.e. the embedded response. Weissenborn, Roeper and de Villiers (1991) found that like their English counterparts, German- and French-speaking children aged between 3;3 and 6;11 also showed knowledge of long-distance questions. The same picture story framework was used with 4–5-year-old Malay-speaking children. At the end of each story, each child was asked either the in situ or the moved question (Aman, 1999). If the child was asked a moved question, and s/he believed that the wh-word was part of the subordinate clause, and then gave the embedded reading, then this constituted a long-distance interpretation. (5a)

In situ question Kakak bilang mak [Ali koyakkan baju dia __(kat mana)_] _(kat mana)_? sister tell mom [Ali tore dress her __(where)__] _(where)__? ‘Where did (Ali’s) sister tell mom Ali tore her dress?’

94

Norhaida Aman

Table 8.4 Proportion of in situ versus moved wh-questions across wh-words Wh-word kat mana ‘where’

macam mana ‘how’

Question structure

Matrix

Embedded

Matrix

Embedded

In situ Moved

6.2 0

93.8 100

14.3 40

85.7 60

(5b)

Moved question Kat manai kakak bilang mak [Ali koyakkan baju dia _ti_] _ ti _? ‘Where did (Ali’s) sister tell mom Ali tore her dress?’

There are two possible responses: (i) Matrix response (modifying the main verb bilang ‘tell’): di dapur ‘in the kitchen’ (ii) Embedded response (modifying subordinate verb koyakkan ‘tear’): di dalam bilik ‘in the room’ Table 8.4 shows the percentage of embedded responses the children gave across the two question structures.3 The results not only show that children can handle both in situ and moved long-distance questions, which involve extractions out of complement subordinate clauses; they actually demonstrated a strong preference for the embedded response. Just like their English, German and French counterparts, Malay children at this age can also process long-distance structures. However, what is different is the strong preference for the embedded response (i.e. the long-distance interpretation) shown by the Malay children involved in the study.4 One possible reason for this finding is perhaps the location (where-question) and manner (how-question) of the reported action or event is more salient than the location and manner of how the event was reported or told. Spontaneous production data show that children are exposed to some long-distance questions. While mothers generally uttered short constructions, they also produced some sentences with embedded clauses, and some longdistance questions. A careful analysis of the long-distance questions revealed 3 4

The count for the three question words was collapsed into one because all three forms have the same meaning. The yang constructions that children produce in naturalistic data included sentences like, Yang ni, which was very productive. These were analysed as having the meaning ‘this one’ rather than a headless relative structure ‘(the one) that is this’.

The acquisition of Malay wh-questions

95

that argument questions were all in situ, while only moved adjunct questions were asked: (6a)

(6b)

In situ Argument Wh Kira tolong mama buat apa? Kira help mom do ‘You help me do what?’ Moved Adjunct Wh Apasal dia suruh bunuh dia? why she tell kill ‘Why did she tell (him) to kill her?’

what

her

Neither of the two children (2;6–3;2 years) uttered complex structures. However, a follow-up session with one of the children when she was 3;8 revealed a number of complex structures in the child’s syntax – 13 complex sentences and two long-distance questions, both of which are in situ argument questions as illustrated in Example (7). This suggests it is at about this age that children are able to produce more complex structures. (7)

Baba, baba, bawak, baba bawak panggil apa? dad dad bring dad bring ‘(The thing that) dad (rides) is called what?’

call

what

Island constraints Otsu (1981), de Villiers et al. (1990), de Villiers and Roeper (1995), Goodluck, Sedivy and Foley (1989) and Goodluck, Foley and Sedivy (1992) have shown that children’s early grammars are sensitive to innate universal constraints on movement, since children (at around age 4) observe wh-islands and prohibit adjunct extraction from both relative clauses and temporal adjunct clauses in English. For example, de Villiers and Roeper (1995) tested 4–5year-old English children on extraction out of relative clauses, as seen in the non-existent embedded response in Example (8). (8)

How did the boy who sneezed ____ drink the milk ____? 0% 94%

Goodluck et al. (1992) tested temporal adjunct islands and found some children committing island violations: (9)

What did the fox eat ____ before whistling ____ ? 3 years 67% 11% 4 years 78% 8.5% Adults 98% 0%

96

Norhaida Aman

Research on children learning other languages such as German, French and Spanish have shown that children obey these constraints on movement (Perez-Leroux, 1990; Weissenborn et al., 1991). Through adults’ grammaticality judgement, Cole and Hermon (1998) found that for an ‘optional wh-movement language’ like Malay, wh-in situ can freely go inside islands like relative clauses, but moved questions like (10b) obey island constraints. (10a)

Grammatical in situ question inside a relative clause [CP [IP Ali [VP pandu [NP kereta [CP yang [IP lelaki itu [VP baiki cam mana]]]]]]]? Ali drive car that man that repair how ‘Ali drove the car that the man repaired how?

(10b)

*‘How i did Ali drive the car that the man repaired t i?’

If Universal Grammar (UG) principles which apply to movement are in fact universal and inviolable, it is predicted that the Malay children should also respect these constraints on movement, known as island constraints (Chomsky, 1986; Ross, 1967) even though Malay differs markedly from these languages in that Malay allows the in situ option for wh-questions, which the other languages lack. The comprehension study testing the children’s knowledge of island constraints showed that, unlike the adult controls and English children in other studies, Malay children (mean age 5;2) gave many responses that appeared to violate island constraints, which are universal principles in UG (Aman, 1999, 2007). Table 8.4 which summarizes the percentage of embedded responses (modifying the subordinate verb) for both in situ and moved questions shows that the children allowed extractions out of islands by giving the embedded response from 27% to 56% of the time, depending on the embedded clause type. Interestingly, the data also show that responses for moved questions pattern rather closely with responses for in situ questions, with the exception of postposed adjuncts. Hermon and Aman (1999) and Aman (1999) argue that the island violations may be the result of a parsing strategy, whereby the children reconstructed the moved wh-questions as in situ questions. According to Hermon and Aman, Malay children rely on the in situ strategy whenever the structure is complex and reconstruct the moved question as an in-situ question. Instead of finding a gap, the children actually move back a wh-phrase to reconstitute it as an in situ question. They have to wait till later

The acquisition of Malay wh-questions

97

in the sentence after all obligatory elements have been attached to the verb to attach the adjunct wh-word. Consider the object relative in Example (11): (11)

Object RC

The main verb is considered first. Even though a gap could be posited here, the adjunct wh-phrase kat mana has to wait until the whole object relative has been parsed in order to get attached after the embedded verb. In this position, the reading modifying the embedded verb is preferred. Since Malay has a way to interpret in situ wh-phrases even inside islands, the child now gets an in situ reading. The issue of islands is thus avoided altogether because wh-in situ questions inside islands are allowed in Malay. In wh-in situ structures, nothing has moved out of the island domains, and hence no constraints are violated. This explanation accounts for the results seen in Table 8.5 where the responses given to moved questions pattern closely with their in situ variants. Aman (1999) argued that this parsing strategy is only available in languages, like Malay, that allow both the in situ and moved wh-questions, in which in situ questions do not violate island constraints. Because the adult controls respected the island constraints regardless of their parsing preferences for the in situ wh-questions, this is taken as evidence that the adults did not reconstruct the wh-phrase by treating fully moved questions as wh-in situ. So when and how will the children acquire the adult grammar, by positing gaps instead of reconstructing the wh-phrase, and respecting the island constraints? One possibility is that when the children’s parser matures, they will treat moved questions in complex structures as involving long-distance chains between the filler and the gap, which must then respect island constraints. Hence, as the children become more proficient in their language, their parser becomes more mature and they have more computational resources available to process syntactically complex structures (wh-questions with islands) instead of reconstructing the wh-phrases. As children receive more exposure Table 8.5 Means of embedded responses given by the children Embedded clause Question structure

Subject relative clause

Object relative clause

Preposed adjunct

Postposed adjunct

In situ Moved

34.3 27

64 56

58 46

82 29

98

Norhaida Aman

to the formal variety of Malay in the educational institutions, they will acquire adult grammar. In school, children are exposed to the formal structures and taught the focused question structure. The authors posit that as children hear more focused structures in both questions and declaratives like Buku yang Mary beli, ‘It was a book that Mary bought’, they will realize that they need to ‘switch’ to the filler-gap strategy, and can no longer rely on the reconstruction strategy. Conclusion The discussion has shown that there was a strong preference in both the input and children’s output to leave the argument wh-word in situ and to move the adjunct question word in spontaneously produced speech. However, young Malay-speaking children know their language allows two options in wh-questions – in situ and moved, and in carefully constructed production tests, children around age 4;2 were able to repeat both question structures equally well, indicating equal competence. The focused question structures that involve a headless relative clause in the subject position of a nominal construction and are more formal and rarely present in naturalistic data, proved to be problematic for the children. Previous research on long-distance question structures involving English children and children learning languages like German, French and Spanish have shown that children obey these constraints on movement known as island constraints. Malay is interesting in that it differs markedly from these languages because it allows the in situ option for wh-questions, which the other languages lack. In Malay, while wh-in situ questions can be present inside islands, wh-movement out of island domains is not allowed. Malay children (at around 6 years old) were found to violate island constraints, which are universal principles in Universal Grammar. However, it was argued that these apparent violations may be the result of a parsing strategy, whereby the children reconstructed the moved wh-questions as in situ questions. Hence, the study on Malay affords researchers with an alternative perspective in looking at the different types of wh-questions, and the issue of island constraints in in situ languages.

9

Marking plurals: the acquisition of nominal number inflection in Marathi Shalmalee Pitale and Vaijayanthi M. Sarma

The difference between one and many is both conceptually and linguistically important in child development. This understanding is reflected most readily in the inflectional morphology (number marking on the noun, subject–verb and other agreement) of the world’s languages and acquisition of number entails knowing how number is morphologically marked or otherwise indicated. Studies in a wide variety of the world’s languages have shown that inflections appear early especially in richly inflected languages and that children induce rules to build the inflected forms (Dhongde & Wali, 2009; Lakshmanan, 2006; Sarma, 1998; Stephany & Voeikova, 2009; Varma, 1979). Morphological acquisition is governed by both universal principles of acquisition (developmental sequence, regularity of affixation, conditioning, etc.) and by specific features of the actual target languages (number of nominal classes or genders, phonotactics, etc.). Marathi,1 an Indo-Aryan language, makes a two-way number distinction, singular and plural. The number feature is synthetically marked on the noun. Two linguistic factors determine the plural forms: (a) grammatical gender, whether a noun is masculine, feminine or neuter, and (b) the final segment of the noun. We begin with a detailed account of nominal number morphology in Marathi based on Damle’s comprehensive grammar (1970) and the linguistic analyses of Kelkar (1958) and Pandharipande (1997). The chapter describes our studies on the acquisition of Marathi plural (with an emphasis on count, inanimate nouns) in the context of acquisition of nominal number inflection in general. Since number inflection in Marathi is influenced by gender, acquisition of this ‘inherent’ nominal property is intertwined with acquisition of number inflection. This interplay between grammatical gender and number varies from language to language and may show no influence of gender (Portuguese, Spanish, French), or complete determination by either gender (Hindi, Swahili) or phonology (Gujarati, Italian), or determination by both phonology and gender in varying degrees of complexity (Russian, German, Marathi and Konkani). 1

Standard Marathi is the official language of the state of Maharashtra and is spoken by over 70 million people.

99

100

Shalmalee Pitale and Vaijayanthi M. Sarma

In modern Indo-Aryan languages, the three gender (masculine, feminine and neuter) system of Sanskrit (preserved in Pali and Prakrit) has been weakened in many ways (Masica, 1993).2 For example, Oriya, Assamese and Bengali have lost the gender distinction, Hindi, Urdu, Punjabi and Rajasthani have moved to a two-gender system, merging neuters with masculine and Nepali, Awadhi and Bhojpuri have attenuated gender and mark only the feminine. Only in the three western languages of Gujarati, Marathi and Konkani is the three-way distinction preserved. Consequently, the interplay between number and nominal gender in these languages and the rules and how they are learnt is of interest both in the context of language acquisition in general and of acquisition of Indian languages in particular. In this chapter, we have adopted and adapted J. Berko’s classic study on the acquisition of the inflectional morphology of English (1958) and B. Spreng’s work on the acquisition of plural morphology in German (2004) to study plural inflections in children acquiring Marathi.3

Plural nouns in Marathi Nouns in Marathi are typically inflected for number and case. The inherent lexical feature of grammatical gender plays an important role in the selection of nominal inflections. Marathi nouns are typically placed in different classes given the three-way gender distinction as well as the phonological shape of the bare or unmarked noun (Pandharipande, 1997). Gender is reflected primarily through agreement (article, adjectival and verbal). Grammatical gender is congruent with natural sex distinctions (male versus female) for animate nouns but is not otherwise uniquely identifiable through phonological or semantic features. In this, Marathi differs from German and acquiring nominal morphology is mediated by the acquisition of nominal gender and phonology. This complicates the task for the language learner, as we shall see.

Classifying nouns in Marathi Singular/bare forms of countable nouns in Marathi may end in a vowel or a consonant. As can be seen in Table 9.1, the final segment does not uniquely determine the gender feature. Thus, a consonant or [u¯] final stem is arbitrarily 2 3

Sanskrit had three numbers (singular, dual and plural) with the dual disappearing by Middle Indo-Aryan period. This is part of a larger project on the acquisition of inflectional morphology in Marathi and Konkani.

The acquisition of nominal number inflection in Marathi

101

Table 9.1 Noun classes in Marathi Plural Singular

Masculine

ə

Ø mitrə ‘friend’ Ø hat ‘hand’

consonant

ī

Ø məɳī ‘bead’ Ø laɖɖu¯ ‘a sweet’

u¯

a

a! e vaɖa!vaɖe ‘mansion’

Feminine

Neuter

C !a /ī a (i) vaʈ ! vaʈa ‘path’ vihir !vihirī ‘well’ ī !a nədī ! nədya ‘river’ Ø or a (i) vəstu¯ ‘thing’ (ii) pisu¯ ! piswa ‘a mosquito’ ÐØ Ð aɭa ! aɭa ‘school’

C !e/ə ghər ! ghəre /ə ‘house’

e/ ə u a

u¯ ! e/ə limbu¯ ! limbe/ə ‘lemon’

e/ə ! ī təɭe/ə ! təɭī ‘pond’ Ø Ð ətru ‘enemy’

See the next section for further discussion.

masculine, feminine or neuter (laɖɖu¯ m, vəstu¯ f, and limbu¯ n). Though neuter nouns show greater shape restriction than do masculine or feminine nouns (as can be seen by the grey cells), the class overlap is significant and phonological shape cannot directly cue nominal gender for a language learner. Affixal change in the plural is visible in only one of the six masculine classes, in two of the four feminine classes and in all three neuter classes. The classes that do not overtly inflect for number (though the tokens are count nouns) are shown with a phonologically null morpheme [Ø] (for example in English sheep!sheep or shrimp!shrimp). The distribution of plural endings is restricted to one of {ø, a, ī, e, ə} and each class selects the plural suffix. This match is not a unique one; thus -ī may mark the plurals of neuter or feminine nouns. We would also like to point out that the rules in Table 9.1 such as C ! e/ə, e/ə ! ī, u¯ ! e/ə, indicate free variation where either variant can be attached to the noun stems in that class. The -e ending was mainly used in formal speech and writing and [ə] elsewhere. This distinction has disappeared in contemporary Marathi. The cells in grey indicate empty classes, for example, there are no feminine or neuter nouns that end in -ə or -u.

102

Shalmalee Pitale and Vaijayanthi M. Sarma

Phonological shape and plural The phonological ending of the noun is an important cue in plural formation and gender assignment, but not unambiguously so. For example, only neuter nouns end in e/ə and show a one-to-one correspondence with the plural form (təɭe/ə – təɭī ‘pond’) and only masculine nouns end in [-u]. However, feminine noun classes overlap with masculine classes and cannot be uniquely identified. Further, feminine nouns ending in [u¯] may change in one of two ways: (a) take the plural affix [-a] or (b) remain unchanged [-ø]. The correspondence between gender and phonological shape is equally unreliable and the plural affix is determined by the inherent gender assigned to a noun. (1)

Masculine Feminine Neuter

ghoʈ moʈ poʈ

! ! !

ghoʈ- ø moʈ-ī poʈ-ə

‘gulp’ ‘a machine’ ‘stomach’

Gender and plural While the assignment of gender is a determinant of the plural as we saw in Example (1), gender is only partially determined by the phonology. Often even gender fails to determine the plural affix, as seen in Examples (2) and (3) below. Stems in classes F1 and F3 are further grouped based on the actual plural affix chosen as evidenced by the rule C !a /ī in Table 9.1 where a consonant-final feminine noun may take either [a] or [ī] as a plural suffix. Nouns with same phonological shape and gender are nonetheless marked for plural differently and gender fails to determine the plural suffix. (2)

Feminine

gath ! gath-ī

‘knot’

but

lath ! lath-a

‘kick’

(3)

Feminine

baju¯ ! baju¯-ø

‘thing’

but

pisu¯! pisw-a

‘mosquito’

The determinants of the plural marker in Marathi include a noun’s (a) phonological ending, (b) grammatical gender, (c) natural gender and (d) the inflection class.4 In the next section, we discuss how this system is acquired and used by young children and adults. Acquisition studies Studies investigating the acquisition of morphology give us insights into the underlying rules of mental grammar and how they operate in language use. A classic example is Jean Berko’s (1958) study which demonstrates that 4

An inflection class is the set of words choosing the same affix for a morphological property in the language.

The acquisition of nominal number inflection in Marathi

103

children and adults apply tacitly known rules of inflectional morphology to new-coined words. Standard linguistic descriptions of English inflectional morphology show that the selection of a particular form of the affix {s, z, ιz} in the plural, possessive or present tense or {t, d, ιd} in the past tense, is phonologically conditioned by the noun/verb. Berko’s famous wug test was designed to determine whether young learners induce this posited system of rules with sufficient exposure to the language and whether it governs their use of inflectional morphology in their production. Preschool and first-grade children between 5 and 7 years of age were presented with a number of nonsense words associated with simple pictures of unfamiliar objects and asked to supply the appropriate inflected forms. The results showed that the children were able to form plurals/past tense using the appropriate allomorphs. The epenthesized affixes were produced later showing possible effects of maturation. The use of coined words ruled out the possibility that the children knew the plurals independently. Berko also demonstrated that regular patterns were favoured over irregular ones. The internalized linguistic system enables the children to create new forms and be productive users. This study constituted the first rigorous experimental proof of the generalized system of rules that young children extract from the language data surrounding them. A second study that informs our work on Marathi is Bettina Spreng’s (2004) analysis of German plurals. German nouns belong to one of three genders (masculine, feminine and neuter) that is reflected in the choice of the definite article but is not overtly marked on the noun. There are five plural suffixes -e, -(e)n, -er, -s, -ø, and the selection of each is determined primarily by the phonological shape of the noun (i.e. number of syllables, shape of the final syllable and the final segment of the stem), and secondarily, by the noun’s gender. Spreng outlines the conditions for each plural type. For example, the ‘e’ type comprises monosyllabic, consonant final nouns of all genders, ø(þUmlaut) type comprises bisyllabic, consonant final nouns etc. Sixty-five monolingual German-speaking children, 3 to 7 years old, were tested on tokens of real as well as coined words. An analysis of the error patterns indicated that, for children, gender disambiguates plural types somewhat. Thus, phonologically similar das Haus (neuter) ‘house’ and die Maus (masculine) ‘mouse’ take two different affixes, ‘-er’ and ‘-e’ respectively. Gender as an abstract category aids acquisition of plural morphology where a single plural class has nouns of all three genders. For example der Fisch (m) ‘fish’, die Maus (f) ‘mouse’ and das Schiff (n) ‘ship’ of the –‘e’ plural type and der Mast (m) ‘mast’, die Uhr (f) ‘clock’ and das Ohr (n) ‘ear’ of the -en plural type. The children never used a plural type for a noun that did not have at least one conditioning factor of

104

Shalmalee Pitale and Vaijayanthi M. Sarma

that class. The main determining factor was phonological shape, but recourse to gender was made whenever possible. The high number of null affix responses, especially for coined words, suggests an avoidance strategy rather than a default rule. Phonology and gender play a crucial disambiguating role in German plural morphology, but not so in Marathi. This difference has consequences for the performance patterns that we discuss in the next section.

