The Oxford Handbook of Language Attrition 9780198793595, 0198793596

Table of contents :
Cover
The Oxford Handbook of Language Attrition
Copyright
Contents
Acknowledgements
List of Abbreviations
The Contributors
Chapter 1: Introduction
Part I: THEORETICAL IMPLICATIONS OF LANGUAGE ATTRITION
Chapter 2: Language attrition and the competition model
2.1 Four puzzles
2.1.1 Puzzle #1: permastore
2.1.2 Puzzle #2: variation across levels
2.1.3 Puzzle #3: childhood forgetting
2.1.4 Puzzle #4: catastrophic interference
2.2 The competition model
2.2.1 Entrenchment
2.2.2 Transfer
2.2.3 Participation, isolation, and identity
2.2.4 Resonance
2.2.5 Decoupling
2.3 Conclusions
Chapter 3: Language attrition and the feature reassembly hypothesis
3.1 Introduction
3.2 FRH in heritage grammars
3.3 Final remarks
Chapter 4: The interface hypothesis as a framework for studying L1 attrition
4.1 Introduction
4.2 The interface hypothesis in first generation L1 Arrtition
4.3 Selectivity of individual L1 attrition
4.4 Is indeterminacy in L1 attrition caudes by transfer or by bilingualism itself?
4.5 Reversibility of indivudual L1 attrition
Chapter 5: Implicatons of the bottleneck hypothesis for language attrition
5.1 Introduction
5.2 A fundamental division based on age: the maturational factor
5.2.1 The grammatical competence of early attriters
5.2.2 The grammatical competence of late attriters
5.3 Conclusion
Chapter 6: A complex dynamic systems perspective on personal background variables in L1 attrition
6.1 Personal background variables in L1 attrition research
6.2 Complex dynamic systems theory and language development
6.3 Complex dynamic systems theory and research methodology
6.4 Complex dymanic systems theory and personal background variables
6.5 Summary and outlook
Acknowledgements
Part II: PSYCHOLINGUISTIC AND NEUROLINGUISTIC APPROACHES TO LANGUAGE ATTRITION
Chapter 7: Introduction to psycholinguistic and neurolinguistic approaches to language attrition
7.1 Introduction and brief historical overview
7.2 Language Attrition as a processing mechanism
7.3 Language attrition as a memory mechanism
7.4 Language attrition as a brain mechanism
7.5 Section outlook
Chapter 8: Language attrition as a special case of processing change: A wider cognitive perspective
8.1 Introduction
8.2 Investigating attrition: research tradition
8.2.1 Representation, change, and the nature of processing
8.2.2 Interfaces
8.3 Unifying knowledge, performance, and development within one architecture
8.3.1 An overview
8.3.2 Modularity and collaboration: key elements of the framework
8.3.3 The developmental perspective
8.3.4 Representations and memory theory
8.3.5 Working memory
8.3.6 Resting levels of activation
8.4 Attrition and interface processing
8.4.1 Acquisition and growth
8.4.2 Interactions across interfaces in acquisition and attrition
8.4.3 Frequency effects
8.4.4 Degrees of forgetting
8.4.5 Inhibition and attrition
8.4.6 Metalinguistic knowledge and attrition
8.5 Concluding comments
Acknowledgements
Chapter 9: Memeory retrieval and language attrition
9.1 Introduction
9.2 Native language change in L2 learning and proficient bilingual language processing
9.3 Retrieval induced forgetting
9.4 The consequences of language immersion
9.5 Implications of retrieval induced forgetting for theories of attrition
9.6 Conclusions
Acknowledgements
Chapter 10: How billingualisn affects syntactic processing in the native language: Evidence from eye movements
10.1 The interactive nature of the bilingual system
10.2 Eye-mevement records as a premier behavioural tool to study L1 language change
10.3 Illustrative cases
10.3.1 Reconfiguration of the L1
10.3.2 Reactivation of the L1
10.4 Conclusion
Acknowledgements
Chapter 11: First language attrition and developmental language disorder
11.1 Introduction
11.2 Overlapping profiles: Monolingual developmental language disorder and bilingual typical development
11.3 Combined effects of developmental language disorder and bilingualism
11.3.1 Delay in the L2
11.3.2 L1 attrition
11.4 Factors influencing L1 attrition in bilingual chindren
11.4.1 Age of exposure to L2 and L1 development
11.4.2 Socio-economic status: parent–child interaction and L1 development
11.4.3 Language status and L development
11.5 Isolating the effect of developmental language disorder
11.5.1 Assessing the L1: comparing an individual child using appropriate norms
11.5.2 Assessing the L1: linguistic markers and valid tools
11.5.3 Possible alternatives: measures that areinsensitive to L1 attrition (or L2 delay)
11.6 Concluding remarks
Chapter 12: Ageing as a confound in language attrition research: Lexical retrieval, language use, and cognitive and neural changes
12.1 Introduction
12.2 Lexical attrition in bilingual adults
12.2.1 Causes of lexical attrition
12.3 Neurophysiological changes in language attrition
12.4 Lexical decline in monolingual older adults
12.4.1 Neurophysiological changes in cognitive ageing
12.4.2 The interaction between age-related declines and L1 attrition
12.5 Language attrition and ageing
12.6 Cognitive and neuropsychological contributions to language attrition and ageing
12.7 Conclusion
Acknowledgements
Chapter 13: Linguistical regression in bilingual patients with alzheiner's didease
13.1 Introduction
13.2 Alzheimer's disease, language, and bilingualism
13.2.1 Alzheimer’s Disease
13.2.2 Language deterioration in monolingual patients with Alzheimer’s Disease
13.2.3 Bilingualism and Alzheimer’s Disease: a topic of growing interest
13.3 Alzheimer's disease and bilingualism
13.3.1 The Babylonian benefit: a controversial issue
13.3.2 Second-language regression in Alzheimer’s Disease
13.3.3 Language mixing in bilingual individuals with Alzheimer’s Disease
13.4 Conclusion
Chapter 14: Electrophysiological approaches to L1 attrition
14.1 Event-related potential and language processing in real-time
14.2 How plastic is the adult brain? event-related potentialsin L2 learners and attriters
14.3 Morpho-syntactic processing in attriters' L1
14.3.1 Grammatical gender violations in German and Dutch
14.3.2 Number agreement violations at various positions in Italian sentences
14.3.3 Relative clauses in Italian immigrants: can one’s L2 render the L1 ungrammatical?
14.4 Lexical-semantic processing in attriters' L1 and L2
14.4.1 A study on confusable words in Italian sentences
14.4.2 Lexical processing in attriters’ L2
14.5 State of affairs, limitations, and future directions
Chapter 15: Neuroimaging perspectives on L1 attrition and language change
15.1 L1 attrition: the phenomenon
15.2 Introducing magnetic resonance imaging methods
15.3 Magnetic resonance imaging as a tool to tackle L1 attrition
15.4 Neural investigations of L1 change beyond extreme L1 attrition
15.5 Future directions in neuroimaging L1 attrition research
15.5.1 Longitudinal designs
15.5.2 Neural oscillations
15.6 Summary and conclusions
Part III: Linguistic factors in language attrition
Chapter 16: Introduction to linguistic factors in language attrition
16.1 Introduction
16.2 The Linguistic manifestations of language attrition
16.3 Knowledge gaps and future research agenda
16.4 Contributions to the current sections
16.5 Conclusion
Chapter 17: Phonetic drift
17.1 Introduction
17.2 Phonetic drift of consonants
17.3 Phonetic drift of vowels
17.4 Phonetic drift of suprasegmentals
17.5 Discussion and conclusion
Chapter 18: Phonetic attrition
18.1 Introduction
18.1.1 Definition of attrition
18.2 Aim of contribution
18.3 What is phonetics?
18.4 Speech learning model (SLM)
18.5 Segmental assimilation in phonetic attrition
18.6 Prosodic assimilation in phonetic attrition
18.7 Segmental dissimilation in phonetic attrition
18.8 Prosodic dissimilation in phonetic attrition
18.9 Global regional or foreign accent in a native language
18.10 Summary of research
18.11 Discussion of implications of phonetic attrition
Chapter 19: Phonological attrition
19.1 Introduction
19.2 Phonological contrasts
19.2.1 Contrasts in the vowel systems
19.2.2 Contrasts among obstruents
19.3 Phonotactic patterns
19.4 Future challenges in the study of phonological attrition
Chapter 20: Morphological attrition
20.1 Introduction
20.2 Background
20.3 Attrition of derivational morphology
20.4 Attrition of Inflectional morphlogy
20.5 Markedness theory
20.6 Regression hypothesis
20.7 Activation threshold hypothesis (ATH)
20.8: Interface hypothesis (IH)
20.9 The 4-M model
20.10 Conclusion
Chapter 21: Lexical attrition
21.1 What is lexical attrition and what does it look like?
21.2 Explanations of lexical attrition
21.3 Factors that affect lexical attrition
21.4 Methods for exploring lexical attrition
21.5 Final Thoughts
Chapter 22: Null and overt pronouns in language attrition
22.1 Introduction
22.2 The linguistic phenomena associated with null and overt pronominals
22.3 Adult L1 attrition in the pronominal domain
22.4 Summary and conclusion
Part IV: EXTRALINGUISTIC FACTORS IN LANGUAGE ATTRITION
Chapter 23: Introduction to extralinguistic factors in language attrition
23.1 Introduction and historical background
23.2 The european graduate network on language attrition
23.3 Attrition as a gradient phenomenon
23.4 Overview of the present section
23.5 Conclusion
Chapter 24: Age effects in language attrition
24.1 Introduction
24.2 Empirical findings on age effects in language attrition
24.2.1 Attrition in the absence of L contact
24.2.2 Attrition under reduced L contact
24.3 Theoretical accounts of age effects in language attritiion
24.3.1 Impediment accounts
24.3.2 A psychosocial account
24.3.3 The maturational account
24.4 Principles concerning age effects in attrition
Chapter 25: The impact of frequency of use and length of residence on L1 attrition
25.1 Introduction
25.2 Memory and the activation threshold hypothesis
25.3 What kind of L1 use?
25.4 The Empirical state of the art
25.4.1 Findings on L1 use
25.4.2 Findings on the impact of length of residence
25.5 Discussion and conclusion
Appendix
Chapter 26: L1 attrition, L2 development, and integration
26.1 Introduction
26.2 The bilingual mind
26.3 The Bicultural mind
26.4 The (assures) role of language in the integration outcomes
26.5 Conclusion
Chapter 27: Language contact and language attrition
27.1 Where does language contact take place?
27.2 Language contact in attrition processes
27.3 Key terms: borrowing, transference, convergence
27.4 Phenomena of language contact: types of transference and convergence
27.4.1 Lexical transference
27.4.2 Semantic transference/conceptual restructuring
27.4.3 Lexico-syntactic transference
27.4.4 Syntactic transference and convergence
27.4.4.1 Syntactic transference
27.4.4.2 Convergence/pivot matching
27.4.5 Phonological transference and convergence
27.4.6 Morphological transference
27.4.7 Pragmatic transference and convergence
27.5 External factors: language awareness and norm orientation in bilingual communities
Part V: Second language attrition
Chapter 28: Introduction to L2 attrition
28.1 Introduction
28.2 Terminological issues
28.3 Historical background
28.4 The baseline problem
28.5 State of the art
28.6 Delving into some of the studies
28.7 How we (try to) explain what we know
28.7.1 The Regression Hypothesis
28.7.2 The Savings paradigm
28.7.3 The Activation Threshold Hypothesis
28.7.4 Critical threshold
28.7.5 Dynamic Model of Multilingualism
28.8 Future research directions
28.9 Outline of part V
Chapter 29: Exploring the impact of extralinguistic factors on L2\FL Attrition
29.1 Introduction
29.2 From indivudual to cluster factors: A dynamic systems theory approach
29.3 Attained proficiency
29.4 The usual suspects: age, attitude and motivation, language contact, and use
29.4.1 Age
29.4.2 Attitude and motivation
29.4.3 Language contact and use
29.5 The great unknown: Context, length of residence/exposure
29.5.1 Context
29.5.2 Length of exposure/residence
29.6 Odds and ends: aptitude, level of education, literacy
29.6.1 Aptitude
26.6.2 Level of education
29.6.3 Literacy
29.7 Incubation
29.8 Discussion and conclusions
Acknowledgements
Chapter 30: Syntax and phonology in L2 attrition: Modularity and resilience
30.1 Introduction
30.2 Critical periods in acquisition and attrition
30.3 Patterns of retention in cases of childhood attrition
30.4 Resilience of phonology and syntax in L2 attrition
30.4.1 Attrition in L2 phonology
30.4.2 Attrition in L2 syntax
30.5 Discussion: disentangling syntax and phonology from lexical attrition
Chapter 31: L2 Lexical attrition
31.1 Introduction
31.2 Hypotheses concerning L2 Attrition
31.3 Memory loss and lexical attrition
31.4 Results from studies addressing L2 lexical attrition
31.4.1 Differences in L2 lexical attrition depending on age
31.4.2 Differences in L2 lexical attrition depending on incubation period
31.4.3 Proposing a definition for catastrophic loss andthe incubation time required to find such loss
31.4.4 Looking at studies with extremely long incubation periods, and finding an additional factor of target language
31.4.5 Which words are lost?
31.5 How Meara's constant decay hypothesis applied to the result of empirical L2 Attrition Research
31.6 Future research
Chapter 32: Attrition studies on japanese returnees
32.1 Introduction
32.2 Early investigatory stage
32.3 Hypothesis-testing stage
32.4 A new stage: Neurolinguistic approaches
Chapter 33: Event-related potentials as metrics of foreign language learning and loss
33.1 Language attrition
33.2 Event-related potentials as metrics of native-language comprehension
33.3 Event-related potentials as metric of foreign language learning and 'forgetting'
33.4 A longitudinal study of foreign language acquisition and attrition
33.4.1 Materials
33.4.2 Behavioural responses
33.4.3 ERPs elicited during the ‘learning phase’
33.4.4 Post-learning ‘attrition phase’
33.5 Discussion
Part VI: Heritage languages
Chapter 34: Introduction to heritage language development
34.1 Introduction
34.2 Bilingual acquisition: from infancy to adolescence
34.3 Language development and age
34.4 Other factors affecting the language development of heritage speakers
Chapter 35: Quantifying language experience in heritage language development
35.1 Introduction
35.2 The effect of specific properties of bilingual experience on (heritage) language development
35.3 How to qualtify bilingual/HL Experience
35.3.1 Conceptual issues
35.3.2 Practical issues
35.4 Survey of available parental questionnaires
35.4.1 Administration
35.4.2 Output
35.4.3 Population
35.4.4 Languages
35.5 A Final note on parental report
35.6 Conclusion
Chapter 36: Intra-generational attrition: Contributions to heritage speaker competence
36.1 Introduction
36.2 Attrition and its trickling down effects across generations
36.3 Empirical support
36.4 Conclusion
Chapter 37: 2L1 Simultaneous bilinguals as heritage speakers
37.1 Introduction
37.2 Language separation and CLI
37.3 AoO
37.4 Input and language dominance
37.5 Distance to the homeland
37.6 Conclusions
Acknowledgements
Chapter 38: Language loss and language learning in internationally adopted children: Evidence from behaviour and the brain
38.1 Introduction
38.1.1 Behavioural attrition
38.1.2 Neural attrition
38.1.3 Attrition summary
38.2 Implications of attrition for later language development/processing/learning
38.2.1 Implications for L1
38.2.2 Implications for L2
38.3 Conclusion
Acknowledgements
Chapter 39: Childhood language memory in adult heritage language (re)learners
39.1 Introduction
39.2 Heritage language loss among immigrant-background children
39.3 Language memory among childhood overhearers
39.4 Language menory amng childhood speakers of a heritage language
39.5 Evidence of childhood language menory among adult internationally adopted as infants
39.5.1 Re-accessing childhood language memory through relearning: the case of Korean-American adopted adults in Minnesota
39.5.1.1 Participants
39.5.1.2 Procedure
39.5.1.3 Language background assessment
39.5.1.4 Phoneme identification task
39.5.1.5 Phoneme production task
39.5.1.6 Results
39.5.1.6.I Language background assessment
39.5.1.6.II Phoneme perception
39.5.1.6.III Phoneme prouction
39.5.1.7 Relearning as one key to accessing international adoptees’ early childhood language memory
39.6 The long-lasting benefits of early childhood language experience
Chapter 40: Language development in bilingual returnees
40.1 Introduction
40.2 When the society language becomes the minority language
40.3 When the heritage language becomes the societal language
40.4 Concluding remarks
Concluding remarks
The coming of age of a research field
Defining language
Theoretical approaches: attrition and Universal Grammar
The balance hypothesis
The tidal waves of research on foreign language skills
Attrition and international adoptees
The unavoidable age factor
Some problematic areas
Units of analysis in attrition research
Incomplete acquisition and incomplete transmission between generations
Attrition and heritage languages
Suprasegmental aspects of language attrition
The attrition of gestures and sign languages
Attrition as a process rather than a state
Attrition at different time scales
To conclude
Acknowledgement
Annotated bibliography
References
Index

Citation preview

OUP CORRECTED PROOF – FINAL, 25/6/2019, SPi

   

LANGUAGE ATTRITION

OUP CORRECTED PROOF – FINAL, 25/6/2019, SPi

OXFORD HANDBOOKS IN LINGUISTICS Recently published

THE OXFORD HANDBOOK OF PRAGMATICS Edited by Yan Huang

THE OXFORD HANDBOOK OF UNIVERSAL GRAMMAR Edited by Ian Roberts

THE OXFORD HANDBOOK OF LANGUAGE AND SOCIETY Edited by Ofelia García, Nelson Flores, and Massimiliano Spotti

THE OXFORD HANDBOOK OF ERGATIVITY Edited by Jessica Coon, Diane Massam, and Lisa deMena Travis

THE OXFORD HANDBOOK OF WORLD ENGLISHES Edited by Markku Filppula, Juhani Klemola, and Devyani Sharma

THE OXFORD HANDBOOK OF POLYSYNTHESIS Edited by Michael Fortescue, Marianne Mithun, and Nicholas Evans

THE OXFORD HANDBOOK OF EVIDENTIALITY Edited by Alexandra Y. Aikhenvald

THE OXFORD HANDBOOK OF LANGUAGE POLICY AND PLANNING Edited by James W. Tollefson and Miguel Pérez-Milans

THE OXFORD HANDBOOK OF PERSIAN LINGUISTICS Edited by Anousha Sedighi and Pouneh Shabani-Jadidi

THE OXFORD HANDBOOK OF ENDANGERED LANGUAGES Edited by Kenneth L. Rehg and Lyle Campbell

THE OXFORD HANDBOOK OF ELLIPSIS Edited by Jeroen Van Craenenbroeck and Tanja Temmerman

THE OXFORD HANDBOOK OF LYING Edited by Jörg Meibauer

THE OXFORD HANDBOOK OF TABOO WORDS AND LANGUAGE Edited by Keith Allan

THE OXFORD HANDBOOK OF MORPHOLOGICAL THEORY Edited by Jenny Audring and Francesca Masini

THE OXFORD HANDBOOK OF REFERENCE Edited by Jeanette Gundel and Barbara Abbott

THE OXFORD HANDBOOK OF EXPERIMENTAL SEMANTICS AND PRAGMATICS Edited by Chris Cummins and Napoleon Katsos

THE OXFORD HANDBOOK OF EVENT STRUCTURE Edited by Robert Truswell

THE OXFORD HANDBOOK OF LANGUAGE ATTRITION Edited by Monika S. Schmid and Barbara Köpke For a complete list of Oxford Handbooks in Linguistics please see pp. –.

OUP CORRECTED PROOF – FINAL, 25/6/2019, SPi

   

......................................................................................................................

LANGUAGE ATTRITION ...................................................................................................................... Edited by

MONIKA S. SCHMID and

BARBARA KÖPKE with

MIRELA CHERCIOV TUĞBA KARAYAYLA MEREL KEIJZER ESTHER DE LEEUW TEODORA H. MEHOTCHEVA SILVINA MONTRUL MARIA POLINSKY

1

OUP CORRECTED PROOF – FINAL, 25/6/2019, SPi

3

Great Clarendon Street, Oxford,  , United Kingdom Oxford University Press is a department of the University of Oxford. It furthers the University’s objective of excellence in research, scholarship, and education by publishing worldwide. Oxford is a registered trade mark of Oxford University Press in the UK and in certain other countries © editorial matter and organization Monika S. Schmid and Barbara Köpke  © the chapters their several authors  The moral rights of the authors have been asserted First Edition published in  Impression:  All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, without the prior permission in writing of Oxford University Press, or as expressly permitted by law, by licence or under terms agreed with the appropriate reprographics rights organization. Enquiries concerning reproduction outside the scope of the above should be sent to the Rights Department, Oxford University Press, at the address above You must not circulate this work in any other form and you must impose this same condition on any acquirer Published in the United States of America by Oxford University Press  Madison Avenue, New York, NY , United States of America British Library Cataloguing in Publication Data Data available Library of Congress Control Number:  ISBN –––– Printed and bound by CPI Group (UK) Ltd, Croydon,   Links to third party websites are provided by Oxford in good faith and for information only. Oxford disclaims any responsibility for the materials contained in any third party website referenced in this work.

OUP CORRECTED PROOF – FINAL, 25/6/2019, SPi

C

................................

Acknowledgements List of abbreviations The contributors

ix x xv

. Introduction



M S. S  B Kö

PART I THEORETICAL IMPLICATIONS OF LANGUAGE ATTRITION . Language attrition and the Competition Model



B MW

. Language attrition and the Feature Reassembly Hypothesis



M T. P, S P-C,  L Sˊ 

. The Interface Hypothesis as a framework for studying L attrition



G C  A S

. Implications of the Bottleneck Hypothesis for language attrition



R S

. A Complex Dynamic Systems perspective on personal background variables in L attrition



C O

PART II PSYCHOLINGUISTIC AND NEUROLINGUISTIC APPROACHES TO LANGUAGE ATTRITION . Introduction to psycholinguistic and neurolinguistic approaches to language attrition B K̈    M K



OUP CORRECTED PROOF – FINAL, 25/6/2019, SPi

vi



. Language attrition as a special case of processing change: A wider cognitive perspective



M S S

. Memory retrieval and language attrition: Language loss or manifestations of a dynamic system?



J A. L  J F. K

. How bilingualism affects syntactic processing in the native language: Evidence from eye movements



P E. D, J R. Vˊ  K, M J,  Á V

. First language attrition and Developmental Language Disorder



E B, T B,  J  J

. Ageing as a confound in language attrition research: Lexical retrieval, language use, and cognitive and neural changes



E H, A L, M K, T M,  L K. O

. Linguistic regression in bilingual patients with Alzheimer’s Disease



M B-D, Fˊ ˊ  G, B Kö,  L L

. Electrophysiological approaches to L attrition



K S  K K

. Neuroimaging perspectives on L attrition and language change



E R, Y P,  M T. D

PART III LINGUISTIC FACTORS IN LANGUAGE ATTRITION . Introduction to linguistic factors in language attrition



M S. S  E  L

. Phonetic drift



C B. C

. Phonetic attrition E  L



OUP CORRECTED PROOF – FINAL, 25/6/2019, SPi



. Phonological attrition

vii



C C

. Morphological attrition



E S

. Lexical attrition



S J

. Null and overt pronouns in language attrition



A̧  G̈  

PART IV EXTRALINGUISTIC FACTORS IN LANGUAGE ATTRITION . Introduction to extralinguistic factors in language attrition



M S. S  M C

. Age effects in language attrition



E B

. The impact of frequency of use and length of residence on L attrition



M S. S

. L attrition, L development, and integration



G̈   Y

. Language contact and language attrition



C M R

PART V SECOND LANGUAGE ATTRITION . Introduction to L attrition



T H. M  B Kö

. Exploring the impact of extralinguistic factors on L/FL attrition



T H. M  K M

. Syntax and phonology in L attrition: Modularity and resilience



K B-H  D S

. L lexical attrition J L-H



OUP CORRECTED PROOF – FINAL, 25/6/2019, SPi

viii



. Attrition studies on Japanese returnees



H T

. Event-Related Potentials as metrics of foreign language learning and loss



L O, I P̈  , J ML,  M Z

PART VI HERITAGE LANGUAGES . Introduction to heritage language development



S M  M P

. Quantifying language experience in heritage language development



S U

. Intra-generational attrition: Contributions to heritage speaker competence



F B, D P  C,  J R

. L simultaneous bilinguals as heritage speakers



T K

. Language loss and language learning in internationally adopted children: Evidence from behaviour and the brain



L J. P, F G,  D K

. Childhood language memory in adult heritage language (re)learners



J S. O, T K- A, S-A J,  R M. L

. Language development in bilingual returnees



C F

Concluding remarks K  B

Annotated bibliography

 

M S. S

References Index

 

OUP CORRECTED PROOF – FINAL, 25/6/2019, SPi

A

......................... .......................................

We are very grateful to Asha Ali and Rachel Ting Yan Kan, who helped us put this volume together and keep track of too many things to list here. We are also deeply indebted to all of the anonymous reviewers whose input was vital in improving the quality of the chapters collected here. Joy Mellor, Kumar Anbazhagan, and Vicki Sunter from OUP were an absolute joy to work with and always responded promptly and kindly to our many (and often grumpy—apologies again, Joy, Kumar, and Vicki!) queries. Finally, we would like to express our gratitude and appreciation to the artist Bruno Catalano and his representative, Emilie Catalano, for kindly allowing us to reproduce a picture of one of his stunning sculptures on the cover of this volume. The field of language attrition has many different parents, but we would particularly like to remember two of the most influential ones here: Richard D. Lambert and Theo van Els, both of whom passed away in . They will be fondly remembered and the field is much poorer for their loss.

OUP CORRECTED PROOF – FINAL, 25/6/2019, SPi

L  

........................................................................

