Crosslinguistic Influence in Multilinguals: An Examination of Chinese-English-French Speakers 9781783096893

This book connects the topics of third language acquisition, reverse transfer and the Interface Hypothesis. It reports o

160 28 715KB

English Pages [152] Year 2016

Report DMCA / Copyright

DOWNLOAD FILE

Polecaj historie

Crosslinguistic Influence in Multilinguals: An Examination of Chinese-English-French Speakers
 9781783096893

Table of contents :
Contents
Acknowledgements
1. Overview
2. Design of the L3 French – L2 English Project
3. Target Structures of the L3 French – L2 English Project
4. Nominal Plural Marking
5. Past Tense Marking
6. Adverb Placement
7. Conclusions: The Way Forward
References
Index

Citation preview

Crosslinguistic Influence in Multilinguals

SECOND LANGUAGE ACQUISITION Series Editors: Professor David Singleton, University of Pannonia, Hungary and Fellow Emeritus, Trinity College, Dublin, Ireland and Dr Simone E. ­Pfenninger, University of Salzburg, Austria This series brings together titles dealing with a variety of aspects of language acquisition and processing in situations where a language or languages other than the native language is involved. Second language is thus interpreted in its broadest possible sense. The volumes included in the series all offer in their different ways, on the one hand, exposition and discussion of empirical findings and, on the other, some degree of theoretical reflection. In this latter connection, no particular theoretical stance is privileged in the series; nor is any relevant perspective – sociolinguistic, psycholinguistic, neurolinguistic, etc. – deemed out of place. The intended readership of the series includes final-year undergraduates working on second language acquisition projects, postgraduate students involved in second language acquisition research, and researchers, teachers and policy-makers in general whose interests include a second language acquisition component. Full details of all the books in this series and of all our other publications can be found on http://www.multilingual-matters.com, or by writing to Multilingual Matters, St Nicholas House, 31–34 High Street, Bristol BS1 2AW, UK.

SECOND LANGUAGE ACQUISITION: 108

Crosslinguistic Influence in Multilinguals An Examination of Chinese–English–French Speakers

Wai Lan Tsang

MULTILINGUAL MATTERS Bristol • Blue Ridge Summit

Dedicated to my much-missed parents, TSANG Tin Kam & KAM Yuk Ching 獻給我永遠懷念的父母親

Library of Congress Cataloging in Publication Data A catalog record for this book is available from the Library of Congress. Names: Tsang, Wai Lan, author. Title: Crosslinguistic Influence in Multilinguals: An Examination of Chinese-EnglishFrench Speakers / Wai Lan Tsang. Description: Bristol: Multilingual Matters, [2016] | Series: Second Language Acquisition: 108 | Includes bibliographical references and index. Identifiers: LCCN 2016031378| ISBN 9781783096886 (hbk : alk. paper) | ISBN 9781783096909 (epub) | ISBN 9781783096916 (kindle) Subjects: LCSH: Multilingualism – China – Hong Kong – Cross-cultural studies. | Languages in contact – China – Hong Kong – Cross-cultural studies. | Language acquisition – China – Hong Kong – Cross-cultural studies. | Intercultural communication – China – Hong Kong – Cross-cultural studies. | Discourse analysis – Social aspects – China. Classification: LCC P115.5.C6 T826 2016 | DDC 404/.2095125 – dc23 LC record available at https://lccn.loc.gov/2016031378 British Library Cataloguing in Publication Data A catalogue entry for this book is available from the British Library. ISBN-13: 978-1-78309-688-6 (hbk) Multilingual Matters UK: St Nicholas House, 31–34 High Street, Bristol BS1 2AW, UK. USA: NBN, Blue Ridge Summit, PA, USA. Website: www.multilingual-matters.com Twitter: Multi_Ling_Mat Facebook: https://www.facebook.com/multilingualmatters Blog: www.channelviewpublications.wordpress.com Copyright © 2017 Wai Lan Tsang. All rights reserved. No part of this work may be reproduced in any form or by any means without permission in writing from the publisher. The policy of Multilingual Matters/Channel View Publications is to use papers that are natural, renewable and recyclable products, made from wood grown in sustainable forests. In the manufacturing process of our books, and to further support our policy, preference is given to printers that have FSC and PEFC Chain of Custody certification. The FSC and/or PEFC logos will appear on those books where full certification has been granted to the printer concerned. Typeset by R. J. Footring Ltd, Derby Printed and bound in the UK by the CPI Books Group Ltd. Printed and bound in the US by Edwards Brothers Malloy, Inc.

Contents

Acknowledgements

vii

1 Overview 1 Focus of the Book 2 Multilingual Hong Kong 2 Third Language Acquisition 4 Second Language Acquisition from the Generative Perspective16 Reverse Transfer in L2A and L3A 19 The Interface Hypothesis in L2A and L3A 23 The Present Study 32 Structure of the Book 33 2

Design of the L3 French – L2 English Project Research Question and Hypotheses Setting of the L3 French – L2 English Project Design of the L3 French – L2 English Project Test Structures Methodological Limitations

35 35 37 39 44 46

3

Target Structures of the L3 French – L2 English Project Nominal Plural Marking in English, Cantonese and French Past Tense Marking in English, Cantonese and French Adverb Placement in English, Cantonese and French Concluding Remarks

48 48 50 51 52

4

Nominal Plural Marking Nominal Plural Marking in English, Cantonese and French Number Agreement and L2A

53 54 54

v

vi Contents

L2 Acquisition of English Nominal Plural Marking by Chinese Learners of English L2 Acquisition of English Nominal Marking by L1 Cantonese – L2 English – L3 French Speakers Discussion: L3 French – L2 English Nominal Plural Marking Implications for Future Research 5

6

7

Past Tense Marking Past Tense Marking in English, Cantonese and French Past Tense Marking and L2A L2 Acquisition of English Past Tense-Aspect Morphology by Chinese Learners of English L2 Acquisition of English Past Tense Marking by L1 Cantonese – L2 English – L3 French Speakers Discussion: L3 French – L2 English Past Tense Marking Summary and Implications for Future Research

56 59 65 72 74 75 77 80 82 94 104

Adverb Placement Adverb Placement in English, Cantonese and French Theoretical Accounts of Adverb Placement Adverb Placement and L2A L2 Acquisition of English Adverb Placement by Chinese Learners of English L2 Acquisition of English Adverb Placement by L1 Cantonese – L2 English – L3 French Speakers Discussion: L3 French – L2 English Adverb Placement Implications for Future Research

106 107 107 110

Conclusions: The Way Forward Summary of the Major Findings Potential Forces Behind Possible Interaction Between L3 and L2 Vulnerability of the Internal Interfaces Pedagogical Implications Areas for Further Research

123 123

110 112 115 121

125 126 128 129

References

130

Index

142

Acknowledgements

The book, based on a study funded by the Research Grants Council of the Government of the Hong Kong Special Administrative Region (RCC-GRFGrant HKU 755111), is the result of the invaluable advice, encouragement and assistance from numerous people I would like to acknowledge here. First, I would like to express my gratitude to the series editors, David Singleton and Simone Pfenninger, for their constructive comments on my book, which have made it a better one. My thanks are also due to the anonymous reviewer for his/her useful and insightful suggestions for all the changes and revisions. I am also thankful to Laura Longworth, Sarah Williams and their colleagues at Multilingual Matters for their professional support and great patience throughout the long process of writing the book. My heartfelt thanks go to two important scholars and mentors on the path of my academic pursuits, who have also kindly written the blurbs for this book. First, I am immensely grateful to Professor Ianthi Tsimpli for guiding me through acquisitional research with her inspirational and astute discussions during my doctoral study. I was always and am still fascinated by the ideas, observations and arguments in her research. I am also much obliged to Professor Stephen Matthews for introducing me to the field of Cantonese/Chinese linguistic and acquisitional research when I was still a research student. He also kindly gave me chances to co-supervise his students who worked on third language acquisition, when I was in a capacity to do that, enabling me to explore this exciting field further. I am also appreciative of his reading a very early draft of a chapter and giving me shrewd comments on various theoretical and acquisitional issues as well as showing me the way to structure a proper book chapter. That helped me much in compiling the other parts of the book. My earnest acknowledgement also goes to Professor Jasone Cenoz for her generous and precious help in writing a blurb for this book amid her busy schedule. vii

viii Acknowledgements

Special thanks go to Dr Henriette Hendriks and Dr Boping Yuan. With their intellectual discussions and pleasant chats, my time at Cambridge as a student and later as a Visiting Scholar was made much more enriching. I would also like to gratefully thank Dr Denis Meyer, who kindly proofread the French versions of two of the experimental tasks and shared with me the learning problems among his French students. My sincere thanks are extended to Professor Agnes Lam for her constant support and encouragement, always making me stronger and more determined. A note of thanks also goes to colleagues at the Centre for Applied English Studies, some of whom kindly helped proofread the English versions of two of the experimental tasks, spread the news of the experiment and recommended potential research assistants to me. Last, I wish to thank Dr Ingrid Leung for stimulating me with her research in third language acquisition and our cheerful professional and personal chats. I would like to recognise the following research assistants and student research assistants for their valuable assistance at various stages of com­ pletion of the book: Rachel Kan, Anna Cheung, Bryant Hui, Janet Zhang and Amy Pao. It was a great pleasure working with them. I am also grateful to a countless array of students at the University of Hong Kong and at the University of Cambridge who participated in the project at different phases in the two universities. My final gratitude goes to my family. I am much indebted to my much-missed parents, TSANG Tin Kam and KAM Yuk Ching, for their unconditional love and sacrifice. They have given me an encouraging and loving family and supported me in all my pursuits. I also owe many thanks to my dear siblings, Shiu Kei, Shiu Kai and Wai Yin, for lending their ears and time to me whenever I was soliciting their linguistic judgements or statistical advice. Without the selfless support and unwavering encouragement from my family, I would never have finished this project.

1 Overview

Global citizenship in the 21st century has entailed one remarkable and widespread phenomenon: multilingualism. As the prefix ‘multi-’ denotes, the world is moving beyond the stage of bilingualism, the state where people speak at most two languages, to the phase in which people speak not only a native language plus a second/foreign language but also some other second/foreign languages. This transition from bilingualism to multi­ lingualism means that people are speaking three or more languages, which enables them to communicate with people from other parts of the world and enhances ‘global’ mobility. How common this multilingual phenomenon is can be illustrated by census statistics in different countries or regions. Hong Kong, being an international hub in Asia, is of no exception in the globally multilingual community. It has been reported that many Hong Kong people are multilingual (a point to return to later in this chapter). The ability of many Hong Kong people to speak at least three languages may be closely related to the multilingual environment of the city. In Hong Kong, English and Chinese are the two official languages which are used in different formal contexts, such as government, law and education. Chinese in turn embraces Standard Written Chinese and Cantonese. Standard Written Chinese is used in written communication, and Cantonese, a variety of Chinese, serves as the vernacular language. This variety of Chinese does share some similarities with the standard Chinese variety (i.e. the variety used as the national language of China, which is commonly known as Putonghua), especially in the realm of lexis and syntax, but the two varieties differ quite a lot in phonology. Besides Cantonese, some other Chinese dialects (e.g. Shanghainese, Chiu Chow and Hakkanese) are also commonly spoken among the immigrants from mainland China. It is in this multilingual context that the project reported in this book was conducted on a group of local multi­linguals with knowledge of Cantonese, English and French. 1

2  Crosslinguistic Influence in Multilinguals

Focus of the Book The research project reported in this book delves into one aspect of language acquisition in multilingual Hong Kong, where Cantonese is the mother tongue (L1) of the majority of the population and where the learning of foreign languages is commonplace. In addition to English, which is usually the second language (L2), largely for historical and economic reasons, more and more people learn other languages, such as French, Spanish, Korean and Japanese. It is therefore easy to come across people in the city with knowledge of at least three languages, which serves as a good venue for research on second language acquisition (L2A) and a related and emerging field, third language acquisition (L3A). The study was based on a group of L1 Cantonese – L2 English – L3 French (L1C–L2E–L3F) multilinguals in Hong Kong. They were advanced or nearnative English speakers and were asked to work on a number of language tasks covering three English structural patterns: (1) nominal plural marking (e.g. He has three carrots), (2) past tense marking (e.g. He mashed the carrots) and (3) adverb placement (e.g. He always buys carrots). The performance of the participants on the three patterns were interpreted in terms of the notion of ‘reverse transfer’ (influence from a later acquired language back to an earlier acquired one, i.e. from L3 French to L2 English in this case), and were explained in terms of both the key concerns in L3A and the tenets of the interface hypothesis proposed by Antonella Sorace (2005, 2011). In the above brief description of the study, three core notions require more detailed discussion before we look at the experimental results of the study: L3A; reverse transfer; and the interface hypothesis. This introductory chapter will discuss these three kernels of the study, which serve as the bases for the specific discussions in the later chapters. Given the close connection between L2A and L3A, a review of the tenets of L2A from the generative perspective is also included. This discussion of the four background areas will be preceded by a mention of the linguistic situation of multilingual Hong Kong in this chapter and followed by a description of the structures and the methodological design of the present project in the next two chapters.

