261 31 5MB
English Pages [276] Year 2015
Crosslinguistic Influence and Crosslinguistic Interaction in Multilingual Language Learning
Also available from Bloomsbury Experimental Research Methods in Language Learning, Aek Phakiti Multilingualism: A Critical Perspective, Adrian Blackledge and Angela Creese Multilingualism: Understanding Linguistic Diversity, John Edwards Social Identities and Multiple Selves in Foreign Language Education, edited by Damian J. Rivers and Stephanie Ann Houghton
Crosslinguistic Influence and Crosslinguistic Interaction in Multilingual Language Learning Edited by Gessica De Angelis, Ulrike Jessner and Marijana Kresić
Bloomsbury Academic An imprint of Bloomsbury Publishing Plc
Bloomsbury Academic An imprint of Bloomsbury Publishing Plc
50 Bedford Square London WC1B 3DP UK
1385 Broadway New York NY 10018 USA
www.bloomsbury.com BLOOMSBURY and the Diana logo are trademarks of Bloomsbury Publishing Plc First published 2015 Paperback edition first published 2017 © Gessica De Angelis, Ulrike Jessner, Marijana Kresic´ and Contributors, 2015 Gessica De Angelis, Ulrike Jessner and Marijana Kresic´ have asserted their right under the Copyright, Designs and Patents Act, 1988, to be identified as the Editors of this work. All rights reserved. No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage or retrieval system, without prior permission in writing from the publishers. No responsibility for loss caused to any individual or organization acting on or refraining from action as a result of the material in this publication can be accepted by Bloomsbury or the author. British Library Cataloguing-in-Publication Data A catalogue record for this book is available from the British Library. ISBN: HB: 978-1-4742-3585-3 PB: 978-1-3500-3648-2 ePDF: 978-1-4742-3587-7 ePub: 978-1-4742-3586-0 Library of Congress Cataloging-in-Publication Data Crosslinguistic influence and crosslinguistic interaction in multilingual language learning/edited by Gessica De Angelis, Ulrike Jessner, Marijana Kresic. pages cm Includes bibliographical references and index. ISBN 978-1-4742-3585-3 (hardback) – ISBN (invalid) 978-1-4742-3586-0 (epub) – ISBN 978-1-4742-3587-7 (epdf) 1. Multilingualism–Study and teaching. 2. Second language acquisition–Study and teaching. 3. Second language acquisition–Social aspects. 4. Education, Bilingual–Cross cultural studies. 5. Multicultural education–Cross cultural studies. 6. Intercultural communciation–Cross cultural studies. 7. Language acquisition–Cross cultural studies. I. De Angelis, Gessica, 1965- editor. II. Jessner, Ulrike, editor. III. Kresic, Marijana, editor. P115.2.C77 2015 404'.2–dc23 2015010280 Typeset by Deanta Global Publishing Services, Chennai, India
IN MEMORIAM of Håkan Ringbom (1936-2015) an outstanding scholar and pioneer in the field of multilingualism and crosslinguistic influence, and a dear friend who will be dearly missed
Contents Notes on Contributors
ix
1
The Complex Nature of Crosslinguistic Influence in Multilingual Learning Gessica De Angelis, Ulrike Jessner and Marijana Kresić 1
2
The Lexicon-Syntax Interface in a 3L1 context: An Experimental Investigation of the Unaccusative/Unergative Distinction in English and Italian Megan Devlin, Raffaella Folli and Christina Sevdali 13
3
Developing Cognitive Strategies Through Pluralistic Approaches Rebecca Dahm 43
4
Can a Background Language Alter the Path of Acquisition of Verb Placement in a Multilingual Context? Evidence from a Longitudinal Study Laura Sánchez 71
5
Item-related Determinants of Cognate Guessing in Multilinguals Jan Vanhove and Raphael Berthele 95
6
Perceptual Training of Novel Speech Contrasts in L3 Acquisition: The Effect of Multilingual Benefit Divya Verma Gogoi, James D. Harnsberger and Caroline Wiltshire 119
7
‘If you know Amharic you can read this’: Emergent Literacy in Multilingual Pre-reading Children Anat Stavans 149
8
Does German Help Speakers of Dutch to Understand Written and Spoken Danish Words? The Role of Non-Native Language Knowledge in Decoding an Unknown but Related Language Femke Swarte, Anja Schüppert and Charlotte Gooskens 173
9
Is A2 in German Better than B2 in French when Reading Danish? The Role of Prior Language Knowledge when Faced with an Unknown Language Karolina Mieszkowska and Agnieszka Otwinowska-Kasztelanic 199
viii
Contents
10 Do Learners Transfer from the Language They Perceive as Most Closely Related to the L3? The Role of Psychotypology for Lexical and Grammatical Crosslinguistic Influence in French L3 Christina Lindqvist 231
11 Crosslinguistic Influence and Metalinguistic Awareness Research: New Evidence and Future Challenges Gessica De Angelis, Ulrike Jessner and Marijana Kresić 253 Index
259
Notes on Contributors Agnieszka Otwinowska-Kasztelanic is an Assistant Professor at the Institute of English Studies, University of Warsaw, Poland. Her main field of research includes language acquisition, bilingual education and crosslinguistic influences in the multilingual lexicon. With Gessica De Angelis, she co-edited a volume entitled Teaching and Learning in Multilingual Contexts: Sociolinguistic and Educational Perspectives. She is currently principal investigator of a project on the linguistic development of Polish bilingual children raised in the United Kingdom. Anat Stavans is an Associate Professor of applied linguistics in the English department and director of the research authority at Beit Berl College and a researcher at the Research Institute for Innovation in Education, Israel. Her research focuses on developmental and educational linguistics, trilingual acquisition and development and cross-cultural and crosslinguistic literacy development. She is the co-author of Multilingualism (with Charlotte Hoffmann 2015) and co-editor of Studies in Language and Language Education (with Irit Kupferbeg 2008). She is a consultant for and developer of intervention programs in multilingual literacy. Recently, she has joined the EU COST Action ‘The New Speakers of Europe’; and she is principal investigator on a research project on predictors of text quality in written expository texts. Anja Schüppert is an Assistant Professor of linguistics at the Department for European Languages and Cultures at the University of Groningen, the Netherlands. Her research focuses on sociolinguistic and psycholinguistic issues with a special interest in Germanic languages, and in particular specializing in Scandinavian languages. One of her major fields of interest is the receptive multilingualism of speakers of Danish and Swedish with regard to their fellow Scandinavian languages. Caroline Wiltshire is an Associate Professor of linguistics at the University of Florida in the United States. Her research and teaching focus primarily on the phonology of Arabic, Romance languages and Indian languages, including both Dravidian and Tibeto-Burman languages. Dr Wiltshire has also published on the phonology of second language acquisition, especially Indian English, in journals such as World Englishes and English World-Wide, as well as in numerous books. Charlotte Gooskens is an Associate Professor of applied linguistics at the University of Groningen, the Netherlands. Her research focuses on two main subjects: (a) Objective and subjective quantitative measurements of linguistic differences between dialects and languages and (b) Communicative aspects of language variation. She is currently principal investigator of the project Mutual intelligibility of closely related languages in Europe: linguistic and non-linguistic determinants.
x
Notes on Contributors
Christina Lindqvist is an Associate Professor and research fellow of Romance languages, especially French, at Uppsala University, Sweden. Her main research interests are crosslinguistic influence, third language acquisition and vocabulary acquisition. Her most recent publication appeared in Eurosla Yearbook in 2014: When Germans begin to learn Swedish. Which is the transfer source for function words, content words and syntax? Christina Sevdali is a lecturer in linguistics at Ulster University. She received her BA from the University of Crete and her MPhil and PhD from the University of Cambridge. Her main research interests include theoretical syntax and the syntax-morphology interface (especially control theory, case and passivization), language change and multilingualism (especially crosslinguistic influence). Her work has been published in Syntax, Journal of Historical Syntax, and she also has upcoming publications in Language and Lingua. Divya Verma Gogoi, PhD, University of Florida, is a linguist and freelance writer. Her main research interests include crosslinguistic influence in speech perception and training, articulatory and acoustic phonetics, third language acquisition and multilingualism. She has co-authored the paper ‘Thai English: Rhythm and vowels’ in English World-Wide. Femke Swarte is a PhD student at the University of Groningen, the Netherlands. Her research focuses on receptive multilingualism in the Germanic language area. She has also published on the mutual intelligibility between minority languages and the position of minority languages in Europe. She has recently co-edited the special issue Minority Languages in a Multilingual Europe in the Frisian journal Us Wurk. Gessica De Angelis is an Assistant Professor of applied linguistics at Trinity College Dublin, Ireland. Her main research interests are in second/third language acquisition, Crosslinguistic influence and multilingualism. She is the author of Third or Additional Language Acquisition (2007) and the co-editor of two volumes: Teaching and learning in multilingual contexts: sociolinguistic and educational perspectives (with Otwinowska 2014) and New Trends in Crosslinguistic Influence and Multilingualism Research (with Dewaele 2011). She is on the executive board of the International Association of Multilingualism and is former vice president (2009–14) of the association. James D. Harnsberger, PhD, is a linguist and speech scientist specializing in speech acoustics and perception, forensic speech science, and second language learning. He currently serves as dean of the University of New Haven – Qatar. Dr Harnsberger has published in the Journal of the Acoustical Society of America, the Journal of Phonetics, and the Journal of Forensic Sciences, among others. He has also served on scientific advisory boards for the US State Department and the Defense Intelligence Agency, and as an expert witness in over twenty criminal and civil cases. Jan Vanhove is an ‘Oberassistent’ at the Department of Multilingualism in Fribourg, Switzerland. His main field of research to date involves closely related language varieties
Notes on Contributors
xi
(his PhD thesis on receptive multilingualism across the lifespan was published in 2014) and he blogs semi-regularly about statistical issues and research design in applied linguistics and multilingualism research at http://janhove.github.io. Karolina Mieszkowska is a PhD student at the University of Warsaw, Poland. Her main fields of research include multilingual metalinguistic awareness and the link between cognition and language in bilingual child development. She is currently involved in two research projects examining the linguistic and cognitive development of Polish bilingual children living in the United Kingdom. Laura Sánchez is an active member of the GRAL research group. She works as a lecturer at the University of Barcelona and as a postdoc researcher at Stockholms universitet. She is primarily interested in third (or additional) language acquisition. Within this field, her main research focuses on the role of cognitive factors (working memory, attention) and learner-internal factors (age, proficiency) in crosslinguistic influence and the development of writing proficiency. Marijana Kresić is an Associate Professor of linguistics in the Linguistics Department of the University of Zadar in Croatia. Her areas of research include second and third language acquisition, multilingualism, German as a non-native language, language and identity and language comparison. She is currently partner and research associate in the project Experimental Morphosyntax of South Slavic languages (EMSS), which investigates agreement from a crosslinguistic perspective on the basis of psycholinguistic experiments. Megan Devlin completed her PhD at Ulster University. Her research interests are crosslinguistic influence, multilingual language acquisition and the acquisition of syntax (her PhD thesis was a case study of crosslinguistic influence in the English of a trilingual child). She has an upcoming publication in Lingua and has published in edited volumes including Advances in Language Acquisition. Raffaella Folli is senior lecturer in linguistics at Ulster University. Her research interests are theoretical and comparative syntax and multilingual acquisition, with a special focus on the syntax-lexicon interface and the syntax-semantics interface. She has published in journals such as Linguistic Inquiry, Journal of Linguistics, Lingua and Trends in Cognitive Science, as well as in several edited volumes. Raphael Berthele is Professor of multilingualism research at the University of Fribourg and co-founder of the Fribourg-based Institute of Multilingualism. His research interests include receptive multilingualism, the linguistic expression of space and motion and variational linguistics, with a particular eye towards multilingualism (see http://tinyurl.com/l99pudr). Rebecca Dahm is an Associate Professor of applied linguistics at the University of Limoges, France. She currently serves as vice dean of the teacher training college and is co-editor of the special volume Strategies, engagement and cognitive ergonomics
xii
Notes on Contributors
as levers for teaching and learning languages of Les cahiers de l’APLIUT. Her areas of research include second and third language acquisition with a special focus on the added value of pluralistic approaches. Ulrike Jessner is a Professor at the University of Innsbruck and the University of Veszprem. She has published widely in the field of bilingualism and multilingualism with a special focus on the acquisition of English in multilingual contexts. She is the co-author of A dynamic model of multilingualism (with Philip Herdina in 2002) and Linguistic awareness of multilinguals: English as a third language (2006). She is also founding editor of the International Journal of Multilingualism and the book series Trends in Applied Linguistics (with Claire Kramsch from Berkeley University). She is a founding member and former president of the International Association of Multilingualism.
1
The Complex Nature of Crosslinguistic Influence in Multilingual Learning Gessica De Angelis, Ulrike Jessner and Marijana Kresić
Research on crosslinguistic influence (CLI) has evolved considerably over the past twenty years due to a clear surge of interest in phenomena arising from languages other than the L1 (see De Angelis 2007). After decades of attention to L1 influence and its role in the second language learning process, researchers’ interest gradually shifted to the interaction of multiple languages in the mind and how such interactions ultimately influence target language development. The awareness of the increased combination of languages and of the potentially multilingual nature of the learning process has led scholars to conceive CLI as a dynamic phenomenon determined and defined by the amount and type of language knowledge held in the mind, and by the use that multilinguals make of such knowledge in production and comprehension. Such a development is also accompanied by the ongoing symbiosis between research on second/third language learning and bilingualism/multilingualism, and between research on natural and instructed bi/multilingual learning (see Jessner 2015). Along with novel evidence of CLI involving languages other than the L1, multilinguals also repeatedly emerged as individuals who seem to develop better language-learning abilities in comparison with monolingual and bilingual learners. In their definition of multilingual proficiency, Herdina and Jessner (2002) call upon the notion of crosslinguistic interaction to explain contact phenomena from both a linguistic and a cognitive perspective (see also Jessner 2003). They refer to increased metalinguistic awareness and metacognitive skills that multilinguals display and describe how these skills are used in language learning (see also Jessner, Megens and Graus, 2015). Metalinguistic awareness in multilingual learners, or multilingual awareness, has been described as a key factor of multilingual proficiency, learning and use (Jessner 2006, 2008). It is an emergent property of the multilingual mind, which not only exerts influence on other variables in the system but also is influenced by other variables in multilingual development at the same time. In recent years, researchers working on multilingualism also extended discussions on the directionality of language transfer, traditionally conceived to be
2
Crosslinguistic Influence and Crosslinguistic Interaction
unidirectional – from one source to one target language. The notion of bidirectionality came to the forefront with Jarvis’s and Pavlenko’s (2002) work. Focusing on narratives in English (L2) and Russian (L1), the authors offered evidence not only of how the Russian L1 initially influences English L2 narratives, but also of how the continued use of the English L2 begins to influence and shape the use of the Russian L1. We now know that when multiple languages interact with one another, the influence arising from such interaction is often cumulative and non-predictable, as described from the perspective of the dynamic systems theory in Herdina and Jessner (2002). Adding a language to the individual’s repertoire sometimes even leads to an overall increase in achievement in languages previously known by the learners (Griessler 2001; De Angelis and Jessner 2012). Aiming to capture such cumulative effects in syntax, Flynn, Foley and Vinnitskaya (2004) proposed a cumulative enhancement model for language acquisition. They argue that the L1 does not play a privileged role in the language acquisition process as other languages concur in influencing target language development. Similarly, focusing on the Determiner Phrase (DP), Leung (2005) maintained that the L3 initial state does not agree with either the Full Transfer/Full Access Hypothesis (Schwartz and Sprouse 1996) or the Failed Functional Features Hypothesis (FFFH) (Hawkins and Chan 1997), two much-cited hypotheses in the literature on syntax and language transfer. Leung claimed that the L3 initial state differs from the L2 initial state because transfer is not restricted to the L1 and it may involve other languages known to the speaker. Further discussions on syntactic development focused on the L2 status and typological proximity. Drawing from earlier proposals on the so-called foreign-language effect (De Angelis and Selinker 2001) and L2 status (Williams and Hammarberg 1998), Bardel and Falk (2007) argue for a strong role of the second language in the syntactic development of the target language (the L2 status factor model). The role of L2 status seems to be quite pervasive when transfer involves three or more languages, and the same factor has been shown to play a role in phonological development as well (Llama, Cardoso and Collins 2010; Wrembel 2010). L2 status is often discussed in relation to typological proximity. We have known from as early as the 1950s (Weinreich 1953) and 1960s (Vildomec 1963) that typological proximity, also referred to as language/typological closeness or distance, plays a fundamental role in CLI phenomena. Kellerman (1977, 1978, 1983) took this claim a step forward when he proposed the notion of psychotypology, which captures learners’ perception of typological proximity and how these perceptions may influence judgements of transferability between a source and a target language. Typological proximity and psychotypology have been strong topics in CLI research for decades and have been shown to be robust predictors of language transfer (Bouvy 2000; Cenoz 2001; De Angelis and Dewaele 2011; Fouser 2001; Kresić and Gulan 2012; Ringbom 2001). Within syntax, typological proximity remains a dominant topic up to the present day (Rothman 2011; Cabrelli Amaro et al. 2012). The presence of more than one language in the mind also raised the question of the level of language proficiency that is necessary for transfer to occur between non-native languages. Evidence was found that even a couple of years of non-native language instruction may influence the development of a third or additional language
The Complex Nature of Crosslinguistic Influence in Multilingual Learning
3
(De Angelis 2005a, 2007), a finding that shows the need to refine our approach to subject selection procedures in empirical research. For a long time, the objective of CLI researchers was to identify what is transferred from the L1 to the target language and why. The addition of other languages inevitably complicated the picture, as evidence on phenomena that can only exist when multiple languages are involved started to accumulate. From the late 1990s, in order to understand how elements from various background languages were used, combined and transferred to the target language, several researchers focused on the difference between content and function words (De Angelis 2005a; Navés, Miralpeix and Celaya 2005; Ringbom 2001; Williams and Hammarberg 1998). Some studies extended the discussion to morphology (Jarvis and Odlin 2000) and to morphosemantic code mixing (Bouvy 2000). The interaction among non-native languages additionally brought to light the unusual phenomenon of learners losing the ability to assign correct language membership to their own production (De Angelis 2005b). External factors, such as the role of the interlocutor in using and mixing known languages in production, were also examined (Grosjean 1992; Dewaele 2001; Cenoz, Hufeisen and Jessner 2003). Given the strong interest in how languages interact in the mind and the multitude of phenomena identified in recent years, the editors of this volume selected a number of contributions that reflect current research trends in the field of CLI and metalinguistic awareness. As each chapter will show, the dynamics of language interactions is shaping current theoretical and methodological considerations and is building novel foundations for discussions that are original and progressive. The following sections provide a brief description of the breadth and depth of the research questions that are being asked, and discuss why the studies included in this volume represent an element of novelty in the field. The first three chapters (Chapters 2, 3 and 4) present longitudinal case studies involving children or adolescents. Chapter 5 introduces a novel approach to the analysis of CLI using random forests as the main statistical tool. The volume continues with a study on cross-language speech perception, one on linguistic landscapes and one on exposure to different scripts in multilingual Israel. The last three chapters focus more closely on the role of background languages and language similarity in comprehension and production. The first longitudinal study is thus presented in Chapter 2 (Devlin et al. 2012, 2013). The study focuses on the acquisition of the unaccusative/unergative distinction using data collected from a trilingual child at the age of 5;5 to 5;7. Whereas in English the unaccusative/unergative distinction is not overtly marked, in Italian it is overtly encoded through auxiliary selection. The study, therefore, focuses on crosslinguistic influence at the lexical-syntax interface. Based on the literature on language acquisition in bilingual children (e.g. Sorace 2011), it is expected that the child will not show any sign of crosslinguistic influence at the syntax-lexicon interface and that the core verbs according to Sorace’s Auxiliary Selection Hierarchy will be acquired robustly. The parents report that they follow the one-parent-one-language rule, so the child is exposed to Italian from her mother and Scottish Gaelic from her father; English is the language of the community and the shared language of the parents. To test the child’s awareness of the unaccusative/unergative dichotomy, the authors used
4
Crosslinguistic Influence and Crosslinguistic Interaction
two grammaticality judgement tasks. The overall aim of the study is then to help us understand the early development of metalinguistic awareness in multilingual children using a combination of languages that is quite rare in the literature: English, Italian and Scottish Gaelic. The second longitudinal study in Chapter 3 (Dahm) investigates the metacognitive strategies developed through pluralistic approaches at the school level in France. Dahm’s work builds upon research on the development of metalinguistic abilities by Bialystok (1987), Cummins (1991), Jessner (2006) and Moore (2006). In this one-year study, twelve-year-old students who are dominant in French and have reached A1 in English are confronted with Dutch, Finnish and Italian. The aim of the chapter is to offer insights into the wide range of cognitive strategies that are applied in the case of learning an unknown language, and to have a closer look at the transferability of strategies. Thus the research questions are concerned with the students’ strategy use and the transfer of strategies. A pluralistic approach to unknown languages (PAUL) is used to discuss the processes involved in the metasemantic problem-solving activities. Additionally, the implementation of learning strategies that are of a facilitative nature in the decoding of a written source is examined. In Chapter 4 (Sanchez), we move on to another impressive longitudinal study on syntax and language learning at a young age. Sanchez examines syntactic transfer with special reference to the head initial features from the L1 or head final features from the L2 to verb placement (VP) in the English L3 using a combination of languages that includes Romance and Germanic languages. Participants were born and raised in Catalonia and were Spanish-Catalan bilinguals with prior knowledge of German, a language with a distinctive head final feature value for VP. The study is theoretically grounded in syntax research where different proposals have been put forward to account for transfer in L3 acquisition, placing emphasis either on typological proximity (Rothman 2010, 2011), L2 status (Bardel and Falk 2007) or on cumulative effects attributed to the learning process (Flynn, Foley and Vinnitskaya 2004). Sanchez additionally evaluates earlier proposals related to transfer of feature values to an L2, such as the Full Transfer/Full Access Hypothesis (Schwartz and Sprouse 1996), the Minimal Trees Hypothesis (Vainikka and Young-Scholten 1996) and the Weak Parametric Transfer Hypothesis (Eubank 1993, 1994). The guiding question in the study is whether the German L2 with its head final value influences the path of acquisition of VP headedness in the English L3 and, if so, how the influence evolves over time. Data were collected yearly over a four-year period, making this one of the few longitudinal studies on L3 syntax in the literature. Chapter 5 (Vanhove and Berthele) reflects widespread concerns about finding suitable research methods for the study of CLI when multiple languages are involved. Research on CLI and multilingualism is often constrained by research methods that are suitable for L1 influence but are perhaps less ideal when multiple languages are involved. Vanhove and Berthele’s chapter offers a substantial contribution to current methodological debates as the authors apply lesser-used statistical models to identify CLI in cognate guessing tasks. Most importantly, they do not focus exclusively on L1-based interactions but attempt to evaluate the role of all previously known languages, hence the L1 as well as the non-native languages. Vanhove and Berthele
The Complex Nature of Crosslinguistic Influence in Multilingual Learning
5
reanalyse data from two independent datasets using random forests as their primary predictive tool. Their aim is to reach an understanding of the impact of item-related characteristics when guessing the meaning of cognates. Previous literature identified the influence of several variables on cognate guessing. The authors set out to examine some of these, such as formal distance, which includes consonantal and word-initial Levenshtein distances, and cognate frequency. Their study is particularly interesting for two reasons. On the one hand, they explore a novel approach for the identification of crosslinguistic interactions in cognate guessing which is predictive rather than explanatory, showing the potential of lesser-used mathematical models for language data analysis. On the other, they offer evidence that the non-native languages are actively used when multilingual individuals complete cognate guessing tasks. In Chapter 6, Gogoi, Harnsberger and Wiltshire report on a study that contributes to our understanding of the role of metalinguistic awareness in third language acquisition and cross-language speech perception. The authors hypothesize that a perceptual training experiment can lead to a multilingual benefit in adult bilinguals, thereby supporting earlier studies highlighting a positive effect in third language acquisition due to prior language experience. But whereas previous studies concentrated on lexical processing and word learning, this study concentrated on the phonological/ phonetic aspect, as evidenced, for example, in earlier studies by Wrembel (2010, 2011) and Wunder (2011). This study on speech perception in multilingual learning offers a novel research perspective in a hitherto rather unexplored area of multilingualism study. The bilingual learners (Bengali-English; Spanish-English) of Malayam started learning their L2 before the age of eight. The monolingual peers are American-English speakers. Both groups had to learn three speech contrasts, all involving a dental/ alveolar-retroflex distinction. The research questions asked concern the possible facilitative effects that limited training could imply for perceptual assimilation patterns and whether the changes in the patterns are linked to the acquisition of non-native contrasts in bi- and monolingual subjects. The authors concentrated on changes in the assimilation patterns of non-native speech contrasts at both the initial perceptual stage and post-training stage by adopting the assimilation types described in the Perceptual Assimilation Model by Best and Tyler (2007). Chapter 7 (Stavans) provides a fascinating account of how linguistic landscapes are also relevant to discussions on CLI and multilingualism. More specifically, Stavans’ study examines the role of early exposure to different writing systems in multilingual children’s literacy development. She focuses on pre-schoolers who are bilingual but live in the multilingual environment of Israel, where three or more languages are used on a daily basis. The study is well-grounded in the literature on writing development and bilingual literacy, and this contribution provides elements of novelty for both multilingual literacy development and CLI research. Stavans explains how children are aware of the routines associated with reading well before they start to read (Snow and Ninio 1986) and initially learn letters by focusing on some of their properties such as size, shape or length (Ferreiro 1983, 1984; Tolchinsky-Landsmann and Levin 1985). Only a few studies are currently available on the association between writing systems and pre-schoolers. One of these (Tolchinsky et al. 2012) has shown how five-yearold speakers of Spanish, Hebrew and Cantonese develop metalinguistic awareness and
6
Crosslinguistic Influence and Crosslinguistic Interaction
visual discrimination abilities that seem to be associated with the writing systems of their own culture. Stavans goes a step forward, with the intention of showing how multilingual pre-schoolers have the additional advantage of being able to transfer knowledge of different scripts between languages and therefore use their background language knowledge to their benefit. Focusing on multilingual and monolingual children’s judgements of readability and the reasons they provide for their judgements, she asks whether they differ in the way they judge alphabetic representations (single, other or mixed) and what the reasons are for their choices. The study convincingly shows the close association between multilingual landscapes and the development of multilingual literacy in pre-school children. In Chapter 8, Swarte, Schüppert and Gooskens investigate the role of the nonnative language German in processing the unknown, but related, language Danish. The authors also explore the question of whether knowledge of a non-native language facilitates receptive multilingualism, a term that refers to communication on the basis of the active use only of the speaker’s L1 and the attempt to understand the interlocutor’s communicative messages in the respective unknown language. The study is situated in the general context of research on CLI in multilingual learning (cf. De Angelis and Dewaele 2011; Aronin and Hufeisen 2009) with a particular focus on receptive multilingualism, which has been particularly investigated for the Scandinavian language context (e.g. Golinski 2007; Gooskens 2007) and with reference to various combinations of Germanic languages (Berthele and Wittlin 2013; Swarte and Hilton 2013). In the context of studies focusing on linguistic and extra-linguistic factors that are relevant for the phenomenon of receptive multilingualism (e.g. Gooskens 2006, 2007), including the role of knowledge of an L2 (Gibson and Hufeisen 2003) in understanding an L3 (Hufeisen and Marx 2007), the study presented in this volume is one of the first and few studies that explore the role of the non-native language German in processing the unknown language Danish. It also represents a contribution to research on the understanding of cognates (Berthele 2011) and non-cognates in unknown languages (Van Bezooijen, Gooskens and Kürschner 2012). Swarte, Schüppert and Gooskens conducted two experiments, one of them using written stimuli and the other one using spoken stimuli, in order to explore the research question of whether knowledge of German as a non-native language is helpful to L1 Dutch speakers in understanding unknown Danish words. They also investigate whether there is evidence for the effect of a foreign language mode (Selinker and Baumgartner-Cohen 1995) in their data. Furthermore, the authors are interested in the possible influence of knowledge of the semantic context or field on the recognition of cognates. The empirical study presented by Mieszkowska and Otwinowska in Chapter 9 aims to investigate mechanisms used by multilinguals when processing an unknown language that is similar to the languages they already know. The authors are also interested in finding out which of the languages from the multilingual’s repertoire are used as sources for CLI and which factors might be responsible for the selection of the respective source language, taking into account different language combinations that might have an impact on the translation of a text in an unknown language. This study clearly illustrates the shift from analysing mere transfer or CLI, mainly from L1 to L2, towards investigating the complex phenomenon of multilingual language acquisition,
The Complex Nature of Crosslinguistic Influence in Multilingual Learning
7
a process that is characterized by the complex and dynamic interplay of the various languages of the learner (Cook 1991, 1992; Herdina and Jessner 2002), by a raised metalinguistic awareness and multidirectional CLI (De Angelis 2007), with various languages acting as source languages. Mieszkowska and Otwinowska’s work represents a novel contribution to the analysis of the mechanisms underlying the selection and use of multiple source languages in decoding a text in an unknown language, with a particular focus on possible factors influencing the activation of previously learnt languages by multilingual, advanced learners of English with Germanic and/or Romance L3-Ln languages. In Chapter 10, Lindqvist reports on a study in which she investigated whether subjects’ perceptions concerning language relatedness represent a factor that is relevant for the choice of the source of CLI in L3 written text production. Data were gathered from learners with L1 Swedish, L2 English and L3 French with the help of a questionnaire exploring psychotypology. Subsequently, lexical and grammatical CLI was analysed in a corpus of written story retelling tasks in French (L3) produced by the same subjects. Earlier research shows that languages that are typologically closer to the target language tend to serve as source languages for CLI (Cenoz 2001; De Angelis 2005a; De Angelis and Selinker 2001; Lindqvist and Bardel 2013). There is also evidence that in instances of Romance languages as L3, other Romance languages tend to serve as sources of CLI (De Angelis 2005a; De Angelis and Selinker 2001; Lindqvist 2009; Lindqvist and Bardel 2013), rather than other languages that are typologically more distant. In the study presented here, Lindquist explores questions related to whether there is an association between the source of CLI (L1 or L2) and the subjects’ psychotypology with respect to Swedish (=L1), English (=L2) and French (=L3). In line with the psychotypology factor, it was assumed that the language perceived to be more closely related to L3 French would serve as a source of transfer in areas of grammar and vocabulary. This brief introduction shows the breadth and variety of topics, questions and language combinations examined in each of these chapters. The last chapter (Chapter 11) is co-written by the three editors and offers an outline of major findings and the overall theoretical and methodological contributions presented in the volume with respect to language interactions, CLI and metalinguistic awareness.
References Aronin, L. and B. Hufeisen, eds (2009), The Exploration of Multilingualism: Development of Research on L3, Multilingualism and Multiple Language Acquisition, Amsterdam and Philadelphia: John Benjamins. Bardel, C. and Y. Falk (2007), ‘The role of the second language in third language acquisition. The case of Germanic syntax’, Second Language Research, 23 (4): 459–84. Berthele, R. (2011), ‘On abduction in receptive multilingualism. Evidence from cognate guessing tasks’, Applied Linguistics Review, 2: 191–220. Berthele, R. and G. Wittlin (2013), ‘Receptive multilingualism in the Swiss army’, International Journal of Multilingualism, 10 (2): 181–95.