AA

age of attrition

ACC

accusative

AD

Alzheimer’s Disease

Af S

affective structure

AGC

corpus of German immigrants in Australia

AL

Applied Linguistics

AMTB

attitude and motivational test battery

ANOVA

analysis of variance

AOA

age at onset of attrition

AoA

age of acquisition

AoO

age of onset

AOR

age of return

APT

Attrition by Processing Theory

ASL

American Sign Language

AaT

age at testing

AT

Activation Threshold

ATH

Activation Threshold Hypothesis

ATM

attitude and motivation

BAT

bilingual aphasia test

BH

Bottleneck Hypothesis

BIA

Bilingual Interactive Active Model

BIA+

Bilingual Interactive Active Model Plus

BOLD

blood-oxygen-level dependent

BP

Brazilian Portuguese

CC

Cognate Correct

CD

close duration/closure duration

CDST

Complex Dynamic Systems Theory

CELF

Clinical Evaluation of Language Fundamentals

CHILDES

Child Language Data Exchange System

CHL

computational procedure for human language

OUP CORRECTED PROOF – FINAL, 25/6/2019, SPi

  

CLI

cross-linguistic influence

CLLD

clitic left-dislocations

COMP

Complementizer

CP

complementizer phrase

CPH

Critical Period Hypothesis

CS

conceptual structure

CSF

cerebrospinal fluid

CT

Chaos/Complexity Theory

DAT

Dative

DLD

Developmental Language Disorder

DM

Distributed Morphology

DMM

Dynamic Model of Multilingualism

DOM

differential object marking (or marker)

DP

determiner phrase

D/P

declarative/procedural

DS

Dynamic System

DSM

Diagnostic and Statistical Manual of Mental Disorders

DST

Dynamic Systems Theory

DTI

Diffusion Tensor Imaging

DV

dependent variable

EC

English-correct sentences

EEG

electroencelphalogram

EFL

English as a Foreign Language

ELAN

early left anterior negativity

EP

European Portuguese

EPP

Extended Projections Principle

ERLAC

emotion-related language choice

ERP

Event-Related (brain) Potential

ESRC

Economic and Social Sciences Research Council

EVT

Ethnolinguistic Vitality Theory

FA

false alarms

FEM

feminine

FF

functional features

FL

foreign language

fMRI

functional Magnetic Resonance Imaging

fNIRS

functional Near Infrared Spectroscopy

xi

OUP CORRECTED PROOF – FINAL, 25/6/2019, SPi

xii

  

FRH

feature reassembly hypothesis

GAMM

generalized additive mixed models

GEN

genitive

GG

grammatical gender

HBP

Heritage Bilingual Paradox

HL

heritage language

HP

high proficiency

HS

heritage speaker

IA

International Adoptees/internationally adopted

IC

Inhibitory Control

ICFLA

International Conference on First Language Attrition

IH

Interface Hypothesis

INC

incubation period

IV

independent variable

JND

just-noticeable difference

L

language

LAD

language acquisition device

LAE

language acquisition effort

LAN

left anterior negativity

LATB

Language Attrition Test Battery

LC

linguistically competent user

LF

logical form

LI

Language Impairment

LID

Language Input Diary

LIFO

last in, first out

LME

language maintenance effort

LOR

length of residence

LTM

long-term memory

LTN

left temporal negativity

MANOVA

multiple analysis of variance

MASC

masculine

MCI

mild cognitive impairment

MEG

magnetoencephalography

MICD

Missing Input Competence Divergence

MOGUL

Modular Online Growth and Use of Language

OUP CORRECTED PROOF – FINAL, 25/6/2019, SPi

  

MP

Minimalist Program

MRI

Magnetic Resonance Imaging

MTLD

Measure of Textual Lexical Diversity

NIRS

Near Infrared Spectroscopy

NOM

nominative

NP

noun phrase

OA

onset age

OPC

overt pronoun constraint

PAS

Position of Antecedent Strategy

Pas

passive

PDP

parallel distributed processing

PET

positron emission tomography

PF

phonological form

PL

plural

PLI

Primary Language Impairment

PLU

plural

POp

Pars Opercularis

POSS

possessive

PRES

present

PRG

progressive

PrG

Precentral Gyrus

PRO

pronoun

PSA-NCAM

polysialic acid-neural cell adhesion molecule

PT

planum temporale

PTr

Pars Triangularis

PV

parvalbumin

PWM

phonological working memory

RH

Regression Hypothesis

RIF

retrieval induced forgetting

RL

recipient language

RP

Retrieval Practice

RP–

Retrieval Practice minus

RP+

Retrieval Practice plus

RT

reaction times

SA

study abroad

xiii

OUP CORRECTED PROOF – FINAL, 25/6/2019, SPi

xiv

  

SAI

short-latency afferent inhibition

SD

standard deviation

SES

socio-economic status

SG

singular

SING

singular

SL

source language

SLA

second language acquisition

SLD

second language development

SLI

Specific Language Impairment

SLM

Speech Learning Model

SLP

speech-language therapist

SOV

subject–object–verb

SPEC

specifier

SSBE

Standard Southern British English

STG

Superior Temporal Gyrus

STS

Superior Temporal Sulcus

SVO

subject–verb–object

TD

typical development/typically developing

TL

target language

TLU

target-like usage

TOEFL

Test of English as a Foreign Language

TOT

tip of the tongue

TTR

type–token ratio

UB

usage-based

UG

Universal Grammar

VFT

Verbal Fluency Task

VOCD

vocabulary diversity

VOT

voice onset time

WM

working memory

OUP CORRECTED PROOF – FINAL, 25/6/2019, SPi

T  ............................................................

Kathleen Bardovi-Harlig, Professor of Second Language Studies, Indiana University, teaches and conducts research on second language acquisition, temporal semantics, pragmatics, formulaic language as it relates to pragmatics, and attrition (in joint work with David Stringer). Bardovi-Harlig and Stringer’s work on attrition has appeared in SSLA, Linguistic Approaches to Bilingualism, and edited volumes. Melissa Barkat-Defradas received her PhD in Language Sciences in  at University of Lyon and received the Young Researcher Award for her work. After a research fellowship at UC Berkeley she integrated the French National Centre for Scientific Research. She is now full-time researcher at The Institute of Evolutionary Sciences of Montpellier (France), where she develops interdisciplinary works that contribute to bridge the gap between linguistics and evolutionary biology. Her recent researches focus on psycholinguistic aspects of language processing in bilingual subjects suffering from Alzheimer’s disease. She is currently interested in the determination of biological and environmental factors that may explain senescence in humans. Fatih Bayram is a post-doctoral research fellow at UiT the Arctic University of Tromsø, funded by the European Union’s Horizon  research and innovation programme under the Marie Skłodowska-Curie Individual Fellowship grant agreement No. . His research covers various types of child and adult bilingualism with a primary focus on heritage language bilinguals. Some of his recent research has appeared in International Journal of Bilingualism and Bilingual Education, Linguistic Approaches to Bilingualism, and Frontiers in Psychology. Elma Blom is Professor at the Department of Special Education at Utrecht University where she teaches about language development. Her research and publications are about developmental language disorder, the parallel development of multiple languages in immigrant children, and the relationship between language and cognition both in language-impaired and multilingual children. Besides theoretical issues, she works on the improvement of diagnostic tools for multilingual children. Tessel Boerma is a linguist with expertise on child language development and impairment. In , Tessel started her PhD at Utrecht University, investigating the linguistic and cognitive development of monolingual and bilingual children with developmental language disorder. Tessel defended her dissertation in October  and now continues her work on language development in clinical child populations at Utrecht University. She is a postdoctoral researcher in the research programme ‘Language impairment in the q. deletion

OUP CORRECTED PROOF – FINAL, 25/6/2019, SPi

xvi

 

syndrome’. Moreover, she is involved in one of the interdisciplinary projects of Dynamics of Youth, entitled ‘The first  critical days of a child’s life’. Kees de Bot graduated from the University of Nijmegen and became Professor of Applied Linguistics in Nijmegen and later in Groningen. His main interests are multilingualism, psycholinguistics, language policy, and more recently the philosophy of Science and the role of evidence in science and how that relates to notions of truth in society at large. He has published in most of the leading journals in applied linguistics. His most recent book is A History of Applied Linguistics –. (Routledge, ). He is now a full Professor of Applied Linguistics at the University of Pannonia in Hungary. Emanuel Bylund is Professor of General Linguistics (Stellenbosch University). His research concerns age effects in language acquisition and attrition, and conceptual representation in bilinguals. He has worked on languages such as Afrikaans, German, isiXhosa, Spanish, and Swedish. Chiara Celata is a researcher of linguistics and speech sciences at Scuola Normale Superiore of Pisa. She has a background in diachronic phonology and phonological acquisition, with particular reference to L Italian. She has taught applied linguistics at the University of Siena and L phonological attrition at CSIC and UIMP, Madrid. Her additional research interests include the sociophonetics of the Romance languages and the study of regional and individual variation according to multi-level acoustic and articulatory approaches. She has authored or co-authored more than fifty papers in scientific journals and volumes, including the editing of international books and journal special issues, in the domain of speech variation, sound change, and bilingual phonology. Gloria Chamorro is Lecturer in Applied Linguistics at the University of Kent. She obtained her PhD in Developmental Linguistics from the University of Edinburgh, and has researched and published on first language attrition, adult second language acquisition, and bilingual education. She has also developed the public engagement project English Language Support for Refugees, which helps unaccompanied refugee minors gain the English language skills they need to integrate into the community and access mainstream education. Charles B. Chang is Assistant Professor of Linguistics at Boston University, where he directs the Phonetics, Acquisition & Multilingualism Lab (PAMLab) and is Affiliated Faculty with the Center for the Study of Asia and the Center for Research in Sensory Communication & Emerging Neural Technology. His research concerns the processing, representation, and development of speech sounds in the context of multilingualism and language contact, the cognitive organization of dual phonological systems, and the cross-language interactions that occur during second-language learning and first-language attrition. Mirela Cherciov obtained her PhD in Linguistics (Department of French) in  from the University of Toronto. Her thesis, titled Between Attrition and Acquisition: the Dynamics between Two Languages in Adult Migrants, was partly funded by a grant

OUP CORRECTED PROOF – FINAL, 25/6/2019, SPi

 

xvii

from the Social Sciences and Humanities Research Council. She has since held positions and taught at York University (Glendon Bilingual Campus) and Brock University. Her research focuses on linguistic and sociolinguistic factors linked to L attrition in adult migrants, bilingual language development across the lifespan, adult L acquisition, and pedagogy of FSL. Michele T. Diaz, PhD, is the Director of Human Imaging and an Associate Professor of Psychology at the Pennsylvania State University. Her research focuses on the neural representation of language and how this representation changes with age. In particular, she is interested in age-related differences in phonological and semantic aspects of language production. Her lab combines behavioural, neuropsychological, and neuroimaging measures to examine the relations between age, cognition, and the brain. She has also investigated contributions of the right hemisphere to language by studying figurative language, discourse, and novelty. Finally, she worked on the Biomedical Informatics Research Network focusing on best practices in neuroimaging and multi-site research. Paola E. Dussias is Professor of Spanish, Linguistics, and Psychology at Pennsylvania State University. The research that she and her students conduct employs methodological tools from linguistics, experimental psycholinguistics, and cognitive neuroscience to examine the way in which bilingual readers and speakers negotiate the presence of two languages in a single mind. They also examine the variables that incur processing costs during the comprehension and production of code-switched language. Their work, supported by grants from the National Science Foundation and the National Institutes of Health, has shown that the native language is open to influence from the second language, and that not all code-switches incur processing costs. Cristina Flores is Associate Professor of German Linguistics in the Department of German and Slavic Studies at the University of Minho/Portugal. Her research focuses on bilingual language development, especially on heritage language acquisition and on language attrition in returnees. She has coordinated several projects on Portuguese–German bilingualism. She is an integrated researcher at the Center for Humanistic Studies of the University of Minho, of which she is also vice-director. Frédérique Gayraud is Professor of Psycholinguistics at Lyon University. Her main research interests include early language acquisition and language decline in Alzheimer’s disease in a lifespan perspective. More specifically, the goal is to compare acquisition and desacquisition processes in order to test the retrogenesis hypothesis according to which degenerative mechanism reverse the order of acquisition in normal development. Another research interest focuses on bilingualism in normal and pathological ageing. Fred Genesee is Professor Emeritus in Psychology, McGill University. He has conducted extensive research on alternative forms of bilingual/immersion education for language minority and majority students, the academic development of at-risk students in bilingual programmes, language acquisition in typically-developing and at-risk pre-school bilingual children, and internationally adopted children. He has published numerous articles in

OUP CORRECTED PROOF – FINAL, 25/6/2019, SPi

xviii

 

scientific journals and magazines and has authored sixteen books on bilingualism. He is the recipient of the Canadian Psychology Association Gold Medal Award, Paul Pimsler Award for Research in Foreign Language Education, and the Canadian Psychological Association Award for Distinguished Contributions to Community or Public Service. Ayşe Gürel holds a PhD in Linguistics from McGill University. She is a full professor at the Department of Foreign Language Education of Boğaziçi University. Her research interests include first language attrition and second language acquisition of morpho-syntactic features as well as neurolinguistic and psycholinguistic aspects of bilingualism. She has publications in international journals and books in her fields of interest and she has been serving in the editorial boards of Second Language Research and Linguistic Approaches to Bilingualism. Among her current work is an edited volume on Second Language Acquisition of Turkish published by John Benjamins. Eve Higby is a postdoctoral researcher in Psychology at the University of California, Riverside. She received her PhD in Speech-Language-Hearing Sciences from the City University of New York Graduate Center. Her research focuses on cross-linguistic influences in bilingualism, cognitive contributions to language processing, and brain-behaviour relationships in bilingualism and ageing. Scott Jarvis is Professor and Chair of the Department of Linguistics at the University of Utah. His research concentrations include cross-linguistic influence and lexical diversity. In the former domain, he has developed frameworks for exploring conceptual transfer and confirming cross-linguistic influence in cases where it is difficult to detect or is confounded by competing variables. In the domain of lexical diversity, he has concentrated on developing a multidimensional construct definition of this phenomenon and an accompanying set of measures that account for how lexical diversity is perceived and how it contributes to the quality of both spoken and written language. Michael Johns is a PhD student in Spanish and Language Science at Penn State University. His research focuses on the integration of psycholinguistic methodologies, such as eyetracking and pupillometry, with insights from sociolinguistic data and bilingual corpora to investigate the processing of code-switched speech. A central theme to his research is the role of input and usage in shaping the processing system, and the interaction between external, social factors, and internal, cognitive factors. Jan de Jong is Associate Professor at the Department of Biological and Medical Psychology of the University of Bergen (Norway). Previously he worked at the University of Amsterdam where he is still a guest researcher. His past research has addressed grammatical symptoms of language impairment in Dutch, linguistic precursors of dyslexia and, most recently, language impairment in a bilingual context. He was Vice Chair of COST Action IS ‘Language impairment in a multilingual society’ and, together with Sharon Armon-Lotem and Natalia Meir, edited Assessing Multilingual Children (), a volume that summarizes the diagnostic tools created in the Action.

OUP CORRECTED PROOF – FINAL, 25/6/2019, SPi

 

xix

Sun-Ah Jun received her PhD in Linguistics from Ohio State University and is Professor at the Department of Linguistics at the University of California, Los Angeles (UCLA). Her research focuses on intonational phonology, prosodic typology, the interface between prosody and the various subareas of linguistics, and language acquisition. She has published the book, The Phonetics and Phonology of Korean Prosody: Intonational Phonology and Prosodic Structure (Garland ) and edited two volumes of Prosodic Typology: The Phonology of Intonation and Phrasing (Oxford University Press,  & ). Tuğba Karayayla obtained her PhD in Applied Linguistics in  from the University of Essex. Her thesis is on structural and lexical development of Turkish spoken as an immigrant and a heritage language in the UK, investigating attrition of Turkish and its development as a heritage language. Her research areas mainly include language acquisition, language attrition, bi- and multilingualism, and heritage language development. She is working as a lecturer at the Department of English Language and Literature, Yıldırım Beyazıt University, Ankara. Kristina Kasparian is a research associate at McGill University (Montreal, Canada) where she studies the neurocognition of language and multilingualism. Her PhD research used event-related-potentials to examine first language attrition in Italian native-speakers living in Canada, compared to Italian second-language learners. Kristina is especially interested in the factors that modulate native-like language processing, in cross-linguistic influence in different populations of multilinguals, and the parallels that can be drawn between L attrition, L learning, and L attrition. Kristina has also worked on the linguistic and sociolinguistic development of internationally adopted children and on cross-linguistic influences at play in adult bilinguals’ reading processes. Merel Keijzer is an associate professor of Applied Linguistics at the University of Groningen, where she also holds a position as Rosalind Franklin Research Fellow. She was the president of the Dutch association of Applied Linguistics and is currently the vice-president of the Young Academy Groningen and a member of the Young Academy (Jonge Akademie) of the Royal Dutch Academy of Arts and Sciences, where she chairs the interdisciplinarity working group. In her work on bilingualism across the lifespan, too, she adopts an interdisciplinarity framework. Focusing on the effects of long-term and short-term bilingual experiences in older adulthood, the research fields she actively tries to bring together are linguistics, cognitive science and psychology, neuroscience, and geriatrics. As part of the dynamics of bilingualism research, she is also active in language attrition research and wrote several papers and contributed to and edited various volumes on the topic. Her work has received funding from the Netherlands Organisation for Scientific Research (NWO). Terry Kit-fong Au obtained her PhD in Psychology from Stanford University. She was Associate Professor at Brown University and Professor at UCLA, and is currently Chair Professor of Psychology and holds the Karen Lo Eugene Chuang Professorship in Diversity and Equity at the University of Hong Kong. Professor Au is an elected Fellow of the Association of Psychological Science and has served as Associate Editor of the journal

OUP CORRECTED PROOF – FINAL, 25/6/2019, SPi

xx

 

Developmental Psychology, and research grant panel member for NIMH, NIH, and the Research Grants Council of Hong Kong. Denise Klein is Director of the Centre for Research on Brain, Language and Music, and Associate Professor in Neurology and Neurosurgery, McGill University. She has been in the Cognitive Neuroscience Unit at the Montreal Neurological Institute since  where she did her postdoctoral fellowship under Brenda Milner after completing her PhD at the University of Witwatersrand in South Africa. Denise’s early work pioneered the use of brain imaging for the study of bilingualism and provided a springboard for current debates about bilingual brain organization. Her current work focuses on finding behavioural and neural biomarkers that predict language-learning success and difficulty. Barbara Köpke is Professor of Neuropsycholinguistics at the University of Toulouse  and head of the Octogone-Lordat Laboratory. Her research involves neuro- and psycholinguistic aspects of language processing in bilingual subjects with specific attention to ‘extreme’ situations such as L attrition, simultaneous interpreting, and aphasia. Her work has appeared in journals such as Linguistic Approaches to Bilingualism and International Journal of Bilingualism, and she is the editor of a special issue of Journal of Neurolinguistics on first language attrition. Marta Korytkowska is currently a doctoral student in the Speech-Language-Hearing Sciences department at CUNY Graduate Center. She is a practising Speech-Language Pathologist in the acute care setting with experience in acute rehabilitation, outpatient, and home care settings. Her main areas of research interest include treatment approaches in bilingual populations with language disorders, bilingual aphasia, and the influence of cognition in recovery from aphasia. Judith F. Kroll is Distinguished Professor at the University of California, Riverside and the former director of the Center for Language Science at Pennsylvania State University. Her research uses the tools of cognitive neuroscience to understand how bilinguals juggle the presence of two languages in one mind and brain. Her work, supported by NSF and NIH, shows that bilingualism provides a lens for revealing the interplay between language and cognition that is otherwise obscure in speakers of one language alone. Tanja Kupisch is Full Professor at the Department of Linguistics, University of Konstanz and Professor II at UiT The Arctic University of Norway. She got her PhD from the University of Hamburg in . She has worked as a researcher and teacher at the universities of Calgary, McGill, Hamburg, Lund, Konstanz, and Tromsø, specializing in first, second and third language acquisition, heritage bilingualism, bilectal acquisition, and language attrition, including child and adult learners. Her work focuses on phenomena such as nominal syntax and foreign accent, including the languages French, Italian, English, German, Turkish, and Russian. Jenifer Larson-Hall is Associate Professor of Applied Linguistics at Kitakyushu University, Japan. Her interest in attrition stems from the effects that age has on the loss of a language as well as the acquisition of it. She undertook attrition research as a quixotic way to discover

OUP CORRECTED PROOF – FINAL, 25/6/2019, SPi

 

xxi

the ideal age at which a young person might be ‘dipped’ for several years into a different language and emerge as something close to a balanced bilingual. After raising four of her own bilinguals she now sees that learning a language is a much bigger and longer project than she had even imagined. Richard M. Lee, PhD is the Distinguished McKnight University Professor in the Department of Psychology at the University of Minnesota, Twin Cities Campus. His research focuses on the psychological aspects of culture, race, and ethnicity and its role in the development, well-being, and mental health of racial and ethnic minority youth and families. Since , he has been studying the transnational and transracial experiences of families and children who were adopted internationally from South Korea. Esther de Leeuw obtained her PhD in Speech and Hearing Sciences in  from Queen Margaret University, Edinburgh. The title of her PhD thesis was ‘When your native language sounds foreign: A phonetic investigation into first language attrition’. She has since held positions at the Centre for Research on Bilingualism at Bangor University, Wales, and at the BBC Pronunciation Unit, in London, UK. She has also spent research sabbaticals at the Lifespan Cognition and Development Lab at York University, Canada, and at the Institute for Phonetics and Speech Processing at the LMU Munich, Germany. She is now Senior Lecturer and Director of the Phonetics Laboratory at Queen Mary University of London, and Associate Editor for the journal Second Language Research. She has published widely on phonetic and phonological dimensions of first language attrition, as well as on speech and bilingualism more generally. Laurent Lefebvre has acquired an expertise in the study of language and executive functioning deficits encountered by patients with neurodegenerative diseases (Alzheimer’s disease, Primary Progressive Aphasia, Vascular Dementia). He is at the head of the UMONS department of Cognitive Psychology and Neuropsychology since . He created some neuropsychological tests (e.g. GREMOTS, DTLA) and cognitive interventions (e.g. Logaatome) specifically dedicated for aged patients with cognitive deficits. Laurent Lefebvre is a member of several scientific and patients’ groups committees in neuropsychology and dementia. He is Full Professor at the University of Mons and Dean of the Faculty of Psychology and Education. Aviva Lerman is currently a doctoral student in the Speech-Language-Hearing Sciences department at CUNY Graduate Center. She is a qualified SLP and holds a Master’s degree specializing in bilingualism and biculturalism. Her research focuses on bilingualism, aphasia, healthy ageing, and dementia. Jared A. Linck holds a PhD in Cognitive Psychology from The Pennsylvania State University. He is a Research Scientist at the University of Maryland Center for Advanced Study of Language (CASL), where he conducts basic and applied research to improve foreign language capabilities. Dr Linck’s research interests include the psycholinguistics of multilingual language processing, second language acquisition, individual differences in learning processes and outcomes, and the impact of both technology use (e.g., web apps,

OUP CORRECTED PROOF – FINAL, 25/6/2019, SPi

xxii

 

virtual reality) and the context of learning (e.g., immersion, online, self-study) on second language learning outcomes. Brian MacWhinney is Professor of Psychology, Computational Linguistics, and Modern Languages at Carnegie Mellon University. His Unified Competition Model analyses first and second language learning as aspects of a single basic system. He has developed a series of thirteen TalkBank open access online databases for the study of language learning, multilingualism, and language disorders. His research topics include methods for online learning of second language vocabulary and grammar, neural network modelling of lexical development, fMRI studies of children with focal brain lesions, ERP studies of between-language competition, and the role of embodied perspectival imagery in sentence processing. Taryn Malcolm is currently a doctoral student and member of Loraine Obler’s Neurolinguistics lab in the Speech-Language-Hearing Science Department at The Graduate Center, CUNY. She has practised as a speech-language pathologist in acute and sub-acute rehabilitation with a focus on neurogenic disorders and respiratory/voice disorders. Her main areas of research interest include bilingualism, bilingual aphasia, and neurological processes underlying acquired language disorders. She is currently working on a research project investigating cross-linguistic influence in speakers of Jamaican Creole following immersion in the environment of their second language, English. Judith McLaughlin received her PhD in Psychology at the University of Washington. After receiving her degree, Dr McLaughlin served as a Research Scientist in the Cognitive Neuroscience of Language lab at the University of Washington for nearly two decades. Among her numerous contributions is the finding that foreign-language learners’ brain activity discriminates between foreign language words and ‘pseudowords’ after just fourteen hours of foreign instruction, even though the learners were at chance when making conscious word/non-word judgements about these stimuli. She is currently a Lecturer in the Department of Psychology at the University of Washington. Teodora H. Mehotcheva earned a double PhD in Applied Linguistics from Pompeu Fabra University, Barcelona, Spain, and University of Groningen, Groningen, the Netherlands. She is an Assistant Professor of Second Language Studies at New Bulgarian University, Sofia, Bulgaria. Her research interests include second and foreign language attrition, and foreign language acquisition and teaching of English. Silvina Montrul is Professor and former Head of the Department of Spanish and Portuguese and Professor in the Department of Linguistics at the University of Illinois at Urbana-Champaign. Her research focuses on linguistic and psycholinguistic approaches to second language acquisition and bilingualism, with particular emphasis on heritage speakers. She is co-editor of Second Language Research and author of The Acquisition of Spanish (Benjamins, ), Incomplete Acquisition in Bilingualism (Benjamins, ), El bilingüismo en el mundo hispanohablante (Bilingualism in the Spanish-speaking world)

OUP CORRECTED PROOF – FINAL, 25/6/2019, SPi

 

xxiii

(Wiley-Blackwell, ), and Heritage Language Acquisition (Cambridge University Press, ), in addition to over a hundred peer reviewed articles and chapters. Kleopatra Mytara (MA, University of Strasbourg) is a PhD student at the University of Toulouse , France and a member of the Interdisciplinary Research Unit Octogone-Lordat (EA ). Her research interests include second and foreign language attrition, second and foreign language acquisition and teaching, particularly French as a foreign language, as well as multilingualism and language control. Loraine K. Obler is Distinguished Professor at the CUNY Graduate Center in the programmes of Speech-Language-Hearing Sciences and Linguistics. Her first book, which opened up the field of neurolinguistics of bilingualism, was The Bilingual Brain: Neuropsychological and Neurolinguistic Approaches to Bilingualism (with M. Albert). More recently she has published articles and books on bilingual aphasia and cross-language study of aphasia. As part of her body of work on the language changes associated with ageing, she is now working on L learning in older adults. Janet S. Oh received her PhD in Psychology from UCLA. She is Professor of Psychology and Senior Director of Institutional Research at California State University, Northridge. Dr Oh’s work focuses on the experiences and development of children who come from linguistic minority backgrounds, including the development of children who are exposed to more than one language, as well as adults who are (re)acquiring their heritage language. Her current research focuses on the development of a three-language immersion model for preschoolers. Conny Opitz is Lecturer in German Studies at Cardiff University and holds a PhD from Trinity College Dublin, where she taught Russian, Linguistics, and Translation. Her main research interest concerns the multilingual development of varied populations of adult learners and users from the perspective of Complex Dynamic Systems Theory. Her publications include a co-edited special issue of the International Journal of Bilingualism entitled ‘Dynamics of first language attrition across the lifespan’ (, with E. de Leeuw and D. Lubińska) and a contribution to the volume Complexity Theory and Language Development: In Celebration of Diane Larsen-Freeman (, eds. L. Ortega and Z. Han). Lee Osterhout received his PhD in Psychology from Tufts University. Dr Osterhout uses electrophysiological methods to study language comprehension in native speakers and in foreign-language learners. His research with native speakers indicated that semantic and syntactic anomalies elicit distinct and independent brain responses. He also conducted some of the earliest work demonstrating changes in brain activity that occur during foreign-language learning. He is a Professor in the departments of Psychology and Linguistics the University of Washington, and a member of the university’s Graduate Program in Neuroscience. Diego Pascual y Cabo is Assistant Professor of Hispanic Linguistics and Director of the Spanish Heritage Language Program at the University of Florida. His primary research interest is heritage speaker bilingualism. Over the past few years, his work on this topic has

OUP CORRECTED PROOF – FINAL, 25/6/2019, SPi

xxiv

 

appeared in several scholarly journals such as Applied Linguistics, Linguistic Approaches to Bilingualism, Heritage Language Journal, and International Review of Applied Linguistics, among others. Silvia Perez-Cortes is Assistant Professor of Spanish at Rutgers University, Camden. She holds a PhD in Bilingualism and SLA from Rutgers University, New Brunswick, and MA in Hispanic Linguistics from the University of Massachusetts-Amherst () and a Bachelor of Arts in Spanish Philology from the University of Barcelona (). Her research interests lie in the areas of bilingual language acquisition in language contact situations. She is particularly interested in analysing the grammatical development of heritage populations (both children and adults) with the objective of exploring how syntax and the lexicon are accessed and represented in the bilingual mind. Lara J. Pierce is a postdoctoral research fellow at Harvard Medical School/Boston Children’s Hospital. She received her PhD in Psychology from McGill University. Her research centres on the effects of early experience on development. In particular, how deviation from ‘typical’ early experiences (e.g., variation in the early language environment, exposure to early stress/adversity) influence the developing brain, and the extent to which the nature and timing of early experiences have a lasting effect on the brain and behaviour. She has received awards from the Canadian Psychological Association, NSERC, Centre for Research on Brain, Language, and Music, and McGill University. Ilona Pitkänen received her PhD in Psychology at the University of Washington. She is currently on the teaching faculty in the Department of Psychology at Bellevue College in Bellevue, WA. Maria Polinsky is Professor of Linguistics at the University of Maryland. Yanina Prystauka received a BA in Translation Studies from Belarusian State University (Minsk, Belarus) in  and an MS in Cognitive Science from the University of Trento (Rovereto, Italy) in . She is currently a PhD student in the Language and Cognition programme at the University of Connecticut. Her research interests include sentence processing, event comprehension, and bilingualism. Michael T. Putnam is Associate Professor of German and Linguistics at Penn State University. He has published widely on topics related to morpho-syntax, syntax, and the syntax–semantic interface in language attrition contexts. He has extensive research experience with moribund varieties of German (i.e., Sprachinseln) spoken throughout the world, and is interested in the role that cognitive ageing plays in language attrition in these speakers. He is broadly interested in how bi- and multilingualism impacts and enriches our understanding of the cognitive architecture underlying the language faculty. Claudia Maria Riehl is Professor for German Linguistics and German as a Foreign Language. She is chair of the Institute of German as a Foreign Language and director of the International Research Unit of Multilingualism (IFM) at the Ludwig-Maximilians-Universität