Multilingual Hong Kong A review of the literature on the linguistic status of Hong Kong revealed two contrary outcomes. On the one hand, despite the advocacy of biliteracy and tri­lingualism by the government, Hong Kong has been depicted in some studies as a largely monolingual society (e.g. Ho & Ho, 2004), with Cantonese being the main medium of communication used by the majority of people in different speech contexts, ranging from daily conversations to spoken government communication (e.g. discussions of government policies

Overview 3

at the Legislative Council) (Bolton, 2011; Li, 2009). The use of English, one of the two official languages, is mostly confined to written communication, education and the workplace, as in the government, newspaper agencies and the legal system (Bolton, 2011). In the educational scenario, since the handover of Hong Kong by Britain back to China in 1997, the use of English as the medium of instruction (MOI) has shrunk from all local schools to 114 schools only, though there has been a recent increase in the use of English as the MOI for different subjects, dependent on the overall English level of the students. Among the general public, the frequency of use of English varies, depending largely on people’s educational level and proficiency in the language: the higher the educational level and the higher the English proficiency, the higher is the frequency of use of the language. All the govern­ ment policies or practices, together with the proficiency in and limited use of English among the people at large, have to a large extent, resulted in the image of a monolingual Hong Kong. Some other studies, on the other hand, describe Hong Kong as a multi­lingual society (e.g. Kirkpatrick, 2007; Li, 2009; Trent et al., 2014). This multilingual status is reflected in the codeswitching and code­mixing behaviour of the Hong Kong people in their daily lives (e.g. Ho, 2007) or even in the government or classroom context (e.g. Li, 2008; Lin, 1990, 2008; Luke, 1998; Yau, 1997). Such codeswitching/codemixing seems to be in­ dependent of one’s proficiency in English, probably due to the widespread use of English among the people, reinforced by the dialogues/conversations in different programmes in the media (e.g. Leung, 2010). The use of simple words or formulaic expressions such as ‘okay’ and ‘hello’ can be heard in daily conversations from time to time. Besides the reports in academic research articles of the use of both Cantonese and English, feature articles in local news­papers and government statistics also lend support to the multilingual nature of the city. For example, in a feature article on an interview with a faculty dean at a local university (Tsui, 2015), students in Hong Kong were described as trilingual. Also, according to the census statistics in 2011, the majority of people in Hong Kong report that they can speak two or more languages (Table 1.1), with the percentage ranging from 29.30% to 34.31%, and the percentage of speakers who spoke four languages dropped to 6.02%. Table 1.1  Hong Kong population aged five years and over by number of languages spoken (2011 census statistics) Number of languages spoken

Percentage of the population (n)

1

30.38% (2,068,451)

2

29.30% (1,994,558)

3

34.31% (2,335,814)

4

  6.02% (409,610)

4  Crosslinguistic Influence in Multilinguals

The languages recorded in the census were principally Cantonese, English and Putonghua. Other Chinese dialects could also be among those languages, especially for immigrants from mainland China. Meanwhile, other world languages, such as French, Korean, Japanese, Italian and Arabic, have gained popularity in the territory, as evidenced by the running of language courses at local schools and universities and private educational institutes. This in turn strongly suggests the possibility of these foreign languages being people’s third or fourth languages – see, for example, Humphrey and Spratt (2008) for a discussion of the motivation of university students to learn different languages. Whether Hong Kong is a monolingual or a multilingual society turns out to be more complicated than it seems and there has been research upholding one status or the other. Meanwhile, as Bolton remarks (2002), Hong Kong seems to be holding the two identities at the same time. Should Myers-Scotton’s (2006: 44) definition of multilingualism be considered – ‘the ability to use two or more languages sufficiently to carry on a limited casual conversation’ – and should different dialects be regarded as distinct linguistic systems, Hong Kong could well be labelled a multilingual society.

Third Language Acquisition Third language acquisition (L3A), a field closely connected to L2A, does not necessarily refer to the language in which the person is third most proficient (Hammarberg, 2001, 2009). The following quote from Cenoz and Jessner (2000) can explain why (with emphasis added): In talk[ing] about L1 and L2 it is implicitly assumed that L1 is the dominant language and that the level of proficiency in L2 must necessarily be lower than in L1. When a third language is acquired, however, the chronological order in which the three languages have been learnt does not necessarily correspond to the frequency of use by or level of competence in the trilingual speaker. (p. x) As the quote implies, the sequence of acquisition does not suffice in identifying the ‘third’ language. Instead, two other variables – frequency of use and proficiency – should be considered as well. This also corresponds to the non-linear definition of L3A proposed by Williams and Hammarberg (2009), who state that L3A represents ‘a cognitive hierarchy between the languages for the user in the current situation’ rather than a ‘language-by-language chronology’ (p. 7). Since the order of acquisition is too simplistic in defining L3 (and L3A), Cenoz (2000: 40) lists the different combinations of L1, L2 and L3 (or even more):

Overview 5

• • • •

after L2 (L1 à L2 à L3) simultaneous acquisition of two languages (L1 à Lx/Ly; Lx/Ly à L3) simultaneous acquisition of three languages (Lx/Ly/Lz) L1 à L2 à L3 à L4.

Given the above possible combinations, it is not surprising to find De Angelis’s (2007) definition of L3, which helps distinguish L3A from L2A: ‘a third or additional language … regardless of whether it is a third, fourth or sixth language’ (p. 10). Similarly, Hammarberg (2010, 2014) advocates the three-tier definition of multilingualism (i.e. L1, L2 and L3). Citing Hufeisen (1998), Hammarberg notes ‘a distinct qualitative difference between the conditions for acquiring the first, second and third language’ (2010: 95), but no similar distinction has been noted when the fourth, fifth and further languages have been studied. Therefore, what is of prime importance is distinguish­ing L3 (indeed, Ln) and L2, and such a distinction has been adopted in different studies (e.g. Berkes & Flynn, 2013; Cheung et al., 2011; Rast, 2010; Rothman & Halloran, 2013; Wunder, 2011; the studies in the 2010 special issue of the International Journal of Multilingualism on L3 phonology). Following this research practice and considering the general multilingual situation in Hong Kong, the present study also defined L3 as the language(s) being acquired after L1 and L2 and as a notion embracing L3, L4 and so forth. Apart from the definition of L3, the study of L3A has been developing principally in two research directions. One centres on various factors behind L3A, and the other hinges on the models or approaches that help explain the process of L3A. These two directions therefore served as the basis for the present L3A study, which investigated whether a later acquired language (L3) would affect an earlier acquired one (L2) (i.e. reverse transfer from L3 to L2). Of particular concern was what structural forces (e.g. typological proximity) and non-linguistic factors (e.g. nature and amount of input) might bring about such reverse transfer, as exemplified by three structures in English. As we shall see, particular factors account for various patterns revealed by the findings of the three structures. Those L3A factors relevant to the present study were also sequenced to show when and how they might result in potential reverse transfer from L3 to L2 (see Chapter 7, where the applicability of the L3A models and approaches to the possible interaction between the two languages is also discussed).

Factors behind L3A Cenoz (2013), while acknowledging the distinctive nature of L3A, points out the possible similarities between L2A and L3A. As a field closely related to (and probably evolved from) L2A, L3A looks into crosslinguistic influences in terms of not only core factors such as age and motivation (Cenoz

6  Crosslinguistic Influence in Multilinguals

& Gorter, 2011a) but also those which have been assumed to be specific to L3A. The review below will first cover four specific factors which were more commonly discussed and then the L2A factors being studied in the L3 contexts. (For a list of other factors specific to L3A, see De Angelis, 2007.) The first factor commonly identified in L3A studies is typology or typological proximity, which refers to ‘the distance that a linguist can objectively and formally define and identify between languages and language families’ (De Angelis, 2007: 22). Cenoz (2001) highlights the potential role of typology in her study on speakers whose L1 is Basque and/or Spanish and whose L3 is English. Through a story-telling task, Cenoz examined their interactional strategies and codeswitching, and identified possible traces of transfer from L1 to L3. L1 Spanish was found to be the source of function words in L3 English, probably because Spanish is ‘closer’ to English than Basque. Similar to Cenoz, Williams and Hammarberg (1998) and ­Hammarberg (2001) also point out the role of typology in their studies based on their two-year longi­tudinal data from an L1 English – L2 German – L3 Swedish speaker. Both of those studies looked into language switches, word search and word construction, morphology, as well as phonology and phonetic settings in the spoken data collected via interviews, discussions and narratives. Interestingly, the speaker’s L3 revealed two distinct roles associated with her L1 and L2: L1 as the ‘external instrumental language’, facilitating the production and acquisition of L3 pragmatics; and L2 as the ‘supplier language’, typo­ logic­ally closer to L3 and therefore enhancing the production and acquisition of new words and phonological patterns in L3. Bardel and Lindqvist (2007) observed a similar role of typology in their introspective study of speakers with the language background L1 Swedish, L2 English/French/Spanish and L3 Italian. Four recordings, including guided conversations and story retelling, were collected, and lexical and phonological patterns in the recordings were examined. The recorded L3 patterns showed different degrees of activation of the previously learnt languages, largely dependent on typological distances of the languages involved. For example, the codeswitching instances in the recordings, which were highly affected by L2 Spanish, were argued to be caused by the similarity between Spanish and L3 Italian. In a subsequent study, Lindqvist and Bardel (2014) found further support for the significant role of typology with recordings collected from speakers with L1 Swedish/Italian, L2 English/French and L3 Spanish. Other than typology, a related factor has been put forward, namely psycho­typology – the similarities between and differences among languages as perceived by learners (Kellerman, 1983). Such perceived similarities and differences might have activated learners’ earlier languages in their acquisition of L3, as indicated in De Angelis and Selinker’s (2001) study. Spontaneous interview data were collected from two adult multi­linguals: an L1 French – L2 English – L3 Spanish – L4 Italian speaker and an L1 English – L2 Spanish – L3 Italian speaker. Their interlanguage production

Overview 7

revealed instances of transfer from L3 to L4 and L2 to L3, and such transfer was both lexical and morphological. Similarly, Bouvy (2000) also noted the potential role of psychotypology in L3A in her examination of an error corpus of oral production by L1 French – L2 German/Dutch – L3 English speakers. Psycho­typology was found to have influenced codemixing, direct borrowing and spelling in the corpus data. A study by Ringbom (2001) also lends support to the possible effect of psychotypology on the number of L1- or L2-driven lexical transfer errors in L3. Lastly, in analysing the questionnaire data collected from the participants of the present study, Tsang (2015a) accounted for the more detailed metalinguistic response from one group of participants in terms of their higher sensitivity to the similarities and differences among three test languages: L1 Cantonese – L2 English – L3 French (CEF). All these different studies thus help showcase how the learners’ perception of the previously and newly acquired languages may affect the process of acquisition.1 Besides (perceived) linguistic similarities or differences, proficiency in L2 or L3 is also considered to affect L3A. Jaensch (2009) studied L1 Japanese – L2 English – L3 German speakers by asking them to complete three tasks: a sentence completion task and two multiple choice tasks (dialogue and story). She observed that the higher the L2 English proficiency was, the more target-like performance was on three structures in L3 German: gender; gender and case; and gender, case and context. With higher proficiency in L2, the speakers developed more cognitive skills and heightened meta­ linguistic expertise, thereby becoming more sensitive to the new features in L3. Jaensch cautioned, though, that the effect of L2 proficiency might be applicable to certain but not all structures: the lower intermediate learners in her study did not show a clear effect of L2 proficiency on L3 for two of the test structures: gender; and gender and case. While the possible connection between L2 proficiency and L3 structural differences deserves further investigation, both L2 proficiency and L3 proficiency seem to matter too. In Cenoz’s (2001) study, as reviewed above, the finding of more instances of transfer among older learners was possibly due to their low L3 English proficiency. Interestingly, Bardel and Lindqvist (2007) and Lindqvist and Bardel (2014) also highlight the potential role of both L2 and L3 proficiency in conditioning codeswitching instances in L3 Italian and L3 Spanish. The last factor is L3 status, as discussed by Williams and H ­ ammarberg (1998), Hammarberg (2001) and De Angelis (2005). Williams and ­Hammarberg (1998) and Hammarberg (2001) maintain that the ‘division of labour’ between L1 and L2 in L3A (as reviewed above) is due to the ‘foreign’ status that the learners have assigned to their L2, leading to more activation of L2 in L3 (which is another ‘foreign’ language) than of L1. The influence of ‘status’ seems to be greater than that of typology, high proficiency and status. De Angelis (2005) also lists association of foreignness as one factor in her study of L1 French – L2 Spanish – L3 Italian and