8
Crosslinguistic Influence and Crosslinguistic Interaction
Best, C. T. and M. D. Tyler (2007), ‘Nonnative and second language speech perception: Commonalities and complementarities’, in O.-S. Bohn and M. J. Munro (eds), Language Experience in Second Language Speech Learning, 13–45, Amsterdam: John Benjamins. Bialystok, E. (1987), ‘Words as things: Development of word concept by bilingual children’, Studies in Second Language Acquisition, 9: 133–40. Bouvy, C. (2000), ‘Towards the construction of a theory of cross-linguistic transfer’, in J. Cenoz and U. Jessner (eds), English in Europe: The Acquisition of a Third Language, 143–56, Clevedon: Multilingual Matters. Cabrelli Amaro, J., S. Flynn and J. Rothman, eds (2012), Third Language Acquisition in Adulthood, Amsterdam: John Benjamins. Cenoz, J. (2001), ‘The effect of linguistic distance, L2 status and age on cross-linguistic influence in third language acquisition’, in J. Cenoz, B. Hufeisen and U. Jessner (eds), Cross-linguistic Influence in Third Language Acquisition: Psycholinguistic Perspectives, 8–20, Clevedon: Multilingual Matters. Cenoz, J., B. Hufeisen and U. Jessner (2003), ‘Why investigate the multilingual lexicon?’, in J. Cenoz, B. Hufeisen and U. Jessner (eds), The Multilingual Lexicon, 1–9, Dordrecht: Kluwer Academic Publishers. Cook, V. J. (1991), ‘The development of multi-competence’, in Proceedings of the 5th Symposium on the Description and/or Comparison of English and Greek, March 27–29 1991, 394–404, Thessaloniki: University of Thessaloniki. Cook, V. J. (1992), ‘Evidence for multicompetence’, Language Learning, 42: 557–91. Cummins, J. (1991), ‘Interdependence of first- and second language proficiency’, in E. Bialystok (ed.), Language Processing in Bilingual Children, 70–89, Cambridge: Cambridge University Press. De Angelis, G. (2005a), ‘Interlanguage transfer of function words’, Language Learning, 55 (3): 379–414. De Angelis, G. (2005b), ‘Multilingualism and non-native lexical transfer: An identification problem’, International Journal of Multilingualism, 2: 1–25. De Angelis, G. (2007), Third or Additional Language Acquisition, Clevedon, UK: Multilingual Matters. De Angelis, G. and U. Jessner (2012), ‘Writing across languages in a bilingual context: A dynamic systems theory approach’, in R. M. Manchòn (ed), L2 Writing Development: Multiple Perspectives, 47–68, Trends in Applied Linguistics Series. Boston: Mouton de Gruyter. De Angelis, G. and J.-M. Dewaele, eds (2011), New Trends in Crosslinguistic Influence and Multilingualism Research, Bristol: Multilingual Matters. De Angelis, G. and L. Selinker (2001), ‘Interlanguage transfer and competing linguistic systems in the multilingual mind’, in J. Cenoz, B. Hufeisen and U. Jessner (eds), Crosslinguistic Influence in Third Language Acquisition: Psycholinguistic Perspectives, 42–58, Clevedon, UK: Multilingual Matters. Devlin, M., R. Folli, A. Henry and C. Sevdali (2012), ‘Clitic dislocation in the absence of clitics: A study in trilingual acquisition’, University of Pennsylvania Working Papers in Linguistics, 18 (1): 40–50. Devlin, M., R. Folli, A. Henry and C. Sevdali (2013), ‘Vulnerable domains and crosslinguistic influence: The view from trilingual acquisition’, in S. Stavrakaki, M. Lalioti and X. Konstantinopoulo (eds), Advances in Language Acquisition, 309–19, Cambridge: Cambridge Scholars Publishing. Dewaele, J.-M. (2001), ‘Activation or inhibition? The interaction of L1, L2 and L3 on the language mode continuum’, in J. Cenoz, B. Hufeisen and U. Jessner (eds),
The Complex Nature of Crosslinguistic Influence in Multilingual Learning
9
Cross-linguistic Influence in Third Language Acquisition: Psycholinguistic Perspectives, 69–89, Clevedon: Multilingual Matters. Eubank, L. (1993/1994), ‘On the transfer of parametric values in L2 development’, Language Acquisition, 3: 185–208. Ferreiro, E. (1983), ‘The development of literacy: A complex psychological problem’, in F. Coulmas and K. Ehlich (eds), Writing in Focus, 277–90, Berlin: Mouton. Ferreiro, E. (1984), ‘The underlying logic of literacy development’, in H. Goelman, A. Oberg and F. Smith (eds), Awakening to Literacy, 154–73, Exeter, NH: Heinemann. Flynn, S., C. Foley and I. Vinnitskaya (2004), ‘The cumulative-enhancement model of language acquisition: Comparing adults’ and children’s patters of development in first, second and third language acquisition of relative clauses’, International Journal of Multilingualism, 1 (1): 3–16. Fouser, R. J. (2001), ‘Too close for comfort? Sociolinguistic transfer from Japanese into Korean as an L 3’, in J. Cenoz, B. Hufeisen and U. Jessner (eds), Cross-linguistic Influence in Third Language Acquisition: Psycholinguistic Perspectives, 149–69, Clevedon, UK: Multilingual Matters. Gibson, M. and B. Hufeisen (2003), ‘Investigating the role of prior foreign language knowledge’, in J. Cenoz, B. Hufeisen and U. Jessner (eds), The Multilingual Lexicon, 87–102, Dordrecht, Boston and London: Kluwer Academic Publishers. Golinski, B. (2007), Kommunikationsstrategien in interskandinavischen Diskursen. [Communication strategies in inter-Scandinavian discourse], Hamburg: Kovač. Gooskens, C. (2006), ‘Linguistic and extra-linguistic predictors of Inter-Scandinavian intelligibility’, in J. van de Weijer and B. Los (eds), Linguistics in the Netherlands 23, 101–13, Amsterdam: John Benjamins. Gooskens, C. (2007), ‘The contribution of linguistic factors to the intelligibility of closely related languages’, Journal of Multilingual and Multicultural Development, 28 (6): 445–67. Griessler, M. (2001), ‘The effect of third language learning on second language proficiency: An Austrian example’, International Journal of Bilingual Education and Bilingualism, 4 (1): 50–60. Grosjean, F. (1992), ‘Another view of bilingualism’, in R. J. Harris (ed.), Cognitive Processing in Bilinguals, 51–62, Amsterdam: North Holland. Hawkins, R. and Y. C. Chan (1997), ‘The partial availability of universal grammar in second language acquisition: The “failed functional features hypothesis”’, Second Language Research, 13: 187–226. Herdina, P. and U. Jessner (2002), A Dynamic Model of Multilingualism: Perspectives of Change in Psycholinguistics, Clevedon, UK: Multilingual Matters. Hufeisen, B. and N. Marx (2007), ‘How can DaFnE and EuroComGerm contribute to the concept of receptive multilingualism? Theoretical and practical considerations’, in J. D. Ten Thije and L. Zeevaert (eds), Receptive Multilingualism, 307–21, Amsterdam and Philadelphia: John Benjamins. Jarvis, S. and T. Odlin (2000), ‘Morphological type, spatial reference, and language transfer’, Studies in Second Language Acquisition, 22: 535–56. Jarvis, S. and A. Pavlenko (2002), ‘Bidirectional Transfer’, Applied Linguistics, 23 (2): 190–214. Jessner, U. (2003), ‘The nature of crosslinguistic interaction in multilingual systems’, in J. Cenoz, B. Hufeisen and U. Jessner (eds), The Multilingual Lexicon, 45–55, Dordrecht: Kluwer. Jessner, U. (2006), Linguistic Awareness in Multilinguals: English as a Third Language, Edinburgh: Edinburgh University Press.
10
Crosslinguistic Influence and Crosslinguistic Interaction
Jessner, U. (2008), ‘A DST-model of multilingualism and the role of metalinguistic awareness’, Second Language Development as a Dynamic Process, Special Issue of Modern Language Journal, 92 (2): 270–83. Jessner, U. (2015), ‘Multicompetence approaches to language proficiency development in multilingual education’, in O. Garcia and A. Lin (eds), Encyclopedia of Language and Education 5, Bilingual Programs, 591–104, New York: Springer. Jessner, U., M. Megens and S. Graus (2015), ‘Crosslinguistic influence in third language acquisition’, in R. Alonso (ed.), Crosslinguistic Influence in Second Language Acquisition, Clevedon: Multilingual Matters. Kellerman, E. (1977), ‘Towards a characterization of the strategy of transfer in second language learning’, Interlanguage Studies Bulletin, 2: 58–145. Kellerman, E. (1978), ‘Giving learners a break: native language intuitions as a source of prediction about transferability’, Working Papers on Bilingualism, 15: 59–92. Kellerman, E. (1983), ‘Now you see it, now you don’t’, in S. Gass and L. Selinker (eds), Language Transfer in Language Learning, 112–34, Rowley, MA: Newbury House. Kresić, M. and T. Gulan (2012), ‘Interlingual identifications and assessment of similarities between L1, L2, and L3: Croatian learners’ use of modal particles and equivalent modal elements’, in D. Gabryś-Barker (ed.), Cross-linguistic Influences in Multilingual Language Acquisition, 63–80, Berlin and Heidelberg: Springer-Verlag. Leung, Y. (2005), ‘L2 vs. L3 initial state: a comparative study of the acquisition of French DPs by Vietnamese monolinguals and Cantonese-English bilinguals’, Bilingualism: Language and Cognition, 8 (1): 39–61. Lindqvist, C. (2009), ‘The use of the L1 and the L2 in French L3: Examining crosslinguistic lexemes in multilingual learners’ oral production’, International Journal of Multilingualism, 6: 1–17. Lindqvist, C. and C. Bardel (2013), ‘Exploring the impact of the proficiency and typology factors: Two cases of multilingual learners’ L3 learning’, in M. Pawlak and L. Aronin (eds), Essential Topics in Applied Linguistics and Multilingualism, Studies in honor of David Singleton, 253–66, Switzerland: Springer International Publishing. Llama, R., W. Cardoso and L. Collins (2010), ‘The influence of language distance and language status on the acquisition of L3 phonology’, International Journal of Multilingualism, 7 (1): 39–57. Moore, D. (2006), ‘Plurilingualism and strategic competence in context’, International Journal of Multilingualism, 3 (2): 125–38. Navés, T., I. Miralpeix and M. L. Celaya (2005), ‘Who transfers more … and what? Crosslinguistic influence in relation to school grade and language dominance in EFL’, International Journal of Multilingualism, 2: 113–34. Ringbom, H. (2001), ‘Lexical transfer in L3-production’, in J. Cenoz, B. Hufeisen and U. Jessner (eds), Crosslinguistic Influence in Third Language Acquisition: Psycholinguistic Perspectives, 59–68, Clevedon, UK: Multilingual Matters. Rothman, J. (2011), ‘L3 syntactic transfer selectivity and typological determinacy: The typological primacy model’, Second Language Research, 27 (1): 107–27, doi: 10.1177/0267658310386439. Schwartz, B. D. and A. Sprouse (1996), ‘L2 cognitive states and the full transfer/full access model’, Second Language Research, 12 (1): 40–72. Selinker, L. and B. Baumgartner-Cohen (1995), ‘Multiple language acquisition: “Damn it, why can’t I keep these two languages apart?”’, Multilingualism and Language Learning, 8 (2): 115–23.
The Complex Nature of Crosslinguistic Influence in Multilingual Learning
11
Snow, C. E. and A. Ninio (1986), ‘The contracts of literacy: What children learn from learning to read books’, in W. Teale and E. Sulzby (eds), Emergent Literacy: Writing and Reading, 116–38, Norwood, NJ: Ablex Publishing Corporation. Sorace, A. (2011), ‘Pinning down the concept of “Interface” in bilingualism’, Linguistic Approaches to Bilingualism, 1 (1): 1–33. Swarte, F. and N. H. Hilton (2013), ‘Mutual intelligibility between speakers of North and West Frisian’, in C. Gooskens and R. van Bezooijen (eds), Phonetics in Europe: Perception and Production, 281–302, Frankfurt am Main: Peter Lang. Tolchinsky, L., I. Levin, D. Aram and C. McBride-Chang (2012), ‘Building literacy in alphabetic, abjad and morphosyllabic systems’, Reading and Writing, 25 (7): 1573–98. Tolchinsky-Landsmann, L. and I. Levin (1985), ‘Writing in preschoolers: An age-related analysis’, Applied Psycholinguistics, 6: 319–39. Vainikka, A. and M. Young-Scholten (1996), ‘Gradual development of L2 phrase structure’, Second Language Research, 12 (1), 7–39. Van Bezooijen, R., C. Gooskens and S. Kürschner (2012), ‘Deens is makkelijker voor Friezen dan voor Nederlanders – feit of fabel?’, in P. Boersma, G. Jensma and R. Salverda (eds), Philologia Frisica, 286–98, Leeuwarden: Fryske Akademy. Vildomec, V. (1963), Multilingualism, Netherlands: A. W. Sythoff-Leyden. Weinreich, U. (1953), Languages in Contact, New York: Publications of the Linguistic circle of New York. No.1. Williams, S. and B. Hammarberg (1998), ‘Language switches in L3 production: implications for a polyglot speaking model’, Applied Linguistics, 19 (3): 295–333. Wrembel, M. (2010), ‘L2-accented speech in L3 production’, International Journal of Multilingualism, 7 (1): 75–90. Wrembel, M. (2011), ‘Cross-linguistic influence in third language acquisition of VOT’, in W.-S. Lee and E. Zee (eds), Proceedings of the 17th International Congress of Phonetic Sciences. August 2011. Hong Kong, 17–21, Hong Kong: City University of Hong Kong. Wunder, E. M. (2011), ‘Crosslinguistic influence in multilingual language acquisition: Phonology in third or additional language acquisition’, in G. De Angelis and J. M. Dewaele (eds), New Trends in Crosslinguistic Influence and Multilingualism Research, 105–28, Bristol: Multilingual Matters.
2
The Lexicon-Syntax Interface in a 3L1 context: An Experimental Investigation of the Unaccusative/Unergative Distinction in English and Italian Megan Devlin, Raffaella Folli and Christina Sevdali
1 Introduction This chapter investigates the acquisition of the unaccusative/unergative distinction in a trilingual (3L1) child. This study is part of Project S (Devlin et al. 2013; Devlin 2015), an extensive longitudinal case study of the simultaneous trilingual acquisition of English, Italian and Scottish Gaelic from birth. Project S investigates the effects that acquiring three languages simultaneously from birth has on the child’s English, with a particular focus on crosslinguistic influence. Previous studies on the spontaneous production of this child (S) have revealed crosslinguistic influence in the production of complex determiner phrases (DPs) (Devlin et al. 2013) and in the production of target-deviant it-doubling constructions (Devlin et al. 2012). This chapter studies the child’s acquisition of the unaccusative/unergative distinction through two grammaticality judgement tasks: one in her English and one in her Italian1. At the time of testing, the child was aged between five years, four months and five years, seven months. In the literature, there is evidence that children acquiring certain languages go through a phase where they have problems with the distinction: for example, Babyonyshev et al. (2001) show that monolingual Russian children aged three and six years six months tend to treat unaccusatives like unergatives. In contrast, Snyder, Hyams and Chrisma (1995), in a study of monolingual Italian children aged two, show that children make almost no mistakes with auxiliary selection and that by three they can distinguish between the two classes of intransitive verbs. Similarly, Costa and Friedmann (2012) report that the Hebrew and European Portuguese children in their study can distinguish unaccusatives from unergatives at an early age, as evidenced by the different word orders adopted for the different verb types. Lorusso, Caprin and Guasti
14
Crosslinguistic Influence and Crosslinguistic Interaction
(2005) also report that Italian children are sensitive to the unaccusative/unergative dichotomy as they use subjects depending on the verb class. They report that children use more post-verbal subjects with unaccusative verbs than both unergative and transitive verbs, which show a preference for pre-verbal subjects. Overall, the literature suggests that, crosslinguistically, we see differences in how early this distinction is acquired and that this is facilitated in languages where the unaccusative/unergative distinction is more transparently encoded in auxiliary selection or word order. The subject of this study is the acquisition of English, a language where the unaccusative/unergative distinction is not overtly marked and one where the distinction is overtly encoded through auxiliary selection. No previous study has tried to see whether the acquisition of the distinction in a speaker of English and Italian shows evidence of crosslinguistic influence, and this is one of the questions this study wants to address. The unaccusative/unergative distinction is crucially related to questions regarding the lexicon-syntax interface. In this field of study, the analysis of verb classes and alternations has been particularly important, as by identifying common syntactic properties belonging to verbs with common semantic characteristics, one can provide support for the hypothesis that some generalizations between the two are indeed possible. In particular, the fundamental issue for the lexicon-syntax interface seems to be whether unaccusativity is semantically determined, and only in consequence of that syntactically represented (Van Valin 1990; Dowty 1990; Levin and Rappaport-Hovav 2000); or the other way round (Rosen 1984; Borer and Grodzinsky 1986; Hoekstra and Mulder 1990; Levin and Rappaport-Hovav 1995; among others). More generally, one important question about argument structure in relation to the lexicon-syntax interface is how the speaker knows which verb takes which argument, and, in relation to the unaccusative/unergative distinction in Italian, which verb takes which auxiliary. Is this due to the lexical meaning of the verb, or is it determined by the syntactic structure in which the verb can be inserted? These questions remain the centre of debate in the field of acquisition; however, there are two prominent hypotheses: the semantic bootstrapping hypothesis and the syntactic bootstrapping hypothesis. The semantic bootstrapping hypothesis (Dowty 1990; Levin and Rappaport-Hovav 1995; among others) proposes that the lexicon is a sub-module of the language faculty, whereby semantic rules and information on semantic roles are made available there. By way of a linking theory, then, the information on different semantic roles is projected from the lexicon and mapped into different places in the syntactic structure. This is in contrast to the syntactic bootstrapping hypothesis, whereby the lexicon is not a sub-module of language; rather, it is an encyclopaedic system of knowledge. With the syntactic bootstrapping hypothesis, the syntactic structure can be freely constructed, while the encyclopaedic knowledge from the lexicon dictates which lexical items can be inserted into which structure. These hypotheses may be further tested when studied in relation to a child acquiring two languages where the unaccusative/unergative distinction is encoded differently. The study of multilingual language development is crucially centred on questions regarding the interaction of the languages as they develop. This study takes as a starting point the view that the interaction between languages can be viewed as crosslinguistic influence, following Serratrice, Sorace and Paoli (2004), Sorace and Serratrice (2009),
The Lexicon-Syntax Interface in a 3L1 context
15
Sorace (2011), Döpke (1998), Müller and Hulk (2001), among many others. The challenge for researchers trying to model this phenomenon is to understand where crosslinguistic influence occurs and why it occurs in some domains rather than others. Additionally, researchers have to try to account for why crosslinguistic influence is not always consistently found across bilingual or multilingual populations. One popular theory of crosslinguistic influence is the vulnerable domains hypothesis. Vulnerable domains in acquisition are those domains that are reported to take children longer to acquire. Müller and Hulk (2000, 2001) follow Platzack (2001) in proposing that the C-domain is the grammatical domain that is vulnerable in acquisition. Its vulnerability lies in the fact that it is the domain where syntax interfaces with other cognitive domains. Müller and Hulk (2001) propose that there are two conditions for crosslinguistic influence: first that there is a structural overlap, and second that the grammatical element is a phenomenon of the C-domain. Following this proposal, we would not expect that crosslinguistic influence would have an effect on the encoding of unaccusativity in S’s language, as there is clearly no overlap in the encoding of unaccusativity in English and Italian, and the phenomena in question are not related to the C-domain. Neither condition is satisfied in this instance, and yet, target deviance is still observed. Müller (2003) observes that crosslinguistic influence does not occur in every domain and that only some grammatical domains are vulnerable to crosslinguistic influence. Sorace’s (2000, 2004, 2005, among others) research in the field of bilingualism has focused on the task of accounting for the selective crosslinguistic influence observed in different bilingual populations. Research spanning over a decade from Sorace et al. (Sorace 2004, 2005; Sorace and Filiaci 2006; Sorace and Serratrice 2009; Serratrice, Sorace and Paoli 2004; Tsimpli and Sorace 2006; among others) examines the existence of crosslinguistic influence in bilingual populations, including L2 adults, older L2 children and simultaneous bilinguals, across various language combinations. Numerous studies from Sorace et al. and from researchers including (Tsimpli and Sorace 2006, 2008; Sorace and Serratrice 2009; Serratrice, Sorace and Paoli 2004; Belletti, Bennati and Sorace 2007; Filiaci et al. 2013; among many others) show that the notion of interface is relevant to understanding the nature of this phenomenon and, in particular, that not all interfaces are equally problematic in acquisition. While this hypothesis was originally formulated to account for non-convergence in adult L2 acquisition, Sorace (2011) has extended it to account for optionality in bilingual first language acquisition (2L1) and L1 attrition. Moreover, Sorace (2011) states that optionality and target deviance are observed in the same structures across the three language groups. In line with Sorace’s hypothesis, interfaces that occur between modules of language are generally referred to as internal interfaces, such as the syntaxsemantics interface, while interfaces generally referred to as external are those that involve language and non-linguistic systems, such as the syntax-discourse interface (for review, see Slabakova, Kempchinsky and Rothman (2012)). The interface hypothesis put forward by Sorace and colleagues is that the external interfaces are the most vulnerable in multilingual acquisition, while internal interfaces such as the syntax-semantic interface and the syntax-lexicon interface should not be affected. While Sorace and colleagues present extensive evidence to support their claim regarding vulnerability at external interfaces, research by Slabakova and Ivanov (2011), Slabakova, Kempchinsky
16
Crosslinguistic Influence and Crosslinguistic Interaction
and Rothman (2012), Rothman (2009), among others, present instances whereby L2 children are acquiring external interface phenomena in keeping with native speakers. Similar findings are reported by Montrul and Ionin (2012) in relation to phenomena at the internal syntax-semantic interface. Montrul and Ionin (2012) examine the use of definite articles in L2 Spanish speakers and Spanish heritage speakers. They report that both L2 Spanish speakers and Spanish heritage speakers make similar mistakes with regard to determiners in generic contexts; that is, they are more likely to accept ungrammatical bare noun phrases (NPs) in Spanish in the acceptability judgement task and the truth-value judgement task. Montrul and Ionin (2012) argue that crosslinguistic influence from the English of the two Spanish-speaking groups causes the target deviant judgements in the interpretation of definite articles in generic contexts in Spanish. This study provides evidence against the invulnerability of the internal interfaces, as crosslinguistic influence occurred in phenomena of the syntax-semantics interface. This chapter will further test the invulnerability of the internal interfaces. This study of trilingual acquisition also allows for the examination of the relationship between language dominance and target deviance. The natural assumption would be that crosslinguistic influence sees the stronger (or more dominant) language affecting the weaker one. While Yip and Matthews (2000), Döpke (2000), among others, argue that crosslinguistic influence does occur in the direction of the strong language to the weak language, Gawlitzek-Maiwald and Tracy (1996), Müller (1998), Müller and Hulk (2001), among others, argue that this is not the case. Given that throughout the case study of S’s language development, English is consistently considered the child’s most dominant language, one would expect that any crosslinguistic influence would go in the direction of English to Italian, as Italian is the weaker language. The final ingredient is related particularly to Italian and to the acquisition of auxiliaries. Sorace (2000, 2004) proposes the Auxiliary Selection Hierarchy (ASH) as a model of acquisition of auxiliary selection. ASH can account for the variable behaviour that has been observed crosslinguistically in auxiliary selection. ASH is a gradient scale of intransitive verbs distinguished by their lexical semantics. The two main semantic components that differentiate the verbs are agentivity and telicity, with the core unergative verbs being the most strongly agentive verbs and the core unaccusative verbs being the most strongly telic. Sorace (1993a, 2000, 2004) states that the core verbs are the verbs that are strictly unaccusative or unergative and show the least flexibility in relation to auxiliary selection, while the periphery verbs are more flexible in relation to auxiliary selection. These verbs also show the most variability. Sorace (2000, 2004) proposes that the verbs classified as core verbs in the ASH are the first verbs to be acquired and are shown to be least problematic in L2 acquisition, while the verbs classified as periphery verbs are acquired later and are more problematic in the acquisition process. This proposal is important because it allows us to hypothesize that if any problems were to arise in the grammaticality judgement tasks in Italian, we would expect that they would occur with periphery verbs. In this chapter, we will test the two hypotheses discussed above in the acquisition of the unaccusative/unergative distinction in English and Italian. Based on the literature, we expect a That the child will not show any sign of crosslinguistic influence at the syntaxlexicon interface
The Lexicon-Syntax Interface in a 3L1 context
17
b That core verbs according to Sorace’s ASH will be acquired robustly and unproblematically, while problems, if any, will arise with periphery verbs. This chapter is organized as follows: in Section 2 we discuss the various ways in which unaccusativity is encoded in English and Italian; in Section 3 we present previous findings on the acquisition of unaccusativity (in monolingual and bilingual acquisition); in Section 4 we present the methodology and results of the current study; and in Section 5 we present a discussion of the results and conclusion.
2 The unaccusative/unergative distinction The unaccusativity hypothesis, as proposed by Perlmutter (1978) and later by Burzio (1981), states that not all one-place predicates have the same underlying structure; instead, there are two types of intransitives: unaccusatives and unergatives. These two verb classes differ in both their syntactic structure and relatedly in their lexical semantics. Although unergative and unaccusative constructions are similar in their surface structure in English, the two constructions differ in their underlying syntax. With unergative verbs, subjects are base-generated in the subject position where they are assigned the agent theta-role. This contrasts with unaccusative verbs, where the subject is initially merged as the object of the verb and thus receives a theme theta-role, before moving to the subject position where it receives nominative case.
1. a. The cat jumped.
b. The vasei broke ti.
Semantically, the two intransitive verb types differ in that the subject of unaccusative verbs is non-agentive; that is, it does not actively initiate the action of the verb, whereas the subject of an unergative verb is typically agentive. Crucially, the distinction of monadic verbs in terms of the base-generated position of their single arguments goes hand in hand with a classification of intransitives on the basis of the aspectual properties of the event they denote. Accordingly, unaccusative verbs are inherently telic, in that the event denoted by the verb has a natural endpoint (2):
2. a. John arrived. b. The fly died.
Unaccusatives
This contrasts with unergatives, which are typically atelic and denote events that do not entail a natural endpoint:
3. a. Mary sang. b. Cathy laughed.
Unergatives
As is noted widely in the literature (Baker 1988; Alexiadou, Anagnostopoulou and Martin 2004; Ramchand 2008; among many others), there is crosslinguistic variation in the way in which unaccusatives and unergatives are encoded in language, and many diagnostics for the differentiation of intransitives have been proposed in
18
Crosslinguistic Influence and Crosslinguistic Interaction
previous research for many different languages by Perlmutter (1978), Burzio (1981), Levin and Rappaport-Hovav (1995), Alexiadou, Anagnostopoulou and Martin (2004), among others.
2.1 English In English, the unaccusative/unergative distinction can be observed in resultative constructions and in past participle constructions. In particular, resultatives can be formed with unaccusative verbs but not with unergative verbs with a bare XP, as in (4) and (5)2:
4. a. The door slammed shut.
Unaccusatives
5. a. *John ran tired.
Unergatives
b. The water froze solid. b. *Mary laughed sore.
In line with Simpson’s (1983) direct object generalization, a resultative phrase can only be predicated on an object. With this in mind, we expect to be able to form resultatives with verbs whose argument is underlyingly an object, that is, unaccusative verbs. Since the argument of unergatives is a surface subject, resultatives should not be available with these verbs. The past participle construction is also a diagnostic of the unaccusative/unergative distinction, as English past participles are grammatical with unaccusative verbs (6) but ungrammatical with unergative verbs as exemplified in (7):
6. a. The broken window. b. The fallen tree. c. The hidden treasure.
7. a. *The waved policeman. b. *The shouted man. c. *The slept baby.
The modification of an object by a prenominal perfect participle was proposed as a diagnostic of the unaccusative/unergative distinction by Perlmutter (1978). The perfect participle in English provides evidence in support of Perlmutter’s (1978) original hypothesis that intransitives can be divided into two classes, as in constructions with the modifying perfect participle it is observed that the subjects of transitive verbs and the objects of unaccusatives share similarities while the subjects of transitive verbs and unergative verbs exhibit similar behaviour. These two constructions are the ones we chose for the English tasks.
2.2 Italian The unaccusative/unergative diagnostic selected for Italian is auxiliary selection in passato prossimo tense. The passato prossimo is a periphrastic tense form consisting
The Lexicon-Syntax Interface in a 3L1 context
19
of the present inflected form of the auxiliary and the past participle derived form of the verb. The unaccusative/unergative distinction is observed in the selection of the auxiliary, with unaccusative verbs selecting the auxiliary essere (to be) and unergative verbs selecting the auxiliary avere (to have), as exemplified in (8) and (9), respectively:
8. a. Il treno è arrivato.
Unaccusative
9. a. Maria ha cantato.
Unergative
The train is arrived. ‘The train arrived.’ Mary has sung ‘Mary sang.’
While auxiliary selection is widely recognized and used as a diagnostic of unaccusative/ unergative distinction, it has been acknowledged that there is variation within languages with respect to the selection of auxiliaries. For instance, while there are verbs in Italian that can consistently select the same auxiliary in all contexts, there are others that can select either auxiliary in different syntactic contexts (Levin and RappaportHovav 1995; Borer 1994; Sorace 2000, 2004; among others). Such verbs are referred to as variable behaviour verbs and are exemplified in (10):
10. a. Gianni ha corso per un’ora. b.
Gianni has ran for an hour. Gianni è corso in casa. Gianni is ran in house. ‘Gianni has run into the house.’
In (10)a, run is an unergative and describes an agentive activity that lasts for a period of time, while in (10)b it selects for essere and denotes an unaccusative change-of-location event. This variability is taken to be problematic for approaches to argument projections, assuming that the lexical meaning is, to a large extent, crucial in determining class membership.
3 The unaccusative/unergative distinction in acquisition As we briefly discussed above, it is widely reported that children appear to be capable of making the distinction between unaccusatives and unergatives at an early age (Van Hout, Randall and Weissenborn 1992; Snyder, Hyams and Crisma 1995; Babyonyshev et al. 2001; Sano, Endo and Yamakoshi 2001; Lorusso, Caprin and Guasti 2005; Lorusso 2007; among others). Despite this general view, it seems that children can experience varying levels of difficulty with the production and comprehension of constructions that are diagnostic of the unaccusative/unergative distinction. Snyder, Hyams and Crisma (1995) observed the spontaneous production of unaccusatives in a French child and three Italian children aged between one year, seven months and two years, eleven months. The study reported that the Italian children made relatively few target deviant productions in relation to auxiliary selection: only
20
Crosslinguistic Influence and Crosslinguistic Interaction
two were observed, and the same child produced both target deviant instances. These target deviant productions consisted of the overextension of the unergative auxiliary have to verbs that require the unaccusative auxiliary be. Snyder, Hyams and Crisma (1995) report that, similarly to the Italian children in the study, the French child made few errors in the selection of auxiliaries. Specifically, it is reported that the French child only produced two errors, which were instances of overextending the have auxiliary to verbs that require be. Based on this (Snyder, Hyams and Crisma (1995)), we conclude that the acquisition of auxiliary selection in child language is a process that occurs early in language development and is generally error free. This contrasts with previous research by Van Hout, Randall and Weissenborn (1992), who report a higher frequency of overextension errors in their French population whereby the have auxiliary is selected with the unaccusative verb. The findings from Snyder, Hyams and Crisma (1995) have been further supported by research on the acquisition of the unaccusative/unergative distinction by Lorusso, Caprin and Guasti (2005) and Sano, Endo and Yamakoshi (2001), where it is shown that the children in their study (monolingual Italian and monolingual Japanese, respectively) accurately distinguish unaccusatives from unergatives verbs as evidenced by the auxiliary selection in the Italian monolingual children (Lorusso, Caprin and Guasti 2005) and the case morphological distinctions in Japanese (Sano, Endo and Yamakoshi 2001). More recently, Costa and Friedmann (2012) examined the acquisition of unaccusativity and A-movement in Hebrew and European Portuguese monolingual children aged between one year, six months and six years, one month. Their study reports that children can clearly distinguish between unaccusative and unergative verbs. Evidence for this comes from the different word orders the children adopt for the unaccusative and unergative verbs and by the possessive datives in Hebrew (Costa and Friedmann 2012). Lorusso’s (2007) study of the L1 acquisition of aspect in Italian brought further evidence in support of children’s ability to differentiate between unaccusative and unergative verbs at an early age. While focusing on the production of the perfective morphology of passato prossimo, Lorusso’s (2007) study reported that Italian children were 98 per cent accurate in their auxiliary selection with unaccusative verbs, while auxiliary selection accuracy rates were lower with unergative verbs. In summary, the key observation from the literature is that children are widely reported to be capable of making the distinction between unaccusative and unergative verbs early in the development process (Snyder, Hyams and Crisma 1995; Lorusso 2007; Costa and Friedmann 2012; among others). Snyder, Hyams and Crisma (1995), Lorusso, Caprin and Guasti (2005), Lorusso (2007) also report that the monolingual Italian-speaking children in their studies acquired auxiliary selection with few instances of target deviance, especially with regard to the matching of the unaccusative verb to the auxiliary essere.
4 The Data The data in this chapter were collected when S was aged between five years, five months and five years, seven months. The parents follow the one-parent one-language rule,
The Lexicon-Syntax Interface in a 3L1 context
21
according to which S is exposed to Italian from her mother, Scottish Gaelic from her father3 and English as the language of the community and the shared language of the parents. The aim of the study was to investigate the child’s awareness of the unaccusative/unergative dichotomy in both English and Italian. To do this, grammaticality judgement tasks were carried out on both languages. Recording sessions of S’s English occurred in the phonetics lab on the Jordanstown campus at the University of Ulster. Also present at the recording sessions was one of the child’s parents,4 although she was encouraged not to participate in any aspect of the recording sessions. Recording sessions of the child’s Italian occurred in a playroom at the child’s home. Present at these occasions were two investigators5 (in line with risk and ethics procedures) and both of S’s parents. All recording sessions were audio recorded with a Dictaphone.
4.1 English grammaticality judgement tasks 4.1.1 Procedure of tasks We used grammaticality judgement tasks (McDaniel, McKee and Smith Cairns 1998; Thornton 1996; Foursha-Stevenson and Nicoladis 2011; among others). In this task, we used hand puppets to provide grammatical and ungrammatical sentences in different contexts. S’s role was to decide whether the sentence produced by puppet one or the sentence produced by puppet two is a grammatical sentence in English.
4.1.2 Materials This experiment was designed in the format of a game, whereby S had to decide which puppet produced a grammatical sentence and which produced an ungrammatical sentence. Every time a puppet got a sentence correct, S rewarded the puppet with a golden coin, and every time a puppet got the sentence wrong, S would punish it with a worm. For each unaccusative/unergative testing condition (past participles and resultatives in English), three sets of ten sentences were presented. Of the ten test sentences, four were grammatical, four were ungrammatical and two were fillers. Each sentence said by the puppet was preceded by a sentence from the experimenter, which set the context for the following sentence (11)6:
11. To get to granny’s cottage, red riding hood had to cross the frozen lake. All grammatical, ungrammatical and filler sentences were said by the puppets in a randomized order throughout each test. Examples of grammatical and ungrammatical sentences from both the perfect participle task (12) and the resultatives task (13) are exemplified below: 12. a. The fallen tree b. *The shouted postman. 13. a. The river froze solid. b. *Dora screamed happy. It is expected then that, if S is aware of the unaccusative/unergative distinction, she would reject the unergative verbs in both the past participle construction in (12b) and the unergative verb in the resultative constructions in (13b).
22
Crosslinguistic Influence and Crosslinguistic Interaction
4.1.3 English results Table 2.1 summarizes S’s results in the two English tasks: Table 2.1 Results of English tasks.
Verb class
Example
Unaccusative The frozen lake English Perfect *Unergative Participle GJT Fillers
Correct Incorrect Overall judgement judgement correct (%) (%) (%) 75
25
*The shouted postman
75
25
The storm appeared
100
0
75
25
91
9
100
0
Unaccusative The glass smashed English into pieces Resultatives *Unergative *Woody sang tired GJT Fillers John drank his tea Across tasks
80
86 84.5
Figure 2.1 English perfect participle results. In the English grammaticality judgement tasks, S accurately judged a total of 80 per cent of perfect participle constructions and 86 per cent of resultative constructions. A breakdown of these results shows that S accurately judged 91 per cent of unergative constructions in resultative tasks as ungrammatical. The rate at which S correctly judged the grammaticality of the unaccusative constructions in the resultative task was 75 per cent. The accuracy rate reported for the unergative constructions in the perfect participle task was 75 per cent, with the accuracy rate for the unaccusative constructions in this task also at 75 per cent. The results show that S had no problems in identifying the fillers as grammatical. This is a reassuring indication that S understood the task at hand.
The Lexicon-Syntax Interface in a 3L1 context
23
Figure 2.2 English resultative GJT results. In comparison to previous grammaticality judgement studies investigating a variety of linguistic structures among bilingual and monolingual children aged between four and six, the average percentage rate termed as ‘high’ ranged in this study from around 70–80 per cent (Van der Lely and Ullman 2001; Foursha-Stevenson and Nicoladis 2011). As can be evidenced from the results of this study, the child displayed exceptionally high accuracy ratings. This would suggest that S appears to have acquired the unaccusative/unergative distinction in her English, despite the fact that the constructions tested are arguably quite hard to process because these particular constructions are less frequent in S’s input. Before moving on to the Italian task, the specific judgement errors made by S in the two English tasks will be examined. In the perfect participle task, the child accepted the following ungrammatical structures:
14. a. *The screamed man. b. *The danced girl. c. *The slept baby.