Marathi experiments We ran two tests, one on real Marathi words and one on non-words, to see how plural formation rules are applied by Marathi-speaking children and adults. The tests involved two tasks: (a) plural formation (singular to plural) and (b) affix-stripping (plural to singular). While gender is not directly marked on nouns, it is marked morphologically through agreement. Therefore, any agreement cues should help a Marathi speaker identify the inherent gender of a noun and choose the corresponding plural form or, conversely, identify the plural affix and place the bare noun in the relevant gender class. In the first study, we asked subjects to (a) supply the plural of the nouns read out to them and (b) strip the affix off plurals to recover the bare noun. Ample gender cues were provided. All the tokens were commonly used, highfrequency nouns. Ten adult subjects (five men and five women, educated native Marathi speakers from Mumbai) and ten children (five boys and five girls, 3 to 7 years old from Shahapur, Maharashtra) participated in the study. The children’s first language was Marathi and their exposure to Hindi and English was minimal. Multiple tokens of real words for the different noun types (see Table 9.2) below were used with adults and simple, coloured pictures of familiar objects (one token per class) with children. The two tasks were run consecutively but with different tokens. Each picture had a written description, where the keyword, indicated by the picture, agrees with co-occurring words (adjective or verb) in the description. For example, along with the picture of gaɽi ‘car’, the written description given in (4) was read out. Children were expected to complete the sentence with an appropriate plural form: (4a)

hī pəha ek gaɽi alī. (Introduce token, agreement with article and verb) dem-f-sg see-imp one car(f) come-f-sg-past Here comes one car

(4b)

hī gaɽi hirvī ahe. (Adjective agreement with green) dem-f-sg car(f) green-f-sg be-sg-pres This car is green

The acquisition of nominal number inflection in Marathi (4c)

hi ajun ek gaɽi alī. (Plural context) dem-f-sg another one car(f) come-f-sg-past Here comes another car

(4d)

ata ahet don ______ now be-pl-pres two now there are two . . .

105

(Test sentence: expect gaɽya)

In (5) we provide an example of affix-stripping: (5a)

hya tin gaɽya alya (Introduce token, agreement with article and verb) dem-pl three cars(f) come-f-pl-past Here come three cars

(5b)

tyatil don gaɽya gelya (Number agreement) dem-abl two cars(f) go-f-pl-past Two cars went off

(5c)

tya gaɽya hirvya hotya (Gender agreement) dem-f-pl cars(f) green-f-pl be-f-pl-past Those cars were green

(5d)

ata urlī ek __________ now left-f-sg-past one Now here left one . . .

(Test sentence: expect gaɽi)

The results for the two test groups in this first study are given in Table 9.2. The overall performance of the adults shows, not surprisingly, that the system of plural marking is in place. Across all classes, the affix-stripping performance is a uniform 100%. In the affix-assigning task, the performance is perfect except for classes F4 and N2. The children’s performance, however, varied across different noun classes and, interestingly, across the two tasks. For the non-changing classes (M1–4, F1 and F2), children correctly produced/analysed all the tokens. Performance was also good for the one changing masculine class (M5). Children (and some adults) had difficulty with consonant final feminine nouns (F4 and F5), as they inflect in one of two ways for plurality. The performance on the affixstripping task for these classes is superior to affix-assigning and suggests that children recognize the bare noun and isolate the plural affix more reliably.5 The uncertainty in the other direction must follow from the arbitrary affix choice that needs to be made. The performance for the three subclasses of neuter nouns is quite poor. Here too, affix-stripping appears easier. The shape of the N1 nouns overlaps with that of M2, F4 and F5 and the shape of the N2 nouns overlaps with that 5

In the next task, we see that the uncertainty with the noun’s form degrades performance on affix-stripping.

106

Shalmalee Pitale and Vaijayanthi M. Sarma

Table 9.2 Results for testing the plural of real words Results (%) Children

Adults

Gender

Sg

Pl

Token

Gloss

Sg!Pl

Pl!Sg

Sg!Pl

Pl!Sg

M1 M2 M3 M4 M5 F1 F2 F3 F4 F5 N1 N2 N3

ə C ī u¯ a a u¯ ī C C C u¯ ə/e

Ø

mitrə hat məɳī laɖu¯ gÐ hoɽa aɭa vəstu¯ paʈī maɭ vihir jhaɽ limbu¯ ɖokə/e

friend hand bead a sweet horse school thing slate garland well a tree lemon head

100 100 100 100 100 100 100 70 30 20 10 10 50

100 100 100 100 100 100 100 100 70 80 80 80 90

100 100 100 100 100 100 100 100 80 100 100 80 100

100 100 100 100 100 100 100 100 100 100 100 100 100

e Ø Ø e/a a a ī ə/e ə/e ī

Table 9.3 Tokens for non-words Singular

Masculine

ə consonant

M1 pu¯mbə M2 tsag

ī u¯ a e/ə

M3 pəɳī M4 maʈu¯ M5 tuɖa

Feminine

Neuter

F4 raʈ F5 ɖhak F3 sakvī F2 bhaku¯ F1 gaʈa

N1 khaɖ

N2 jambu¯ N3 bhəɭə

of F2. The degradation in performance could be linked to this class overlap and the difficulty in assigning nominal class unambiguously. The nonoverlapping N3 class shows significantly improved performance but is still somewhat problematic. In the next study, we tested the subjects’ knowledge of plural formation using non-words. We used pictures from Berko’s study of unfamiliar objects which were associated with non-words such as those listed in Table 9.3 for different noun classes (except -u ending masculine nouns which are restricted to animate nouns borrowed from Sanskrit).6 We used the same tasks here as in 6

From CHILDES (http://childes.psy.cmu.edu/topics/).

The acquisition of nominal number inflection in Marathi

107

the test for real words with similar accompanying descriptions with agreement cues as shown in (6) and (7) below: (6a)

hī pəha ek sakvī alī. (Introduce the token, agreement with article and verb) dem-f-sg see-imp one sakvī(f) come-f-sg-past Here comes one sakvī

(6b)

hī sakvī hirvī ahe. (Adjective agreement) dem-f-sg one sakvī(f) green-f-sg be-sg-pres This sakvī is green

(6c)

hi ajun ek sakvī alī. (Plural context) dem-f-sg another one sakvī come-sg-past Here comes another sakvī

(6d)

ata ahet don ________ now be-pl-pres two Now there are two. . .

(7a)

hya tin sakya alya (Introduce the token, agreement with article and verb) dem-pl three sakvī-f-pl come-f-pl-past Here come three sakvya

(7b)

tyatil don sakvya gelya (Number agreement) dem-abl two sakvī-f-pl go-f-pl-past Two sakvya went off

(7c)

tya sakvya hirvya hotya (Gender agreement) dem-f-pl sakvī-f-pl green-f-pl be-f-pl-past Those sakvya were green

(7d)

ata urlī ek ________ now left-f-sg-past one Now here left one. . .

(Test sentence: expect sakvya)

(Test sentence: expect sakvī)

From the results in Table 9.4, we see that both children and adults appear to have no difficulty with the non-changing classes (M1–4, F1 and F2). However, consonant-final feminine classes (F4 and F5) and the three neuter classes remain problematic. Spreng’s conclusion that some plural types are easier than others and those that are difficult remain so for both adults and children seems to hold true for Marathi. Children’s performance for the tokens simulating the neuter classes is particularly poor and these classes appear difficult for both groups. When in doubt, subjects (children more than adults) typically left the nouns unmarked.7 Other errors were restricted

7

This is also seen in German, what Spreng deems an avoidance strategy when confronted with uncertainty.

108

Shalmalee Pitale and Vaijayanthi M. Sarma

Table 9.4 Results for testing of plurals for non-words Results (%) Children

Adults

Gender

Sg

Pl

Token

Sg ! Pl

Pl ! Sg

Sg ! Pl

Pl ! Sg

M1 M2 M3 M4 M5 F1 F2 F3 F4 F5 N1 N2 N3

ə C ī u¯ a a u¯ ī C C C u¯ ə/e

no change

pu¯mbə tsag pəɳī maʈu¯ tuɖa gaʈa bhaku¯ sakvī raʈ ɖhak khaʈ jambu¯ bhəɭə

100 100 100 100 80 100 100 60 10 0 20 0 0

100 100 100 100 20 100 100 40 10 0 0 0 0

100 100 100 100 80 100 100 100 30 20 80 20 30

100 100 100 100 100 100 100 80 40 20 40 10 20

e no change a a ī ə/e ə/e ī

to the potential suffixes available for a particular gender class. Despite gender cues, children and adults sometimes chose an alternate affix from a noun class of the same gender or an affix that may apply to multiple genders. For example, the errors for ‘u¯’ ending tokens are restricted to -a(F), -ə/e(N) or ø(M). The results of the children’s test group did not show any age specific differences. Unlike the real words’ test, the two tasks did not show a great difference in performance. Where there is a difference, affix-assigning is better than affixstripping. This may appear counter-intuitive at first since once the affix is given it should be a simple matter to recover the bare noun (wʌgz!wʌgþz). Recall however that in Marathi there are several plural affixes which are not uniquely matched to a class by phonological shape or gender. Unlike the previous task with common nouns and where affix-stripping performance was better than affix-assigning, here the bare noun is not known and the affix could legitimately be seen as a part of the root. Thus, performance degrades in both tasks. The results of both the real word and non-word tests showed similar, poor performance for consonant-ending feminine nouns and neuter nouns. The performance for affix-stripping is better than affix-assigning for real words, but for coined words, affix-stripping is as problematic as affix-assigning. Knowing or identifying the plural affix does not seem to make recovery of a novel base noun and assigning it to a nominal class any easier.

The acquisition of nominal number inflection in Marathi

109

Conclusion Clearly, a linguistic system is in place fairly early, which enables the children to produce novel forms. Our results parallel those on the acquisition of English and German and show that morphological rules are induced by children and these rules are applied to unfamiliar words to a significant extent. The system, however, has several built-in uncertainties because the match between the phonological shape and gender of nouns is not unique and also because the plural affix selection is not uniquely determined by any single property. Marathi makes an interesting study for this reason and because there are very few languages that behave in this way. As was mentioned in the beginning, Marathi along with Gujarati and Konkani forms an interesting group of languages that still maintain a threeway, arbitrary gender contrast. Since neuter nouns are somewhat problematic for both children and adults, even for known words, it is important to consider if there is pressure on this class to merge with the other two classes. This merger is seen in many of the other languages of the Indo-Aryan family. Our unpublished work also shows that overall, neuter nouns are less frequent in the language and of the three neuter classes, N2 and N3 are very marginal. This may have an effect on overall performance of participants in these tasks. Outstanding issues that will further our understanding of Marathi are (a) the frequency of the noun classes and its effect on acquisition and use of the affixes, (b) the extent to which new words are classified as neuter, testing the robustness of this class, and (c) the systems as they occur in Konkani or Gujarati and their use by those native speakers, both adults and children. Our ongoing research seeks to address these questions for a more complete account of the acquisition of inflection in these languages.

10

Issues in the acquisition of Tamil verb morphology Vaijayanthi M. Sarma

Introduction Tamil is a South Dravidian language with an ancient literary tradition and is spoken in the southern Indian state of Tamil Nadu, the Union Territory of Puducherry, Singapore and parts of Sri Lanka by over 65 million native speakers. It is a diglossic language with many regional and caste-based vernacular variants and a ‘high’ or codified variety that is used in writing, education, television and other media.1 Tamil is a morphologically rich language and is of particular interest given the agglutinating nature of the language. The verb is the repository of key grammatical functions and can be, structurally and functionally, highly complex. This, and the fact that verbs are also the key determinants of argument structure, makes the Tamil verbal system particularly interesting in the context of language acquisition. From the early work of Berko (1958) and Brown (1973) to more recent work such as Slobin (1985) and Bittner, Dressler and Kilani-Schoch (2003), verbal inflectional morphology is critical to furthering our understanding of language acquisition. This research informs us of the range of cross-linguistic features that are encoded in inflectional systems and how the components of morphology, syntax, semantics and phonology interface. Studying the acquisition of these features allows linguists and psycholinguists to deduce the mechanisms that drive language learning, and to distinguish between universal and language-specific patterns in acquisition. Given the central role of verbal inflection, specifically in Tamil and more generally in language acquisition, this chapter analyses the verbal morphological system of Tamil, the functions that are expressed by the verbal complex and their appearance and use in first language acquisition. Grammatical knowledge of the Tamil verbal system includes knowing the forms of the verbs (finite, non-finite and infinitival), agreement inflection 1

Within child language acquisition, longitudinal and cross-sectional studies focus entirely on the spoken variants and there are no studies of the formal variety or how it influences the vernacular styles.

110

Issues in the acquisition of Tamil verb morphology

111

(for three persons, two numbers and three genders in the 3rd person), argument structure including case relations, and also the associated auxiliary elements that express tense, aspect, attitude, reflexivity, voice, mood, negation, embedding, etc. We focus our attention on the issue of agreement, that is, between the subject and the verb which is tied to the identification of subject, object and other grammatical roles, case assignment (nominative, accusative or other), the use of different auxiliaries and how they are stacked, and the various finite and non-finite forms of the verb. The acquisition data discussed here is from the recorded speech of a single Tamil-speaking child, Vanitha (V) (CHILDES database MacWhinney, 2000; Narasimhan, 1981).2 The recordings begin with the child at 9 months of age and end 2 years later, at 33 months. There are 25 sessions in total, each about an hour long. The recordings are of natural speech with the child interacting primarily with her parents.3

Agreement and case assignment Tamil is an SOV language that has rich case and agreement morphology (Asher, 1982; Lehmann, 1989). The case forms in Tamil are marked by either a case affix or a bound postposition; certain postpositions take case-marked noun phrases (NPs) as complements.4 There are nine cases in Tamil and they are listed in (1) along with the corresponding morphemes and their variants. (1)

Case Nominative Accusative Dative Locative Instrumental Sociative Ablative Genitive Benefactive Vocative

Case affix (unmarked) -ai -ukku, -kku, -ku -il -aal -ooDu -irundu -in, -adu, ø

Postposition

-iDam -uDan (nom. or loc. NP) -uDaiya -aaga (dative NP)

-ee (consonant final stems)

The case morphemes are affixed to the noun stem, after the number marker (unmarked in the singular and marked by –gaL in the plural), the oblique suffix and certain euphonic increments.

2 3 4

We limit the discussion to only this data set since it is readily available on CHILDES. This corpus has been analysed in other papers (Vaidyanathan, 1988, 1991; Sarma, 1995). There are also a number of free postpositions in Tamil but they are not relevant here.

112

Vaijayanthi M. Sarma

The verb agrees with the nominative subject for person and number in all persons and for gender only in the 3 person.5 The agreement suffixes are given in (2). (2)

Person 1st person 2nd person 3rd person

Singular -een, –e˜e˜ -aay, -ee -aan, a˜a˜ (m) -aaL, -aa (f) -adu, -um (n) -aar (honorific)

Plural -oom, o˜o˜ (inclusive and exclusive) -iirgaL,-iiŋga -aargaL, -aaŋga, -aa (m., f., honorific) -ana (n)

In the following, we discuss different sentence types in Tamil and the kinds of grammatical properties that follow from the verb morphology as well as their occurrence in the V-corpus. Sentences have functional (tense, agreement, complementizer, etc.) as well as lexical categories operating within them. We assume that the (binary) features of these functional nodes (as seen in many analyses in Generative Grammar) characterize different sentences and determine the verb’s co-occurring arguments. The implicit understanding is that these grammatical mechanisms drive both the adult and child grammars. Beyond this we do not discuss the syntactic processes or the syntactic structures.6

Finites Finite sentences show inflectional reflexes of the functional categories on the predicate. The verb (head) determines the subject and object(s) (arguments), their thematic roles and the morphological case that they bear. In (3) KupiDə, call is a transitive verb requiring a caller and one who is called. The order within the verb phrase (VP) shows that Tamil is right-headed. The direct object bears the accusative case and the subject the nominative case. We also see inflectional markers on the verb for tense, person, gender and number. One much-discussed property of languages with such verbal inflections is the occurrence of null subjects. This is especially relevant in the context of acquisition as we will see later in this chapter.

5

6

We draw a distinction between the actual case-marking on an NP (its morphological case) and the structural/configurational position it occupies (abstract case, as in Marantz, 1991). Readers may refer to Sarma (2002) for discussion on syntactic operations.

Issues in the acquisition of Tamil verb morphology (3)

kiran kuupD-r-aa enn-a7 (kiran kuppiD-ar-aa enn-ai) Kiran(f)-N call-pres-3sf me-A Kiran is calling me. [TP[NP kiran] [VP enna kupDr-aa]]

113

(2;4;5) (adult)

In finite intransitives (unergative or unaccusative), the verb selects only one argument and bears the tense and agreement markers as illustrated in (4). (4)

naa (naan I-N I will cry.

ayi-v-een azhu-v-een) cry-fut-1s

(2;4;5) (adult)

In some sentences, the subject bears a non-canonical morphological case. Typically, we find the dative–accusative (DA) and the dative–nominative (DN) configurations, and for at least one verb muDi ‘to be able’ the instrumental– nominative pattern. Predicates that require the DA configuration include verbs of mental experience such as teriya ‘to know’ and puriya ‘to understand’, verbs of emotional experience such as piDikka ‘to like’ and verbs of physical or biological experience such as pasikka ‘to be hungry’ and valikka ‘to pain, to hurt’. (5)

ramanu-kku Rama(m)-D Rama likes Sita.

siitaav-ai Sita(f)-A

piDikk-um like-3sn

As seen in (5), the verb bears the default agreement features of the 3rd person, neuter, singular, which are not shared by the subject or the object. However, the dative-marked NP, despite the non-canonical morphological case, is the subject. This can be demonstrated by the following two standard tests for subjecthood: (i) anaphor binding and (ii) subject control (PRO) (Sridhar, 1979). In Tamil, an object anaphor can only be bound by a subject and no other argument. The fact that the anaphor in (6) must be bound by the dativemarked NP (Rama) is therefore very strong evidence of the dative NP’s subjecthood. (6)

7

raman-ukku Rama(m)-D Ramai likes selfi.

tann-ai self-A

piDikk-um likes-3sn

The word order is not canonical and in fact the object has been displaced. See Sarma (2003) for discussion on word order and language acquisition.