OUP CORRECTED PROOF – FINAL, 25/6/2019, SPi

 

xxv

München. Her research interests are sociolinguistic and cognitive aspects of multilingualism, language contact, minority languages and language policy, multiliteracy, and second language teaching. Professor Riehl is the author of many titles, among them the German-speaking introduction into language contact research, Sprachkontaktforschung. Eine Einführung (rd edn, Narr, ). Eleonora Rossi’s primary research interest focuses on language processing and bilingualism. She got her MA in Speech Pathology at the University of Padua (Italy) with a thesis on two case studies of bilingual aphasia. She followed this line of work with her PhD in Linguistics at the University of Groningen (The Netherlands), in which she investigated priming effects and language production in agrammatic aphasic speakers. Her postdoctoral work took her to Penn State (US) where she worked with Judith Kroll and Giuli Dussias on the neurocognition of bilingualism. Her recent work has utilized Event-Related Potentials (ERPs), functional and structural magnetic resonance imaging (MRI) data to unveil the neural bases of bilingual language control, code-switching, and first language attrition. Dr Rossi is currently Assistant Professor in the Linguistics Department at the University of Florida, and Research Associate at UC Riverside. Jason Rothman is Professor of Linguistics at UiT the Arctic University of Norway. He also has a fractional research appointment at the Universidad Nebrija in Madrid. He is the Executive Editor of the journal Linguistic Approaches to Bilingualism (LAB) as well as Editor of the book series ‘Studies in Bilingualism’ (SiBIL). Rothman’s research focuses on language acquisition and processing in children and adults in addition to the interface between domain general cognition and language. Some of his recent research has appeared in Bilingualism: Language and Cognition, Journal of Experimental Psychology: Learning, Memory and Cognition, Studies in Second Language Acquisition, Proceedings of the National Academia of Science (PNAS), Language, Cognition and Neuroscience, and Frontiers in Psychology. Liliana Sánchez is Professor of Spanish at Rutgers, The State University of New Jersey. She has published Bilingualism in the Spanish-speaking World with Jennifer Austin and Maria Blume (Cambridge University Press, ), The Morphology and Syntax of Topic and Focus: Minimalist Inquiries in the Quechua Periphery (John Benjamins, ), and QuechuaSpanish Bilingualism: Interference and Convergence in Functional Categories (John Benjamins, ) as well as articles in journals such as Linguistic Approaches to Bilingualism, Bilingualism: Language and Cognition, Lingua, International Journal of Bilingualism, Probus, Studies in Second Language Acquisition, among others. She has also co-edited the volumes Information Structure in Indigenous Languages of the Americas () and Romance Linguistics  (). She is a co-founder of the PhD Program in Bilingualism and Second Language Acquisition at Rutgers. Monika S. Schmid is Professor of Linguistics at the University of Essex, having previously held positions at the Vrije Universiteit Amsterdam and the Rijksuniversiteit Groningen. Her research focuses on various aspects of first language attrition and has been published in journals such as Bilingualism: Language and Cognition and Applied Psycholinguistics. She is the author of several books including Language Attrition (CUP, ) and First

OUP CORRECTED PROOF – FINAL, 25/6/2019, SPi

xxvi

 

Language Attrition, Use, and Maintenance: The Case of German Jews in Anglophone Countries (Benjamins, ). Elena Schmitt is a Professor of Applied Linguistics and Director of the TESOL/Bilingual Education Program at Southern Connecticut State University in New Haven, Connecticut. A native of Russia, she was educated at Moscow State Linguistics University and received her PhD at the University of South Carolina. Her teaching focuses on linguistic theory and its practical applications to educating English learners. Her personal bilingual experiences continue to ignite her research interests in the fields of childhood bilingualism, bilingual education, language attrition/loss, classroom interaction, teacher education, and technology-assisted teaching. Michael Sharwood Smith is Emeritus Professor at Heriot-Watt University, Edinburgh, Honorary Professorial Fellow at Edinburgh University and a founding editor of Second Language Research. His main interests are what links language representation, language processing, and language development. To this end, he is working with John Truscott on the Modular Cognition Framework to integrate diverse explanations of language acquisition, attrition, and processing within a general account of human cognition as a whole drawing on current thinking in different branches of cognitive science. His more recent books include Introduction to Language and Cognition () and, with Truscott, The Multilingual Mind (). Roumyana Slabakova is Professor and Chair of Applied Linguistics at the University of Southampton, UK, and a Research Professor II at UiT The Arctic University of Norway. Her research interest is in the second language acquisition of meaning, more specifically phrasal-semantic, discourse, and pragmatic meanings. Her monographs include Telicity in the Second Language (Benjamins, ) and Meaning in the Second Language (Mouton de Gruyter, ). She co-edits the journal Second Language Research (SAGE) and is a founding editor of Linguistic Approaches to Bilingualism (Benjamins). Her textbook Second Language Acquisition was published by Oxford University Press in . Antonella Sorace is Professor of Developmental Linguistics at the University of Edinburgh. She has published widely on bilingualism and language learning over the lifespan, language typology, and gradience in natural language, bringing together methods from linguistics, experimental psychology, and cognitive science. She is also committed to public engagement and is the founding director of the research and information centre Bilingualism Matters. Karsten Steinhauer is Professor of Cognitive Neuroscience and Director of the Neurocognition of Language Lab at McGill University in Montreal. He and his students investigate the brain’s electrical activity (EEG) to uncover how the human mind/brain processes language in real-time, and what changes when we learn a second language or lose our first one. His original research has been published in high-impact journals such as Nature Neuroscience, PNAS, and the Journal of Cognitive Neuroscience, and has been supported by

OUP CORRECTED PROOF – FINAL, 25/6/2019, SPi

 

xxvii

all three federal agencies in Canada, by the Canada Research Chair programme, and by many prestigious studentships (e.g., Tomlinson, Vanier, PBEEE, and Fulbright). David Stringer, Associate Professor of Second Language Studies Indiana University, researches the acquisition of syntax and lexical semantics. He is interested in universal aspects of word meaning that play a role in grammar across languages. Other areas of research interest include World Englishes (especially in South Asia and Sub-Saharan Africa), language attrition (in joint work with Kathleen Bardovi-Harlig), and biocultural diversity (linking language revitalization in indigenous cultures to the conservation of ecosystems). His interests extend to all formal aspects of second language acquisition and multilingualism, and he works with students on a wide range of projects. Hideyuki Taura is Professor of Applied Linguistics at the Graduate School of Language Education and Information Science, Ritsumeikan University, Japan. He obtained his MA and PhD from Macquarie University, Sydney. His research interests include () bilingual language acquisition and attrition from both linguistic and neuro-linguistic perspectives, () early bilingual cognitive development, () bilingual narrative development, () EFL teaching in Japan, () novice English teachers’ professional development in China, and () translanguage teaching for heritage language education. Sharon Unsworth is Associate Professor in the Department of Linguistics, Department of English Language and Culture and the Centre for Language Studies at Radboud University in Nijmegen. Her research interests include child bilingualism and second language acquisition, both in naturalistic and instructed contexts, and she has authored articles in internationally renowned journals such as Applied Psycholinguistics, Bilingualism: Language and Cognition, and Language Acquisition, as well as numerous book chapters. Jorge R. Valdés Kroff is Assistant Professor of Spanish and Linguistics at the University of Florida. He is the director of the Bilingual Sentence Processing Lab which uses behavioural and eye-tracking methods to investigate how bilinguals adapt their parsing preferences to better anticipate upcoming code-switches. His recent work focuses on testing whether bilinguals tap into production asymmetries to guide comprehension and recruit greater engagement of cognitive control to rapidly integrate upcoming code-switches. Álvaro Villegas is Assistant Professor in the Department of Modern Languages and Literatures at the University of Central Florida, where he teaches Spanish Linguistics. He earned his PhD in Spanish Linguistics and Language Sciences at The Pennsylvania State University in . His research focuses on Psycholinguistics and Bilingualism, with a special interest in the production and comprehension of heritage speakers. He also works with bilingual communities in Orlando to inform parents and educators about the benefits of bilingualism and to promote bilingual education. Gülsen Yılmaz is currently a Post-Doc at the Humboldt University of Berlin, working with Prof. Artemis Alexiadou and her research group. Her current research focuses on the variability of the article system in heritage language grammar. She holds a PhD in Linguistics from Groningen University, the Netherlands. She did her MA in foreign

OUP CORRECTED PROOF – FINAL, 25/6/2019, SPi

xxviii

 

language education. Her research includes theoretical and experimental methods studying bilingual development with a focus on psycholinguistics aspects of language acquisition and attrition. She taught language attrition, contact linguistics and experimental methods at undergraduate and graduate levels. Margarita Zeitlin is a PhD candidate in the Department of Psychology at the University of Washington. Her research interests include language learning and the relationship between language and other cognitive domains.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

  ......................................................................................................................

 ......................................................................................................................

 .    ö

I has become a time-honoured tradition for overviews and introductory chapters on language attrition to begin by pointing out the recency of the field. Most commonly this is done by referring to the time period elapsed since the publication of Lambert & Freed (eds.) in . We considered adhering to this tradition, or alternatively referring to the (much more impressive) eighty years since Einar Haugen published his seminal essay ‘Language and Immigration’ in , which not only—to the best of our knowledge—coined the term ‘attrition’ but also anticipated many future findings (for more details on this essay, see Schmid & de Leeuw, Chapter , this volume). We eventually decided, however, to use a much more personal milestone: as this volume is going into print we are coming up to the twentieth anniversary of a cold November afternoon in the small Dutch town of Veldhoven where the Third International Conference on Maintenance and Loss of Minority Languages () took place. At this conference two PhD students were presenting their first ever posters summarizing the initial findings of their respective studies on German L attrition. This was how the two authors of this introduction first met. We got to talking and it soon became evident that, while the scope, methodology, theoretical framework, and results of our two research projects were entirely different, we shared the same concerns regarding the field of attrition studies, which not long afterwards we summed up as follows: [ . . . ] research is still very much divided on the question of whether any such thing as adult L attrition exists in the first place. Such differences are likely to be the outcome of inconsistencies in the methodology used in the collection and interpretation of the data—and not necessarily of actual differences in the linguistic repertoire of the speakers analysed by these different studies [ . . . ]. It has been pointed out that there is great variance in the amount of attrition that individual studies have found, but it has not, so far, been investigated in what way these differences are linked to the methodology of collecting and analysing data. [We hope to] reconcile some of these differences, and develop a sound methodological framework for further studies in language attrition. For this, it appears crucial to reconcile the different disciplines involved, to take advantage of the strength of the linguistic, sociolinguistic and psycholinguistic approaches and to join them in an interdisciplinary frame able to account for all types of variation which render comparisons between different studies difficult or impossible. (Call for Papers for the First International Conference on First Language Attrition, which was held at the VU Amsterdam in 2002)

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



  ö

This was the research agenda that we felt should guide language attrition studies into its next phase. We knew that it was an ambitious one. What we did not know at the time was to what extent we were fortunate to come to the topic at the right time and place, with circumstances as well as the personal generosity of many individuals fortuitously conspiring to support our efforts. In the first instance, the networking aspect of our agenda benefitted enormously from the fact that, at that time, it was still possible to bid for relatively small sums of money to support stand-alone scientific events. Today, in the rare cases where conference and workshop funding opportunities still exist, one must usually come up with series of linked events, preferably to be held in different countries—an endeavour that is simply not feasible for early career researchers with limited professional networks and experience. These funds—generously awarded by organizations such as the Netherlands Organisation for Scientific Research (NWO), the Dutch Royal Academy of Sciences, as well as our own institutions (the VU University Amsterdam, the University of Toulouse, and the University of Groningen) allowed us, between  and , to not only organize two international conferences and five meetings of the European Graduate Network on Language Attrition but to provide financial support to a very substantial number of postgraduate students who would not otherwise have been able to attend these events. With the help of a Seminar Series grant from the British Economic and Social Sciences Research Council (ESRC), this was followed between  and  by a series of eight events focusing on interdisciplinary aspects of language attrition research, including the Third International Conference on Language Attrition (see https://languageattrition.org/attrition_seminars/). The focus of the initial series of meetings was on the development of a common methodology for language attrition studies. What we suggested was that each study, no matter what its eventual aim, theoretical framework, or specific linguistic interest, should adopt a common baseline methodology with respect to the collection of background features and language proficiency data, in order to make it possible to benchmark findings from different languages and different settings against each other. The tools we developed comprised a range of questionnaires on personal background, language use and attitudes, and self-evaluated proficiency, as well as the use of a C-Test, Verbal Fluency tasks, and a short passage of semi-structured discourse elicited through a film retelling task. The instruments, along with instructions on how to use, code, and analyse them, were made available in a number of languages on languageattrition.org, and are described in detail in Schmid (). They have since formed the basis for a large number of studies (see also Schmid & Cherciov, Chapter , this volume). While the exchange of ideas, methods, and results that took place at these events and the personal links and contacts thus came to form the backbone of language attrition research over the next years, the personal friendships and mentor/mentee relationships that developed from them became its heart. Kees de Bot, who himself has undoubtedly been the single most important figure in not only creating and shaping the field of L attrition research but leading it into the twenty-first century, remarks on this in the concluding chapter to this volume. Where previously attrition research had been characterised by the solitary existence of the PhD-student, spending a large part of her time doing fieldwork and, typically, having little opportunity and less money to go to conferences and make contacts in the international scientific community. (Köpke & Schmid, , p. )

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi





PhD students, postdocs, and young researchers like ourselves began to feel part of the wider community of applied linguists, with the bonus of a distinctive identity as an attrition researcher. This is how not only the co-editors of the present volume but also many of its contributors began their career, and it is only partly due to the support from the abovenamed funding bodies, and in much greater measure to the generosity, kind-heartedness, openness, interest, and sound advice of established senior researchers such as Kees and too many others to name here. The second attendant favourable circumstance for our ambitions to advance language attrition from its niche status to an integral part of research on bilingual development were technological and practical innovations which over the past two decades have made it possible and feasible to conduct investigations on a previously unimaginable scale and level. Our efforts began around the time when the analysis of spoken data (an indispensable cornerstone to attrition research) made the quantum leap from painstaking (and, as we know from personal experience, unspeakably tedious) hand-coding to freeware tools such as the ones developed in the CHILDES-project or ELAN. The exponential increase in size and affordability of computer memory furthermore made it feasible to work with much larger audio, or even video, corpora, not to mention the availability of fully tagged corpora often comprising millions of words of the target languages, allowing much more precise benchmarking. Experimental methods, too, developed at breathtaking speed—it was not so long ago that even reaction-time software was exorbitantly expensive, while today many linguistics departments have their own eyetrackers and/or EEG scanners. Second language acquisition research had only just begun to appropriate all of these methodologies at the time that the authors of this chapter first met, so we are delighted that attrition research has since begun to avail itself of all of them, as many of the chapters in the present volume demonstrate. Alongside such methodological and practical issues, we felt that another key area in need of attention was the theoretical underpinnings of attrition research. This became the focus of the second attrition conference (held, again at the VU Amsterdam, in ), at which we attempted to bring together more theoretically-deductive and more empirical-inductive approaches to language attrition, and also to interest researchers from neighbouring areas of linguistics—such as contact linguistics or ethnolinguistic vitality theory—in applying their insights and theories to language attrition data. These efforts are ongoing, and we are delighted that some of the theoretical chapters in the first section of the current volume (such as the ones by MacWhinney, Putnam et al., and Slabakova, Chapters , , and , respectively) represent novel theoretical takes on language attrition. All of these developments have helped language attrition research to ‘reach its adulthood’ (de Bot, Concluding remarks, this volume). The increase in empirical work has demonstrated beyond any doubt that attrition is not, as had originally been assumed, an ‘extreme’ form of development, experienced by a small minority of bilinguals with decades of no or very little exposure, but a form of language development experienced from the early stages of second language acquisition and thus, in all likelihood, common to all bilinguals (Schmid & Köpke, a). The present Handbook is testimony to the fact that attrition is now very much part of the mainstream of applied linguistics. We mentioned above a second series of seminars (financed by the ESRC), in which the focus turned towards interdisciplinarity. We would like to suggest that this is indeed what should be on the agenda next for the field of language attrition: having consolidated itself as

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



  ö

an area of investigation with a firm sense of identity, a solid theoretical and methodological foundation, and a wide base of empirical evidence, we feel that it is time to turn our attention again to the fact that language attrition is only ever one side of the coin of bilingual development. Limiting the perspective of investigations to what happens in the native language of bilinguals ignores at least half of the process. The same is true, we feel, for investigations of second language acquisition which, in turn, ignore language attrition. Our agenda, then, for the next phase of research in bilingual development should be to argue most strongly that, whether the object of an investigation is L acquisition or L attrition, ignoring the other language is to bias the results and limit the potential for understanding what actually happens when two languages interact within the same mind.

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

  .............................................................................................................

THEORETICAL IMPLICATIONS OF LANGUAGE ATTRITION ............................................................................................................. Edited by

 .    ö

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

  ......................................................................................................................

      ......................................................................................................................

   

T chapter explores ways in which the Competition Model can illuminate our understanding of language attrition. We begin by formulating some of the key puzzles that arise from empirical studies of language attrition. After this initial problem description, we will consider how the cognitive and social processes included in the Competition Model can help in the formulation of solutions to these various puzzles.

. F 

..................................................................................................................................

.. Puzzle #1: permastore It is easy to imagine that language attrition arises simply from disuse of a speaker’s first language (L). The idea is that, for any given ability, you must ‘use it or lose it’ (Pinker, , p. ). Experimental psychology has provided abundant evidence indicating that unused memory traces fade over time (Ebbinghaus, ). In some accounts, forgetting arises through the weakening of a memory trace by changes at the level of the synapse. In other accounts, forgetting arises through interference of old memories with new memories. In either scenario, failure to repeatedly access and use the old memory is viewed as the cause of forgetting and problems with retrieval (Paradis, ). Problems of this type also arise in people suffering from dementia or anomia. However, there is also evidence that adults can maintain linguistic representations for decades without active use. For example, Bahrick (b) and Bahrick & Phelps () demonstrated that vocabulary learned in college Spanish classes was maintained after as much as fifty years without significant maintenance or reuse. This type of ‘permastore’ corresponds to what we find in people who continued using L through adolescence, but then ceased its use during adulthood. These speakers retain their ability to use their

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi





L despite years of disuse (Schmid & Dusseldorp, ). We can refer to these speakers as ‘maintainers’. We can contrast people in this group with ‘attriters’ who show a greater loss of L forms. Within this group, we can contrast ‘strong attriters’ who have undergone the greatest amount of attrition with ‘weak attriters’ who show less attrition, but still not full maintenance. What is puzzling for standard psychological theories of forgetting is that even the strongest attriters maintain a large amount of their L (Steinkrauss & Schmid, ). We can also distinguish a third group of adult second language (L) learners that maintains use of L in various aspects of daily life. Because this group suffers minimally from attrition, we can refer to them as ‘late bilinguals’. There are no sharp cutoffs between these three groups; rather they represent various points along a continuum of degree of language attrition. Understanding why a speaker ends up at a given point along this continuum constitutes a further aspect of this puzzle.

.. Puzzle #2: variation across levels Both attriters and maintainers show the impact of an increasingly stronger L on a somewhat weakened L. These effects can reveal themselves through the inappropriate use of L words, expressions, and structures while speaking an L. In such cases, we can say that the L and the L are competing for activation and that the L form is used because the L form is weaker. There are at least five ways in which this competition can play out. . First, intrusion of an L form when speaking L may arise from problems retrieving the L form. If the activation of the L form is insufficient, the L form will win out in the horse race for activation and production (Ratcliff, ). This effect underscores the extent to which L remains active, if somewhat repressed, during production of L. . In a related case, the weaker activation of the L form can lead to a total failure to recall the required form. In such cases, we see omissions, hesitations, and other disfluencies without actual intrusion of an L form. . When L and L share words or constructions, it can happen that both sources of activation support the use of the same form (Tokowicz & MacWhinney, ). In such cases, intrusions may be scarcely noticed. This is particularly likely in closelyrelated languages, such as Norwegian and Danish. . In other cases, it may be that the L form is slightly more appropriate in the given context than the L form. This type of transfer from L resembles what we see in cases of lexical insertion during code-switching (Poplack, ; Myers-Scotton, ) in which a speaker will insert an L word into an L matrix sentence because it seems like the mot juste or most appropriate form in a particular context. It would be a mistake to think of this use of an L form as indicative of a weakened L alternative. . If a speaker’s L has marked dialect features, the contrast between those features and the features of the standard version of L can further complicate the competition between L variants and L. What may appear to be an error in L may simply be a reliance on a non-standard form. In some cases, the form in question may derive from an earlier standard L form that has now fallen out of use. Such competitions between alternative forms can lead to delays or errors in retrieval.

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

     



What complicates matters further is that these alternative forms of competition and transfer work in different ways across the language levels of output phonology, input phonology, lexicon, collocation, syntax, and pragmatics. For output phonology, Schmid (a) has shown that maintainers are perceived as having the accent of native speakers, although strong attriters may eventually lose their native accent. We know that second language learners show a type of backward transfer in terms of shifts in the details of their control of phonetic features, such as voice onset time (VOT) for stop consonants (Major, ) during the process of L learning. The fact that this happens so quickly during early stages of L learning is surprising, given the relative resistance of L phonology to loss in language attrition. However, the variations that L learning forces on L phonetics are not so extreme as to push L sounds out of the normal range; rather, they represent deviations within the overall range of potential L sounds. Basic syntactic patterns also seem resistant to loss during attrition. However, attriters can begin to reduce their use of marked orders and formats, avoiding some of the more complex L forms they would have used in earlier years. These adjustments amount to minor losses of L forms and fluency (Schmid & Beers Fägersten, ). Complete abandonment of L syntax under the pressure of L learning seems to be minimal, even in strong attriters. The relative resistance of phonological and syntactic structures to attrition contrasts with greater attrition in vocabulary, idioms, and collocations. There are three theoretical approaches to explaining this contrast. A basic usage-based or frequentist account predicts that the least frequent L words should be the ones most subject to L intrusion. A further consequence of this would be that the first forms that are lost are also the ones that were learned last during L acquisition (Wixted, ). However, in a careful study of frequency effects on these intrusions, Schmid (a) finds only marginal effects for lexical or collocational frequency. A second account, based on the theory of between-language competition, suggests that intrusions arise from forces such as the appropriateness of the L form, contextual support for code-switching, and structural similarities between L and L. A third approach focuses on the contrast between procedural and declarative memory systems (Ullman, c). Here, the idea is that entrenchment operates most strongly for systems such as output phonology and word order patterning that have undergone proceduralization through basal ganglion and striate functioning (Stocco et al., ), whereas lexical and collocational patterns that are more responsive to hippocampal consolidation are more susceptible to intrusion. Figuring out which of these three accounts for the contrasting effects of linguistic levels is operative in a given case is a major challenge for studies of language attrition.

.. Puzzle #3: childhood forgetting Of the four puzzles related to language attrition, the third is the most perplexing. This puzzle arises from the observation that children who are adopted into a new family in a new country before the age of  show a nearly complete loss of the ability to use their L (Pallier et al., ; Ventureyra et al., ). A basic psycholinguistic characterization of the process of entrenchment would predict that this should not happen. If a child is younger than age  when coming to a new country, one could argue that their language had not yet fully consolidated and was therefore vulnerable to complete language loss. However, by age ,

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi





a child will have spent six years making continual daily use of the L in fully appropriate social contexts. Based on Bahrick’s (b) findings regarding the entry of memories into a ‘permastore’, these children should have thoroughly consolidated or ‘overlearned’ their L by the age of . The type of natural learning that children experience is particularly well suited to the development of long-term consolidation (Pashleret al., ; Pavlik & Anderson, ). Moreover, studies of monolingual first language learners demonstrate entrenchment well before the age of . For example, Brooks & Tomasello () showed strong entrenchment effects for item-based patterns in causative constructions in English by age . These considerations indicate that late adoptees should retain a significant amount of L knowledge in permastore. At the very least, this knowledge should facilitate rapid relearning or recognition of L forms (MacLeod, ). However, studies of adoptee’s L relearning during adolescence have provided only weak evidence of any retention of L sensitivities, as revealed through somewhat more accurate relearning of some phonetic contrasts (Au et al., ; Oh, Au, & Jun, ; Oh, Jun, et al., ; Park, ; Singh et al., ). However, even these results may be due to motivational factors, rather than stored memories. There are two solutions that have been offered to this puzzle of childhood forgetting. The most popular solution is the Critical Period Hypothesis (CPH), which attributes childhood L loss to the expiration of a critical period for language acquisition. However, that theory was intended to account for language learning, not language loss. Nonetheless, it might be possible to relate early language loss to neurodevelopmental processes that impact both the ability to learn and the ability to lose a language. For example, declines in susceptibility to loss could be related to increases in the myelination of cortical layers. On this account, a high level of neuronal plasticity facilitates new learning, but also permits quick loss of new patterns. The theory of Universal Grammar offers a related viewpoint on this issue, particularly as elaborated in terms of ‘biolinguistics’ (Boeckx & Grohmann, ). On this account, L learning involves the setting of a series of parameters. Problems with resetting these parameters are supposed to be at the root of difficulties adult learners have acquiring L. In accord with the CPH, parameter setting by children is labile and incomplete, allowing for reversal and modification during childhood, but not afterwards. The actual neuronal implementation of this version of the CPH account could conceivably rely on the same neuronal mechanisms used to support other accounts based on entrenchment and plasticity. A second approach focuses on changes in the social and emotional supports needed for language maintenance. Pulvermüller & Schumann () argue that first language learning links in more directly to emotional roots than does second language learning. In addition, we need to consider the extent to which adoption and immigration can break off links between language and the original social and experiential contexts to which linguistic forms refer. Moreover, childhood forgetting may arise from the involvement of additional neuronal circuitry supporting changes in attention and motivation. We will consider these two accounts—critical periods versus abrupt contextual change—in further detail later.

.. Puzzle #4: catastrophic interference The fourth and final puzzle derives as a further complication from the third. This puzzle involves the fact that parallel distributed processing (PDP) connectionist models of second language acquisition demonstrate ‘catastrophic interference’ or complete loss of L at the

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

     



beginning of learning of a second language. This occurs because L knowledge overwrites the connection weights established during L learning (Zevin, ). The fact that adult maintainers demonstrate minimal language loss argues strongly against models predicting such catastrophic loss. At the same time, developers of these models could point to catastrophic loss in children as evidence in support of their account. There are also other neural network models, such as DevLex (Li et al., ) that avoid the problem of catastrophic interference through reliance on learning through self-organizing feature maps. However, those models then have no ready account of the phenomena of language loss by children.

. T C M

.................................................................................................................................. Taken together, these four puzzles indicate that no single-process model will be sufficient to account for the observed phenomena in language attrition. A model that combines the mechanism of entrenchment with a model of transfer would come close to explaining some of the puzzles, particularly if it incorporates a theory of differences between linguistic levels. However, a full account of these phenomena will require still further detail regarding social and motivational inputs, contextual changes, internal processing mechanisms for consolidation, and neurological foundations of plasticity and reorganization. The Competition Model (MacWhinney, b, ) is designed to provide an integrated approach to these issues, as well as to additional issues in online language processing, first and second language learning, bilingualism, code-switching, and language loss in aphasia. The model holds that language, as a fundamentally biological system, emerges from processes of competition. Like other emergent biological systems, language structure derives from properties and interactions unique to hierarchically organized levels of processing that interoperate across differential timescales. Implications of the emergentist constructs of competition, levels, and timescales are further explored in MacWhinney (a). This view of language structure as emerging from competing motivations (MacWhinney et al., ) is based on mechanisms that have been documented by experimental psychology, cognitive linguistics, sociolinguistics, typology, neuroscience, conversation analysis, and social research. However, we do not yet understand how these competing component processes fit together cooperatively into a functional system. This means that use of the model to account for patterns of language attrition requires extensive consideration of the ways in which competing motivations and forces are integrated. For the study of language attrition, the most important aspects of Competition Model analysis are those relating to the interplay between risk factors and protective factors. Risk factors are forces that operate to reduce smooth language learning and functioning. Support factors are forces that work to promote fluent language functioning. The model specifies the risk factors of entrenchment, transfer, over-analysis, and isolation, and the support factors of resonance, decoupling, chunking, and participation. The processes of over-analysis and chunking apply mostly to the first stages of language acquisition, so we will not consider them here. Let us consider how the other six factors (entrenchment, transfer, isolation, participation, decoupling, and resonance) operate in the case of language attrition.