8  Crosslinguistic Influence in Multilinguals

L1 English – L2 Spanish – L3 Italian speakers. Data from interviews, translation tasks and reflection in diary entries indicated some form of ‘system shift’, where L2 lexical items were regarded as part of L3 and the learners were not aware of the source of the items (i.e. L2). De Angelis explains that such transfer of lexical items could be driven by the learners’ assignment to any non-native language (no matter whether it is L2 or L3) the status of foreignness. In other words, similar to another L3A factor, psychotypology, learners’ judgement or perception of the languages is involved in the L2/ L3 status factor: psychotypology relates to the learners’ judgement of the similarities or differences between the languages concerned, and the L2/L3 status factor concerns the learners’ perception of the ‘foreignness’ of the languages.2 Besides the above four L3A-specific factors, two core L2A factors, which underpin two influential hypotheses – the critical period hypothesis and the input hypothesis – have also been examined in the L3 context: age; and the role of input or length of exposure to the target language.3 Noting that the age factor has not been much studied in L3A, Cenoz (2001) put age as one of the concerns in her L3A study of a group of 90 elementary and secondary students in the Basque country with Basque and/or Spanish as their L1 and English as their L3. The students fell into three age groups – grade 2, grade 6 and grade 9 – and were asked to tell a picture story in English. Their stories revealed differences between the younger group and the older groups. Interestingly, contrary to the expectation that the older students would be in a better position to perceive the similarities and differences between the languages they knew and therefore had a more native-like fluency, their younger grade 2 peers were found to be subject to less crosslinguistic influence in their L3 English. In other words, the younger learners of English made fewer instances of linguistic transfer from L1/L2 to their L3 and were more accurate in their productions. While the differences do imply a possible influence of the age factor in the acquisition of an L3, Cenoz accounted for the ‘unexpected’ outcome by suggesting that the ‘limited’ proficiency of English across all these comparatively young L3 learners of English might have affected them, and a longitudinal study would probably help unveil the nature of the age factor in L3A further. While the age factor has been examined as it is in L2A in the context of L3A, the other core L2A factor – input – has been interpreted in a more complex way in L3A research. Basically, input covers both naturalistic input and classroom input. In L2A, L1 is in general associated with naturalistic input and L2 with classroom input. Unlike L1 and L2, L3 can be related to both naturalistic and classroom input, depending on where, when and how it is acquired. As reviewed above, L3 can be simultaneously acquired with L1 and L2, possibly making it one of the speaker’s ‘first languages’. On the other hand, L3 may be acquired later than L1 and/or L2, and L3 may be a non-native language, which may be acquired in a naturalistic environment

Overview 9

(e.g. in the form of immersion in the place where the target language is used extensively) or an instructed context (i.e. in the classroom). Connecting L3 with the classroom context in turn carries a number of implications, such as metalinguistic awareness, as in Herdina and Jessner’s (2002) model (to be reviewed in the next sub-section), and potential synergy and ‘collaboration’ among the languages learners know and the medium of instruction in a classroom (see the review by Jessner, 2008). Slabakova and García Mayo (forthcoming) further draw our attention to the role of comprehensible linguistic input in L3A. In their study, two trilingual groups, one L1 Basque – L2 Spanish – L3 English and the other L1 Spanish – L2 Basque – L3 English, were tested on their judgement of topicalisation in context, null objects and generic/specific object pronouns in English. The responses from both groups in a grammaticality judgement task indicated that they did not accept or reject topicalised structures in a target manner, ‘were much more accurate’ in rejecting unacceptable null object patterns (p. 15) and rejected significantly more specific null objects than generic null objects. Their responses did not only help the researchers examine the expectations of four existing L3A models but also suggested two other possible forces behind making native judgement: negative evidence and frequency of use of the structure in the linguistic input. The significance of the two forces, according to Slabakova and García Mayo, may be particularly related to the non-native judgement of English topicalisation among the participants: negative evidence, which is not available in the linguistic input, is needed to help them reject the presence of resumptive pronouns in English topicalised patterns; and the low frequency of use of topicalisation in English makes the structure more challenging to learners. As far as the present study was concerned, L3 classroom input appeared to play a more determining role than did L3 naturalistic input, in that French is not widely used in Hong Kong and the CEF participants acquired or were exposed to French in a classroom context. Interestingly, the possible role of exposure to explicit L3 instruction in a classroom context was highlighted by Tsang (2015a), where the CEF participants of the present study were examined in terms of their responses in a biodata questionnaire. The questionnaire was used to examine the language learning repertoire of all the participants of the study, and one section required them to compare the languages they knew, for example Cantonese, English and French for the CEF participants, and Cantonese and English for the L1 Cantonese – L2 English (CE) participants. The CEF participants, in comparison with their CE peers, were found to be more likely to give more specific or more detailed comments on the similarities or differences between Cantonese and English. What is more, the two groups seemed to differ in terms of how they perceived the language distance between Cantonese and English. These differences might be attributed to the more enhanced crosslinguistic experience as enriched by their exposure to explicit instruction in their L3 French classes.

10  Crosslinguistic Influence in Multilinguals

Table 1.2 summarises the L2 and L3 factors as commonly discussed or cited in L3A studies. Table 1.2  L3 factors and some related sample studies L3A factors

Sample studies

Typology/typological proximity

Cenoz (2001); Williams and Hammarberg (1998); Hammarberg (2001); Bardel and Lindqvist (2007); Lindqvist and Bardel (2014)

Psychotypology

De Angelis and Selinker (2001); Bouvy (2000); Ringbom (2001)

L2/L3 proficiency

Jaensch (2009); Cenoz (2001); Bardel and Lindqvist (2007); Lindqvist and Bardel (2014); Rothman (2014)

Association of foreignness De Angelis (2005); Hammarberg (2001); Williams and of L2/L3 (or, in a broader Hammarberg (1998) sense, status of L2/L3) Age

Cenoz (2001); Herdina and Jessner (2002); Jessner (2006)

Input/exposure to L3

Cenoz (2001); De Angelis (2015); Rothman (2014); Tsang (2015a); Slabakova and García Mayo (in press)

As we shall see in the discussion chapters on the three test structures, some of the above factors appeared to have played a role in bringing about possible influences from L3 to L2 in one way or another. Depending on the nature of the test structures, some of the factors seemed to be more influential than others.

Models of L3A With various factors identified as potential driving forces behind L3A, recent advances in the field of L3A centre on different approaches explaining how L3 acquisition proceeds. The review below, which is not intended to be exhaustive, covers a selection of L3A models or approaches which have been influential and are relevant to our understanding of possible crosslinguistic influence in L3A: Flynn et al.’s Cumulative Enhancement Model (2004), Bardel and Falk’s L2 Status Factor Model (2007), Rothman’s Typological Primacy Model (2010), Slabakova’s Scalpel Model of third language acquisition (2016), Herdina and Jessner’s Dynamic Model of Multi­lingualism (2002) and Cenoz and Gorter’s ‘Focus on Multilingualism’ approach (Cenoz & Gorter, 2011a, 2011b; Cenoz, 2013). The first four are from the generative-cum-psycholinguistic perspective, with the first three working on the initial state of L3A (see also Jaesnch, 2013; Rothman & Halloran, 2013) and the last one moving beyond the initial stage and covering various

Overview 11

less examined forces behind L3A. Herdina and Jessner’s one is from the more purely psycholinguistic perspective and Cenoz and Gorter’s from the sociolinguistic perspective (for a review of different L3A approaches from the psycholinguistic perspective, see Jessner, 2006). Reviewing all these models in this introductory chapter, regardless of whether they are from the generative perspective or not, is necessary in that they all help shed light on a range of factors/concerns which are crucial to our discussion of potential interaction between L2 and L3 in the discussion chapters.

Cumulative Enhancement Model Flynn et al.’s Cumulative Enhancement Model (CEM) (2004) draws our attention to the effect of acquisition sequence on L3A. The model is based on a comparison of the data collected from L1 Kazakh – L2 Russian – L3 English adults, L1 Kazakh – L2 Russian – L3 English children, L1 Spanish – L2 English adults and L1 Japanese – L2 English adults. Their production of three kinds of English restricted relative clauses (RRCs) was examined via an elicited imitation task: lexically headed with semantic content (e.g. The boy who kicked the ball scored a goal); lexically headed without any semantic content (e.g. The person who kicked the ball scored a goal); and free relative (e.g. Whoever kicks the ball will score a goal). The L3 adult subjects were found to show the same pattern of production of the three tested RRCs as their L1 Spanish – L2 English counterparts (rather than the L1 Japanese – L2 English speakers). Also, the L3 adult subjects produced more lexically headed relative clauses while the children used far more free relatives. These instances of contrastive performance, according to Flynn et al., suggest two points: first, any previously learnt language other than L1 (L2 Russian in this case) can influence L3 acquisition (e.g. L3 English); second, simultaneous or near-simultaneous acquisition of L2 Russian and L3 English among the children distinguished them from their adult counterparts. This in turn supports the cumulative effect of different linguistic experience in language acquisition, implying that learning languages other than L1 (L2 in this case) and the ordering of learning a second or third language can affect L3A in a positive manner. The cumulative effect in L3A, which largely hinges on the positive role of linguistic development, has been supported in some more recent studies. For example, Berkes and Flynn (2012a, 2012b) provide further support for the CEM with empirical findings obtained from 36 L1 Hungarian – L2 German – L3 English speakers. They focused on three types of English relative clauses, as in Flynn et al.’s (2004) study: lexically headed specified relative clauses; lexically headed unspecified relative clauses; and free relatives. Through an elicited imitation task, they examined the responses to the stimulus sentences from the L3 speakers and a group of L1 German – L2 English speakers (as a control group), and three distinctive patterns revealed differences between the two groups of speakers. First, the L3 group attained similar mean numbers of correct responses on the three types of relative clauses,

12  Crosslinguistic Influence in Multilinguals

but the L1–L2 group showed a significantly higher mean number of correct responses on free relatives that those for the other two types of relative clauses. Second, analyses of subject–object conversion errors between the two groups of speakers further indicated different developmental paths between the two groups: the L3 group made fewer errors than their L2 counterparts. Lastly, the L3 group had a higher correction score in the task than the L2 group. In light of these three distinctive patterns, Berkes and Flynn highlight the role of the multiplying influence of language acquisition and thus the relevance of the CEM in accounting for the difference between the L3 group and the L2 control group.