The child also made target deviant judgements in the opposite direction, rejecting grammatical structures such as
15. a. The fallen tree.
b. The cooked food.
In the grammaticality judgement task with resultatives, the target deviant judgements made by the child included the rejection of grammatical structures such as the following:
16. a. The river froze solid.
b. The table appeared spotless. c. The door slammed shut.
Again, this target deviance was made in the opposite direction, with the acceptance of one ungrammatical structure:
17. * Winnie the Pooh jumped red.
24
Crosslinguistic Influence and Crosslinguistic Interaction
Looking at the English results overall, it appears that S does not seem to have a problem with one particular verb class over another; instead she makes an equal number of target deviant judgements with unergative verbs as she does with unaccusatives. Even within a class, she makes target deviant judgements with various types of verbs. So, for example, with unergatives, there are problems both with motion verbs (dance, jump) and with agentive activities (scream, sleep and slam). Similarly, S makes target deviant judgements both with verbs that function exclusively as unaccusatives (fall, appear) and with alternating change-of-state verbs (freeze, cook). Finally, S makes target deviant judgements across both the perfect participle and the resultative tasks, a fact that clearly shows that her problems cannot be attributed to her having difficulties with one particular test item over another. Bearing in mind what was observed in her English results, S’s Italian results are presented below.
4.2 Italian grammaticality judgement tasks 4.2.1 Procedure of tasks in Italian Again, we used grammaticality judgement tasks (McDaniel, McKee and Smith Cairns 1998; Thornton 1996; Foursha-Stevenson and Nicoladis 2011; among others) to test the unaccusative/unergative distinction in Italian. These tasks differ slightly from the English tasks as the diagnostic utilized in the Italian grammaticality judgement tasks is auxiliary selection. S’s mother carries out these tasks using a hand puppet for the first task and a competition format in the second task.
4.2.2 Materials The first grammaticality judgement task in Italian consists of a puppet that wants to learn English. The mother plays the role of the puppet. The puppet utters both grammatical and ungrammatical sentences in turn and the child’s role is to help the puppet learn Italian by deciding whether each sentence said by the puppet was correct or incorrect. Each test contains four grammatical constructions, four ungrammatical constructions and four variable behaviour verbs, all of which are read out in a randomized order. Italian variable behaviour verbs are included in this task to allow us to test whether the ASH (Sorace 2000, 2004) has an effect on S’s grammatical judgements of auxiliary selection in Italian. It should be recalled that these are verbs that are acceptable with both essere (to be) and avere (to have) in different contexts. Below are examples of grammatical sentences (a), ungrammatical sentences (b) and variable behaviour verbs (c, d) included in this study:
18. a. Biancaneve ha dormito tanto!
Snow White has slept lots! b. *Biancaneve è dormito tanto! Snow White is slept lots. c. Il bambino è gattonato in casa. The child crawled into the house. d. La rana ha saltato. The frog jumped.
25
The Lexicon-Syntax Interface in a 3L1 context
As previously discussed, auxiliary selection in Italian is indicative of the underlying structure of a verb, since an unergative verb selects the auxiliary avere while the unaccusative verb selects the auxiliary essere. Thus the grammaticality judgement tasks will be indicative of the child’s comprehension of the unaccusative/unergative distinction if the child correctly identifies the ungrammatical constructions. The second Italian grammaticality judgement task has a similar procedure, but the task takes the form of a competition between S’s parents: the child’s mother and father take it in turns to utter grammatical and ungrammatical unaccusative and unergative sentences. Essentially the parents use the same verb and present it to the child with different auxiliaries, as shown below:
19. Parent 1: Mickey è congelato (Mickey is frozen)
Child responds Parent 2: *Mickey ha congelato (Mickey has frozen) Child responds
The task of the child is to judge whether the sentence presented is grammatical or ungrammatical, to award the correct parent and to tell the other parent who had uttered a wrong sentence that they were wrong.
4.2.3 Italian task results Recall that previous research conducted by Snyder, Hyams and Crisma (1995) suggests that monolingual Italian children aged between two and two years, seven months are able to distinguish unaccusative from unergative verbs in choosing the correct auxiliary each time. Such claims are further supported by Guasti (2004). Table 2.2 and Figure 2.3 summarize the results for S’s Italian tasks: Overall, the results of this task indicate that, despite her age at the time of testing, the child falls outside the general pattern of development shown by her monolingual peers in relation to the unaccusative/unergative distinction, making target deviant judgements 33.5 per cent of the time, even though Italian is a language that she uses accurately and naturally on a daily basis. This percentage is just outside what is termed Table 2.2 Results of Italian tasks.
Italian Auxiliary Selection
Correct Incorrect Overall judgement judgement correct (%) (%) (%)
Verb class
Example
Unaccusative
il vaso è rotto– the vase broke
58
42
Unergative
Il bambino ha riso – the baby laughed
75
25
75
25
Alternating verbs
66.5
75
26
Crosslinguistic Influence and Crosslinguistic Interaction
Figure 2.3 Italian auxiliary selection GJT results. as high according to Van der Lely and Ullman (2011) and Foursha-Stevenson and Nicoladis (2011). It is also clearly not as high as her English results. In relation to the variable behaviour verbs tested in the Italian tasks, S shows correct acceptance of both the essere and avere auxiliaries 75 per cent of the time. Interestingly, she does so even in the task where answering correctly implies modifying her typical response type by accepting that both parents are right and hence that there is no winner or loser. This shows that she understands the tasks and engages with it fully but also that the choice of auxiliary is not random for her; with these verbs, she modifies the typical response she has been trained to give so as to reconcile with her intuition that the verbs can occur with both auxiliaries7. Looking at the target deviant judgements in more detail, eight of the unergative verbs presented were non-alternating verbs, which can only occur with have, while four of the unergative verbs were alternating. Sorace (2000) reports that native Italian speakers have strong intuitions about the core, non-alternating verbs in their grammar and overwhelmingly reject structures with be. For example, sorridere (‘smile’) obligatorily selects have:
20. Winnie ha sorriso.
Winnie has smiled.
However, the child in this study accepted the following ungrammatical sentences, despite the fact that the verbs in question were non-alternating unergatives and are thus considered ‘core’ verbs:
21. *Winnie è sorriso.
*Winnie is smiled.
22. *Tiger è cantato. *Tiger is sang.
Also with one unergative, S classified the verb dormire (‘sleep’) as variable and therefore accepted it with both auxiliaries:
23. Biancaneve *ha/è dormito.
Snow White *has/is slept.
The Lexicon-Syntax Interface in a 3L1 context
27
Turning to unaccusative verbs, eight of the verbs presented were non-alternating verbs selecting the auxiliary be, while four were alternating. S accepted the correct auxiliary 58 per cent of the time. While this may indicate that S is delayed in comparison to her monolingual peers, who arguably master auxiliary selection very early, it is also possible that in this instance S is performing at chance. It is clear from the results of Italian unergative verbs and the English tasks that S understands the task and can distinguish between unaccusatives and unergatives; however, based on the results from the task involving Italian unaccusative verbs, it is not possible to definitely propose that she experiences more problems with unaccusatives rather than unergatives in Italian. Closer observation of the task results sheds more light on the types of target deviant judgements made with unaccusative and unergative verbs. S made two types of target deviant judgements with unaccusatives. In some cases, she chose the wrong auxiliary, that is, have:
24. a. Merlino è caduto. b. 25. a. b.
*Merlino ha caduto. Merlin is/*has fallen. Mickey è congelato. *Mickey ha congelato. Mickey is/*has frozen.
In other cases she accepted both auxiliaries as grammatical:
26. Babbo Natale è/*ha arrivato.
Father Christmas is/*has arrived.
27. Barbapapá è/*ha apparso.
Barbappa is/*has appeared.
With variable behaviour verbs such as gattonare (‘crawl’), the child correctly accepted both auxiliaries with this construction – when they were presented in either unaccusative or unergative frames, respectively – 75 per cent of the time. Looking at Italian results collectively, the following types of target deviant judgements are observed: i. Both auxiliaries or just be with (core) unergatives ii. Only have with variable behaviour verbs iii. Both auxiliaries or just have with (core) unaccusatives.
4.3 Overall results Below is a summary of the main results from the tasks. What follows is a comparative analysis of the English and Italian tasks. i. S appears to have acquired the unergative/unaccusative distinction both in her English and in her Italian. ii. Her English results are high, although not yet at ceiling. iii. Her Italian results are not as robust as her English results. iv. There is considerable overlap between the lexical verbs that S miscategorizes in the English and in the Italian tasks.
28
Crosslinguistic Influence and Crosslinguistic Interaction
4.4 A comparative look at the results Looking at the results comparatively, it can be said that although S seems to have acquired the distinction in both languages, she is better in English than in Italian. The most striking thing that remains to be discussed is that there is a significant overlap between the two languages in the verbs with which she makes target deviant judgements. This is shown in Table 2.3, where the verbs judged incorrectly in both languages are given in bold. The verbs in bold are the ones where S made target deviant judgements both in the English and the Italian task. Let us consider a couple of examples for each of the languages. For English, the table shows that S classed ‘*The danced girl’ incorrectly as grammatical, and ‘The river froze solid’ incorrectly as ungrammatical. In Italian, the auxiliary in parentheses indicates the one selected by the child. When both auxiliaries are given, it means she accepted both. Let us discuss a couple of examples in detail: S incorrectly accepted as grammatical ‘Tiger è cantato’ and again incorrectly ‘Mickey ha
Table 2.3 Breakdown of verbs incorrectly judged in English and Italian tasks
English Past Participle Task
Verb construction
Child’s judgement
The danced girl
*grammatical
The slept baby
*grammatical
The screamed man
*grammatical
The fallen tree
*ungrammatical
The frozen lake
*ungrammatical
The cooked food
*ungrammatical
English Resultative Task Winnie the Pooh jumped red
Italian Auxiliary Selection Task
*grammatical
The river froze solid
*ungrammatical
The table appeared spotless
*ungrammatical
The door slammed shut
*ungrammatical
[be] Cantare – sing
*grammatical
[be] Sorridere – smile
*grammatical
[be/have] Dormire – sleep
*grammatical with have
[have] Cadere – fall
*grammatical
[be/have] Arrivare – arrive
*grammatical
[be/have] Apparire – appear
*grammatical with have
[have] Congelare –freeze
*grammatical
[have] Saltare – jump
*grammatical
The Lexicon-Syntax Interface in a 3L1 context
29
congelato.’ The stars in the final row, therefore, are used to represent the target deviant judgements in S’s judgements. As Table 2.3 clearly demonstrates, there is an overlap in the verbs incorrectly judged by S across the tasks in the two languages: sleep, fall, freeze, jump and appear. It is important to note again that the tasks were done on different days and in different settings (the university vs. the child’s home), and that they involved different elicitation techniques and different experimenters (in the case of Italian, the child’s parents). This fact excludes any possibility of the child influencing her choices knowingly. Instead, we argue that the existence of a strong overlap in problematic verbs for S in the different grammaticality judgement tasks is very interesting as it points to some kind of influence between the grammars/lexicons of the two languages. In the following section, we attempt to account for this main finding.
5 Discussion Let us remind ourselves of the main results from the English and the Italian tasks. i. S has acquired the unergative/unaccusative distinction both in English and Italian. ii. Her Italian less developed than her English. iii. Neither language has been acquired perfectly (unlike her monolingual peers in both languages, who master the distinction very early). iv. The problem does not lie with one verb class, as both unergatives and unaccusatives have similar grammaticality levels. v. Non-alternating unaccusative and unergative verbs are vulnerable. vi. There is a strong overlap between the verbs miscategorized in both English and Italian. In comparison to the near-perfect overall grammaticality judgements across both English tasks (86 per cent), the overall number of correct grammaticality judgements in the Italian tasks was lower (66.5 per cent). Having considered previous studies (Snyder, Hyams and Crisma 1995; Lorusso, Caprin and Guasti 2005), which report that monolingual Italian children’s production of auxiliary selection in relation to the essere/avere distinction is acquired with few problems from a young age (from one year, seven months in one child in Snyder, Hyams and Crisma (1995) study), the expectation in relation to the grammaticality judgement task is that S would have few problems distinguishing the grammatical from the ungrammatical auxiliary verb matches. Surprisingly, this is not the case, and yet S does remarkably well with variable behaviour verbs. This is important, as it seems to indicate that the child is well aware of the distinction between different verb classes in Italian and that she also knows that while some verbs can occur with both auxiliaries, others are only grammatical with one. With this in mind, we will examine the target deviant judgements produced by S in more detail by comparing them to the most prominent extant hypothesis regarding auxiliary selection in Italian, Sorace’s ASH.
30
Crosslinguistic Influence and Crosslinguistic Interaction
5.1 Italian target deviant judgements and ASH Sorace’s ASH distinguishes Italian verbs into seven classes:
6. Change of location
Change of state Continuation of a pre-existing state Existence of state Uncontrolled process Controlled process (motional) Controlled process (non-motional) (Sorace 2004: 256)
Selects be (least variation)
Selects have (least variation)
If we look at the verbs that S judged incorrectly in the Italian tasks, we see that they fall into four of the above classes on either side of the continuum; in particular, there are examples in both core classes (e.g. i and iv) and periphery classes (e.g. ii and iii) and in fact more target deviant judgements occur with the former:
7. i. Change of location – cadere (fall), arrivare (arrive)
ii. Change of state – apparire (appear), congelare (freeze) iii. Controlled motional process – saltare (jump) iv. Controlled non-motional process – cantare (sing), dormire (sleep), sorridere (smile)
We now move on to discuss the various classes in more detail. The core verbs that are problematic for S in the Italian grammaticality judgement task are cadere, arrivare, dormire, cantare and sorridere. These verbs are the ones that are the least variable and strongly identifiable as either unaccusative and unergative. Recall that according to the auxiliary selection hypothesis, core verbs should be acquired first and in a more unproblematic way. This is clearly not the case. Although the periphery verb classes show more variability within and across languages, in Italian their auxiliary selection is still reported to be consistent (Sorace 2000, 2004; Bard, Frenck–Mestre and Sorace 2010). The results from this study show that S’s responses do not seem to be guided by the auxiliary selection hierarchy. This is something that would be worth investigating further with larger studies. In particular, the verb arrivare is unaccusative and belongs to the change-of-location class of verbs. This class of verbs is the least flexible on the scale, and the selection of the auxiliary essere with these verbs is strictly adhered to in Italian. This verb class is defined by the inherent telicity of the verbs, and the verbs in this class consistently select essere regardless of agentivity or other aspectual features:
8. a. Sono arrivati ospiti. b.
Are arrived guests. ‘The guests arrived.’ Sono arrivati ospiti per ore e ore. Are arrived guests for hours and hours. ‘Guests arrived for hours and hours.’ (Sorace 2000: 864, ex. 4)
The Lexicon-Syntax Interface in a 3L1 context
31
Despite this, S treats arrivare as a variable verb, as exemplified by her acceptance of both auxiliaries with the verb in the grammaticality judgement task:
9. Babbo Natale è/ha arrivato.
Father Christmas is/has arrived.
In the task, S incorrectly judged the essere + cadere (‘fall’) combination as ungrammatical. This is particularly puzzling for two reasons: First, S makes this target deviant judgement in both languages (recall she judges the fallen tree as ungrammatical); second, both arrive and fall are exceptional change-of-state verbs that do not have a causative variant in either language. In consequence, this means that in Italian it would never occur with avere since it cannot be used as a transitive verb. On the opposite side of the continuum, S also makes target deviant judgements with some strictly unergative verbs such as dormire, for which she also accepts both auxiliaries. The verb dormire belongs to the class of verbs called controlled nonmotional process verbs. The processes these verbs refer to are normally agentive. Since agentivity is a distinguishing factor in the unaccusative/unergative dichotomy, it is, once again, surprising that S would treat this verb as a variable behaviour verb. S makes target deviant judgements with two other strongly unergative verbs, sorridere (‘smile’) and cantare (‘sing’), which she treats as unaccusatives (but not as variable behaviour). The fact that these unergatives are used wrongly but in different ways is interesting, as it seems to suggest that S is not using semantic aspects of the auxiliary selection hypothesis to classify these unergatives as unaccusatives or variable behaviour verbs. If she were showing sensitivity to the auxiliary selection hypothesis, we would expect her to treat these verbs the same, but she does not. Moving to the verbs in the middle of the hierarchy, Sorace (2000, 2004) also reports that while the majority of change-of-state verbs in Italian select the auxiliary essere, apparire (‘appear’) shows variation in native speakers’ judgements and is sometimes accepted with avere. Sorace (2000, 2004) proposes that the variation observed in the Italian native acceptability is a result of the semantics of the verb itself and its degree of telicity. In accordance with these findings, S accepted both auxiliaries with it:
10. a. Barbapapà è apparso.
Barbapapà is appeared. b. *Barbapapà ha apparso. Barbapapà has appeared.
Above we have counted the occurrence of apparire with avere as a mistake, but this discussion could suggest that S considers the verb apparire as a variable behaviour verb rather than an unaccusative verb, which obligatorily selects essere. This also means that if we were to classify this verb as potentially able to take avere, the percentage correct for the unaccusative tasks would be higher. A similar target deviant judgement is observed with the verb congelare (‘freeze’). This is also a periphery verb. It belongs to the change-of-state verb class, which, as previously discussed, is identifiable by the variability of telicity within the verb class.
32
Crosslinguistic Influence and Crosslinguistic Interaction
While the verb congelare predominantly selects essere, S incorrectly judges the auxiliary avere and verb congelare combination as grammatical:
11. *Mickey ha congelato.
Mickey has frozen.
Also, it is worth mentioning again that S makes two target deviant judgements with freeze in the English task too (‘the frozen lake’ is judged as ungrammatical, ‘the river froze solid’ is also considered ungrammatical). We return to this below in the concluding section. Finally, the verb saltare (‘jump’) is also a controlled motional event verb, which takes the auxiliary avere; however, the verb can indicate change of location, and in this case it selects essere. Thus the verb saltare is an instance whereby S is exposed to two syntactic options in her input. In the task, S is exposed to an unaccusative use of the verb (in conjunction with an in x PP) but she non-standardly chooses avere as the correct auxiliary:
12. *Winnie ha velocement saltato nel fosso.
Winnie has quickly jumped in the ditch.
In summary, from the Italian grammaticality judgement tasks, we have observed that S makes target deviant judgements with both periphery and core verbs.
5.2 English target deviant judgements in relation to ASH While Sorace’s hypothesis is developed in relation to Italian, it is meant to have crosslinguistic predictive power in relation to the class of verbs that are likely to be problematic for the distinction and the acquisition of it. It is therefore useful to compare S’s target deviant judgements in the English tasks to the ASH classes listed above. It turns out that S’s target deviant judgements fall into six of the seven classes identified by Sorace:
13. i. Controlled motion process – dance (unergative)
ii. Controlled non-motional unaffecting process – scream (unergative) iii. Controlled non-motional process – sleep (unergative) iv. Change of location – fall (unaccusative) v. Change of state – appear (unaccusative) vi. Change of state (alternating) – freeze, cook, slam (unaccusatives).
This finding is unsurprising in consideration of S’s target deviant judgements in her Italian task, as once again it shows that S’s target deviant judgements are not restricted to one verb class. The following section examines the target deviant judgements further in relation to the syntax and semantics of the verb.
5.3 The syntax and the semantics of verb We showed above that there is considerable overlap between the verbs that are problematic for S in English and in Italian. The overlapping verbs are presented in the shaded areas in the table below:
The Lexicon-Syntax Interface in a 3L1 context Table 2.4 The verbs with which S made target deviant judgements in the English task. Target deviance
Verb type
Status
Scream
Controlled non-motional unaffecting process
Unergative
Dance
Controlled motion unaffecting process
Unergative
Slam
Change of State
Unaccusative – alternating
Cook
Change of State
Unaccusative – alternating
Freeze
Change of State
Unaccusative – alternating
Appear
Change of State
Unaccusative – exceptional
Fall
Change of Location
Unaccusative – alternating
Jump
Controlled motional process
Unaccusative – alternating
Sleep
Controlled non-motional unaffecting process
Unergative
Table 2.5 The verbs with which S makes target deviant judgements in Italian. Target deviance
Verb type
Status
Congelare Freeze
Change of State
Unaccusative – alternating
Apparire Appear
Change of State
Unaccusative – exceptional
Cadere Fall
Change of Location
Unaccusative – alternating
Saltare Jump
Controlled motional process
Unaccusative – alternating
Dormire Sleep
Controlled non-motional unaffecting process
Unergative
Sorridre Smile
Controlled non-motional unaffecting process
Unergative
Cantare Sing
Controlled non-motional unaffecting process
Unergative
Arrivare Arrive
Change of Location
Unaccusative
33
34
Crosslinguistic Influence and Crosslinguistic Interaction
So far we have seen that S is clearly better at identifying the unaccusative/ unergative distinction in English than in Italian. It is interesting to observe that, even though the English task is considered to be much harder for S as the two testing constructions are not as frequent in her input as the Italian testing constructions, she still performs better in the grammaticality judgement task in this language. Also, it appears that S has acquired the phenomenon of split intransitivity in both languages as evidenced by her performance in English and her accepting both auxiliaries for variable behaviour verbs in Italian, despite the higher complexity of this kind of response as discussed above. One other thing to note is that she makes just about the same amount of overgeneralizations (in English, this is where an ungrammatical structure is wrongly accepted as grammatical; and in Italian, where the two auxiliaries are wrongly accepted for a verb) and miscategorization judgements (in English, this is where a grammatical structure is judged as ungrammatical; and in Italian, where the wrong auxiliary is accepted for a verb). Finally, there is considerable overlap between the lexical verbs that S makes target deviant judgements with in English and Italian. As we discussed above, the phenomenon of split intransitivity has been at the heart of the debate on the relationship between the lexicon and syntax. Essentially, the puzzle it poses is the following: on the one hand, if the notion of unaccusativity is semantically determined, then we expect unaccusative verbs to be both stable and consistent across languages and within a language. Thus, the existence of verbs behaving like unaccusatives in one language and like unergatives in another, along with the existence of variable behaviour verbs (within the same language), would be problematic for such an approach. On the other hand, if unaccusativity is due to a verb appearing in a certain syntactic configuration, it can be asked why there are classes of alternations as opposed to complete productivity in verb flexibility, and, moreover, why class alternations are somewhat stable across languages. The issue at the heart of our study is the very one that, in the 1980s and 1990s, led to the development of two opposing theories regarding the relationship between verb meanings and their syntactic realization in early acquisition. The first view, which is closer in spirit to lexicalist approaches, is the semantic bootstrapping hypothesis (Grimshaw 1981; Pinker 1984; among others). This theory argues that the child has access to a certain amount of lexical-semantic information by exposure to uses of the verb in actual situations, and from this knowledge he or she is able to acquire information on the syntax of the argument structure of verbs. Thus, the syntactic information regarding the distribution of arguments is derived from the lexical semantics of individual verb forms. On the other hand, syntactic bootstrapping (Gleitman 1990, inter alia) views the primary acquisition of syntactic knowledge, and in particular the structural information regarding the placement of arguments, as fundamental in guiding the acquisition of the lexical meaning of verbs. Clearly, the debate between these two positions mirrors the debate between lexical versus constructional approaches to the lexicon-syntax interface, which was discussed briefly in the introduction. One crucial prediction of the semantic bootstrapping approach is that some meaning components will have privileged status over others. For example, Pinker (1989) and Gropen et al. (1991) argue that children display earlier sensitivity
The Lexicon-Syntax Interface in a 3L1 context
35
to change-of-location than to change-of-state verbs. Accordingly, we would expect that miscategorization/overgeneralization judgements should occur more often in certain classes over others. The syntactic bootstrapping hypothesis, on the other hand, is compatible with constructionalist views of argument projection, where syntactic derivation drives computation and hence verbs mean what the syntax allows them to mean. The prediction of this type of approach is that in acquisition we should witness protracted and unconstrained target deviance combined with massive variation in the syntactic behaviour of verbs (Sorace 2004; Borer 2004). How do our results fit these two opposing views and their respective predictions? We have seen that S makes target deviant judgements across languages, across the unaccusative/unergative divide and, most importantly, across verb classes (change of state, change of location, controlled motion, etc.) in both English and Italian. This clearly shows that she does not have a problem with one specific syntactic configuration, but that admittedly she inserts verbs in various syntactic frames, at times giving rise to alternations that are not possible in her languages, and at other times rejecting correct structures. This is not specific to a verb class, nor can a meaning component be said to take precedence over another, thereby explaining her results in either language. Moreover, these non-standard judgements occur late in her acquisition process. All of this seems to be more compatible with a syntax-driven view of the acquisition process and, in turn, of the interface between lexicon and syntax. One final puzzle we still have to discuss relates to the most striking aspect of our findings, namely the overlap between problematic verbs in the two languages. One could argue that this is evidence that S lets semantics guide her syntax: verbs whose syntax is not yet acquired perfectly in one language (possibly the weaker one) can affect the syntax of the same verbs in the other language. In turn this would point to crosslinguistic influence (Müller and Hulk 2001; Sorace and Serratrice 2009; Stevenson and Nicoladis 2011; among others) from Italian to English, where the delay in the acquisition of unaccusativity in S’s Italian affects her English. Following the semantics-driven acquisition approach, and given the fact that S makes target deviant judgements with the same verb classes in English and Italian, this might suggest that she has only one language system for the languages in her triad. The unitary language system hypothesis (Leopold 1978; Volterra and Taeschner 1978; among others) proposes that children have one unitary lexical-syntactic system at the beginning of their language development. While this school of thought was popular in the 1970s as a way of accounting for language-mixed utterances in the productions of bilinguals, it is now widely accepted that bilingual children have two separate language systems from the beginning of their language development (Genesee 1989; Meisel 1989; De Houwer 1990; among others). Research by Montanari (2009; 2010) proposes that this is also true in trilingual acquisition. It is unlikely that S has one language system at this stage of development, not only because of the previous and widely accepted view that children have different language systems for each language, but also because the unitary language system hypothesis proposes that multilingual children have a single language system only in the initial stage of language development (Volterra and Taeschner 1978). Given that these grammaticality judgement tasks occur when S is aged between five years, four months and five years, seven months, it is reasonable to
36
Crosslinguistic Influence and Crosslinguistic Interaction
assume that, if there were a unitary language stage, S would have passed through this development stage before the age of five years, four months. S also shows evidence of language separation throughout her language development. Another drawback of the semantics-driven language acquisition is that it does not address the question of why S’s acquisition of unaccusativity in Italian is so behind. In contrast, it appears that the first problem is naturally explained and in fact predicted by a syntactic bootstrapping approach, according to which, as previously discussed, a protracted period of overgeneralization is to be expected. The child has syntactic frames and inserts verbs into these, giving rise to alternations that are ruled out in the adult language. What, then, about the overlap? How can syntactic bootstrapping help us make sense of this? One of the crucial aspects of syntax-driven approaches to acquisition of alternations and alternations in general is that meaning alone cannot drive computation, and that meaning is a by-product of syntax. There is nothing intrinsic about the meaning of sleep that explains why it should not be possible to use this verb in a causative frame and give rise to a structure such as Dora slept the baby or the slept baby (which is one of the mistakes S makes). The fact that S makes target deviant judgements with the same verbs in both English and Italian could then be viewed as confirmation of the fact that syntax is sufficiently prominent in the derivation of structure in one language to take over the way the verb is used in the other language. In general, this would mean that S has not yet acquired the proper meaning of these verbs, for example that sleep/dormire cannot be used as unaccusative (possibly derived from a change-of-state transitive) and that the overgeneralizations she makes in one language are then transferred to the other.
5.4 Crosslinguistic influence and the interface hypothesis We have argued that the overlap between problematic verbs in the two languages seems to be related to the notion of crosslinguistic influence from Italian, the child’s weaker language. This seems to be at odds with the interface hypothesis put forward by Tsimpli and Sorace (2006), Sorace and Serratrice (2009), among others, since in this study we found evidence of crosslinguistic influence in the acquisition of a lexicon-syntax interface phenomenon. According to the interface hypothesis, this interface should not be susceptible to crosslinguistic influence since it is a language-internal interface. This study is not the first to observe vulnerabilities at language-internal interfaces; such findings have also been observed by Liceras (2011), among others. Furthermore, given S’s advanced stage of language development and the fact that English is consistently considered the child’s most dominant language in this case study, these phenomena present evidence against a language-dominance approach to crosslinguistic influence (Yip and Matthews 2000; Döpke 2000; among others) as they show target deviance in the child’s stronger language as well as her weaker language.
5.5 Conclusion This chapter reports on the acquisition of unaccusativity by an English-Italian-Scottish Gaelic 3L1 child, aged between five years, four months and five years, seven months. The
The Lexicon-Syntax Interface in a 3L1 context
37
data were collected through three types of elicitation tasks: acceptability judgements of resultative and perfect participle constructions (English) and auxiliary selection in passato prossimo tense (Italian). While the trilingual child in this study is shown to have acquired the unaccusative/unergative distinction in both her English and Italian, the target deviant judgements produced by the child revealed an overlap in the verbs that were problematic in the two languages. The grammaticality judgement tasks showed that S had more problems with grammaticality judgements in Italian than in English (66.5 per cent vs. 86 per cent respectively). Also the verbs that were problematic fall into several semantic classes, making a lexical-semantic-driven analysis of target deviance difficult to sustain. In contrast, we have argued that a model according to which children have knowledge of the projection of arguments independently of the semantics of verbs is best able to account for our data.
Notes 1 Scottish Gaelic is not examined in this chapter as the unaccusative/unergative distinction is not overtly marked nor encoded in Gaelic. As a result, it is not expected that Scottish Gaelic would have an effect on the phenomenon under investigation in this chapter. 2 Unergatives can only form resultatives if they are accompanied by a reflexive, as in John ran himself tired or John laughed himself sore. 3 The father is also a speaker of Italian, albeit a non-native one. During the test, the mother and the father had an equal number of grammatical and ungrammatical sentences. 4 One of the child’s parents attended every data-recording session at the university in keeping with risk and ethics guidelines at the University of Ulster. Neither parent was encouraged to take part in the recording sessions. This precaution was taken to reduce the influence that the presence of another language (by way of the parent’s language) has on the production of S’s English. 5 Special thanks to Deborah Henderson for her assistance in the development of and participation in the elicitation tasks. 6 For a full list of tested items see Appendix 3. 7 In order to neutralize the possible influence on her judgements of the child’s knowledge that her mother, as a native speaker of Italian, will have a better standard of Italian than her father, the parents randomly said grammatical and ungrammatical constructions using different auxiliaries.
References Alexiadou, A., E. Anagnostopoulou and E. Martin (2004), The Unaccusativity Puzzle: Explorations of the Syntax–Lexicon Interface, Oxford: Oxford University Press. Babyonyshev, M., J. Ganger, D. Pesetsky and K. Wexler (2001), ‘The maturation of grammatical principles: Evidence from Russian unaccusatives’, Linguistic Inquiry, 32: 1–43. Baker, M. (1988), Incorporation: A Theory of Grammatical Function Changing, Chicago: University of Chicago Press.
38
Crosslinguistic Influence and Crosslinguistic Interaction
Bard, E., C. Frenck–Mestre and A. Sorace (2010), ‘Processing auxiliary selection with Italian intransitive verbs’, Linguistics, 48 (2): 325–61. Belletti, A., E. Bennati and A. Sorace (2007), ‘Theoretical and developmental issues in the syntax of subjects: Evidence from Near–Native Italian’, Natural Language and Linguistic Theory, 25: 657–89. Borer, H. (1984), Parametric Syntax: Case studies in Semitic and Romance Languages. Foris, Dordrecht. Borer, H. (2004), ‘The grammar machine’, in A. Alexiadou, E. Anagnostopoulou and Martin Everaert (eds), The Unaccusativity Puzzle: Explorations of the Syntax-lexicon Interface, 288–331, Oxford: Oxford University Press. Borer, H. and Y. Grodzinsky (1986), ‘Syntactic cliticization and lexical liticization: The case of Hebrew Dative clitics’, in H. Borer (ed.), The Syntax of Pronominal Clitics, Syntax and Semantics 19, 175–215, New York: Academic Press. Burzio, L. (1981), ‘Intransitive verbs and Italian auxiliaries’, Doctoral diss., Massachusetts Institute of Technology. Costa, J. and N. Friedmann (2012), ‘Children acquire unaccusatives and a–movement very early’, in M. Everaert, M. Marelj and T. Siloni (eds), The Theta System, 354–78, Oxford: Oxford University Press. De Houwer, A. (1990), The Acquisition of Two Languages from Birth: A Case Study. Cambridge University Press. Devlin, M. (2015), ‘Crosslinguistic influence in the English of a trilingual child: A case study of trilingual acquisition’, Ph.D. – diss., University of Ulster. Devlin, M., R. Folli, A. Henry and C. Sevdali (2012), ‘Clitic dislocation in the absence of clitics: A study in trilingual acquisition’, University of Pennsylvania Working Papers in Linguistics, 18 (1): 40–50. Devlin, M., R. Folli, A. Henry and C. Sevdali (2013), ‘Vulnerable domains and cross – linguistic influence: The view from trilingual acquisition’, in S. Stavrakaki, M. Lalioti and X. Konstantinopoulo (eds), Advances in Language Acquisition, 309–19, Cambridge: Cambridge Scholars Publishing. Döpke, S. (1998), ‘Competing language structures: The acquisition of verb placement by bilingual German–English children’, Journal of Child Language, 25: 555–84. Döpke, S. (2000), ‘Generation and retraction from crosslinguistically motivated structures in bilingual first language acquisition’, Bilingualism: Language Cognition, 3: 209–26. Dowty, D. (1991), ‘Thematic proto–roles and argument selection’, Language, 67: 547–619. Filiaci, F., Sorace, A. and M. Carreiras (2013), ‘Anaphoric biases of null and overt subjects in Italian and Spanish: A cross-linguistic comparison’, Language and Cognitive Processes, 1–19. Foursha–Stevenson, C. and E. Nicoladis (2011), ‘Early emergence of syntactic awareness and cross–linguistic influence in bilingual children’s judgements’, International Journal of Bilingualism, 15 (4): 521–34. Gawlitzek-Maiwald, I. and R. Tracy (1996), ‘Bilingual Bootstrapping’, Linguistics, 34: 901–26. Genesee, F. (1989), ‘Early bilingual development: one language or two?’, Journal of Child Language, 16: 161–79. Gleitman, L. (1990), ‘The structural sources of verb meaning’, Language Acquisition, 1: 3–55. Grimshaw, J. (1981), ‘Form, function and the language acquisition device’, in C. L. Baker and J. J. McCarthy (eds), The Logical Problem of Language Acquisition, Cambridge, MA: The MIT Press.