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi





.. Entrenchment The basic unit of Competition Model analysis is the cue. Cues are relations between forms and functions (de Saussure /). Cues can involve units on any level of linguistic structure—phonology, lexicon, or grammar. If a cue correctly predicts a function during comprehension or selects out the correct expression during production, then it is high in reliability. Competition Model studies (MacWhinney, ) have shown that, during early stages of language learning, cue frequency or availability is more important than reliability. However, over time, cue reliability becomes the major determinant of cue strength. Cue availability and reliability are estimated through corpus-based counts (McDonald, ; Ellis et al., ), whereas cue strength is estimated through experimental studies that place different cues into competition (MacWhinney & Bates, ). A core finding of these studies is that cue reliability closely predicts cue strength. Like other models that are implemented in the neural network framework, the Competition Model assumes that cues are in continual competition and that this competition is determined through cue strength. Entrenchment arises from the continual use of a cue. When a cue is reliable, strength is increased; when it is unreliable, strength is decreased (McDonald & Heilenman, ). Entrenchment from this process of cue strengthening must play a central role in language maintenance and attrition. However, the four puzzles we have discussed suggest that entrenchment cannot work alone, and that its operation may be quite complex. Recent work in neurobiology has begun to uncover some of the mysteries of entrenchment in sensory areas. Examining the neuronal basis of amblyopia (poor visual acuity) arising from a lazy eye, researchers (Takesian & Hensch, ) have demonstrated an interplay between five neurobiological forces that lead to the entrenchment of a cortical region and those that can reopen that area for new learning. . Initial brakes. Before the onset of entrenchment, the polysialic acid-neural cell adhesion molecule (PSA-NCAM) prevents the opening of a cortical area to learning. . Triggers. A sensitive period opens when molecular triggers (such as orthodenticle homeobox 2, BDNF)—in response to sensory input—promote parvalbumin (PV) cell maturation, turning on GABA (gamma-aminobutyric acid) circuit functioning. GABA circuit maturation may operate by suppressing responses to intrinsic activity, allowing a focus on extrinsic input (Sale et al. ; Van Versendaal et al., ; Toyoizumi et al., ). . Learning. Learning then involves physical pruning and homeostatic regrowth of synapses by tPA, TNFα, and protein synthesis. . Terminal brakes. A consolidated or entrenched state is maintained by the production of functional molecular brakes, such as Lynx1 and NgR1, and physical brakes such as PNN and myelin, which may limit plasticity by preventing further structural changes. These brakes rely on the same processes that created changes during the plastic period, essentially preserving these changes. . Reopeners. The reopening of cortical areas to new inputs can be achieved by lifting the brakes through epigenetic manipulation (HDAC inhibitors) or heightened attention (neuromodulators such as 5-HT, Ach, NE). The operation of this system of excitation and inhibition across sensory levels varies based on the timing of maturation of the PV cells that trigger the opening of the learning

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

     



period. The variability in these distributions of PV cells could help us understand variations in entrenchment and maintenance across language levels. Because entrenchment can be reversed through attentional processes, we should think of this system as one supporting variable entrenchment, rather than an irreversible critical period. We should also note that research to date has focused on the operation of these brakes and reopeners in sensory systems. How they operate in motor areas and higher cortical regions is not yet clear, although the neurobiological components are common across areas.

.. Transfer Entrenchment constitutes a risk factor for L learning, because it restricts the availability of cortical areas for new learning, forcing the learner to engage in transfer of L forms to L. At the same time, entrenchment serves as a protective factor promoting L maintenance, as opposed to L loss. However, once L has become established as the primary means of daily communication, transfer from an increasingly stronger L to L begins to operate against the forces of L entrenchment. Because of the emphasis it places on interactive activation, the Competition Model holds that, for initial L learning, ‘everything that can transfer will’. In the domain of phonology, whenever an L phoneme can be mapped onto an L sound, learners will attempt to transfer that form and slowly correct for the inevitable mismatch. For lexicon, the meanings of L words will transfer, although the sounds typically will not. In morphology, transfer is seldom possible except between closely-related languages. In syntax, transfer will impact collocations and even basic word order. In each of these areas, what will transfer from L is the unmarked pattern. Studies of transfer in L learning largely support these predictions. There is evidence that they also apply to language attrition, as an increasingly stronger L begins to transfer patterns to L. For example, Tóth () compiled an inventory of transfer patterns in Hungarian and German L speakers in the San Francisco Bay Area. He could identify an L English source for each erroneous form he observed. The patterns of language transfer found in such studies also closely resemble those reported in cases of simultaneous childhood bilingualism (Döpke, , ; Yip & Matthews, ). Looking in greater detail at areas of particular susceptibility to transfer, some studies (Seliger, ; Ribbert & Kuiken, ) have suggested that transfer is particularly strong in areas where uses of an L form constitute a subset of uses of a similar, but more general, L form. This finding is in accord with the Competition Model expectation that the degree of interference from L to L will depend on the closeness of the mapping, as well as the markedness of forms in L and L (Eckman, , ). Because of its emphasis on competition, interaction, and variation, the Competition Model also predicts back transfer from L to L, as well as partial merger of grammar in certain areas. Such effects have been observed for syntactic structure (Liu et al., ; Dussias, ; Dussias & Sagarra, ), lexicon (Malt et al., ), and phonology (Major, ; Mennen, ), sometimes arising after only a few weeks of exposure to L (Chang, ). Thus, when studying speakers of two or more languages, we must always keep in mind the possibility that there will be dynamic interactions between the languages (Cook, a). However, such interactions do not, in themselves, constitute full language attrition.

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi





.. Participation, isolation, and identity For successful L learning, the single greatest protective factor is the ability to participate actively in L social groups without having to rely on switches to L. Conversely, the greatest risk factor for L learners is the threat of isolation from L groups. De Bot & Clyne (, ) have described situations in which immigrants give up on attempting to integrate with the L community and revert to reliance on L. However, there are also cases in which immigrants integrate so thoroughly with the L community that they cease daily use of L. This is the situation that gives rise to Puzzle # regarding the maintenance of an unused L in permastore. Schmid () conducted an examination of attrition effects in Jews who escaped to Anglophone countries from Germany between early adolescence and young adulthood at various times during the Fascist pogroms. If they escaped before the onset of the racial discrimination laws, these people tended to maintain a high level of functional use of German, along with a basic acceptance of their German heritage. However, people who escaped after the November  pogrom and the beginnings of deportations tended to reject their German language and heritage. Schmid found that the single best predictor of degree of language attrition in this group was the time of leaving Germany. Although this is an extreme case, it still serves to reveal the importance of fundamental cultural identification as either a risk or protective factor for language maintenance. There are two ways in which cultural identity can play out on a daily level. For those who have decided to reject their L identity, there can be active suppression of L words and thoughts. This type of suppression of L can be accompanied by enthusiastic use of new L forms both in conversation and in inner speech, formulation, and planning. The other way in which cultural identity can play out in everyday life is through the organization of activities in L and L. Schmid (a) reports that ongoing use of L in a professional context is a good predictor of a minimal level of L attrition. For example, the need to use L for translation or writing requires ongoing attention to correct use of L forms. Conversely, if L usage is confined to less demanding environments such as the home or if it is being used in code-switched interactions, then relatively more attrition will occur. The link between identity and language maintenance may also help us understand Puzzle # regarding children’s forgetting of L. For many adoptees, life in the L setting may have been relatively unstable and harsh. Moreover, the new L environment provides little support for or even recognition of L. Because of this, adoptees may have also decided to actively repress their L. Burling () presents a rather poignant case of this type. During a two-year-long field trip in the hills of Myanmar, his son acquired Garo and English simultaneously. On the airplane flight home to the States, the -year-old boy attempted to speak to the flight attendant in Garo and she only replied in English. After that, he never used a word of Garo again.

.. Resonance In second language learning, the effects of L entrenchment can be countered through the process of resonance. This process is supported by resonant connections between the hippocampus and cortical areas, leading to the consolidation of memories and linkages

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

     



in the cortex. The hippocampus has connections to all cortical areas, each arranged in distinct subareas. When a new experience or episode is encountered, the hippocampus uses these connections to maintain an ongoing interaction or conversation with these cortical areas. This is not a one-shot activation of the cortex, but a process of ongoing maintenance of cortical activity (Nadel et al., ) through reentrant connections (Wittenberg et al., ). These patterns of activation go beyond mere consolidation of individual episodes. Relying on additional control from medial prefrontal areas, the consolidation process can link new experiences with old concepts into increasingly rich conceptual networks (Schlichting et al., ; Zeithamova et al., ). This method of building up new relational networks from new experiences can be particularly important for consolidation of language structures such as words and collocations, as specified in the Competition Model theory of learning from item-based patterns (MacWhinney, ). For people who are trying to maintain their L in an L environment, resonance can rely both on continued L input from conversation and reading, as well as internal use of L for formulation and planning. Resonance can operate both passively and actively. Its passive operation can be detected during rest (Karni & Sagi, ) and sleep (Lindsay & Gaskell, ). Its active function arises when learners use orthography to activate phonology (Share, ) or when they derive word meanings from analysis of derivational and inflectional patterns (Presson et al., ). There is also a second form of consolidation that is important for motor and sequential procedures. This system utilizes frontal-striatal and basal ganglion structures (thalamus, striatum, cingulate, precuneus, substantia nigra, globus pallidus) in coordination with the cerebellum to coordinate and proceduralize smoothly functioning action sequences for speech and syntax (Graybiel, ; Hinaut & Dominey, ). Although attriters may have developed these patterns in earlier years, they may need to reactivate these processes to maintain fluency. These two systems for resonant consolidation allow the language learner to avoid the potential problem of catastrophic interference. Although the L and L may share some phonological and semantic resources, the connections between linguistic elements are structured to maintain a separate pattern of connectivity, even if they reside in neighbouring areas of the cortex. The role of resonance in language learning helps us understand Puzzle # and the problems faced by international adoptees. During their first years of language learning, all children work intensely on the learning of systems for audition and motor planning. These systems rely on statistical learning (Thiessen & Erickson, ), rather than resonance, for consolidation. These are the systems that show the clearest entrenchment during L learning. On the other hand, resonance is centrally involved in the learning of the higher linguistic levels of lexicon, morphology, collocation, syntax, clause combination, narrative, dialogue, and discourse. Learning on these levels requires support from resonant connections based on episodic encounters with objects, events, and situations. For the youngest learners, these connections depend heavily on direct support from specific situations, people, and objects. Later, as children start to rely increasingly on inner speech, language forms resonate more and more with one another independent of the environmental context. For example, - and -year-olds can recite only a few short songs, but when they are a few years older they will repeat much longer songs with ease.

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi



 

If a child is removed from the original language context before being able to make extensive internal resonant connections, their L will be missing the crucial match to the situational context that it needs for support. The result of this rapid and forced environmental change is a type of linguistic amnesia that impacts systems depending on resonance more than systems depending on statistical learning. Resonance may also be involved in the phenomenon of permastore. Bahrick (b) emphasized the fact that his subjects made minimal use of Spanish after the end of their courses in college. However, because these forms were learned through internal resonant processes initially, rather than through direct contact with environmental experiences, there may have been enough ongoing resonant links between these forms and other L forms to show evidence of retention even decades later.

.. Decoupling Successful second language learning depends on the decoupling of L from its dependence on L (Kroll et al., ). To the degree that a speaker can use resonance to construct each language as a (partially) separately activated neural circuit, the effects of transfer can be minimized. To achieve decoupling, an L learner must combine inhibition of L with resonant activation of L. In addition to the underlying cognitive processing involved in resonance, learners and maintainers may need to establish social and attitudinal methods for decoupling their two languages. If use of L can be localized to contexts that require minimal competition with L, there can be greater decoupling and maintenance of the two systems. At this point, the second language learner is functioning in a more truly bilingual fashion. The fact that lexical semantics and idiomatic expressions are particularly subject to intrusion from L can be related to the fact that L learning relies so heavily on L resources in this area. In effect, the two languages derive their meanings from similar experiences in the world. However, even here, careful use of resonance and decoupling can pull apart forms based on fine-grained distinctions. Phonological processes must also involve shared resources between L and L. However, in this case, L has experienced earlier consolidation through the fronto-striatal proceduralization system and fine-tuning involves the development of recombinations of these gestures to support L.

. C

.................................................................................................................................. Language is a dynamic, emergent system (Van Geert & Verspoor, ), grounded on complex inputs from biology and society. As such, we should not be surprised that theories of language acquisition and loss must refer to a wide range of processes to understand language patterning across individuals, groups, language levels, and language change. To solve the four puzzles involved in language attrition, we need to refer to various combinations of these processes. The conclusions we reach here are very close to those developed in Steinkrauss & Schmid ().

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

     



Entrenchment is certainly involved in the initial consolidation of L. However, as we now know, entrenchment is not irreversible. If life events force a person to abandon their L, they can avoid use of that language, emphasize decoupling and resonance in L, and actively suppress internal use of L. This type of abandonment of L is most easily achieved by young learners of L. For them, higher linguistic levels are particularly vulnerable to the loss of specific contextual supports for the L world. Speakers who do not seek to abandon their L will still need to deal with the pressures of transfer from an increasingly robust L. The exact nature of these transfer phenomena depends on the strength and accessibility of the L pattern, as well as the nature of the match between L and L. Because learners can make use of resonant connections to decouple L from L, they can avoid the problem of catastrophic interference. In effect, each language becomes a separate functional circuit, while still sharing common resources at the level of articulatory output and semantic reference. Because it is difficult to fully decouple languages at these levels, they remain susceptible to at least some transfer from L or L–L merger. By comparing the processes determining the shape of language attrition with those involved in L and L acquisition, as well as language loss in aphasia and dementia, we can gain an increasingly deep understanding of the emergent structure of human language.

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

  ......................................................................................................................

       ......................................................................................................................

 . ,  -,   ´ 

. I

.................................................................................................................................. A bilingual acquisition phenomena, attrition, defined as the loss of access to linguistic features/structures previously acquired (Schmid, a), is one that forces theoretical linguists to reconsider the nature of the acquisition of atomic units of language and the way in which they are assembled and accessed throughout a bilingual individual’s lifespan (Paradis, ). In this chapter we extend to attrition contexts a particular attempt to understand the processes that may lead to syntactic restructuring previously proposed to account for second language acquisition and for heritage language development: the feature reassembly hypothesis (hereafter, FRH; Lardiere, , , , ). The FRH owes its origins to the initial theoretical developments in the Minimalist Program (MP; Chomsky,  et seq.), where ‘a linguistic expression (π, λ) of L satisfies output conditions at the PF and LF interfaces’ (Chomsky, , p. ).1 In this framework, formal features are seen as indices on lexical items and larger syntactic objects that allow generated structures to be interpreted at external interfaces (i.e., PF and LF). Throughout the course of a derivational process, the operation known as feature checking takes place, which enables the legibility of linguistic expressions. Under the assumption that formal features and feature checking are universal elements of the CHL,2 a core

1

L = Language; PF = phonological form; and LF = logical form. CHL is the computational procedure for human language (Chomsky, ). This is roughly compatible with Hauser, Chomsky, & Fitch’s () Narrow Faculty of Language (FLN). 2

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

     



notion of the underlying linguistic competence of particular grammars (I-language) consists of knowing which features are assigned to which units and when and how they are interpreted and valued by external interfaces (commonly by means of the operation Spell-Out). According to these initial Minimalist assumptions, first language (L) acquisition proceeds through stages of acquiring the necessary functional projections, the formal features associated with lexical items and larger syntactic objects, and the additional steps of knowing when these features are to be valued with appropriate morphophonological reflexes at the PF-interface. The notion that the same processes take place in second language (L) acquisition has led to the postulation of the FRH. Seminal proposals by Lardiere (, , , ) and colleagues (see the contributions in Liceras, Zobl, & Goodluck, , for a general overview) laid the foundation for research based on the proposal of a feature reassembly operation, where the feature configuration of the L is adjusted to conform to the mapping of syntactic features onto morphological forms found in the L grammar. This research programme has been popular and successful in the generative L literature to date (Domínguez et al., ; Guijarro-Fuentes, ; Cho, ; Gil & Marsden, ; Spinner, ). As nuances have entered Minimalist theorizing since its inception, the FRH has also been updated to accommodate to these changes, with most contemporary studies adhering to the model of grammar commonly assumed in Distributed Morphology (DM; Embick & Noyer, ). Additionally, following initial proposals by Slabakova ( and Chapter , this volume), the question of whether or not the notion of feature is a more effective and appropriate atomic unit of linguistic competence than traditional parameters has also been raised. If language attrition is understood on a basic level to result in the loss of structural mappings, simpler structures—including stylistic variation—and paradigmatic levelling (Schmid, a), the FRH can also be extended to model these situations. Putnam & Sánchez () advance the proposal that the FRH can account for instances of incomplete acquisition3 and language attrition by appealing to the lack of activation (Levy et al., ; Linck et al., ; Seton & Schmid, ) of the recessive L grammar system over the lifespan of individuals. This approach differs from the notion of incomplete acquisition in that the latter assumes that at some point the process of acquisition is interrupted due to diminished input (Montrul, a). Incomplete acquisition has been argued to account for differences in ultimate attainment between heritage bilinguals living in a language contact situation and monolingual speakers with access to formal instruction (Montrul, ). More recently, the notion of incomplete mastery originally proposed by Berman () to characterize the variable nature of first language acquisition has been put forth as an explanation for the outcomes observed in heritage bilinguals (Montrul, b). Following Putnam & Sánchez (), we assume an alternative account to this view and propose to characterize variability across heritage speakers (HS), in particular those who experience attrition, as differences in activation for comprehension and production purposes rather than as incomplete acquisition.

3

For an alternative conceptualization, see Kupisch & Rothman ().

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi



, -,   ´ 

A study that shows how this alternative view analyses differences in production and comprehension is Perez-Cortes (). This study found evidence of restructuring of subjunctive features and their corresponding morphological forms among HS of Spanish living in the US modulated by differential levels of activation of Spanish. The acceptability of indicative and infinitive forms in contexts where subjunctive was required, for instance, was more frequent among HS who, despite exhibiting high levels of proficiency, displayed lower levels of activation of Spanish. These effects extended not only to acceptability, but also to production and interpretation tasks, as seen in the following examples obtained from an Acceptability Judgement task () and a Sentence Completion task (): () El gato tiene hambre y quiere que los ratones *salen del agujero The cat has hunger and wants that the mice leave[ppIND] the hole ‘The cat is hungry and wants the mice to leave the hole’ (Perez-Cortes, , p. ) () Bob Esponja quiere que Patrick *se viajar a Hawaii antes Bob Esponja wants that Patrick CL to travel[INF] to Hawaii before ‘Bob Esponja wants Patrick to travel to Hawaii before him’ (Perez-Cortes, , p. ) In these examples, the use of subjunctive forms [salgan, viajen] should have been obligatory due to lexical selection and disjoint reference. However, participants with lower levels of Spanish activation accepted—and produced—indicative () and infinitival () forms in these contexts, suggesting difficulties in remapping the modal feature selected by the matrix verb onto the corresponding morphological form (subjunctive). The emergence of indicative and infinitive as alternatives to the subjunctive was also observed in heritage bilinguals with low proficiency in the minority language, and can be explained by adopting a feature reassembly approach. In order to successfully master obligatory mood selection in Spanish, HS who are dominant in English are required to remap the interpretable feature [M], typically hosted by (for)-to infinitival complements (‘Spongebob wants Patrick to travel to Hawaii before him’), onto an inflected subjunctive form (Iverson et al., ; Kempchinsky, ). The results obtained in Perez-Cortes’ () study indicate that the process of feature reassembly leads to a wide variety of systematic outcomes largely modulated by bilinguals’ frequency of language activation and proficiency. HS who experienced decreased activation of Spanish and/or low levels of proficiency had more difficulties remapping the interpretable modal feature onto an inflected form, accepting and producing infinitival forms such as the one illustrated in () over their indicative counterparts (). As their use of Spanish for comprehension and production purposes increased, so did their ability to interpret, accept, and produce subjunctive as a marker of directive force and disjoint reference. Additionally, and as initially proposed by Putnam & Sánchez (), this fluctuating process of restructuring affected HS’s productive and receptive abilities in a very specific manner. In bilinguals with higher levels of Spanish activation and proficiency, morphological optionality was limited to production where they exhibited difficulties in retrieving subjunctive morphology. Despite these difficulties, they were able to interpret the absence of subjunctive morphology as ungrammatical, and its presence as a marker of disjoint reference. Those with lower levels of activation and proficiency, however, presented additional complications associating subjunctive to disjoint reference contexts, suggesting that the effects of decreased activation had also affected their grammatical representations.

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

     



Although the empirical focus of the FRH in language attrition contexts generally involves heritage grammars (Domínguez, ; Putnam & Sánchez, ; Cuza & Pérez-Tattam, ), there is nothing a priori that would prevent its application to processes involving the second language acquisition of a majority language by heritage speakers, or to multilingual populations. When analysing the linguistic development of HS, the FRH offers one possible way to address what Bayram et al. (Chapter , this volume) call the Heritage Bilingual Paradox (HBP): ‘how is it possible that native child naturalistic acquisition, albeit in heritage language contexts, can result in large disparities in grammatical competence between HS and monolinguals as well as across subsets of heritage speakers?’ Here we explore and build upon the core inner-workings of the FRH in the context of the development of heritage grammars across the lifespan as initially proposed by Putnam & Sánchez (). Crucially, we wish to emphasize that we shift the common narrative of incomplete acquisition in HS to the alternative of restructuring as has been posited in attrition studies (Polinsky, ; Yager et al., ; Yager, ) under the FRH. In what follows we focus on how the FRH can account for heritage grammars and on the strengths and weaknesses of this model in the face of evidence adduced from additional shifts and debates in the theoretical literature as well as advances on bilingual grammars from the perspective of cognitive neuroscience.

. FRH   

.................................................................................................................................. Putnam & Sánchez () propose a model of heritage grammar development that incorporates activation of features for comprehension and production (see Putnam & Schwarz, ; Yager et al., ; Cuza & Pérez-Tattam, ; Perez-Cortes, ; and Yager, , for extensions of this proposal). In this model, the mapping of syntactic features onto morphological forms may be differentially affected for comprehension versus production purposes. In top-down processes of integration for comprehension, some heritage speakers may be unable to access a representation that includes a mapping of syntactic features and morphology that resembles that of a monolingual or dominant speaker’s grammar. Others may be able to access such representations in comprehension tasks but the mapping would not be reflected in production. An example of such difficulties is provided in Perez-Cortes () in an utterance from an HS’s grammar with low proficiency in Spanish in which the modal feature is remapped to a periphrastic expression that can potentially be used in other structures to convey directive force instead of resorting to the subjunctive to do so: () El rey quiere que los príncipes *tien-en que dec-ir las The king wants COMP the princes have- that say- the noticias importantes. news important ‘The king wants the princes to have to say the important news.’ Those with the highest levels of proficiency may access such mappings in comprehension and also be able to show evidence of them in their production (Perez-Cortes, ). In this respect, HS show parallel developmental paths to those found in other groups of

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi



, -,   ´ 

attrited speakers who may similarly experience differential effects in comprehension and production (Seton & Schmid, ). This claim finds additional support in research by Gollan et al. (), which indicates that the activation of lexical items in comprehension shows different frequency effects than those for production. If, as suggested by Gollan et al. (), comprehension is primarily constrained by pragmatic and semantic factors, frequency effects related to the activation of lexical items and their related meanings will play an important role in this domain. Production, on the other hand, involves the integration of conceptual, syntactic, and phonological features (Bock & Levelt, ) and some level of automatization in accessing language units resulting from such integration. Given this complexity, production among heritage speakers is expected to result in more frequencysensitive stages. In this respect, comprehension contrasts with production, thus presenting significant challenges to bi/multilinguals leading to additional language dominance effects in this domain. What is more, if, following scholars such as Brysbaert & Duyck (), who claim that the competition of lexical items in the bilingual mind involves both languageinternal as well as cross-linguistic competition, information associated with lexical items exists in conflict with one another. We address this point in the final section (see also Kroll & Stewart’s () Revised Hierarchical Model). Putnam & Sánchez () present the sketch of a model of how language attrition would take place, making use of Minimalist tools. In their account, they assume that linguistic knowledge consists of sets of abstract features (Chomsky, , et seq.): (i) functional features (FFs); (ii) phonological form features (PF-features); and (iii) semantic features: [T]he acquisition of heritage language grammars involves a change in the frequency of the amount of input of the L, with extensive, if not exclusive, L input in the earliest stages of acquisition slowly shifting to L dominance at some point prior to ‘ultimate attainment’ of the L grammar. This process is thus characterized by changes in the frequency of exposure (i.e., from more to less) or by a limited exposure to lexical items that instantiate some values for FFs in language A (i.e., the L/heritage language). Increased exposure to lexical items in language B (i.e., the socially-dominant L) and lower frequency of exposure to lexical items in language A may result in lower levels of activation of certain FF values in language A. In turn, lower activation may result in their gradual replacement by the functional values in language B but in association with PF and semantic features of lexical items in language A. (Putnam & Sánchez, , p. )4

On this view, there is no ‘incomplete acquisition’ (pace Montrul, ) of the L/heritage grammar that takes place due to insufficient input, as the acquisition process is never truly interrupted. Rather, according to this proposal, heritage language acquisition and further developments (i.e., what is commonly referred to as language attrition) of the L/heritage grammar over the course of the lifespan are the result of the remapping of featural content (FF) to different semantic and PF-features. Abstracting away from age-related effects that have been appealed to in order to create a dichotomy between incomplete acquisition and language attrition, Putnam & Sánchez’s model interprets these effects as largely involving

4 This model can also subsume cross-linguistic effects (i.e., transfer) that are both facilitative and nonfacilitative in nature (see Jarvis & Pavlenko, ; Jarvis, ).

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

     



the same phenomenon (i.e., remapping and restructuring of linguistic information), thus viewing both of these processes as epiphenomenal. To account for the various degrees of interference and restructuring, Putnam & Sánchez establish different stages and their (potentially lasting) impact on the competence grammar of HS (adopted from Putnam & Sánchez, , pp. –): Stage : Transfer or re-assembly of some FFs from the L grammar to L PF and semantic features which may coincide with the activation of L lexical items on a more frequent basis from the standpoint of linguistic production; Stage : Transfer or re-assembly of massive sets of FFs from the L to the L PF and semantic features, while concurrently showing significantly higher rates of activation of L lexical items than L lexical items for production purposes (i.e., this group of bilinguals might code-switch more than those in the previous situation); Stage : Exhibit difficulties in activating PF and semantic features (as well as other FFs) in the L for production purposes but are able to do so for comprehension—especially in high frequency lexical items; and Stage : Have difficulties activating PF features and semantic features (as well as other FFs) in the L for both production and comprehension purposes.5 Attrition phenomena in adult learners represent an ideal area to test the descriptive adequacy of this model and its stages, because unlike young heritage speakers, adult attriters’ restructuring processes usually take place after having established specific feature values in their L and their mappings onto PF and semantic meaning. Furthermore, their level of activation of bundles of features is assumed to have been high prior to L acquisition. This can allow us to study those feature values that diverge in the L and the L and address the extent to which activation has an impact on the valuation of features.