114

Vaijayanthi M. Sarma

The matrix subject also determines the referential features of the null subject of an embedded non-finite verb, as can be seen in (7),8 where the dative pronoun controls the embedded subject. (7)

ena-kku kudiray-ai ooTTa piDikk-um I-D horse-A to drive like-3sn I like to ride a horse. [TP ena-kkui [VP [TP PROi [VP kudiray-ai ooTTa]] piDikk-um]]

The DN configuration is of greater interest since the object agrees with the verb as in (8). Such agreement is generally taken to be a feature of subjecthood. The verb can also carry tense morphology. This pattern is seen in a closed class of predicates expressing existence or possession such as iru ‘to be’ and uL ‘to exist’, need or obligation such as poodu ‘be enough’ and veNDu ‘want’ as in (9), in certain complex verbs such as ñaabakam vara ‘to remember’ and kaadu kekka ‘to hear’ and finally, in some modal auxiliaries. (8)

raman-ukku naalu Rama-D four Rama had four sons.

(9)

ee-kku (ena-kku I-D I want food.

mam-mam mam-mam food-N

pasanga sons-N veeN-um veeND-um) want-3sn

iru-nd-aanga be-past-3p (2;9;4) (adult)

The dative-marked NP is the subject and not the nominative object as is shown in example (10), where the anaphor is bound not by the object but the dative NP. The oblique nominal also determines the reference of the null subject of the embedded verb as in (11). (10)

raman-ukku oru payyan tan Rama-D one boy-N self-G Ramai has one boyj in selfi/*j’s village.

(11)

raaman-ukku tan payan-ai aDikka tevaipaT-T-adu Rama-D self-G boy-A to beat be necessary-past-3sn It was necessary for Rama to beat his boy. [TP raaman-ukkui [VP [TP PROi [VP tan payan-ai aDikka]] tevaipaTT-adu]]

uur-ile village-L

iru-kk-aan be-pres-3sm

NP-V agreement does not uniquely signal the subject. The subject position is determined by its structural position. Since these are common verbs, a child

8

The internal structure of the sentence is also given.

Issues in the acquisition of Tamil verb morphology

115

learning Tamil must also be able to deal with these split properties of the subject as well as the case configuration determined by each verbal lexeme. Tamil also permits optional drop of the accusative case marker as can be seen from (12) and (13). (12)

ena-kku ellaam-ø (ena-kku ellaam I-D everything-N I know everything.

teriy-um teriy-um) know-3sn

(2;9;4) (adult)

(13)

eekku adu (ena-kku adu I-D that-N/that-A I am hurting there (It hurts me).

valik-r-adu valikka-ra-du) hurt-pres-3sn

(2;1;18) (adult)

The object-NP has either been incorporated into the verb to form a complex verb form or is interpreted as non-specific as illustrated by (14) and (15). (14)

ena-kku talaiy-ai I-D head-A My head is hurting.

(15)

ena-kku talai I-D head-ø I have a headache.

valikk-ara-du hurt-pres-3sn valikk-ara-du hurt-pres-3sn

(Case-drop)

Additionally, the object cannot be a [þrational] noun. We do not pursue this further, but it is important to keep separate a DA configuration that surfaces as DN and a true DN configuration. Finally, the verb can be in any tense but, as it does not agree with the dative subject, the verb surfaces with default values of agreement (as in (14) and (17)) when the object bears accusative case and agrees with the object when it is truly nominative (as in (8) and (9)). ! ! !

(16)

[α person] [β number] [γ gender]

(17)

ee-kku rattam (ena-kku rattam I-D blood-N I am bleeding.

[–speaker, –addressee]9 [singular] [neuter] va-r-du va-ra-du) come-pres-3sn

(2;1;15) (adult)

There are also some intransitive, dative subject predicates as illustrated by (18).

9

The default values are also seen in negatives (Ramadoss & Amritavalli, 2007).

116

Vaijayanthi M. Sarma

(18)

ee-kku (ena-kku I-D I am cold.

kulur-du kulur-u-du) cold-pres-3sn

(2;1;18) (adult)

In the next section we look at the V-corpus to determine the range of structures that are attested and the ‘correctness’ of use of the properties of case, agreement etc.

Finite sentences in the corpus The tacit assumption, thus far, has been that the processes and properties that underlie adult grammar are also a part of the child’s developing grammar and further, that some manifestation of them ought to be discernible in the performance data. This full-competence hypothesis (Poeppel & Wexler, 1993) has been argued to be simpler than structure-building or structurematuring approaches (Radford, 1990). This question concerns primarily the degree of deviance of the child’s performance from the adult’s performance, and for V, it is practically zero. The child is approximating the adult’s speech very successfully and her performance is consistent with an analysis where the grammatical know-how is in place at a very early stage (Sarma, 1995). A summary of the production data for agreement marking on the verb is given in Figure 10.1. The line tracings show the use of various agreement

60 2;1;15 50

2;1;18 2;2;28

40

2;4;5 30

2;4;22 2;6;7

20

2;9;4 10 0 1s

2s

3sm

3sf

3sn

1pl

2pl

3mpl

3fpl

Figure 10.1. Production across different person/number/gender in the corpus.

3npl

Issues in the acquisition of Tamil verb morphology

117

Table 10.1 Match between predicate and subject case Overt NPs only

Nominative case

Dative case

Other

DA/DN predicates NA predicates

0 257

30 0

0 0

250 200 150

Negative-default

No. of 100 instances

Quirky-default ‘Real’ agreement

50 0 1sg

2sg

3sm

3sf

3sn

1pl

2pl

3mpl

3fpl

3npl

Agreement forms

Figure 10.2. Default versus real agreement.

forms from the last seven transcripts (between the ages of 25 and 33 months) when the use of agreement is most prolific. The agreement forms include the use of both overt agreement and default agreement markers. The x-axis indicates the various agreement feature combinations (as permitted in Tamil) and the y-axis indicates the number of instances of each form in these transcripts. The use of the 1st and 3rd person far exceeds all other combinations at all ages. The 3rd person, neuter, not surprisingly, is far more frequent than either masculine or feminine. Recall that the default agreement on verbs is the ‘3sn’. Figure 10.2 separates the instances of default agreement from actual subject– verb agreement for the third person, neuter. V’s production data has the complete range of sentences discussed above.10 Her grammatical competence encompasses the differences between various predicates and what they entail in terms of case and agreement. The distinctions between default and real agreement are also present. In the corpus, there are no errors of case marking as can be seen in Table 10.1. The subject case is strictly determined by the choice of predicate. 10

Raghavendra and Leonard (1989) also demonstrate the remarkable competence of children with verbal forms.

118

Vaijayanthi M. Sarma

Null subjects The null argument or pro-drop phenomenon is a property of languages with rich agreement.11 In Tamil, pro is licensed in all case-marked (structural or inherent) NP positions as shown in (19–22). Our attention will be focused mainly on subjects and objects, which are in a structural relationship with the verb. (19)

naa (pro) (naan (pro) I-N I spilt (that).

(20)

(pro) idula oo-kku ((pro) idula una-kku (I-N) this-L you-D (I) will give you (that) in this.

(21)

(pro) (pro) kuLppaaT-r-een ((pro) (pro) kuLuppaaTTa-r-een (I-N) give a bath-pres-1s (I) am giving (it) a bath.

(22)

(pro) ((pro)

koTT-in-een koTT-in-een) spill-past-1s

(2;1;18) (adult)

(pro) (pro)

ta-r-een12 ta-r-een) give-pres-1s

(2;2;28) (adult)

(2;1;18) (adult)

(pro) (pro)

vaaN-aam (2;9;4) veeND-aam) (adult) want-neg-3sn (I) don’t want (that). (Lit: (X) doesn’t want (Y))

In (19) the object NP has been dropped, while in (20) and (21) both the subject and the object NPs have been dropped. The identification of the subject is signalled by the agreement suffix on the verb, 1st person, singular. In (22), a dative case subject and the nominative object have been dropped and the verb bears default agreement features. pro is characterized in the same manner as other NPs and carries case, agreement and categorial features but is phonologically empty. Its reference is fixed by context or by an antecedent in an appropriate position. The structural relations of the antecedent and pro pairs are like those between an antecedent and a pronoun.13 In the previous section we saw the extent of V’s knowledge with respect to agreement and case assignment. Figure 10.3 summarizes the corpus and shows the proportion of use of overt subjects and objects (referential NPs and pronouns) to pro in subject and object positions.

11 12 13

In Rizzi’s theory (Rizzi, 1986) pro is licensed in a Spec-Head relation to ‘strong’ AGR. We have shown the pro in the canonical object position – before the verb. We ignore here the position of adjuncts and other postpositional phrases. That is, governed by Condition B of the Binding Theory.

Issues in the acquisition of Tamil verb morphology

119

2.5 Nominative

2

Accusative Dative

1.5 Proportion of overt NPs to pros

1

0.5

0 2;1;15

2;1;18

2;2;28

2;4;5

2;4;22

2;6;7

2;9;4

V’s age

Figure 10.3. Overt NPs versus pro.

As we might expect, pro is amply attested in the corpus in subject and object positions. Since the use of pro is a discourse-governed function, we expect the use of pro to vary unpredictably. We find that this is indeed so, and V’s competence also extends to pro. Imperatives One of the striking features of any corpus is the use of the ‘bare’ verb or the verb stem – a verb without overt inflectional morphology. This has led to hypotheses about the acquisition of agreement or the lack of it at early stages of development (Radford, 1990) and is particularly hard to prove or disprove in languages without rich agreement. In the V-corpus there are a number of bare verb stems. However, these inflectionally impoverished forms can be shown to be the imperative forms of the verb. Context offers one kind of evidence, and the imperative is uttered in the discourse context of an order or request and only those verbs that can be so employed are utilized such as give, beat, go away, put down, come, etc. V uses the imperative forms frequently with accompanying gestures (for example, an outstretched hand when asking for something). The meaning is quite clear from context as is illustrated in (23). (23)

(Situation: Vanitha: Vanitha:

Father pinches Vanitha) eey! kiLLaada! eey! aDippeen

(Situation: Vanitha: Father: Vanitha:

Father tugs Vanitha’s hair) eey eey. Translation: ennaaccu? Translation: eey! en mayira Translation: puDikkaada!

Translation: Translation:

(2;4;5) Hey! Don’t pinch! Hey! (I) will spank (you) Hey! Hey! What happened? Hey! Don’t hold my hair

120

Vaijayanthi M. Sarma

Table 10.2 Imperative forms in Tamil Affirmative

Negative

Verb (infinitive)

Gloss

Singular

Plural

Singularb

Plural

tara poga vara paarkka tuukka

to to to to to

taa poo vaa paaru tuukku

taanga poonga vaanga paarunga tuukkunga

tar-aat-ee pog-aat-ee var-aat-ee paarkk-aat-ee tuukk-aat-ee

tar-aat-iinga pog-aat-iinga var-aat-iinga paarkk-aat-iinga tuukk-aat-iinga

give go come see carry

a

a

There is an appended second person honorific suffix and a plural marker -nga. This form can be interpreted either as singular, honorific imperative or as the plural imperative. b The singular imperative negative form includes the negative morpheme -aat and a euphonic suffix -ee.

The second kind of evidence is syntactic and includes subcategorization properties of the individual verb, case and agreement inflection creating a continuum between imperatives and the grammatical competency described earlier. Tamil distinguishes between singular and plural imperative forms in the negative and affirmative as shown in Table 10.2 for a few verbs. The imperative is marked for number and, by default, the second person agreement feature as well. This is especially clear in the plural negative imperative form where the honorific/person suffix is identical to the inflectional suffix for second person plural (see (2)). These forms are not a simple concatenation of a bare verb and a negative auxiliary but are uniquely derived. (24)

eDukk-aad-a (eDukk-aad-ee) take-neg-euph ‘Don’t take!’

(2;2;28) (adult)

There are several instances of the use of the imperative in the corpus (see 24).14 The subject or the person being ordered/requested is always the second person and is seldom expressed. When expressed, it is present in the vocative case with the accompanying intonational peak to capture the attention of the person as in (25). (25)

14

appaa (appaa Papa-V Papa! Don’t cry!

azh-aad-a azh-aad-ee) cry-neg-euph-imp.sg.

(1;9;16) (adult)

Fifty-five instances of the imperative before the child is 21 months old.

Issues in the acquisition of Tamil verb morphology

121

Apart from the unexpressed subject, the verbs are seen with a full range of complements, including embedded sentential complements. In (26) and (27) the object bears accusative case. The complex verb in (27) also contains an aspectual auxiliary. In (28) we find a finite sentential complement of the verb and in (29), a non-finite sentential complement. (26)

anda pustakatt-a (anda pustakatt-ai that book-A Take that book!

eDu eDu) take-imp.sg.

(2;4;5) (adult)

(27)

appa inda pant-a (appa inda pant-ai Papa-V‵ this pant-A Papa! Take these pants off!

(28)

naa enna (naan enna I-N what-A See what I have [PRO [CP

(29)

nii eendu pooy eDuttiNDu vaa (2;6;7) (nii ezhundu poyi eDuttu-koNDuvaa) (adult) You-V rising going bring-have-pple come-imp.sg. You get up and go bring it and come! [nii [PRO [[PRO ezhundu][[PRO poyi] [[PRO eDuttu-koNDu] vaa]]]]]

awttuDu avuttu-viDu) removing-finish-imp.sg.

(2;9;4) (adult)

iidi iru-kk-een paaru (2;6;7) ezhudi iru-kk-een-(nu) paaru) (adult) writing be-pres-1s-that see-imp.sg. written! naan enna ezhudi irukk-een-(nu)] paaru]

The internal structure and of an imperative construction is practically identical to the finite constructions discussed earlier. The main difference is the null subject which inherits its reference from the contextually salient entity, the addressee. The data in the corpus bear out V’s competence in the use of these forms. In the next section, we discuss a final range of structures that are involved in subordination of clauses that are also well-attested in the corpus. Infinitivals and verbal participles Non-finite sentences/clauses in Tamil include infinitivals and verbal (nonadjectival) participles (VBP) and are used in clause-embedding. Predicates vary in whether they take infinitival or VBP complements. There is no overt subject and no tense and agreement inflection on the verb in a VBP. One of the most common uses of the VBP is in auxiliation. The auxiliary (aspectual and attitudinal15) bears the inflections and the main verb appears 15

We borrow these labels from Lehmann (1989).

122

Vaijayanthi M. Sarma

in the VBP form. These structures are well-instantiated in the corpus as illustrated by (30)–(32). (30)

(pro)

(31)

poliis (pro) uvvaa (poliis uvvaa police-N (X) hurt The police(man) was wounded (X).

(32)

(pro) maavu (maavu (I-N) flour-ø I have made flour.

caaviy-a (caaviy-ai (I-N) key-A I threw away the key.

tuukki tuukki lift

eenju-TT-een erinju-viTT-een) throw(vbp)-left-1s

(2;2;28) (adult)

paNNi-TT-aan paNNi(vbp)-viTT-aan) do-left-3sm

paNNi-rukk-een paNNi(vbp)-irukk-een) make-am-1s

(2;1;15) (adult)

(2;4;5) (adult)

The auxiliary is affixed to the main verb and has no lexical function. The arguments are determined by the main verb16 and other material cannot interrupt the main verb-auxiliary sequence. The auxiliary is a reflex of aspectual features that need to be expressed. In (30) and (31), V uses the perfective aspect and in (32), she uses the progressive aspect. Of greater interest is the use of VBPs in sub-ordination and co-ordination and the use of infinitival complements. We saw one instance of embedding while discussing imperatives (29). Interpretively, the subject of each predicate refers to the same entity, the one to whom the order is given, and all the embedded verbs are interpreted as imperatives. If the matrix verb carried the 3rd person plural agreement and a past tense, perfective aspect feature, then all the embedded verbs would be interpreted likewise (see (33)). (33)

They got up and (they) went and (they) brought (it) and came.

Infinitivals occur most frequently as complements to modal verbs. Unlike aspectual auxiliaries, it is the modal verb which is the main verb and determines subject case. (34)

naa alloo solla poo-p-r-een (naan hallo solla poga-por-een) I-N hello to say to go-go-pres-1s I am going (in order) to say ‘hello’.

(35)

(pro)

16

kay ambi-kka-num (kay alambik-koLLa-veND-um) I-D hand wash-to have-want-3sn I want to wash (my) hands.

A DA/DN main verb will take a dative-marked subject.

(2;1;18) (adult)

(2;4;22) (adult)

Issues in the acquisition of Tamil verb morphology (36)

kaDay-kku pooy (kaDay-kku pooy

123

ena-kku jem ena-kku jem

vaangi-NDu varaN-um (2;1;18) vaangikvara-veND- (adult) koNDu um) store-D goI-D gem-ø buy-vbpcomevbp have-vbp must-3sn Going to the store I want to buy gems and come.

In (34), the embedded infinitival clause alloo solla is itself embedded within another infinitival poga ‘to go’. The matrix verb ‘want’ (in (35) and (36)) takes a dative-subject (enakku) and surfaces with default (3sn) agreement (-um). Its complement is an infinitival clause with an infinitival verb, alambikoLLa ‘to wash’ in (35) and vara ‘to come’ in (36). In (36) there are further embeddings that are in the VBP form (pooy, vaangi) for co-ordination. The subject of all the embedded clauses is construed with the matrix subject ‘I’. The tense information is constant across the sentence. From the examples we see that the V-corpus has a number of finites and non-finites and that the entire range of verb inflections are attested in her production data. This is not surprising when we understand that the underlying mechanisms are fundamentally the same and V’s competence across a range of structures follows from the acquisition of the same verbal processes. Theoretical discussions on language acquisition have broadly separated themselves into approaches that argue for a progressive building of grammatical structure with experience and one where full competence is assumed with maturation accounting for delays in the appearance of certain processes. The acquisition data from Tamil shows us that when inflections emerge they do so quite early in development and in an all-at-one-go manner and in a wide variety of (indirectly related) grammatical processes, which seem to most directly support a full-competence hypothesis. Conclusion Our intention at the start of the chapter was to discuss aspects of developmental syntax with a primary focus on verbs, and the associated properties of case, agreement and argument structure. Children need to acquire the forms of verbs and inflections, as well as the associated auxiliary elements to express grammatical features such as tense, aspect, attitude, mood etc. In languages like Tamil, where the bulk of the structural information is carried by the verbal group, many other phenomena such as reflexivity, voice and mood also follow from the verbal complex. A Tamil native speaker’s competence in the verbal system constitutes the nucleus around which the remaining grammatical properties of the target language are constructed. The child’s competence in the verbal system is therefore critical to arriving at a fully developed grammar.