L2 Status Factor Model Further to Williams and Hammarberg’s (1998) and Hammarberg’s (2001) studies, Bardel and Falk (2007) proposed the L2 Status Factor Model to account for post-verbal sentence negation in the initial state of L3 Swedish or Dutch. Recordings, in the form of ‘semi-spontaneous speech’ (p. 461), were collected from two groups of L3 Swedish/Dutch learners in a classroom context: L1 V2 (verb second) – L2 non-V2 – L3 V2 (e.g. L1 Dutch – L2 English – L3 Swedish); and L1 non-V2 – L2 V2 – L3 V2 (e.g. L1 English – L2 German – L3 Dutch). Significant differences were noted between the two groups in terms of the number of post-verbal negated instances: the L1 non-V2 – L2 V2 – L3 V2 group produced more instances of post-verbal negation than the L1 V2 – L2 non-V2 – L3 V2 counterpart; the L1 V2 – L2 non-V2 – L3 V2 group, on the other hand, produced more postverbal negated instances with non-thematic verbs (e.g. copulas and modals). Given that the L1 V2 group did not resort to the same V2 pattern in their L3 Swedish/Dutch, Bardel and Falk stress the important role of L2 in L3A, and state that typological proximity is applicable only to L2–L3 transfer, not L1–L3 transfer, a claim that is contrary to Flynn et al.’s (2004) hypothesis. In a later study, Falk and Bardel (2011) highlighted the applicability of the L2 Status Factor Model to L3 intermediate learners by examining L3A of German object pronoun placement in both main and subordinate clauses among two groups of L1 English – L2 French (L1E–L2F) and L1 French – L2 English (L1F–L2E) speakers. The two groups were asked to complete a timed grammaticality correction judgement task, where they judged 60 sentences with German object pronouns and provided corrections for any ungrammatical sentences. Differences were noted between the two groups in their judgements and corrections of the test sentences, though both groups were found to have acquired grammatical object pronoun placement in L3 German subordinate clauses. The L1F–L2E group performed better in judging the grammatical and ungrammatical object pronoun placement in L3 German main clauses but the L1E–L2F group was more accurate in judging the ungrammatical object pronoun placement in L3 German subordinate

Overview 13

clauses. Such differences, according to Falk and Bardel, support the L2 Status Factor Model: the L1F–L2E group was assisted by their L2 English in judging all sentences, except ungrammatical object pronoun placement in German subordinate clauses, while the L1E–L2F group was helped by their L2 French in performing better in judging the ungrammatical object pronoun placement in German subordinate clauses. The researchers further explicate six cognitive and sociolinguistic variables of the L2 status factor, namely age of onset, outcome, learning situation (natural vs classroom setting), metalinguistic knowledge, learning strategies, and awareness of the learning process. (For a discussion of the L2 Status Factor Model from a neurolinguistic perspective, see Bardel & Falk, 2012.) In addition to Falk and Bardel (2011), Bono (2011) supports the role of L2 in L3A. Bono studied a group of elementary and intermediate L3 Spanish speakers whose native language was mostly French and L2 mostly English. Spontaneous oral data were collected from these L1 French – L2 English – L3 Spanish speakers in a classroom context and analysed in terms of three types of language switches: pragmatic switches, lexical insertions and metalinguistic switches (i.e. switching from content to form of words). The L3 elementary group was found to produce more pragmatic and meta­ linguistic switches, but both elementary and intermediate groups produced a similar proportion of lexical insertions among all the switches. While the switches showed the activation of the speakers’ L1, more than half of the lexical insertions, according to Bono, were driven by the ‘foreign’ status of the speakers’ L2 English.

Typological Primacy Model Stimulated by Flynn et al.’s (2004) CEM and Bardel and Falk’s (2007, 2012) L2 status factor model, Rothman (2011) advanced the Typological Primacy Model (TPM) to account for syntactic transfer at the initial stages of L3A. The TPM acknowledges that it is possible to have L1 or L2 transfer (unlike in the L2 status factor model) and such transfer can be positive or negative (unlike in the CEM). What determines the transfer options seems to be (psycho)typology of structural similarities, and transfer can therefore take place selectively. The model is based on data collected from 60 L1 Italian – L2 English – L3 Spanish and L1 English – L2 Spanish – L3 Portuguese speakers. They were asked to work on adjectival interpretation and placement in either Spanish or Portuguese in a semantic interpretation task and a context-based collocation task. The performance of the two groups was similar in the two experimental tasks, and also similar to that of a control group of native speakers. This confirmed the predictions of the TPM and highlights the important role of (psycho)typology. Rothman (2014) further elucidates the TPM by clarifying and expounding on a number of concepts germane to the model. In particular, he

14  Crosslinguistic Influence in Multilinguals

explicates how transfer from L1 or L2 to L3 takes place in the initial stages as a result of a multilingual’s subconscious processing of linguistic cues and analyses of structural/typological similarities of the languages concerned. To determine to what extent L1, L2 and L3 are similar or different – which he terms ‘overall typological proximity’ (p. 6) – a multilingual subconsciously assesses four types of cues in the three languages (p. 7): lexicon; phonological/phonotactic cues; functional morphology; and syntactic structure. Such assessment of the input will help the multilingual to decide on which similar or even identical mental representations of the cues from L1 or L2 are to be transferred to L3. (For a list of studies supporting the TPM, see Rothman & Halloran, 2013; Amaro et al., 2015.)

Scalpel Model of L3A The last generative perspective to be reviewed here is that of Slabakova (2016), where a number of driving forces behind L3A are proposed. The Scalpel Model probably evolved from her 2012 Modular Transfer Hypothesis, which explores the application of four L2A theories to explain L3A transfer as reported in three studies. The four L2A theories were: the Interpretability Hypothesis (Hawkins & Hattori, 2006; Tsimpli & Dimitrakopoulou, 2007), the Interface Hypothesis (Sorace, 2000, 2003, 2011), the Feature Re­assembly Hypothesis (Lardiere, 2007, 2008) and the Bottleneck Hypothesis (Slabakova, 2006, 2008). The three studies were by Chin (2008), Foote (2009) and Montrul et al. (2011). Slabakova reviews these three L3A studies and discusses various problems in applying the four L2A theories to them (e.g. no predictions about intermediate L3 learners, no explanation of the successful or unsuccessful acquisition of interpretable aspectual features. She further points to the broad distinctions between internal and external interfaces in the Interface Hypothesis and that between interpretable and un­interpretable features in the Interpretability Hypothesis. In light of the observation that none of the theories helps explain the transfer from both L1 and L2 to L3 and the roles of different L3A factors (e.g. psychotypology), she proposes the Modular Transfer Hypothesis. The hypothesis emphasises the transfer of ‘significant parts of grammar’ (p. 137) of L1 or L2 in L3A. Such ‘significant parts’ are determined by the ‘intrinsic difficulty of the linguistic property’ (p. 137). Should there be both (1) direct mapping between the form and meaning of inflectional morphology and (2) consistent L3 input, then transfer to L3 is more likely. The term ‘direct mapping’ denotes the match between form and meaning, and ‘consistent L3 input’ implies ample linguistic evidence for the learner to process in the input of the target language. The linguistic complexity of the target structure is thus one of the L3A factors examined in Slabakova’s (2016) Scalpel Model. The main tenet of the model is that transfer from L1/L2 to L3 can go beyond the initial

Overview 15

stages of L3A in the mode of ‘selective transfer ’, depending on the nature or ‘complexity’ of the structures involved. Whether the target structure is complex or not is subject in turn to both its own linguistic properties and relevant ‘cognitive and experiential’ factors, such as the nature of input and activation and the availability of positive/negative evidence. What is more, both L1 and L2 can play a role in influencing the acquisition of the target L3 in a positive or negative manner. To support these claims, a number of recent studies are cited (e.g. Bruhn de Garavito & Perpiñán, 2014; Fallah et al., 2016; García Mayo & Slabakova, 2015). Slabakova states that the Scalpel Model makes similar or even the same conjectures as L2A theories such as Lardiere’s Feature Re­assembly Hypothesis (2009) and her own Bottleneck Hypothesis (Slabakova, 2008) and L3A models such as the ones reviewed above. As we shall see in the discussion of the findings of the present study, both Lardiere’s Feature Reassembly Hypothesis and Slabakova’s Scalpel Model of L3A may be applied to examine potential reverse transfer from L3 to L2 in the present study. Lardiere’s analysis could be used to examine the more ‘internal’ or ‘micro’ interaction among the structures in the languages concerned, that is, via an examination of the intrinsic structural or linguistic features. Meanwhile, Slabakova’s Scalpel Model, which goes beyond the initial state of L3A among the CEF participants, might be adopted at a more ‘global’ or ‘macro’ level, where other ‘cognitive and experiential’ factors might be explored to account for possible reverse transfer noted in the structures involved.

Dynamic Model of Multilingualism To acknowledge the complexity of L3A and to highlight the presence of crosslinguistic interaction (CLIN) in a multilingual’s mind, Herdina and Jessner (2002) and Jessner (2006) put forward the Dynamic Model of Multilingualism (DMM), which is elucidated by the formula below (Jessner, 2006: 33): LS1 + LS2 + LS3 + LSn + CLIN + M = MP where LS = language system; CLIN = crosslinguistic interaction; M = a multilingualism factor; MP = multilingual proficiency. The interaction between different languages, which can be in the form of transfer or codeswitching, is assumed to enhance metalinguistic awareness and cognitive skills (part of M), leading to ‘multilingual proficiency’ (MP), the outcome of continuous changes in the language learning process. The significant role of metalinguistic awareness in L3A is further illustrated in Jessner’s (2014) study and substantiated in her latest account with Megens and Graus (Jessner et al., 2016). The two groups of English learners in Western Austria in Jessner’s 2014 study differed in terms of their age and

16  Crosslinguistic Influence in Multilinguals

Jessner argues that metalinguistic awareness could affect not only younger learners but also more mature ones. In the 2016 account, Jessner and her colleagues corroborate the importance of metalinguistic awareness in L3A by referring to a number of studies on crosslinguistic interaction, thereby pinpointing the distinctive nature of L3A.

Focus on Multilingualism Endorsing the importance of studying the whole linguistic background of a multilingual, Cenoz and Gorter (2011a, 2011b) and Cenoz (2013) put forward the ‘Focus on Multilingualism’ approach. The approach highlights the ‘soft boundaries’ (Cenoz & Gorter, 2011a: 306) among the languages in the development of the competence of a multilingual: the languages a multilingual knows interact with each other in the acquisition process. Analysing writing data from a group of L1 Basque/Spanish – L2 Spanish/ Basque – L3 English speakers, Cenoz and Gorter noted some strong correlations in the writing scores (both overall scores and component scores) across the three languages: Basque, Spanish and English. For example, a higher score in grammar (as one component) in one piece of L2 Basque writing was found to be positively correlated with a higher score in grammar in a piece of L3 English writing. Instances of transfer among the three languages were also observed in their writing (e.g. from L2 Spanish to L1 Basque, or from L2 Basque to L3 English). No particular language was found to play a more dominant role than the other two. Which language to be used (L1, L2 or L3) hinges on the proficiency of the learner in the languages, the message(s) being conveyed, and other factors, such as ‘attitudes [to] or perceptions of the different languages’ (p. 306). The discussion of the findings thus calls for the need to consider all the languages concerned in examining language learning and use among multilinguals.

Second Language Acquisition from the Generative Perspective Given the close connection between L3A and its predecessor, second language acquisition (L2A), this section presents a brief review of L2A from the generative perspective, ending with how key concepts in L2A can be interpreted when three languages are involved (i.e. L3A).

A brief review of generative L2A Research on L2A in the generative tradition has centred on: (1) the availability and role of Universal Grammar (UG) in the acquisition of a

Overview 17

non-L1; and (2) the possibility of crosslinguistic influence from the learners’ L1 to the target L2 (i.e. possible transfer from L1 to L2). Table 1.3 sets out the possible sets of relations between these two phenomena and lists the relevant studies. Table 1.3  Access to Universal Grammar (UG) and L1 transfer in the generative tradition UG access

L1 transfer

Key studies

Full

Full

White (1989); Schwartz (1998); Schwartz and Sprouse (1994, 1996, 2000 )

Full

Partial

Vainikka and Young-Scholten (1994, 1996a, 1996b); Eubank (1993/94, 1996)

Full

None

Epstein et al. (1996, 1998); Flynn (1996); Flynn and Martohardjono (1994)

Partial

Full

Hawkins and Chan (1997); Tsimpli and Roussou (1991)

Partial

Partial

Beck (1997); Eubank et al. (1997)

None

Full

Bley-Vroman (1989, 2009); Schachter (1988)

Three possibilities for the availability of UG in L2A have been put forward: full access, partial access and no access. Studies advocating full UG access (e.g. Schwartz, 1998; Vainikka & Young-Scholten, 1994, 1996a, 1996b; White, 1989) highlight two points. One is that both L2A and L1A involve UG in a similar manner; in other words, UG is open to and facilitates both acquisitional processes. The other point is that an L2 learner can still get access to UG even when his/her L1 does not possess the target L2 features or properties. For example, L1 Chinese learners, without the agreement features, can still gain access to UG in acquiring the corresponding features in English. It is the second point which some researchers (e.g. Beck, 1997; Hawkins & Chan, 1997) find controversial. Those researchers endorse partial UG access, explaining that UG properties which are not available in the learner’s L1 will no longer be available in L2A. Take the L1 Chinese learners as the example again: they will not be able to access UG in learning English agreement features because such features are not present in their L1. Research supporting no UG access (e.g. Bley-Vroman, 1989, 2009), on the other hand, rejects the role of UG in L2A and upholds the stance that L2A is not the same as L1A and L2 is acquired by means of general problem-solving abilities. The three degrees of UG access – full, partial or none – can also apply to possible transfer from L1 to L2. The notion of ‘full transfer’ implies that the entire L1 grammar, including all the abstract properties and features, serves as the base of L2A (e.g. Schwartz & Sprouse, 1994, 1996, 2000). Unlike the full transfer stance, ‘partial transfer’ models (e.g. Vainikka & Young-Scholten,

18  Crosslinguistic Influence in Multilinguals

1994, 1996a, 1996b; Eubank, 1993/94, 1996) hinge on the assumption that either functional categories (e.g. agreement) or feature strength of functional categories (e.g. strong or weak feature of agreement) are absent or ‘inert’ in the initial state of L2A, resulting in partial transfer of the L1 grammar. Last, no L1 transfer (e.g. Epstein et al., 1996, 1998) highlights the disconnection between the L1 grammar and the development of the L2 grammar; in other words, L2A starts with UG instead of the learners’ L1 grammar. However, as White (2000) reviews, the combination no transfer with full UG access has not gained any support in the field, whereas ample instances of L1 transfer have been noted.