. F 

.................................................................................................................................. As we have demonstrated here, the FRH continues to serve as a viable option to capture language attrition and language restructuring generalizations (e.g., Lardiere, , ; Slabakova, ). Although the FRH was born under the auspices of the initial proposal of the MP, recent developments in this research programme call for a radical reduction of the number of formal features as well as the outright removal of the latter from the CHL (see, e.g., Boeckx, ; Eguren et al., ). Recent models that adopt the position that morpho-syntax is due to the realization of cues or features at the PF-interface (see, e.g., Grimstad et al., ; Alexiadou, Lohndal, et al., ; and Riksem et al., to appear, for exoskeletal approaches to language mixing) are also compatible with the core tenets of the FRH we outline above. Future applications of the FRH in attrition and restructuring 5

Although this original proposal focused exclusively on morpho-syntactic information and its link with semantic and pragmatic meaning, there is nothing in principle that would prevent an extension to the phonological domain.

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi



, -,   ´ 

contexts can also be extended to proposals that adopt a parallel architecture of cognition underlying the language faculty (Jackendoff, , ; Mayer & Sánchez, ; Putnam, ). Under such assumptions, attrition and restructuring will be due to either the loss and/or reduction in connections among sub-levels or the introduction of new mappings. The possibility of uniting some version of the FRH with a parallel architecture is promising for a number of reasons: First, since the notion of ‘interface’ has been difficult to define in this research programme (Sorace, ; Montrul, a), a parallel architecture that focuses on the word as an interface rule would allow more fine-grained distinctions in grammatical descriptions. Second, this flexibility in the description of grammatical units is compatible both with the micro-cue approach (Lightfoot, ; Westergaard, ) and MacWhinney’s (Chapter , this volume) competition-based model. Third, the parallel architecture allows for a more detailed documentation of the two languages in conflict and the resulting grammatical representations of these duels. Evidence from cognitive neuroscience abounds showing that the conflict of competing forms is the norm rather than the exception (Kroll, Bobb, & Wodniecka, ; Kroll, Bobb, Misra, & Guo, ; Melinger et al., ). These ‘competing forms’ can be construed as correlates to features, which is the basic strategy adopted by Amaral & Roeper () in their proposal concerning the interaction of simultaneously conflicting linguistic information. Adopting this notion of conflicting grammatical representations interacting with executive and inhibitory psychological function results in a reclassification of language attrition proper, where attrition is perhaps more appropriately understood as the strengthening of neural activities and connections of the L/Ln grammars rather ‘than the actual degradation of the connections which represent the L’ (Seton & Schmid, , p. ; see also, e.g., Hulsen, ; Pallier et al., ; Schmid, ; Seidenberg, ). Additional evidence for this hypothesis can be adduced from the observation that receptive bilinguals (i.e., those who can comprehend a heritage grammar but encounter significant difficulties in producing forms in the same grammar or cannot produce them) maintain some degree of competence with respect to inflection morphology (see, e.g., Sherkina-Lieber, ). The application of a parallel model opens the door to gradient and probabilistic approaches to language acquisition, change, and attrition (Smolensky, ; Pater, ; Smolensky et al., ; Bousquette et al., ; Goldrick et al., a,b; Putnam & Klosinski, ). Irrespective of the formalism one chooses to model the restructuring of linguistic information in bilingual grammars, the FRH continues to serve as a viable option. As a scientific paradigm, the FRH is parasitic upon other core theoretical assumptions of a particular framework. Putnam & Sánchez’s () proposal to link the concept of formal feature with the Activation Threshold (Paradis, ; Ullman, b, ; see also Köpke & Keijzer, Chapter , this volume) has led to nascent research that also employs this version of the FRH. This research programme has yet to be fully tested in attrition studies; however, both the notion of feature reassembly and the stages proposed can be empirically investigated in the growing area of attrition research.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

  ......................................................................................................................

                                  ......................................................................................................................

    

. I

.................................................................................................................................. T Interface Hypothesis (IH) was originally proposed by Sorace & Filiaci () to explain the non-convergence and optionality revealed in very advanced adult second language (L) learners in the comprehension and production of certain structures. The original proposal suggested that those language structures that are sensitive to conditions involving an interface between syntax and other cognitive domains, such as pragmatics, are more difficult to be acquired completely than structures that do not involve such an interface. The IH has evolved over time from assuming a dichotomy between ‘narrow’ syntax and ‘interface’ structures to a more fine-grained differentiation among types of interface conditions (‘internal’ versus ‘external’; Sorace, ; Tsimpli & Sorace, ), opening the way for research on the multiple cognitive mechanisms involved in acquiring and processing structures sensitive to different interface conditions. Specifically, ‘external’ interface conditions, such as those operating at the syntax–pragmatics interface, involve the integration of ever-changing contextual and pragmatic information while processing language; ‘internal’ interface conditions, on the other hand, involve the integration of semantic or morphological information, which is not sensitive to external contextual changes and relies on well-rehearsed processing mechanisms. The IH was later extended to early bilingual acquisition and individual first language (L) attrition, suggesting that interface structures are less likely to be acquired completely for the former and more likely to undergo attrition for the latter. In relation to L attrition, the current hypothesis further proposes that individual L attrition affects only the ability to

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



  

process interface structures but not knowledge representations themselves (Sorace, , ). It is important to emphasize that this prediction relates to first generation individual attrition in speakers who have acquired the L completely before the onset of attrition and not to second generation attrition in heritage speakers, for whom the acquisition of the L may be incomplete or divergent, depending on the quantity and quality of input received. The IH has been supported by a large body of research exploring cross-linguistic influence effects for different interface structures in diverse bilingual groups. For example, many studies have been conducted on the acquisition of null versus overt pronouns in bilingual children (e.g., Paradis & Navarro, ; Serratrice et al., ; Argyri & Sorace, ; Sorace et al., ; Serratrice et al., ), advanced adult L learners (e.g., Belletti et al., ; Lozano, ; Rothman, ) and L attriters (e.g., Chamorro, Sorace, & Sturt, ; Montrul, a; Tsimpli et al., ). The present chapter will focus on first generation individual attrition from the point of view of the IH, and particularly on recent research that provides evidence of the selectivity and reversibility of individual L attrition.

. T I H    L 

.................................................................................................................................. In line with findings in bilingual L acquisition and adult L acquisition, recent research on L attrition also supports the IH, revealing that the structures at the syntax–pragmatics interface are the most prone to undergo attrition, causing ‘emerging optionality’ in the attrited speakers. Sorace (a) tested anaphora resolution in Italian near-native speakers of English under L attrition and found that they overgeneralized overt pronouns in Italian to contexts in which native speakers of Italian would use a null pronoun due to the influence from English. Sorace further established a connection between L acquisition and L attrition given that both English near-native speakers of Italian and Italian L attriters overextended the use of overt pronouns in Italian as a result of the influence from English. Tsimpli et al. () also reported attrition effects in a group of Greek and Italian nearnative speakers of English in relation to subject pronouns. They tested the production and interpretation of null versus overt pronouns in Greek and Italian using a picture verification task to elicit the participants’ preference for the subject or the object antecedent with each pronoun. During the experiment, participants were presented with three pictures and an ambiguous sentence like the ones in (), and they were asked to choose the picture or pictures that correctly matched the meaning of the sentence. () a. Quando lei attraversa la strada, l’anziana signora saluta la ragazza. ‘While she crosses the street, the old woman greets the girl.’ b. Quando pro attraversa la strada, l’anziana signora saluta la ragazza. ‘While (she) crosses the street, the old woman greets the girl.’ c. L’anziana signora saluta la ragazza quando lei attraversa la strada ‘The old woman greets the girl when she crosses the street.’ d. L’anziana signora saluta la ragazza quando pro attraversa la strada ‘The old woman greets the girl when (she) crosses the street.’

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

     



Consistent with Sorace’s (a) results, attrition effects were found for Italian attriters on the interpretation of the overt pronoun, for which they showed more indeterminacy in their antecedent preferences than the control group of Italian monolinguals. In contrast, no attrition was revealed with the null pronoun, for which both groups of attriters preferred the subject as its antecedent, in line with both control groups of Greek and Italian monolinguals. Interestingly, the monolingual control group in this study did not perform categorically, showing a tendency towards more variable interpretation of the overt pronoun. Tsimpli et al. () also investigated the interpretation of preverbal versus post-verbal subjects in a group of native Greek L speakers of English. In this study, participants were also presented with three pictures and a pair of sentences like the ones in (), in order to compare the speakers’ interpretation of preverbal subjects, which are usually understood as ‘old’ information (i.e., topic), as in (a), versus post-verbal subjects, which can be ambiguously understood as ‘old’ or ‘new’ information, as in (b). () a. I gitonisa mu ston trito orofo apektise dhidhima. Xtes vradhi my neighbour on the third floor had twins last night ena moro ekleje. one baby cry-. ‘My neighbour on the third floor had twins. One of the twins was crying last night.’ b. I gitonisa mu ston trito orofo apektise dhidhima. Xtes vradhi my neighbour on the third floor had twins last night ekleje ena moro cry-. one baby. ‘My neighbour on the third floor had twins. A baby (one of the twins or some other baby) was crying last night.’ The results from this experiment also revealed attrition effects, given that Greek attriters showed significantly more indeterminacy when interpreting preverbal subjects in comparison with the control group of Greek monolinguals. Based on the results obtained from these two structures, the authors concluded that attrition affects properties at the syntax–pragmatics interface but not syntactic features obeying conditions internal to the grammar. Gürel (a) also investigated the L attrition of null and overt pronouns in Turkish L learners of English and found language attrition to be selective. Turkish has two overt pronouns, o ‘s/he’ and kendisi ‘self ’, and a null pronoun. Whereas all three pronouns can occur in subject position, as in (), and in object position, as in (), only the overt pronoun kendisi and the null pronoun, but not the overt pronoun o, can refer to the subject of the main clause when they occur in object position, as () illustrates. () O/kendi-si/pro Londra’ya git-ti (S)he/self-/pro London- go- ‘(S)he went to London.’ () Buraki o-nu*i/j/kendi-si-nii/j/proi/j begˇen-iyor Burak (s)he-/self--/pro like- ‘Buraki likes him*i/j/selfi/j/proi/j’

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



  

This study involved three tasks (a written interpretation task, a truth-value judgement task, and a picture identification listening task) that tested the participants’ interpretation of sentences like () and (). Gürel (a) reported that the interpretation of the overt pronoun o in Turkish was influenced by English because attriters appeared to treat this Turkish overt pronoun as if it was the English overt pronoun (i.e., co-referential with the subject antecedent). In contrast, the interpretation of the null pronoun and the overt pronoun kendisi did not show attrition effects. Gürel interpreted these results under Paradis’ () Activation Threshold Hypothesis (ATH) (see also Köpke & Keijzer, Chapter , this volume). The ATH establishes a correlation between the frequency of use of a language element and its availability (or activation) to the speaker. In particular, it proposes that when an item is not used, the threshold of activation would rise, and when it is used, the threshold of activation would be low. Therefore, a linguistic item that has not been frequently used would have a high activation threshold and it would be difficult to activate, which would lead to the attrition of the item. This suggests that different language elements, depending of their frequency of use, would have different thresholds of activation, so that some would be more likely than others to undergo attrition. More specifically, the ATH predicts that L attrition will occur when an element in the L with a high activation threshold (i.e., disused) has a corresponding ‘competing’ element in the L with a lower activation threshold (i.e., used more frequently). Gürel’s results are predictable under the ATH because it is the Turkish overt pronoun o, which is in competition with the English overt pronoun, the one that shows attrition due to its disuse in Turkish and frequent use in English, but the overt pronoun kendisi or the null pronoun in Turkish, which do not have a competing item in the L, do not show attrition effects. The ATH is also partly compatible with the most recent version of the IH, in that it provides a processing explanation, which is, however, only focused on L–L cross-linguistic effects. Otheguy et al. () investigated attrition effects in the use of Spanish pronouns by different Spanish-speaking communities in the United States. They analysed the use of pronouns using data from a corpus of , verbs extracted from interviews conducted with the six largest Spanish-speaking communities living in New York City and originated in six different Latin American countries. In order to analyse the use of overt pronouns, speakers were divided in terms of their dialect regions, ‘Caribbean’ (newcomers from the Dominican Republic, Puerto Rico, and Cuba) versus ‘Mainlanders’ (newcomers from Ecuador, Colombia, and Mexico), and in terms of their generation, ‘newcomers’ (those who arrived in New York after the age of  and had lived there for a maximum of five years) versus ‘born and/or raised in New York’ (those who were born in New York or arrived before the age of ). The results from this study revealed that overt pronouns were used more frequently by the Caribbean speakers than by the Mainlanders. More importantly, the ‘born and/or raised in New York’ group showed a significantly higher rate of overt pronouns than the newcomers, which revealed the influence from English in the use of overt pronouns in the Spanish of both Caribbean and Mainlanders speakers living in New York City. L attrition effects have been found in languages with different inventories of anaphoric forms. Wilson () and Wilson et al. () used the visual world eye-tracking methodology to investigate the online processing of German anaphora with demonstratives and pronouns by English-speaking L learners of German and German-speaking L learners of English under L attrition. Participants were presented with a set of pictures while they

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

     



heard a pair of sentences like the ones in () and were asked to answer a yes/no question that revealed their antecedent preferences for the pronouns. Similar to the distribution of null and overt pronominals in null-subject languages, personal pronouns in German (er, sie, es) refer to the subject antecedent whereas demonstrative pronouns (der, die, das) refer to the object antecedent. () a. Der Kellner erkennt den Detektiv als das Bier umgekippt the- waiter recognizes the- detective as the beer tipped-over wird. Er ist offensichtlich sehr fleißig. is he- is clearly very hard-working ‘The waiter recognizes the detective as the beer is tipped over. He is clearly very hard working.’ b. Der Kellner erkennt den Detektiv als das Bier umgekippt the- waiter recognizes the- detective as the beer tipped-over wird. Der ist offensichtlich sehr fleißig. is he- is clearly very hard-working ‘The waiter recognizes the detective as the beer is tipped over. He is clearly very hard working.’ c. Den Kellner erkennt der Detektiv als das Bier umgekippt the- waiter recognizes the- detective as the beer tipped-over wird. Er ist offensichtlich sehr fleißig. is he- is clearly very hard-working ‘The waiter is recognized by the detective as the beer is tipped over. He is clearly very hard working.’ d. Den Kellner erkennt der Detektiv als das Bier umgekippt the- waiter recognizes the- detective as the beer tipped-over wird. Der ist offensichtlich sehr fleißig. is he- is clearly very hard-working ‘The waiter is recognized by the detective as the beer is tipped over. He is clearly very hard working.’ The results from these studies revealed that while L learners performed similarly to German native speakers with pronouns, they showed indeterminacy with demonstratives, revealing no clear preference for the object as their antecedent. Similarly, attriters showed more attrition effects with demonstratives than with pronouns in comparison to the control group of monolinguals, also revealing no clear preference for a specific antecedent. The results also showed that the degree of attrition effects depended on the attriters’ length of residence in the UK. The literature reviewed in this section supports the claims of the IH that structures at the syntax–pragmatics interface are likely to undergo attrition. There are also studies, mostly on L acquisition, exploring interface structures that have revealed mixed results with regards to the IH, with some findings being consistent and some being inconsistent with the IH (e.g., Slabakova & Montrul, ; Rothman & Iverson, ; Judy & Rothman, ; see Leal Méndez et al., , on heritage speakers). Crucially, many of these studies typically use offline measures involving metalinguistic tasks, which are not the most suitable to test the predictions of the IH concerning the processing of structures at the syntax–pragmatics

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



  

interface, rather than the mental representation of these structures (see Section . below). While a detailed review of these studies is beyond the scope of this chapter, they are a reminder that more research is needed—on a wider range of structures and based on multiple experimental methods—to precisely identify the boundaries on the applicability of the IH.

. S   L 

.................................................................................................................................. The previous section presented research on L attrition that supports the predictions of the IH, as they reveal attrition effects with structures at the syntax–pragmatics interface. This selectivity of L attrition as well as Sorace’s (, ) proposal that individual L attrition affects only the ability to process interface structures but not knowledge representations was tested in two recent studies: Chamorro, Sorace, & Sturt (), which investigated a structure involving an external interface, the syntax–pragmatics interface (i.e., subject pronouns), and Chamorro, Sturt, & Sorace (), which investigated a structure involving an internal interface, the syntax–semantics interface (i.e., the personal preposition). Chamorro, Sorace, & Sturt () investigated whether pronominal subjects, a structure at the syntax–pragmatics interface, would undergo attrition in L Spanish under prolonged exposure to the L English and, if so, whether these effects occur at the processing or representational level. Therefore, the interpretation and processing of overt versus null subject pronouns in Spanish was investigated using an offline naturalness judgement task and an online eye-tracking-while-reading task, where participants were presented with anaphora in which the antecedent preferences were predicted using Carminati’s () Position of Antecedent Strategy (PAS). Carminati proposed the PAS for Italian intra-sentential anaphora and it postulates that null pronouns are generally assigned to the subject antecedent, as in (a), whereas overt pronouns are generally assigned to the object antecedent, as in (b). The PAS was shown by Alonso-Ovalle et al. () to also apply to Iberian Spanish. () a. Quando Marioi ha telefonato a Giovannij, proi aveva appena finito di mangiare. ‘When Mario has telephoned Giovanni, (he) had just finished eating.’ b. Quando Marioi ha telefonato a Giovannij, luij aveva appena finito di mangiare. ‘When Mario has telephoned Giovanni, he had just finished eating.’ Based on Carminati’s PAS, participants in Chamorro, Sorace, & Sturt () were presented with intra-sentential semantically-neutral forward anaphora like the ones in (), where the grammatical number of the antecedents was manipulated such that the pronoun could only refer to either the subject or the object antecedent. This led to sentences where the pronoun agreed in number with the pragmatically infelicitous antecedent as predicted by the PAS, as in (a) and (d), and sentences where the pronoun agreed in number with the pragmatically felicitous antecedent, as in (b) and (c).

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

     



() a. La madre saludó a las chicas cuando ella cruzaba una calle the mother greeted- to the girls when she crossed- a street con mucho tráfico. with much traffic ‘The mother greeted the girls when she crossed a street with a lot of traffic.’ b. Las madres saludaron a la chica cuando ella cruzaba una calle the mothers greeted- to the girl when she crossed- a street con mucho tráfico. with much traffic ‘The mothers greeted the girl when she crossed a street with a lot of traffic.’ c. La madre saludó a las chicas cuando pro cruzaba una calle the mother greeted- to the girls when pro crossed- a street con mucho tráfico. with much traffic ‘The mother greeted the girls when she crossed a street with a lot of traffic.’ d. Las madres saludaron a la chica cuando pro cruzaba una calle the mothers greeted- to the girl when pro crossed- a street con mucho tráfico. with much traffic ‘The mothers greeted the girl when she crossed a street with a lot of traffic.’ On the other hand, Chamorro, Sturt, & Sorace () tested whether L attrition also affects structures involving internal interfaces or only those structures that involve external interfaces such as subject pronouns. In order to do so, they investigated the interpretation and processing of a syntax–semantics interface structure, the Spanish personal preposition a or differential object marking (DOM), to explore whether structures sensitive to semantic conditions also undergo L attrition. This structure differs from pronominal subjects in that its use does not depend on context, but is conditioned by semantic factors such as the animacy and/or specificity of the direct object. In line with Chamorro, Sorace, & Sturt (), this study also investigated whether any attrition effects revealed with this structure would be related to inconsistent or inefficient processing in real time or to a change in the attriters’ L knowledge representations (i.e., in their L grammatical competence) by implementing the same offline and eye-tracking measures as the pronoun study. The DOM, which occurs in Spanish but not in English, establishes that some direct objects must be introduced by a dative preposition, a ‘to’ in the case of Spanish. The presence or absence of this preposition depends on the type of direct object. Generally speaking, in Spanish, a direct object must be marked with the dative preposition if it is animate and specific, as () below shows, but inanimate direct objects, independently of the specificity, must not be preceded by the preposition, as in (). () a. María vio al1 niño esta mañana. María see- to-the kid this morning ‘María saw the kid this morning.’ 1

Note that al is the contraction of the preposition a and the masculine singular definite article el.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



   b. *María vio el niño esta mañana María see- the kid this morning ‘María saw the kid this morning.’

() a. María vio una película/la película esta mañana. María watch- a movie/ the movie this morning ‘María watched a movie/the movie this morning.’ b. *María vio a una película/la película esta mañana. María watch- to a movie/ the movie this morning ‘María watched a movie/the movie this morning.’ Based on this, participants in Chamorro, Sturt, & Sorace () were presented with sentences like the ones in (). Each item consisted of a simple sentence which contained a specific direct object, either animate or inanimate, which led to sentences where the animate object was correctly introduced by the preposition, as in (b), or ungrammatically lacked the preposition, as in (a), and sentences where the inanimate object correctly lacked the preposition, as in (c), or was ungrammatically introduced by it, as in (d). () a. *Juan defendió el conductor que fue despedido. Juan defend- the driver that was fired ‘Juan defended the driver that was fired.’ b. Juan defendió al conductor que fue despedido. Juan defend- to-the driver that was fired. ‘Juan defended the driver that was fired.’ c. Juan defendió el argumento de forma efusiva Juan defend- the argument in way effusive ‘Juan defended the argument in an effusive way.’ d. *Juan defendió al argumento de forma efusiva. Juan defend- to-the argument in way effusive ‘Juan defended to the argument in an effusive way.’ Unlike the experimental items in Chamorro, Sorace, & Sturt (), in these experimental items, whether the personal preposition should be used or not is completely clear and does not require participants to consult the discourse context: if there is an animate direct object, then the preposition must be used, and if the direct object is inanimate, the preposition must not be used, regardless of any context. The same group of Spanish native speakers under L attrition was tested in both studies to investigate two questions: () whether the IH would correctly predict attrition effects with the structure involving an external interface (i.e., pronominal subjects) but not with the structure involving an internal interface (i.e., the DOM); and () whether any attrition effects revealed occurred at the processing level (i.e., in the online eyetracking-while-reading task) but not at the representational level (i.e., in the offline naturalness judgement task). The group of attriters tested in both studies consisted of twenty-four Spanish native speakers from Spain who had been residing in the UK for a minimum of five years and were near-native speakers of English. This group was compared to a control group of monolinguals, which consisted of twenty-four Spanish

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

     



native speakers from Spain who had recently arrived in Edinburgh and had very little knowledge of English. Chamorro, Sorace, & Sturt’s () results on pronominal subjects reveal that attrition occurs with this structure at the syntax–pragmatics interface and that attrition affects online sensitivity rather than offline judgements. The offline judgement data shows no significant differences between the group of attriters and the group of monolinguals, with both groups showing equal sensitivity to the pronoun mismatch (i.e., anaphora containing an overt pronoun when a null pronoun is appropriate or anaphora containing a null pronoun where an overt pronoun is appropriate) when rating anaphora offline. Results from the eye-tracking experiment revealed that, in online reading, the group of monolinguals was reliably more sensitive than the attriters to the pronoun mismatch, as the monolinguals showed a significant mismatch sensitivity with pronominal subjects (i.e., significant Pronoun by Antecedent interaction effects in their reading times in the critical region, where the pronoun occurs, as well as in the post-critical region), whereas attriters did not reveal online sensitivity to the pronoun mismatch (i.e., no significant interaction effects were shown in any of the regions) and performed significantly differently from monolinguals. In contrast, Chamorro, Sturt, & Sorace’s () results on the DOM revealed no attrition effects with this structure at the syntax–semantics interface. The offline results showed an equal sensitivity to DOM violations for both groups, with all participants correctly differentiating the grammatical sentences in which the animate object was preceded by al and those in which the inanimate object was preceded by el from the ungrammatical sentences in which the animate object was preceded by el and those in which the inanimate object was preceded by al. Similarly, the eye-tracking results showed very early sensitivity to DOM violations (i.e., both groups showed significant interaction effects of Animacy by Article in first-pass reading time in the critical region, which is the earliest possible point), and this sensitivity was of an equal magnitude across both groups, with no significant differences revealed between the groups. This suggests that both groups were sensitive to the mismatching conditions when processing the DOM in real time (i.e., when the animate object was incorrectly preceded by el or when the inanimate object was incorrectly preceded by al). Considering the findings from both studies, Chamorro, Sorace, & Sturt () and Chamorro, Sturt, & Sorace () provide further support for the IH and the selectivity of individual L attrition, as it was shown that attrition affects structures at syntax–pragmatics interface, such as pronominal subjects, but that structures requiring the satisfaction of semantic conditions, such as the DOM, do not undergo attrition, either at the processing or the representational level. Crucially, these studies provide supporting evidence for Sorace’s (, ) proposal, as it was revealed that attrition affects online sensitivity with structures at the syntax–pragmatics interface but not knowledge representations.

. I   L        ?

.................................................................................................................................. In line with the findings discussed in Sections . and . for L attrition, research on L acquisition also reveal that advanced adult L learners show optionality in the production

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



  

and/or interpretation of anaphoric forms (e.g., Sorace & Filiaci, ; Belletti et al., ; Lozano, ; Rothman, ). Another aspect that these L studies have in common, as well as the ones on L attrition reviewed above, is that they all investigate speakers who have English as the other language. Therefore, one could argue that the optionality revealed with interface structures by these L learners and L attriters may be due to interference from the language that does not have a choice of anaphoric forms (i.e., English) on the language that has such a choice, as some of these studies have concluded. However, anaphoric forms have also been found to cause optionality in bilingual speakers of two typologically similar languages, not only in speakers of two null-subject languages (Bini, ; Lozano, ; Margaza & Bel, ; Mendes & Iribarren, ; de Prada Pérez, , ; Sorace et al., ), but also in speakers of two Germanic languages (Juvonen, ; Ellert, ), bimodal bilinguals (Bel et al., ), and unimodal bilinguals (Chen Pichler & Koulidobrova, ). These results suggest that the optionality revealed in L attrition and L acquisition with interface structures cannot be due only to interference effects from English, but may be also related to the cognitive effort of handling any two languages in real time (Sorace, , ). In addition, the convergences between L attrition and L acquisition suggest that both languages of late bilingual speakers may be affected by cognitive changes due to accommodating an L and that having two plausible anaphoric options may lead to processing difficulties when speakers compute pronoun-antecedent mappings, both in production and comprehension. It has been proposed that these difficulties could be attributed to bilingual speakers’ reduced efficiency when integrating information from different domains in real time and updating the mental discourse model when needed, as a side effect of the need to exercise inhibitory control to avoid interference from the language not in use (see Green, ; Costa et al., ; Green & Abutalebi, , on the role of inhibitory control; and Sorace, , on possible consequences for integration). Therefore, if the effects of attrition do not involve language change at the representational level, but rather affects the cognitive strategies to access and use this knowledge in real time, one may predict that these effects are not irreversible but may be sensitive to the frequency and recency of exposure to the L. The question of the reversibility of individual attrition was investigated by Chamorro, Sorace, & Sturt (), which will be discussed in the following section.

. R   L 

.................................................................................................................................. Another important issue in individual L attrition is the question of whether attrition effects are permanent or can be reversed. Chamorro, Sorace, & Sturt () also explored this question by investigating whether attrition effects with pronominal subjects can decrease or disappear with recent re-exposure to the attriters’ L. In order to investigate this, Chamorro, Sorace, & Sturt () tested a second group of Spanish L attriters using the same experimental items and the same offline and online tasks as the ones described in Section .. In line with the other group of attriters, this second group of attriters also consisted of twenty-four Spanish near-native speakers of

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

     



English who had been living in the UK for a minimum of five years, but they had been exposed exclusively to Spanish in Spain for a minimum of a week right before they were tested. This ‘exposed’ group was compared to the other group of attriters and the control group of monolinguals to explore whether attrition effects with structures at the syntax– pragmatics interface can decrease or disappear after recent exposure to L input. The results obtained from this exposed group revealed that attrition effects are not permanent but decrease as a result of L re-exposure. The offline judgement data revealed that this group was not significantly different from the attriters or the monolinguals, with all groups showing equal sensitivity to the pronoun mismatch when rating anaphora offline (i.e., all three groups revealed significant Pronoun by Antecedent interaction effects in their ratings and no significant differences were revealed when these groups were compared). Results from the eye-tracking experiment revealed that, in online reading, the monolingual and exposed groups are reliably more sensitive than the attriters to the pronoun mismatch, as they show significant Pronoun by Antecedent interaction effects in their reading times with these sentences in the critical region (i.e., the region where the pronoun occurs) as well as in the post-critical region, while the attriters do not show any significant interaction effects in any of the regions. These results can be explained following Paradis’ () ATH, which predicts that attrition diminishes with frequency and recency of exposure to the L. In line with this hypothesis, the results obtained for the exposed group did not reveal attrition effects with pronominal subjects, since this group, unlike the attriters, showed a reliable sensitivity to the pronoun mismatch when processing this interface structure in real time. Moreover, when they were compared to the monolinguals, no significant differences between the two groups were revealed, which suggests that attrition effects diminish after recent exposure to the L. Taking together all the results from Chamorro, Sorace, & Sturt () and Chamorro, Sturt, & Sorace (), these findings reveal that individual L attrition is selective and occurs with structures that involve an external interface (i.e., the syntax–pragmatics interface) but not with structures that involve an internal interface (i.e., the syntax– semantics interface). They also reveal that attrition effects on these structures are more likely to be revealed in tasks tapping real-time processing rather than offline metalinguistic tasks such as acceptability judgements. Crucially, Chamorro, Sorace, & Sturt () provide evidence for the question raised in this section, as it was revealed that individual L attrition affects online sensitivity with structures at the syntax–pragmatics interface rather than causing a permanent change in speakers’ L knowledge representations. This reveals that bilinguals are sensitive to input changes and provides further support for Sorace’s (, ) proposal that first generation individual attrition affects only the ability to process interface structures but not knowledge representations themselves.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

  ......................................................................................................................