11

Fast mapping of novel words in bi/multilinguals Vishnu K. K. Nair, Sunil Kumar Ravi, Sapna Bhat and Shyamala K. Chengappa

Introduction Fast mapping is a term coined by two eminent child-language researchers, Carey and Bartlett (1978), to explain how children acquire vocabulary so rapidly during their early developmental years. According to traditional developmental psychology literature, fast mapping is a process that enables children to create lexical representations for the unfamiliar words they encounter. Throughout the course of word learning, children rapidly construct the meaning for new words they encounter (Dollaghan, 1985). For example, in one of the earlier experiments conducted with 3–4-year-old children, Carey and Bartlett (1978) observed that when exposed to a novel referent, children were able to map some information concerning the novel referent after just a single exposure. To explain this process further, Carey and Bartlett suggested two separate stages for word learning: a fast mapping phase and a slow mapping phase. Fast mapping is defined as the rapid or fast construction of unfamiliar knowledge by children during the process of novel word learning (Carey & Bartlett, 1978). Fast mapping is considered to be the preliminary phase in the long-term process of lexical acquisition (Dollaghan, 1985). Carey and Bartlett (1978) observed that during the fast mapping phase, children construct a partial or incomplete knowledge of the phonology, semantics or syntax of the new word that they encounter. In the later or slow mapping phase, children form an association between the phonology, semantics and syntax through exposure to the word in various communicative contexts, so that eventually they develop an adult-like formulation (Carey & Bartlett, 1978; Dollaghan, 1985; Gray, 2006; Kan & Kohnert, 2008). Studies in the 1980s and 1990s examined fast mapping and its role in enriching the vocabulary of children, in particular during early language development (e.g. Dollaghan, 1985; Markman, 1989; Golinkoff et al., 1992). Most of these studies concluded that children in their early developmental years are able to fast map a novel word, and so facilitate early lexical development (e.g. Markman, 1989; Golinkoff et al., 1992; Gray, 2006). 124

Fast mapping of novel words in bi/multilinguals

125

Over the last decade, researchers have focused on how children with language impairment map a newly encountered word with their limited attention and memory skills. Many studies have examined fast mapping abilities in children with language impairment, especially in children with hearing impairment, specific language impairment, Down’s syndrome and autism (e.g. Dollaghan, 1985; Gray, 2006; McDuffie, Yoder & Stone, 2006; Parish-Morris et al., 2007; Luyster & Lord, 2009). Evidence obtained from these studies points towards the fact that word learning difficulties in children with language impairment is primarily due to reduced fast mapping abilities. Other studies have attempted to investigate the effect of factors such as phonological memory and vocabulary in the ability to fast map a novel word. For example, Gray (2006) assessed fast mapping abilities and its relation with phonological memory and existing receptive vocabulary in 53 children diagnosed with specific language impairment (SLI) and 53 typically developing children aged between 3 and 6 years. The experiment was carried out in three phases. First, the experimenter named each of the eight items (four target and four common words) for each child. In the second phase, the experimenter administered a comprehension probe for each of the target and common objects and in the last phase a production probe was administered for each object. Each correct response was given a score of 1. The results indicated that the performance on the fast mapping task differed significantly only in children over 5 years of age with the typically developing children outperforming the SLI children. They concluded that neither phonological memory nor existing receptive vocabulary predicted fast mapping ability in either group of children. The majority of the studies on fast mapping have focused on how typically developing children and children with language impairment map novel words. Very few studies have investigated fast mapping skills in children growing up in bilingual and multilingual contexts. Studying fast mapping skills in bi/multilingual children can be a good indicator of their word learning skills with respect to each language. It can also in principle improve our understanding of how lexical representations are created and stored by these children. Fast mapping in bi/multilingual children Research directly investigating bi/multilingualism and fast mapping from a cross-linguistic perspective are few in number. Kan and Kohnert (2005) investigated the ability of young monolingual children to fast map new word forms, and concluded that age and existing vocabulary knowledge can influence children’s ability to fast map a novel word. However, a later study on bilingual Hmong–English-speaking children, by the same authors, concluded

126

Vishnu K. K. Nair et al.

that fast mapping performance was not related to age or existing vocabulary knowledge. They also compared the results of receptive and expressive fast mapping between the L1 (Hmong) and L2 (English) and found that receptive fast mapping probes were comparable in the L1 and L2. However, performance of the participants was greater in the expressive fast mapping probes in the L1 than L2, although less than half of the participants were able to accurately name newly introduced items in their L1 (Kan & Kohnert, 2008). The results also suggest cross-language relationships between the L1 receptive vocabulary and L2 expressive vocabulary and between the L1 expressive fast mapping and L2 receptive scores on both vocabulary and fast mapping tasks. These results suggest that bilinguals and monolinguals, as they have different learning histories, respond differently to novel word learning tasks. Nair et al. (2010) investigated fast mapping skills in bilingual and multilingual adults and concluded that there was no difference between the performance in the L1 and L2. However, the performance of multilinguals in the L1, L2 and L3 was found to be jointly determined by language proficiency, exposure and spoken usage. Variables that influence bi/multilingual fast mapping There are several factors that are crucial to consider in relation to bi/multilingual language performance in most tasks including fast mapping. Phonological memory and existing vocabulary knowledge can be a strong predictor for fast mapping ability (e.g. Gray, 2006). The general process of memory can significantly influence word learning. How do bi/multilinguals organize memory for lexical storage or language learning in the brain? The concept of shared and separate memory representation in bilinguals may possibly explain how memory is represented in bi/multilinguals (Ijalba, Obler & Chengappa, 2004). This proposal states that bilinguals may store their languages either in a shared memory system or in a different memory system for each language. However, currently there is no research to support whether bi/multilingual subjects use shared or separate memory systems during a fast mapping task. According to Ijalba et al. (2004), lexical processing relies heavily on the declarative memory system, which is anatomically represented by the medial and temporo-parietal structures in both the left and right hemispheres. Second-language learning heavily relies on this system and therefore may tend to rely more on declarative memory. These authors also report that languages, that are learnt in a more structured and less natural setting are represented in the cerebral cortex to a greater extent than the first language (refer to Ijalba et al. (2004) for a more detailed review and explanation of these models). However, with no existing research findings, it may be premature to comment on the role of declarative memory during a fast

Fast mapping of novel words in bi/multilinguals

127

mapping task in bi/multilinguals. The role of other cognitive variables such as attention, executive function and working memory in fast mapping also needs to be explored further. Another important variable that may affect fast mapping performance is the age of language exposure and manner of language acquisition in a bi/multilingual individual. In the Indian context, the majority of children are either simultaneous or successive bilinguals. Simultaneous exposure of mother tongue, father tongue and the regional language is a very frequent occurrence in urban India. Hence, a child may already be exposed to two or three languages in the oral context of the home environment by the time he/she enters school. Specific studies comparing the performance of fast mapping in simultaneous and successive bilinguals (early and late bilinguals) can give us greater insights into the manner in which language acquisition and exposure influences lexical learning. Language proficiency and existing vocabulary knowledge are two other critical factors that can influence fast mapping performance in the L1, L2 and L3 of bi/multilinguals. However, in the Indian context assessing linguistic proficiency and existing vocabulary knowledge can be difficult due to the lack of standardized bi/multilingual proficiency tools or vocabulary measures to collect such information in each language involved. A cross-linguistic study on fast mapping In the following section, we report the results of a fast mapping study conducted on bi/multilingual children. Given the fact that word learning is mediated through fast mapping, we adopted a typical methodology used to investigate fast mapping and lexical learning abilities in bi/trilinguals. A fast mapping paradigm typically involves selection of a number of novel words and their referents. The words are introduced to the children in natural language situations and then probes are used to evaluate retention ability through a comprehension and production task. Although there is no clear consensus regarding the number of exposures needed for fast mapping, most of the studies introduce the novel words for a very brief time ranging from single to multiple (three to four) exposures. In the present study, we phonologically manipulated the structure of the novel word and introduced these words to bilingual children through a natural story narration. Initially, we decided to probe fast mapping after a single exposure; however, piloting revealed no fast mapping after one exposure in bi/trilinguals. Hence, the novel words and their referents were presented to participants on five separate occasions. We selected Tulu–Kannada–English trilinguals and Kannada–English bilinguals (two Indian language communities) for the study. The participants

128

Vishnu K. K. Nair et al.

of this study were sequential bi/trilinguals who learn the majority language from birth (L1) and use it as a mother tongue for primary functions in their society. They begin to learn a second language (L2) – and, in the multilingual case, a third language (L3) – from early childhood and use the L2 (and where applicable L3) for formal functions in their society (similar to the definition used by Jia, Kohnert & Collado (2006) for Spanish–English–Dutch trilinguals). These children had acquired the L1 primarily in the home environment whereas the acquisition of the L2 and L3 was learned in a more formal, educational setting at about 5 years of age. In the Southern Karnataka state of India, where this study was conducted, Kannada is the dominant language. It is spoken by over 40 million people. Tulu is a widely spoken language with 1.5 million speakers, and although it is not the dominant language, it is a robust feature of the linguistic landscape of Southern Karnataka. Its speakers use Tulu with relatives and friends and Kannada as a spoken language in institutional settings, and English is used to meet certain formal and educational needs. It is relatively easy to find Tulu–Kannada–English trilingual speakers and Kannada–English bilingual speakers, who differ only in their language-learning histories and are otherwise comparable. The data was elicited from 20 trilingual children, who spoke Tulu as the L1, Kannada as the L2 and English as the L3, and 20 bilingual children whose L1 was Kannada and L2 was English. All the subjects were schoolchildren aged from 8 to 10 years. The subjects were assessed for their proficiency in all languages spoken using the International Second Language Proficiency Rating scale (ISLPR: Wylie & Ingram, 2006). Twenty-four novel nonsense words (eight words in each language) were constructed to assess fast mapping skills in these children. Initially, a nonsense word stimuli pool was created with sound combinations typically found in Kannada, Tulu and English, which conformed to the phonological characteristics of these languages and were pronounceable. The novel words used in the study were selected after being validated by two linguists. Eight novel words in each language were grouped into four pairs; 24 pictures were selected on the basis that the participants were unlikely to be very familiar with them. Each of these pictures was connected to one novel word. A total of 12 short stories were devised with four stories in each language. Each pair of words was embedded in the story in such a way that no two novel words occurred within a single sentence. For example, the English novel word penears was associated with a picture of a synagogue – an image that the participants were unlikely to be familiar with. The novel words hugura (Kannada) and jeppula (Tulu) were associated with pictures of an avalanche and a submarine respectively. A word-learning task was employed, which consisted of five sessions. In the initial phase, these novel words were introduced by a live story narration

Fast mapping of novel words in bi/multilinguals

129

in conjunction with a picture presentation. The novel words appeared in every narrative session, thus maximizing the participants’ exposure to these words. The stories were repeatedly presented in each language for the next four sessions. Each participant was assessed for the acquisition of these novel words following the five phases of the story narration. The assessment tasks consisted of a referent identification and picture-naming task. In the referent identification task, a set of three picture choices was provided. The picture choices included the target referent and two distracters, a semantically related referent and an unrelated referent. The target referent was then presented in auditory mode via a loudspeaker and the participant was instructed to point to the picture that corresponded to the auditory stimulus. In the picture-naming task, the target referent was presented via a laptop computer and the participant was requested to name it. Naming and referent identification scores for bilinguals in picture-naming and referent identification were very similar for both the L1 and L2 (Kannada–English). The bilingual children had equivalent high levels of proficiency (S: 4 level proficiency on ISLPR) in both Kannada and English. However, the Tulu–Kannada–English trilinguals exhibited a very different pattern of results. Naming scores in English (L3) were better than in Tulu (L1), which in turn was better than in Kannada (L2). The trilingual children were more proficient in the L3 than L1 and L2 (S: 4 level proficiency for L3; S: 3 level proficiency for L1 and L2). The greater the proficiency in the L3 accounts for the better naming scores for English (L3) than Tulu (L1) or Kannada (L2). As proficiency increases, there may be a larger association between lexical forms and semantics, which leads to faster lexical retrieval (Kroll & Stewart, 1994; Kroll & de Groot, 1997). Interestingly, the referent identification scores across languages did not vary significantly across multilinguals. The repeated language comprehension task should have facilitated rapid lexical access in the bi/multilinguals. This may be the reason for similar identification scores across languages. In addition, we also examined the error patterns for the naming task in all the languages of the children. The percentage of word repetition errors for the participants in both the groups was calculated. A detailed analysis of word error types in each language was conducted (see Table 11.1). This was performed using the Li and Williams (1990) checklist, an extended version of the error categorization system devised by Kohn and Goodglass (1985). The types of error observed in the L1 (Kannada) were syllabic repetitions (3.75%), syllabic substitutions (3.75%), additions (1.87%), part word repetitions (0.62%), rejection errors (0.62%) and reduplications (1.25%). Percentages of syllabic repetitions and substitutions were greater in the L1; no final consonant deletion or phonemic omission errors were observed. The error types in the L2 included syllabic repetitions (1.25%), syllabic

130

Vishnu K. K. Nair et al.

Table 11.1 Analysis of word errors in the L1 and L2 Word errors

L1 (%)

L2 (%)

Syllabic repetition Syllabic substitution Addition Related word error Part word repetition Rejection error Reduplication Final consonant deletion Phonemic omission

3.75 3.75 1.87 — 0.62 0.62 1.25 — —

1.25 1.25 0.62 2.5 1.87 0.62 3.12 6.25

Table 11.2 Analysis of word errors in the L1, L2 and L3 Word errors

L1 (%)

L2 (%)

L3 (%)

Syllabic repetition Syllabic substitution Addition Related word error Part word repetition Rejection error Reduplication Final consonant deletion Phonemic omission

4.75 3.25 1.55 — 0.42 1.25 0.62 — —

3.75 3.5 1.85 — 0.25 2.25 0.425 — —

1.55 0.82 0.25 2.5 1.87 0.62 — 4.25 5.85

substitutions (1.25%), additions (0.62%), part word repetitions (1.87%), rejections (0.62%), final consonant deletions (3.12%) and phonemic omissions (6.25%). Final consonant deletion and omission errors were more frequent in the L2 (English). Word-error analysis in the particular languages showed maximal percentages for syllabic repetitions (4.75% and 3.75%), syllabic substitutions (3.25% and 3.5%) and addition errors (1.55% and 1.85%) in Tulu (L1) and Kannada (L2) respectively (see Table 11.2). Part word repetition (0.42% and 0.25%) and rejection errors (1.25% and 2.25%) also occurred in the L1 and L2. However, errors such as final consonant deletions and phonemic omissions were not observed in either Kannada or Tulu. In English (L3), the highest percentages were observed in the case of omission errors (5.85%) and final consonant deletion errors (4.25%), followed by part word repetitions (1.87%). Interestingly, irrespective of language proficiency, both bilinguals and trilinguals exhibited similar kinds of error patterns. Syllabic repetition, substitution and addition errors were found to have the highest scores in Kannada

Fast mapping of novel words in bi/multilinguals

131

and Tulu. There were a greater proportion of omissions and final consonant deletion errors for English. This finding could be interpreted as a reflection of the major structural differences between English and the other languages. Correspondingly, the similar proportion of error types in Kannada and Tulu may be due to the close linguistic similarity (cognate) between these languages. Phonemic deviation had the highest score; such deviations occur more often in a second language. This could be due to weaker phonological representation knowledge of the second language. However, little is understood about error patterns during the learning of novel words in the case of Indian languages and it would be premature to do more than suggest the possibility that the L2 phonological system may be weaker than the L1 system. Theoretical standpoints on fast mapping and directions for future research Bi/multilingual fast mapping research will shed more light on how lexical representations are mediated and stored in a bi/multilingual speaker. However, the majority of premises for this research are based on Western studies. Indian and Western bilingual populations differ in many aspects of individual and societal bilingual practices. Thus, results from these Western studies cannot be directly generalized to Indian bilingual populations. Instead, they help us to explore the possibilities of larger and more focused dimensions of research in relevant domains of fast mapping specific to Indian multilingual contexts, which itself is a varied and complicated terrain. Language dominance as a confounding variable also needs to be accounted for. Even though both populations have English in their repertoire, it may be dominant in one context but not in the other. Language dominance can also vary across the lifespan, changing over time. A previously less dominant language may become more dominant in an environment where it is used more frequently. This is especially true in the Indian linguistic context where individuals are frequently on a bilingual continuum. India is a country where ordinary people in their everyday lives are often exposed to novel words in languages in which they have relatively low proficiency, for example, among people from the lower socio-economic status who change their jobs on a daily basis. It is important to understand that fast mapping of novel words is not just an idiosyncratic reflection of a person’s linguistic history, but that generalizations are possible and involve such factors as language knowledge, proficiency, degree of exposure and opportunities of frequent conversational usage. It is thus imperative to achieve some understanding of bilingual contexts and strategies that come into play when bilinguals encounter new words.

132

Vishnu K. K. Nair et al.

The studies reviewed in this chapter highlight the importance of conducting further research on fast mapping in bilinguals and multilinguals. As we work towards the establishment of clinically useful norms for bi/multilingual societies, it is essential to improve our understanding of different linguistic repertories and the effects they have on a person’s word perception ability for each of the respective languages. The results of such studies will also be invaluable in relation to setting treatment goals for language-impaired people from bi/multilingual backgrounds. Novel word learning is an interesting phenomenon and has various research and clinical implications. Indian contexts are predominantly bi/multilingual and so offer ideal contexts to investigate the effect of bi/multilingual factors on the ability to fast map novel words. Factors such as proficiency, language usage patterns of the community and linguistic input across languages as well as influence of the media may differentially influence fast mapping skills. These contributing factors need to be examined in greater depth in the future.

12

Studies on the acquisition of morphology and syntax among Malay children in Malaysia: issues, challenges and needs Rogayah A. Razak Introduction

This chapter will survey studies conducted on the acquisition of morphology and syntax among Malay children in Malaysia. Studies will be described so as to give an account of how research in this area has progressed and aspects of research which are still lacking will be highlighted. Issues and challenges will also be raised and the future needs in terms of research and its applications will be elaborated on.