The Lexicon-Syntax Interface in a 3L1 context
39
Gropen, J., S. Pinker, M. Hollander, R. and Goldberg (1991), ‘Affectedness and direct objects: The role of lexical semantics in the acquisition of verb argument structure’, Cognition, 41: 153–95. Guasti, M. T. (2004), Language Acquisition: A Linguistic Perspective, Cambridge, MA: The MIT Press. Hoekstra, T. and R. Mulder (1990), ‘Unergatives as copular verbs: Locational and existential predication’, The Linguistic Review, 7: 1–79. Leopold, W. (1978), ‘A child’s learning of two languages’, in E. Hatch (ed.), Second Language Acquisition: A Book of Readings, 23±32, Rowley, MA: Newbury House. Levin, B. and M. Rappaport-Hovav (1995), Unaccusativity: At the Syntax–Lexical Semantics Interface, Linguistic Inquiry Monograph 26, Cambridge, MA: The MIT Press. Liceras, J. (2011), ‘Beyond (or besides) interfaces?’, Linguistic Approaches to Bilingualism, 1 (1): 54–7. Lorusso, P. (2007), ‘The acquisition of aspect in L1 Italian’, in A. Belikova, L. Meroni, and M. Umeda (eds), Proceedings of the 2nd Conference on Generative Approaches to Language Acquisition North America (GALANA), 253–64, Somerville, MA: Cascadilla Proceedings Project. Lorusso, P., C. Caprin and M. T. Guasti (2005), ‘Overt subject distribution in early Italian children’, in A. Brugos, M. R. Clark–Cotton and S. Ha (eds), BUCLD29: A Supplement to the Proceedings of the 29th Annual Boston University Conference on Language Development. Available online: http://www.bu.edu/linguistics/BUCLD/supp29.html. Meisel, J. (1989), ‘Early differentiation of languages in bilingual children’, in K. Hyltenstam and L. Obler (eds.), Bilingualism Across the Lifespan: In Health and Pathology, 13±40. Cambridge: Cambridge University Press. McDaniel, D., C. McKee and H. Smith Cairns (1998), Methods for Assessing Children’s Syntax, Cambridge, MA; London: The MIT Press. Montrul, S. and T. Ionin (2012), ‘Dominant language transfer in Spanish heritage speakers and second language learners in the interpretation of definite articles’, Modern Language Journal, 96: 70–94. Müller, N. (1998), ‘Transfer in bilingual first language acquisition’, Bilingualism: Language and Cognition, 1 (3): 151–71. Müller, N. (2003), (In)vulnerable Domains in Bilingualism, Amsterdam: Benjamins Publishing. Müller, N. and A. Hulk (2001), ‘Crosslinguistic influence in bilingual language acquisition: Italian and French as recipient languages’, Bilingualism: Language and Cognition, 4 (1): 1–21. Perlmutter, D. (1978), ‘Impersonal passives and the unaccusative hypothesis’, in Proceedings of the Fourth Annual Meeting of the Berkeley Linguistic Society, 157–89, Berkeley: University of California. Pinker, S. (1984), Language Learnability and Language Development, Cambridge, MA: Harvard University Press. Pinker, S. (1989), Learnability and Cognition: The Acquisition of Argument Structure, Cambridge, MA: The MIT Press. Platzack, C. (2001), ‘The vulnerable C–domain’, Brain and Language, 77: 364–77. Ramchand, G. (2008), Verb Meaning and the Lexicon: A First Phase Syntax, Cambridge: Cambridge University Press. Rosen, C. (1984), ‘The interface between semantic roles and initial grammatical relations’, in D. M. Perlmutter and C. Rose (eds), Studies in Relational Grammar 2, 38–77, Chicago, IL: Chicago University Press.
40
Crosslinguistic Influence and Crosslinguistic Interaction
Rothman, J. (2009), ‘Understanding the nature of early bilingualism: Romance languages as heritage languages’, International Journal of Bilingualism, 13: 155–63. Sano, T., M. Endo and K. Yamakoshi (2001), ‘Developmental issues in the acquisition of Japanese unaccusatives and passives’, in Proceedings of the 25th Annual Boston University Conference on Language Development, 668–83. Serratrice, L., A. Sorace and S. Paoli (2004), ‘Crosslinguistic influence at the syntax – pragmatics interface: subjects and objects in English–Italian bilingual and monolingual acquisition’, Bilingualism: Language and Cognition, 7 (3): 183–205. Simpson, J. (1983), ‘Resultatives’, in L. Levin, M. Rappaport and A. Zaenen (eds), Papers in Lexical–functional Grammar, 143–57, Bloomington: Indiana University Linguistics Club. Slabakova, R. and I. Ivanov (2011), ‘A more careful look at the syntax–pragmatics interface’, Lingua, 121: 637–51. Slabakova, R., P. Kempchinsky and J. Rothman (2012), ‘Clitic–doubled left dislocation and focus fronting in L2 Spanish: A case of successful acquisition at the syntax–discourse interface’, Second Language Research, 28: 319–43. Sorace, A. (1993a), ‘Incomplete vs. divergent representations of unaccusativity in non–native grammars of Italian’, Second Language Research, 9: 22–47. Sorace, A. (2000), ‘Gradients in auxiliary selection with intransitive verbs’, Language, 76 (4): 859–90. Sorace, A. (2004), ‘Gradience at the lexicon–syntax interface: evidence from auxiliary selection and implications for unaccusativity’, in A. Alexiadou, E. Anagnostopoulou and M. Everaert (eds), The Unaccusativity Puzzle: Explorations of the Syntax–Lexicon Interface, 243–68, Oxford: Oxford University Press. Sorace, A. (2011), ‘Pinning down the concept of “Interface” in bilingualism’, Linguistic Approaches to Bilingualism, 1 (1): 1–33. Sorace, A. and F. Filiaci (2006), ‘Anaphora resolution in near–native speakers of Italian’, Second Language Research, 22: 339–68. Sorace, A. and L. Serratrice (2009), ‘Internal and external interfaces in bilingual language development: Beyond structural overlap’, International Journal of Bilingualism, 13 (2): 195–210. Snyder, W., N. Hyams and P. Crisma (1995), ‘Romance auxiliary selection with reflexive clitics: Evidence for early knowledge of unaccusativity’, in E. Clark (ed.), Proceedings of the 26th Annual Child Language Research Forum, 127–36, Stanford: Centre for the Study of Language and Information. Thornton, R. (1996), ‘Elicited production’, in D. McDaniel, C. McKee and H. S. Cairns (eds), Methods for Assessing Children’s Syntax, 77–101, Cambridge, MA: The MIT Press. Tsimpli, I. M. and A. Sorace (2006), ‘Differentiating interfaces: L2 performance in syntax– semantics and syntax–discourse phenomena’, in D. Bamman, T. Magnitskaia and C. Zaller (eds), BUCLD30: Proceedings of the 30th Annual Boston University Conference on Language Development, 653–64, Somerville, MA: Cascadilla Press. Van der Lely, H. and M. Ullman (2001), ‘Past tense morphology in specifically language impaired and normally developing children’, Language and Cognitive Processes, 16 (2–3): 177–217. Van Hout, A., J. Randall and J. Weissenborn (1992), On the Acquisition of Unaccusatives, Paper presented at the Boston University Conference on Child Language Development, 25 October 1992.
The Lexicon-Syntax Interface in a 3L1 context
41
Van Valin, R. D. (1990), ‘Semantic parameters of split intransitivity’, Language, 66: 221–60. Volterra, V. and T. Taeschner (1978), ‘The acquisition and development of language by bilingual children’, Journal of Child Language, 5: 311–26. Yip, V. and S. Matthews (2000), ‘Syntactic transfer in a Cantonese-English bilingual child’, Bilingualism: Language and Cognition, 3: 193–208.
3
Developing Cognitive Strategies Through Pluralistic Approaches Rebecca Dahm
1 Introduction Over the last thirty years, many researchers have explored the advantages related to bi/multilingualism. Bialystok (1987) was able to demonstrate the cognitive advantages presented by bilinguals as well as the increased ability of bilingual children to solve problems. Subsequently, Cummins (1991) stressed the interdependence of language skills in both languages in his theory of the common underlying proficiency: He considered that the use of two or more languages develops metalinguistic skills1 that can be regarded as a comprehensive monitoring system related to all the languages that the multilingual speaker knows. Finally, the research led by Herdina and Jessner (2002) on multilinguals helped highlight the transversal nature of these benefits. They confirm the existence of a unique system, which they consider both specific and global. Additional research (Bono 2008a; Cenoz, Hufeisen and Jessner 2001; Clyne 2003; Ringbom 2007) has underlined these advantages in the context of the heightened ability of multilinguals to reflect upon the languages belonging (or not) to their multilingual repertoire. Gajo (2001), Herdina and Jessner (2002) and then Moore (2006) determined the nature of this strategic advantage as being metalinguistic. As it is now commonly admitted that metalinguistic awareness is an essential component of multilingual competence, it seems interesting to adapt this hypothesis to the French institutional context, and more specifically to the ‘strictly monolingual’2 population. Is it possible to enable these students to develop such cognitive advantages without benefitting from an authentic multilingual situation? Can the strategies thus being used be transferred, de facto, to the learning of English as a foreign language (L2)? To explore these research questions, I decided to confront French monolingual learners, who had benefited from four years of basic instruction in English, with pluralistic approaches based upon unknown languages (PAUL). Such an approach partially relies upon the definition given by Candelier (2008) and published on
44
Crosslinguistic Influence and Crosslinguistic Interaction
the FREPA website (European Centre for Modern Languages 2007): ‘The term “pluralistic approaches to languages and cultures” refers to didactic approaches which use teaching/learning activities involving several (i.e. more than one) varieties of languages or cultures.’ Furthermore, in PAUL sessions, students are in turn confronted with metasemantic, metasyntactic and metaphonological tasks3: They will need to build upon their prior knowledge of (in this case) two different linguistic systems (L1 French and L2 English) so as to solve the problem they are presented with. Finally, PAUL sessions are led in a systematic and regular way so as to provide comparable results. This chapter discusses the cognitive processes involved in the metasemantic problem-solving activities and therefore examines the implementation of learning strategies facilitating the decoding of a written source. After exploring the theoretical background (taxonomy of learning strategies, underlying processes, crosslinguistic influence and the transfer of learning strategies), I will expose the research design and methodology being used before analysing and interpreting the data.
2 Theoretical background 2.1 Taxonomy of learning strategies For Weinstein and Mayer (1986), learning strategies aim to facilitate learning and require a consciousness on behalf of the learner. Rabinowitz and Chi (1987) consider that strategies cannot be ‘strategic’ without a certain degree of consciousness. Therefore, they can no longer be considered as strategies as soon as they are implemented automatically. This is the reason why I will focus on conscious efforts made by the students to solve a problem, thus complying with the continua ‘from conscious to unconscious,’ as defined by Finkbeiner (2005: 60). These strategies will be analysed by studying the operations students ‘consciously’ implement when performing tasks that will enable them to solve the problem. For O’Malley and Chamot (1990: 8), learning strategies can be divided into three categories: metacognitive strategies, cognitive strategies and socio-affective strategies. The focus of this chapter is on cognitive strategies and their underlying processes. Although the framework is the taxonomy of cognitive strategies as proposed by O’Malley and Chamot, I would like to adapt it according to additional theoretical information and the specificity of my experiment, focusing on four types of strategies. The necessary adaptations will be explained below.
2.1.1 Translation strategy For O’Malley and Chamot (1990: 120), implementing the translation strategy covers the use of the first language as a base for understanding and/or producing the second language. It is not, per se, an immediate word-for-word translation; it can also represent a connection between a linguistic element from the L3 with an element derived from
Developing Cognitive Strategies Through Pluralistic Approaches
45
the L1 or the L24. So this variable should be identified on the basis of indicators generally reflecting a comparison with the L1 and/or L2. Nevertheless, unlike O’Malley and Chamot, I would rather restrict the translation strategy to the direct activity of translation. It seems, in fact, essential to distinguish this strategy from the strategy of comparison.
2.1.2 Comparison strategy The strategy I choose to call ‘comparison’ differs from the strategy of translation by the degree of consciousness that its implementation requires. When learners notice similarities between two languages, they show a metalinguistic activity (Gombert 1990), which is by definition conscious. They clearly express their reliance on their prior knowledge of the L2 or L1 as their answers reveal the identification of the linguistic proximity. Even if the underlying processes of translation can include comparison procedures (setting up L3–L2 connections in order to understand the new data before translating), the expressed consciousness needs to be taken into account as it plays a major part in the transferability of the strategy. If the student does not deconstruct the performed task consciously, he will not be able to activate the necessary schemata when faced with a situation that only presents structural similarities (Holyoak and Koh 1987).
2.1.3 Inferencing strategy O’Malley and Chamot (1990: 120) define inferencing as follows: ‘Using available information to guess meanings of new items, predict outcomes, or fill in missing information.’ I therefore consider that the indicator for this strategy is the dependence on several elements of the sentence to infer meaning. The decision was made to rely upon this rather narrow definition of inferencing so as to be able to compare the results.
2.1.4 Deduction strategy O’Malley and Chamot (1990: 119) consider deduction as ‘applying rules to understand or produce the second language or making up rules based on language analysis.’ Therefore, the reliance on a principle or rule is seen as an indicator of this strategy. For example, relying upon uppercase letters to recognize proper nouns can be considered a deduction. The aforementioned strategies have been adapted and defined so as to provide a clear theoretical framework, thereby enabling the subsequent analysis. Even though the descriptions may appear restrictive and only describe certain aspects of the announced strategies, they are neither mutually exclusive nor do they pretend to encompass the entire scope.
2.2 Underlying processes To be able to properly describe strategy use, one needs to take into account the underlying processes, which are a ‘series of actions or steps taken in order to achieve a particular end’ (Oxford Dictionaries).
46
Crosslinguistic Influence and Crosslinguistic Interaction
2.2.1 Top-down and bottom-up processes Following Howard (1985: 28), I choose to distinguish between low-level (bottom-up) processes and high-level (top-down) processes when studying the strategy of deduction. Once again, only certain aspects of the theory will be examined so as to facilitate the analysis. The identification of uppercase letters referring to proper nouns or the association of numbers with age will testify to the implementation of a bottom-up deduction strategy (which starts from the input), while relying on world or academic knowledge outside the input refers to a top-down deduction strategy. Considering bottom-up or top-down processes is one way of analysing the processes underlying the stated strategies. However, sometimes learners indicate one strategy while implementing another one.
2.2.2 Implicit and explicit metalinguistic activity Processes can be considered as explicit or implicit. Ellis (2008: 965) uses the term implicit knowledge to refer to the knowledge of language that is ‘intuitive and tacit,’ and explicit knowledge to indicate the ‘knowledge of rules and items in an analysed form’ (2008: 962). This distinction makes it possible to underline the situations when attention to language surfaces. I would like to borrow these terms so as to be able to properly term the way strategies are implemented by learners. Indeed, strategies can be consciously used and will thus be directly stated on the group sheets. However, sometimes the learner states a certain strategy while he ‘unconsciously’ resorts to another one. Using the term ‘unconsciously’ remains hazardous as the results do show a metalinguistic activity. Therefore, the term ‘implicit’ seems more appropriate.
2.3 Crosslinguistic influence or transfer? When an individual engages in language learning, one can easily surmise that both prior linguistic knowledge and prior learning experience will play a part in this complex cognitive learning task. Therefore, transfer will not only be examined from a linguistic point of view but will also be explored from a cognitive point of view. Both theoretical fields will be drawn upon: First, the strategies used by the students to learn how to understand a text in an unknown language will be analysed, taking into account any existing crosslinguistic influence. Hence, the concept of psychotypology, the part played by the level of proficiency as well as the L1 and Ln factors, will be explored so as to understand the origin of the linguistic transfer. Secondly, transfer of learning will be examined so as to understand how learning strategies can be used in a different context.
2.3.1 Crosslinguistic influence and multilingualism In the field of multilingualism, Herdina and Jessner (2002: 29) defined ‘transfer’ as a ‘basically predictable static or monotonous phenomenon of the transfer of (the same) structures of L1 to L2’. It has also been called ‘crosslinguistic interaction,’ ‘interference,’ or more frequently ‘crosslinguistic influence’ (Sharwood-Smith and Kellerman 1986),
Developing Cognitive Strategies Through Pluralistic Approaches
47
which can be considered as a ‘super-ordinate term’ (De Angelis and Selinker 2001: 42) and covers a wider spectrum of languages. As Sharwood-Smith (1994: 194) puts it, ‘crosslinguistic influence’ also ‘means the influence of any “other tongue” known to the learner on that target language’. In this chapter, I will strive to understand how ‘nonnative language knowledge is … used and, more broadly, how it affects the acquisition and production process as a whole’ (De Angelis 2007: 17).
2.3.1.1 The concept of psychotypology Kellerman (1977) found it necessary to complete the theory of linguistic transfer using a psychological approach so as to be able to understand what makes language learners identify certain target structures with their L1. He thus developed (and was followed by many others) the concept of psychotypology or perceived language distance, which takes into account the fact that the subjective assessment of the typological proximity between languages can change learners’ behaviour in lexical transfers (Deyrich 2007; Ellis 1994; Kellerman 1979, 1983, inter alia). This concept plays a major part in crosslinguistic influence since In any learner’s attempt to acquire a new language, language distance is ultimately in the eye of the beholder. Research indicates that when everything else is equal, transfer will most likely result from a learner’s judgment (made consciously or unconsciously) that particular structures in a previously learned language are quite like – if not the same as – structures in the target language. (Odlin 1989: 142).
It therefore seems essential to try and understand the learner’s judgement regarding the structures of the new language (or target language) they are being confronted with, and to analyse the impact of such a consideration on his ability to understand the target language.
2.3.1.2 Part played by the level of proficiency Studies on crosslinguistic influence in multilingual contexts have shown that the aforementioned phenomenon may not only come from an L2 that the learner knows well (Singleton 1987; Williams and Hammarberg 1998) but can also spring from lesserknown L2s (De Angelis 2005; Rivers 1979; Selinker and Baumgartner-Cohen 1995). The level of proficiency can be relatively low, as one to two years of formal education seem to be enough to significantly affect the production and development of the target language (De Angelis 2007: 34). One can therefore predict that the students observed in this study will be able to rely on their L2, since they have been studying English for four years and have reached an A1 level.
2.3.1.3 The L1 and Ln factors Williams and Hammarberg (1998) underline the importance of the L1’s instrumental role, since the L1 tends to be largely used when resorting to metalinguistic activities. Bardel and Falk (2007), however, consider that the ‘L2 factor’ can be more important than the typological proximity between the L1 and L3, as learners rely more upon
48
Crosslinguistic Influence and Crosslinguistic Interaction
their L2 (Hammarberg 2001), whatever the typological factor. The L2 factor is ‘a desire to suppress L1 as being “non-foreign” and to rely rather on an orientation towards a prior L2 as a strategy to approach the L3’ (Hammarberg 2001: 36–7). For reasons of clarity in a multilingual context, I choose to follow Bono (2008b) and call this the Ln factor.
2.3.2 The transfer of strategies Crosslinguistic influence plays a major part in the transfer of linguistic items. However, this research focuses on the transfer not only of lexical items but also of learning strategies in multilingual contexts. The initial findings related to linguistic transfer are based on the traditional contrastive analysis hypothesis developed by Lado (1957). Interestingly, research in psychology adds to the characterization of transfer. Gick and Holyoak (1983) and then Holyoak and Koh (1987) consider it possible for a person who has first induced a schema from initial examples to solve novel problems that can be categorized as instances of this schema. However, if two situations are drawn from disparate domains, learners will mainly rely upon structural similarity to make use of the source analogue. Lots of discussions surround the term ‘structural similarity,’ which can involve building upon ‘structural isomorphism between base and target domains’ (Kokinov and French 2003: 114) or ‘finding a correspondence between the conceptual structures of the two domains compared’ (Gentner and Markman 2006: 1–2), the latter being the view explored in this chapter. It seems therefore possible to consider that strategies may be transferred inasmuch as the student recognizes structural similarities in the given tasks so as to resort to the adequate strategy. In the context explored here, it seems important to study whether the acquired skills can be generalized to apply to other contexts. I will thus explore the ‘“transfer of Learning” [which] refers to how previous learning influences current and future learning or performance, and how past or current learning is applied to similar novel situations’ (Haskell 2001; Speelman and Kirsner 2005). Indeed, one of the main preoccupations in the classroom is the transfer of learning, which is not merely the application of learning but the learning of something new.
2.3.2.1 The taxonomy for levels of transfer In order to determine the impact of connecting past learning to new situations, Haskell (2001) devised a taxonomy for levels of transfer, building upon the research led by Gagné (1985), Gick and Holyoak (1987) and Perkins and Salomon (1988), inter alia. He considered that for transfer to take place, levels 4 (near transfer), 5 (far transfer) and 6 (displacement or creative transfer) needed to be reached. These levels can be described as follows (Calais 2006: 3): Level 4: Near transfer Near transfer occurs when we transfer previous knowledge to new situations closely similar to, yet not identical to, initial situations. Transferring our
Developing Cognitive Strategies Through Pluralistic Approaches
49
experiences associated with driving a car with a manual transmission to driving a truck with a manual transmission reflects an example of near procedural transfer. Level 5: Far transfer Far transfer entails the application of learning to situations entirely dissimilar to the initial learning. This level of transfer of learning reflects analogical reasoning. For example, learning about logarithms in algebra and applying this knowledge in assessing the growth of bacteria in microbiology. Level 6: Displacement or creative transfer Displacement or creative transfer results in the creation of a new concept because of the interaction of the newly perceived similarity between the new and the old. This type of learning involves more than the mere insight that something is similar to something else.
Haskell also classified transfer according to the different types of knowledge upon which transfer is predicated.
2.3.2.2 Types of transferable knowledge Anderson (1980, 1993) developed the adaptive control of thought (ACT) skillacquisition theory, mainly building upon the distinction of declarative and procedural knowledge. As Mehrnoosh, Manijah and Daryoush (2013: 2) put it, Declarative knowledge is added into the memory through chunks and is activated and subsequently acquires strength as a result of frequent use, i.e. practice. Proceduralization takes place when declarative knowledge is transformed into procedures, production rules, for performing a skill. Production rules also require practice to acquire activation and strength. Practice at this stage results in automatization of skill.
Strategic knowledge (Flavell 1979; Weinstein and Mayer 1986) has also been regularly used by scientists working with cognition since it mainly involves knowledge of our own cognitive processes, occurring during self-monitoring of our progress when attempting to learn. Schoenfeld (1985) and Marzano et al. (1988) explored conditional knowledge, which is the knowledge or awareness of when to apply knowledge in a context-appropriate manner. Haskell (2001) added a fifth category, which is theoretical knowledge: It consists of ‘our understanding of various explanatory connections regarding phenomena, cause and effect, and in-depth level relationships’ (Calais 2006). However, declarative knowledge seems to be essential for successful transfer as it establishes the preconditions for the four other types. This is confirmed by Mehrnoosh, Manijah and Daryoush (2013: 14) when they consider that ‘availability of speeded explicit declarative knowledge, or at least partially acquired proceduralized knowledge can enable learners to generalize it to unfamiliar contexts.’
50
Crosslinguistic Influence and Crosslinguistic Interaction
3 Research questions Taking into account the research on crosslinguistic influence and transfer of learning, I will address the following research questions: 1. Which strategies do students resort to when trying to decode a text in an unknown language? What do these strategies reveal about crosslinguistic influence and underlying processes? 2. Which learning strategies are transferable from one language to another, and under what conditions?
4 Methodology 4.1 Participants The research project was implemented by five lower secondary school teachers with students from year seven, aged twelve to thirteen (136 students over five forms), throughout the year 2011–12 in the regions of Limousin and Aquitaine (France). These students had studied one foreign language (L2 English) for four years. Out of the total sample of 136 students, only 88 are included in the study, as they are perfectly monolingual in their family environment (cf. bottom page note 2). So as to be able to understand the implemented strategies, twenty-two groups of four were set up. Peer-to-peer explanations of a linguistic feature are said to be beneficial for second language (L2) development (Hulstijn and Hulstijn 1984; Sorace 1985; Webb 1989). I can therefore hypothesize that the verbalization of their understanding of the target form should help students become aware of the implemented strategy. The groups were set up according to various parameters (gender, learner of Latin, other L2, willingness to learn other languages, interest in reflexive practice, perception of personal grammar skills, results of pre-test) extracted from both the semantic pre-test and the pre-survey (see below) so as to be heterogeneous, yet of comparable structure.
4.2 Design 4.2.1 Experimental design The strategy study is part of a larger experiment based on PAUL sessions and is thus set up in such a way as to confront students successively with three unknown languages (henceforth referred to as Ln): The first language is Dutch (chosen because of its typological proximity with English), the second is Italian (close to French), and the last is Finnish, an agglutinative language that presents no immediate similarity with any language they know. The experiment includes, for each language, three successive sessions of metasemantic, metasyntactic and finally metaphonological activities. The sessions follow the same model so as to confer a certain systematicity to the regular exercise and to allow the comparison of the results. They take place on a monthly basis.
Developing Cognitive Strategies Through Pluralistic Approaches
51
In this chapter, I will explore the results from the first session involving each language, which rely on metasemantic activities: Students are to solve the problem of access to meaning in the texts presented below. Text in Dutch:
Text in Italian:
Text in Finnish:
The three texts, while addressing different subjects (all chosen according to the A1 level of the CEFR), have similar structural characteristics: The students can rely upon eight cognates5 (some are repeated) and can complete nine inferencing strategies and six deductions (three bottom-up deductions and three top-down deductions).
52
Crosslinguistic Influence and Crosslinguistic Interaction
All three texts and accompanying worksheets have the same instructions: ‘What do you understand?’ and ‘Explain how you proceeded to understand this piece of information.’ No specific guidance is given by the teacher so that the results represent the strategies naturally implemented by students, without being influenced by the teacher. Organization of the PAUL sessions Each fifty-minute session is organized in three different phases: 1. The pupils have about five minutes of individual work for them to focus on the task. 2. The group work lasts for twenty-five minutes and requires the students to share their findings. By doing so, the students not only validate their own results but also enhance their understanding of the texts. They are also required to reflect upon the strategies they have implemented by discussing how they managed to reach the meaning. It is a ‘reasoning-gap activity’ as defined by Prabhu (1987). 3. The teacher leads a synthesis during which both the meaning and strategies are discussed.
4.2.2 Research design The study followed a quasi-experimental, pre-test and post-test mixed-method design. The PAUL sessions were administered by teachers of English and took place once a month, instead of their ordinary English class. Each one of them also taught English to another year-seven form (same age group) with a similar number of students, who then make up the control group. Both groups underwent a semantic pre- and post-test: Students were led to reflect upon a text in English, level C1 (their ordinary CEFR6 target level being A2), and were asked to explain how they managed to access meaning. These tests were completed with pre- and post-surveys, so as to extract relevant additional data (i.e. group organization, opinion, etc.).
4.3 Corpus For each meta-semantic PAUL session, students filled in group sheets; the results recorded on the group sheets come from a consensus reached collaboratively through the comparison of individual ideas. The sheets were also compared to the recordings and transcriptions of the interactions, so as to determine which strategies were explicit and which were implicit. There are thus twenty-two group sheets and twenty-two transcriptions of the interactions for each language (Ln). The corpus is also composed of the semantic pre- and post-tests and post-surveys, for both experimental and control groups.
4.4 Analysis Two types of analyses were used in the quasi-experimental strategy study: 1. Qualitative analysis ●● A first qualitative analysis of the group sheets enabled the classification of the strategies explicitly mentioned by the students.
Developing Cognitive Strategies Through Pluralistic Approaches ●●
53
A second qualitative analysis was carried out of both the group sheets and the transcripts of the group interactions; the question, ‘How did you manage to understand [the above-mentioned information]?’, made the students verbalize their thoughts and was thus used as a data elicitation method. For reasons of brevity, only some excerpts of the qualitative analysis will be given (see Table 3.1).
2. Quantitative analysis ●● A quantitative analysis of the qualitative results from the group sheets was carried out so as to be able to consider the frequency of use of each type of strategy that the students declared to have implemented (Figure 3.1). These results were compared to an ideal projection, as could have been achieved by a multilingual expert7 who would reach a precise and complete understanding of the proposed texts. The graph representing these quantitative data (Figure 3.2) allows a better understanding of what strategies are either most attractive or least easily implemented by students. ●● A statistical analysis (Table 3.1) of the explicit strategies (organized according to the methodology proposed by Finkbeiner et al. (2012: 67)) enables us to visualize examples of the strategy, the percentages of use, the mean values and the standard deviation (SD). ●● An additional study (Table 3.2) of the number of groups having implemented these strategies is useful so as to complement some of the results. Consequently, I can discuss what strategies are most readily used. The analysis of the evolution of practices over the three sessions should enable us to check if the mere contact with foreign languages triggers a consciousness of the implemented strategies. Finally, these results were triangulated with the results from the semantic pre- and post-tests and the post-survey so as to inform the transfer of strategy use to the L2 or to other extracurricular situations.
5 Results In order to determine which strategies students readily resort to when trying to decode a text in an unknown language, I examined the frequencies of the strategies explicitly mentioned by the students (Figure 3.1). I then compared these results to the strategies an ‘expert multilingual’ would have used; the results can be visualized in Figure 3.2. When observing Figure 3.2, one can see positive values (related to strategies of comparison and translation), which means that the implementation led by pupils is higher than the initial projection made by an expert, thus signalling the most readily used strategies. The negative values (related to inferencing and deduction) hint that these strategies are less implemented than is possible. So as to have a finer analysis of the implemented strategies and their frequency of use, Table 3.1 gives examples of these strategies and also shows the mean values and standard deviation (SD).
54
Crosslinguistic Influence and Crosslinguistic Interaction 60% 50% 40% 30% 20% 10% 0%
Comp Comp Comp Comp GB F lat/ ital Span Dutch S1 55% 10% 0% 0% Italian S1 9% 54% 1% 1% Finnish S1 4% 58% 1% 0%
Infer.
Ded –
10% 9% 14%
12% 4% 10%
Ded + Transl 8% 7% 12%
5% 15% 1%
Déd –
Déd +
Trad
Figure 3.1 Synthesis of the strategies declared by the students. 30% 20% 10% 0% –10% –20% –30%
Com p GB
Com pF
com p ital
Com p Esp
Infer.
Diff. NLS1
20%
6%
0%
0%
–25%
2%
–5%
2%
Diff. ItS1
5%
23%
1%
1%
–28%
–8%
–5%
11%
Diff. FinnS1 0%
27%
1%
0%
–25%
–3%
–1%
1%
Figure 3.2 Differential between the students’ use of explicit strategies and the expected results.
5.1 Most readily used strategies: comparison and translation A number of interesting observations can be made from Table 3.1 when considering the most readily used strategies (comparison and translation). The source language changes according to the language distance it presents with the Ln. There are thus, not surprisingly, a great number of comparisons with English when solving the access to meaning of a Dutch text (55 per cent, mean 4.33, SD = 1.68) and of comparisons with French when trying to understand a text in Italian (54 per cent, mean = 4.63, SD = 1.98). The standard deviation is quite high in both cases because of the differing value scale used by the students, but the hypothesis of psychotypology seems to be worthwhile examining.
Table 3.1 Frequency of strategies used by students during the metasemantic task (sample with n = 88). Strategy definition
Dutch
Example
Mean
SD
% use
4.33
1.68
55
1.5
0.67
9
1.2
0.45
4
1.7
1.17
10
4.63
1.98
54
4.33
1.25
58
‘Rossi, bianchi = rouge, blanc. C’est très logique car ça ressemble au français et ça ressemble à l’espagnol’.
N/S
N/S
1
[It’s very logical because it is similar to French and to Spanish.]
N/S
N/S
‘Ca ressemble à l’anglais. Hallo ressemble à Hello (façon de l’écrire)’. [It looks like English. Hallo is like Hello (the way you write it).]
Italian
‘Nome. Ressemble au nom d’anglais = name’. [Is similar to the English name = name.]
Finnish
‘Musiikkia: fait penser à l’anglais music’ [makes us think of English music.]
Comparison F Comparing new language items to French (L1)
Dutch
‘zijn naam. Nous avons *penser à “son nom” en français’. [Made us think of ‘son nom’ in French.]
Italian
‘Il y a écrit grandi = grand ressemble au français’. [You can read grandi = grand like in French.]
Finnish
‘musiikkia: musique. Cela ressemble au français’. [musiikkia: music. Looks like French.]
Comparison Spanish Comparing new language items to Spanish
Italian
55
(Continued )
Developing Cognitive Strategies Through Pluralistic Approaches
Comparison GB Comparing new language items to English (L2)
Language
56
Table 3.1 (Continued) Strategy definition
Language Finnish
Inferencing Using available information to guess meanings of new items
Dutch
Mean
SD
% use
‘Myös. Ma comme mio en Italien’
N/S
N/S
2
[My like mio in Italian.]
N/S
N/S
‘Nous avons penser à “parents” parce que c’est logique dans la phrase’
1.7
0.95
10
1.75
0.89
9
2.75
2.04
14
1.64
0.37
12
[We thought of ‘parents’ because it’s logical in the sentence.] Italian
‘Si chiama Minnie. On a trouvé avec le contexte’. [I found it thanks to the context.]
Finnish
‘Kyllä, minä soitan. Soitan pianoa. Oui, j’en joue. je joue au Piano. On a trouvé grâce au français et au sens des mots’. [We found it thanks to French and the meaning of the words.]
Deduction bottom-up (D-) Applying rules to understand the Ln. Associating capital letters or numbers to understand the Ln
Dutch
‘Anneke a 12 ans. Il y avait marqué 12. C’est de la déduction’. [Anneke is 12. There was written 12. It’s a deduction.]
Italian
‘Paperino et Pippo amici Minnie. Les amis de Minnie -> Majuscules alors *nom *propre’. [Capital letters so it’s a proper noun.]
1.2
0
4
Finnish
‘Soitatko musiikia? Fais-tu de la musique? On a trouvé grâce à la ponctuation (pour les questions)’.
1.14
0.36
10
[We found it thanks to the punctuation (for the questions).]
Crosslinguistic Influence and Crosslinguistic Interaction
Comparison Italian Comparing new language items to Italian
Example
Dutch
‘Je te dis au revoir de la part de ma famille. Logique par rapport à la lettre’.