A move from L2A to L3A in the generative tradition From the above L2A review, two concepts stand out: (1) access to UG and (2) crosslinguistic influence (or ‘transfer’). While much L2A research has either supported or refuted the two concepts and the argument is still ongoing, a recent study (Ding et al., 2016) has corroborated the notion of UG with neural data, and suggests that UG is accessible in language acquisition. Should that be the case, UG is likely to be accessible in L3A as well. What is more likely to remain controversial (or in fact to become more controversial) is the possibility of and mode of transfer/crosslinguistic influence among the three languages concerned. The possibility of transfer refers to whether transfer takes place (i.e. ‘transfer effect’ vs ‘no transfer effect’). The two options for the mode of transfer/crosslinguistic influence are, as above, ‘full’ and ‘partial’. Such a distinction between full transfer and partial transfer seems less crucial in L3A than in L2A, however; Leung (2006) states that both the full and the partial transfer L2A models may predict that either L1 or L2 influences L3, resulting in the two transfer paths shown in Figure 1.1, which represent the possible interactions among the three languages in L3A. First, L1 influences L2 and L2 affects L3, but no transfer is noted from L1 to L3. Alternatively, L1 influences both L2 and L3, and L2 does not transfer to L3. These paths of interaction among the languages, according to Leung, may exist in L3A no matter whether the entire or partial grammar of the earlier languages influences that of the later ones. In light

L1

L1

L2

L2

L3

+

Figure 1.1  Possible paths of transfer in L3A

L1

L3

Overview 19

of Leung’s remark, the present L3A study focused on whether and how the three languages concerned interact with each other, and no assumption was made about whether there would be full or partial transfer between the languages involved.

Reverse Transfer in L2A and L3A The acquisition of L2, as reviewed above, focuses on ‘forward transfer’, that is, crosslinguistic influence from L1 to L2. However, some studies have pointed out the possibility of ‘reverse transfer’, that is, an influence of L2 on L1. This possibility is explained in Cook’s (2003) ‘Integration Continuum’, where the notion of multi-competence is highlighted. Multi-competence is defined as ‘the compound state of a mind with two grammars’ (Cook, 1991: 112), resulting from the ‘knowledge of two or more languages in one mind’ (Cook, 1991; Cook, 2003: 2). In relation to multi-competence is the Integration Continuum (Cook, 2002), which is a composition of three models: separation, integration and interconnection (Figure 1.2).

separation

LA

LB

interconnection

LA LB

integration

LA + LB

Figure 1.2  Cook’s Integration Continuum of possible relationships in multi-competence in two languages (LA and LB)

At one end is the separation model, in which there is basically no interaction between the two languages (LA and LB). The languages are in ‘watertight compartments’ and crosslinguistic transfer does not exist. However, the model seems less tenable because there has been evidence of possible interaction, such as L1-driven non-target forms in L2. At the other end of the continuum is the integration model. It suggests the presence of a single system, which in turn means that there is no distinction between LA and LB. Again, this claim is not tenable, in that learners can always tell the languages apart. Last, the interconnection model, which is in the middle,

20  Crosslinguistic Influence in Multilinguals

Table 1.4  Studies of multi-competence examined in Cook (2003) Studiesa

Research foci

Languages

Research tools

Laufer

Collocational knowledge

L1 Russian; L2 English

Correctness judgement task; written composition

Pavlenko; Dewaele and Pavlenko

Lexicon and semantics; morphosyntax; linguistic framing

L1 Russian; L2 English

Oral narrative

Cenoz

Requesting behaviour

L1 Spanish; L2 Basque; L3 English

Discourse completion test

Jarvis

Lexicosemantics and general idioms

L1 Finnish; L2 English

Natural language use; elicited but unguided data; guided data; metalingustic judgements; selfreported data

Porte

Codemixing and blending

L1 English; L2 Spanish

Group discussion

Murphy and Pine Inflectional morphology

L1 English; L2 nonEnglish; L1 nonEnglish; L2 Englishb

Production task; forced choice task; reading task; grammaticality judgement task

Balcom

Middle constructions

L1 French; L2 English

Grammaticality judgement task

Cook, Iarossi, Stellakis and Tokumaru

Word order; case; animacy; agreement

L1 Japanese/ Spanish/Greek; L2 English

Judgement task

a

Studies are reported in separate chapters in the collection edited by Cook (2003).

There is no specific mention of the non-English languages involved. Instead, the participants are described as those with ‘at least one other language – which was also spoken in the home’ (Cook, 2003: 150) or as ‘a heterogeneous group of bilinguals varying in their L1’ (Cook, 2003: 160). b

points to partial interaction between LA and LB as driven by shared or overlapping aspects of language; it is in this model that interaction between languages can trigger both forward transfer and reverse transfer. The notion of multi-competence was examined in a number of studies reported in various chapters in the volume edited by Cook (2003) and Table 1.4 summarises the foci, languages concerned and research tools used in those studies. The experimental studies illustrate how reverse transfer can be examined via different means, and Kecskes and Papp (2003) discuss various methods and techniques that can be deployed to demonstrate reverse

Overview 21

transfer in a foreign language environment (i.e. language learning mostly in a classroom context). They highlight ‘change in conceptual fluency’ (Kecskes & Papp, 2003: 253) as an indicator of reverse transfer; it can be measured by tasks which focus on content rather than language structures, such as picture-writing tasks, free writing tasks and summary writing tasks. In analysing the collected data, attention should be paid to three areas which reflect such fluency: appropriate structural form (the complexity and nature of clauses), lexical quality (the variation and sophistication of the words), and cognitive functioning (the use of modality and metaphorical ex­pressions).4

Reverse transfer in L3A Reverse transfer has also been recently examined in three studies involving L1 Cantonese speakers in the L3A context. Hui (2010) investigates the use of English relative clauses among a group of L1 Cantonese – L2 English – L3 French speakers. In a written picture elicitation task, the participants were asked to respond in English to some questions about 24 pictures. Compared with a group of L1 Cantonese – L2 English speakers, the L3 French group used more full subject-extracted relative clauses (e.g. the man who wears sunglasses) than reduced relative clauses with present participle (e.g. the man wearing sunglasses) in their responses. They also inserted relative pronouns in object-extracted relative clauses (e.g. the man who we all know, vs the man we all know) in their responses more often. Last, non-human antecedents were accompanied by ‘who’ (e.g. the horse who Tom is riding). Hui concluded that such performance was caused by transfer from L3 to L2, that is, reverse transfer, in that full subject- or objectextracted relative clauses are far more common than the reduced ones with present participles in French. Cheung et al. (2011) studied the judgement on and use of English past simple tense among a group of L1 Cantonese – L2 English – L3 German speakers. In light of two possible renditions of the German past form (past simple or present perfect), it was hypothesised that L3 German might affect the learners’ L2 English past simple tense. A free writing task and an acceptability judgement task were used to verify the hypothesis, and differences were found between the L3 German group and the L2 English control group. The L3 group produced non-target present perfect instances in the English writing task, while the control L1 Cantonese – L2 English group did not produce any present perfect instances. The L3 group also displayed a higher acceptance rate of non-target English present perfect sentences in the judgement task. The performance of the L3 group in both tasks thus indicated possible traces of negative transfer from L3 German to L2 English, perhaps driven by factors such as actual and perceived language distance, recency of use/acquisition and frequency of use.

22  Crosslinguistic Influence in Multilinguals

Matthews et al. (2014) explored possible ‘anti-transfer’ from L3 German back to L2 English. Free writing collected from an L1 Cantonese – L2 English – L3 German group in their 2011 study was examined, with special attention to the ungrammatical use of the English present simple tense in referring to past events. A comparison of the writing of the L3 German participants with that of a group of L1 Cantonese – L2 English speakers showed that the L3 German group produced fewer instances of ungrammatical use of the present simple tense in the past than the L1 Cantonese – L2 English group. Matthews et al. account for such difference in terms of the possibility of ‘anti-transfer’, where L3 German might have had a counteracting effect against the possible negative transfer effect from L1 Cantonese. Under such circumstance, the learners’ L3 German possibly played a vital role in expunging the ungrammatical forms caused by their L1 Cantonese in the free writing. From the above, it can be seen that both forward and reverse transfer is possible, which corresponds to Pavlenko and Jarvis’s (2002) concept of ‘bi­directional language transfer’ in their L1 Russian – L2 English study. Similarly, Cenoz (2001: 2) highlights this possibility in L3A: bi-directional relationships can take place in third language acquisition: the L3 can influence the L1 and be influenced by the L1 (L1↔L3) and crosslinguistic influence can also take place between the L2 and the L3 (L2↔L3). Given (1) such chance of forward and reverse transfer in L2A as well as (2) the expectations derived from the full and partial L2A transfer models, the notion of transfer in L3A results in a number of possible paths of transfer, as set out in Figure 1.3. With the forward transfer paths explained earlier, attention is paid to the other two types of transfer path: (1) reverse transfer paths and (2) both forward and reverse paths. The first possible reverse transfer path concerns the influence of L3, the latest acquired language, on L2, an earlier acquired language, and then from L2 to L1, the learner’s L1 (i.e. L1 ß L2 ß L3). Another possibility is where L3, the latest acquired language, influences the learner’s L1 too (i.e. L1 ß L2; L1 ß L3). The last possible path involves L3 influencing both L1 and L2, leading to two reverse transfer paths. While these three possibilities are all one-way transfer paths (with arrows pointing in the same direction), the presence of three languages suggests the possi­ bilities of forward and reverse transfer between the languages concerned. As Figure 1.3c shows, there are two possible forward and two possible reverse transfer paths: (1) L1 influences L2 (i.e. forward transfer) and L3 influences L2 (i.e. reverse transfer), and (2) L1 is influenced by L2 (reverse transfer), which also influences L3 (i.e. forward transfer). It is these two forward and two reverse transfer paths (Figure 1.3b and c) that the present study was particularly interested in.

Overview 23

L1

L1

L2

L2

L3

L1

+

L3

(a) Forward transfer paths (Figure 1.1)

L1

L2

L1

L2

+

L1

L3

L1

L3

+

L2

L3

L3

(b) Reverse transfer paths

L1 L1

L2 L2

L3 L3

(c) Forward and reverse transfer paths Figure 1.3  Possible paths of forward and reverse transfer in L3A

The Interface Hypothesis in L2A and L3A The Interface Hypothesis (IH) centres on the concept of ‘interface’. As pointed out by Sorace in her keynote article (2011: 6), the term ‘interface’ can be defined as ‘syntactic structures that are sensitive to conditions … [which] have to be satisfied … for the structure to be grammatical and/ or felicitous’. The definition implies that an interface involves a syntactic structure and the conditions relevant to the structure. Should the conditions be related to morphology (e.g. plural marking in English), the interface concerned is the syntax–morphology interface. Should the conditions concern discourse (e.g. null arguments in Greek), the interface is the syntax– discourse interface. Interfaces broadly fall into two groups: internal and