Background Though this chapter covers studies which were carried out in Malaysia, we will briefly mention studies carried out elsewhere in the region, namely Singapore and Indonesia. This is to illustrate the diverse nature of research conducted and also to contribute to the overall picture of developmental Malay morphology and syntax studies in the region. Aman (2007 and Chapter 9) examined the acquisition of wh-questions among 30 Singapore Malay preschool children (with a mean age of 4;2) with ages ranging from 3;4 to 4;6, who were tested using experimental (elicited imitation tasks, picture-story comprehension task) and longitudinal methodologies (spontaneous production tasks). Aman examined children’s understanding of three different types of wh-questions, that is, questions which involved moved wh-words, wh-in-situ (wh-word in base position) and focus questions. The children’s knowledge of long-distance movement and the role of island constraints were examined. Aman found that children between ages 4;5 and 6;5 seemed not to respect island constraints on wh-movements. It was argued that the island constraints are the result of a processing effect in which in-situ wh-questions are not subject to islands, but instead are primes for the responses in fully moved questions. Children used relative clause marker ‘yang’ in both headless (e.g. Yang colour orange panjang itu apa? ‘What (is) that (thing that is) long (and) orange (in colour)?’) and headed relative clauses (e.g. Nanti kita baca buku 133

134

Rogayah A. Razak

yang lain ‘We (will) read a book that (is different than this) later’). This was attributed to the filler-gap strategy rather than the reconstruction strategy. Yassir Tjung (2006) compared the use of relative clauses in Jakarta Malay, particularly the subject–object asymmetry, among 20 adult subjects and three children. Yassir found that the majority of relative clauses produced by children, as with adults in naturally occurring speech, tended to be subject relative clauses (e.g. Kucing yang ditangkap sama anak jadi kuning ‘The cat that is being caught by a child is turning yellow’) rather than object relative clauses (e.g. Kucing yang anak tangkap jadi kuning ‘The cat that a child is catching is turning yellow’). Yassir concluded that the greater processing cost of object relative clause was alleviated by the ease of the passivization process. Work on developmental literacies on local languages such as Malay, Mandarin and English are actively being investigated at the National University Singapore. Jalil and Rickard Liow (2008) examined how home language influenced early spelling in Malay children in Singapore and Jakarta. In the former, the group was said to be diglossic due to the non-standard Malay variety used, and in the latter case the group were monolingual standard Malay speakers. A word-spelling test, which comprised 51 high-frequency Malay words taken from the Dewan Bahasa Pustaka Malaysia Literary and Language Agency database comprising primary education textbooks, was used. The words were recorded by a female native Malay speaker who used standard Malay pronunciation. They found that early spellings among diglossic Singaporean Malay children are speech-based, therefore affirming that the exposure to non-standard phonology influenced the kinds of errors made by the children. They explicated the spelling process through the concept of redintegration, that is, restoration of the whole of something from fragments using long-term representation to support short-term recall. Vowel substitution errors were made: for (e.g., pelik ! pelek ‘strange’), and for (e.g., kebun ! kebon ‘garden’). They concluded that diglossic children based their spellings on their own non-lexical phonological representations because they made phonologically plausible errors for the non-standard form even when the target words were dictated to them by an adult speaker using a standard form of the language. Early linguistic experience was found to have influenced the development of phonological representations and hence literacy skills of the Malay children in Singapore. In another study, word recognition in English and Malay was investigated by Yap, Rickard Liow, Jalil and Faizal (2010). Words were taken from a 2.14-million-word corpus by Dewan Bahasa Pustaka based on newspaper articles in Berita Harian, a major Malay daily in Malaysia. A lexicon database composed of variables (such as letter length, syllable length, phoneme length, morpheme length, word frequency, orthographic and phonological neighbourhood size) for 9,592 Malay words was compiled. Measures of word length, word

Acquisition of morphology and syntax among Malay children

135

frequency, neighbourhood size and lexical decision task performance on word recognition were then found to be significant measures in memory processing when learning Malay words. This Malay lexicon database is a crucial and essential tool for researchers conducting research in this area. In Malaysia, research on developmental morphology and syntax (e.g., Omar, 1988; Simanjuntak, 1990; Mohamad Noor, 2002; Arshad & Subramaniam, 2006) have been sporadic, involved small numbers of children, were often in the form of anecdotal accounts and were not focused enough to contribute to the explanatory specifics of language acquisition. These studies provided a general picture of Malay child language development, and at best show that the language abilities of Malay children are varied and increase with age. Furthermore, most studies show substantial individual variations and as such were not able to be used to make generalizations, thus failing to provide normative data for Malay children. However, there are studies (e.g. Simanjuntak, 1983; Long, 1993; Tan, 1999; Razak et al., 2010, 2011) which have employed quantitative coupled with qualitative methods to analyse the data collected. These studies employed data-collection methods consisting of either spontaneous speech or elicited responses from structured tasks or data collected in experimentally designed tasks. These studies have significantly increased our understanding of Malay child language development. This chapter will describe these studies in detail. Due to this limited research, there is a dearth of resources particularly for the purposes of determining milestones and ascertaining normative data in Malay child language. The need now is to establish milestones that will indicate at which age Malay children acquire certain sounds, morphemes and words, phrasal and clausal categories, and even the developmental milestones of semantics and pragmatics. Information about milestones in language acquisition is important as a basis for researchers, speech-language pathologists, teachers and parents to use as a yardstick to determine whether a child is developing typically. For example, it would be beneficial to have a Malay equivalent of Brown’s (1973) work on the acquisition of English morphology. This would be useful for developmental studies and for studies on language impairment including studies determining clinical markers for Malay morphology. Acquiring this complex morphology is particularly challenging for Malay children with specific language impairment (SLI). The unavailability of this kind of data is a stumbling block for studies on SLI. Another pressing need is for normative data for the assessment and management of communication impairments. This situation has led to efforts by Razak, Madison, Siow and Aziz (2010) and also Razak, Aziz, Lim and Jin (2011) to employ methodologies that combine the collection of crosssectional developmental data together with the development of relevant language assessment tool prototypes. These combined methodologies,

136

Rogayah A. Razak

utilizing elicited and spontaneous data, are one means of solving the dire need for typical developmental data for the management of atypical development or communication difficulties in speech therapy clinics. It is also a step towards contributing to a body of knowledge on the psycholinguistics of first language acquisition of Malay. Approaches and methodologies Previous studies on the acquisition of Malay have adopted various approaches, namely longitudinal case studies, cross-sectional studies and experimental studies. These different types of studies will be reviewed. Longitudinal studies Simanjuntak (1987) carried out a longitudinal study investigating the phonological, syntactic and semantic acquisition of three Malay children from 10 months to 3;6 years old. He found that the children started to combine words at 2 years old following the holophrastic stage. The fact that Malay language is a pro drop language was reflected in the children’s utterances, for example Mak, makan ‘Mum, eat’ where the object noun phrase is dropped from the verb phrase makan ‘eat’ and Nak ikut ‘want (to) follow’ where the subject noun phrase is dropped. Simanjuntak used pivot grammar to describe the syntactic constructions of one of the 2-year-old children, as presented in Table 12.1. Table 12.1 Pivot (P) and open class (O) structure in utterances No.

Utterances

Meaning of utterance

1

Mami bayang (P) (O) Gloss: Mother pray Beg mami (P) (O) Gloss: Bag mother Ampu mbang (P) (O) Gloss: Light plane Ai makan (P) (P) Gloss: Come eat Mau ai apu (P) [(P) (O)] Gloss: Want drink iced

‘Mummy prays’

2

3

4

5

‘Mummy’s bag’

‘The plane’s lights’

‘(Please) come (and) eat’

‘(I) want iced water’

Acquisition of morphology and syntax among Malay children

137

The utterances in Table 12.1 demonstrate the emerging syntactic system of a 2-year-old, which consisted of combinations of the (P)(O) and (P)(P) patterns, and variation of both patterns (P)(P)(O) on which utterances are built. These words are early word forms as some were partial syllabic forms. He noted that by 3 years 6 months, the child had a good grasp of the syntactic constructions. Tan (1999) carried out an extensive longitudinal study on the acquisition of wh-questions and yes/no questions among five children between the ages of 18 and 48 months over a period of 2 years 6 months by recording the language used at each weekly home visit. Parents were asked to report when their child started using new question words. The objectives of the study were to determine the age at which Malay children started to use question words, the order of acquisition of question words, and the structure of the questions. The first wh-word to appear was mana ‘where’, apa ‘what’ and siapa ‘who’ at 20–21 months for two children and between 26 and 30 months for three children. This was followed by kenapa ‘why’ (31 months) and macam mana ‘how’ (34 months). The last wh-word to be acquired, but rarely used, was berapa ‘how much’ (35–6 months). Further expansion of the usage of wh-words was evidenced by the use of preposition þ wh-word and wh-word þ particle. Examples of the first type are dekat siapa ‘near whom’, dengan siapa ‘with whom’, dengan mana ‘which one’, untuk apa ‘for what’, macam apa ‘like what’, kat mana ‘where at’ and macam mana ‘how’. The expanded forms were not evident in children before 28 months. An example of wh-word þ particle is apakah ‘what þ particle’ in Emak buat apakah? ‘What is mother doing?’ produced by a 26-month-old child. The yes/no questions, on the other hand, were preceded by intonation questions followed by intonation with particles such as ah, kah, tah and dah. Negation particles such as (bu)kan and tak emerged later to convey the meaning of whatever the child already knew but needed confirmation on. Another formal structure used in yes/no questions was the modal auxiliary boleh ‘can’ as in bolehkah, bolehtah, bolehkan (28–38 months). For the yes/ no questions there also appeared to be variability among subjects. Kader and Tan (2002) discussed the stages of syntactic development that children in their study demonstrated; the progression from one word to up to more than three words. They concluded that the developing syntax corresponded with the developing cognitive abilities of the children. Cross-sectional studies Cross-sectional studies are significant due to the number of participants involved and the generalizations which can be obtained from the phenomena studied. We will discuss one important cross-sectional study on the

138

Rogayah A. Razak

acquisition of Malay affixes among Malay preschool children. Long (1993) conducted a cross-sectional study on the development of affixes among preschool children. She used a combination of direct observation of spontaneous language use by children at home and at their preschools with their parents/ caregivers and with their teachers respectively. Direct observation of a structured language task was also conducted to analyse the understanding (knowledge) and production (use) of the verbal and nominal Malay affixes among 5- and 6-year-old Malay preschool children. Twelve children and their respective parents or caregivers were selected as subjects. The study hypothesized that children’s patterns of acquisition of affixes could be influenced by the adults’ use of affixes. The conversations children had with their parents or caregivers and teachers were analysed on the morphological processes used, for example simplification of the affixes by the children or the total absence of affix use referred to as bare forms. Adults’ conversations directed to children were also examined. They found that preschool children, in general, were able to use the majority of verbal and nominal affixes. Only a small group of affixes were not used at home or in school such as loan prefixes and suffixes, nominal infix -em-, verbal infixes and verbal circumfix ke-. . .-an. According to Long (1993), preschool children at 5 and 6 years old were not able to master the affixes as they had not yet reached Piaget’s concrete operational stage, which occurs at about 7 years old. Morphological development is considered to be an ongoing developmental process for the preschoolers and would be completed when they reached the final year of the primary schooling at 12 years old. Among the affixes, it was shown that bare verbals and nominals had the highest mean score over the other affixes. Examples given were as illustrated below: (1)

Abang gunting kertas. (absence of prefix meN- on the verb) Brother cut paper ‘Brother is cutting the paper.’

(2)

Emak dia masuk.

(absence of prefix meN- on the verb and absence of object NP)

Mother her enter ‘Her mother entered (the room).’ (3)

Dia ada mainan. He has toy ‘He has a toy.’

(absence of the prefix peR- on the noun)

Examples of affixes (underlined) used by the children is given below. (4)

Dia terjatuh dari basikal. He Aff-fall from bicycle ‘He accidentally fell from his bicycle.’

Acquisition of morphology and syntax among Malay children (5)

139

Dia baling dengan penyepit. He throw with Aff-pince ‘He threw (something) with a pincer.’

Children tended to use words in their bare root forms compared to their derived forms. This is in line with the assumption that children tend to learn unmarked forms earlier than marked forms. Interestingly, this pattern could also be seen among adult native speakers in their interaction with their children. However, the mean scores for verbal and nominal affixes used at home seemed to be higher than those used in school. There were also some nominal and verbal affixes which were used at home but not in school. These affixes were the nominal affixes such as the circumfixes ke-. . .-an, peN-. . .-an, peR-. . .-an, the affixes -is, -wan, ke- and infix -er-. The verbal affixes on the other hand are prefixes di-, ke-, circumfixes di-. . .-kan, me-. . .-i, peR-. . .-kan, meN-. . .-kan, beR-. . .-an and di-. . .-i. Generally, girls tended to be more proficient in using affixes and had higher mean scores than boys, but the difference between boys and girls did not reach significance. Only certain affixes such as the bare verbals, and prefixes beR- and teR- were significantly different, whereas for other affixes such as the suffix -kan, -i, prefixes meN-, peR- and se-, the differences were not significant. Similarly, there were differences in mean scores between age ranges but generally the differences were not found to be consistent nor significant. Children also tended to use nominal and verbal forms in simple sentences with the SV and SVO sentence patterns (mean 328.92, SD 138.91) significantly more than compound derived sentences (mean 25.25, SD 14.52). The most frequent active sentences used by children seemed to be sentences with the prefix mem- omitted. Preference seemed to be for the unmarked use of the active sentence (e.g. Adik beli selipar instead of Adik membeli selipar ‘Small brother bought a pair of slippers’). Long suggested that the reason behind this might be that bare active verbs were easier to process by children both at the receptive and expressive levels. Young children would normally use affixes in the intransitive simple (as in Example 10) or compound sentences. The compound sentences used were typically a combination of compound sentences (6–7) or with a question (8). (6)

Jika kita pandai ambil hati, kita akan disayangi. If we clever take heart we will be Aff-love ‘If we are good at pleasing others, we will be loved.’

(7)

Kami berasa gembira pada hari ini kerana hari ini hari penyampaian hadiah. We Aff-feel happy on day this because day this day giving presents ‘We feel happy today because today is prize-giving day.’

(8)

Ni alat pembesar suara kah? This equipment Aff-magnifier voice Q-PRT ‘Is this a microphone?’

140

Rogayah A. Razak

Passive sentences used by children were found to be generally quite limited. When they did use passives, they tended to use passives with bare verbals with the accompaniment of the auxiliary-like verb kena ‘get’ or with the prefix teR- rather than with the other passive verb forms like di-, ber- or ke-. . .-an. (9)

Orang ini kena tangkap. Man this got catch ‘This man got caught.’

(10)

Payung terbuka. Umbrella Aff-open ‘The umbrella was opened.’

Long noted that children tended to use more bare verbals in simple sentences compared to complex sentences. Long’s (1993) study made a significant contribution to our understanding of morphological development of affixes among preschool children. It also demonstrated that a rich home environment is more conducive to children using affixes compared to the school environment, in spite of the didactic nature of schools. Her discussion on the role of adult speech in enriching children’s language and the inclusion of caretaker talk puts this study firmly on the forefront of Malay language acquisition research in the 1990s. Studies on the development of assessment tools Increasingly, in the area of communication disorders in Malaysia, there is a need for typical developmental child language data, which functions as a yardstick or baseline from which to assess atypical language development in children. Nevertheless, due to the disparate and sporadic child language studies conducted prior to 2000, there was limited information available that could be utilized towards this goal. Razak et al. (2010) described the development of a Malay Preschool Language Assessment Tool (MPLAT) prototype designed to assess the language abilities of 101 Malay preschool children between 4;0 and 6;11 years of age. The MPLAT is a prototype assessment tool that measures receptive, expressive use of language and early literacy skills. It has six subtests, namely picture vocabulary, grammatical understanding, sentence repetition, relational meaning, referential meaning and early literacy. In constructing the MPLAT, receptive vocabulary, grammatical structures, types of sentences (e.g. wh-questions, yes/no questions, commands, passives, negation) and early literacy skills were included. The objective was to create a language assessment tool that included important elements of semantic knowledge, receptive language and literacy skills as well as grammatical elements that Malay children use in their daily language interactions.

Acquisition of morphology and syntax among Malay children

141

Table 12.2 Mean and standard deviation by age group for total MPLAT raw scores Age

N

Mean

Standard deviation

4;0-4;5 4;6-4;11 5;0-5;5 5;6-5;11 6;0-6;5 6;6-6;11

7 7 7 7 7 7

37.36 39.71 46.86 66.82 97.43 105.64

8.58 8.18 8.27 13.92 11.26 20.68

The scores on the MPLAT tested on 42 children were found to increase with age (refer to Table 12.2). The correlation between age and total scores was found to be statistically significant (r ¼ 0.85, df ¼ 40, p ಕನ (kana) (a non-word) ಮಠ (mata) (temple) -> ಮರ (mara) (tree) ಹಸು (hasu) (cow) -> ಅಸು (asu) (a non-word)

Literacy in Kannada

189

The reading and dictation errors of VN were mostly confusion between the aksharas: ಸ read as ನ and ಠ read as ರ, where there is a difference of one dot, and ಹ for ಅ, which is confusion with the akshara as well as sound. Even though these errors can be classified as visual errors, they could also be interpreted as orthographic errors. Nevertheless, these cases show a clear dissociation between decoding and comprehension even in Kannada, an alphasyllabic language. Previously such dissociation had been shown mainly in English. Joshi and colleagues interpreted their findings based on the Componential Model of Reading (CMR) (Aaron et al., 2008; Joshi & Aaron, 2000). CMR is an extension of the Simple View of Reading (SVR) (Gough & Tunmer, 1986; Hoover & Gough, 1990) which shows a clear dissociation between decoding and comprehension. Alexia in a Kannada–English bilingual Alexia is the loss of a once well-developed reading skill. In one of the earliest studies, Karanth (1981) reported a case, NR, a 57-year-old, right-handed male, who suddenly lost his ability to read words in both English and Kannada. After 6 months of tutoring in both English and Kannada, NR was able to read words and sentences, albeit slowly. His performance and self-report indicated that he was better in reading English than Kannada, even though he had been exposed to Kannada longer than English before the onset of alexia. Karanth explained NR’s difficulty with reading Kannada as due to the orthographic complexity of akshara, despite its being a highly transparent language. Kannada instruction Even though the akshara – the basic unit of writing in Kannada – is arranged for instruction in an orderly fashion, with the vowels arranged first and then the consonants, an informal survey of Kannada instruction in elementary schools showed that generally there is no systematic instruction utilizing the linguistic characters of vowels and consonants. Instead, the emphasis is more on rote memorization and repeated tracings of akshara and words. The basic format of akshara, with vowels and consonants, is introduced in Grade 1, and by the end of Grade 2 students are expected to master the complex syllable matrix of Kannada orthography. However, as Nag (2007) noted, even in Grade 4, children have problems learning the complex syllable patterns of Kannada. To examine whether the transparency of Kannada orthography can be utilized in teaching English as a second language, Nishanimut, Johnston, Joshi, Thomas and Padakannaya (in press) taught three groups of students in Grade 5 different methods of reading in English. One group was taught with

190

R. Malatesha Joshi

the traditional synthetic phonics instruction with no reference to Kannada orthography and a second group was taught with the Samveda Synthetic Phonics Program (SSPP), which used a variation of the synthetic phonics instruction making use of the Kannada akshara–sound correspondence to teach English letters and words. For example, the teacher introduced the letter ‘l’, pronounced its sound /l/ and represented the letter in Kannada akshara. The students then saw the letter highlighted in an English word (e.g. lion) and heard it pronounced by the teacher. The students then wrote the English letter, the Kannada akshara and the English word containing the target letter. Their performance was compared to a control group, which did not use systematic phonics instruction. After 5 weeks of instruction, students in the two synthetic phonics instruction performed significantly better than the comparison group; however, the group taught by SSPP performed significantly better than both the groups. The results of this study demonstrated the importance of using the knowledge of first language (L1) in teaching English as a second language. Additionally, the transparent nature of Kannada facilitated learning of English as many sounds of English can be written in Kannada also. Cheung, McBrideChang and Tong (2011) carried out a series of studies involving Chinese (morphosyllabic writing system) and Korean (syllabic/alphasyllabic writing system) children learning to read in their native languages and in English (alphabetic writing system). They concluded that ‘alphabetic English reading is cross-linguistically facilitated only if written Chinese or Korean is taught in an alphabetic way, so that phoneme-level phonological awareness in promoted’ (Cheung et al., 2011, p. 184). .