                  ......................................................................................................................

 

. I

.................................................................................................................................. T Bottleneck Hypothesis (BH; Slabakova, , , ) was proposed to identify properties and constructions that are harder or easier for learners to acquire in their second language (L). It is predicated on the generative linguistics notion attributing a special place to functional morphology in grammar and acquisition. Functional morphemes, both bound (inflectional) as well as free, carry grammatical meanings that radically change the overall meaning of a sentence. No sentence meaning is complete without taking the contributions of functional morphology into account. The following example illustrates the division of labour between lexical items and functional categories. Suppose we want to combine the lexical items Jane, eat, and sandwich into a message with this meaning: ‘Jane is consuming a sandwich at the moment of speaking.’ It is a rare language that would just string the words together with no grammatical morphemes and without taking word order into account. In English, for example, the meanings of , , and  noun are imparted by the indefinite determiner a; the aspectual meaning   is reflected in the progressive -ing; the temporal meaning of   and agreement with a third person singular subject are captured by the auxiliary is.1 The final makeup of the sentence Jane is eating a sandwich carries a complete meaning (its truth value) composed of lexical and grammatical meanings that are considerably removed from the raw assembly of lexical verb and nouns. These additional meanings that anchor the event in time and space are contributed by the functional morphemes.

1 I use small letters to set out grammatical meanings with predominantly semantic content, as opposed to purely grammatical meanings such as subject–verb agreement.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

     



The grammatical and semantic features expressed by functional morphology, such as the ones mentioned above as well as , , etc., are universal in the sense that they have to be expressed in every language, in one way or another. However, languages differ with respect to how these universal meanings are expressed. For example, Mandarin Chinese does not express the past time of a state or an event with a dedicated morpheme, while English does. Korean and Russian are languages without a definite article. The lack of overt exponents does not mean that the respective languages cannot express those meanings, but that they express them by other means, including leaving them for the context to fix. An important feature of functional categories is their multi-functionality. In languages such as English and many others, functional morphemes can encode multiple grammatical meanings. Linguists express this state of affairs by postulating that a number of grammatical features reside (are checked) in the same functional category. In the sentence Gwen often eat-s fish, the agreement morpheme -s signals  ,   and agreement with a third person subject. However, two other grammatical meanings are also captured by features in the same functional category Tense: the obligatory subject (*
often eats fish) and the lack of verb movement over the adverb and negation (*Gwen eats often fish, *Gwen eats not fish). Thus, multi-functional Tense in English carries at least five features, three semantic features, and two formal syntactic features. In addition to being indispensable in carrying grammatical meanings, the functional morphemes are the main locus of linguistic variation. As per the Minimalist Program assumptions (Chomsky, ), the core syntactic operations (e.g., Merge, Agree, etc.) are the same in all natural languages, hence they are also universal just as the grammatical meanings. Parametric differences are no longer attributed to the syntax of two languages differing in constrained ways. If syntactic operations are universal, then variation among languages can only be located in the functional lexicon, the part of the lexicon where the functional morphology resides, and at the two interfaces with meaning and form (or expression). How about semantics? Just like syntax, semantics is postulated to contain only universal operations (e.g., functional application, lambda abstraction, etc.) which allow languages to translate sentence syntactic structure into sentence meaning, or truth value. As mentioned above, grammatical meanings such as , , and  are reflected in the functional lexicon. We can think of each lexical item, including functional and open-class ones, as combining information of three types: semantic, phonological, and morpho-syntactic (Jackendoff, ). To illustrate with a verb such as drink, when we know the verb, we know that it means ‘to consume or imbibe a liquid’; we know it is made up of a unique sequence of sounds [drıŋk]; we also know that it is a transitive verb that most of the time takes an object, and that this object may be dropped but is implied (e.g., when used to characterize a person, He drinks does not mean He drinks water). This view of the language architecture (see Slabakova, , chs  and ) informs views and predictions about language acquisition in general. In addition to learning the equivalents of the native open-class lexical items, acquiring the functional morphology with all its features secures the skeleton of the grammar. Potentially speaking, the three types of information encoded in lexical items can be decoupled in acquisition: learners may not

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi





learn them all at the same time. For instance, one can imagine a speaker knowing the basic meaning of drink but not understanding the significance of the implied object (that it can only refer to alcohol). While formal syntactic features are universal, their UniversalGrammar-provided values are reflected in the specifications of the functional morphology and need language-specific input to be fixed. Grammatical meanings (, ) can also be uncoupled in acquisition, especially if they are complex ( and  ). For example, the usage of the progressive tense with stative verbs to convey temporary events (He is being honest) may be acquired later than its primary meaning of ongoing event (He is playing). In summary, formal syntactic features and grammatical meanings reflected in a functional category can be dissociated in acquisition, as the features can be learned one by one, or in bundles. While this view of the language architecture is theory-specific, there is experimental evidence to support the dissociation between functional morphology and syntax. White (, pp. –), following the idea of morphological separation (Beard, , ; Halle & Marantz, ), provides convincing evidence for the so-called syntax-before-morphology view. She argues that L acquisition data from children and adult learners point to the following dissociation: The morphological expressions of the functional category Tense are being produced with accuracy between around % and % in obligatory contexts, while the syntactic reflexes of the same functional category (nominative subject, subject remains in VP) are observed with close to a % accuracy. It seems that the invisible syntactic knowledge is in place much earlier than the obligatory suppliance of the morphological expressions. Slabakova (, ) argued that universal semantic knowledge is also in place without much effort on the part of learners. To take a specific example, Slabakova () examined whether learners of L Chinese know which sentences are to be interpreted as encoding a past action, if there is no overt morphological expression of Tense. It turns out they do, and even at intermediate levels of proficiency. These and many other studies (see Slabakova, , for a review) supply evidence that syntactic and semantic knowledge come before fully accurate realization of the inflectional morphology. In addition, functional morphology is often expressed by complex paradigms (e.g., subject–verb agreement in Slavic and Romance languages), which have to be learned and accessed as separate lexical entries and assembled online both in comprehension and production. Since there is an element of effort-intensive lexical learning in acquiring paradigms, inexperienced learners (children, L learners) at the beginning of the acquisition process often assume one form to be the representative of the whole morphological paradigm, or a default form. Paradigmatic forms present more difficulty in processing language, as opposed to single-form expressions. While there are solid linguistic explanations of why a certain member of a paradigm can become a default, the phenomenon of default forms in principle suggests that not all members of a paradigm are equal, lexically learned at the same time and processed equally easily. When non-default forms are required by the linguistic context, errors are more likely. Finally, some functional morphology has only grammatical functions and not much communicative import. Take for example the agreement between a noun and its modifying adjective. Whether the agreement morpheme is present on the adjective or not, the noun phrase is still comprehensible. Such a situation might apply processing pressure for omissions or supplying default forms.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

     



The BH, whose application to attrition I will explore in this chapter, is based on comparing degrees of success or difficulty in the L acquisition of three modules of the grammar: syntax, semantics, and morphology. Its main tenets, as specified in (i), (ii), and (iii), are both theoretically motivated and supported with acquisition facts. i. Functional morphology is more challenging for learners than syntax or semantics, which are universal. This claim is founded on the minimalist view that functional morphology encodes variation among languages, and hence is the locus of what has to be acquired. In addition, processing and learning considerations also point to heightened difficulty. ii. The syntactic representations attained by learners may be accurate even though the associated overt morphology may be lacking. iii. The compositional semantic representations employed by learners are accurate. In this chapter, I will apply this language acquisition hypothesis to a different population with the ensuing different outcomes. The purpose is trying to account for language loss, not language gain. Together with recent treatments (see Schmid, b, for an overview), I will assume that language attrition represents a reanalysis of the grammar over the course of a lifetime in the absence of abundant, consistent input, and in this reanalysis, the language architecture and inherent complexity of grammatical features are an essential factor that affects the process. When a fully acquired language is used under reduced input conditions and under pressure from another language in a bilingual environment, at issue is which areas of grammar remain resilient and likely to be spared, and which ones are vulnerable to attrition? The BH can provide a useful lens to these issues.

. A     :   

.................................................................................................................................. Recent research on language attrition in the lexicon as well as in grammar has led to the conclusion that we can no longer treat language attrition as a unitary phenomenon. In addition to the high amount of individual variation due to personal, motivational, and experiential factors, it appears that the age of onset of bilingualism is of utmost importance. With all the necessary caveats needed, when we average over a multitude of people’s linguistic experiences, we may be describing two distinct processes. One would be the heritage speaker experience: child learners growing up in homes where the native language is spoken, but in countries with another majority language. Very commonly, the heritage language is the individual’s first language chronologically, a mother tongue learned in a naturalistic setting in a family environment. The majority language comes in second. However, being used at school, with peers, and in the wider society, the majority language often becomes dominant and has a dramatic effect over the native language, setting off language attrition (Montrul, ; Benmamoun et al., a). Broadly defined, the exposure

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi





to the dominant societal language comes either at school age or at the age of immigration, but crucially before the age of puberty (around ) (see Bylund, Chapter , this volume, as well as Kupisch & Rothman, , for more discussion). In our context, these speakers could be called ‘early attriters’. Summarizing recent research, Benmamoun et al. (a) argue that loss of functional structure is evident in this population. Another population exhibiting language attrition is adult immigrants. Very often, they are the parents of the heritage language speakers described above. Exposure to the L is sustained and intense, and may be accompanied with a sharp decrease in exposure to the native language. In our context, the defining characteristic of this group is that they have left the country where their mother tongue is spoken after the age of puberty, and sometimes much later. This ensures that their native language grammar was completely mature and stable before the onset of (potential) attrition. For the purposes of this chapter, I will call these ‘late attriters’. The current evidence suggests that changes in the core grammar, or in the functional category structure, is not much in evidence in this population. So, while we are pondering the issue of why is it the case that certain areas of the grammar are so resilient whereas other areas are malleable and subject to change throughout the life of a speaker, we should also keep in mind that the age of onset of attrition can change the answer dramatically (Schmid, ; Montrul et al., ).

.. The grammatical competence of early attriters Heritage speaker competence and performance are notoriously diverse and fluid, so generalizations are to be attempted with caution. Nevertheless, generalizations are indispensable in checking theoretical predictions. Benmamoun et al. (a), a keynote article digesting decades of research on heritage language speakers, summarizes and interprets data from several areas where heritage speakers appear to diverge from monolingual native speakers: phonology, lexical knowledge, morphology, syntax, case marking, and codeswitching. In order to check the predictions of the BH against these pooled findings, we shall examine the areas of functional morphology and syntax in some more detail. The following citations come from Benmamoun et al. (a). Inflectional morphology in languages that exhibit robust morphological systems, including regular and irregular paradigms, is particularly vulnerable in heritage languages. (Benmamoun et al., a, p. ) Morphological deficits in heritage languages are asymmetric; they seem to be more pronounced and pervasive in nominal morphology than in verbal morphology. (Benmamoun et al., a, p. ) Agreement is the functional feature that undergoes the highest erosion. (Benmamoun et al., a, p. )

A number of studies have demonstrated significant loss of subject–verb agreement in heritage speakers (as compared to monolinguals): Håkansson (); Anderson (); Polinsky (); Rothman (); Albirini et al. (). For example, Albirini et al. () recorded Egyptian and Palestinian heritage speakers with English as the dominant language and found up to .% of verb agreement errors and up to % nominal agreement errors.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

     



Speakers experienced particular difficulty in creating agreement chains between the verb and plural or feminine subjects, as well as where more arcane rules of grammar dictate, for example, singular agreement with post-verbal conjoined subjects (p. ).2 Another example comes from Hungarian, where the possessed noun morphologically agrees with the possessor in person and number. Bolonyai () reports an error rate of % for this property among Hungarian speaking children in the US between the ages of  and . She explicitly argues for the selective nature of morphological variability. Erosion of agreement is particularly striking given that other verbal categories seem to be more resilient, but is predicted by the BH. It is particularly striking that high percentages of agreement errors occur in null subject languages (Arabic in this case), where the agreement affix identifies the missing subject. The lack of semantic significance explanation presents a good account of the Hungarian agreement attrition. Since the possessed and the possessor are both present in the noun phrase, dropping the agreement morpheme does not incur communication loss. In addition to agreement, other functional categories morphologically expressed on the verb also create difficulties. For example, aspect vulnerability has been studied widely in heritage Russian, because aspect is such a prominent feature of the language and because it is intricately linked to lexical knowledge in a complex way (Slabakova, ). Russian grammatical aspect is a functional category encoding purely grammatical features but their expression is an amalgamation with derivational morphology. Perfective-marking prefixes comprise some eighteen to twenty derivational morphemes, and prefix–verb combinations have to be learned individually. Among many others, Laleko () described aspectual errors that heritage Russian speakers made in free oral production, and while they can be classified as aspectual errors, they also constitute wrong lexical choices. In order to understand the process, the reader must keep in mind that lexical aspect is a semantic property of verbal predicates including the telicity distinction. Unlike lexical aspect, grammatical aspect is expressed through inflectional morphology on the verb (Smith, ).3 Polinsky (, a, ) proposed the Lexical Aspect Hypothesis (see also Pereltsvaig, ), according to which verbal aspectual morphology (i.e., perfective prefixes) in heritage Russian encodes lexical, rather than grammatical aspect distinctions: ‘most verbs become either lexicalized perfectives or lexicalized imperfectives’ (Polinsky, , p. ). This functional lexicon reanalysis predicts that telic verbs will mostly appear in perfective forms while atelic verbs will surface as imperfectives. These predictions were supported by a vocabulary translation experiment (Polinsky, ) and by a forced production task (Laleko, ). The BH would explain this lexicalization process by the close link between the lexicon and functional morphology, and the need for these particular functional morphemes to be learned one by one with their appropriate verbs (Slabakova, ). Early attriters simplify their lexicon by fusing the prefix and the verbal root. Even if

2

It is also interesting to note that there were significant differences among the heritage Arabic speakers. Egyptian Arabic speakers showed a greater loss of agreement than Palestinian Arabic speakers. This divergence may have both structural and sociolinguistic explanations. 3 To eat a sandwich is a telic VP because it has a potential endpoint, while She was eating a sandwich is a clause encoding imperfective viewpoint aspect, since the speaker focuses attention on the process, not on the endpoint of the event. English sentences mark both lexical aspect (telicity) and viewpoint aspect.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi





attriters know that a prefixed verb signals a complete event and treat the prefix as separable from the root, they still might produce the wrong prefix for any verb. Staying among verbal inflectional distinctions, it is to be noted that some are much more vulnerable than others. As discussed above, the challenge of Russian aspect is based on its lexical rather than grammatical difficulty. Frequency of usage may be another factor explaining differential difficulty. Let us compare tense with subjunctive mood in Romance, which differ in frequency of occurrence. While both grammatical features are expressed with complex paradigms, tense occurs in every sentence whereas subjunctive mood is rarer and needs a specific context (Iverson et al., ). Not surprisingly, while mood distinctions are subject to significant attrition (e.g., subjunctive in heritage Spanish), tense distinctions are quite robust, even in a weak heritage language (Montrul, ). Next, we shall look at nominal morphology, which does not have much semantic effect on the interpretation of the sentence: gender, case, and differential object marking (DOM). In some languages, such as Slavic, Germanic, and Romance, all nouns are categorized in lexical classes of grammatical gender. Note that this categorization does not carry any informational function, because most nouns refer to inanimates, so gender marking is not linked to natural sex but is purely grammatical. Gender represents another vulnerable feature illustrating the lexical burden of functional morphology, both as noun categorization and as gender agreement within the noun phrase. Montrul et al. (a) found agerelated effects in the marking of heritage Spanish gender agreement, while Polinsky (b) discovered that heritage Russian speakers had restructured the formal cues used to assign Russian nouns to gender classes. In the latter study, this restructuring was related to general proficiency in Russian. Case marking also displays attrition and loss, especially in constructions that allow optionality, such as the genitive of negation in Russian (Leisiö, ). DOM in Spanish, a grammatical phenomenon with complex semantic distinctions (see examples in later sections), is frequently deficient among heritage speakers (Montrul & Bowles, ). In sum, early attriters are challenged by inflectional morphology when it is expressed by large paradigms involving affix learning and online assembly (agreement), when marking a grammatical category implicates a lexically fixed affix–verb combination (aspect) or individual exponence (gender) and when the distinction is rarely used (mood). In contrast to difficulties with functional morphology, early attriters rarely have issues with basic syntax, such as word order in simple clauses or verb movement. There is not much work on low-level syntactic properties because researchers focus their attention on much more complex syntax, but at least some evidence on accurate verb raising in heritage Spanish is available (Bruhn de Garavito, ). Among the complex syntactic characteristics most commonly considered in current research are A-bar movement regularities4 (O’Grady, Lee, & Choo, ; O’Grady, Kwak, et al., ; Lee-Ellis, ; Polinsky, ), wh-movement (Montrul et al., b), and binding (Kim et al., ; Keating et al., ). An interesting case in point is presented by Montrul et al. (b), which studied a range of interrogative sentences and extractions out of main and embedded clauses. The researchers tested early bilinguals at three proficiency levels and compared them to similar

4 A-bar movement involves extraction of a phrase from its position of merging and case marking to another position in the sentence, usually to the left periphery of the sentence.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

     



proficiency adult-onset L learners of Spanish. Both the L learners and the heritage speakers demonstrated solid knowledge of constraints on wh-movement, even in constructions where transfer from English would lead the speakers astray. The only area where they differed from native speakers was subject and object extraction from embedded clauses. These findings were interpreted as pointing to correct underlying representations in all L learners and heritage speakers but difficulties in processing such long sentences. The general conclusion of this and comparable studies is that basic syntactic operations are not vulnerable to attrition even in early bilinguals, while their reduced practice of the heritage language may lead to superficial processing deficiencies. How about knowledge of semantics and appropriateness in discourse context? The BH predicts that comprehension resulting from semantic composition of already established grammatical features and constructions is not going to be vulnerable. Studies on such properties are rare, but they do exist. Leal Méndez et al. () tested whether heritage speakers at two proficiency levels and adult L-ers recognize the meaning and appropriate context of clitic left-dislocations (CLLD) and focus fronting in Spanish. The English equivalent to CLLD, topicalization, does not double the dislocated object with a clitic. CLLD is a very frequent feature of Spanish. The specific semantic constraint tested was the semantic freedom of the antecedent–dislocate connection: the dislocated clitic-doubled constituent can be identical, a subset or even a part of the discourse antecedent. This semantic freedom is universal for all topicalizations, hence the BH would predict that early attriters would not have any problem with comprehending it. The example below illustrates the part–whole relationship between the dislocated constituent (the table legs) and the antecedent (the table). () Context: What shall we do with the table? It is too big! Mira, las patas, las doblas así . . . look the legs Cl. fold..sg this-way ‘Look, you can fold the legs like this . . . ’

Part/whole

While attriters might make errors with the grammar of CLLD by not providing the cliticdoubling, they should not be less accurate on sentences such as () compared to sentences in which the dislocated element and its antecedent are identical. A felicity judgement task ascertained that the learners and heritage speakers were no different from the monolingual natives in their comprehension of this semantic freedom. The Interface Hypothesis (IH; Sorace, ) was initially proposed to address language attrition and later expanded to other bilingual conditions (see also Chamorro & Sorace, Chapter , this volume). The predictions of the IH and the BH differ: whereas the former hypothesis sees most challenge at the interface of grammar and external modules such as context, the latter hypothesis places the main burden of difficulty on the functional morphology. While there is substantial evidence for the IH coming predominantly from null and overt subject usage, a range of experiments on other interface constructions has not encountered heritage speaker deficiencies, challenging its predictions. For example, the same experiment cited above, Leal Méndez et al. (), in addition to checking a semantic constraint, also established that the heritage speakers had no problem with choosing the context-appropriate construction with the clitic-doubling. Furthermore Leal et al. () looked at a very rare construction, clitic right dislocation, which is only acceptable in highly

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi





circumscribed contexts such as the affective context illustrated below. Note that the object is right-dislocated in (a), but is doubled by a pre-verbal clitic: () Context: Mariana and Omar are siblings and they inherited several pieces of furniture after their father died. Because Mariana has a larger house, she kept the majority of them. The problem is that many pieces of furniture were damaged, and Mariana initially didn’t know what to do with them. One day, Omar goes to visit her; he thinks he recognizes a chair in the living room. O: Veo que por fin decidiste arreglar la silla de papá. ‘I see that you finally decided to fix dad’s chair.’ (a) M: No. La regalé, la silla odiosa esa. Lo que arreglé fue el sofá que tanto me gustaba. no CL I-gave the chair ugly that. CL that fixed was the sofa that so me pleased (b) M: No. # Regalé, la silla odiosa esa. Lo que arreglé fue el sofá que tanto me gustaba. no I-gave the chair ugly that. CL that fixed was the sofa that so me pleased ‘No, I gave that hideous chair away. What I did repair was the sofa that I liked so much.’ The experiment used a judgement task where speakers had to rate the felicity of (a) or (b) in the given context. Results established that Spanish-dominant and heritage bilinguals were equally selective with the contexts in which they allowed clitic right dislocation. No group differences were in evidence. The researchers argued that these findings challenge the IH, at least in offline tasks where there are no time constraints. To bring all these disparate threads together, the comparison between early attriters’ knowledge of functional morphology, basic syntactic operations and semantic composition constitutes support for the BH, in the sense that functional morphology is much more vulnerable to attrition than the latter two areas of the grammar. However, in summarizing apparent knowledge of different properties, it also emerged that grouping linguistic properties and their acquisition on the basis of gross modules may be over-simplistic. Some functional morphology such as agreement seems to be more vulnerable than other functional morphology such as tense marking, with various features in between these two extremes. In other words, the functional morphology is not monolithic but can be selectively attrited. A tentative explanation based on the grammatical status, complexity of the grammatical features, and the evidence for their expression runs like this: If a grammatical feature does not affect meaning-making and comprehension, but captures purely syntactic effects (such as subject–verb agreement and DOM), then this feature is vulnerable. Seemingly redundant morphology is not retained in heritage grammars but simplified, possibly leading to a restructuring of the overall grammar (see also Seliger, ). If a piece of functional morphology has a semantic or discourse-pragmatic effect and the evidence for it is rarely seen in the language, then that feature may also be vulnerable (mood, DOM). If functional morphology is conflated with derivational morphology as in Russian aspect, it is likely to be restructured. One-to-one form–meaning mappings that are clearly and abundantly evidenced in the linguistic input (such as tense marking) endure the best. This explanation is compatible with the spirit of the Interpretability Hypothesis (Tsimpli & Dimitrakopoulou, ) because it elevates meaning-related features as more resilient than form-related ones.5 The Interpretability Hypothesis (Tsimpli & Dimitrakopoulou, ) has been proposed to explain the inability of L learners to reach native-like levels of competence. The hypothesis builds on the 5

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

     



.. The grammatical competence of late attriters Let us now turn our attention to later attriters, the immigrants who moved to the country of their L well after puberty. Studies documenting attrition of functional morphology among later attriters are rare, but do exist. For example, Ribbert & Kuiken () documented attrition of the infinitival complementizer among Germans who moved to the Netherlands as university students. There is both overlap and difference in the use of Dutch and German infinitival complementizer um/om: Dutch om is a superset of the contexts where German um is used. Furthermore, Dutch has a high number of contexts where om can be used optionally. The German speakers were asked to rate the grammaticality of pairs of sentences with and without a complementizer. Their preferences differed significantly from other German speakers who were not in contact with the Dutch language. Still staying within functional morphology and its interpretation, selective attrition is attested in late attriters, just like in early attriters. Gürel (a) provides a very interesting example. She looked at a rare morpho-syntactic phenomenon with semantic effects: the binding of overt, null, and reflexive pronouns in Turkish embedded clauses as exemplified in (a) below. The overt third person pronoun o, when it is embedded clause subject, cannot refer back to the main clause subject, while the null pronoun (pro) can. The English overt pronoun he in (b) has no such restriction. In the examples, the sub-index ‘i’ refers to the main subject while ‘j’ means that the pronoun has some other discourse-provided antecedent. (a)

Buraki [o-nun*i/j / kendi-si-nini/j / proi/j zeki ol-duğ-u]-nu Burak s/he- self-- pro intelligent be--- düşün-üyor think- ‘Buraki thinks that [he*i/j / selfi/j / proi/j is intelligent]’

(b)

Buraki thinks that hei/j is intelligent.