1.25
0.30
8
1.83
0.53
7
1.73
0.52
12
0.8
1.41
5
3
2.14
15
0.50
0
1
[I’m saying goodbye on behalf of my family. Logical since it is a letter.] Italian
‘Walt Disney. Animateur de dessin animé’. [A cartoon animator.]
Finnish
‘*Sa ressemble à un dialogue ou une pièce de théâtre’. [It looks like a dialogue or a theatre play.]
Translation (verbatim) Rendering ideas from one language to another in a relatively verbatim manner
Dutch
‘Il a une petite sæur qui s’appelle Sarah’. [He has got a little sister whose name is Sarah.]
Italian
‘Chi sono io? Qui suis-je?’ [Who am I?]
Finnish
‘Rap mussiikkia: musique rap’. [rap music]
Developing Cognitive Strategies Through Pluralistic Approaches
Deduction top-down (D+) Building on external knowledge to understand the Ln
57
58
Crosslinguistic Influence and Crosslinguistic Interaction
The results regarding Finnish are quite surprising: 58 per cent of the implemented strategies are comparisons with French (mean = 4.33; SD = 1.25). Even though the mean results are quite high, they do not support the psychotypology hypothesis. As a matter of fact, the students mainly relied upon the cognates present in the text; they only established the common genetics of these words without establishing a linguistic proximity between the language systems. However, sometimes students did not rely on their L1 or L2 but on other languages. During the PAUL session on Italian, 1 per cent of the strategies (implemented by two groups) are a comparison with Spanish, a language they have not yet studied. The students have noted the Roman origin of the languages and draw parallels regarding lexicon. In one case, the reference is accurate (but not explicitly mentioned): ‘Rossi, bianchi = rouge, blanc. C’est très logique car ça ressemble au français et ça ressemble à l’espagnol’. [It’s very logical because it is similar to French and to Spanish.] The other group inaccurately compared the language to Spanish, of which they have no knowledge: ‘molto: ça ressemble à l’espagnol *moltum’ [it is similar to Spanish *moltum]. One can thus consider that, despite their lack of knowledge of the language, the awareness of language families (related to the concept of psychotypology) has enabled them to make such a comparison. The results regarding translation strategies show a specific profile; when observing PAUL sessions in Dutch (5 per cent; Mean = 0.8; SD = 1.41) and Finnish (1 per cent; Mean = 0.50; SD = 0), pupils do not seem to favour this strategy as both the mean value and the standard deviation are low. It also appears that the translation strategy is less used during the last PAUL session, on Finnish. The results for Italian, however, show a higher reliance upon this strategy even though the value scale varies from one group to another (15 per cent; Mean = 3; SD = 2.14).
5.2 Least readily used strategies: Inferencing and deduction The differential measurement (as shown in Figure 3.2) between potentially useful strategies and strategies actually implemented highlights the difficulty in resorting to inferencing and deduction. The inferencing strategy is the least frequently used strategy, but the results show a slight increase over the three sessions, as the highest inferencing rate appears with the Finnish language (14 per cent; Mean = 2.75; SD = 2.04). Students seem to have developed their ability to resort to this strategy. When considering the mean results regarding deduction strategies (whether bottom-up or top-down), they appear rather homogeneous. This is due to the type of texts that were submitted. All three contain proper nouns and/or numbers, which enable the learner to rely directly on the input so as to deduce meaning (bottom-up processes, indicated as ‘Ded−’). They can also rely upon external knowledge, whether it is related to the world or academic (‘Ded+’, top-down processes). One can note the specific situation regarding Italian: The bottom-up processes are less readily implemented in this language. Sometimes, having a more complete understanding of the groups’ behaviour (Table 3.2) usefully complements the analysis of each strategy.
Developing Cognitive Strategies Through Pluralistic Approaches
59
Table 3.2 Percentage of groups having implemented each strategy Comp Comp Comp GB F (%) Latin (%) (%)
Comp Infer. Ded- Ded+ Transl Span (%) (%) (%) (%) (%)
NLS1
% groups
95
41
0
0
45
55
45
23
ItS1
% groups
50
91
9
9
36
27
32
36
FinnS1 % groups
23
100
9
0
55
64
59
9
When observing the percentage of groups having implemented each strategy (Table 3.2), one can see that more and more groups resort to deductions, whether they are bottom-up (55 per cent of the groups resort to this strategy when observing Dutch and 64 per cent when observing Finnish) or top-down (from 45 per cent to 59 per cent). Once again, one can note the differing use of strategy when the students reflect upon Italian, results that will need to be discussed.
6 Discussion When taking the use of strategy into consideration and relating it to the literature at hand, a certain number of remarks regarding the research questions can be made.
6.1 Research question 1: Students’ strategy use Analysing the preferential use of learning strategies makes it possible to examine the role of psychotypology during comparison phenomena as well as the effects of the L1 and/or L2 factors. To complete the analysis and to better understand the choices of the students, the quantitative analysis of the implemented strategies will be complemented by a quantitative analysis showing the number of groups that have made use of the various strategies. This triangulation should allow us to better understand student practices in solving the problem of access to meaning. The students mainly resort to comparison and translation strategies, while strategies of inferencing and deduction seem to be more difficult to implement. Comparison is the most frequently used strategy by the greatest number of groups, but the source language depends on a great number of factors.
6.1.1 Effects of psychotypology Students mainly resort to comparisons with languages that present the greatest linguistic proximity: When observing the Dutch language, 55 per cent of the implemented strategies are comparisons with English. The high rate of comparisons with English, a language typologically close to Dutch (used by 95 per cent of the students) favours the hypothesis of psychotypology (Kellerman 1979). This
60
Crosslinguistic Influence and Crosslinguistic Interaction
is confirmed by the comparison rate of 54 per cent between Italian and French, comparisons made by 91 per cent of students. It seems therefore possible to confirm that the implementation of a strategy of comparison will first of all occur, in a privileged manner, according to the perceived typological proximity between the target language and the source language. When students are confronted with a language that is typologically close to the L1 (Italian compared to French), they tend to rely primarily on comparisons and translations that are readily available strategies (69 per cent of the implemented strategies). It seems that this proximity has an inhibitory effect on the spontaneous implementation of the inferencing and deduction strategies (the lowest rate of the three languages, without guidance from the teacher). The perceived typological proximity to the L1 could therefore affect the transferability of these strategies. This echoes the observation made by Deyrich (2007), who considers that the subjective evaluation of the typological distance between languages changes the behaviour of the learners in lexical transfers.
6.1.2 Effects of the L1 Factor The observation of the results related to the use of comparison strategies when accessing the meaning of the text in Finnish can highlight a predominance of comparisons with French (58 per cent), while only 4 per cent of the strategies are comparisons with English. As Finnish is a Finno-Ugric language with very little resemblance to French or English, the only cognates are internationalisms. However, internationalisms are characterized by their existence in many languages. In addition, the presented internationalisms (pianoa, musiikkia, Internetistä, kitaraa) belong to the students’ vocabulary and had been studied during the English course. The preferential use of comparisons with French by 100 per cent of the students enables us to consider that, regarding lexicon, the L1 factor (Rothman, Iverson and Judy 2011) plays an essential part when there is no typological proximity. This L1 factor, that is to say the preferential use of the mother tongue, may also be influenced by the assumption of Matthey and Véronique (2004): They consider that since the L1 is used to talk about the strategy, it may influence the choice of the source language in an unconscious way, regardless of the target language. The aforementioned L1 factor is confirmed when some students need to indicate the translation into French, even though they first translated the Ln (Dutch) into the L2 (English). It seems that the English language repertoire is not yet sufficiently stabilized for it to reach the status of reference language. The L1 factor could play a major role for students who have not yet reached a minimum level of competence in the L2; if they had, the L2 could have competed with the L1.
6.1.3 Effects of the Ln factor Even if students resort to comparisons when the target and source languages are perceived as typologically close, the L1 factor seems to prevail without any typological proximity. Based on the research of Williams and Hammarberg (1998), Bardel and Falk (2007) posit that when a learner comes into contact with an unknown language
Developing Cognitive Strategies Through Pluralistic Approaches
61
(L3 or Ln), he will be more inclined to use a language other than his L1. Trévisiol (2006: 7), building on the work of Williams and Hammarberg (1998), states that ‘L2s are better candidates than the L1 to be activated as a provider of early language acquisition’8 because of: 1. the fundamental difference between the acquisition mechanisms for L1 and L2. In the acquisition of L3, the same L2 acquisition mechanisms are reactivated, which in turn could reactivate other possible L2s. 2. the desire to erase the L1 which is perceived as inherently ‘non-foreign’ and therefore not suitable as a strategy for acquiring another LE9. The language provider would then be the language that fulfils the requirements of competence, typology, recent use of language and its status as L2, which corresponds to typical findings (for a complete review, please report to the comprehensive study led by De Angelis (2007)). Although only 9 per cent of the groups have made comparisons with third languages (other than L1 or L2) during the three sessions, it is interesting to note that they are spread among six different groups. Using comparison strategies with Latin or Spanish during the sessions involving accessing the meaning of a text in Italian can be justified by psychotypology and the concept of language families. However, two groups performed comparison strategies with Italian (1 per cent of the implemented strategies) during the session on Finnish. Italian was the second unknown language presented in the context of the experiment. And yet, one cannot regard these sessions as involving the teaching/learning of a new language since no active use of Ln (Finnish) occurred. It is likely that due to the absence of perceived typological proximity, the L2 factor (Williams and Hammarberg 1998) played a major role; one could consider, expanding on the concept of ‘foreignness’ proposed by De Angelis (2005), that students have identified the Finnish language as ‘foreign’ so that they have associated it with languages they have an awareness about, without, however, having studied them. For this author, sometimes a trilingual speaker addresses an element from the L2 as belonging to the L3. It can therefore be unconsciously included in L3 production since it is associated with a foreign language. One can imagine such a phenomenon occurring in the implementation of comparison strategies; students build on ‘foreign’ languages (regarded as ‘foreign’ to their usual linguistic environment) to access the meaning of a text in a novel unknown language. Students could thus establish stronger cognitive links between the various non-native systems in use and the mother tongue. Although the frequency rate is low, this result still raises the hypothesis of the L2 factor (which is also named the Ln factor so as to encompass all additional languages). Since 9 per cent of the groups have implemented a comparison strategy with third languages, it does not seem possible to discard the influence of the Ln factor on the choice of source language. One can therefore rightly raise the question of the status of English as an L2.
6.1.4 Special status of English as an L2 Taking into account the L2 factor theory (Williams and Hammarberg 1998), English as the L2 of these students should also benefit from the L2 effect: Students should
62
Crosslinguistic Influence and Crosslinguistic Interaction
have resorted to their knowledge of the English language when confronted with an unknown language, all the more so when there is no linguistic proximity. However, the results here do not seem to support such a hypothesis. It seems that English has attained a special status within the language system of the students. Even if it has been proven that with multilinguals a transfer can take place from a language that has been studied for only a year or two (Rivers 1979; Selinker and Baumgartner-Cohen 1995; De Angelis 1999, 2005), the results here seem to underline the fact that these ‘monolingual’ (in their home environment) students have not yet begun to behave in a similar way to a multilingual, even though they have been studying English for four years. The results of these authors only seem to be verified with a recently studied (or observed) Ln which can be used as a source language when observing an Ln+1. When considering the results in the aforementioned experiment, an assumption can be made about the special status of the languages taught at school; students do not seem to consider such languages in the same way as an unknown language that is presented to them. Therefore, one can assume that the teaching/learning of a language can also play a role in the selection of the source language; learners do not consider languages they study at school and unknown languages they discover during PAUL sessions in the same way. They can, however, reactivate certain skills developed during L2 teaching/ learning situations.
6.1.5 Building upon skills developed during the learning of the L2 (English) The inferencing strategy is the least frequently used strategy (see Figure 3.2); however, the results show a slight increase over the three PAUL sessions. Only 10 per cent of the declared strategies used to access the meaning of the text in Dutch involve inferencing, coming down to 9 per cent for Italian, even though the frequency raises to 14 per cent for Finnish (see Figure 3.1). For Bialystok (1980), the inferencing strategy can only be used if the student has achieved a minimum level of competence in his L2. In this experiment, students are confronted with unknown languages and are therefore very far from the minimum level of proficiency in these languages. And yet, some students manage to implement this strategy in a conscious way. It is therefore possible to hypothesize that they rely on the skills developed during their L2 learning. So, following the theory defended by Bono (2008b), it is noticeable that the learning dynamics related to the school environment can have a major impact on the transfer of linguistic elements, since learners tend to rely on their previous experience in learning a foreign language. In the present experiment, students who manage to implement an inferencing strategy have therefore probably developed this strategy while learning English (L2). It is quite remarkable that nearly half of the groups (44 per cent) who implemented an inferencing strategy when accessing the meaning of the text in Finnish had never been able to use this strategy before. One can therefore assume that this strategy has been used to overcome the lack of transparency presented by a language typologically distant from the L1 and/or L2.
Developing Cognitive Strategies Through Pluralistic Approaches
63
6.1.6 Relationship between top-down and bottom-up processes The graph showing the difference between the reported and expected strategies (see Figure 3.2) also underlines a deficit of the deduction strategy, although to a lesser extent. This deficit is particularly pronounced for Italian, as it is at 9 per cent for bottom-up deductions and 6 per cent for top-down deductions. During the PAUL session on Dutch, 55 per cent of the students were able to identify textual elements contributing to their understanding of the text (see Table 3.2). However, only 27 per cent were able to implement bottom-up deduction strategies for Italian against 64 per cent for Finnish. A similar distribution for top-down deduction strategies can be observed: 45 per cent of the students use this strategy to access the meaning of the text in Dutch while only 32 per cent use it for Italian. This rate exceptionally rises to 59 per cent during the PAUL session on Finnish. Students who cannot readily infer meaning may rely upon contextual and co-textual elements because their decoding abilities are underdeveloped. They might thus seek to compensate for problems of immediate accessibility of the text by setting up topdown processes. This concurs with the research led by Perfetti (1985), who focused on reading comprehension, as well as with the work by Tsui and Fullilove (1998) on spoken English. However, the results appear to contradict the theory of the bottom-up dependency elaborated by Field (2004: 367) in reference to Cummins’ research (1979); the latter considered that students need to have a minimum level of proficiency in order to be able to use high-level processes. This view is confirmed by Clarke (1980), who states that the poor readers’ attention is so focused on decoding that they are unable to transfer to the L2 the top-down processes they would naturally resort to in their mother tongue. Stanovich (1980), however, considers that the relationship between top-down and bottom-up information is controlled by an interactive-compensatory mechanism. The results here seem to confirm this hypothesis since it appears that topdown information are used in a compensatory way, not to reinforce the meaning that has already been extracted from a text but to restore the parts that have not been fully understood. One can thus conclude by saying that, in this study, the choice of strategy mainly depends on the perceived linguistic distance between the source language and target language (leading to the favoured strategies of comparison and translation). It will then vary according to the textual elements at hand (for inferencing and top-down or bottom-up deduction processes). However, the least readily implemented strategy is inferencing. Considering the transfer of these strategies should allow a better understanding of the results at hand.
6.2 Research question 2: Transfer of strategies So far, the findings here have shown what kind of strategies are more or less implemented and some of the underlying processes that might prove useful for a better understanding of the implementation. These processes also play a part in the transferability of strategies as they will enable the student to activate the schemata they have previously been confronted with.
64
Crosslinguistic Influence and Crosslinguistic Interaction
6.2.1 Influence of the L2 on the transfer of strategies The least immediately accessible strategy is inferencing. Students use this strategy when confronted with a language typologically close to the studied L2; one can assume that this strategy was implemented during the learning of English and has been transferred to the new situation. Bono (2008b) notes that the learning dynamics related to the school environment can have a major impact on the transfer of linguistic elements, as learners tend to rely upon previous experience acquired during the learning of a foreign language. The results here tend to confirm that learning dynamics also influence the transfer of strategies. It seems that the skills developed during the learning of the L2 are transferable to an unknown language, provided a minimum threshold of competence has been reached in the L2. However, this strategy is less used during the PAUL session on Italian since the latter does not appear as having triggered the resort to strategies used during the learning of the L2. The perceived typological proximity with the L1 could affect the transferability of these strategies, thus agreeing with the concept of psychotypology (Kellerman and Sharwood-Smith 1986). By expanding this hypothesis (drawn from observations made during production activities) to our situation (comprehension activities), it seems possible to assert that the learning strategies developed during the L1 are transferred to the Ln that presents a psychotypological proximity. Therefore, when students face languages they consider close to the L1 (French), they mainly rely upon comparison or translation strategies, which are immediately accessible strategies. It appears that the proximity to the L1 may have an inhibitory effect on other strategies such as inferencing and deduction, when not guided by the teacher. That is why, contrary to what has proven useful in research on transfer of lexical or morpho-syntactic items, the L1 may negatively affect the transfer of strategies. This research shows that more and more groups resort to deduction strategies, whether they are bottom-up or top-down process oriented. Even though the number of groups resorting to inferencing does not increase, the overall use of inferencing strategies has improved. This tends to demonstrate that inferencing is probably the strategy that most requires training and is not easily transferred from one language to another, contrary to the strategy of deduction.
6.2.2 Type of transferred knowledge In one of my previous articles (Dahm 2014), I analysed the type of knowledge that is being mobilized during PAUL sessions. The results showed that the strategies of comparison and translation mainly relied upon the conscious mobilization of declarative knowledge, whereas strategies of inferencing and deduction also required procedural knowledge, which is more difficult to mobilize. Students sometimes resorted to deduction by building upon their knowledge of grammatical rules. The fact that they were sharing this knowledge with their peers in the group turned their procedural knowledge into explicit declarative knowledge. It thus seems possible to confirm that the availability of explicit declarative knowledge is essential to the transfer of learning strategies; by becoming aware of their procedural, conditional knowledge, students are able to verbalize the implemented learning strategies which enable them to generalize the new knowledge to unfamiliar contexts.
Developing Cognitive Strategies Through Pluralistic Approaches
65
6.2.3 Characterization of the situations allowing transfer The three sessions that have been analysed present a structural similarity through their organization, implementation and main objective (access to meaning problemsolving). Even if the characteristics of the contingent situations do differ somewhat, their surface features are similar enough to allow spontaneous recognition of an analogue situation, which is an essential feature enabling transfer, as shown by Holyoak and Koh (1987). It is, indeed, essential for a transfer to take place – that the learner is able to select the source analogue. It is the structural similarity between the different PAUL sessions that enabled the students to gradually increase their consciousness of the implemented strategies and to resort to the inferencing strategy during the last PAUL session in Finnish. When students are confronted with repetitive PAUL sessions, one can consider that they are transferring previous knowledge to new situations that are closely similar (surface features similarity). Near transfer seems to have occurred when reflecting upon the second or third unknown language (Italian and Finnish). However, so as to be able to analyse far transfer and/or creative transfer, the surface features need to differ. This can be studied by taking into account the results of the post-tests and survey.
6.2.4 From near transfer to creative transfer The analysis of the post-survey underlines the fact that only 27 per cent of the pupils declare having personally used the learning strategies developed during the PAUL sessions in the English class, since the teachers never explicitly referred to them. However, the comparison of the semantic pre- and post-tests10 (of both the experimental and control group) allows us to make a few remarks. In both tests, pupils were asked to understand difficult texts in English (level C1 of the CEFR) and to explain what they understood. The results highlight the transfer of learning strategies to the L2 as existing, mainly for the comparison strategy (+8 per cent for the experimental group/16 per cent for the control group) and for the inferencing strategy (+21 per cent for the experimental group/+15 per cent for the control group). Students also seem to be more willing to try and summarize the text, since 16 per cent of the pupils in the experimental group (against 6 per cent of the control group) proposed a summary of the meaning of the text in the post-test. However, this change of attitude leads to greater risk taking, making them sometimes infer contradictory meaning. Despite a lack of similarities between the test situations and the PAUL sessions, students seem to have transferred their learning strategies to a certain extent, however unconsciously. It seems possible, therefore, to consider that a certain amount of far transfer has taken place, since the strategies were applied in situations quite dissimilar to the initial learning. The post-survey shows that 37 per cent of the students declare having used the learning strategies outside school to try and understand new information in a different context. For instance, they have tried to understand labels on food products or leaflets in foreign airports. Even though no explicit link has been established, one can consider that for some of the students, creative transfer took place. They have been able to activate their metalinguistic skills autonomously, without any obvious similarities between the learning situations.
66
Crosslinguistic Influence and Crosslinguistic Interaction
7 Conclusion Through these pluralistic approach sessions on languages, where there are no learning goals, students have built abstract and unconscious schemata enabling them to solve the problem and thus develop cognitive strategies. Most of these strategies were successfully transferred from one session to another, as students managed to establish connections between different linguistic situations. For the strategic benefits to be consciously transferred to the L2 they are studying, it seems necessary to make the students perform an activity that allows them to organize their experiences so as to bridge the gap between pluralistic approaches and the subject L2. This could be facilitated by introducing problem situations rooted in the English course that have structural similarities to the PAUL sessions that have been presented. It is, indeed, essential for a transfer to take place that students subjectively recognize common characteristics to the source task and target task. Only by doing so will they be able to fully benefit from pluralistic approaches and eventually increase creative transfer, whether consciously or unconsciously (Odlin 1989).
Notes 1 Metalinguistic skills are strategies that are applied, either consciously or automatically, to an oral or written linguistic interaction to allow one to think about language and a linguistic message, to analyse a message and to control language processing within the communicative culture (Bialystok, 1987). 2 I chose to specifically study those students who declared (in the pre-experiment survey) that they had no contacts in their direct environment with languages other than their L1 French and the L2 (English) being taught at school. The term ‘typically monolingual’ is a shortcut used to distinguish this school population from the multilingual students who declare being in contact with languages other than French (and English). However, one cannot consider that these children have no knowledge at all of the existence of other languages. 3 The tasks can be metasemantic if their purpose is to reflect on the meaning of the text, metasyntactic if the student is encouraged to reflect on grammatical elements, or metaphonological when he is led to reflect upon phonology. 4 We choose to name the different languages the learner is presented with according to a chronological order: the L1 therefore is his mother tongue, the L2 the first foreign language, the L3 the second one, etc. 5 ‘Having the same linguistic derivation as another’ (Oxford dictionary, consulted on 10 November 2013. Available online: http://www.oxforddictionaries.com/definition/ english/cognate). 6 CEFR: Common European Framework of Reference for Languages (Council of Europe 2001). 7 Three multilingual experts (speaking at least three languages different from the ones presented in the PAUL sessions) tried to solve the access to meaning problem and stated the strategies they used. The texts were then subsequently modified so as to present a similar number of implemented strategies, thus becoming comparable, whatever the language of the PAUL session. This only gives us an idea of what strategies seem most plausible for use, and is in no case a prescriptive mode.
Developing Cognitive Strategies Through Pluralistic Approaches
67
8 Our translation of Trévisiol (2006: 7): ‘les L2 sont de meilleurs candidats que la L1 pour l’activation en tant que langue fournisseur en début d’acquisition’. 9 Our translation of Trévisiol (2006: 7): ‘1. la différence fondamentale entre les mécanismes d’acquisition pour la L1 et la L2. Dans l’acquisition d’une L3, les mêmes mécanismes d’acquisition qu’en L2 sont réactivés, et ils réactiveraient à leur tour d’autres L2 éventuelles. 2. le désir d’effacer la L1 qui est perçue comme intrinsèquement ‘non étrangère’, donc non adéquate comme stratégie d’acquisition d’une autre LE’. 10 See Appendix 1 for the results.
References Anderson, J. R. (1980), Cognitive Psychology and its Implications (2000), 5th edn, New York: Worth. Anderson, J. R. (1993), ‘Problem solving and learning’, American Psychologist, 48 (1): 35. Bardel, C. and Y. Falk (2007), ‘The role of the second language in third language acquisition: the case of Germanic syntax’, Second Language Research, 23 (4): 459–84. Bialystok, E. (1980), On the relationship between formal proficiency and strategic ability. Presented at the Annual Meeting of TESOL, March 1980, San Francisco, CA. Bialystok, E. (1987), ‘Words as things: Development of word concept by bilingual children’, Studies in Second Language Acquisition, 9: 133–40. Bono, M. (2008a), ‘“Quand je parle en langue étrangère, je parle anglais”. Conscience métalinguistique et influences interlinguistiques chez les apprenants plurilingues’, in M. Candelier (ed.), Conscience du plurilinguisme. Pratiques, représentations et interventions, 93–107, Rennes: Presses Universitaires de Rennes. Bono, M. (2008b), ‘Ressources plurilingues dans l’apprentissage d’une troisième langue : aspects linguistiques et perspectives didactiques’, Ph.D. diss., Université Sorbonne Nouvelle – Paris 3 école doctorale 268 – langage et langues: description, théorisation, transmission, Paris. Calais, G. J. (2006), ‘Haskell’s taxonomies of transfer of learning: Implications for classroom instruction’, National Forum of applied Educational Research Journal, 20 (3): 1–8. Candelier, M. (2008), ‘Approches plurielles, didactiques du plurilinguisme: le même et l’autre’, Les Cahiers de l’Acedle, 5 (1): 65–90. Cenoz, J., B. Hufeisen and U. Jessner (2001), Cross-linguistic Influence in Third Language Acquisition, Clevedon: Multilingual Matters. Clarke, M. (1980), ‘The short circuit hypothesis of ESL reading – or when language competence interferes with reading performance’, Modern Language Journal, 64: 203–09. Clyne, M. (2003), Dynamics of Language Contact, Cambridge: Cambridge University Press. Council of Europe (2001), Common European Framework of Reference For Languages, Cambridge, UK: Cambridge University Press. Cummins, J. (1991), ‘Interdependence of first- and second-language proficiency in bilingual children’, in E. Bialystok (ed.), Language Processing in Bilingual Children, 70–89, Cambridge: Cambridge University Press. Dahm, R. (2014), ‘Les approches plurielles, vecteurs du développement de la compétence métalinguistique des collégiens’, Cahiers de l’APLIUT, Vol. XXXIII n°2/2014.
68
Crosslinguistic Influence and Crosslinguistic Interaction
De Angelis, G. (2005), ‘Multilingualism and non-native lexical transfer: Aan identification problem’, International Journal of Multilingualism, 2 (1): 1–25. De Angelis, G. (2007), Third or Additional Language Acquisition, Clevedon: Multilingual Matters. De Angelis, G. and L. Selinker (2001), ‘Interlanguage transfer and competing linguistic systems in the mutlilingual mind’, in J. Cenoz, B. Hufeisen and U. Jessner (eds), Crosslinguisitc Influence in Third Language Acquisition: Psycholinguistic Perspectives, 42–58, Clevedon: Multilingual Matters. Deyrich, M.-C. (2007), Enseigner les langues à l’école, Paris: Ellipses. Ellis, R. (1994), The study of Second Language Acquisition, Oxford: Oxford University Press. Ellis, R. (2008), The Study of Second Language Acquisition, 2nd edn, Oxford: Oxford University Press. European Center for Modern Languages (2007), ‘Pluralistic approaches to languages and cultures’, FREPA/CARAP a Framework of Reference for Pluralistic Approaches. Available online: http://carap.ecml.at/Keyconcepts/Pluralisticapproachestolanguagesandcultures/ tabid/2683/language/en-GB/Default.aspx (accessed 5 January 2011). Field, J. (2004), ‘An insight into listeners’ problems: Too much bottom-up or too much top-down?’, System, 32: 363–77. Finkbeiner, C. (2005), Interessen und Strategien Beim Fremdsprachlichen Lesen: Wie Schülerinnen und Schüler Englische Texte lesen und Verstehen, Tübingen: Narr. Finkbeiner, C., M. Knierim, M. Smasal and P. Ludwig (2012), ‘Self-regulated cooperative EFL reading tasks: Students’ strategy use and teachers’ support’, Language Awareness, 21 (1–2): 57–83. Flavell, J. H. (1979), ‘Metacognition and cognitive monitoring: A new area of cognitivedevelopmental inquiry’, American Psychologist, 34: 906–11. Gagné, E. D. (1985), The Cognitive Psychology of School Learning, Boston: Little, Bown and Company. Gajo, L. (2001), Immersion, bilinguisme et interaction en classe, Paris: Didier. Gentner, D. and A. B. Markman (2006), ‘Defining structural similarity’, The Journal of Cognitive Science, 6 (1): 1–20. Gick, M. L. and K. J. Holyoak (1983), ‘Schema induction and analogical transfer’, Cognitive Psychology, 15: 1–38. Gick, M. L. and K. J. Holyoak (1987), ‘The cognitive basis of knowledge transfer’, in S.-M. Cormier and A. Giddens (eds), La constitution de la société, Paris: Presses Universitaires de France. Gombert, J. E. (1990), Le développement métalinguistique, Paris: Presses Universitaires de France. Hammarberg, B. (2001), ‘Roles of L1 and L2 in L3 production and acquisition’, in J. Cenoz, B. Hufeisen and U. Jessner (eds), Cross-linguisitc Influence in Third Language Acquisition: Psycholinguistic Perspectives, 21–41, Clevedon: Multilingual Matters. Haskell, R. E. (2001), Transfer of Learning: Cognition, Instruction and Reasoning, San Diego: Academic Press. Herdina, P. and U. Jessner (2002), A Dynamic Model of Multilingualism. Perspectives of Change in Psycholinguistics, Clevedon: Multilingual Matters. Holyoak, K. J. and K. Koh (1987), ‘Surface and structural similarity in analogical transfer’, Memory and Cognition, 15 (4): 332–40. Howard, D. (1985), Cognitive Psychology, New York: Macmillan. Hulstijn, J. and W. Hulstijn (1984), ‘Grammatical errors as a function of processing constraints and explicit knowledge’, Language Learning, 34 (1): 23–43.
Developing Cognitive Strategies Through Pluralistic Approaches
69
Kellerman, E. (1977), ‘Towards a characterization of the strategy of transfer in second language acquisition’, The Interlanguage Studies Bulletin, 2 (1): 58–145. Kellerman, E. (1979), ‘Giving learners a break: native language intuitions as a source of predictions about transferability’, Working Papers on Bilingualism, 15: 37–57. Kellerman, E. (1983), ‘Now you see it, now you don’t’, in S. Gass and L. Selinker (eds), Language Transfer in Language Learning, 112–34, Rowley, MA: Newbury House. Kellerman, E. and M. Sharwood-Smith (1986), Crosslinguistic Influence in Second Language Acquisition, Oxford: Pergamon Press. Kokinov, B. and R. M. French (2003), ‘Computational models of analogy-making’, in L. Nadel (ed.), Encyclopedia of Cognitive Science. Vol. 1, 113–18, London: Nature Publishing Group. Lado, R. (1957), Linguistics Across Cultures: Applied Linguistics for Language Teachers, Ann Arbor: University of Michigan Press. Marzano, R. J., R. S. Brandt, C. S. Hughes, B. F. Jones, B. Z. Presseisen, S. C. Rankin and C. Suhor (1988), Dimensions of Thinking: A Framework for Curriculum and Instruction, Alexandria, VA: Association for Supervision and Curriculum Development. Matthey, M. and D. Véronique (2004), ‘Trois approches de l’acquisition des langues étrangères: enjeux et perspectives’, Acquisition et interaction en langue étrangère, 21: 203–23. Mehrnoosh, F., Y. Manijah and N. Daryoush (2013), ‘Proceduralization, transfer of training and retention of knowledge as a result of output practice’, International Journal of research Studies in Language Learning, November 2013: 1–14. Moore, D. (2006), ‘Plurilingualism and strategic competence in context’, The International Journal of Multilingualism, 3 (2): 125–38. O’Malley, J. M. and A. U. Chamot (1990), Learning Strategies in Second Language Acquisition, Cambridge: Cambridge University Press. Odlin, T. (1989), Language Transfer: Cross-Linguistic Influence in Language Learning, Cambridge: Cambridge University Press. Perfetti, C. A. (1985), Reading Ability, Oxford: Oxford University Press. Perkins, D. N. and G. Salomon (1988), ‘Teaching for transfer’, Educational Leadership, 46 (1): 22–32. Prabhu, N. S. (1987), Second Language Pedagogy, Oxford: Oxford University Press. Rabinotwitz, M. and M. T. Chi (1987), ‘An interactive model of strategic processing’, in S. J. Ceci (ed.), Handbook of Cognitive, Social and Neuropsychological Aspects of Learning Disabilities, 83–102, Hillsdale, NJ.: Erlbaum. Ringbom, H. (2007), Cross-linguistic Similarity in Foreign Language Learning, Clevedon: Multilingual Matters. Rivers, W. M. (1979), ‘Learning a sixth language: An adult learner’s daily diary’, The Canadian Modern Language Review, 36 (1), 67–82. Rothman, J., M. Iverson and T. Judy (2011), ‘Introduction: Some notes on the generative study of L3 acquisition’, Second Language Research, 27 (1): 5–19. Schoenfeld, A. H. (1985), Mathematical Problem Solving, New York: Academic Press. Selinker, L. and B. Baumgartner-Cohen (1995), ‘Multiple language acquisition: “Damn it, why can’t I keep these two languages apart?”’, Language, Culture and Curriculum, 8 (2): 115–21. Sharwood-Smith, M. (1994), Second Language Learning: Theoretical Foundations, London: Longman. Sharwood-Smith, M. and E. Kellerman (1986), ‘Crosslinguistic influence in second language acquisition: An introduction’, in Kellerman, E. and M. Sharwood-Smith (eds), Crosslinguistic Influence in Second Language Acquisition, 1–9, New York: Pergamon Press.
70
Crosslinguistic Influence and Crosslinguistic Interaction
Singleton, D. (1987), ‘Mother and other tongue influence on learner French: A case study’, Studies in Second Language Acquisition, 9: 327–46. Sorace, A. (1985), ‘Metalinguistic knowledge and use of the language in acquisition poor environments’, Applied Linguistics, 6 (3): 239–54. Speelman, C. P. and K. Kirsner (2005), Beyond the Learning Curve: the Construction of Mind, Oxford: Oxford University Press. Stanovich, K. E. (1980), ‘Toward an interactive-compensatory model of individual differences in the development of reading fluency’, Reading Research Quarterly, 16: 32–71. Trévisiol, P. (2006), Influence translinguistique et alternance codique en français L3. Rôles des L1 et L2 dans la production orale d’apprenants japonais. Acquisition et interaction en langue étrangère, (24): 13–43. Tsui, A. and J. Fullilove (1998), ‘Bottom-up or top-down processing as a discriminator of L2 listening performance’, Applied Linguistics, 19: 432–51. Webb, N. M. (1989), ‘Peer interaction and learning in small groups’, International Journal of Education, 13 (1): 21–40. Weinstein, C. E. and R. E. Mayer (1986), ‘The teaching of learning strategies’, in M. C. Wittrock (ed.), Handbook of Research on Teaching, 3rd edn, 321–7, New York: Macmillan. Williams, S. and B. Hammarberg (1998), ‘Language switches in L3 production: Implications for a polyglot speaking model’, Applied Linguistics, 19 (3): 295–333.