Conclusion The nature of Kannada and a selection of research studies involving Kannada orthography is presented in this chapter. Kannada orthography is considered to be an alphasyllabic script with a fairly consistent relationship between the written unit (akshara) and the sound. Even though the orthography is transparent, literacy acquisition is still a difficult task, perhaps for various reasons. Poor teaching methods, the large number of students in each classroom and lack of books in school as well as at home may be some of the challenges facing literacy development of children in Kannada. Additionally, the complex alphasyllabic nature of the akshara of Kannada could also contribute to the literacy problem. Based on the studies on dyslexics, phonological aspects and the complex nature of aksharas seem to be important determining factors in the literacy development of students in Kannada. However, in cases of bilingual dyslexia and hyperlexia, the performance on Kannada and English literacy tasks was similar perhaps due to the inherent presence of phonetic elements in akshara. An exploratory study of synthetic phonics instruction

Literacy in Kannada

191

revealed that making use of the sounds of Kannada akshara helped in learning English as a second language. Even though some progress has been made in exploring the nature of Kannada orthography, literacy development, literacy problems and literacy breakdown, more research needs to be conducted. One of the areas that needs to be explored is the development of standardized assessment materials exploiting the nature of Kannada aksharas. Instructional materials and procedures would be important areas for further research as well.

18

Reading in Tamil: a more alphabetic and less syllabic akshara-based orthography Bhuvaneshwari B. and Prakash Padakannaya

Tamil belongs to the southern branch of the family of Dravidian languages, a family consisting of around 26 languages native to the Indian subcontinent. Tamil is spoken by around 65 million people worldwide (Asher & Annamalai, 2002). It is primarily spoken in Tamil Nadu, a southeastern state of India, and in the northern and eastern parts of Sri Lanka. Tamil is also spoken by migrant communities in Malaysia, Singapore, Mauritius, Fiji, the United Kingdom and the United States, among other countries. Tamil is diglossic (Ferguson, 1959), with a marked difference between the forms used in writing and formal speaking, and in informal conversations. Tamil script is akshara-based, which typically represents language at the (orthographic) syllable level. It is a relatively opaque system with a distinctly different visuospatial organization (Nag, 2007), as all the elements of akshara are written as in-line characters in contrast to other akshara-based orthographies. Tamil has fewer graphemes compared to other Brahmi-derived scripts. A proficient reader makes use of conventions as well as sentence context and prior lexical knowledge while decoding Tamil. This chapter discusses features specific to Tamil orthography and their implications for reading research, pedagogy and remediation. Tamil script The Tamil script is believed to have been derived from the Brahmi script. Asokan Brahmi of the third century BCE is the common source from which all Indian scripts, both Indo-Aryan and Dravidian, evolved. It was an alphasyllabic script with diacritics used for vowels occurring in post-consonantal position. The writing system was based on the concept called akshara or ‘graphic syllable’, which has a vowel as the final constituent (Krishnamurthi, 2003). According to Bright (1996), alphasyllabaries are a group of orthographies that represent sounds at the level of the syllable but have distinctive features to indicate subsyllabic information. He defines alphasyllabic scripts as those scripts in which ‘each consonant-vowel sequence is a unit, called an aksara, [and] in which the vowel symbol functions as an obligatory diacritic to 192

Reading in Tamil

193

Table 18.1 A comparison of the characteristics of a traditional alphabet and a syllabary Alphabet

Syllabary

Distinct graphic elements for vowels and consonants Graphic units/subunits correspond to individual phonemes rather than to words and syllables

Primary unit is the graphic syllable Syllabic units are in most cases indivisible in the sense that at least some of their component parts, i.e. the secondary or ‘diacritic’ vowel signs, cannot stand alone

the consonant’ (Bright, 1996, p. 492). Salomon (2000) described alphasyllabaries as classes of characters that are neither strictly alphabetic, in that they do not represent a single sound unit or phoneme, nor strictly syllabic, in that they do not stand for a single and indivisible syllabic unit. He further describes the features of alphasyllabic scripts as follows:
The physical graphic unit is the syllable, typically of the types V, CV, CCV, etc.
An unmarked consonantal graph is understood to have an automatic or ‘inherent’ vowel (in the Indian scripts, the so-called ‘short a’) following it, unless an explicit mark for another vowel overrules the implied neutral vowel (e.g. in Tamil, a ‘pulli’ or an overdot diacritic mutes the inherent vowel).
Vowels other than the inherent vowel, when following a consonant, are indicated by the addition of an extra ‘diacritic’ sign, which is typically attached directly to the consonantal character.
Vowels that do not follow a consonant (i.e. word-initial vowels or the second vowel in a V–V sequence) are represented by separate graphs, namely ‘full’, ‘initial’, or ‘independent’ vowel signs. Alphasyllabic scripts share the characteristics of a traditional alphabet as well as a traditional syllabary. This is illustrated in Table 18.1.

Tamil phonology The Asokan Brahmi script was adapted to suit Tamil phonology in two respects: graphemes for non-occurring phonemes such as aspirated stops were omitted and graphemes for characteristic Tamil phonemes such as |zh| were added (Lehmann, 1998). Many of them were borrowed from Grantha, the southern variant of Brahmi script. Tamil has 10 vowels including 5 short and 5 long vowels and 2 diphthongs. Tamil consonants are classified into three

194

Bhuvaneshwari B. and Prakash Padakannaya

types (|vallinam| meaning hard consonants, |mellinam| meaning soft consonants and |idaiyinam| meaning middle-level consonants) with 6 letters in each type, thus making a total of 18 consonants. The hard consonants include க (k, g), ச (ch), ட (T, D), த (t, d), ப (p, b), ற (r), soft consonants include ங (ng), ஞ (nj), ண (N), ந (n), ம (m), ன (n) and the medial consonants include ய (y), ர (r), ல (l), வ (v), ழ (zh), ள (L). The 12 vowels (including diphthongs) and 18 consonants combine to produce 216 consonant–vowel alphasyllabographs. All consonants have an inherent vowel; diacritic markers or secondary vowels in the form of ligatures on the sides of consonants are used to lengthen this inherent vowel or add other vowels to the consonant. The inherent vowel can be nullified by using an overdot diacritic called ‘pulli’. Apart from the 12 vowels, there is a special character called ‘aytham’ – represented as three dots (ஃ) as in a shield (and hence the name). It is neither a pure vowel nor a pure consonant. This letter is used to glottalize the sounds that it combines with. It does not usually occur with words of Tamil origin. It is mostly used to modify Tamil syllabographs to represent foreign sounds such as /f/, (e.g. an ‘aytham’ followed by ‘p’ would represent ‘f’, as in ‘fees’, which is written as ‘aytham followed by |pi:s|’ ஃபீஸ்). Some consonants occurring in the nontraditional positions in loan words are represented by letters borrowed from the Grantha script. These include |s|-ஸ, |dz| – ஜ, |sh|- ஷ, |ksh|- கஷ, |h|- ஹ and |sri| – ÿ. The matrix of Tamil akshara is presented in the appendix. Some of the features of the Tamil phonological system are distinctly different from other Indian languages, and are reflected in its orthography. Tamil does not differentiate between voiced and unvoiced consonants (e.g. |p| and |b| are represented by the same letter ‘ப’). Voicing of these sounds is governed by specific rules. Plosives are unvoiced if they occur in the initial position of the word or if they are doubled (geminated) in medial positions. They are voiced when they occur in medial position. Since the word final position always contains a vowel, there is no question of plosives occurring at the final position. If they do occur as pure consonants between phrases (at word ends), they are unvoiced. Consider the word பட° |padagu|; the plosive in the initial position is unvoiced and the plosives in the medial position are voiced. To take another instance, கப்பல் – |kappal|, the medial plosive here is unvoiced because there is gemination. The voiced and the unvoiced plosives in Tamil seem to be in a complementary distribution when analysed phonologically (Asher & Keane, 2005). Tamil orthography does not make any distinctions between aspirated and unaspirated consonants as native Tamil words do not contain aspirated consonants. Another notable feature is that consonant clusters are restricted in native words. There are no clusters in the word initial or final positions. Medial clusters are restricted to combinations of liquids and/or nasals with stops, which occur only in syllable offsets (Steever, 1998). Modern Tamil has all the characters written in line. In all

Reading in Tamil

195

other Brahmi derived scripts in India, the vowel diacritics as well as the consonants occurring as consonant conjuncts are spatially arranged on all sides of the consonant (i.e. left, right, above, or below). All the conjunct consonants are written in Tamil using ‘pulli’. Tamil morphology and syntax Tamil has an ‘agglutinating’ or ‘concatenating’ morphology. Words are formed by concatenating morphemes one after the other. Nouns and verbs inflect for tense, person, number, gender, etc. The resultant words formed by concatenating morphemes are functionally different. Case suffix markers and post-positions added to word stems result in changes in grammatical categories thereby bringing about meaning changes. Tamil sentences generally follow a subject–object–verb pattern. However, Tamil is a relatively free word-order language, and the constituents of the sentence can be shifted to provide emphasis, without affecting the semantics. Irregularities in Tamil orthography Some of the vowel diacritic markers are non-linear, which means that the written order of phonemic elements is different from their order in the spoken form (e.g. |ke| is written as ெக). The diacritic marker for |e| appears before the consonant |k|, while in the spoken form the consonant comes first followed by the vowel. The diacritic marker for vowel /o/ has a compound form in the sense it is represented jointly by |e| and |a:| diacritics appearing before and after the consonant as in ெகா. A notable feature of Tamil orthography, as mentioned earlier, is the absence of differentiation between unvoiced and voiced consonants. No distinction is made between aspirated and unaspirated consonants as well, because native Tamil words do not have aspirated consonants. Therefore a single grapheme க can refer to |k|, |kh|, |g| as well as |gh|. The grapheme க also refers to |h| – when it occurs in medial position in some words. The other five of the six hard consonants are also pronounced/read in various ways depending on their position in the word. The grapheme ச is pronounced as |s|, |ch|, |j|. The grapheme ட is pronounced as |T| or |D|; த is pronounced as |t| or |d|; ப is pronounced as |p| or |b|; ற is pronounced as trilled |r| or |tr|. This oneto-many mapping of grapheme to phonemes becomes a potential source of confusion for novice readers (and sometimes for skilled readers too) while reading. This is especially so for pronouncing loan words, transliterated words, or in some instances, even non-words. For example, பாவம் is read as |pavam| if read using the rules governing hard consonant pronunciation in Tamil. However, if it is read as the loan word from Sanskrit, it is read as

196

Bhuvaneshwari B. and Prakash Padakannaya

|bhavam|. Semantically these two words are very different. காந்தி is another example to show the confusions arising due to a single grapheme representing many phonemes. This word should be read as |gandhi|; however, it is sometimes read as |kandi|. Another characteristic feature of Tamil that is different from the other Dravidian languages is the way consonant conjuncts are represented. There are very few consonant conjuncts in Tamil and they seldom occur in word-initial position. When they occur in other positions, each consonant of the cluster is written separately as a pure consonant with the over-dot diacritic (‘pulli’) and the last consonant in the cluster is written without a ‘pulli’. An exception to this is the clusters borrowed from the Grantha script கஷ |ksha| and ÿ - |sri|. Reading and spelling in Tamil Tamil, as discussed in the previous sections, is an alphasyllabary with a relatively opaque orthography in comparison to other Brahmi scripts. Reading in Tamil is influenced by several factors. One of the factors is related to the large number of visual symbols that need to be mastered. As the grain size of akshara is relatively large, the number of symbols in the orthography is also correspondingly large. These include syllabographs representing pure vowels, their diacritic forms, consonants with inherent shwa, consonants with vowel diacritics, etc. Further, letter knowledge in English would include learning letter names as well as letter-name–sound associations (the sound that each letter stands for). However, children learning to read Indian alphasyllabaries do not need to learn letter-name–sound associations as letter names and sounds are the same. Even though Tamil has fewer alphasyllabographs compared to other Brahmi scripts, learning the correspondence between letters and sounds is challenging, as Tamil orthography has one-to-many correspondences. There are instances where the graphemic sequence is not congruent to the phonemic sequence, and cases of non-linearity. In Tamil, non-linearity is present in certain consonant–vowel sequences. While the phonemic sequence of some consonant vowel combinations is consonant þ vowel, the graphemic sequence is vowel þ consonant, that is, the diacritic marker vowel is placed before the consonant. For example, |ke| is written as ெக, which has the vowel diacritic appearing before the consonant. This kind of non-linearity causes additional cost in processing time as evidenced by Blessy (2007). Blessy, in her study, compared the time taken for reading non-words with and without non-linearity, controlling for all other extraneous factors in a group of skilled readers. The results showed a significant effect of stimulus type. That is, the presence of discrepancy between the spatial positioning and temporal sequencing of phonemic elements resulted in significantly higher response latencies.

Reading in Tamil

197

The results are in concurrence with a similar study in Hindi (Vaid & Gupta, 2002). Such results also suggest that these alphasyllabographs are processed more like an alphabetic script. If the readers were to process it syllabically this non-linearity of phonemic elements should not interfere with reading. In a study on phonemic segmentation in Grade 5 Tamil-speaking children, Wali, Sproat, Padakannaya and Bhuvaneshwari (2009) found that children committed more errors while deleting the initial phoneme from a non-word beginning with a non-linear diacritic marker. The children deleted the vowel sound (denoted by vowel diacritic appearing before the consonant) instead of the first heard consonant sound. This again supports the view that the phonemic elements in the alphasyllabary are represented independently of the akshara that they form. Oral vocabulary is a significant factor in decoding and word learning in opaque/deep orthographies (Share, 2008). In languages with transparent orthographies, this factor is not as significant. With respect to Tamil, oral vocabulary becomes important while reading loan words or transliterated words. Also, since there is no distinction made between voiced and unvoiced plosives as well as aspirated and unaspirated consonants, oral vocabulary and knowledge of the specific spoken forms become important factors that influence decoding. Dialectal variation is another factor that can influence reading, particularly the way in which voiced/unvoiced and fricative/affricate distinctions are made. Tamil is diglossic, and has various dialects. There are significant differences between the forms that are used in writing and speaking. At the phonological level, nasalization, monophthongization, vowel lowering, assimilation and lateral deletion cause differences between these two forms (Matiki, 2010). There are morphological differences too between these forms. Morphemes such as quotative particles, conditional clause markers and perfective aspect suffixes appear in spoken Tamil as bound forms while in written Tamil they appear as free forms (Matiki, 2010). There have been a few studies on reading disorders in Tamil. Aaron (1982) concluded from his study that children learning to read Tamil do have difficulties with inflections and suffixes. Joshi and Aaron (2006), in a crosslinguistic study between Tamil and English, found that Tamil children did not make many spelling errors in Tamil but committed more spelling errors in English. In another study, they compared the spelling performance among two groups of Tamil children, one studying in an English medium, another studying in a Tamil medium and a group of younger children in the USA (controlling for the number of years of exposure to English in school). Tamil-medium children were taught English from Grade 6 onwards while English-medium children were exposed to English from Grade 1. They observed that learning English first as a written language in school helps

198

Bhuvaneshwari B. and Prakash Padakannaya

children learn English spellings but the phonology (not morphology or diglossic nature) of Tamil causes them to make different kinds of spelling errors in English. A study by Akila (2000) on the relationship between phonological awareness and reading in Tamil found that phonemic as well as syllabic awareness is significantly related to reading performance. This is in accordance with other similar studies reported on Indian alphasyllabaries, which suggest that akshara are processed phonemically sometimes and syllabically at other times (Padakannaya & Ramachandra, 2011). These conditions for phonemicand syllabic-level processing of akshara need to be examined further. Reading models and Tamil Reading is an intriguing cognitive process illustrating neuronal plasticity in that there is a recycling of the existing brain centres and circuits originally meant for other sensorimotor functions occurring for reading. Reading is not a spontaneous skill but a skill that needs to be learned. The major challenge in learning to read is to be able to map between orthography, phonology and semantics. Despite the vast amount of work done in this field, particularly in English, there are a number of issues that still remain unresolved. The most interesting debate in the literature has been on the ‘dual route’ and ‘connectionist’ (or ‘single route’) models. In a nutshell, ‘dual route’ theories claim that readers construct lexical entries either by ‘assembling’ phonology from grapheme-to-phoneme correspondences and then looking up the phonological output or by directly constructing a lexical entry (addressed phonology) based on the orthographic input (Besner, 1990; Coltheart, 1978; Paap & Noel, 1989). Assembled phonology is vital for reading novel words, lowfrequency words and non-words. On the other hand, the phonology strategy is not used when reading high-frequency words and irregular words. ‘Connectionist’ models contrast with dual route theories in their mechanisms, because they postulate parallel activation along networks of distributed sublexical information. These models propose that visual word identification (reading) is an emergent property of parallel activation networks (Seidenberg & McClelland, 1989; Van Orden, Pennington & Stone, 1990). Words are merely ‘identified’ as an automatic consequence of patterns of phonological and graphic input features. However, these models are largely Anglocentric. Questions about the universal applicability of these models have been raised given that the scripts of the world differ in terms of the extent to which they depend on the GPC rules for processing (Karanth, 2003). Padakannaya and Mohanty (2004) outlined a psycholinguistic framework to account for reading in Brahmi-derived scripts. They suggested that the default strategy in readers of Indian languages is most likely through akshara-based

Reading in Tamil

199

assembled phonology. However, the relatively less transparent orthography of Tamil probably would show more activation of the lexical route. Wellplanned naming and priming studies (or even imaging studies) in Tamil should shed more light on this issue. Tamil orthography is a unique Indian alphasyllabary, as mentioned in the previous sections. It has more phonemes than the akshara characters. It is certainly more opaque than any other Indian Brahmi-derived orthography. Tamil is also diglossic, with a considerable difference between the spoken and written forms of the language. There are non-linear arrangements of phonemic markers. Tamil is also probably the most alphabetic-like system among the Indian languages, as almost all the characters in Tamil appear in line. These unique features of Tamil make it a useful choice for crosslinguistic studies within Indian languages. Results from comparisons across writing systems (for example between alphabetic and morphosyllabic systems) often cannot be compared due to basic differences between the systems. In many ways, Tamil provides the best link between alphasyllabaries and alphabetic systems (such as Roman script) and also between alphasyllabaries and syllabaries (such as Japanese Kana) for delineating the orthographic independent universal aspects of literacy.