In two interpretation tasks where speakers had to choose between the pronoun antecedent being the main clause subject and a third person not mentioned in the sentence, native speakers overwhelmingly chose the third person, the only available interpretation. However, Turkish attriters living in the USA and Canada chose the wrong antecedent between % and % (on the different conditions). They still preferred the third person as antecedent, choosing it above %; however, their knowledge in this respect was evidently shaky. In contrast to the overt pronoun o, the attriters maintained their knowledge of the nominative reflexive kendisi and the null pronoun. It is also interesting to note that this selective attrition is parallelled exactly by selective acquisition among near-native L learners of Turkish (Gürel, ). It appears that the null pronoun and kendisi are acquired completely and resist attrition, while the antecedent choices of o in embedded subject position create problems for both near-native speakers and late attriters. Binding is among the most complex properties of language, relying on syntactic features captured by Minimalist Program interpretable–uninterpretable feature distinction and postulates that interpretable features are acquirable in the L while purely syntactic uninterpretable features are not, if they are not already established in the native language and thus transferable. Note also the explanation provided above is in contrast to that proposed by Tsimpli et al. (), which gives rise to the IH.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi





functional morphemes with semantic import, and frequently discourse-pragmatic import as well. The prohibition of main clause antecedents for the overt pronoun o, but not for the other two pronominal elements, is eroded under the influence of English where such prohibition does not exist. In one sense, this restructuring is a simplification of the grammar. In checking the predictions of the BH, it is informative to look at Schmid (), a study comparing a number of late attriters with very advanced L speakers, matched in proficiency and looking at a number of linguistic properties. The attriters were German native speakers living in Canada, while the advanced L learners and controls were living in Germany at the time of testing. The data comes from two spontaneous speech samples per subject, a film retelling task, and a semi-structured autobiographical interview, in which errors were calculated per , words. The range of functional morphemes tested included case, gender, plural, articles, pronouns, tense morphology (the weak/strong distinction), auxiliaries, and subject–verb agreement; while syntactic properties included verb-second and discontinuous word order, as well as word order in subordinate clauses. Schmid reported the error counts for each speech sample separately (her table , p. ). The late attriters were non-distinguishable from the monolingual native speakers for all of the properties tested, functional morphology and syntax alike. The BH cannot offer an explanation for their linguistic performance just because there is nothing to explain: these German immigrants to Canada did not demonstrate any attrition. In contrast to the findings in Schmid (), Montrul et al. () did attest attrition of functional morphology. This a large-scale study of DOM examined in heritage speakers and their parent generation, thus early and late attriters, of Spanish, Hindi/Urdu, and Romanian. DOM is among the most complicated aspects of functional morphology, where object marking depends mainly on animacy and specificity, but aspect, topicality, agentivity, and affectedness may also influence it. The example in () comes from Hindi. () Mira-ne ramesh-ko dekhaa. Mira-erg Ramesh- saw ‘Mira saw Ramesh.’ (*Mira-ne ramesh dekhaa.) The three languages under investigation work largely (although not entirely) in parallel with respect to DOM’s semantic restrictions, but while the Spanish marker is a (not audible after a vowel), Hindi and Romanian markers are syllabic (ko and pe, respectively), so better audible. The findings are complicated. The heritage speakers accepted ungrammatical DOM omission with animate and specific objects to higher degrees than they accepted omission of indirect object or locative marking. Thus, they were simplifying their grammar in the direction of redundancy elimination (compared to a locative, DOM is largely redundant, as context certainly distinguishes specific from non-specific objects). Thirteen (out of twenty-one; %) late attriters of Mexican Spanish had lost DOM in the USA, while Hindi and Romanian speakers were not attrited for this property at all. While these results are difficult to explain straightforwardly, the linguistic complexity of the phenomenon is the prime suspect. In addition to the functional morphology being largely redundant, the markers are multi-functional in the sense that they express other grammatical meanings as well, producing an ‘opaque form-meaning mapping’ (Slabakova, ; O’Grady et al., ). The phonological salience mentioned above cannot be the sole

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

     



factor explaining the group differences. The authors make a convincing argument that some language-internal structural factors such as clitic-doubling in Romanian and lack of definite articles in Hindi may be a cause of differences. Ultimately, DOM markers have to be considered as functioning in tandem with the rest of the grammar and being affected by the holistic linguistic experience. The authors conclude that ‘the structural properties of the DOM markers together with the syntax of definiteness and specificity in each language (more than external factors) seem to account for the degree of DOM erosion in each language’ (Montrul et al., , p. ). Turning to knowledge of syntax and of interpretation now, the findings of an influential study, Tsimpli et al. () are worth reviewing here. The study investigated production and comprehension of overt and null preverbal and post-verbal subjects among late attriters of Italian and Greek living in an English-speaking environment. A robust finding was that the syntax of subjects and interpretation of null subjects were not attrited, while overt subject interpretation choices were more vulnerable because they depended on consideration of discourse-pragmatics as well as syntax. These and other findings were conceptualized as the IH (see more on this hypothesis in Chapter ). Another recent study, Flores (), compared knowledge of syntax and discourse in four age groups of German-Portuguese bilinguals living in Portugal, for whom German was the minority language.6 Flores’ subject pool allowed her to make very precise distinctions based on age of loss of contact with the attriting language. She examined a narrow syntax property, verb placement in main and embedded clauses, and a syntax–discourse interface property, object expression under topichood. The speakers who had lost contact with German before the age of  exhibited around % errors on both properties. Those who had moved to Portugal between the ages of  and  were markedly different: while their errors on topic object drop were in the % range, their word order errors were negligible. The third group of adult returnees (at ages –) and the group of children who were still in contact with German made no errors on either property. The author concluded that narrow syntax was more likely to suffer from attrition if contact with German was lost before age  to , while object expression continues to be vulnerable to reduced input until the age of . Late attriters did not exhibit any attrition of these properties. Let us take stock of the main findings for late attriters. While attrition of grammatical features is much rarer among this population (Schmid, ), some functional morphology does suffer attrition (Ribert & Kuiken, ; Montrul et al., ). Purely syntactic properties such as verb placement (Flores, ) and the syntax of null and overt subjects (Tsimpli et al., ) are much more robust and do not appear to undergo attrition. Finally, interpretive knowledge seems to be vulnerable only if it is calculated at the syntax– discourse interface (Tsimpli et al., ). At the same time, Flores’ () adult attriter group did not demonstrate such vulnerability. The BH is largely borne out by such findings. If anything is vulnerable in the grammars of late attriters, it is the functional morphology and not pure syntax. The additional factors we saw at work in early attriters, namely linguistic complexity of the property under investigation, redundancy of the functional morphology marker, opaqueness of form–

6 Flores’ child control group was the exception, because these subjects were still living in Germanspeaking countries.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi





meaning mapping, as well as usage frequency hence activation, go a long way to explaining most of the findings we discussed in this section. At the same time, the hypothesis, as it is formulated now, may be too large-grained to provide a sufficient explanation of particular selective attrition, such as the one discovered by Gürel (a), where a restriction of one pronominal form is relaxed, or by Montrul et al. (), where one group of late attriters is affected but not others.

. C

.................................................................................................................................. Examining findings from early and late attrited speakers invites fundamental questions on what the grammar of these speakers look like. In other words, what are the linguistic representations that allow not altogether erroneous, but ‘shaky’ or ‘rusty’ performance on some properties? The BH is useful in identifying the essential grammatical complexity of the functional morphology, especially with many-to-many features-to-form mappings, as the main reason. In comparison, the core syntactic operations and the compositional principles of putting the meaning of a message together are resilient in the face of suboptimal input, just because they are universal and are used in the dominant L as well as in the affected L. While the BH addresses the issue of why certain properties are differentially vulnerable in attrited grammars, a related important question is how this state of grammatical knowledge operates in individual speakers’ mind/brain. Reminiscent of Yang’s () variationist learning approach, a recent generative proposal, Amaral & Roeper (), specifically endorses and explains the mechanism of how all the grammars of a multilingual can arise and function in tandem. ‘[A]ny language contains properties of several recognizable language types, i.e. the grammar of a language L can have elements that form sub-grammars compatible with L, L, Ln’ (Amaral & Roeper, , p. ). According to this view, L/Ln acquisition is a natural proliferation of sub-grammars, a process of grammar expansion that is certain to affect the native language right back. Multilingual linguistic competence is an amalgamation of sub-grammars coming from the previously acquired languages, equipped with some sort of differentiation mechanism, such as a tagging device for the rules and lexicon of grammar A or grammar B. If one rule, in the attrition situation the L rule, is used much more intensively, its activation level is likely higher in the sense that it will be invoked much faster and with less effort. The L rule is not truly gone and it is still tagged as a L rule, but summoning it may be more effortful and costly. This is essentially a processing view of grammar usage (see also Sharwood Smith & Truscott, ; and Sharwood Smith, Chapter , this volume), which explains how grammatical rules, while they are in place, may appear ‘rusty’ and ‘shaky’ in performance.

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

  ......................................................................................................................

                                 ......................................................................................................................

 

. P    L  

.................................................................................................................................. I is fair to say that it has become standard practice in L attrition research to collect information on background variables, for the purposes of describing and homogenizing research groups, on the one hand, and to assess the extent to which these variables may influence the outcome of attrition processes, on the other. While certain variables, such as gender and age of bilingualism, are reported in most studies, Köpke & Schmid () recommended that L attrition studies minimally take into account ‘age at onset of L acquisition (simultaneous/early/late bilingual); age at onset of L attrition (pre‑/post puberty); time since onset of attrition; level of education; attitudes; frequency, amount and settings of use of the attriting language’ (pp. –). This proposal was taken up in the work of the Graduate Research Network on First Language Attrition (henceforth the Network; Schmid, b). The Network sought to strengthen the empirical basis of L attrition research by developing a set of instruments, the Language Attrition Test Battery (LATB; Schmid, ; cf. https://languageattrition.org/), for use across diverse language contexts, theoretical frameworks, and research objectives. The LATB included a sociolinguistic questionnaire, which elicited data on the following variables and was deployed in a number of studies (Keijzer, ; de Leeuw, ; Dostert,

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi





; Cherciov, ; Lubińska, ; Opitz, a; Varga, ; Yılmaz, ) (variables in italics were included in a subgroup of studies): • Personal background: gender; citizenship; marital status; number of children/grandchildren • Education and socio-economic status (SES): highest educational level; professional/ vocational training; previous and current jobs • Migration history: age at migration/arrival/onset of attrition (AOA), length of residence (LOR), age at testing (AaT) • Language learning history: number of languages (L1(s) +); age of acquisition/learning; variety of L1(s) spoken; motivation for acquiring/learning additional languages; past maximal and present proficiency (L1(s) +); language learning aptitude • Language use: frequency of use with family members (partner, children, parents) and friends; mode of use (face to face/telephone/Skype/letters; directionality); size of network; use in various domains (church, expat networks, shopping, pets, professional use), engagement with various media (TV/radio, newspaper, tapes/CDs/ videos/DVDs) • Language attitudes: towards intergenerational transmission (encourage children, correct children, regrets, relevance for contact with grandparents); towards language and culture of both communities (Attitude and Motivational Test Battery); towards own bi-/multilingualism The Network and related studies treated the elicited background variables as potential predictor, or independent, variables (IVs) for language proficiency outcomes (dependent variables, DVs), some taking a more exploratory approach, and others more specifically testing particular hypotheses concerning the relative importance of different variables. Thus, Bylund studied the impact of ‘age of reduced L contact’ (Bylund, , a,b) as well as language aptitude (Bylund et al., ; Bylund & Ramírez-Galan, ) in his maturational investigation of L attrition. Other studies focused on AOA, language use/ contact, aptitude, and attitude (e.g., Keijzer, ; Opitz, , a; de Leeuw et al., ; Lubińska, ; Cherciov, ; de Leeuw, Mennen, & Scobbie, ). Attempts to find meaningful correlations between IVs and DVs generally did not lead to conclusive results, with few variables emerging as definite predictors in their own right, and large parts of variability remaining unaccounted for (Schmid, Chapter , this volume; Schmid et al., ; and for L attrition, Mehotcheva & Mytara, Chapter , this volume). Schmid writes: ‘So far the findings from quantitative investigations of L attrition and external factors appear somewhat unsatisfying, and often contradictory’ (a, p. ). While maturational variables are sometimes viewed as the single indisputable predictor (Ammerlaan, , p. ; Köpke & Schmid, , p. ; Bylund, a, p. ; Schmid et al., , p. ), even they do not ‘predict’ attrition outcomes in the strict sense of the word, since participants with similar language biographies show substantial variability in the degree with which they succeed in maintaining their L (Schmid, a, a). Instead, variables that were assumed to be crucial, such as education level or contact, turned out to be considerably less influential than expected, or at least not to confirm simple cause–effect relationships (Schmid et al., ; Opitz, b). For example, Schmid (a) discusses two participants who despite a ‘nearly identical’ background showed ‘dramatic differences [ . . . ]

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

     



in their linguistic data’, and concludes that in order to account for these differences we ‘have to look elsewhere’ rather than the background variables (p. ). It is reasonable to assume that the elusiveness of strong correlations and significant effects may in some part be linked to methodological issues, such as the specifics of the samples drawn, and difficulties in operationalizing and reliably measuring certain variables (e.g., language use; Schmid, ). As mentioned above, the lack of comparable designs amongst the extant empirical studies had in fact been the motivation for elaborating a common methodology within the Network. Thus, in order to avoid the problematic distinction between L attrition and nonacquisition, many studies, including the Network studies, investigate adult migrants who are well past puberty, thereby limiting the potential impact of AOA. Other variables, for example SES and education level, are also often controlled in participant recruitment, to ensure the homogeneity of the population and in view of the particular elicitation tasks planned during an investigation. Indeed, Schmid (a) advises: It is therefore of vital importance that you try to eliminate any impact of those predictors which you do not want to investigate. That is, there should be as little variation as possible along those factors in your sample. (p. )

This, necessarily, constrains the potential explanatory power of these variables. With single factors failing to produce unanimous results, and some of the variables, for example SES, gender, and education level, apparently inextricably confounded, some authors suggested that it might be profitable to look at clusters of variables (e.g., Köpke, ; Opitz, b, a) or ‘compound factors’/‘composite determinants’ (Schmid & Dusseldorp, ) and covariation. As an additional consideration, clustering variables also has a practical benefit in that the required number of participants is lower than if the variables were studied individually. This line of research has been somewhat more fruitful, in that some variables do seem to show an effect in the presence of another variable. De Bot et al. () and Schmid (b), for example, studied the interplay of LOR and frequency of L use. In de Bot et al., LOR was a factor only for participants with little contact with the L. Schmid, on the other hand, found LOR to be a predictor both for participants with the least, and those with the highest amount of contact, which she interprets as evidence for different sources of attrition for the two sets of participants— that is, reduced access in the former, and exposure to a changed variety in the latter. Schmid & Dusseldorp (), in their multivariate study of extralinguistic factors, found evidence for ‘several reliable factor groups’ (p. )—that is, sets of variables with high internal consistency, categorized as: (a) general factors (biological age, sex, education level); (b) factors pertaining to bilingual populations (age at onset of bilingualism/at emigration, attitudes and (ethnic affiliation); and (c) factors specific to attrition (contact with and use of the L). However, these factor groups were shown to have relatively weak predictive power. While studying variable clusters is undoubtedly a promising avenue, the findings by de Bot et al. () and Schmid (b) point to an alternative explanation for the difficulties in establishing meaningful correlations between background variables and the outcomes of attrition. This proposal rests on the nature of language as a complex dynamic system, and, as this chapter argues, consequently on the properties and interactions of the background variables themselves (Opitz, b).

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi





At a very basic level, some variables, such as age, education, or language background, are relatively fixed or change in a predictable measurable way, while others, such as language use and attitudes, are conceivably quite variable over time. This renders conventional analyses rather problematic. In addition, the findings by de Bot et al. and Schmid reveal non-linear interactions between variables, as theorized in Complex Dynamic Systems Theory (CDST), which also pose a challenge for traditional, means- or regression-based, statistical analyses. In the following subsections, I will present a CDST view of the challenges associated with studying background variables in L attrition.

. C D S T   

.................................................................................................................................. CDST was initially introduced to the study of second language acquisition (SLA) in the guises of Chaos/Complexity Theory (CT) by Larsen-Freeman (e.g., ; Cameron & Larsen-Freeman, ), and Dynamic Systems Theory (DST) by Van Geert (e.g., , ) and de Bot and collaborators (e.g., de Bot, ; de Bot et al., a,b), respectively. It is now agreed that despite CT and DST’s differing origins and emphases, their tenets and implications are sufficiently compatible for the purposes of research into language development to refer to both theories jointly as CDST (Larsen-Freeman, ), a convention adopted in this chapter. The advancement of CDST to the study of language is directly related to the failure of traditional approaches to provide more convincing answers for how language develops and changes. As Larsen-Freeman et al. () put it, CDST responds to the ‘general unease with neat, straightforward, linear and monodimensional solutions’ which is discernible in much scientific endeavour in applied linguistics and beyond, and ‘the recognition that reality is usually messy and often unpredictable, since it is defined by multiple, diverse and interdependent factors’ (p. ). Later on in the text, the authors, in view of the available evidence from empirical research, argue that ‘[t]he traditional scientific method, which is based on analysis, isolation, and the gathering of complete information about a phenomenon, is incapable of dealing with [the] complex interdependencies’ present in language development (p. ) and that it ‘might be time to consider a radical change of scientific paradigm’ (p. ). The core of this changed perspective consists of moving away from the (natural and at times warranted) desire to reduce complexity (Larsen-Freeman et al., ; LarsenFreeman, ), towards embracing the interconnectedness, dynamism, and non-linear interaction of elements within the language system, and of the system with its environment. Complex dynamic systems are in constant flux, yet generally stable, continually adapting to and creating new conditions through self-organization and emergence (Cameron & Larsen-Freeman, ). Language development is viewed holistically as being shaped by ‘interrelated patterns of experience, social interaction, and cognitive processes’, and language acquisition, use, and change are considered different ‘facets of the same system’ (Becker et al., , p. ; see also Ellis, ). This view implies that L and other languages may be considered embedded sub-systems operating within the same context and under the same constraints. It also implies that

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

     



attrition is as much a feature of language development as is acquisition. Or put differently: ‘Linguists who convincingly want to account for the enigma that is human language have to develop models and explanations that are able to accommodate incomplete, transitory, unstable, changeable and fossilized languages’ (Larsen-Freeman et al., , p. ). A first formal description of language development including L attrition from a CDST perspective was presented in Herdina & Jessner’s () Dynamic Model of Multilingualism (DMM) (see also Jessner, ; Herdina & Jessner, ). But, having been ‘instrumental in placing [language attrition] on the bilingual map’ (Larsen-Freeman et al., , p. ), it was de Bot in particular who made the case for CDST’s applicability to attrition research, arguing that development entails both growth and reduction, which are affected by the same forces and constraints (e.g., de Bot, , ). In , a special issue of the International Journal of Bilingualism on ‘Dynamics of First Language Attrition across the Lifespan’ (Opitz et al., ) presented empirical research from a dynamic perspective. In their introduction, de Leeuw, Opitz, and Lubińska () discuss the propositions that ‘L attrition /as such/ is evidence for the view that L develops dynamically across the lifespan’ and that ‘L attrition itself is dynamic’ (p. ). In the same issue, Schmid et al. () discuss methodological issues arising from CDST for attrition research, while Herdina & Jessner () reflect on the implications of attrition for CDST. Attrition is thus ‘negative’ or inverse language growth (Jessner, , p. ), which becomes noticeable in the absence of conditions that would normally support the acquisition or maintenance of a language, such as input in that language. As a result of the interconnectedness of the components of the language system, adaptations to environmental or internal changes in one part of the system potentially have consequences in other parts and at whole-system level. For example, increased use of a language in a new linguistic environment may lead to retrieval difficulties in other languages due to their inhibition and lowered activation level. This strongly suggests a need to consider language attrition vis-à-vis the development of the overall system. However, the bulk of L attrition research to date, with the exception of Opitz (a, a), Cherciov (), and Yılmaz (, Chapter , this volume), has studied changes in the attriting language only, without much recourse to the other languages, which may be an additional source of the divergent findings. While it is recognized that the future behaviour of a complex dynamic system ‘sensitively’ depends on the initial conditions prevailing at a given time (de Bot et al., , p. ; de Bot & Larsen-Freeman, , p. ), such adaptations are neither linear (i.e., they may be disproportionate to an initial impact) nor directly causal since self-organization gives rise to emergent, qualitatively novel, properties (de Bot & Larsen-Freeman, , pp. –). Moreover, the variables that are associated with such adaptations are themselves subject to complex interactions with a multitude of other variables, and each variable’s impact is thought to change over time (Larsen-Freeman, , p. ). For language development this implies that, on the one hand, the proficiency level in a language matters for the subsequent speed and developmental trajectory of that particular language, and the language system overall. This is potentially observable in differential rates of attrition in first versus second language attriters, mature versus developing first-language attriters, though inter-individual variability even within groups is a common finding, as shown below. On the other hand, non-linear cause–effect relationships are at play when, for example, language learners experience some feature of that language suddenly and seemingly without effort ‘falling into place’. The non-linearity of changes in individual language systems, and

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi





their vulnerability to minor changes, was also demonstrated in Meara’s () computer simulation of language attrition (see also Larson-Hall, Chapter , this volume). These characteristics of complex dynamic systems lead to ‘variation both in and among individuals’ (de Bot et al., ) and ‘at all levels of linguistic organization’ (Cameron & Larsen-Freeman, ; Becker et al., , p. ). Inter- and intra-individual variability are indeed a robust finding in both acquisition and attrition studies, with occasional extreme examples like the two participants described in Schmid’s (a) study. As discussed in the next subsection, it has been argued that it is this variability that should be the proper concern of researchers studying language development. Ultimately, these properties also lead to an inherent unpredictability of the outcomes of change processes, which means that research from a CDST perspective is more concerned with retrodiction—that is, the attempt to ‘predict’ a previous state from the present (Cameron & Larsen-Freeman, ; Verspoor, de Bot, & Lowie, ). Notwithstanding the intrinsic dynamism of complex systems, language systems do not develop entirely randomly, with different periods showing more or less stability and change. Moreover, there is a considerable amount of evidence of patterns in language development, such as U-shaped development, which is compatible with CDST tenets (Cameron & LarsenFreeman, ; Lenzing, ). A relevant concept in this context is the notion of ‘attractor states’. An attractor state has been defined as ‘a critical value, pattern, solution or outcome towards which a system settles down or approaches over time’ (Newman, , cited in Hiver, , p. ). Metaphorically speaking, they are ‘pockets’ of (relative) stability within the topography of ‘state space’, the hypothetical landscape of all the possible outcomes of a complex dynamic system (Larsen-Freeman & Cameron, ; Hiver, , p. ff ). Since language systems are open and dynamic, such stability is never absolute—there may be ‘stability but never stasis’ (Larsen-Freeman et al., , p. ). Importantly, these developmental patterns are neither brought about by external forces, nor by a central agency, nor do they in themselves exert any ‘attraction’ or bias for developmental outcomes. Instead, they result purely from the self-organization of the system (de Bot & Larsen-Freeman, , p. ), in response to inputs including feedback and perturbations, such as changes in linguistic environment, and increased or decreased opportunities to use a language. Developing systems overall are thought to follow positive and negative growth curves akin to those modelled for biological growth under constant conditions (Herdina & Jessner, ). Such growth curves entail slower growth at the extreme ends of the acquisition and attrition processes, where adaptions to external and internal variables are initially absorbed by the system, before the process ‘takes off ’, as it were. These adaptations become noticeable as ‘increased scatter’ or optionality in performance, before qualitative changes occur in the system and a different stage is reached (Herdina & Jessner, , p. ff; Larsen-Freeman & Cameron, ). Indeed, high variability has been associated with subsequent learning (Van Dijk et al., ). When viewed close up and over short intervals, the developmental trajectories for individual properties vary from one property to another, and can appear rather erratic. However, at larger timescales and higher levels of granularity, what may have looked like a negative trend (e.g., more errors on a particular feature) may in fact be part of a generally positive change (Van Geert, ; de Bot, , ). Conversely, a system that appears to be stable at a certain point may in fact be experiencing internal restructuring which ultimately becomes visible as changed proficiency.

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

     



. C D S T   

.................................................................................................................................. As we have seen, CDST emphasizes the interconnectedness of language subsystems within the overall system, and their context-dependence, the perpetually changing, variable, and emergent nature of these (sub-)systems, as well as the non-linearity and basic non-predictability of developmental processes. These properties present a considerable problem for theory and empirical research in language acquisition and attrition since they run counter to some long-standing traditions of doing research. These, in their quest for generalizability, have tended to ignore individual variability, traditionally seen as ‘noise’ (Thelen & Smith, ; de Bot et al., a; Cameron & Larsen-Freeman, ), and resorted to idealization. CDST in particular critiques traditional research designs and conventional statistics. The cornerstone of SLA research has been a reliance on (generally) cross-sectional investigations comparing (ideally large) groups of participants, while the statistical procedures applied to data serve to neutralize differences between individuals through averaging. However, as Larsen-Freeman (e.g., , , ) and others have pointed out, group averages do not allow researchers to make inferences about individuals and therefore have limited value in understanding actual language development (see also Verspoor & Van Dijk, ; Lowie & Verspoor, ). Moreover, most commonly used statistical tests rely on problematic assumptions concerning the properties of samples and data (normal distribution, independence, homoscedasticity) as well as the nature of the relations between variables (i.e., linear and potentially causal). If language is indeed a complex dynamic system, the assumptions of linearity and Gaussian distributions are fundamentally violated (e.g., Larsen-Freeman, ) making the application of conventional statistics inappropriate for many questions (Lowie & Verspoor, ; Larsen-Freeman, ). This does not mean that statistics has to be discarded entirely, since it may be used for questions for which the assumptions hold, but new statistical methods need to be developed that are capable of ‘examining the flexible, transient, dynamic aspects of learner language which emerge from use’ (Larsen-Freeman et al., , p. ). One route has been to improve statistical tests so that they are less affected by violations of assumptions, and to work with data in novel ways. This includes new non-parametric tests (e.g., Anderson, ; Terpstra & Magel, ), Bayesian and Paretian statistics (Andriani & McKelvey, , cited in Larsen-Freeman, , p. ), ‘robust statistics’ (Larson-Hall, ), estimation (e.g., ‘new statistics’; Cumming, ), resampling and simulation methods (e.g., bootstrapping, Monte Carlo), combined with data smoothing techniques (e.g., moving averages). (For overviews and applications in the context of second language development (SLD), see Verspoor, Lowie, & Van Dijk, ; and Verspoor, Lowie, Van Geert, et al., .) Another approach has been a move towards different types of research design and data, in which statistical tests are less relevant (e.g., Larsen-Freeman & Cameron, ). From a CDST perspective, ‘the study of variability as a factor in its own right’ is a paramount requirement (Larsen-Freeman et al., , p. ; Van Dijk et al., ; Verspoor & Van Dijk, ). This, in turn, revalidates the individual or small-scale case study, with change,

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi





variability, and dense qualitative sampling at the centre of attention, and calls for longitudinal studies of language development: For [CDST] researchers, this change in perspective has implied a radical change in the way in which language is regarded, which implied a shift in focus from the internal structure of a closed module to change over time of an embedded and embodied self-organizing system. It also required a shift in the empirical approach, from a focus on the grand sweep of factors affecting the language module to more qualitative, fine-grained longitudinal studies. (Larsen-Freeman et al., , p. )

The current challenge lies in implementing these imperatives in empirical work. Studies into SLD conducted from a CDST perspective include Verspoor et al.’s () reanalysis of a longitudinal investigation of an advanced learner’s L negation; Larsen-Freeman’s () study of the writing and oral production of five Chinese learners; Caspi’s (; described in Verspoor & Van Dijk, ) study of four learners’ vocabulary learning; Polat & Kim’s () case study of untutored language development; and Opitz & Smyth’s () study of writing development in ab-initio learners (see also Opitz, ). An impressive example in the area of L motivation are the studies presented in Dörnyei et al.’s () volume. In the area of attrition, no studies have, to my knowledge, been designed in response to CDST requirements as yet, but a number have adopted a dynamic perspective in their analysis and discussion of results. These include two doctoral dissertations (Cherciov, ; Opitz, a), and empirical contributions by de Leeuw, Opitz, and Lubińska (); Opitz (); Cherciov (); and Ecke & Hall () to the aforementioned special issue of the International Journal of Bilingualism (Opitz et al., ). As Larsen-Freeman et al. () argue, ‘[d]ata-driven variability studies’ and computer simulations have already proven to be ‘powerful tools’; however, they are ‘still in their infancy compared to the wealth of research techniques and statistics available to “old school” approaches’ (p. ). Given that longitudinal research is associated with practical disincentives, such as the time it is necessary to invest and the possible loss of participants, it is unlikely that traditional ways of doing language research will be eclipsed by CDSTinformed research any time soon. Indeed, despite their criticism of reductionism (LarsenFreeman, ), and contrary to some researchers’ perceptions (e.g., Erdocia, ), the proponents of CDST are not intent on waging war, but argue for the complementarity of both approaches: ‘CT/DST researchers should acknowledge the value of more traditional research and traditional researchers should acknowledge the continuity of human cognition and the emerging complexity of language’ (Larsen-Freeman et al., , p. ).

. C D S T    

.................................................................................................................................. The previous subsection has shown that CDST provides a challenge for research design as well as analysis. In view of the concern of this chapter, it also raises the question of what this perspective means for the study of background variables in first language attrition.