24  Crosslinguistic Influence in Multilinguals

external. Internal interfaces cover the interaction between syntax and other language-internal domains (e.g. semantics), while external interfaces relate to the interaction between syntax and language-external domains such as the cognitive domain and world knowledge (e.g. pragmatics). The IH was first advanced by Sorace and Filiaci (2006) in their study of a group of near-native L2 Italian speakers whose L1 was English. The syntactic structures concerned Italian pronominal subjects in complex sentences and were considered to be at the interface between a linguistic domain (syntax) and a cognitive domain (discourse). Through a picture verification task, the L1 English – L2 Italian speakers were observed to show interpretation of Italian null subjects similar to those revealed by native Italian speakers. However, a different picture was displayed in the structures with overt subjects. While the native speakers preferred ‘extralinguistic referent[s]’ (Sorace and Filiaci, 2006: 359) for overt subject pronouns (i.e. a referent as implied by the context rather than the subject of the sentence) the L2 speakers opted for the matrix subjects of the complex sentences, implying the non-native-like interpretation of overt pronouns among the learners. In light of the asymmetry noted in the learners’ interpretations of null and overt subjects, Sorace and Filiaci argued that the syntax–discourse interface (as exemplified by extralinguistic referencing) is more vulnerable, even among near-native learners, where ‘residual first language (L1) effects, in­ determinacy or optionality’ can be found. They further propose two factors behind the difficulties at the syntax–discourse interface: learners’ ‘level of knowledge representations’ and the availability of ‘processing resources’ for learners (Sorace & Filiaci, 2006: 340). In a later account, Sorace and Serratrice (2009) highlight the respective roles of processing resources and ‘structural overlap’ (p. 196) in the syntax– discourse and syntax–semantics interfaces by comparing the results from one study on overt/null pronominals in Italian/English and from another on definite articles in Italian/English. They also address five factors which result in the distinction between internal and external interfaces (pp. 198–199, abridged below): (1) underspecification of interpretable features; (2) crosslinguistic influence; (3) processing limitations; (4) input received by bilingual speakers; (5) bilingualism effects. The first two, which concern the structural aspects of the language, are interrelated: interpretable features, such as animacy (e.g. animate ‘who’ vs inanimate ‘which’ in English wh-interrogatives), may be underspecified (i.e. not fully developed) in the learners’ L2 possibly because of the absence of the corresponding features in their L1. The remaining three factors, which relate to language learning and use, highlight the role of the availability and

Overview 25

processing of language resources among the L2 learners in a bilingual context. Presumably, the bilingual learners, unlike monolinguals, are in a complex language environment, where they are exposed to different language forms. Owing to the potential competition among these forms, the L2 learners showed slower processing of their comprehension and production of the two languages. Since its introduction, the IH has stimulated much discussion from different perspectives about L2A among near-native learners. A number of concerns were covered in a special issue of Linguistic Approaches to Bilingualism (2011), and some problematic issues were highlighted which prompt further scrutiny of the hypothesis: • conceptual issues of the hypothesis – circularity in defining ‘interface’ (Duffield: 35ff.), the distinction between narrow syntax and an interface (Gurel: 40ff.), the nature of topic and focus on the syntax–discourse interface (Slabakova: 89ff.); • the ‘computational complexity’ of different structures within and across interfaces (Hopp: 45; Pires & Rothman: 74ff.); • the nature of the ‘end-state’ in L2A (Lardiere: 51ff.); • the role of input in determining the difficulty of an interface (Paradis: 68ff.; Pires & Rothman: 74ff.); • ‘optionality’ among native speakers (Prevost: 79ff.). The above concerns, as well as the diverse arguments and findings about the IH and its internal/external interfaces, might have explained the revised version of the IH presented by Sorace (2012), when she revisited the notion of narrow syntax/internal interfaces and acknowledged the potential complexity of narrow syntax and internal interfaces: ‘narrow syntactic properties and internal interfaces can be computationally complex and resource consuming too, as Gurel observes’ (Sorace, 2012: 210). In other words, not only the external interface but also the internal one can constitute a challenge to learners. This revised version of the IH, which assigns a ‘new’ status to narrow syntax and internal interfaces, thus corresponds with the findings of many empirically based generative studies, where L2 learners were reported to have produced non-target language structures which belong to the so-called internal interface. Therefore it is justifiable to look into the acquisition of any structures/features on not only the external but also the internal interfaces. As just stated, the IH has fostered many studies of different interfaces involving multifarious language pairings (e.g. Anderssen & Bentzen, 2013; Belletti et al., 2007; Cuza, 2013; Ivanov, 2012; Jegerski et al., 2011; LilloMartin & de Quadros, 2011; Shin & Cairns, 2009; Tsimpli & Sorace, 2006; Wilson, 2009; Wilson et al., 2009; Yamada, 2009; Yuan, 2013, to name just a few). Instead of reviewing all the above studies, the following will discuss the major claims of various pioneering or core studies on the IH and illustrate

26  Crosslinguistic Influence in Multilinguals

how research has been conducted to examine the hypothesis with different methods and language pairings. Among the most cited studies on the IH is that of Tsimpli and Sorace (2006), which focuses on both the syntax–discourse and the syntax– semantics interfaces. They investigated the acquisition of Greek overt subject pronouns by 27 near-native Russian learners of Greek. These 27 participants were put into three groups according to their length of residence in Greece and thus exposure to Greek. In Greek, focusing involves verb raising and is at the syntax–semantics interface, while the usage of overt subject pronouns is discourse-regulated and is therefore at the syntax–discourse interface. The data collected from oral interviews of the three groups of participants revealed alignment between the use of null subjects of the learners and that of the native Greek speakers. On the other hand, just like Sorace and Filiaci’s study of Italian learners, the learners of Greek were observed to have overused overt subject pronouns, regardless of their length of stay in Greece, which in turn is an indication of their non-native pronominal usage. Meanwhile, the learners seem to have native-like performance in their verb-raising patterns as associated with focusing, even though they were at the beginning of their stay in Greece. The distinctive patterns between subject overt pronouns and focusing in the data, as Tsimpli and Sorace point out, suggest possible differences between the syntax–discourse and syntax–semantics interfaces. A similar picture is revealed in Yamada’s (2009) study of the acquisition of Japanese pronominals by five English native speakers and five Korean native speakers. She explored the extent to which ‘optionality’, as discussed in the IH, is present in Japanese, which uses both null and overt pronouns. In a forced written elicitation task, the two learner groups and the control group showed the same pattern of use of null and overt subject pronouns. However, the English group displayed optionality in the pattern of use of object pronouns, whereas the Korean group performed similarly to the native group. While Yamada proposes that Park’s (2004) account better explains the performance of the English group from a syntactic perspective, the discrepancy between overt and null Japanese objects among the English learners can be treated as another illustration of the problematic nature of the syntax–discourse interface. The vulnerable nature of the syntax–discourse interface is also supported by data from L1 acquisition of sign language. Lillo-Martin and de Quadros examine the pattern of role shift in sign language, which concerns description of events with ‘a particular character’s point of view’ (Lillo-Martin & de Quadros, 2011: 624) and involves both sentence-level and discourse-level constraints. In total, 2200 spontaneous items were collected from two deaf children (aged between 1 year 7 months and 2 years five months) whose native languages were American Sign Language and Brazilian Sign Language. The two children were found to be able to mark role shift with first-person

Overview 27

verb agreement by eye gaze and facial expressions, implying their knowledge of an existence of an operator for a covert point of view in sign language. However, they did not produce any ‘shifted overt first-person pronouns’ (p. 633), which could be accounted for by their young age and resulting lack of ability to acquire different speech acts. The researchers argued that while the children managed to display certain sentence-level features (e.g. point of view operator), they were still not native-like in some specific speech contexts by not making the referent explicit or keeping the point of view of the referent constant across sentences. Their findings thus lend support to the vulnerability of the syntax–discourse interface. Meanwhile, some other studies have noted results contrary to the expectations of the IH (e.g. Cuza & Frank, 2011; Donaldson, 2012; Ivanov, 2012; Yuan, 2010; Zhao, 2012, to name just a few). The results are opposite in two ways: (1) the near-native/advanced L2 learners did not show any difficulties on the external interfaces, and (2) the internal interfaces were still problematic for near-native/advanced L2 learners. For example, Slabakova, Rothman and Kempchinsky (2011) adopt another perspective to show the controversial nature of the syntax–discourse interface. Using Duffield’s (2003, 2005, as cited in their work) division between ‘underlying competence’ and ‘surface competence’, Slabakova et al. illustrate how the judgements of L2 Spanish learners could diverge from that of native speakers’ in four Spanish dislocation structures in the syntax–discourse interface. They found that the L2 learners revealed a judgement pattern similar to the one depicted in the literature (e.g. Lopez, 2009, as cited) but the native speakers displayed an unexpected pattern. They argue that the native speakers might have reanalysed the given sentences and therefore showed a different pattern. The unexpected performance of the native speakers prompted Slabakova et al. to point out that the reliance of native speakers’ judgements in the examination of the nature of an interface deserves some further thought, which itself suggests further challenges in pinpointing what an interface is. Similarly, Donaldson (2012) expressed concern about the perceived vulnerability of the syntax–discourse interface. He examined the use of two French cleft structures, c’est (it’s) and avoir (have), among 10 L1 English speakers whose French reached the near-native level. In French, the two expressions are used respectively to indicate focus and introduce new referents in a discourse, rendering them properties of the syntax–discourse interface. Both experimental and spontaneous conversational data were collected from the group of L2 French speakers. Their performance on two experimental tasks, one on the felicity of given sentences and one on their preferences for the sentences, was found to be similar to that of another group of L1 French speakers. Neither was a significant difference noted between the learner and native groups in terms of the use of the two cleft structures in the spontaneous production. In light of the lack of significant differences between the L1 English – L2 French group and the native French group,

28  Crosslinguistic Influence in Multilinguals

Donaldson concluded that it is possible for learners to reach a native-like level on the syntax–discourse interface. Concerning the nature/status of the internal interfaces, Slabokova (2010), upholding the Bottleneck Hypothesis (Slabokova, 2008, 2009), highlights how challenging morphology can be to the learners in some structures on the syntax–semantics interface, one example being aspectual mismatch between English and languages such as German. Slabakova and Ivanov (2011) compared two studies on the acquisition of clitic left dislocation by near-native L2 Spanish learners (Valenzuela, 2005, 2006) and advanced L2 Bulgarian learners (Ivanov, 2009). Although the two studies investigated comparable structures in their respective languages, they differed in terms of the conclusions reached: the syntax–discourse interface was problematic even among near-native Spanish learners, while the syntax–discourse interface was learnable among advanced Bulgarian learners. Slabakova and Ivanov (2011) reinterpreted Valenzuela’s results as some form of ceiling effect and overgeneralisation, and pointed out that the syntax–discourse interface might not be as difficult as it is perceived and that the seeming distinction between the external and internal interfaces needs more careful examination. White (2011) also argues for the challenging nature of the syntax–semantic interface, by citing research done on the L2 acquisition of Korean, Japanese or Chinese. As another example of its problematic nature, she refers to some robust discussions of those difficult structures involving syntax and morphology, namely Lardiere’s (2007, 2008, 2009) numerous discussions of the acquisition of English tense marking by L1 Chinese – L2 English subject Patty. Some empirical studies have also supported White’s remarks by showing how problematic the internal interfaces can be for learners. For example, Cuza and Frank’s (2011) study revealed potential difficulties of the syntax– semantics interface for heritage speakers in the USA. They investigated the acquisition of Spanish double-que in embedded questions, which belong to the syntax–semantics interface, among a group of heritage speakers. The participants were asked to complete three tasks: an oral sentence completion task, an acceptability judgement task and a preference task. Compared with a group of native Spanish speakers in the study, the heritage speakers were found to produce fewer double-que constructions, to accept ungrammatical instances without que more often, and to prefer both grammatical and ungrammatical usage of double-que. In light of the non-native-like performance of the participants, Cuza and Frank discuss the role of transfer from the majority language (i.e. English) and bilingual effects (i.e. input and use of Spanish in the English-speaking environment) in the acquisition of the syntax–semantics structure. Likewise, Androutsopoulou et al. (2010) investigated the syntax–­ morphology interface by focusing on the acquisition of plural inflection on quantifiers in Spanish by French native speakers. The key concern was the