Appendix: The alphasyllabaries of Tamil

(A) TAMIL ALPHABET CHART

‘

’ ‘ ’

‘

’

Source: www.docstoc.com 200

Reading in Tamil

201

(B) CONSONANTS, VOWELS AND COMPOUND FORMS FOR TAMIL

Source: en.wikipedia.org

19

Akshara–syllable mappings in Bengali: a language-specific skill for reading Shruti Sircar and Sonali Nag

Writing systems Writing systems differ in the phonological units they represent. In alphabetic scripts, the basic unit represented by a grapheme is essentially a phoneme although the nature of this correspondence can vary. In languages such as Finnish, every grapheme is realized by only one phoneme and every phoneme represents one grapheme, making these transparent orthographies. Conversely, in opaque orthographies, a phoneme can be realized by different graphemes (e.g. compare /u:/ in ‘to’, ‘too’, ‘two’), and a grapheme by many different phonemes (e.g. the letter [a] in ‘late’, ‘above’, ‘cat’ and ‘car’). In syllabic scripts such as Cherokee written units represent syllables. In contrast, the symbols in alphasyllabic scripts are orthographic syllables but can also be segmented into phonemes (Nag, 2011). In this chapter we examine how sound–symbol mappings influence the ability to read and process phonological units. As we will see, the unique properties of the Bengali writing system make it particularly useful for examining this question.

Phonological and orthographic structure of Bengali Bengali has 7 oral vowels (/i/, /u/, /e/, /o/, /æ/, /ɔ/, /a/), 4 semi-vowels (/j/, /w/, /e̯ /, /o̯ /) and 30 consonants (Ray, Hai & Ray, 1966). Many diphthongs are possible and necessarily consist of one semi-vowel, though only two diphthongal symbols /oi/ and /ou/ are represented in the symbol register. All vowels can be nasalized. Vowel deletion (like schwa deletion) is common in the language, particularly in word medial and final positions. This phenomenon is governed by rules of syllabification, and in marginal cases by word etymology, phonotactic constraints and morphological compositions. Bengali has 16 canonical syllable patterns (Sarkar, 1986), which in descending order of frequency are CV, CVC, V, VC, VV, CVV, CCV, CCVC, CVVC, CCVV, CCVVC, CCCV, CCCVC, VVC and CCCVV. The CV syllables constitute 54 per cent of the whole language (Dan, 1992), while CCCV, CCCVC and CCCVV are infrequent. Consonant clusters occur in 202

Akshara–syllable mappings in Bengali

203

onset but not coda positions, unless they are in loan words (e.g. English /pænts/ or Urdu /goʃt/ ‘meat’). Bengali allows a large set of consonant clusters in word medial positions, particularly in monomorphemic words. The Bengali script, like all other Indic scripts, originated from ancient Brahmi. The fundamental organizing principle of the script is the (orthographic) syllable. An orthographic syllable – akshara – consists of a consonantal core with the vowel arranged around the core. The akshara represent /V/, /Ca/ and /CV/ syllables, the complex /CCV/ and /CCCV/ syllables, as well as the phonemic /C/ (Nag, 2011). Vowels and consonants have a primary and a secondary form; the representation by either is rule-based, thereby making this aspect of the Bengali akshara system consistent. Elsewhere in the symbol set, the coming together of consonants can result in a new akshara where the component segments are either visually recognizable (কþক¼ , ka þ ka ¼ kka) or a fused form, with little scope for applying rule-based combinatorial principles to construct the akshara (কþত¼ , kaþta¼kta). Any application of the construct of consistency needs to account for such intra-orthography variations. Further, the alphabet-focused definition of consistency (e.g. Frost, Katz & Bentin, 1987) as 1 : 1 mapping is also limited and does not cover the 1 : 2 representations of the akshara system (Nag, Treiman & Snowling, 2010). The Bengali orthography is moderately transparent. The vast majority of CV akshara map to their corresponding spoken CV syllables (thus a 1 : 1 mapping of orthographic syllable and phonological syllable) except in the following cases: (a) The vowel এ [e] is pronounced as both /e/ and /æ/ (for e.g. ekA1 /ækɑ:/ ‘alone’ or bega /beg/ ‘velocity’); (b) Consonant clusters are often pronounced as geminates irrespective of the second consonant (e.g. satYa /sotto/ ‘truth’, padmA /pɔddɑ:/ ‘lotus’); (c) the Bengali script has two symbols each to represent the vowel sounds /i/ and /u/, and more than one grapheme for the consonant sounds /ʃ/, /dʒ/ and /n/. In addition, inconsistency arises when /Ca/ akshara in word medial and final positions have vowel deletions (e.g. the ra in দরবার darabAra /dɔrbɑr/ ‘court’; and ka নাক nAka /nɑ:k/) (more examples are illustrated in Figure 19.1).2 But also confusingly, orthographic representations of intervocalic consonants may be read as two full consonants (/Ca–CV/ akshara sequences) or as a ligatured half-symbol (a /CCV/ akshara). Taken together, there are two implications for the young learner: the written language has rules about akshara sequences specific to particular words (graphotactic rules), and there are several instances of inconsistency. 1 2

Bengali graphemes have been represented here using ITRANS and the pronunciations are written in IPA. The inherent vowel of Bengali akshara is pronounced as /ɔ/ or /o/ but for easy reference we shall denote it as [a].

204 1.

Shruti Sircar and Sonali Nag Bengali: Pronunciation Child’s reading:

/pikniker/ /pik niker/

/ /

h h

/p : e/ /p : e/

urir/ urir/

‘beside the picnic basket’

2. Bengali: Pronunciation: Child’s reading:

/ma:tite/ /ma:tite/

/pa:/ /pa:/

h

h

/t ukte/ /t ukte/ h h /t uk te/ /t uk te/

‘stamping his feet on the floor’

/pikniker/ → /pik niker/ h / t ukte/ → /th uk te/ (error with akshara–syllable mapping in medial /Ca/ akshara) boy, 7;10 years, Grade 2

Figure 19.1. Sample of reading errors associated with akshara–syllable mismatch.

We now turn to the results of a study conducted on primary school children to examine the phonological processes that underpin learning to read in the Bengali alphasyllabary, with a special focus on the mapping between the orthographic syllable and the phonological syllable (akshara–syllable mappings).

The study Our study was conducted in five schools in Kolkata (India) and we present descriptive data of 109 7- to 10-year-old children. Typically developing children were recruited from Grade 2 and Grades 3–4,3 and poor readers were recruited from Grades 3–4 on the basis of a screening battery comprising tests of initial phoneme identification, word reading in context and reading comprehension. Bengali was the home and neighbourhood language for all children. From Grades 3–4, 57 typically developing readers participated (TDG3), and 15 children whose scores fell in the lowest 15% on the screening battery were placed in the poor readers group (PR). We also report profiles of 37 typically developing children from Grade 2 (TD-G2). 3

Grades 3 and 4 showed substantial overlap in attainments. On reading accuracy Grades 3 and 4 had the same median. Hence we have chosen to not break up the analyses by grade for these children.

Akshara–syllable mappings in Bengali

205

We assessed akshara knowledge and reading accuracy on words and nonwords based on items chosen for their /Ca/, /CV/ and /CCV/ characteristics. In addition, aligned with our interest in akshara–syllable mapping, we assessed phonological processing. Children manipulated target syllables and phonemes in non-words, in either initial or final positions. Akshara knowledge and word recognition Research in other akshara languages such as Kannada and Malayalam has shown us that learning akshara, particularly the complex, later taught and less frequently encountered akshara, is a prolonged process (Nag, 2007; Tiwari, Nair & Krishnan, 2011). On a Bengali akshara recognition task, we found a similarly extended acquisition phase with some akshara still not learnt by Grade 4. Most errors were on the akshara for consonant clusters. For the /CCV/ akshara on the list, the mean success rate for the typically developing Grade 3–4 children was 70.2%, whereas for Grade 2 children and the poor readers, accuracy was 60.5% and 54.6% respectively. Both the latter groups struggled to give a blended response for consonant clusters, tending instead to ‘spell out’ each sound segment of the akshara. Clusters like [pra] that can occur word-initially and medially and are frequent were recognized more [nta] that occur only word accurately than less frequent clusters like medially. Apart from the complex akshara, however, children in our survey demonstrated mastery of a large number of akshara. This profile of mastery is perhaps reflective of the methods of akshara instruction in the schools in our study. The participating schools followed a phase-wise scheme, teaching /Ca/ first and /CV/ and /CCV/ akshara later. Akshara instruction focused on helping children pull apart akshara, so that they could perceive the components – the base consonant and the vowel signs (Kishalaya, 2004). The gains from such explicit teaching of phonemic markers within akshara are that the children are able to use combinatorial rules and rapidly gain mastery on recurrent features that are applicable to most akshara. Examples of recurrent features are the predictable use of the primary or secondary form of a phoneme and the predictable position of the vowel form in CV akshara irrespective of the base consonant. According to Nag’s model of akshara learning (Nag, 2011), the children are using a strategic approach to akshara recognition. Thus while the less familiar symbols appeared to elicit segmental analysis of the markers within the syllable block (the ‘spelling’ of the akshara), the familiarity of the common akshara appears to allow the symbols to be reliably processed as undifferentiated blocks. On the reading task, accuracy on words was unaffected by akshara complexity and variable akshara–syllable mapping, except for words with inconsistency of /CVCoCV/ spoken words written as /CVCaCV/ akshara

206

Shruti Sircar and Sonali Nag

sequences.4 Our word list contained frequently encountered /CVCoCV/ – /CVCaCV/ contrasted words. Eight per cent of Grade 2 children but none of the poor readers misread such words (see Figure 19.1). For the older children, practice may have bootstrapped recognition; children knew when to read an akshara with or without the inherent vowel [a]. Should the children have encountered unfamiliar and infrequent words, the inconsistent akshara– syllable mapping would have impacted their reading accuracy, a hypothesis we confirm through an analysis of errors on the non-word reading task. In all groups of children, 6% of the errors in non-word reading were lexical (e.g. the non-word pram was read as the word prem ‘love’ or as pratham ‘first’) and 94% of errors were phonological, whereby consonants in the /CCV/ akshara were reversed in reading (e.g. rognA /rɔɡna:/ became rongA /rɔnɡɑ/) or geminated (rogga). But the more interesting interactions were seen between lexical knowledge and decoding when akshara–syllable mappings were ambiguous (see Table 19.1, Panel B for examples). Recall that, given the arbitrary nature of mappings in word-medial position, /CVCaCV/ non-words can be read with or without the inherent vowel. Three non-words [DoTakA, kuTalo, piraShA] were analysed for the nature of phonological analogies used for decoding. Older and more proficient readers read DoTakA and kuTalo without the inherent vowel [a] (as /dɔtkɑ/ and /kutlo/) analogous to the early-acquired familiar words maTka ‘pot’, paTka ‘firecrackers’ and jhaTka ‘a bolt’; but read piraShA always with the inherent vowel (as /piroʃɑ:/) similar to DhAronA ‘assumption’, preronA ‘inspiration’ and ShironAm ‘title’. Many younger children (36%) and poor readers (33%) did not use these analogies, and instead read the medial akshara of DoTakA and kuTalo by sounding out the inherent vowel [a]. These trends in non-word reading reveal for us the possible decoding strategy that may be employed for identifying unfamiliar words with similar opacity within the medial akshara–syllable mappings. While it is not unexpected that inconsistencies in akshara–syllable mappings will be difficult for younger readers and poor readers, our analysis gives us clues of why these difficulties occur. Less skilled readers appear to be faithful, perhaps overly so, to the orthographic information. They may not productively use phonological analogies for generalizations nor actively draw upon them during learning. Phonological processing The complexity of the syllable unit has been shown to be a factor influencing phonological processing (e.g. Nag, 2007). In this section, we examine 4

For representation of word-medial sequences, the written form of CC (orthographic syllables) is italicized and presented as /CVCaCV). The corresponding spoken form (phonological syllables) when carrying a vowel deletion is represented as /CVCoCV/.

Akshara–syllable mappings in Bengali

207

Table 19.1 Examples of children’s responses on non-word reading, syllable substitution and phoneme segmentation tasks Panel A: Non-word reading (girl, 7;4 years, Grade 2) Child’s response Item

Orthographic representation

Phonetic representation

েকফ কূ টেলা পীরশা লেরা েহাকু ল

kefa kuTalo piraShA lerA hokula tipno Telpi

/kef/ /kutlo/ /poriʃɑ:/ /lorɑ:/ /holuk/ /tipon/ /tepli/

Panel B: Syllable substitution (girl, 7;7 years, Grade 2) Item

Position; unit of manipulation Initial; েচ /tʃe/ Initial; জা /dʒɑ:/ Final; ন /no/ Final; চ /tʃ/

/ta:ɡno/ /gokti/ /ʃɔpti/ /bindɑ/

Child’s response /tʃeɡno/ /dʒɑ:kti/ /ʃɔpno/ /bincho/

Panel C: Phoneme segmentation (girl, 7;1 years, Grade 2; boy, 9;2 years, Grade 4) Item খাস /k ɑ:s/ েপক /pek/ /ɡlɑ:t/ /brɔl/ h

Child’s response (Grade 2)

Child’s response (Grade 4)

/k ɑ:/ þ /sɔ/ খা þস /pe/þ /kɔ/ পে þ ক /ɡlɑ:/ þ /tɔ/ þ ট /brɔ/ þ /lɔ/ þ ল

/khɑ/ þ /ɑ/ þ /sɔ/ খা þ আ þ স /pe/ þ /e/ þ/kɔ/ পে þ এ þ ক /ɡɔ/ þ /lɑ/ þ /tɔ/ গ þ লা þ ট //brɔ/ þ /rɔ/ þ /lɔ/ ব þ র þ ল

h

children’s processing of simple and complex syllables for segmentation, deletion and substitutions. Children’s responses are given in Figure 19.2. First, in keeping with universal trends, we found phoneme processing to be lower than syllable processing, with a corresponding increase in phonemic processing as children gained skills in decoding accuracy. Second, in syllable tasks, the complexity of the syllable appears to have some effect on performance (Figure 19.2, Panel A). On the phoneme tasks, the complexity effects were more evident. Manipulation of word initial phonemes in CVC syllables was easier than CCVC syllables for all grade levels and the poor readers (Figure 19.2, Panel B). Third, we found the phoneme processing in poor readers to be lower than Grade 3–4 children, but similar to Grade 2

208

Shruti Sircar and Sonali Nag

Panel B

86.7% 87% 85.3%

85.2% 80.6% 78.6%

CVCV

CVCCV

Types of syllables TD-G2

TD-G3-4

Percent correct responses

Percent correct responses

Panel A

78.2% 57.4%

60.5%

61% 34.3%

CVC

34.7%

CCVC

Types of syllables PR

TD-G2

TD-G3-4

PR

Figure 19.2. Children’s success rate in syllable processing (Panel A) and phoneme processing (Panel B) on simple and complex syllables.

children, suggestive of a delay in the development of phonemic awareness (Figure 19.2, Panel B). Syllable processing and akshara–syllable mappings On a segmentation task, all children had 95% accuracy on non-words with CV and CVC syllables, with errors among some younger children and poor readers located around longer non-words. But beyond the attainments in syllable processing, we were also interested in examining individual akshara–syllable mappings, and the segmentation of intervocalic consonant clusters (CVCCV). Segmentation could occur either as CVC–CV or CV–CCV. In terms of mapping, the former would require an intra-akshara split, so that the first consonant emerges as the coda of the first syllable and the second as the onset of the second syllable. In the latter, the CCV akshara becomes the onset of the second syllable. So, if segmentation was phonological, the CVC–CV parse would be preferred, and if the segmentation was mediated by orthography, the CV–CCV parse would be favoured. In our study, children overwhelmingly preferred the CVC–CV parse. One interpretation of these trends is that akshara–syllable mappings do not seem to interfere with the syllable segmentation process. Furthermore, we found that children’s preferences appeared to be modulated by the phonotactic constraints of the language. In standard Bengali, clusters with sonorant þ obstruent/sonorant sequences (e.g. /gn/, /bd/, /pt/, /kl/, /rm/, /rn/, /rp/, /rt/, /rg/, /rd/, /np/, /nt/, /nd/ /mb/, /lm/ and /lp/) occur only word medially, but an obstruent þ obstruent/ sonorant cluster (e.g. /kr/, /kl/, /gr/, /gl/, /tr/, /dr/, /dhr/, /nr/, /pr/, /pl/, /br/, /bhr/,

Akshara–syllable mappings in Bengali

209

/mr/ and /ml/) occurs both word initially and medially (Kar, 2010). Children’s segmentation showed a preference for the CVC–CV parse (above 85%) when the medial CC could not occur as an onset cluster for the second syllable (e.g. kitnA, nubjA, chorti). However, when the medial CC could occur as onset cluster for the following syllable (e.g. makrA, leddhi, librA), the use of a CVC–CV parse dropped to around 50%. Children appeared to be using their knowledge of phonotactic constraints of the language to guide syllable-level processing. On the substitution tasks, akshara–syllable mappings impacted performance, particularly when the final syllable needed to be manipulated. Again our interest was in CVCCV sequences and whether there was a phonological splitting of the second akshara or an akshara-by-akshara manipulation (see Table 19.1, Panel B). We found that when asked to substitute the first syllable of Tagno with [che], 75.1% of Grade 2, 85.6% of Grade 3–4 readers and 50.6% of poor readers substituted the CV [Ta] rather than the CVC [Tag]. The resultant non-word was chegno, not ‘cheno’. Similarly, in a substitution task with the final syllables, 43.7% of Grade 2 children, 63% of Grade 3–4 and 12% of poor readers chose to substitute the final CV rather than CCV, that is, they substituted the last syllable of Shopti, so that the resultant non-word was Shopno. Again, the preference was for a phonological manipulation rather than an akshara-by-akshara manipulation. From this preliminary descriptive analysis, our findings suggest that the type of akshara–syllable mapping appear to make a difference to certain syllable manipulations. Phoneme processing and akshara–syllable processing We next examine two tasks requiring manipulation of target phonemes in CVC and CCVC non-words. On the CVC non-words, segmentation accuracy of Grade 2 children was at 63.6%, Grade 3–4 children at 81.3% and poor readers at 55.3%. On CCVC non-words, accuracy rates were 49.6%, 71.2% and 45.5% respectively. The trend suggests greater ease with phonemic manipulations on the simpler CV syllable than the complex CCVC syllable. Also, in contrast to the roughly equivalent performance across grades on syllable processing tasks, we found a stepwise developmental pattern on the phoneme tasks (Figure 19.2). Furthermore, two types of segmentation errors on CCVC non-words stood out for their possible association with the characteristics of the writing system. First, more than 35% of children in Grades 2 and Grade 3–4 and /brɔl/ into ba þ ra þ la 45% of poor readers segmented CCVCs such as and not b þ r þ a þ l. Also, rather than segmenting into phonemic units, children appeared to represent the inherent vowel [a] in their segmentation

210

Shruti Sircar and Sonali Nag

Percent incorrect responses

(corresponding to the /Ca/ akshara in the writing system). Moreover, vowels with orthographic arrangements misaligned to the spoken syllable interfered with the segmentation sequence. For example, 12% of Grade 2 children (but none of the Grade 3–4 children) segmented েপক /pek/ as /e/, /p/ and /k/, arguably because the vowel diacritic for /e/ ে orthographically precedes /p/ প. Importantly, the poor readers group in our study did not make such errors, suggesting that the influence of the visuospatial arrangements of the Bengali akshara on phonological processing is not straightfoward. Other evidence of akshara–syllable mapping being a factor in developing phonemic skills comes from the initial phoneme deletion task (deletion of /t/ from /teb/ and /p/ from /preʧ/). Among CVCs, 51.5% of Grade 2 children, 75.08% Grade 3–4 children and 66.6% of poor readers could delete the initial phoneme. The accuracy rates dropped for CCVCs, to 37.5%, 58.95% and 36% respectively. Grade 2 children deleted the whole initial akshara (Ca; CV) in 22% of all instances of inaccurate segmentation, while such errors were less than 10% in Grade 3–4 children and poor readers (see Table 19.1, Panel C). In CCVC words, the frequent error types were deleting either the whole CCV akshara or CC segment, and accurate dropping of the C phoneme, but with downstream change through vowel reduction. The first error was more common in Grade 2, while older learners more often deleted CC segments (Figure 19.3). The third error type of vowel reduction is particularly interesting, where the vowel is substituted with the inherent vowel; thus, the deletion of the initial consonant in preCh /pretʃ/ becomes /rɔtʃ/ rather than /retʃ/. It is plausible to consider this error type as a /CCV/ manipulation reverting to a /Ca/ akshara– syllable mapping rather than the more accurate /CV/ response.