4
Can a Background Language Alter the Path of Acquisition of Verb Placement in a Multilingual Context? Evidence from a Longitudinal Study Laura Sánchez
1 Introduction The extent to which the transfer of functional categories and their feature values is permitted, especially at the initial stages of acquisition of an L2 or an L3, has been the focus of much research. This is essentially the case within the principles and parameters theory and also within MOGUL (modular online growth and use of language, Sharwood Smith and Truscott 2005, 2006, 2008; Truscott and Sharwood Smith 2004), which combines activation-based psycholinguistic explanations with mainstream theories of language transfer (Sánchez 2011; Sharwood Smith and Truscott 2005, 2006). Things become more complicated when two or more non-native languages come into contact, especially at the very beginning of the L3/Ln acquisition process, a moment that is commonly referred to as the initial state (ex. Foote 2009; García Mayo and Rothman 2012; Leung 2003, 2005; Schwartz and Eubank 1996) or the multilingual initial state (Cabrelli, Iverson and Rothman 2008). A critical issue in the investigation of transfer in multilingual contexts is whether the L1 steady state at the initial state of acquisition of an Ln (i.e. a third or additional language, henceforth L3A) reflects the L1 steady state or, alternatively, ‘the steady state of a previously acquired (inter)language’ (Leung 2005: 40). Hence, different explanations have been proposed in order to determine both the source language of transfer and the extent of this transfer, as will be seen in the forthcoming paragraphs. In the field of L3A, the specific proposals that have been made with the aim of determining the source and extent of transfer are the L2 status model (Falk and Bardel 2010, 2011; Bardel and Falk 2007), the typological primacy model (Rothman 2010, 2011) and the cumulative enhancement model (Vinnitskaya, Flynn and Foley 2003; Flynn, Foley and Vinnitskaya 2004). The L2 status hypothesis has its roots in the second language factor (also known as the ‘L2 factor’ or the ‘status of the interlanguage’ factor), which refers to the general tendency to activate an L2 over the L1 and to transfer from this language (Dewaele 1998; Hammarberg 2001; Leung 2007; Shanon 1991; Williams and Hammarberg 1993, 1998) and explains transfer in various linguistic domains such
72
Crosslinguistic Influence and Crosslinguistic Interaction
as lexis (see contributions in Cenoz, Hufeisen and Jessner 2001, 2003) and morphology (Cabrelli, Flynn and Rothman 2012; Sánchez in press). As regards transfer in syntax, Falk and Bardel (2010) defined the L2 status hypothesis as the transfer of the underlying morphosyntax of an L2 to the syntactic representation of the L3. Thus, they further claim, ‘syntactic structures are more easily transferred from L2 than from L1 in the initial state of L3 acquisition’ (Bardel and Falk 2007: 459; see also Falk and Bardel 2011), thereby superseding the effect of (psycho)typology. This superseding effect of the L2 is not recognized in the typological primacy model (TPM), according to which the role of typological proximity and structural similarity is more determining in the occurrence of transfer (Rothman 2010, 2011). The Achilles’ heel in Rothman’s model is its failure to make predictions for the occurrence of transfer between typologically unrelated languages. Moreover, the languages investigated within this framework are confined exclusively to Spanish and Brazilian Portuguese, which are very similar in general terms and almost perfectly congruent in a number of syntactic properties. Therefore, the vast typological relatedness between Spanish and Brazilian Portuguese is a limitation of the model. Given that Italian is systematically the L1 of the learners investigated, while Spanish or Brazilian Portuguese are the L3s in these studies, it comes as no surprise that the L2 English is not the supplier of morphosyntactic information. As in the studies by Rothman above, Montrul, Dias and Santos (2011) also include Spanish and English in their research design. However, these designs differ in the status of these two languages. Whereas in Rothman (2010, 2011) the L1 is Spanish and the L2 is English, in Montrul, Dias and Santos, the order is the reverse (L1 Spanish, L2 English). Not surprisingly, learners in the two studies preferred Spanish over English as the source language of transfer irrespective of the position that Spanish occupied in their linguistic repertoire (see also Rothman and Cabrelli 2010). This was taken by the authors as evidence in support of rejecting the L2 status hypothesis. A somewhat different view is adopted in the cumulative enhancement model (CEM), which lends just partial support to the role of the L2. The authors claim that ‘the last learned language determines the next language learned in some sense’ (Vinnitskaya, Flynn and Foley 2003: 13), which shows some overlap between the factors L2 status and recency (i.e. recency of acquisition). However, their model assumes only positive influence, as it states that all languages in the linguistic background of a learner are alike candidates to be providers of grammatical properties (Flynn, Foley and Vinnitskaya 2004: 5) because the process of language learning is cumulative. The results of the transfer found in the language triad L1 Kasakh, L2 Russian and L3 English cannot be accounted for by the L2 status hypothesis, but the typology factor cannot be played down either. This is because the L2 and the L3 are relatively closer to each other than the L1 and the L3 or the L1 and the L2, at least in terms of verb phrase (VP)-headedness. In particular, whereas the L1 is a head final language, both the L2 and the L3 are head initial. Therefore, the predominant role of the L2 in the studies conducted in CEM cannot be explained by the L2 factor as much the same way as by typology. Mixed results for L2 status are also reported in Jaensch (2011) and in Weigl (2000). In turn, support for a privileged role of the L1 is given only in two studies, namely, Na Ranong and Leung (2009) and Hermas (2010). In the case of Na Ranong and
Can a Background Language Alter the Path of Acquisition
73
Leung (2009) this is not unexpected, given that the language constellation investigated was L1 Thai, L2 English, L3 Chinese, where the L1 and the L3 bear greater resemblance to each other than the L2 and the L3. Somewhat more unanticipated are the findings in Hermas (2010) on the acquisition of the verb movement parameter in L3 English by adult Arabic-French bilinguals. While the L3 disallows Subject-Verb-AdverbObject scrambled orders, the L2 French allows them. Crucially, the learners’ L1 Arabic allows Subject-Adverb-Verb-Object orders. Therefore, the influence of the L1 might be caused by a structural overlap in the feature examined, regardless of the larger similarities between the L2 French and the L3 English at other linguistic levels. Further support for the L2 status hypothesis on the occurrence of transfer in L3A is found in two studies on the acquisition of features related to V2, namely Sellner (1994) and Yuan (2000). Yuan (2000) confirmed the advantage of L2 status over typology, as concluded from the absence of raising with thematic verbs in the L3 Chinese of French learners with prior knowledge of L2 English. Along similar lines, Sellner (1994) found evidence of transfer from L2 English in the position of the verb in VP in L3 German embedded clauses produced by Japanese university learners. As mentioned in the opening lines of this chapter, transfer at the (multilingual) initial state is a critical issue. From this it follows that different extents of transfer are permitted, as licensed by the availability or non-availability of functional categories associated with feature strength (that is, inflectional phrase or IP and complementizer phrase or CP) that can be transferred from the L1 or from a prior non-native language. Schwartz and Sprouse (1994, 1996) and Sprouse and Schwartz (1998) adhere to a strong view of transfer, namely, the absolute L1 influence hypothesis (i.e. Schwartz 1998) or full transfer/full access model (FTFA). The FTFA, which also underlies MOGUL, postulates that ‘the entirety of the L1 grammar’ (Schwartz and Sprouse 1996: 40) is transferred from the native language, including functional projections. In turn, the entire L1 or L2 grammar constitutes the initial state, which is only restructured when the learner is confronted with target language input that does not match the grammar transferred from the L1 or the L2 (Foley 1999; Schwartz 1998; Schwartz and Eubank 1996; White 2000). Under the FTFA, both lexical and functional projections are available in the L2 or L3 grammar from early on as a result of L1 and/or L2 transfer and full access to parametric values (Leung 2003, 2005, 2006, 2007). Another strong view of transfer is the one adopted in Hawkins and Chan’s (1997) and Hawkins’ failed functional features hypothesis (FFFH), which postulates that functional projections are acquired later than lexical projections. This hypothesis comfortably accommodates optionality in early grammars, that is, the optional manifestation of functional categories transferred from the L1 (Sorace 2003: 141) or an L2. According to weaker views of transfer (Eubank 1993, 1994 1996; Vainikka and Young-Scholten 1994, 1996), functional categories are absent from early grammars, because at the initial state they would not be projected in learners’ interlanguage (due to inert features and lack of verb movement). Hence, at the initial state, only the headedness of the VP (that is, head initial or head final) would be transferred. In Vainikka and Young-Scholten’s (1994) minimal trees hypothesis (MTH) and Eubank’s (1993, 1994) weak parametric transfer hypothesis (WPTH), functional projections do not transfer to early grammars. Vainikka and Young-Scholten (1994) argue that evidence for transfer of functional features would be found if learners transferred a
74
Crosslinguistic Influence and Crosslinguistic Interaction
head final IP or CP; that is, if the data could confirm the presence of inflection or complementizer material (i.e. forms with tense or agreement marking) at the end of the sentence (p. 311), which was not the case in the German interlanguage of Turkish speakers they analysed. Crucially, all of the weak and strong views of transfer reviewed in the preceding paragraphs acknowledge the transfer of VP headedness (but cf. Epstein, Flynn and Marthohardjono’s 1996 Full Access/No Transfer view). If so, whether it is transferred from the L1 or the L2 is an issue that has not been addressed in the literature on L3A. Thus, no specific hypotheses have been put forward predicting the availability and transfer of functional projections and feature strength. Instead, in the acquisition of an L3, the underlying assumption is that the entirety of the grammar is transferred at the initial state, as derived from the FTFA, and preferably from an L2 (Bardel and Falk 2007; Falk and Bardel 2010, 2011; Grüter and Conradie 2006; Leung 2003, 2005, 2006, 2007, 2008; Sánchez 2011). Following the MTH, another possibility would be that transfer comes both from the L1 and the L2, that is, a case of combined influence (De Angelis 2007) as in the L3 German of L1 American-English learners with various Romance languages as L2s (Vainikka and Young-Scholten 2002). More importantly, in language pairs and triads that include different headed languages, the transfer of head final VPs represents a solid base for the argument of an alteration in the path of acquisition of a given target language, especially where there are parametric differences between source and target languages (Gass and Ard 1980; Rutherford 1983, 1984; Schachter 1974, 1983; Schachter and Rutherford 1979). The FTFA, MTH and WPTH hypotheses described above argue in favour of such an alteration, at least at the initial stages of acquisition of an L2 or an L3. As regards VP headedness, the FTFA would predict the transfer of the headedness of head final VPs along with functional projections (Sánchez 2011). In turn, the MTH and WPFT would claim that ‘if the VP headedness in the learners’ native language does not match the headedness of the TL, then acquisition can be seen to proceed in two substages’ (Vainikka and Young-Scholten 1996: 160); that is, a transfer stage where the learner adopts the headedness of VP from the L1 or an L2, and an adjacent stage where the first signs of a functional projection can be found even if it is not yet productive. From a linguistic perspective, this approach to transfer is based on an interpretation of transfer and developmental influences not as opposite processes but as interacting ones, and they result in the overgeneralization of forms or structures (Bohnacker 2006; Bohnacker and Rosén 2007a, b, 2008; Corder 1974; Schachter and Rutherford 1979; Selinker, Swain and Dumas 1975). The interplay between transfer and overgeneralization is echoed in the following quotation: It is not reasonable to classify all errors as results of cross-linguistic errors as there might be developmental errors that are reflecting the characteristics of the language acquisition process or overgeneralizations. As stated by Jordens (1977, cited in Kellerman 1992) the distinction between an overgeneralization and a transfer error in any case is a vacuous one, since it is logically possible for one and the same form to have multiple psychological origins, which to some extent reinforce each other (Çiğdem 2006: 60, emphasis added).
Can a Background Language Alter the Path of Acquisition
75
This interplay between transfer of parametric values and overgeneralization of structures may cause an alteration in the path of acquisition of VP headedness. Indeed, empirical evidence of transfer affecting VP headedness caused by a contrast in head directionality has been reported in the literature (see below) in various language pairs, including English and German, as in the study that will be presented in Section 2. The main parametric difference between English and German lies in the feature value of the functional categories IP and CP. Whereas in English this value is strong, in G erman it is weak (White 2000), and the contrast has consequences for the directionality of headedness. In terms of headedness, German is commonly classified as head final1 and English as head initial. As a consequence, complements in English are post-verbal (i.e. they are placed after the head of the VP because the government linearization in the syntax of these languages is unidirectional), while in German they are pre-verbal. These parametric values account for a range of differences in deep and surface structure. As a result of the head final value in German, most complements are preverbal, as seen in the discontinuous verb placement or verb separation (SEP) in main clauses (ex. 1–4), and verb final (VFINAL) in embedded clauses (ex. 5–6). The SEP rule in main clauses separates the finite and non-finite verbs in periphrastic constructions. It affects thematic verbs such as tense (ex. 1–2) and modal (ex. 3–4) auxiliaries in various structural domains. Next, VFINAL in embedded clauses moves the VP to the end of the clause. In periphrastic constructions, the non-finite form precedes the finite form (as in ex. 6). The following examples illustrate the surface structures associated with SEP and VFINAL: 1. ‘Peter hat gerade einen Brief geschrieben’ (Peter has just written a letter) 2. ‘Peter ist ins Kino gegangen’ (Peter has gone to the cinema) 3. ‘Peter will ins Kino gehen’ (Peter wants to go to the cinema) 4. ‘Peter kann ins Kino gehen’ (Peter can go to the cinema) 5. ‘… daß Peter heute ins Kino geht’ (that Peter goes to the cinema today) 6. ‘… daß Peter heute ins Kino gegangen ist’ (that Peter has gone to the cinema today) In the acquisition of L2 German by native speakers of L1 English, evidence of transfer involving SEP and VFINAL has been reported in the literature (Fathman and LoCoco 1989; Felix and Weigl 1991; Kaltenbacher 2001; Legenhausen 1985; Schwartz and Sprouse 1998; Weigl 1999). This is true in early grammars (e.g. Green and Hetch 1985; Hetch and Green 1993) and more mature ones (e.g. Zydatiß 1975). Furthermore, the verification of transfer of this kind should be taken as suggestive of an alteration in the path of acquisition VP headedness (Ard and Homburg 1983; Müller 2002; Paradis and Genesee 1997; Schachter 1983; Zobl 1980a, b, 1983, 1984, 1986, 1992), and it is
76
Crosslinguistic Influence and Crosslinguistic Interaction
consistent with the argument that transfer can be manifest in the number of rules that have to be traversed in the transition from one rule to another (Zobl 1980a, b, 1982). The following examples illustrate transfer of SEP and VFINAL from L1 German in the L2 English production of fifteen- and sixteen-year-olds. They are adapted from Zydatiß (1975) and Green and Hetch (1985). The asterisk indicates ungrammatical uses, and VPs that show transfer from L2 German are rendered in italics. 7. 8. 9. 10.
* ‘You often cannot so much buy’ * ‘On a motor cycle you can only two persons and a little bag pack’ * ‘Now I know why the man in number 6 so hardly fought’ * ‘You have it in your last letter write’
The same pattern of transfer as in L1 German learners of L2 English has been observed in L1 Turkish learners. The language pair L1 Turkish/L2 English is of relevance here because it corroborates the transfer pattern described above for the German-English pair. In particular, the fact that both language pairs share the same combination as regards VP headedness strengthens the argument of transfer and lends support to strong views of transfer both in children (Haznedar 1997) and adults (Schwartz 1998; Schwartz and Sprouse 1994, 1996; Sprouse and Schwarz 1998; also Hawkins and Chan 1997). While the German-English language pair has been investigated in second language learners, hardly any work has been done with third language learners, with the exception of Sánchez (2011, in press). In this study, Sánchez investigated the activation and transfer of feature values in Spanish/Catalan learners of L3 English with prior knowledge of L2 German. In determining the source language of influence, as Foote (2009: 90) notes, ‘transfer may come from either the L1 or the L2, depending on language typology, or more specifically, whether the languages share the particular features and feature values in question’ (emphasis added). In this sense, the language combination Spanish-Catalan/German/English is empirically pertinent because the L1s Spanish/Catalan and the L3 English share the head initial value, in contrast to the final value of L2 German. In addition to this, this combination makes it possible to test the proposals made within the L2 status, TPC and CEM. First of all, because the L1s and the L3 share the same feature values, the expectation would be that learners transferred from their L1s rather than from the L2, as predicted by the TPM. Besides, transfer from the L2 would be caused by negative influence and go against the predictions of the CEM. However, the evidence gathered by Sánchez (2011) does not point in this direction. The transfer encountered in Sánchez (2011), which focused on the initial stages of acquisition of L3 English of 154 third and fourth graders, lent full support to the L2 status hypothesis. This was seen in the high incidence of overgeneralization of SEP and VFINAL attested in the L3 English of the learners investigated. To be more precise, 89.5 per cent of the learners transferred, while only 10.5 per cent did not. Upon closer inspection, Sánchez found that they transferred up to 95 per cent in main clauses with periphrastic constructions (ex. 11–14) and 85.4 per cent in embedded clauses (ex. 15–18). Moreover, when all the contexts in main and embedded clauses were considered together, transfer was observed in 92.6 per cent of the clauses. Examples 11 to 18 illustrate the kind of transfer found, including functional projections (as seen in
Can a Background Language Alter the Path of Acquisition
77
the transferred head final IPs at the end of embedded clauses in examples 17 and 18). Examples 11 to 14 exhibit the transfer (i.e. overgeneralization) of SEP from L2 German in L3 English sentences. In turn, the transfer found in embedded clauses is illustrated in the overgeneralization of VFINAL in embedded clauses shown in examples 15 to 18. However, example 16 is an exception, because the order overgeneralized in this embedded clause was not VFINAL but SEP (as in main clauses). 11. 12. 13. 14. 15. 16. 17. 18.
* ‘The dog have the picnic ating’ * ‘The dog is he brekfast eat’ * ‘Janet and Pedro must the breakfast doing’ * ‘but the dog Rex will with Joe and Anna going’ * ‘when Susi a man write’ * ‘because the dog has the dinning eat’ * ‘that Bello with them are’ * ‘When they in the mountains were’
This section has reviewed the most important proposals that have been offered in order to explain transfer in L3A and to identify and predict the source language of this transfer. At the same time, it has presented specific hypotheses on the transfer of feature values along with empirical findings that validate them, while highlighting the scarcity of L3 research in this direction. The study described in the following section contributes to this line of research, and it expands the scope of previous investigation by examining the longitudinal development of transfer.
2 The present study: Research question In light of the need to validate different proposals that have been put forward to explain the occurrence of transfer of feature values in L3 acquisition, the study presented now is intended to examine the overtime development and readjustments that take place in relation to the transfer of head initial features from the L1 or of head final values from the L2 in instructed learners of L3 English. Before presenting the research question that guides the study, it is necessary to clarify the use of the terms ‘L1,’ ‘L2’ and ‘L3’ here. The learners who took part in the study, born and raised in Catalonia, were simultaneous bilinguals from birth (see more exhaustive description in Section 3). Therefore, both Spanish and Catalan were considered to be their native languages (that is, the learners had two L1s). Thereafter, following Falk and Bardel’s (2010: 61) definition of a ‘true L2’ as the first non-native language, German was the L2 of these learners. According to De Angelis’ (2007: 4) difference assumption, the critical difference is between the acquisition of an L2 and an L3, but not between an L3 and an L4, L5, L6, etc. (for a similar argument see also Hufeisen 2003 and Neuner, Erlenwein and Hufeisen 2006). In compliance with this assumption, she defines the L3 as any language ‘beyond the L2 without giving preference to any particular language’ (De Angelis 2007: 4). Consistent with these considerations, the L3 in this study was English, which coincided also with the L3 in terms of the chronological order of acquisition.
78
Crosslinguistic Influence and Crosslinguistic Interaction
The research question guiding the study asks whether the L2 German might influence the path of acquisition of VP headedness in the L3 English as a result of the transfer of head final values from this language and the overgeneralization of SEP and VFINAL. If so, it further asks how transfer from the L2 evolves over time in the L3 interlanguage of these learners in the short term, that is, within the course of the first 200 hours of instruction in this language. This research question is formulated as follows: Does L2 German affect the path of acquisition of VP headedness in the L3 English of Spanish/Catalan bilingual learners in an instructed setting? If so, how does transfer affecting VP headedness evolve over time in the L3 interlanguage of these learners within the first 200 hours of instruction?
3 Participants In order to answer this question, participants were recruited who were acquiring the L3 English in a formal context in a situation of ‘minimal input’ (i.e. four hours of instruction per week following Larson-Hall 2008). These participants (63 in number) were primary and secondary school-age learners (ranging in age from nine to fifteen). Their L1s were Spanish and Catalan and they were bilingual to different extents, as found out in a questionnaire that was administered to them with the purpose of gathering information on their language history (see Section 4). They are part of a larger corpus from the BAF and BAFiA projects2 (see Sánchez in press). All participants were learning L2 German as a second language in a school context that followed a content and language integrated programme where some subjects were taught in the L2 German (such as history, geography, maths or arts). Some of the participants combined their formal instruction in L2 German with naturalistic exposure in the form of short stays or summer camps in Catalonia or in a German-speaking country. As for the learning of L3 English, the acquisitional setting corresponds to a situation of foreign language acquisition with input restricted exclusively to curricular exposure received at school. That is, learners did not have any extracurricular formal instruction or naturalistic exposure to the L3 English outside school prior to or during the data collection. Because the research question inquired into the over time development of the transfer of head final values, the research design of the study had to be longitudinal. Bearing this in mind, the study is longitudinal (four years), and data from the learners (see Table 4.1) was collected once a year. A one-year interval is a common betweentest administration time period in research with instructed learners, and this interval can even be two years in studies that extend over long periods of time, as in the sevenyear study in the compilation of studies by Muñoz (2006), also with Spanish/Catalan bilinguals. In the present study, data were collected during the month of May in four consecutive years since the students’ first year of English learning, after they had been exposed to 33, 66, 165 and 198 hours of formal instruction to this language, exclusively through curricular exposure at school. At the time of the first data collection (T1), learners were 9.9 years old. Data were collected a second time (T2) when they were
79
Can a Background Language Alter the Path of Acquisition Table 4.1 Overview of the participants: Distribution across data collection times
Longitudinal study
T1
T2
T3
T4
INPUT
33 hrs
66 hrs
165 hrs
198 hrs
Age
9.9
10.9
11.9
12.9
N
63
53
28
9
T1 = Time ‘1’ of data collection in the longitudinal study T2 = Time ‘2’ of data collection in the longitudinal study T3 = Time ‘3’ of data collection in the longitudinal study T4 = Time ‘4’ of data collection in the longitudinal study
10.9, and then again at the ages of 11.9 (T3) and 12.9 (T4). The participants in the study are presented in Table 4.1. The input received through instructional time at each data collection and the age of the learners is also offered in this table. Even though the initial number of participants was sixty-three, only data from a subsample of them (9 in number) could be included in the four-year longitudinal analysis. This was because, as learners grew older, they used to take extracurricular lessons or naturalistic exposure during stays abroad, as indicated in the questionnaire. Such learners were removed from subsequent analysis, which explains the decrease in the number of learners across data collection times in Table 4.1.
4 Instruments and procedure Two instruments were used at each data collection. On the one hand, the elicitation technique was a storytelling task called ‘The Dog Story,’ adapted from a narrative picture story in Heaton (1966) and suitable for the investigation of transfer (Sánchez and Jarvis 2008). It was based on visual stimuli in a picture series that comprised six panels, and it was time-controlled (ca. ten to twelve minutes). The task was not familiar to the learners, and it was administered to them in class to all intact groups. They were given a sheet with the six pictures of the story and a separate white sheet to write the story. The participants were not allowed to ask questions related to the vocabulary of the story, or to use a dictionary, grammar book or similar reference tools. On the other hand, all potential participants filled out a questionnaire in order to identify and disregard candidates who did not satisfy the requirements of the study. In essence, the main criteria for inclusion in the study were the requirement that students be bilingual from birth in Spanish and Catalan; the chronological order of acquisition of the L2 German and the L3 English; the starting age for the learning of the L3; and lack of extracurricular exposure to this language outside school (so as to control for input). The questionnaire employed in the present study was inspired by an already existing model used in sociolinguistic studies (i.e. Baker and Prys Jones 1998). A version adapted from their bilingual questionnaire was validated by the author to satisfy the needs specific to the multilingual acquisition setting analysed (see Sánchez 2015). With this in mind, a set of questions was designed that surveyed the use of
80
Crosslinguistic Influence and Crosslinguistic Interaction
Spanish and Catalan in the learners’ home and family environments, and of German and English outside school. Both the storytelling task and the questionnaire were administered in class time and carried out under the supervision of their teacher and the researcher (in this case, the author of this study). For the storytelling task, learners were instructed not to describe the pictures but rather to try and explain the story as a whole. They had no time for preparation, but they had the series of pictures in front of them while doing the task. Moreover, they were explicitly informed that the test would not count as a class score.
5 Data analysis With the purpose of ascertaining whether the head final values of L2 German transferred to the L3 English and affected the path of acquisition of this language, the OV surface structures derived from the application of the SEP and VFINAL grammatical rules from L2 German were the targeted linguistic structures selected for analysis. The analytical procedure employed in the analysis of transfer from L2 German involved, first and foremost, the identification of linguistic favourable contexts, that is, contexts that were empirically relevant to observing the position of the verb in VP (ex. 19–21). These are main clauses with periphrastic constructions (ex. 19), and embedded clauses both with synthetic and periphrastic VPs (ex. 20 and 21, respectively). In contrast, non-empirically-relevant contexts were main clauses with synthetic VPs (as in example 22), where SEP could not be transferred. The next step after the identification of these contexts was the observation of the learners’ performance in these contexts. Thus, occurrences of transfer at each data collection time were identified, quantified and submitted to statistical treatment. The object of relevance in the present study was not as much the amount of transfer (i.e. the size) as the verification of its occurrence (i.e. learners’ propensity to transfer). Hence, the occurrence of transfer was measured using a binary logistic variable, that is, a dichotomous categorical variable that was used to keep a record of the propensity to transfer in each learner. More precisely, this dependent variable measured the transfer behaviour of each learner in terms of the absolute presence or absence of transfer from L2 German, coded as ‘1’ when transfer occurred in a favourable linguistic context (ex. 23) and as ‘2’ when it did not. 19. 20. 21. 22. 23.
‘The dog has eaten the food’ ‘because the dog ate the food’ ‘because the dog has eaten the food’ ‘The dog ate the food’ ‘The dog has he food eaten’
This analysis was complemented by another dependent variable that registered, in quantitative terms, the incidence of transfer in learners who did transfer. The raw occurrences of transfer for each participant were noted down on the basis of counts measured in a continuous variable (i.e. size or magnitude of transfer). To have a more
81
Can a Background Language Alter the Path of Acquisition
precise picture of the size of transfer, this continuous variable was later transformed into a new variable that calculated the percentage of transfer (and lack thereof) over the total amount of favourable linguistic contexts. This quantitative analysis allowed the calculation of the ratio of transfer from the L2 of each participant. Indeed, this ratio was meant to represent an index of intra-learner variability, which offered a perspective of the overall spectrum of coexisting structures (transferred and non-transferred) within each participant, and the extent of inter-individual variation.
6 Results This section reports the results of the analysis of transfer at the four data collection times. The report begins with a presentation of the raw numbers and percentages of occurrence of transfer, along with the results of the various statistical tests run on the data to compare learners’ performance at each data collection time. This is followed by a thorough description of the occurrences of transfer observed and their development over time in the course of the first 200 hours of instruction. The examination of linguistic favourable contexts is presented first. Out of the sixtythree participating learners (see Table 4.1), forty created linguistic favourable contexts at T1, thirty-eight at T2, twenty-five at T3, and eight at T4, as shown in Table 4.2. Let us remember at this point that with increasing age, a great number of the participants began taking extra English classes or taking part in stay abroad programs, which explains the decrease in the number of participants at each subsequent data collection time. Table 4.2 summarizes the distribution of learners who transferred and its extent at each data collection time. Table 4.2 Longitudinal development of transfer across the four data collection times T1
T2
T3
T4
63
53
28
9
Mean age at data collection time
9.9 yrs
10.9 yrs
11.9 yrs
12.9 yrs
Accumulated instructional time
33 hrs
66 hrs
165 hrs
198 hrs
Learners who create linguistic favourable contexts
40
38
25
8
Learners who transfer
38 (95%)
36 (95%)
20 (89%)
3 (38%)
Learners who do not transfer
2 (5%)
2 (5%)
5 (20%)
5 (62%)
Instances of transfer
69 (87.3%)
81 (87.1%)
35 (44.9%)
6 (28.6%)
No instances of transfer
10 (12.7%)
12 (12.9%)
43 (55.1%)
15 (71.4%)
n=
82
Crosslinguistic Influence and Crosslinguistic Interaction
As can be seen in Table 4.2, the learners’ performance was nearly identical at T1 and T2, regarding both their propensity to transfer and the size of the overall transfer in their production. However, between ages eleven and twelve (corresponding to collection times T2 and T3), their transfer behaviour experiences a change, which is even more marked between ages twelve and thirteen. More specifically, whereas the percentage of transfer at T1 and T2 is around 95, it decreases to 89 per cent at T3. At T4 it experiences a sudden drop, and the percentage of no transfer is 62 per cent. This pattern of decrease is even more obvious in the size of transfer. From T2 to T3, the percentage of transfer diminishes to nearly half (i.e. 87.1 per cent vs. 44.9 per cent, at T2 and T3, respectively). The drop from T3 to T4 is also substantial, from 44.9 per cent to 28.6 per cent. The means and standard deviation of the propensity to transfer and the size of this transfer are presented in Table 4.3. The observations in the preceding paragraph were corroborated by various statistical analyses run on the data. The drop in the propensity to transfer between T2 and T3 is significant, as indicated by a Wilcoxon signed-ranks test (z = 2.00, p = 0.046). In actual fact, this was the only significant difference across the four data collection times. As for the size of transfer, a paired samples T-test confirmed that the decrease in percentage of transfer between T2 and T3 was significant as well (p = 0.017), and also the difference between T1 and T3 (p = 0.026). For the sake of completeness in the description of the data, the quantitative account of the results concludes with a summary of the patterning of transfer affecting main and embedded clauses. This summary is offered in Table 4.4: Table 4.3 Means and standard deviation in transfer occurrence PROPENSITY
SIZE
Mean
StD
Mean
StD
T1
1.05
.221
87.5000
27.6707
T2
1.05
.226
86.7589
26.6805
T3
1.20
.408
50.1778
39.3565
T4
1.63
.518
20.4167
30.8831
Table 4.4 Overtime development of transfer in the study3 MAIN Transfer Raw T1
42
%
EMBEDDED NO transfer
Raw
91.3%
4
% 8.7%
Transfer
NO transfer
Raw
%
Raw
%
19
76%
6
24%
T2
55
86%
9
14.1%
23
92%
2
8%
T3
24
39.35%
37
60.7%
7
46.7%
8
53.3%
T4
2
16.7%
10
83.3%
3
37.5%
5
62.5%
Can a Background Language Alter the Path of Acquisition
83
Upon closer examination, a discriminating pattern emerges across structural domains in that transfer affected primarily perfective have, especially in main clauses. This pattern was particularly evident in comparison both with modal auxiliaries and the tense auxiliary be in main clauses, and also with synthetic verb phrases and periphrastic constructions in embedded clauses. Indeed, the incidence of transfer with perfective have (ex. 24–27) increases at each data collection time, in contrast to what is observed with auxiliary be in constructions involving the present continuous, the past continuous or the ‘going to’ future (ex. 27). To illustrate this point, in what follows I present and discuss longitudinal data from some learners that suggest that the nature and amount of transfer were different in distinct linguistic environments. The following instances of transfer were produced by BF at T1 and T2 (24–26), against lack thereof in (27). The examples reproduce the words in the same way in which they were rendered by the learner, even if they contain mistakes (spelling, morphology, etc.). 24. 25. 26. 27.
* ‘The dog hat4 a Brot eat’ * ‘The dog has the sandwich eat’ * ‘because the dog the sandwich eat has’ ‘She and he are going to the park to have a picking’
BF T1 BF T2 BF T2 BF T2
In a similar vein, ML shows evidence of transfer with perfective have at T2 and T3. At T2, he does not create any context for progressive tenses using auxiliary be, but he does create it at T3, where no trace of transfer is found (29–32). At this data collection time, though, transfer is observed with perfective have (as in 28). 28. 29. 30. 31. 32.
* ‘… but the dog has all the eat ate’ ‘Rigoberto and Bruna are going to make a Picknick’ ‘They are going to see the plan’ ‘… the dog is going to go to the basket’ ‘They are going to go to the mountain’
ML T3 ML T3 ML T3 ML T3 ML T3
Exactly the same development of transfer is observed in several other learners, as for example in AS or LG. In AS, the kind of transfer with perfective have is exemplified in (33–34), against no transfer in (35–37): 33. 34. 35. 36. 37.
* ‘The dog hatte5 die sandwiches eat’ * ‘Shally have the hands hoch gemachen6’ ‘Paul and Lisa are going to do a Picknick’ ‘The dog is going in the Basket’ ‘They are going to eat the food’
AS T1 AS T2 AS T3 AS T3 AS T3
Likewise, at T2 another learner (LG) created contexts only for perfective have, where she transferred systematically in examples such as (38) and others. At T3, in turn, she continued to exhibit selective transfer in the same structural domains (39) and linguistic environments (40–41). 38. 39. 40. 41.
* ‘The dog have the food eating’ * ‘… but they not here the dog has the brackfast eat’ * ‘The children are bad whay her dog the brackefast eating’ * ‘When she he and she in the mountain was’
LG T2 LG T3 LG T3 LG T3
84
Crosslinguistic Influence and Crosslinguistic Interaction
and also lack thereof (42–45): 42. ‘Lisa and Tom was preparing a brackefast for eaten a in a mountain’ 43. ‘The dog was eaten the brackefast in the corb bastket’ 44. ‘Tom and Lisa was going whi with the corb’ 45. ‘they was staring a brackefast’
LG LG LG LG
T3 T3 T3 T3
Another example is observed in the production of DL at T2 (46) and T3 (47–48): 46. * ‘He has all the food eats’ 47. ‘The dog is going to springen jumping in the basket’ 48. ‘They are going to the picnic on the mountains’
DL T2 DL T3 DL T3
This behaviour has also been observed in more instructed learners at T3 and T4 (ex. 49–54 below). This is for instance the case of CT, whose transfer is illustrated in (52): 49. 50. 51. 52. 53. 54.