Overview 29

way the L2 learners’ knowledge of plural marking on quantifiers in Spanish relates to that of syntactic patterns of quantifier determiner phrases (DPs). In forming quantified DPs, both French and Spanish allow direct construction (DC) (i.e. the quantifier immediately followed by the noun) or de-­construction (i.e. the quantifier immediately followed by partitive de and then the noun), depending on the number agreement between the noun and the quantifier. Should there be agreement between the noun and the quantifier (e.g. a plural noun and a qualifier with overt plural marking such as muchos, ‘many’), a DC is constructed. If there is an absence of agreement (e.g. for quantifiers, such as algo, ‘some’, with no number features), de-­construction is formed. Twenty-eight L2 intermediate and advanced Spanish learners completed two tasks on four quantifiers, namely mucho (‘many/much’), tanto (‘so many/much’), demasiada (‘too many/much’) and bastante (‘enough’): a grammaticality judgement task and an elicited production task. While the intermediate and advanced learners differed in their patterns of rejecting uninflected quantifiers, both groups were found to have performed worst for sentences with bastante. In both tasks, they were less able to reject the uninflected plural bastante or produce plural bastante with overt plural marking. This led Androutsopoulou et al. to conclude that there seemed to be a scale of quantifiers among the L2 learners, in that they had not performed the same for all the tested quantifiers, which implies the problematic nature of the syntax–morphology interface. Yuan (2010) studied the L2 acquisition of Chinese wh-words (e.g. shenme, ‘what’) by English and Japanese native speakers who had different levels of proficiency in Mandarin Chinese (beginner, post-beginner, inter­mediate, post-intermediate, advanced). Crosslinguistically, while wh-words in Chinese are existential polarity words (EPWs) licensed in different domains, Japanese wh-words need the attachment of a particle ka or mo to become EPWs and English wh-words cannot function as polarity items. This crosslinguistic difference in turn implies challenges to Chinese learners whose first language is Japanese or English. From an acceptability judgement task, it was found that: (1) the connection between wh-words and their licensers was established mostly only with the post-intermediate and advanced English and Japanese groups; (2) the relationship between wh-words and some licensers was not established even among more proficient learners (e.g. the relation between wh-words and the licensing negators was available to more proficient learners but that between wh-words and the licensing A-not-A pattern was not); and (3) the less proficient Japanese groups performed better in some patterns than their English counterparts (suggesting developmental differences). While these findings suggest the problematic nature of the syntax–semantics interface among the learners, they draw our attention to the distinctive behaviour of each structure on an interface, some being more challenging to the learners than others. In light of the implications of the results which run contrary to the expectations of the IH,

30  Crosslinguistic Influence in Multilinguals

Yuan proposed a ‘variable-dependent approach’ to the study of an interface and points out four factors which can make a linguistic item vulnerable on an interface: the nature of the item; the status of the item in the target grammar (‘licensing powers’ of the licensers in this case; Yuan, 2010: 254), the nature of input learners receive; and crosslinguistic influence. Similar to Yuan, Zhao (2012) examined L2 Chinese in terms of the interpretation of overt and null embedded arguments. The overt argument under investigation was pronominal ta (‘he/she/it’) and the null argument was of two types: ∅ziji ‘self ’ and ∅topic. According to Zhao, ta belongs to the lexicon–syntax and syntax–semantics interfaces; ∅ziji ‘self ’ is a pure syntactic category; and ∅topic is on the syntax–discourse interface. Two groups of Chinese learners whose first language was English were asked to complete a picture judgement task to indicate the acceptability of the given sentences. The advanced group was found not to differ from the native control group in the judgement of all three test structures, although the higher-intermediate group showed differences in their interpretation of ∅topic in the subject position. While highlighting the role of crosslinguistic transfer and input, Zhao argued that both the internal and external interfaces are acquirable, but the external interfaces can be subject to delays in acquisition. Another study that supports Yuan’s (2010) variable-dependent perspective on the IH is that by Yuan and Dugarova (2011), who examined wh-topicalisation with four D(iscourse)-linked wh-words, namely na ‘which,’ shenme (‘what’), sheide (‘whose’) and shei (‘who’), among 19 English learners of Chinese. Through a grammaticality judgement task, the English learners were found to able to accept all of the topicalised D-linked wh-words except shei ‘who.’ In light of their acceptance of three of the topicalised D-linked wh-patterns and rejection of the topicalised non-Dlinked wh-patterns (e.g. *Shei ni xihuan; ‘Who do you like?’), Yuan and Dugarova argue that the learners’ judgements were unlikely to be subject to ‘unsuccessful integration of syntax and discourse’ (p. 168) or processing difficulties. They explain that the L2 learners possibly were not able to notice the difference between shei (‘who’) and the other three wh-words. Shei, unlike the other three wh-words, is a noun phrase rather than a determiner phrase which can link the wh-phrase (e.g. na bu dianying; ‘which film’) to the discourse. For shei to be linked to the discourse, an adverb (zui; ‘most’) is needed. In other words, what is problematic in the L2 grammar is some kind of representation vulnerability. Their study thus suggests two points about the IH: (1) the external interface may not be as vulnerable as reported (because of the learners’ ability to accept and reject most of the test structures), and (2) different patterns on one interface can behave differently among L2 learners. Yuan’s studies echo what White (2011) and Slabakova and Ivanov (2011) have pointed out: the possible intra-interface variation in relation to the vulnerability of an interface, suggesting that the difficulty of an interface

Overview 31

may not be taken across the board (as it seems to be). These researchers raise the issue of whether the problematic or unproblematic nature of an interface, as reported, is just a reflection of a particular linguistic structure at the interface rather than all structures at the interface concerned. In other words, different linguistic phenomena at one interface can be subject to different levels of difficulty. Extending the impact of the IH on L2A, Slabakova and García Mayo (2013) draw our attention to the applicability of the hypothesis to L3A too.5 They studied the acquisition of three English structures on the syntax–discourse interface – topicalisation, focus fronting and left dislocation – among three groups of speakers in the Basque community: (1) L1 Basque – L2 Spanish – L3 English; (2) L1 Spanish – L2 Basque – L3 English; and (3) L1 Spanish – L2 English. The participants completed an off-line acceptability task with audio-stimulus sentences, and their judgement of the three test structures showed different degrees, nature and sources of transfer. While the participants’ judgement of focus fronting showed successful acquisition, that of topicalisation indicated a possible negative as well as positive influence of Spanish as their L1 or L2 on their L3 English in different ways and that of left dislocation suggested a positive influence of Spanish. The L3 findings thus contrast with the prediction of the IH that the syntax–discourse interface presents difficulty for near-native learners. Slabakova and García Mayo also attribute their findings to: (1) structural similarities/differences between Spanish, Basque and English; and (2) the input effect – ‘the relative frequency of the various constructions in the input’ (Slabakova & García Mayo, 2013: 15). From the above review, it can be seen that various issues have been raised about the nature of interfaces and there may not be a neat distinction in terms of the acquirability of the two interfaces as proposed in the IH. Much effort has also been made in linking the IH to L2A, and the connection between the IH and L3A has started to gain attention. While most IH studies focusing on the external interfaces highlight the problematic nature of the interfaces in advanced learners’ grammar, some other studies have also revealed the potential difficulty of the internal interfaces for learners. Much research has been conducted on the external interfaces while that on the internal interfaces is rather lagging behind. This in turn implies the need for more research on both the external and the internal interfaces, focusing on different structures and learners of various language backgrounds. It is against this background that the present study examined three structures on the internal interfaces in the context of L3A. To end this review of the IH, it is important to state one caveat here: by no means was the present study intended to validate the IH, since such validation would require much more specific methodology in examining the distinction between internal and external interfaces or computational complexity. Instead, the IH serves as a means of categorising the test structures

32  Crosslinguistic Influence in Multilinguals

in terms of different structural properties within the internal interfaces (e.g. syntax–semantics vs syntax–morphology). The focus of the study is on how reverse transfer takes place at these different structures within the internal interfaces.

The Present Study Drawing on the notions of ‘interface’ and ‘reverse transfer’ in L2A, the present study addresses the possible role of L3 French in the acquisition of English as an L2. It has two major concerns: (1) the degree to which L3A will bring about a positive or negative transfer effect on L2A; and (2) the way in which an L3 interacts with an L2 and/or even an L1 on different interfaces as identified in L2A. These two research concerns served as the underpinnings of the main research question of the study: Does the acquisition of a later acquired language (in this study, French) have any effect on the reception and production of an earlier acquired non-native language (in this study, English)? Two terms in the research question require some clarification: ‘later acquired’ and ‘earlier acquired’. The presence of the two terms may give the impression that the present study focused primarily on the sequence of acquisition of L2/L3 and factors related to time or duration of acquisition (e.g. age). While the sequence of acquisition or the time/duration factor are crucial to our understanding of crosslinguistic interaction and therefore definitely worth examining, the availability of CEF participants in the research context of the present study did not allow the recruitment of a representative sample of participants with different groups of exactly identical sequences of acquisition of L2 and L3 or starting age of acquiring L2 or L3. Therefore, following the definition of L3 adopted in the present study (i.e. L3 being the third or any later language(s) acquired after L2), the two terms ‘later acquired’ and ‘earlier acquired’ had to be taken in a very broad or general sense, referring to ‘the third language’ (i.e. L3) and ‘the second language’ (i.e. L2) respectively in the present account. Also, all the L2A/L3A factors, as reviewed in this chapter, were elicited in the discussions of the findings of the three test structures whenever relevant. In search for the answer(s) to the research question, a comparison was made between the CEF group and three control groups, one of which was a group of L1 Cantonese – L2 English (CE) participants. These CE participants did not have any knowledge of languages which are typologically similar or closer to English. This made them different from the CEF group (who had some knowledge of French, a language typologically closer to English than is Chinese). In identifying possible reasons behind the observations

Overview 33

made, various L3A factors, as reviewed above (e.g. proficiency and [psycho] typology), were examined. The present study was intended to contribute to the field of language acquisition, particularly second and third language acquisition, in three respects: theoretical, methodological and pedagogical. First, the study, to the author’s best knowledge, was the first to address the notion of ‘reverse transfer’ in the L1 Chinese – L2 English – L3 French context. Also, the analysis of the data led to: (1) an interpretation Lardiere’s Feature Re­ assembly Hypothesis in the L3 context; (2) an exploration of possible factors behind potential crosslinguistic interactions among L1, L2 and L3 (following the line of the Scalpel Model of L3A); and (3) speculation about the possible sequence of the identified factors as exemplified by the CEF participants in the present study, which might be regarded as an application of the Scalpel Model. Concerning the methodological contribution, the present study exemplified what tools could be designed and used to examine reverse transfer at different levels. Lastly, the study offers some pedagogical suggestions on teaching learners with knowledge of more than two languages for educators’ reference.

Structure of the Book After this introductory discussion of the theoretical issues germane to the present research project, the design of the study and the nature of the three target structures are explicated in the next two chapters. The research findings in relation to the three test structures involving different interfaces are presented and discussed in the three subsequent chapters. Each of the chapters starts with a discussion of the structure and the interface concerned. The chapter then moves onto the acquisitional aspect of the study by looking first into the L2 acquisition of the structure by Chinese learners of English and then by the CEF participants in the research project. The discussion of the CEF participants in the core chapters presents the participants’ performance in terms of their reception and production of the structures concerned, offers explanations in terms of different factors in L3A and L2A, and interprets the findings from the perspective of the Interface Hypothesis. The concluding chapter presents a summary of the main findings and suggests directions for further research.

Notes (1) One other form of learners’ perception, namely perception of correctness in L3, is also argued to exert a possible influence on L3A (see De Angelis, 2005). (2) Williams and Hammarberg (1998) also refer to recency of use in explaining the activation of L2 German in their L3 data (the learner’s recency of use of German in

34  Crosslinguistic Influence in Multilinguals

Germany immediately before her move to Sweden). However, as discussed by De Angelis (2007: 36), recency of use is a controversial L3A factor, in that some studies have reported the activation of previously learnt languages which had not been ‘recently used’. (3) As some background reading for the two factors, readers can refer to L2A accounts such as the following: the age factor (Birdsong, 1999; Muñoz, 2014; Muñoz & Singleton, 2011; Singleton, 2001; Singleton & Lengyel, 1995; Singleton & Ryan, 2004) and the role of input or length of exposure to the target language (Gass, 1997; Krashen, 1985; Muñoz, 2014). (4) See Kecskes and Papp (2000) for a more detailed discussion of how reverse transfer took place among a group of L1 Hungarians and how such transfer can be examined with different indices. (5) To the best of my knowledge, Slabakova and García Mayo’s study is the first and only empirical one addressing the IH in the context of L3A. An earlier account by Slabakova (2012), as reviewed above, also connects the IH with L3A by discussing the results of three selected L3A studies in terms of the postulates of the IH and three other L2A generative proposals.

2 Design of the L3 French – L2 English Project

After the discussion of the theoretical and acquisitional backgrounds of the present study in the previous chapter, the details of the design of the present L3 French – L2 English project are presented. The coming sections explicate the following five areas: • • • • •

the research question and hypotheses; the setting of the L3 French – L2 English project; the design of the project; the test structures; the methodological limitations of the study.

Research Question and Hypotheses Drawing on the notions of ‘interface’ and ‘reverse transfer’ in L2A, the project addressed the possible role of L3 French in the L2A of three English structures, with two major concerns: (1) the degree to which L3A will bring about a positive or negative reverse transfer effect on L2A; (2) the way in which an L3 interacts with L2 and/or even L1 on an internal interface as identified in L2A. These two research concerns served as the underpinnings of the main research question of the study: Does the acquisition of a later acquired language (i.e. French) have any effect on the reception and production of an earlier acquired non-native language (i.e. English)?