40% 32.4% 24.0% 12.6%

16% 11.2%

7% 2.7% 2.6% CCV

CC

5.3%

Vowel reduction Types of errors

TD-G2

TD-G3-4

PR

Figure 19.3. Phoneme deletion errors in CCVC non-words.

2.1% 1.3% No response

Akshara–syllable mappings in Bengali

211

Conclusion In this chapter we have given a descriptive analysis of the nature of associations between the orthographic syllable (the akshara) and the phonological syllable. We have looked at the relative difficulty in decoding when akshara– syllable mismatches occur and the ways in which akshara–syllable mappings differentially impact syllable and phoneme processing. Some of our key findings are as follows:  In syllable and phoneme processing, both less and more skilled readers show a phonological rather than an akshara-by-akshara effect.  In the processing of ambiguous akshara–syllable mappings, less skilled readers are more seriously impacted. They show difficulty at the phonological as well as the orthographic level.  Potentially influential factors in the processing of akshara–syllable mappings include knowledge of phonotactic constraints of the language, use of phonological analogies and frequency effects. These findings are a useful first step (i) for designing experiments to study specific cognitive processes when akshara-syllable mappings differ and (ii) for gaining a greater understanding of the role of orthographic and sublexical processing in Bengali reading. We began this chapter with the thesis that linguistic and orthographic diversity serves as a natural laboratory for identifying the interactions between language-specific properties and the written form. The skills that are involved in learning akshara–syllable mappings, we think, are a good example of such unique interactions.

20

Diversity in bilingual children’s spelling skill development: the case of Singapore Susan Rickard Liow

Much less is known about spelling than reading although both skills are essential for academic achievement. Children learning to spell in English face a difficult task because the mappings between speech sounds (phonology) and letters (orthography) are less predictable than in other alphabetic systems (see Cook & Bassetti, 2005; Seymour, Aro & Erskine, 2003; Share, 2008). Proficient spelling development appears to depend on a balance between visual and phonological processing (Ehri, 1997; Seymour, 1997; Vellutino et al., 2007), a distinction referred to as lexical and sublexical processing in dual-route models of skilled reading and spelling (see Coltheart et al., 2001; Houghton & Zorzi, 2003). In dual-route and dual-foundation models, spelling is achieved by looking up the word’s orthography using the memory-based lexical route which is sensitive to the target word’s frequency and its orthographic neighbourhood, and/or by encoding the target word’s phonology using the rule-based sublexical route, which is sensitive to regularity in the mappings between phonological and orthographic units. The consensus from extensive work on unilingual English-speaking children’s literacy development is that sublexical processing ability, particularly phoneme awareness, is the best predictor of reading and spelling skills (e.g. Byrne, 1998; Caravolas, Hulme & Snowling, 2001; Rayner et al., 2001; Vellutino et al., 2004). This finding has important implications for normal classroom practices, self-teaching strategies (Shahar-Yames & Share, 2008) that involve applying sound–letter rules for new words, and for the design of intervention programmes (Hatcher, Hulme & Ellis, 1994). Nevertheless both a bilingual child’s early aural linguistic experience at home (see Caravolas & Bruck, 1993; Durgunoglu & Oney, 1999; Jalil & Rickard Liow, 2008; Verhoeven, 2000), and the degree of overlap between the orthography and phonology of their two languages (Bialystok, 2007; Durgunoglu & Hancin-Bhatt, 1993; Wang & Geva, 2003), could alter the balance between the lexical and sublexical processing. These additional factors raise an important question for teachers and researchers: Are there differences in the cognitive–linguistic processing of bilingual children that influence their reading and spelling acquisition in English? 212

Bilingual children’s spelling skill development

213

Singapore’s multilingual environment offers an opportunity to start tackling this complex question with a minimal risk of cultural or pedagogical confounds. This chapter summarises a series of empirical studies exploring the processes that contribute to spelling in English for three subtypes of bilingual children who were all following the same curriculum. Early spellings are speech-based (Jalil & Rickard Liow, 2008), and the three main languages (English, Mandarin and Malay) are dissimilar in terms of phonology as well as orthography. These unusual setting conditions make it especially likely that there would be differences between bilingual subtypes in phonological awareness, as well as subsequent spelling development. Nonetheless each of the studies should be of interest to those working with underachieving language-minority children elsewhere (see Han (2012) for a recent review). Contrasting home languages Approximately half the families in Singapore use English at home and it is the main medium of instruction in government kindergartens and schools. Children are also expected to learn to read and write in a second language, typically the one spoken at home. For the purposes of research, the mix of home and school languages results in three types of bilingual (English-L1/ Mandarin-L2, Mandarin-L1/English-L2 and Malay-L1/English-L2), groups which are determined by using measures of receptive vocabulary in both languages and information provided by parents. Spoken Mandarin is morpho-syllabic with few consonant clusters and only one consonant coda (see Hua, 2002). English and Mandarin have almost the same number of consonants, but only nine of them occur in both languages, and the differences between their orthographies are even more extreme. Young children must learn even high-frequency Chinese characters by memory rather than rule because most are opaque with respect to their pronunciation (Rickard Liow, Tng & Lee, 1999). Spoken Malay has much simpler syllable structures than English and is highly inflected. The majority of stems are at least bisyllabic (C1V1C2V2C3: see Prentice, 1987) and there are very few consonant clusters or dipthongs. Malay (Rumi) is one of the most transparent alphabetic orthographies with almost one-to-one correspondence between phonemes and graphemes (see Yap et al., 2010). Except for , and , which are found only in foreign loan words, Malay words comprise the same consonants as English and many of the same phoneme–grapheme correspondences apply. Thus, exposure to Malay makes it more likely that children learn to apply rule-based sublexical skills when learning to spell English words (see Durgunoglu (2002) on cross-linguistic transfer). Research on cross-linguistic application of phonological awareness has been mostly concerned with writing systems (see Bialystok, 2007) but it is

214

Susan Rickard Liow

becoming clear that differences in the spoken language do influence spelling development. For example, children who learn Greek (Harris & Giannouli, 1999) or Turkish (Durgunoglu & Oney, 1999) develop syllable awareness earlier than children who speak English (Liberman et al., 1974). This is probably because their home languages have relatively simple syllable structures and well-marked syllable boundaries in speech (see also Cheung et al., 2001; McBride-Chang & Ho, 2005). This finding has implications for children in any bilingual setting where the phonological structures of the two languages are different. In Singapore, the contrasts are especially marked and they make it likely that the Mandarin-L1 and Malay-L1 kindergarten children will develop strong syllable awareness before they are exposed to print, but also that they might have poorer phoneme awareness compared to their English-L1 classmates. In other words, differences in the phonological structure of spoken English, Mandarin and Malay are likely to influence the development of phonological awareness in the young bilingual children, and this could have consequences for subsequent literacy development. Development of phonological awareness in kindergarten Yeong and Rickard Liow (2012) examined the development of English syllable and phoneme awareness in English-L1/Mandarin-L2 and Mandarin-L1/ English-L2 4- to 5-year-old pre-readers using a longitudinal design with ethnic Chinese children (n ¼ 35 English-L1; n ¼ 35 Mandarin-L1) over three 6-month intervals. The curriculum in government-aided kindergartens in Singapore comprises 3 hours of instruction in English and 1 hour in the mother tongue (Mandarin) each day regardless of the child’s home language. Sessions typically involve storytelling and basic numeracy skills plus some teaching of reading and spelling letter names and high-frequency words using a look–say or look–write lexical strategy. Neither Hanyu Pinyin (Romanised phonemic scheme for Chinese), nor English letter sounds (phonics approach), are used systematically by kindergarten teachers. The contrast between the phonological structures of English and Mandarin was expected to result in qualitative as well as quantitative differences in the development of phonological awareness between the English-L1 and the Mandarin-L1 children. Those from predominantly English-speaking homes would have been exposed to more clusters of phonemes within syllables, whereas those from predominantly Mandarin-speaking homes would have been exposed to simple CV structures with few consonant clusters. There is a progression in sensitivity from larger to smaller units with age (Anthony & Francis, 2005; Carroll et al., 2003): syllable awareness is usually developed before the age of 4 years in English-speaking children, whereas phoneme awareness is often delayed until children have started to read and write

Bilingual children’s spelling skill development

215

(Ziegler & Goswami, 2005). To examine the predicted dissociations in bilingual children’s phonological awareness across languages, equivalent syllable and phoneme deletion tasks were developed and administered in English and Mandarin. The syllable awareness task comprised multimorphemic words from both languages because Mandarin is morpho-syllabic (e.g. ‘say sunset without saying sun’), and the phoneme awareness task involved isolation and deletion of the first phoneme in a single CVC unit for consonants common to English and Mandarin (e.g. ‘what is the first sound in mess?’ then ‘say mess without saying the /m/ sound’). For both languages, the children performed better on syllable awareness than phoneme awareness regardless of language background or language tested. However, there was evidence of time-course disparities between the groups that are consistent with differences in the children’s early linguistic experience at home. In English, the English-L1 children’s performance was better in phoneme awareness at all three time-points, and the Mandarin-L1 children’s syllable awareness only became equivalent by Time 3. In Mandarin, the English-L1 children’s phoneme awareness, but not their syllable awareness, was also significantly better at all three time-points. Importantly, the cross-lagged correlations suggested that the Mandarin-L1 children required more exposure to oral English before they could develop phoneme awareness in either language. These results are consistent with previous work showing children from China are poorer than English-speaking unilinguals on a phoneme-onset deletion task (McBride-Chang et al., 2004). From the Singapore data it appears that English-L1 children retain an advantage in phoneme awareness for at least a year in kindergarten. Of course this advantage is likely to be reinforced by their continued greater exposure to the more complex syllables of English, and it should therefore have an impact on their early spelling skills. Early spelling skills in kindergarten Rickard Liow and Lau (2006) explored whether the phonological representations of Mandarin-L1 children are delayed by extending Treiman, Cassar and Zukowski’s (1994) cloze-procedure flaps task (e.g. wa_er, is it a letter or a letter in the word water?). The performance of 6-year-olds is interesting because they often try to spell words using phonologically plausible combinations of letters and letter names (see Treiman, 1993) before being taught conventional spellings. For inexperienced readers, spoken language is the main resource for encoding words into writing, and so phonological awareness will depend on the features of the home language. Treiman et al. (1994) reasoned that English-speaking children who can make use of phonological processing will (incorrectly) choose the letter rather than

216

Susan Rickard Liow

for flapped words. Consistent with predicted differences in metalinguistic awareness for bilinguals, the knowledge used by the Singaporean children depends on the language they speak at home: Malay-L1 children (n ¼ 29) children showed the greatest use of sublexical–phonological processing, the Mandarin-L1 children (n ¼ 21) showed the greatest use of lexical– orthographic processing, and the English-L1 children (n ¼ 29) appeared to use both routes. To better understand why Mandarin-L1 children’s early aural/oral exposure might make them less well equipped for sublexical–phonological processing, Yeong and Rickard Liow (2010) focussed on the underlying phonological representations that influence early spelling skills. The term phonological representation refers to the long-term store of the sounds in a particular language that support reading and spelling development (Perfetti, 1992) as well as vocabulary acquisition (Baddeley, Gathercole & Papagno, 1998). The study compared the performance of 5- to 6-year-old Mandarin-L1/English-L2 children (n ¼ 23) with that of English-L1/Mandarin-L2 (n ¼ 27) children on a 60-item single-letter cloze spelling task. The spelling list comprised high- and low-frequency English words with word-initial and word-final target phonemes that were either common to Mandarin and English (/f/ and /p/), or found only in English (/v/ and /b/). Dictated targets included ‘view’ (high frequency English-only, initial position and ‘nerve’ (low frequency Englishonly, final position). Even though spelling accuracy on the control common phoneme /m/ was matched, the Mandarin-L1 children’s performance for /v/ and /b/ was significantly poorer than that of their English-L1 classmates on low-frequency words, that is, Mandarin-L1 children had insufficient exposure to the English phonemes that are not present in their home language. This finding is also consistent with previous work and explains, in part, why Chinese-L1/ English-L2 learners’ phonological awareness for English remains less well developed than that of their unilingual age peers (see also Leong et al., 2005; Wang & Geva, 2003). For teachers, the Singapore data may be cause for concern given that most of the work on unilingual English-speaking children strongly emphasises the importance of phoneme awareness for normal reading and spelling development (e.g., Byrne, 1998; Caravolas et al., 2001; McBride-Chang, 1995; Rayner et al., 2001). The logical conclusion is that some second-language learners are obliged to learn to spell without the full repertoire of phonological representations and phoneme awareness. As noted earlier, the phonics approach to teaching literacy skills is not routinely employed in Singapore kindergartens, so it was not clear whether (self-taught) phonological awareness would be the best predictor of spelling in either the English-L1 or the Mandarin-L1 bilingual children. To try and unravel the influence of teaching methods from the underlying

Bilingual children’s spelling skill development

217

cognitive–linguistic processes arising from home language experience, Yeong and Rickard Liow (2011) then conducted a longitudinal design comparing English-L1 (n ¼ 50) and Mandarin-L1 (n ¼ 50) children’s spelling of whole words over a critical 6-month phase. Again, the children were 5- to 6-year-olds who were attending the same kindergarten classes and both language background groups were tested on parallel versions of English and Mandarin tasks as predictors at Time 1, and their spelling scores were then analysed using an experimental task administered at Time 2. The main predictors were chosen with reference to previous work with young children and included letter knowledge (e.g. Caravolas, 2004), rapid automatised naming (RAN digits: see Jongejan, Verhoeven & Siegel, 2007), and verbal short-term and working memory (see Caravolas et al., 2001; Ouellette & Se´ne´chal, 2008), as well as syllable and phoneme awareness (Ziegler & Goswami, 2005). The main dependent variable was the child’s spelling sophistication score which was computed by examining the proximity of individual spelling responses to the corresponding target word in terms of total number of phonemes represented, level of orthographic representation, whether the salient letter is represented, and the presence of intrusion errors. Spelling sophistication scores are a sensitive means of exploring underlying processing especially in young children when the likelihood of an error is high (see Tangel & Blachman, 1992; Treiman & Bourassa, 2000). Yeong and Rickard Liow’s (2011) first aim was to compare the phonological knowledge of English-L1 children with that of Mandarin-L1 children in both languages, as well as English letter knowledge, and the second aim was to explore the contributions of the English and Mandarin predictors to spelling development for the two groups. Given the marked contrast in the two groups’ linguistic experience, different Time 1 predictors were expected to contribute to spelling sophistication at Time 2. The results showed that performance on the tasks of syllable and phoneme awareness depended on children’s first language and the language of the task: Mandarin-L1 children were poorer on phoneme awareness than their English-L1 classmates (see also McBride-Chang et al., 2004), and they showed slower rates of rapid naming. Both groups also apply language-specific knowledge during verbal short-term memory STM tasks more effectively in their respective L1 (see Adams & Gathercole, 2000). However, after controlling for non-verbal IQ, age, vocabulary and Wide Range Achievement Test 4 (WRAT 4) (Wilkinson & Robertson, 2006) spelling accuracy at Time 1, regression analyses showed that phoneme awareness was the strongest predictor of spelling sophistication for Singaporean English-L1 children, as it is for unilingual English-speaking children (e.g. Caravolas et al., 2001). Phoneme awareness was also a significant predictor of spelling sophistication for the Mandarin-L1 children, but for this group both syllable awareness and letter-sound knowledge also accounted

218

Susan Rickard Liow

for unique variance. Again, the results provide evidence of processing differences among bilinguals but suggest that phonological representations and rulebased sublexical knowledge are important for early spelling in English. This conclusion, however, raised the question of whether subtype differences are transient or have a long-lasting impact on spelling proficiency in primary school. Spelling development in primary school For bilingual children, the balance between visual–lexical and phonological– sublexical foundations for spelling in English (Seymour, 1997) depend on the transparency of the target orthography (see Seymour et al. (2003), on reading), the extent of any cross-linguistic transfer during processing, and the child’s oral language experience at home (Rickard Liow & Lau, 2006; Treiman et al., 1997; Yeong & Rickard Liow, 2011). Research on Spanish– English bilinguals living in the United States (e.g. Rolla San Francisco et al., 2006) shows that English spelling errors are influenced by the predictable phonology–orthography mappings in Spanish. Thus a greater reliance on the phonological–sublexical foundation should be apparent in Singaporean Malay-L1 children because Rumi is even more transparent than Spanish (Yap et al., 2010). In contrast, exposure to languages with much less predictable mappings than English, such as Mandarin, should enhance visual–lexical processing in ESL bilinguals. If relatively limited aural/oral experience in English delays the development of phoneme awareness in Singaporean Mandarin-L1 children (Yeong & Rickard Liow, 2011), good visual–lexical skills might compensate (see McBride-Chang et al. (2011) on visual skills). Three experiments have been directed at predictable differences in the balance between the visual–lexical and phonological–sublexical foundations for spelling in bilingual primary school children. First, Rickard Liow and Poon (1998) compared the performance of three subtypes (n ¼ 57, 8- to 9-year-old Mandarin-L1, English-L1 and Indonesian-L1 children) on homophone decision tasks as well as spelling accuracy for English words and non-words. Like Malay, Indonesian has transparent phoneme-grapheme mappings. As expected, the Indonesian-ESL children had developed better sublexical processing abilities and performed better on the homophone task and non-word spelling than students from either English-speaking or Mandarin-speaking language backgrounds. Second, Rickard Liow and Tng’s (2003) work with 8- to 9-year-old children showed that English-L1 children were significantly better than their Mandarin-L1 classmates at non-word spelling as well as English word reading. Interestingly, scores on these two tasks were highly correlated for English-L1 children (n ¼ 24, rho ¼ 0.614, p