‘The dog is going to the picnic’ ‘He is going to the mountain’ ‘the dog is eating the sandwiches and the breat’ * ‘the dog have the breat friss’ ‘Sam and Laura are going to a mountain’ ‘The dog is going with they’
CT T3 CT T3 CT T3 CT T4 CT T4 CT T4
In a similar way, PM creates no contexts for examination of transfer at T1 and T2. At T3, transfer is again observed only with perfective have (55) but not in other structural domains (e.g. modal auxiliaries as in 56) and linguistic domains (e.g. VFINAL with synthetic VPs as in 57). At T4, VP is head initial in all linguistic environments and structural domains (58–60). 55. 56. 57. 58. 59. 60. 61.
* ‘… found her dog in the bag and ther has all the food eat’ ‘Pedro and Sara will have a picknik’ ‘When the two brothers arrive at the park’ ‘Lucia and Max are making a picknick’ ‘When they was saw the map’ ‘When the look into the picknick’ ‘Patt have eat all the food!’
PM T3 PM T3 PM T3 PM T4 PM T4 PM T4 PM T4
At T3 this pattern of transfer is observed also in other learners such as EM: 62. * ‘… and he has all the food ate’ 63. ‘The dog is going on the basket’ 64. ‘when they look in the basket’
EM T3 EM T3 EM T3
Despite the fact that transfer with perfective have is by far the most widespread trend, there are a few exceptions to this general tendency, whereby learners transfer in periphrastic constructions with auxiliary be (65) and with SVP in embedded clauses (66). Crucially, this learner (RC) transfers every time he has the opportunity to do so
85
Can a Background Language Alter the Path of Acquisition
from the moment his linguistic resources are sufficient so as to produce embedded clauses and/or CVPs. 65. * ‘The dog are the sandwiches eating’ 66. * ‘When the kids in the mountains are’
RC T3 RC T4
In addition to this somewhat domain-specific transfer, there were a few examples where transferred orders involving VP headedness co-existed with non-transferred orders in main clauses with perfective have. Here are a few examples at T2 (LG, 68–69), at T3 (VG and JP, in 69–71 and 71–72, respectively) and at T4 (CT, in 73–74). 67. 68. 69. 70. 71. 72. 73. 74.
* ‘He and she have a picnic prepariet’ ‘… and he and she have go to the mountanes to a picnic’ * ‘The dog has the food eating’ ‘… but the dog has eat the food’ * ‘He has the picknick eat’ ‘He is in the basket and have eat all the picknick’ * ‘They were unlucky the dog have the breat frisst7’ ‘He have eat the breat and Jam’
LG T2 LG T2 VG T3 VG T3 JP T3 JP T3 CT T4 CT T4
At T3, the kind of co-existence or optionality in transfer and lack thereof with progressive tenses is worthy of note. Whereas the learner does transfer in (75), he does not in (76). At T4, he never transfers in this structural domain (76–77). Moreover, his production at this time is certainly more elaborated and includes the use of ‘going to’ to express [+ANTERIOR] temporal reference (78). 75. 76. 77. 78.
‘The kids are the breakfast and then the dog eating the breakfast’ ‘The kids going to the mountains’ ‘The kids are going to the mountains’ ‘The dog is going to eat the food’
CT CT CT CT
T3 T3 T4 T4
This optionality, though, is kept to a minimum because the use of final VPs from the L2 in an L3 clause with a progressive tense was not very productive. The interlanguage of GF illustrates this. At T1, his attempt to construct a clause in present continuous fails. While he is able to attach the ‘-ing’ morpheme to the thematic verb, the auxiliary verb is not supplied (79). In this sheer example of present continuous, VP is head final. The example in (80) presents a more baffling instance. The action expressed by the VP ‘is … eat.’ which, moreover, exhibits transfer of VP headedness, refers to an action that happened in the near past but is not explicitly shown in the visual stimuli of the task. To be more precise, the temporal reference of the clause is [+ANTERIOR]. This leads us to have doubts as to whether (80) is really a failed attempt to construct a present continuous, or an ill-formed present perfect. Indeed, at T2 he successfully builds a clause in present perfect to make reference to the same action (81). Furthermore, two more remarks favour this interpretation. First, at T2 his interlanguage does not show a single occurrence of the -ing morpheme. Secondly, he does not provide any cue that suggests he is in the process of acquiring the linguistic devices necessary for the
86
Crosslinguistic Influence and Crosslinguistic Interaction
expression of the past simple (regular or irregular), neither of past participle forms or of the present simple (82–83). 79. 80. 81. 82. 83.
* ‘Tomas and Nana to the mountain going’ * ‘The dog is the breakfast eat’ * ‘The dog have the sanwish eat’ ‘This family go to breackfast’ ‘The dog do’
GF T1 GF T1 GF T2 GF T2 GF T2
In all likelihood, the same explanation may account for the production of JG and ES at T2 among others, as in examples 84 and 85, respectively: 84. * ‘the dog is all geeat8’ 85. * ‘Dog is sandwich eat’
JG T2 ES T2
The description of the results so far has concentrated on transfer involving tense auxiliaries. However, the data also displayed evidence of transfer in constructions with modal auxiliaries. In this study, transfer involving modal auxiliaries was more often observed from T2 onwards (see 88 below), although there are some exceptions. For example, at T1, NB produces periphrastic constructions only in main clauses, and these involve perfective have (86). At T2, her interlanguage exhibits evidence of both tense and modal auxiliaries. The orders provided always exhibit transfer (87–88): 86. * ‘The dog has hat the breakfast eat’ 87. * ‘The dog has does it eat!!!!’ 88. * ‘We can in the garden walk’
NB T1 NB T2 NB T2
Notwithstanding, some occurrences of transfer with modal verbs were documented occasionally at the onset of L3A in some longitudinal learners. The examples below illustrate transfer in learners who transfer the order of modal verbs already at T1: 89. 90. 91. 92. 93.
* ‘They want a picknik meik’ * ‘Bern and Ashly wolen9 one pick nik do’ * ‘The dog want the breakfast eat’ * ‘One day two brothers wellen10 am picknick walking’ * ‘The dog welle the food eat’
SF T1 DL T1 ML T1 VG T1 VG T1
As anticipated above, though, such occurrences were more frequent at T2 or T3 (ex. 94–97), as a minimal knowledge of the TL is a developmental prerequisite for the emergence of modal verbs in the learners’ L3 interlanguage. 94. 95. 96. 97.
* ‘Luisa and Pedrito’s dog will the breakfast eat’ * ‘Last summer Tobbi and Anni will a Picknick mach’ * ‘They willen a Picknick eat’ * ‘we must in the hose “gehen” ’:
ML T2 RM T2 RM T2 OF T2
In this section I have presented, described and illustrated the evidence of transfer of head final values from L2 German assembled in the corpus. Moreover, the section has explained the way these transferred orders pattern across data collection times,
Can a Background Language Alter the Path of Acquisition
87
and has portrayed their development over time. Next, the main implications of these findings are discussed in Section 7, which also provides some guidelines for future investigation.
7 Discussion The question guiding this study asked whether the head final values from L2 German would transfer to L3 English and influence the path of acquisition of VP headedness in this language. The examination of any influence of the L2 on the path or route of acquisition of L3 English caused by transfer from the L2 was done by studying a language triad in which head initial and head final languages were combined. Specifically, the L2 but none of the other languages (native and target languages) was head final and exhibited an SOV order in the targeted linguistic structures SEP and VFINAL. The answer to this question was confirmatory, as indicated by the varied array of qualitative and quantitative analyses performed on the longitudinal data of the participants. To put it another way, the overgeneralization of SEP and VFINAL in the L3 English interlanguage of the learners confirmed the occurrence of transfer. For the present purposes, this is taken as supporting evidence for claims in the literature review that a background language can indeed alter the path of acquisition of a given TL. In this study, the alteration was two fold. First, it consisted in the introduction of verb discontinuity in L3 English, which was transferred from the L2 German rule SEP. Secondly, this alteration was seen in the introduction of VFINAL from L2 German in L3 English. Hence, in this study, transfer was to a great extent a form of overgeneralization of orders from the L2 German (SEP for the most part) and in consequence of underproduction of target-like SVO orders. The assertion that feature values were indeed vulnerable to the effects of CLI confirms the claims in the generative theories on the transfer of feature values reviewed in Section 1 (FT/FA, FFFH, MTH and WPTH). As a matter of fact, the high incidence of transfer found in the data and the presence of transferred head final IPs and CPs at the end of embedded clauses lend support to the FT/FA hypothesis. In turn, this evidence demonstrates the relevance of transfer at the initial state of L3 acquisition, and the interaction of transfer and developmental influences. Furthermore, it runs counter to claims that overlook transfer of feature values (e.g. Epstein, Flynn and Marthohardjono 1996), including VP headedness. In this sense, in the earliest stage of L3, the learners’ grammar was consistently head final, and it showed systematic transfer of the values under examination here. In fact, learners seemed to treat the L3 as a head final language, as in their L2 German. The extremely high percentages of transfer at T1 (after thirty-three hours of instruction) and also at T2 (after sixty-six hours of instruction) suggest that transfer affecting VP headedness was not an idiosyncratic feature of these learners. Instead, it suggests that transfer was a regular pattern in the L3 early grammars of these learners and a characteristic trait at some point of their grammatical development. In this sense, it is useful to recall that learners' propensity to transfer the head final value of the L2 German at T1 and T2 was almost 100 per cent and was identical at
88
Crosslinguistic Influence and Crosslinguistic Interaction
the two data collection times (95 per cent). Likewise, the overall representativeness of transfer in terms of the raw frequency of transferred orders was also nearly identical at T1 and T2. In other words, within the first sixty-six hours of instruction, learners’ performance was homogeneous and markedly head final. By T3, both the propensity and size of transfer decreased. However, the rate of this decrease differed in both of them. While learners’ propensity to transfer was only slightly lower (from 95 per cent at T2 to 89 per cent at T3), the decrease in the representativeness of transfer decreased ostensibly to nearly half the occurrence (viz. from 87.1 per cent to 44.9 per cent from T2 to T3, respectively). Consequently, the difference between T2 and T3 would not be so much the categorical presence or absence of transfer as the frequency with which transferred orders occurred in the data, which was much higher at T2 than at T3. The smaller size of transfer at T3 (and to a lesser extent, also at T4) indicates the co-existence, in the L3 interlanguage, of grammars from the L2 German and the L3 English. This kind of optionality was found only from T3 onwards. In the previous data collection times, optionality was minimal because the alteration in the path of acquisition introduced by the L2 left no room for co-existence of mutually exclusive orders. In addition to this, the transition from T3 to T4 caused a steady rise in the activation of the head initial values of the TL and a progressive falling off or decay in those of the L2. Finally, at T4, the decrease of transfer was conspicuous. This is true both of propensity to transfer (89 per cent to 38 per cent) and of its size (44.9 per cent to 28.6 per cent) from T3 to T4, respectively. In conclusion, the most feasible interpretation of these findings is that the head initial value of the target language (L3 English) was nearly unavailable at the initial state of L3A, which basically corresponded to the data at T1. More importantly for the present purposes, the identification of L2 German as the source language of influence lends support to the L2 status hypothesis. Likewise, the choice of German (L2) over Spanish and Catalan (L1s) is taken as evidence not to support the TPM. Because the L1s and the L3 (but not the L2) are typologically closer as regards VP headedness and because they share the feature values examined, the prediction of the TPM would be that the L1s become the source languages of influence. However, as we have seen, this was not the case. Similarly, the data here suggest that transfer was negative, which did not corroborate the facilitative effects predicted by the CEM. An intriguing pattern encountered in the data at all collection times was the high incidence of transfer that involved perfective have. This pattern may be accounted for in two ways. The first argument would be the objective formal and phonological similarity existing between inflected forms of this auxiliary across the four languages of this study, namely English (have), German (haben), Spanish (haber) and Catalan (haver). Even though all of these forms might be regarded as neighbour words11, this small difference in pronunciation might lead learners to perceive inflected forms of ‘have’ in English (L3) and German (L2) as closer to each other than to Spanish or Catalan (L1s). If this was the case, then these forms would foster transfer at the level of lexical syntax (Hall et al. 2009). For the time being this can only remain a matter of speculation, but it is reasonable to consider that activation of syntax may run parallel to lexical activation (Falk and Bardel 2010; Sánchez 2011). By the same token, the objective dissimilarity between L3 English and L2 German in constructions involving progressive tenses, where the auxiliary be is needed, might
Can a Background Language Alter the Path of Acquisition
89
probably discourage transfer. As this linguistic environment corresponds to the present continuous tense, which does not exist in L2 German, it might have been more difficult for learners to transfer it. Thus, there is no landing site in the L3 for the structure transferred from the L2, as implied in Kellerman’s (1995) ‘transfer to nowhere’ principle. Nonetheless, the similarity argument cannot fully explain why the alteration in path of acquisition of VP headedness in L3 English was caused by the L2 (German) rather than the L1s (Spanish or Catalan). In truth, given that the structural properties selected for investigation were different in the L2 and the L3 and in the L1s and the L2, but similar in the L1s and the L3, it could have been wise to expect learners to transfer from the L1s rather than the L2. In any case, a further insight into this is beyond the scope of the present chapter (but see Sánchez 2011, in press). Future research should address this issue in more depth, while trying to obtain a more comprehensible understanding of the complex interaction between the various languages of the multilingual learner.
Notes 1 As Beck (1998: 319) points out, the general picture is ‘that English prohibits overt verb raising whereas German requires it.’ See theoretical arguments on the potential mixed headedness of German in Abraham (1992), Fanselow (1987) and also Haider (1993). 2 The author is grateful to Grant FFI2013-47616-P from the Spanish Ministry of Education. 3 To be consistent with the analysis of SEP in main clauses, Table 4.4 does not include instances of transfer where synthetic VPs in main clauses were placed clause-finally such as ‘a boy and a girl a pik nik eat’ or ‘Dog foot eat.’ 4 L2 German hat is the equivalent of L3 English has. 5 L2 German hatte is the equivalent of L3 English had. 6 The form ‘gemachen’ is an attempted past participle in L2 German (gemacht), which is equivalent to L3 English ‘made’. 7 L2 German fristt is the equivalent of L3 English devours. 8 The form ‘geeat’ is an attempted past participle in L2 German (gegessen), which is equivalent to L3 English ‘eaten’. 9 L2 German wollen is the equivalent of L3 English want. 10 The form ‘wellen’ is an attempted L2 German infinitive (wollen), which as indicated in the preceding footnote, is equivalent to L3 English ‘want.’ 11 ‘Neighbour words’ are words that overlap in spelling, orthographic and phonological features except for one letter (e.g. Dijkstra, Grainger and Van Heuven 1999; Kroll et al. 2008; Schwartz, Kroll and Díaz 2007).
References Ard, J. and T. Homburg (1983), ‘Verification of language transfer’, in S. Gass and L. Selinker (eds), Language Transfer in Language Learning, 157–76, Rowley: Newbury House. Baker, C. and S. Prys Jones (1998), Encyclopedia of Bilingualism and Bilingual Education, Clevedon: Multilingual Matters. Bardel, C., and Y. Falk (2007) The role of the second language in third language acquisition: The case of Germanic syntax, Second Language Research 23 (4), 459–84.
90
Crosslinguistic Influence and Crosslinguistic Interaction
Bohnacker, U. (2006), ‘When Swedes begin to learn German: From V2 to V2’, Second Language Research, 22 (4): 443–86. Bohnacker, U. and C. Rosén (2007a), ‘Transferring Information-Structural patterns from Swedish to German’, in A. Belikova, L. Merioni and M. Umeda (eds), Proceedings of the 2nd Conference on Generative Approaches to Language Acquisition North America (GALANA), 27–38, Somerville, MA: Cascadilla Proceedings Project. Bohnacker, U. and C. Rosén (2007b), ‘How to start a V2 declarative clause: Transfer of syntax vs. information structure in L2 German’, Nordlyd, 34 (3): 29–56. Bohnacker, U. and C. Rosén (2008), ‘The clause-initial position in L2 German declaratives: Transfer of information structure’, Studies in Second Language Acquisition, 30 (4): 511–38. Cabrelli, J., M. Iverson, Y. Judy and J. Rothman (2008), ‘What the start of L3 tells us about the end of L2: N-drop in L2 and L3 Portuguese’, in H. Chan, E. Kapia, and H. Jacob (eds), A Supplement to the Proceedings of the 32nd Boston University Conference on Language Development. Available online: http://www.bu.edu/bucld/ files/2011/05/32-Cabrelli.pdf. Cabrelli, J., S. Flynn and J. Rothman (2012), Third Language Acquisition in Adulthood, Philadelphia: John Benjamins. Cenoz, J., B. Hufeisen and U. Jessner (2001), Crosslinguistic Influence in Third Language Acquisition: Psycholinguistic Perspectives, Clevedon: Multilingual Matters. Cenoz, J., B. Hufeisen and U. Jessner (2003), The Multilingual Lexicon, Dordrecht: Kluwer Academic. Çiğdem, S. (2006), Third Language Acquisition. Turkish-German Bilingual Students’ Acquisition of English Word Order in a German Educational Setting, Berlin: Waxman. Corder, S. P. (1974), ‘Error Analysis’, in J. Allen and S. P. Corder (eds), Techniques in Applied Linguistics, 122–54), London: Oxford University Press. DeAngelis, G. (2007), Third or Additional Language Acquisition, Clevedon: Multilingual Matters. Dewaele, J. M. (1998), ‘Lexical inventions: French interlanguage as L2 versus L3’. Applied Linguistics, 19 (4): 471–90. Dijkstra, T., J. Grainger and W. Van Heuven (1999), ‘Recognition of cognates and interlingual homographs: The neglected role of phonology’, Journal of Memory and Language, 41 (4): 496–518. Epstein, S., S. Flynn and G. Marthohardjono (1996), ‘Second language acquisition: Theoretical and experimental issues in contemporary research’, Brain and Behavioural Sciences, 19 (4): 677–714. Eubank, L. (1993/1994), ‘On the transfer of parametric values in L2 development’, Language Acquisition, 3 (1): 185–208. Eubank, L. (1994), ‘Optionality and the initial state in L2 development’, in T. Hoekstra, and B. Schwartz (eds), Language Acquisition Studies in Generative Grammar, 369–88, Amsterdam: John Benjamins. Eubank, L. (1996), ‘Negation in early German-English interlanguage: More valueless features in the L2 initial state’, Second Language Research, 12 (1): 73–106. Falk, Y. and C. Bardel (2010), ‘The study of the role of the background languages in third language acquisition. The state of the art’, International Review of Applied Linguistics, 48 (2–3): 185–219. Falk, Y. and C. Bardel (2011), ‘Object pronouns in German L3 syntax: Evidence for the L2 status factor’, Second Language Research, 27 (1): 59–82.
Can a Background Language Alter the Path of Acquisition
91
Fathman, A. and V. Lococo (1989), ‘Word Order Contrasts and Production in Three Target Languages’, in J. Dechert and M. Raupach (eds), Transfer in Language Production, 125–50, Norwood, NJ: Ablex Publishing Corporation. Felix, S. and W. Weigl (1991), ‘Universal Grammar in the classroom: The effects of formal instruction on second language acquisition’, Second Language Research, 7 (2): 162–81. Foley, C. (1999), ‘Knowledge of the syntax of operators in the initial state. The acquisition of relative clauses in French and English’, Glot International, 4: 11–14. Flynn, S., C. Foley and I. Vinnitskaya (2004), ‘The Cumulative-Enhancement Model for Language Acquisition: Comparing Adults’ and Children’s Patterns of Development in First, Second and Third Language Acquisition of Relative Clauses’, International Journal of Multilingualism, 1 (1): 3–16. Foote, R. (2009), ‘Transfer in L3 acquisition: The role of typology’, in Y. I. Leung (ed.), Third Language Acquisition and Universal Grammar, 48–80, Clevedon: Multilingual Matters. García Mayo, P. and J. Rothman (2012), ‘L3 morphosyntax in the generative tradition: The initial stages and beyond’, in. J. Cabrelli, S. Flynn, and J. Rothman (eds), Third Language Acquisition in Adulthood, 9–32, Philadelphia: John Benjamins. Gass, S. and J. Ard (1980), ‘L2 data: Their relevance for language universals’, TESOL Quarterly, 16 (4): 443–52. Green, P. and K. Hetch (1985), ‘Native and non-native evaluation of learners’ errors in written discourse’, System, 13 (2): 77–97. Grüter, T. and S. Conradie (2006), ‘Investigating the L2 initial state: Additional evidence from the production and comprehension of Afrikaans-speaking learners of German’, in R. Roumayana, S. Montrul, and P. Prévost (eds), Inquiries in Linguistic Development: In Honor of Lydia White, 89–114, Amsterdam: John Benjamins. Hall, C., D. Newbrand, P. Ecke, U. Sperr, V. Marchand and L. Hayes (2009), ‘Learners’ implicit assumptions about syntactic frames in new L3 words: The role of cognates, typological proximity and L2 status’, Language Learning, 59 (1): 153–203. Hammarberg, B. (2001), ‘Roles of L1 and L2 in L3 production and acquisition’, in J. Cenoz, B. Hufeisen, and U. Jessner (eds), Cross-linguistic Influence in Third Language Acquisition, 21–41, Clevedon: Multilingual Matters. Haznedar, B. (1997), ‘L2 acquisition by a Turkish-speaking child: Evidence for L1 influence’, in E. Hughes, M. Hughes, and A. Greenhill (eds), Proceedings of the 21st Annual Boston University Conference on Language Development, 245–56, Somerville: Cascadilla Press. Hawkins, R. and Y. C. Chan (1997), ‘The partial availability of universal grammar in second language acquisition: The “failed functional features hypothesis”’, Second Language Research, 13 (3): 187–226. Heaton, B. (1966), Composition Through Pictures, London: Longman. Hermas, A. (2010), ‘Language acquisition as computational resetting: Verb movement in L3 initial state’, International Journal of Multilingualism, 7 (4): 343–62. Hetch, K. and P. Green (1993), ‘Muttersprachliche Interferenz beim Erwerb der Zielsprache Englisch in Schülerproduktionen aus Deutschland, Frankreich, Italien, Schweden, und Ungarn’, Fremdsprachen lehren und lernen, 22: 35–56. Hufeisen, B. (2003), ‘L2, L2, L3, Lx – alle gleich? Linguistische, lernerinterne und lernerexterne Faktoren in Modellen zum multiplen Spracherwerb’, in N. Baumbartern, C. Böttger, M. Motz, and J. Probst (eds), Übersetzen, Interkultureele Kommunikation. Spracherwerb und Sprachvermittlung- das Leben mit mehreren Sprachen, 97–109, Festschrift für Juliane House zum 60, Geburstag.
92
Crosslinguistic Influence and Crosslinguistic Interaction
Jaensch, C. (2011), ‘L3 acquisition of German adjectival inflection: A generative account’, Second Language Research, 27 (1): 83–105. Kaltenbacher, M. (2001), ‘Universal Grammar and Parameter Resetting in Second Language Acquisition, Frankfurt: Peter Lang. Kellerman, E. (1995), ‘Crosslinguistic influence. Transfer to nowhere?’ Annual Review of Applied Linguistics, 15: 125–50. Kroll, J., S. Boo, M. Misra and T. Guo (2008), ‘Language selection in bilingual speech: Evidence for inhibitory processes’, Acta psychologica, 128 (3): 416–30. Larson-Hall, J. (2008), ‘Weighing the benefits of studying a foreign language at a younger starting age in a minimal input situation’, Second Language Research, 24 (1): 35–63. Legenhausen, L. (1985), ‘Typische Fehler im Bereich der englischen Satzkomplemente’, Der fremdsprachliche Unterricht, 75: 184–96. Leung, Y. I. (2003), ‘Failed Features versus Full Transfer Full Access in the Acquisition of a Third Language: Evidence from Tense and Agreement’, in J. Liceras (ed.), Proceedings of the 6th Generative Approaches to Second Language Acquisition Conference (GASLA 2002), 199–207, Somerville, MA: Cascadilla Proceedings. Leung, Y. I. (2005), ‘K2 vs. L3 initial state: A comparative study on the acquisition of French DPs by Vietnamese monolinguals and Cantonese-English bilinguals’, Bilingualism: Language and Cognition, 8 (1): 39–61. Leung, Y. I. (2007), ‘Third language acquisition: Why it is interesting to generative linguists’, Second Language Research, 23 (1): 95–114. Leung, Y. I. (2008), ‘The verbal functional domain in L2A and L3A. Tense and agreement in Cantonese-English-French interlanguage’, in J. Liceras, H. Zobl, and H. Goodluck (eds), The Role of Formal Features in Second Language Acquisition, 379–403, Mahwah: Lawrence Erlbaum. Montrul, S., R. Dias and H. Santos (2011), ‘Clitics and object expression in the L3 acquisition of Brazilian Portuguese. Structural similarity matters for transfer’, Second Language Research, 27 (1): 21–58. Müller, N. (2002), ‘Crosslinguistic influence in early child bilingualism Italian/German’, Eurosla Yearbook, 2: 135–54. Muñoz, C. (2006), Age and the Rate of Foreign Language Learning, Clevedon: Multilingual Matters. Neuner, G., S. Erlenwein and B. Hufeisen (2006), Deutsch als Zweite Fremdsprache, München: Goethe Institut. Paradis, J. and F. Genesee (1997), ‘On continuity and the emergence of functional categories in bilingual first language acquisition’, Language Acquisition, 6 (2): 91–124. Na Ranong, S. and Y. I. Leung (2009), ‘Null objects in L1 Thai-L2 English-L3 Chinese: An empiricist take on a theoretical problem’, in Y. I. Leung (ed.), Third Language Acquisition and Universal Grammar, 162–91, Clevedon: Multilingual Matters. Rothman, J. (2010), ‘On the typological economy of syntactic transfer: Word order and relative clause high/low attachment preference in L3 Brazilian Portuguese’, International Review of Applied Linguistics, 48 (2/3): 245–73. Rothman, J. (2011), ‘L3 syntactic transfer selectivity and typological determinacy: The typological primacy model’, Second Language Research, 27 (1): 107–27. Rothman, J. and J. Cabrelli (2010), ‘What variables condition syntactic transfer? A look at the L3 initial state’, Second Language Research, 26 (2): 189–218. Rutherford, W. (1983), ‘Language typology and language transfer’, in S. Gass and L. Selinker (eds), Language Transfer in Language Learning, 358–70, Rowley: Newbury House. Rutherford, W. (1984), ‘Description and explanation of interlanguage syntax: State of the art’, Language Learning, 34 (3): 127–55.
Can a Background Language Alter the Path of Acquisition
93
Sánchez, L. (2011), ‘ “Luisa and Pedrito’s dog will the breakfast eat”: Interlanguage transfer and the role of the second language factor’, in G. DeAngelis and J.-M. Dewaele (eds), New Trends in Crosslinguistic Influence and Multilingualism Research, 86–104, Clevedon: Multilingual Matters. Sánchez, L. (2015a) ‘Activation and blending in third language acquisition: Evidence of crosslinguistic influence from the L2 in a longitudinal study on the acquisition of L3 English’. Bilingualism: Language and Cognition, 18: 252–69. Sánchez, L. (in press) Transfer in English as an L3: Empirical evidence from Catalonia. In C. Bardel, Y. Falk, and C. Lindqvist (eds), Tredjespråksinlärning. Sánchez, L. and S. Jarvis (2008), ‘The use of picture stories in the investigation of crosslinguistic influence’, TESOL Quarterly, 42 (2): 32933. Schachter, J. (1974), ‘An error in error analysis’. Language Learning, 24 (2): 205–14. Schachter, J. (1983), ‘A new account of language transfer’, in S. Gass and L. Selinker (eds), Language Transfer in Language Learning, 32–46, Amsterdam: John Benjamins. Schachter, J. and W. Rutherford (1979), ‘Discourse function and language transfer’, Working Papers on Bilingualism, 19 (1): 3–12. Schwartz, B. (1998), ‘On two hypotheses of “transfer” in L2A: Minimal trees and absolute L1 influence’, in S. Flynn, G. Martohardjono, and W. O’Neil (eds), The Generative Study of Second Language Acquisition, 35–9, Mathway: Lawrence Erlbaum Associates. Schwartz, B. and L. Eubank (1996), ‘Introduction: What is the L2 initial State?’ Second Language Research, 12 (1): 1–5. Schwartz, B. and R. Sprouse (1994), ‘Word order and nominative case in non-native language acquisition: A longitudinal study of (L1 Turkish) German interlanguage’, in T. Hoekstra and B. Schwartz (eds), Language Acquisition Studies in Generative Grammar: Papers in Honor of Kenneth Wexler from the 1992 GLOW Workshops, 317–68, Philadelphia: John Benjamins. Schwartz, B. and R. Sprouse (1996), ‘L2 cognitive states and the “full transfer/full access” model’, Second Language Research, 12: 4–72. Schwartz, A., J. Kroll and M. Diaz (2007), ‘Reading words in Spanish and English: Mapping orthography to phonology in two languages’, Language and Cognitive Processes, 22 (1): 106–29. Selinker, L., M. Swain and G. Dumas (1975), ‘The interlanguage hypothesis extended to children’, Language Learning, 25 (1): 139–52. Sellner, M. (1994), ‘Japanisch L1, Englishc L2 und Deutsch L3: Zur Frage des OV- und VO Transfers’, in D. W. Halwachs, C. Penzinger, and I. Stütz (eds), Sprache, Onopatopöit, Rhetorik, Namen, Idiomatik, Grammatik: Festschrift für Prf. Dr. Karl Sornig zum 66, 245–58, Geburstag, Graz: Grazer Linguistische Monographien 11. Shanon, B. (1991), ‘Faulty language selection in polyglots’, Language and Cognitive Processes, 6 (4): 339–50. Sharwood Smith, M. and J. Truscott (2005), ‘Stages or continua in second language acquisition: A MOGUL solution’, Applied Linguistics, 26: 219–40. Sharwood Smith, M. and J. Truscott (2006), ‘Full Transfer Full Access: A processingoriented interpretation’, in S. Unsworth, T. Parodi, A. Sorace, and M. Young-Scholten Paths of Development in L1 and L2 Acquisition, 201–16, Amsterdam: John Benjamins. Sharwood Smith, M. and J. Truscott (2008), ‘MOGUL and Crosslinguistic Influence’, in D. Gabrys (ed.), Morphosyntactic Issues in Second Language Acquisition Studies, 63–85, Clevedon: Multilingual Matters. Sorace, A. (2003), ‘Near-nativeness’, in C. Doughty and M. Long (eds), Handbook of Second Language Acquisition, 130–51, Oxford: Blackwell.
94
Crosslinguistic Influence and Crosslinguistic Interaction
Sprouse, R. and Schwartz, B. (1998) In defense of full transfer in German-English and French-English interlanguage: Comparative L2 acquisition research, in Proceedings of the Annual Boston University Conference on Language Development 22, (pp. 726–36). Somerville: Cascadilla Press. Truscott, J., and Sharwood Smith, M. (2004) ‘Acquisition by processing: A modular perspective on language development’, Bilingualism: Language and Cognition 7 (2): 1–20. Vainikka, A. and M. Young-Scholten (1994), ‘Direct access to X’-theory: Evidence from Korean and Turkish adults learning German’, in T. Hoekstra and B. Schwartz (eds), Language Acquisition Studies in Generative Grammar: Papers in Honor of Kenneth Wexler from the 1991 GLOW Workshops, 317–45, Amsterdam: John Benjamins. Vainikka, A. and M. Young-Scholten (1996), ‘The early stages in adult L2 syntax: Additional evidence from Romance speakers’, Second Language Research, 12 (2): 140–76. Vainikka, A. and Young-Scholten, M. (2002), ‘Restructuring the CP in L2 German’, in Proceedings of the Annual Boston University Conference on Language Development 26, (pp. 712–22). Somerville, MA: Cascadilla Press. Vinnitskaya, I., Flynn, S., and Foley, C. (2003) The acquisition of relative clauses in a third language: comparing adults and children, in J. Liceras et al. (eds.), Proceedings of the 6th Generative Approaches to Second Language Acquisition Conference (pp. 340–5). Somerville: Cascadilla Press. Weigl, W. (1999), ‘Die Verbposition im Englisch Deutscher Gymnasten. Ein Indiz für die Wirksamkeit der Muttersprache’, Beiträge zur Fremdsprachenvermittlung, 36: 25–41. Weigl, W. (2000), ‘Verb und Subjekt in Französisch G3’. Beiträge zur Fremdsprachlichenvermittlung, 37: 42–61. White, L. (2000), ‘Second language acquisition: From initial to final state’, in J. Archibald (ed.), Second Language Acquisition and Linguistic Theory, 130–55, Oxford: Blackwell. Williams, S. and B. Hammarberg (1993), L1 and L2 Influence in L3 Production: Evidence from Language Switches, Stockholm: Stockholm University. Williams, S. and B. Hammarberg (1998), ‘Language switches in L3 production: Implications for a polyglot speaking model’, Applied Linguistics, 19 (3): 259–333. Yuan, B. (2000), ‘Is thematic-verb raising inevitable in the acquisition of a nonnative language?’ In S. Howell, S. Fish, and T. Keith-Lucas (eds), Boston University Conference on Language Development, 797–807, Somerville: Cascadilla Press. Zobl, H. (1980a), ‘Developmental and transfer errors: Their common bases and (possibly) differential effects on subsequent leaning’, TESOL Quarterly, 13 (4): 469–79. Zobl, H. (1980b), ‘The formal and developmental selectivity of L1 influence on L2 acquisition’, Language Learning, 30 (1): 43–57. Zobl, H. (1983), ‘L1 acquisition, age of L2 acquisition, and the learning of word order’, in S. Gass and L. Sellinker (eds), Language Transfer in Language Learning, 176–96, Rowley: Newbury House (Revised edition). Zobl, H. (1984), ‘Uniformity and source-language variation across developmental continua’, in W. Rutherford (ed.), Language Universals and Second Language Acquisition, 185–218, Amsterdam: John Benjamins. Zobl, H. (1986), ‘Word order typology, lexical government, and the prediction of multiple graded effects in L2 word order’, Language Learning, 36 (1): 159–83. Zobl, H. (1992), ‘Prior linguistic knowledge and the conservation of the learning procedure: Grammaticality judgements of unilingual and multilingual learners’, in S. Gass and L. Selinker (eds), Language Transfer in Language Learning, 176–96, Amsterdam: John Benjamins. Zydatiß, W. (1975), ‘Eine Taxonomie von Fehlern in der englischen Wortstellung’, Linguistik und Didaktik, 6: 304–24.