35

36  Crosslinguistic Influence in Multilinguals

In view of the language background of the L1 Cantonese – L2 English – L3 French (CEF) participants, the literature on L2A of the selected English structures by L1 Chinese/Cantonese – L2 English (CE) speakers, the interface hypothesis (IH) and the notion of ‘reverse transfer’, the following two hypotheses were made in response to the above research question. Hypothesis 1 relates to research concern (1) and the possibility of reverse transfer involving an L3: • Hypothesis 1. Under the influence of their L3 French, the CEF participants will differ from their CE peers in terms of their reception and production of three English structures. Hypothesis 2 relates to research concern (2) and the possible vulnerability of the English structures as highlighted in L2A studies and the IH: • Hypothesis 2. Under the influence of their L1 Chinese, both CEF and CE participants will judge the selected English structures differently from the native controls and produce more non-target instances of the English structures than the native controls. Note that these hypotheses hinge principally on a comparison between two of the four groups of participants in the project: the CEF and CE groups. As explained in the next section (on the setting of the project), the main difference between these two groups of L1 Cantonese speakers concerned possession of knowledge of the French language, which could serve as one possible explanation for any difference observed between the two groups in the assigned language tasks, enhancing the examination of possible reverse transfer of L3 French to L2 English. Besides the CE groups, two control groups were recruited for the project: L1 French – L2 English (FE) speakers and L1 English (E) speakers. The FE group was involved in the comparison with the CEF and CE groups so as to examine the extent of any possible L3 French reverse transfer. Meanwhile, the native controls (E) were recruited to follow the research practice of the IH (explained below). Therefore, in the course of the research, the following specific questions on cross-group comparisons were also addressed whenever appropriate:

CEF vs CE/FE/E • Did the CEF speakers and CE speakers correspond with each other in their reception and production of the three English structures? • Did the reception and production of the CEF speakers correspond with those of the FE group?

Design of the L3 French – L2 English Project  37

• Did the reception and production of the CEF speakers correspond with those of the E speakers?

CE vs FE/E • Did the reception and production of the CE speakers correspond with those of the FE speakers? • Did the reception and production of the CE group correspond with those of the E group?

Setting of the L3 French – L2 English Project The L3 French – L2 English project took place in one of the universities in Hong Kong. The main medium of instruction and communication is English, and all university correspondence is in English (or in English and Chinese). The students attend lectures/seminars/tutorials which are conducted in English, and are required to submit their assignments in English. Language courses in French, German, Thai and Japanese are offered to the students but those languages are mainly used intensively during class time rather than on the whole campus. Four groups of participants were recruited, one of which was the ex­ perimental group and three of which were the control groups: • Experimental group: L1 Cantonese – L2 English – L3 French (CEF, n = 48). • Control groups: ààL1 Cantonese – L2 English (CE; n = 45); ààL1 French – L2 English (FE, n = 20); ààL1 English (E; n = 20). At the time of the project, all four groups of participants were reading a non-linguistic or non-language-teaching degree at the local university in Hong Kong. This could lower the possibility that they possessed any linguistic knowledge which might influence their performance in the assigned tasks. Table 2.1 presents the basic data on the four groups. For a detailed discussion of the language learning background of the CEF and CE groups, see Tsang (2015a). The CEF group, which was the experimental group, consisted of L1 Cantonese – L2 English speakers who were students of French at the university. They did not have any knowledge of another Romance or Germanic language (e.g. Italian and German), but they could know languages which are not related to French, such as Thai and Japanese. Through a French proficiency test (to be explained below), the CEF participants were further

38  Crosslinguistic Influence in Multilinguals

Table 2.1  Study participants CEF (n = 48)

CE (n = 45)

FE (n = 20)

E (n = 20)

Gender Male Female

n = 14 n = 34

n = 20 n = 25

n = 6 n = 14

n = 6 n = 14

Average age

20.69

20.82

20.9

21

Average no. of years learning English 16

17

11



Average no. of years learning French







 2

categorised into three proficiency groups based on their performance in the test: beginner (CEF-L; n = 17); intermediate (CEF-M; n = 21); advanced (CEF-H; n = 10). The CE group, the first control group, comprised L1 Cantonese – L2 English speakers who were students at the same university but who did not have knowledge of any Romance language. The background questionnaire showed that some of them did have knowledge of languages other than Cantonese and English, meaning that they could be L3 learners of languages other than Cantonese, English or French (as well as other typologically related or similar ones). In other words, one main difference between the CEF and CE groups concerned possession of some knowledge of French, which is germane to the research focus and the properties of the test structures of the L3 French – L2 English project. The FE group, which served as the second control group, consisted of native French speakers and were exchange students at the same university. Again, they were recruited so as to facilitate a comparison with the CEF (and CE) group in terms of any trace of reverse transfer from L3 French to L2 English. If the L1F–L2E data were significantly different from the L1C– L2E–L3F and L1C–L2E data, L1 Cantonese might have been playing a more dominant role in L2 English; if the L1F–L2E data were not significantly from those of L1C–L2E–L3F but were significantly different from the L1C–L2E data, the L3 French group (i.e. CEF), behaving like the L1 French group (FE), might be subject to the influence of the French language. Last, the native controls, the E group, were needed to examine the performance of the CEF group on the internal interfaces. The inclusion of the natives basically followed the tradition of L2A studies on the interface hypothesis (e.g. Yuan, 2013), where the native group would usually serve as the control or benchmark group in analysing the performance of the learners. According to the IH, whether one interface is vulnerable or not depends on the non-target performance of the learners, which is usually compared with the performance of native speakers (as in studies such as those reported by Belletti et al., 2007; Montrul, 2010; Yuan, 2013, to name just a few). Therefore, the present project included the native group so as

Design of the L3 French – L2 English Project  39

Table 2.2  English proficiency of the participants: Placement test scores of the CEF, CE and FE groups CEF

CE

FE

Average score (full score = 60)

47.81

46.91

45.10

Standard deviation

 4.71

 5.91

 7.80

CEF vs CE: Z = 0.571; P = 0.901 FE vs CEF: Z = 0.091; P = 0.038

Table 2.3  English proficiency of the three CEF sub-groups: Placement test scores CEF-H (n = 10)

CEF-M (n = 21)

CEF-L (n = 17)

CEF (all) (n = 48)

Average score (full score = 60)

49.2

47.86

46.94

47.81

Standard deviation

 4.32

 4.42

 5.31

 4.71

CEF-H vs CEF-M: Z = 0.378; P = 0.538 CEF-M vs CEF-L: Z = 0.680; P = 409 CEF-L vs CEF-H: Z = 1.46; P = 0.227

to ensure the methodology of the present study would be comparable with that of previous studies on the IH. Tables 2.2 and 2.3 display the performance of the participants on the Oxford placement tests (UCLES 2001), used to determined their proficiency in the English language. A two-sample Kolmogorov–Smirnov test found that the scores of the three L2 English groups, CEF, CE and FE, were not significantly different from each other, suggesting that their English levels were comparable. The comparison of the scores of the CEF sub-groups in the English placement test did not reveal any statistical difference either.

Design of the L3 French – L2 English Project With the aim of revealing the possible reverse transfer effect from L3 French to L2 English via triangulation, the present study employed three timed offline experimental tasks: a free writing task, a grammaticality judgement-correction task and a multiple-choice cloze task. The first two focused on the receptive knowledge of the target structure among the participants and the last one concerned the productive aspect of the language structures concerned. The participants were asked to finish the writing task first so that their free production would not be affected by the input from the other two

40  Crosslinguistic Influence in Multilinguals

experimental tasks, which probably drew their attention to a number of language structures. Besides these three experimental tasks, two proficiency tests, one in English and the other in French, and a background questionnaire were used.

Timed grammaticality judgement-correction task A timed grammaticality judgement-correction task, which consisted of 50 pairs of sentences, was used to retrieve learners’ judgement on five English structures, three being the test structures (to be explained in the next main section of this chapter) and two being distractors. For each structure, five pairs of sentences were grammatical and five ungrammatical. In each pair of sentences, the first introduced a context and the second was the ‘real’ test sentence with the target structure. Two examples for the past simple tense are shown below: (1) Little Kevin stayed at home last night. He finished all his homework before dinner. (2) *A few days ago, I bumped into Jane. But I have not recognised her at first sight. In each of the above two pairs, the first sentences present a past context, as indicated by the time adverbials (last night, a few days ago) and the tense forms. Without any time indicator in the second sentence, the tense should be the same as the one in the first sentence, namely past simple. Therefore, the pair in (1) is grammatical whereas the one in (2), with the use of the present perfect form, is not. Before conducting the experiment, the judgement-correction task was copy-edited by experienced English teachers who were also native English speakers. Then it was piloted among two group of university students who were L1 English speakers (n = 20) and L1 Cantonese speakers (n = 20). Also, two versions with different orderings of the pairs of sentences were devised to minimise any ordering effect on the participants’ judgement or correction. During the task, the participants were told that they were going to read 50 pairs of sentences presented in natural contexts and decide whether the sentences were grammatical by circling one of the numbers on a Likert scale from ‘1’ to ‘4’ (see Figure 2.1). They also knew that the number chosen represented the degree of grammaticality of the sentence (1 and 2 under the category of ‘ungrammatical’ and 3 and 4 under ‘grammatical’). If they gave a particular sentence a rating of 1 or 2, they were required to make any amendments to justify their rating. This could tap into why they rejected the sentence and facilitate the subsequent analysis of the rating. All participants managed to finish the task within the given 15 minutes.

Design of the L3 French – L2 English Project  41

Please rate the following sentences according to the scale by circling the appropriate number. If you choose options 1 or 2, remember to make necessary amendments to the sentence. Ungrammatical

Grammatical

The new books on the shelf are for display only. They will be on sale next week.

1

2

3

4

It’s Halloween this coming Monday. Would you like to buy a few pumpkin for the kids?

1

2

3

4

A new policy has been released. It has upset many residents in the city.

1

2

3

4

Figure 2.1  Timed grammaticality judgement-correction task

In the analysis of the judgement data, the overall mean ratings of the test sentences were calculated. To ensure that the ratings genuinely reflected the judgement of the participants as far as possible, any target rating plus non-target amendment, or non-target rating plus target amendment, was excluded from the analysis. The Wilcoxon signed rank test was run to discern any statistical significance in the mean ratings of the sentences within each participant group. The Mann–Whitney U-test and the Kruskal– Wallis one-way analysis of variance were run to compare the means of the four participant groups.1

Timed multiple-choice cloze task The other receptive task was a 15-minute multiple-choice cloze task. Following the practice adopted by researchers in vocabulary acquisition, the multiple-choice cloze task, where the participants were asked to identify or recognise the target responses, was considered to be a means of tapping into the learners’ receptive knowledge of the target structures. Just like the grammaticality judgement-correction task, the cloze task was piloted among the native speakers and L1 Cantonese speakers, together with the grammaticality judgement-correction task. The cloze, which was about the sale of human organs, comprised 30 multiple-choice blanks (five questions for each of the three test structures and three distractors). The choices differed in terms of tense (e.g. present perfect vs past simple), number (i.e. singular vs plural), adverb placement (e.g. before or after the main verb), agreement (i.e. singular verb vs plural verb) and position/type of relative pronouns (e.g. animate vs inanimate). An extract from the cloze is shown below:

42  Crosslinguistic Influence in Multilinguals

Local religious and cultural ④ valuation/ values/ value ⑤ play often/ often play an important role in the supply of organs. In Japan, people’s beliefs about brain death mean that there are not many donors inside the country. In the past, rich Japanese who wanted transplants asked gangs of criminals (called the ‘body mafia’) to find kidney donors in other countries. About ten years ago, journalists ⑥ had found/ found/ have found/ were finding one of the ⑦ gang/ gangsters/ gangs working through an important medical center in Boston, USA. Again, two versions with different orderings of the choices were designed. The participants were asked to choose the most appropriate choice to complete the cloze. The respective overall mean scores of the four groups in the cloze task (out of a possible total score 30) were: CEF = 25.94, CE = 25.11, FE = 26.4 and E = 27.4. The Kolmogorov–Smirnov test was conducted to test for any significant differences among the four participant groups in terms of their overall scores on the task and their scores for the respective test structures. No statistical difference was found in the overall scores between any groups, except the E group and the CE group: the CE group scored significantly lower than the E group (Z = 1.510, p