5
Item-related Determinants of Cognate Guessing in Multilinguals Jan Vanhove and Raphael Berthele
1 Introduction Formal similarities to a known language make it considerably easier to learn a new one. More specifically, the learning curve is smoothed if language learners are able to make use of the crosslinguistic similarities provided by cognate pairs (e.g. Carton 1971; De Groot and Keijzer 2000; Lotto and De Groot 1998; Odlin 1989; Ringbom 2007). These are etymologically related translation equivalents that often bear a certain degree of formal similarity, for example, Dutch appel and English apple or Swedish försvinna and German verschwinden ‘to disappear’. It is especially in comprehension that cognate relations, if perceived, can bring about positive interlingual transfer (Ringbom 2007). In fact, formal similarities between related languages can be so pervasive that they give rise to receptive multilingualism (Braunmüller and Zeevaert 2001), a constellation in which readers or listeners are able to understand a language variety (Lx) without having ‘officially’ learnt or acquired it. Additionally, a series of learning materials aiming to foster receptive skills in related languages rely on the presence of such crosslinguistic similarities (e.g., Hufeisen and Marx 2007; Klein and Stegmann 2000; Müller et al. 2009). For crosslinguistic cognate relationships to be of actual help, it is of course essential that foreign language learners or readers and listeners engaging in receptive multilingualism be aware of them (see Otwinowska-Kasztelanic 2011 for a concise discussion) – a condition that is not always met automatically. In order to investigate the factors that affect how well participants can identify crosslinguistic cognate relationships, several studies have made use of what we will call cognate guessing tasks. These are purposefully reductionistic tasks in which participants are presented with isolated Lx stimuli with cognates in known languages and are asked to translate them. Since the role of con- and co-textual cues is eliminated, participants need to engage in interlingual inferencing (Carton 1971) and draw on their linguistic and metalinguistic knowledge in order to guess the meaning of the Lx stimuli. Cognate guessing tasks consequently yield insights into the factors that determine the probability with which interlingual transfer and inferencing processes produce the correct outcome
96
Crosslinguistic Influence and Crosslinguistic Interaction
(i.e. a correctly identified cognate relationship) when the roles of target language and extra-linguistic knowledge are reduced to a minimum.1 Such factors fall into two broad categories. On the one hand, the accurate recognition of cognate relationships in an Lx depends on participant-related variables. Of particular interest for our present purposes, multilinguals often draw on their knowledge of several languages when confronted with a text in an unknown language (e.g. Gibson and Hufeisen 2003; Singleton and Little 1984; Wenzel 2007). With respect to cognate guessing tasks specifically, a large multilingual repertoire is often associated with better task performance, particularly if it includes well-developed competences in language varieties that are related to the Lx (Berthele 2008, 2011; Berthele and Lambelet 2009; Swarte, Schüppert and Gooskens 2013; but see Van Bezooijen, Gooskens and Kürschner 2012, for a study failing to find such a multilingualism effect). On the other hand, the likelihood with which cognate relationships can be identified is also affected by item-related variables, for example the formal similarity between an Lx word and a cognate in a known language. However, the participants’ knowledge of foreign languages is not systematically taken into account when modelling cognate guessing success in terms of item-related variables: typically, these measures are computed with respect to the L1 alone. This state of affairs can presumably partly be ascribed to two facts. First, the number of variables quickly becomes too large when several predictors have to be considered with respect to several potential supplier languages (the ‘small n, large p’ problem). Second, these variables are often characterized by properties that are unfavourable for statistical analyses, particularly by strongly skewed distributions and a high degree of collinearity (see Fox and Weisberg 2011, Chapter 6). Researchers consequently need to make rather arbitrary decisions about which variables to include with respect to which potential supplier languages so that the set of predictors remains of a manageable size. Relatively recent developments in statistics, however, are better suited to cope with multidimensional datasets characterized by skewed and collinear variables and offer new opportunities to analyse such datasets in a principled way. In this chapter, we make use of one such recent development, random forests (Breiman 2001), in order to (re)analyse data from two independent but similar studies in which multilinguals took part in written cognate guessing tasks. Our aim is to assess the impact of a selection of item-related characteristics on the likelihood with which the meaning of Lx cognates can be guessed correctly. For both studies, Germanspeaking Swiss participants (L1s: Swiss-German dialect and Standard German) with some knowledge of at least French and English (typically the first and second foreign language, respectively2) were asked to translate written stimuli from other Germanic languages (Danish, Dutch, Frisian and Swedish). Many of these stimuli have translation-equivalent cognates in French and English, warranting the inclusion of item-related variables with respect to these languages in the analysis. On the basis of the literature (see Berthele 2011; Van Heuven 2008; Vanhove 2014; see Section 1 for further references), we identify four types of such variables whose impact on cognate guessing success can be investigated in the present chapter. We then operationalize each type of variable with respect to Standard German, French and English.3 Rather than assuming that cognate guessing success can most parsimoniously be modelled by including variables with respect to one, two or all three of these common languages,
Item-related Determinants of Cognate Guessing in Multilinguals
97
this decision is taken in an exploratory, data-driven way. By analysing data from two similar but independent studies using different participants and different stimuli, we nevertheless ensure that our exploratory findings are empirically robust. In the spirit of ‘open science’, the data and computer code used for the analyses have been made available at http://dx.doi.org/10.6084/m9.figshare.763246.
2 Explanatory variables considered 2.1 Overall formal distance In order to make an ‘interlingual identification’ (Weinreich 1953), participants need to perceive the crosslinguistic similarity between the stimulus and the potential transfer base. While the perception of crosslinguistic similarity is not a function of objective formal similarity alone (Kellerman 1977, 1983), it is generally speaking more likely to occur when the formal overlap between the stimulus and the potential transfer base is high. The degree of such formal overlap is consequently a key variable in receptive multilingualism. In receptive multilingualism studies, the formal overlap between an Lx word and one of its cognates in a known language is typically measured by means of the Levenshtein algorithm (see examples below). This algorithm is used to compute the smallest number of deletions, insertions and substitutions necessary to transform one string into another. The Levenshtein distance is defined as the total minimal operation cost; higher Levenshtein distances indicate less formal overlap. An example is given in Figure 5.1, which shows how the Dutch string oorzaak ‘cause’ can be transformed into its German cognate Ursache with a minimal total transformation cost of seven operations. These raw Levenshtein distances are then normalized in order to account for differences in word length between cognate pairs, typically by dividing them by the length of the longest possible least-cost alignment (as recommended by Heeringa 2004: 130–32). For the example in Figure 5.1, this yields a length-normalized Levenshtein distance of 7 ÷ 9 = 0.78. In what follows, ‘Levenshtein distance’ refers to the lengthnormalized distance measure.
Figure 5.1 Example of a Levenshtein distance computation. The Dutch orthographic string oorzaak ‘cause’ can be transformed into its German cognate Ursache with a minimum of seven operations. For this alignment, two deletions (D), two insertions (I) and three substitutions (S) were used. Other least-cost alignments are possible, but this is the longest one.
98
Crosslinguistic Influence and Crosslinguistic Interaction
Levenshtein distances between Lx words and their cognates in a related L1 have been found to be respectable predictors of cognate guessing accuracy in the spoken modality (e.g. Gooskens, Kürschner and Van Bezooijen 2011; Kürschner, Gooskens and Van Bezooijen 2008; Van Bezooijen and Gooskens 2005; but see Berthele 2011). As for the written modality, Berthele and Lambelet (2009) administered a cognate guessing task featuring twenty-nine isolated Romansh and Romanian words to 140 French- and Italian-speaking Swiss students. They found that the number of correct translation attempts per word was correlated with the orthographic Levenshtein distance between the Lx word in question and its French or Italian cognate (r = 0.32). In a similar task involving twenty-eight isolated Danish and Swedish words presented to 163 Swiss-German participants, Berthele (2011) failed to find a significant negative association with the Levenshtein distances between the Scandinavian words and their Standard German cognates (r = 0.14). However, such a correlation was present when the Levenshtein distances with respect to English, the participants’ L2 or L3, were considered (r = 0.42). Lastly, Vanhove and Berthele (2013) presented 181 Germanic (Danish, Dutch, Frisian and Swedish) words to ninety-eight SwissGerman participants. They found that while Levenshtein distances between the Lx words and their L1 (German) counterparts were negatively associated with the stimuli’s intelligibility, the fit of their model was markedly better when they also took into consideration the participants’ L2 and L3 (French and English). They did this by computing Levenshtein distances with respect to French and English as well and by then taking as the predictor variable whichever of three distances computed (Lx–German, Lx–French, Lx–English) was the lowest for each stimulus. Together, Berthele’s (2011) and Vanhove and Berthele’s (2013) findings indicate that even when the participants’ L1 is closely related to the Lx, Lx cognate guessing may be modelled more accurately when taking into account the possibility that the participants make use of multiple supplier languages (for the notion of supplier language, see Williams and Hammarberg 1998). In addition to orthographic formal similarity, cognate guessing success in the written modality may also be affected by the participants’ conception of the cognates’ spoken form. Several authors (e.g. Berthele 2011; Möller and Zeevaert 2010; Wenzel 2007) noted that participants tend to self-pronounce written Lx stimuli on the basis of familiar or assumed grapheme-phoneme correspondences in their search for a plausible translation. Evidently, such self-pronunciations do not necessarily correspond to the stimuli’s actual target language pronunciation. This allows for a substantial degree of inter- and intra-individual variability: Not all participants will self-pronounce a given stimulus in the same way and one participant may consider several plausible articulations. As a result, the effect of ‘assumed phonetic distance’ was deemed to be too complex to model adequately in our quantitative analyses.
2.2 The importance of consonants In its crudest form, the Levenshtein algorithm weighs all operations equally; that is, a vowel–vowel substitution has the same operation weight as a consonant–consonant substitution. However, a few studies suggest that Lx intelligibility is more severely affected by consonantal differences than by vocalic differences between the Lx and
Item-related Determinants of Cognate Guessing in Multilinguals
99
the L1, L2, … , Ln. Gooskens, Heeringa and Beijering (2008) computed Levenshtein distances between a spoken standard Danish text and its renderings in seventeen other Scandinavian language varieties. They then computed the share of these distances that was due to vowel insertions, deletions and substitutions (vocalic Levenshtein distance) and to consonant operations (consonantal Levenshtein distance). In the example in Figure 5.1, four operations are associated with vowel slots and three with consonant slots; the length-normalized vocalic and consonantal Levenshtein distances would be 4 ÷ 9 = 0.44 and 3 ÷ 9 = 0.33, respectively. Gooskens, Heeringa and Beijering (2008) found that the intelligibility of the seventeen Scandinavian language varieties to speakers of standard Danish correlated more strongly with the consonantal than with the vocalic Levenshtein distances (r = 0.74 vs r = 0.29), although the correlation with the overall phonetic Levenshtein distances was stronger still (r = 0.86). Similarly, Berthele (2011) found consonantal differences vis-à-vis German and English to be more detrimental to the intelligibility to Swiss-Germans of Danish and Swedish targets compared to vocalic contrasts for both spoken and written stimuli. These results corroborate qualitative findings by Möller (2011), who reported that Germanspeaking students show greater tolerance towards vocalic than towards consonantal differences in a written cognate guessing task. A greater degree of flexibility vis-à-vis vocalic discrepancies has also been reported in so-called word reconstruction tasks in the L1 (Moates and Marks 2012). The implication is that consonants may contribute more to the subjective transparency of cognate relationships than do vowels, possibly due to their greater information value (see Van Heuven 2008: 53). This paper provides an additional quantitative evaluation of whether consonants should indeed be weighed more heavily in models of written cognate guessing. It does so by considering the consonantal Levenshtein distances between the Lx stimuli and their cognates in the three potential shared supplier languages. Our logic is that, if our participants are indeed particularly sensitive to the consonant frames, then these consonantal Levenshtein distances should offer explanatory power above and beyond the full orthographic Levenshtein distances also considered in the model.
2.3 The importance of word beginnings Participants may not only let their cognate guesses be guided more by the consonantal skeleton of the Lx stimulus than by its vowels; they may also be particularly sensitive to word beginnings in cognate guessing tasks (Berthele 2011; Möller 2011; Möller and Zeevaert 2010; see also Müller-Lancé 2003; for L1-related findings pointing to a privileged role of word beginnings, see Broerse and Zwaan 1966; Johnson and Eisler 2012; Scaltritti and Balota 2013). Here, too, the implication is that certain word parts contribute more to the perceived similarity of cognate relationships than do others. In this chapter, we therefore investigate whether word-initial discrepancies between the stimuli and their cognates can help to account for between-item differences in cognate guessing success above and beyond what can be explained by the full formal distances between the stimuli and their cognates. To this end, we will consider the word-initial distances between the Lx stimuli and their cognates in all of the three potential supplier languages. Here, too, our logic is that a higher reliance on word beginnings on the part
100
Crosslinguistic Influence and Crosslinguistic Interaction
of our participants should result in explanatory power of these word-initial distances above and beyond what is offered by the full orthographic Levenshtein distances.
2.4 Cognate frequency The recognition of words in known languages is affected by their corpus frequencies, such that high-frequency words are generally easier to recognize than low-frequency words (e.g. Lemhöfer et al. 2008, and references in, e.g. Brysbaert et al. 2011). Van Heuven (2008) suggests that Lx stimuli with high-frequency known cognates may similarly be easier to decode correctly. The impact of cognate frequency on auditory cognate guessing was investigated empirically by Kürschner, Gooskens and Van Bezooijen (2008) and was found to be negligible. Vanhove and Berthele (2013), by contrast, not only found that L1 cognate frequency was positively associated with written cognate guessing task performance but also that a joint measure comprising the aggregate frequency of the stimuli’s German, French and English cognates was an even better predictor. This may indicate that multilingual participants are sensitive to frequency information from their different languages when participating in a written cognate guessing task.
3 Method In order to determine the set of item-related variables that can most parsimoniously model written cognate guessing success in multilinguals, we adopt a highly exploratory approach, as we discuss below. As a guard against spurious findings, we turn to two independent but similar datasets on cognate guessing tasks (Dataset 1: Vanhove and Berthele 2013; Dataset 2: Vanhove and Berthele 2015) and reanalyse them separately from one another but along the same lines. The datasets feature different stimuli and participants and the cognate guessing tasks were administered in a different way. Crucially, however, we assume that the effects of the four variables outlined in Section 2, if at all present, are comparable in the two datasets, both in terms of their direction (positive or negative) and effect sizes. Consequently, if a particular effect is not found in both datasets, we will have to conclude that it is not robust with respect to the stimuli chosen or the participants recruited and has to be left out of the final models on the grounds of parsimony. We first discuss the specifics of both datasets. Then, we explain how we operationalized the variables discussed in the previous section. Lastly, we briefly present the statistical tools used for our analyses.
3.1 Description of the datasets 3.1.1 Dataset 1 The first dataset was collected for a study by Vanhove and Berthele (2013) that aimed to assess to what extent findings regarding cognate guessing are dependent on the Lx in question.
Item-related Determinants of Cognate Guessing in Multilinguals
101
Participants. Ninety-eight eligible participants were recruited to participate in a cognate guessing task.4 All were native speakers of a Swiss-German dialect; they did not have knowledge of any Germanic language apart from Swiss-German dialects, Standard German and English; and they were not language experts (such as linguistics students or interpreters). Before the cognate guessing task, participants filled out a short questionnaire and rated their reading and listening skills in English, French and any other languages according to the six-level assessment grid for the Common European Framework of Reference for Languages (CEFR).5 We refer to Table 5.1 for the key descriptive statistics. Apart from Swiss-German dialects, Standard German, English and French, some participants reported knowledge of Italian (44), Spanish (38), Turkish (5), Portuguese, Russian (4), Greek (2), Albanian, Arabic, Bulgarian, Cantonese, Kurdish, Korean, Macedonian, Polish, Romansh and Czech (1). Task. The cognate guessing task was a paper-and-pencil task and consisted of four lists, each featuring fifty words without context in a Germanic language (Danish, Frisian, Dutch and Swedish). Of these words, 181 had a German, English or French translation-equivalent cognate, which could in principle render them intelligible to readers without prior competences in the target language. These ‘target words’ were selected from a list with 384 frequent words compiled by the research group Mutual intelligibility of closely related languages (University of Groningen, PI: Charlotte Gooskens; see e.g. Gooskens, Kürschner and Van Bezooijen 2011; Kürschner, Gooskens and Van Bezooijen 2008). For details about the selection procedure, we refer to the studies cited. For the present analyses, one target word (eftermiddag ‘afternoon’) was left out of consideration as it is not a full cognate to either English (afternoon) or German (Nachmittag), but rather a mixture form of both. This left 180 target words for the analyses. The vast majority of these words had a German
Table 5.1 Description of the two studies. Dataset 1
Dataset 2
n participants (of which women)
98 (65)
148 (85)
Age range
16–72
10–86
Mean age (SD)
34 (15)
42 (21)
Mean self-assessed English reading skills (SD)
4.0 (1.2)
3.5 (1.3)
Mean self-assessed French reading skills (SD)
3.3 (1.3)
3.2 (1.4)
180
45
n items Lx
Danish, Dutch, Frisian, Swedish
Note. Self-assessments were based on the CEFR grid (A1 = 1, C2 = 6).
Swedish
102
Crosslinguistic Influence and Crosslinguistic Interaction
cognate (175), about half had an English cognate (84) and about a third had a French cognate (57). Four to five words per list could not be translated without some pre-existing knowledge of the target language, for example Frisian heit ‘father’ and Danish seng ‘bed’. Participants able to correctly translate more than two such words per language would have been assumed to have had too much pre-existing lexical knowledge of the language in question for our purposes and would have been excluded from the analyses (see Kürschner, Gooskens and Van Bezooijen 2008, for a similar filtering function of noncognates). None of the participants was able to do so, however. For the full list of stimuli and their German, English and French translation-equivalent cognates, we refer to the online appendix that is available from http://dx.doi.org/10.6084/m9.figshare.763246. We informed the participants that all 200 words were nouns and asked them to translate these into German. They were not forced to provide an answer in case none came to mind. The order of target languages was counter-balanced between participants (Danish–Dutch–Swedish–Frisian and the reverse) in order to prevent learning or fatigue effects from skewing the results. Each sheet clearly indicated what language the stimuli were in. Every translation was marked as correct or incorrect. Note that ‘reasonable’ but incorrect translations were scored as incorrect, too. Thus, a participant’s attempt to translate the Danish word kylling (‘chicken’) as ‘murder’ (from ‘killing’) was rated as incorrect.
3.1.2 Dataset 2 The second dataset was collected for a subproject of the Multilingualism through the lifespan project (SNSF-130457). The goal of the subproject on cognate guessing was to track the lifespan trajectories of cognate guessing skills and to establish which cognitive and linguistic developments any such age trends could be attributed to (see Vanhove and Berthele 2015; Vanhove 2014). To this end, a total of 167 Swiss-Germans were recruited. Four of them experienced technical difficulties during the cognate guessing task; their data will not be analysed. Participants. All participants self-rated their English, French and other language skills on the six-level CEFR scale. None of the participants reported any knowledge of Swedish or of related North Germanic languages or were language experts. Beside Swiss-German dialects, Standard German, English and French, some participants reported knowledge of Italian (85), Spanish (57), Portuguese (7), Tagalog (5), Serbian (4), Hungarian, Romansh (3), Arabic, Cebuano, Dutch, Greek, Russian, Swahili (2), Bahasa, Catalan, Czech, Hebrew, Romanian, Tamil, Telegu, Thai and Turkish (1). Of the 163 participants that did not experience technical difficulties, we excluded a further fifteen participants because they did not provide self-ratings corresponding to any of the six CEFR levels for French or English, leaving a sample of 148 participants. We refer to Table 5.1 for the key descriptive statistics. Task. We presented the participants with fifty individual isolated Swedish words in random order on a computer screen. Forty-five of these words had German, English or French translation-equivalent cognates; five words had no cognate. Of the
Item-related Determinants of Cognate Guessing in Multilinguals
103
forty-five target words, forty had a German cognate, twenty-seven an English cognate and ten a French cognate. We refer to the online appendix (http://dx.doi.org/10.6084/ m9.figshare.763246) for a list of the stimuli and their German, English and French cognates. As for Dataset 1, we had decided a priori to exclude participants who were able to correctly translate more than two words without cognates from the sample; none of the participants were able to do so, however. Participants were asked to indicate whether they thought that they might be able to translate the word presented into German. If so, a text box appeared in which they could enter their translation. Stimuli remained on screen until the participants indicated whether they would attempt a translation. Every translation was marked as correct or incorrect.
3.2 Quantification of predictors 3.2.1 Formal distance We computed the overall, consonantal and word-initial Levenshtein distances between each stimulus and its German, English and French counterparts as presented in the online appendix, that is, nine distance measures per stimulus. Before the Levenshtein distance computations, the orthographic strings were converted to lowercase. The German letter ß, which is not common in written Standard German in Switzerland, was represented as ss. We computed the overall Levenshtein distances using a binary algorithm that weighted all operations (insertions, deletions, substitutions) equally. Graphemes differing only in their diacritics, for example ö and o in öppna–open, were considered identical.6 Only vowel–vowel and consonant–consonant mappings were allowed. For this purpose, we considered the graphemes a, e, i, o, u and y as well as their diacritical variants to be vowels and all other graphemes to be consonants. We length-normalized these raw overall distances by dividing them by the length of the longest lowestcost alignment. For stimuli lacking a cognate in a given source language, we set the respective overall Levenshtein distance to the maximum, that is, one. The consonantal Levenshtein distances were computed similarly to Gooskens, Heeringa and Beijering’s (2008) operationalization. We first counted the number of consonant operations in the overall Levenshtein alignments and length-normalized these counts by dividing them by the length of the overall Levenshtein alignment. The consonantal Levenshtein distances for stimuli lacking a cognate in a given source language were set to one. Lastly, we computed the Levenshtein distances between the word beginnings of the stimuli and those of their German, French and English cognates. Word beginnings were operationally defined as up to and including the first consonant or consonant cluster. Thus, the word beginnings in the cognate pair oorzaak–Ursache (see Figure 5.1) are oor and Ur, which differ from each other in two slots. The word-initial distances were length-normalized by dividing them by the length of the overall Levenshtein alignment. In our example, this yields a word-initial Levenshtein distance of 2 ÷ 9 = 0.22. The word-initial Levenshtein distance for stimuli lacking a cognate in a given source language was set to one.7
104
Crosslinguistic Influence and Crosslinguistic Interaction
3.2.2 Cognate frequency As cognate frequency measures, we extracted the word frequencies per 1,000,000 words of the German and English translation-equivalent cognates of the Lx stimuli from the subtlex-de (Brysbaert et al. 2011) and subtlex-us (Brysbaert and New 2009) databases, respectively. Word frequencies per 1,000,000 words for the French cognates were extracted from the Lexique 3 database (New et al. 2007). These three databases, which are freely available from http://crr.ugent.be/programs-data/ subtitle-frequencies and http://www.lexique.org, are highly comparable in that they are derived from similar corpora (film and television subtitles) that were compiled according to similar principles. Unlike the subtlex databases, however, Lexique 3 distinguishes homographs by part of speech. For full comparability with the subtlex frequencies, we aggregated the Lexique 3 frequencies over homographs. For the English cognate frequencies, we summed over the US and British spellings as well (e.g. the frequencies of dike and dyke were combined). All frequencies were logarithmically transformed in order to prevent items with extremely high frequency counts from exerting undue influence on the analyses. We added one to every frequency count in order to deal with zero frequencies, the logarithm of which would otherwise be undefined: log-frequency = ln(frequency + 1).
3.3 Random forest-based variable importance We identified four classes of predictors that could help to account for variance in the intelligibility of the written cognates to our participants: (a) overall Levenshtein distance, (b) consonantal Levenshtein distance, (c) word-initial Levenshtein distance and (d) cognate frequency. In each class, we have three predictors each, one for each potential supplier language under consideration (German, English and French). Twelve predictors is obviously too many to consider jointly in a regression that models the intelligibility of merely 180 (Dataset 1) and 45 (Dataset 2) items, respectively. This problem is compounded by the fact that the predictors show a high degree of multicollinearity. Thus, we need to select from our set of potential predictors the ones that can best explain the intelligibility of the cognates in our dataset while accounting for the collinearity between our predictors. A popular solution for selecting variables for a regression model is to enter the variables in a stepwise regression model. However, among other problems, stepwise regression models are sensitive to the order in which the variables are entered and are particularly affected by correlated predictors (see Whittingham et al. 2006, for references and discussion). Instead, we turned to a fairly recent development in statistics that is better able to handle such cases: random forests (Breiman 2001). For reasons of space, we can only briefly describe this algorithm below; for more detailed yet accessible introductions, we refer to Strobl, Malley and Tuz (2009) and Tagliamonte and Baayen (2012).
3.3.1 The rationale of random forests Random forests are ensembles of classification or regression trees. Classification and regression trees seek to explain the variance in an outcome variable by recursively
Item-related Determinants of Cognate Guessing in Multilinguals
105
partitioning the data by means of binary splits so as to reach ever purer (i.e. more uniform with respect to the outcome variable) nodes. Tree models are flexible quantitative tools in that they can easily cope with interacting predictors, nonlinearities and a multitude of predictors relative to the number of cases. They do, however, suffer from two severe drawbacks in particular. First, they are highly unstable: Small changes in the data can and often do result in dramatically different tree models. Second, they are piecewise constant: Even continuous relationships between the outcome variable and the predictor variables are broken down into dichotomies. A popular solution to these problems is to grow not one tree but several hundreds of trees. By randomly resampling from the original set of cases, ‘new’ datasets are created for which new, different trees can be grown. Due to the random fluctuations between the training datasets, the ensemble as a whole is much more robust than a single tree and the hard-cut boundaries that are characteristic of single trees are smoothed. In order to grow even more diverse trees, the set of possible predictors can be reduced randomly as well. For instance, we can specify that at each stage, only four out of twelve predictors are to be taken into consideration. This approach, the defining feature of random forests (Breiman 2001), ‘allows predictor variables that were otherwise outplayed by their competitors to enter the ensemble’ (Strobl et al. 2008: 307) and may reveal subtle interaction effects that would have remained hidden if a handful of variables had otherwise dominated the tree growing process.
3.3.2 Estimating variable importance Random forests often produce excellent prediction rates, but unlike single trees, they are difficult to visualize and interpret, effectively making them ‘black boxes’. To gain some insight into which variables are important, we can compute variable importance measures, for example the so-called ‘permutation importance’. This importance measure is derived by computing how much the overall prediction accuracy of the random forest decreases when the values of a given predictor are randomly permuted, thereby breaking the association between the predictor and the outcome variable. The more important a predictor is in a random forest ensemble, the more this permuting will affect the ensemble’s overall prediction accuracy. Such an approach, however, may overstate the importance of correlated variables, which is why Strobl et al. (2008) proposed the ‘conditional permutation importance’. For this measure, the intercorrelations between predictor variables are also taken into account by means of a conditioning scheme, thereby reducing the effect of collinearity on the permutation scores (see Strobl et al. 2008, for technical details). The variable importance measures thus computed reflect more closely the partial effects of each variable.
3.3.3 Software and settings Random forests were grown using the cforest() function in the party package (Hothorn et al. 2013) for R (R Core Team 2013). The individual trees were grown
106
Crosslinguistic Influence and Crosslinguistic Interaction
on subsamples consisting of 63.2 per cent of the cases of the original dataset (see Strobl et al. 2007). Each forest consisted of 1,000 trees (i.e. ntree = 1,000) and four randomly selected predictors were considered at each split (i.e. mtry = 4). We then computed the conditional permutation importance measures for the predictors using party’s varimp() function. If the significance (i.e. the p-value) of the association between a given predictor and another covariate was lower than 0.80 (the default), the relevant covariate was included in the predictor’s conditioning scheme. Runs with different settings converged to the same results as those reported here; different runs with the same settings likewise yielded similar results. All data and computer code are available at http://dx.doi.org/10.6084/m9.figshare.763246, allowing interested readers to reproduce the analyses or run alternative analyses.8
4 Results The percentage of correct translations per stimulus ranged from 0 per cent to 100 per cent in Dataset 1 (mean SD: 49 31 per cent) and from 1 per cent to 93 per cent in Dataset 2 (mean SD: 43 29 per cent). Thus, neither outcome variable was characterized by floor or ceiling effects.
4.1 Variable importances We grew two random forests (one for each dataset) modelling the proportion of correct translations in terms of the twelve variables discussed above and computed the variables’ conditional permutation importance scores. These importance scores are presented in Figures 5.2 and 5.3 for Datasets 1 and 2, respectively. (Note, however, that the permutation scores cannot be compared between studies in absolute terms; Strobl et al. 2009.) The dotted vertical lines provide conservative indications of the extent to which permutation scores can differ from zero due to randomness alone. Variables with importance scores to the left of this line are effectively irrelevant predictors in the random forest. Both Figures 5.2 and 5.3 reveal that the overall Levenshtein distance between the Lx stimuli and their German cognates is by far the most important predictor of Lx stimulus intelligibility. Additionally, the overall Levenshtein distance with respect to English also appears to be a predictor of stimulus intelligibility in both datasets. The overall Levenshtein distance with respect to French, by contrast, only seems to play a marginal role in Dataset 1, whereas its importance in Dataset 2 is negligible. Additionally, both kinds of partial Levenshtein distances – consonantal and wordinitial Levenshtein distances – play marginal roles at best in Dataset 1 and can be considered to be irrelevant in Dataset 2. Lastly, the corpus frequencies of the stimuli’s German and English cognates emerge as potential predictors in both studies; the frequency of the stimuli’s French cognates does not seem to affect Lx stimulus intelligibility.
Item-related Determinants of Cognate Guessing in Multilinguals
107
Figure 5.2 Conditional permutation importances for a random forest (mtry = 4, ntree = 1,000) grown on Dataset 1 (180 items). The dotted vertical line provides a conservative indication of the extent to which the importance scores differ from zero due to random fluctuations; variables with permutation scores to the left of this line can be considered irrelevant in the random forest. It thus seems that Lx stimulus intelligibility is primarily affected by item-related measures with respect to German and English, whereas the role of French is marginal at best. This may indicate that our participants tend to draw on their knowledge of both German and English when engaging in a written cognate guessing task. Hypothetically, participants may tend to evaluate the orthography of an Lx stimulus with respect to both its German and English cognates simultaneously. For each stimulus, they may then tend to prefer whichever of the German and English words considered that shows the smallest orthographic distance to the stimulus as the most plausible translation candidate. This would imply that it is not necessarily the Levenshtein distance with respect to German that is the best predictor of written-item intelligibility, but rather whichever of the German and English Levenshtein distances that happens to be the lower one. Furthermore, it need not be the language-specific frequencies that play a role in cognate guessing; when a stimulus has cognates in both German and English,
108
Crosslinguistic Influence and Crosslinguistic Interaction
Figure 5.3 Conditional permutation importances for a random forest (mtry = 4, ntree = 1,000) grown on Dataset 2 (45 items). The dotted vertical line provides a conservative indication of the extent to which the importance scores differ from zero due to random fluctuations; variables with permutation scores to the left of this line can be considered irrelevant in the random forest. it may be the (possibly weighted) sum of both cognates’ frequencies that is the better predictor of cognate guessing accuracy. In order to take these possibilities into account, we computed ‘Germanic’ Levenshtein and frequency variables. For each stimulus, the overall Germanic Levenshtein distance equals whichever of the German or English Levenshtein distance is the lower one. The consonantal and word-initial Germanic Levenshtein distances are derived from the alignments associated with this overall Germanic Levenshtein distance. The Germanic cognate frequency is the mean of the stimulus’s German and English cognate frequencies, which was then logarithmically transformed (ln(mean frequency + 1)). These four Germanic variables were added to the original twelve in order to grow another pair of random forests (mtry = 4, ntree = 1,000), on the basis of which new sets of conditional permutation importances were computed. These permutation scores are presented in Figures 5.4 and 5.5 for Datasets 1 and 2, respectively.
Item-related Determinants of Cognate Guessing in Multilinguals
109
Figure 5.4 Conditional permutation importances for a random forest (mtry = 4, ntree = 1,000) grown on Dataset 1 (180 items). The variables with respect to German and English were collapsed into ‘Germanic’ variables. The dotted vertical line provides a conservative indication of the extent to which the importance scores differ from zero due to random fluctuations; variables with permutation scores to the left of this line can be considered irrelevant in the random forest.
In both cases, the overall Germanic Levenshtein distance clearly emerges as more important than its language-specific counterparts and Germanic cognate frequency pips the language-specific frequencies.
4.2 Regression modelling The only two kinds of predictors that were found to have non-zero variable importance scores across the two datasets are the overall Levenshtein distances and the logtransformed corpus frequencies of the stimuli’s cognates. In both cases, the ‘Germanic’ variables, in which the information with respect to German and English was collapsed, outperform the language-specific variables in terms of importance score. In order to
110
Crosslinguistic Influence and Crosslinguistic Interaction
Figure 5.5 Conditional permutation importances for a random forest (mtry = 4, ntree = 1,000) grown on Dataset 2 (forty-five items). The variables with respect to German and English were collapsed into ‘Germanic’ variables. The dotted vertical line provides a conservative indication of the extent to which the importance scores differ from zero due to random fluctuations; variables with permutation scores to the left of this line can be considered irrelevant in the random forest.
render their effects more interpretable, we fitted cognate guessing success in terms of these two variables using regression techniques. As before, the data from the two studies were fitted separately to gauge the robustness of our results across similar but different, independent datasets. Translation accuracy (correct–incorrect) was modelled in logistic linear mixedeffect models with random intercepts for both participants and stimuli and with overall Germanic Levenshtein distance and log-transformed Germanic cognate frequency as fixed-effect predictors (for introductions to logistic linear mixed-effect models geared towards language researchers, see Baayen 2008, Chapter 7; Jaeger 2008).9 In order to account for individual-level differences in the effects of these variables (some participants may be more sensitive to Levenshtein distance or corpus frequency than
Item-related Determinants of Cognate Guessing in Multilinguals
111
Table 5.2 Fixed-effect estimates of the logistic linear mixed-effect model for Dataset 1. Estimate SE
p (z-test)
(Intercept)
0.067 0.157
0.671
Overall Germanic Levenshtein distance
5.4
Variable
Germanic cognate frequency (log)
0.8
0.34 0.09