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

     



According to CDST, personal background variables, like other system components, are characterized by non-linear trajectories and interactions, as well as unique initial conditions and specific combinations, which lead to the inherent unpredictability of outcomes for the language system. Therefore, viewing variables in isolation and treating them as ‘predictors’ becomes theoretically untenable. Byrne (, cited in Larsen-Freeman & Cameron, ) contends: ‘Let us understand clearly, once and for all, that variables don’t exist. They are not real’, thus declaring ‘death to the variable’ (p. ). The inadequacy of conventional approaches based on the identification of a single factor, such as age of onset or maturational constraints, to determine the outcome of language development within a multilingual system is due to the questionable assumption that single factors can be identified as the determining ones, that they can function as predictors of language development and that the growth and decay of languages within a multilingual system can be meaningfully studied in isolation (Herdina & Jessner, , p. ). In other words, adopting a CDST perspective to attrition requires a fundamental rethinking of the very notion of predictor variables. The problem is compounded by the ‘statistically possible’ (Van Geert & Steenbeck, , p. , cited in de Bot & Larsen-Freeman, , p. ), but theoretically untenable separation of learners (and their variables) from context (the ‘boundary problem’; Larsen-Freeman, , p. ; see also Cameron & Larsen-Freeman, ; King, ). Since every dynamic system is constantly interacting and interfacing with others and with the environment organically, the notions of ‘inside’ a system and ‘outside’ a system are never simple or uncontested. (Cilliers, , p. , cited in Larsen-Freeman et al., , p. )

A further practical problem arises from the fact that, despite the efforts and assurances of the traditional scientific method, it is impossible to anticipate all factors and collect complete information (de Bot & Larsen-Freeman, ). This is not to say that CDST is blind to the operation of factors; on the contrary, the study of the impact of ‘language-internal and language-external determinants’ is very much part of the changed research agenda, which is afforded by the methodologies outlined above (Larsen-Freeman et al., , p. ). However, these variables are not static, but emergent from interaction, with some acting as ‘precursors (features whose development is a necessary condition for the development of others)’, and others as ‘dependents (features that rely on precursors)’ (p. ). In other words, these variables cannot be divorced from the learner or the developmental process (Larsen-Freeman, , p. ). Far from simply ‘individualizing’ a universal acquisition/attrition process, they are constituent components of an essentially situated process. If language emerges from and in the intentional interaction with one’s environment, this perspective implies that, of the various variables listed in the first subsection, language use/ amount of contact is seen as a crucial variable (Herdina & Jessner, ; de Bot, Lowie, & Verspoor, a,b; Ellis, ). Larsen-Freeman & Cameron (a) argue that ‘a language at any point in time is the way it is because of the way it has been used’ and ‘would go so far as saying that every use of language changes it in some way’ (p. ). Thus, changes in natural input and opportunities to practise one’s languages should be reflected in the degree of acquisition/attrition experienced by individuals, albeit not in any deterministic, linearly causal way.

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi





Moreover, language use is intuitively a prime example of a dynamic factor which interacts with multiple other internal and external resources and variables (all also potentially interacting with one another), such as age, language aptitude, memory capacity, and motivation on the one hand, and space and invested time on the other (de Bot & LarsenFreeman, , pp. –). These resources, modelled jointly as ‘language maintenance effort’ (LME) and ‘language acquisition effort’ (LAE) in DMM (Herdina & Jessner, , pp.  and ), are required by the language system for development, that is growth or continued stability (de Bot et al., , p. ), and insufficient resources ‘will eventually result in language attrition’ (Herdina & Jessner, , p. ). They are also seen as potentially limited, and there may be competing demands from different parts (e.g., different languages) of a multilingual system for them, thus constraining potential development. To date, the importance of language use for attrition has more often been asserted than tested (Schmid, ). Empirical studies (e.g., Schmid, , b; Opitz, ) using traditional statistical means have found that language use is very hard to operationalize and measure; it is a multi-faceted factor in that it comprises active use and passive exposure, more or less personally relevant use, use in monolingual and bilingual modes, etc., which certainly has some part to play in the discovered effect being very moderate. It is as a result of the interplay between variables/resources that the tension between the inherent unpredictability of the language system, and the patterns discernible in language development may be reconciled. Larsen-Freeman () posed an important question and concern in CDST endeavours: ‘If we reject simple linear causality, do we have to give up on generalizability?’ (p. ). Given the disjuncture between individual trajectories and group averages highlighted above, and the fact that an exclusive focus on the particular is not very satisfactory, she offers three potential solutions: generalizing at system level, generalizing to theory, and finally, generalizing to potential (pp. –). With respect to the latter, she has suggested it may be possible to identify particular configurations of systems, ‘archetypes’ with ‘signature dynamics’ (Larsen-Freeman, ). Taking up this suggestion, a preliminary investigation of attrition profiles based on a comprehensive set of predictor variables found limited evidence for patterns in the attrition outcomes, amidst great variability between participants and counterintuitive patterning of variables for some participants (Opitz, a, b; see also Opitz, , for a first engagement with profiles based on a subset of variables). Thus, frequent L use was often associated with better L maintenance, knowledge of more foreign languages was associated with higher L attainment, the male participants on the whole were outperformed by the females; however, a number of individuals with similar profiles showed a divergent pattern for each of these variables. It appears that personal background variables do ‘conspire’ in the emergence of particular trajectories and profiles, but these outcomes are in no way predetermined.

. S  

.................................................................................................................................. The present exposition of the relevance of CDST to attrition research has argued that the current stalemate in studying the impact of personal background variables, amongst others, can plausibly be connected to the nature of language as a complex dynamic system, and the

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

     



associated properties of the so-called predictor variables. This changed perspective has recently generated considerable interest, though the uptake has, perhaps tellingly, not happened in a linear fashion (Larsen-Freeman, ). Seen by some as potentially paradigm-shifting (Larsen-Freeman & Cameron, ; Verspoor, de Bot, & Lowie, ; Larsen-Freeman, ) or even ‘the quintessential future approach to human action, cognition and behavior, including language’ (Van Geert, , p. ), it has focused our attention on ‘developmental processes in which acquisition and attrition of all of a speaker’s languages are logically merged into the developing language system’ (Larsen-Freeman et al., , p. ). CDST has introduced a very necessary shift away from the one-size-fits-all endeavour to establish general, if not universal, dependencies between variables and outcomes, to a focus on systemic individual variability, a predisposition to look for connections and appreciating diversity, crucially restoring dynamism to language development (Larsen-Freeman et al., , p. ). This, in turn, has opened up many challenges and questions, be they theoretical, methodological, or analytical, and has set in train a process of expanding the tool kit of acquisition and attrition research. However, while it is to be anticipated that CT/DST approaches to language development will contribute significantly to the methodology of applied linguistics in the future, the research techniques available to investigate change over time are recent and have not yet been used to their full potential. (Larsen-Freeman et al., , p. )

CDST, despite the optimism of its proponents concerning CDST’s potential to transform research into language development, has also drawn a certain amount of criticism. Some researchers believe it is a fashion or bandwagon, and indeed there is a danger of notionally adopting the terminology of complex dynamic systems without truly engaging with CDST’s methodological and analytical imperatives. Another problem is connected to the fact that empirical evidence for language as a complex dynamic system is still fairly limited, and to the question of how research findings should relate to theory (the chicken-and-egg problem). While CDST grew out of theoretical positions that did not seem to match the available empirical evidence, and the inconclusiveness of many empirical findings, researchers nevertheless need to be mindful not to treat their findings as a priori evidence, but to consider alternative explanations. A final point of criticism concerns CDST’s status as theory, with many rejecting this notion, and others questioning the very applicability of a theory originating in the physical sciences to the study of humans (Al-Hoorie, , cited in Larsen-Freeman, , p. ). Indeed, Larsen-Freeman () writes: ‘In some ways, the fact that “theory” is in the name of Complex Dynamic Systems Theory (CDST) is unfortunate’ (p. ), and goes on to explain that CDST has more far-reaching consequences than would normally be expected with a new theory. Larsen-Freeman et al. () also note that the word ‘theory’ in CT/DST does not refer to one grand unified theory but, rather, represents a range of theoretical insights that call into question some of the assumptions that have been used to inform and to structure research for many years. (p. )

As Erdocia () noted, there are different uses of CDST in language development, ranging from ‘soft approaches’, or metaphorical use, through the notion of CDST as a

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi





metatheory (‘medium approach’), to ‘hard’ and ‘radical approaches’, which respectively propose or accept language as a complex dynamic system and adapt their methodology accordingly. Larsen-Freeman () acknowledges the diverse uses of the concept (paradigm, metatheory, theoretical framework), preferring to see CDST as a transdisciplinary metatheory (Larsen-Freeman, , p. –), which is the view also adopted in this chapter. As such, CDST is particularly compatible with usage-based, emergentist, and constructivist approaches to language acquisition and attrition (Holland, ; de Bot & LarsenFreeman, ; Larsen-Freeman, ), while critiquing deterministic theories, as discussed above, and, in its rejection of the tenet of linear causality, ‘undoing’ the very notion of ‘good theory’ as one that ‘describes, explains and predicts’ (de Bot & Larsen-Freeman, , p. ). It is an understandable (and important) desire for humans to strive for simplification and objectivity (Larsen-Freeman et al., , p. ); nevertheless, this desire needs to be balanced with the need to take into account prevailing complexities. As this chapter has shown, the critique appears justified where the study of personal background variables in attrition is concerned: so far the notion of ‘predictor variables’ being in fact predictive has not been borne out in empirical research, and the available evidence seems to point to non-linear dynamic interactions and complex relationships between variables and other components. These findings, in turn, can be better accounted for from a CDST perspective.

A I would like to thank the editors Monika Schmid and Barbara Köpke, as well as Mirela Cherciov for their very constructive feedback on earlier versions of this chapter, and for their hard work in bringing this volume about.

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

  .............................................................................................................

PSYCHOLINGUISTIC AND NEUROLINGUISTIC APPROACHES TO LANGUAGE ATTRITION ............................................................................................................. Edited by

 ö   

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

  ......................................................................................................................

         ......................................................................................................................

 ö   

. I    

.................................................................................................................................. T field of language attrition has long been dominated by linguistic, applied linguistic, and sociolinguistic approaches, and psycho- or even neurolinguistic approaches have had some difficulties in becoming established. However, the idea that language attrition is governed by processes and mechanisms that are fundamental in the human mind/brain has steadily become a pillar underlying most attrition research. One of the first questions to be raised from a psycholinguistic perspective was whether attrition is a competence or performance issue. In other words, does attrition affect linguistic ‘knowledge’ or rather ‘control of that knowledge’ (Bialystok & Sharwood Smith, ) or ‘online processing of knowledge’ (Sharwood Smith & Van Buren, )? While our present conception of competence and performance has radically changed since the early publications, one of the major discoveries of language attrition research has certainly been that the answer to this question seems to depend considerably on the age at which a modification of language use arises (commonly referred to as the ‘age of acquisition of the second language (L)’ or AoA, see also Bylund, Chapter , this volume). One of the most consistent findings in language attrition research is that there may be profound changes in the linguistic system of a speaker who reduces the use of a language as a child (i.e., pre-puberty migrants) or even stops using it altogether (e.g., in adoptees, see Pierce, Genesee, & Klein, Chapter , this volume), whereas in migrants who have grown up mostly monolingually (i.e., post-puberty migrants), language attrition, at least as far as the first

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi



ö  

language (L) is concerned, manifests itself mainly at the level of online processing of the L (see Köpke & Schmid, a; Schmid, ; Schmid & Köpke, a; Bylund, Chapter , this volume). Most of the early attrition studies were mainly interested in how attrition at the competence level affected the linguistic system, and attrition phenomena were examined mostly in order to test the predictions of different linguistic theories with respect to the vulnerability of specific structures. In a similar way, attempts have been made to compare attrition to other processes of language change and language development, as was proposed by de Bot & Weltens () in a seminal paper that invited researchers to consider attrition as one of numerous processes of language development through the lens of either the recapitulation or the regression hypothesis. Attrition studies based on the regression hypothesis, which in its original form stipulated mirror symmetries between child language acquisition and language loss in aphasia (Jakobson, ; see also Barkat-Defradas et al., Chapter , this volume) have yielded interesting findings with respect to the vulnerability of different linguistic levels at different times in non-pathological attrition (e.g., Keijzer, a). But however interesting and valuable these approaches are, documenting mirror symmetries between child language, attrited systems and/or pathological language loss itself does not have explanatory power over why these similarities occur. Hypotheses like regression have to be placed in a psycholinguistic framework taking into account cognitive factors such as competition in order to gain explanatory power. Research into cognitive processes and brain mechanisms underlying attrition has only slowly emerged. It was not before , more than twenty years after the first publications on language attrition, that a collection of papers focusing specifically on neurolinguistic aspects of attrition was published (Köpke, a). However, recent evolutions in theories of language development (e.g., Dynamic Systems Theory (DST); see also Opitz, Chapter , this volume) and intensification of research on bilingual language processing as an overarching field, to which attrition also belongs, reflect the current interest into psycho- and neurolinguistic processes and mechanisms involved in bilingualism, including language attrition. The aim of the present section is to provide the reader with an overview of research that takes into account the neurofunctional and cognitive bases of language and how they inform attrition research. Before presenting an outlook on the different chapters assembled in this section, we will outline the main concepts and hypotheses on the basis of psycho- and neurolinguistic studies of attrition. More specifically, we will present research considering language attrition either as a matter of language processing, of memory, or of brain mechanisms. While many of these issues still remain speculative, we will discuss what they may entail for attrition and how attrition data in turn could help to better understand bilingual language processing and brain mechanisms.

. L     

.................................................................................................................................. The interaction of different languages during online processing is a central issue in research on bilingualism. The concept of cross-linguistic influence (CLI) was introduced by Sharwood Smith in , with the intention of replacing the notion of transfer—frequently

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

   



used in second language acquisition (SLA) research—by one with a less behaviourist connotation. While Sharwood Smith stated explicitly that CLI should be taken to refer to both influence of the L on the L and vice versa, the latter aspect has often been ignored in practice (see the discussion in Schmid & Köpke, a). But most importantly, the concept of CLI was meant to go well beyond the idea of interference between two language systems: CLI happens because different languages are processed through the same mechanisms, which leaves space for interactions of different types. Attrition, in this conception, is a direct product of the concomitant processing of two languages (Sharwood Smith, ). This idea was also developed in Green’s  activation, resource, and control model. This model of bilingual language processing was very innovative at that time with respect to several aspects, two of which are particularly interesting for attrition. First of all, the model distinguished between selected, active, and dormant languages. The selected language is the one controlling ongoing verbal activity and the active language plays a role in verbal activity even though it is not selected. That languages which are not selected but are active may intervene in the processing of the selected language has, since Green’s  model, been amply demonstrated for virtually every linguistic level (see Schmid & Köpke, a, for an overview). But what might have been the most interesting concept for attrition research was the idea that there may be dormant languages that are stored in long-term memory but do not intervene currently in ongoing language processing. Although Green’s model thus directly pertains to attrition, it has hardly ever been implemented in attrition research (but see Ammerlaan, ). It must, however, also be noted that no direct evidence has yet been found indicating that there might be a difference between active languages that do intervene in ongoing language processing and dormant languages that do not. In fact, most current bilingual processing models point to non-selectivity of all language systems regardless of their status (cf. Dijkstra & Van Heuven, ). The non-selectivity aspect was, however, also incorporated in another construct underlying Green’s () model, and has crucially shaped attrition research: the idea that there is constant competition between a speaker’s languages. Indeed, the idea of competition between languages clearly provided grist to the mills of attrition research and drew attention to notions such as activation and inhibition, both of which are now seen as pillars in bilingualism research and are perhaps most clearly attested in work on lexical access, for example in lexical decision tasks (for a review, see Van Assche et al., ). Green’s  model and his later Inhibitory Control (IC) model (, see below) paved the way in this respect. The concept of activation in particular was further developed in the Activation Threshold Hypothesis (ATH) proposed by Paradis (e.g., , ), initially mostly to explain nonparallel language recovery patterns in bilingual aphasia. Based on the hypothesis that the two languages of a bilingual are organized in two neuro-functionally independent subsystems within the language system, the ATH holds that interaction between the languages is largely determined by the frequency of activation of each language and each item that is part of the language, and by the time elapsed since its last activation. Following Paradis (e.g., , p. ), the Activation Threshold (AT) of a given item or language is lowered each time the item (or the language) is activated. Lowering of the AT enhances facility of activation and lowers cognitive costs associated with activation. However, if the item (or the language) is not activated again, the AT will slowly be raised, and the item (or the language) will become more and more difficult (and costly) to activate. This framework clearly states that facility of activation is directly linked to frequency of use, and as such the

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi



ö  

ATH is of specific interest for attrition research, where one language is markedly less used in a speaker. The first study to test the predictions of the ATH with respect to first language attrition was Köpke (), who investigated two groups of thirty German migrants in France and in Anglophone Canada respectively. This study’s main focus was the application of the ATH for lexical versus grammatical processing. Köpke’s findings suggested that grammar is not affected by frequency of activation to the same degree as the lexicon and these results were also discussed in relation to the related declarative/procedural model (Paradis, ; see Section . below for a description of the model). Gürel (a), on the other hand, focused precisely on grammatical processing in a study of the binding strategies for overt and null pronouns in L Turkish of twenty-four Turkish-English bilinguals. Her findings showed, in line with her hypothesis, that attrition affected grammatical processing in a selective manner; following the ATH, Gürel’s prediction was that competition due to differences in frequency of use can only arise when there is a competing structure in the L. In the absence of such similarities, competition will not arise. This was borne out by Gürel’s data: Turkish pronouns that have a correspondence in English tended to be processed as in English, while Turkish pronouns without an English equivalent did not show signs of L influence or attrition. Later, Schmitt () investigated the prediction of the ATH using data from five early Russian-English bilingual speakers who migrated to the US between the ages of  and  years and were adults at the time of testing. Spontaneous speech data collected in informal sessions were analysed both from a language contact perspective inspired by models proposed by Myers-Scotton (e.g., ) and using the perspective of the ATH. Schmitt concluded that while theories on language contact were informative in predicting the vulnerability of different types of grammatical morphemes, the ATH allowed her to account for frequency effects in lexical items. This points to the possibility that competition may arise in a more generalized manner between lexical items, whereas, in the grammatical domain, competition remains restricted to very specific conditions, as demonstrated by Gürel (see also Dussias et al., Chapter , this volume). What is furthermore interesting in Schmitt’s findings is the implicit conclusion that attrition phenomena are hard to capture in one single framework. Rather, integrated combinations of different frameworks are needed in order to explain the complexity of attrition. The straightforward relationship between frequency of use and attrition, as suggested by the ATH, has been seriously questioned in recent years (see Schmid, Chapter , this volume, for an in-depth discussion). Instead, growing attention has been paid to competition in addition to frequency, as suggested by Gürel (a). Competition is, of course, not only operating in attrition, but is a force at work in all bilinguals through the parallel activation of the two languages at nearly all linguistic levels (for recent reviews, see Kroll, Dussias, et al., ; or Schmid & Köpke, a). This is why neurolinguistic models of bilingual language processing—with at their foundation the influential IC model (Green, )—focus on the resolution of competition as the main specificity of bilingual language processing. But the precise conditions for the occurrence of competition still need to be specified more precisely. A comparison of the results from recent studies on bilingualism and attrition suggests that attrition is not qualitatively different from other manifestations of interaction between the languages in bilinguals. Rather, where ‘normal’ competition between the bilingual’s

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

   



languages ends and attrition starts seems merely a matter of degree (e.g., Schmid & Köpke, b; Köpke, ). Indeed, that attrition is a phenomenon inherently tied to processing mechanisms is an option further developed in this volume by Sharwood Smith, Linck & Kroll, and Dussias et al. (Chapters  to , respectively), for instance. Whether crosslinguistic interaction in language processing may, ultimately, lead to a restructuring of linguistic competence remains an open question. What is clear, however, is that such a question cannot be resolved without taking into account the memory structures that are supposed to underlie such linguistic competence.

. L     

.................................................................................................................................. Extending language processing accounts, cognitive scientists have investigated how the human brain deals with incoming information: they have explored how we learn, process, and how and where new information is retained, including what we know about words, and grammatical rules. But whereas a substantial body of work has detailed how languages are learned and thus how new linguistic memories are formed, notably less attention has been paid to the other side of the coin: how languages are unlearned. Ecke () explored how established psychological theories of forgetting (such as distortion, decay, and cue dependency) can be used to explain linguistic attrition (see Schmid, Chapter , this volume). But attrition research can in turn also aid the understanding of the link between memory systems and language representation and processing. Specifically, the declarative/procedural (D/P) memory model of long-term memory has frequently been invoked in an attrition context (Paradis, ; Ullman, a). Ullman (a, p. ) defines declarative memory as being involved in the learning and use of factual knowledge (i.e., facts and events) that can typically be verbalized, so this knowledge is explicit. Procedural memory, by contrast, is involved in the learning and use—or rather control—of cognitive and motor skills that are sequenced in nature. This type of memory is typically automatized and therefore mostly implicit. In relation to language, lexical knowledge is typically labelled declarative, with grammar being assigned to procedural knowledge, but—interestingly in the context of attrition—there is a difference between earlier versus later learned languages in bilingual settings: whereas lexical knowledge is declarative in both the L and the L, grammar knowledge of the first language is mostly automatized and so procedural while L grammar mostly remains declarative (Clahsen & Felser, ; Köpke, ). It has even been claimed that declarative knowledge can never become procedural, not even in very advanced L learners (Paradis, ). This, of course, has consequences for language attrition. It has been hypothesized that especially in postpuberty migrants not much grammatical interference from the L to the L will be witnessed, as both rely on distinct memory systems, but that L and L vocabulary items are likely to show considerable interference (Köpke, ), a claim that has been substantiated by empirical attrition research. Attrition itself can also feed back into theorizing within the D/P model: How does declarative versus procedural knowledge interact in attriters? Is it possible that procedural grammatical knowledge becomes weakened in

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi



ö  

attriters to the point where declarative knowledge takes over? And what role does age at emigration play in this? This taps into the evidence presented above that post-puberty migrants mostly show language processing difficulties in their L but no changes in their linguistic repertoires per se. Earlier work on attrition in the context of memory systems has tended to rely on memory metaphors of misplacing language structures or words, ranging from file cabinets to storage bins (cf. Ecke, ). Discussions regarding the D/P model, too, presuppose different storage facilities, although this has never been empirically substantiated. While neuropsychological evidence does point to different neural substrates subserving procedural and declarative memory, the question of storage is not without controversy. Underlyingly, all memory metaphors view the mind as a processor, much like a computer, where files can be stored, replaced, and deleted. This idea has recently been completely rejected, framed by a Radical Embodied Cognition framework (cf. Chemero, ). Instead, changes in language skills within that framework are best captured as stemming from a direct interaction between organisms and the world they live in: depending on a speaker’s environmental input, language learning, unlearning, or even relearning occurs (Keijzer & de Bot, ). Although language use and environment are generally seen as important external predictors of attrition (cf. Köpke, ; Schmid, a), they are not commonly viewed as inherent determiners of the attrition process. But from a Radical Embodied Cognition perspective, languages (words and structures) are never truly lost but are rather associated with or embedded in a given context and can be recalled by triggers pertaining to that environment. There is empirical attrition work to substantiate this claim, most iconically work that has been done on the savings paradigm. First introduced by MacLeod (), the savings account makes predictions about which type of linguistic information is retained after periods of non-use of newly learned words. More specifically, MacLeod found meaning to be retained to the detriment of word forms. In other words, semantic information rather than phonological information tends to be ‘saved’. Contrary to other memory metaphors, the savings paradigm works on the basis of input and associations. It was first applied to language attrition research by de Bot & Stoessel (), who investigated the retention of L Dutch after years of non-use by two German adults who had spoken Dutch for decades following a move to the Netherlands but at the time of testing had not used the language for a long time. Crucially, de Bot & Stoessel also added a relearning task and found that the experimental participants scored significantly better on the relearning task than a control group who learned Dutch for the first time. This and a series of studies following it (e.g., Hansen et al., ; Schneider et al., ) not only confirmed that residual (especially lexical) L linguistic knowledge is present in memory, but that it can be retriggered based on input. In other words, associations are reactivated upon changing input patterns. Retrieving linguistic information in memory in the context of attrition is, of course, also closely linked to the concepts of activation and inhibition outlined above. The theory of retrieval induced forgetting (RIF; cf. Anderson & Spellman, ), basing language retrieval and failure on associations and input, combines both perspectives. It has its roots in the cognitive psychological theory of catastrophic interference (McClosky & Cohen, ) and has relatively recently come to be applied to attrition (see also Larson-Hall and Linck & Kroll, Chapters  and , this volume, respectively). Under this theory, it is assumed that when one retrieves a piece of information, related knowledge is inhibited to avoid

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

   



interference and—with time—such local inhibition can spread to global inhibition of a language system. A classic RIF paradigm consists of three phases: familiarization, practice, and testing. Levy et al. () looked at attrition data invoking RIF. Their goal was to test the theory’s prediction with a sample of L American students taking a semester of Spanish at college. They found that these students became progressively slower to name an L English item after the same item had been named ten times in Spanish. Interestingly, when the word prompt was semantic in nature (e.g., naming the word snake upon seeing venom and cued with the initial word letter s . . . ), only a facilitation effect occurred; the RIF effect instead characterized phonological prompts (i.e., presenting a prompt that rhymed with the target word; the same word snake but this time with the phonological prompt break— s . . . ). Although the numerically small effect of this study has been critiqued (Runnqvist & Costa, ), this research has offered an important step in understanding the nature of forgetting linguistic information and most notably the nature of language storage and retrieval without having to resort to memory metaphors. In terms of theorizing, the (ir)reversible nature of forgetting can be examined more closely: within the RIF framework, L will be reactivated upon more L input. Combining insights from the savings account and RIF, it is evidently crucial to not only examine language decline that results from inhibition, but also to relate this to how that language developed in the first place and also whether it can be relearned and the circumstances that facilitate such relearning, or reactivation.

. L     

.................................................................................................................................. In recent years, the progress made in neurosciences has allowed us to reconsider a number of ideas about language and the brain that had been taken for granted for decades. While notions based on neurophysiological foundations, such as activation and inhibition for instance, have underpinned psycholinguistic research on attrition from the beginning, current insights provided by the neurosciences have hardly been taken into account yet in applied linguistics and specifically in research on language attrition. Research with monolinguals has allowed the development of very detailed models of the neural substrate and the time course of language processing (e.g., Hickok & Poeppel, ; Indefrey, ; Friederici, ; Price, ; Hickok, ). However, these are not, or only marginally, referred to in research on the processing of multiple languages in the bi- or multilingual speaker. Similarly, growing evidence from the neurosciences strongly supports the idea of neural multifunctionality, defined as the constant and dynamic interaction of nonlinguistic functions—that is, cognitive, affective and praxic activity, with neural networks specialized for language (Cahana-Amitay & Albert, ). Such multifunctionality has specifically been investigated with respect to the role of executive functions in language processing in speakers with and without language disorders and these functions have been shown to not only interact closely in cognition, but also to rely largely on common neural networks that are shared by language processing and executive functions (Zappalà et al., ). Despite the extensive debate on executive function benefits in bilinguals

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi



ö  

(e.g., Kroll & Bialystok, ), little to no reference is made to such insights in research on bilingualism or attrition. The most important concept in this context is that of neural plasticity. Recent research in neuro-rehabilitation, for instance, has challenged the notion of brain plasticity, which in turn questions the foundation of the critical period hypothesis (or related frameworks to account for age effects in language development). Evidence from neuroscience clearly demonstrates that brain pathways are adapted when new skills are learned and that rapid functional and structural changes may occur not only in children but also in adults. Studies have shown that increased grey-matter density, associated with skilled behaviour, can be measured after short periods of training in people learning to juggle (Draganski et al., ; Driemeyer et al., ), receiving training for working memory (Takeuchi et al., ) or music (Osterhout et al., ). But most importantly for our purpose, structural changes attributed to neural plasticity are also observed when people are involved in learning new speech sounds (Golestani et al., ) or exposed to a new language (Mecelli et al., ; White et al., ). Findings from Osterhout et al. () suggest that Event-Related Potentials (ERPs) allow us to see changes arising as a function of classroom L instruction in the brain’s electrical activity. Recent findings based on analyses of functional and effective connectivity in brain imaging data suggest that it is also possible to capture successful L learning using such techniques (Li & Grant, ). As also suggested by these neurofunctional data, the early stages of learning a new task are critical and involve the greatest changes, a challenging finding for our conception of attrition, generally based on the idea that attrition is a long term process. It is not known yet whether such neurofunctional impact on the L also has consequences on the existing L at these early stages of L immersion. One can only speculate what the consequences of attrition on brain structure might be since studies on brain plasticity have for the moment focused exclusively on the skills to be learned (e.g., language acquisition) and not on the reorganization of previously mastered skills (e.g., L processing). The picture is probably a complex one since it has also been suggested that expertise in multilingual language and control functions may reduce grey-matter density (Elmer, Hänggi, & Jäncke, ). For the moment, thus, we have only behavioural evidence for (temporary?) changes in processing the L during the first stages of learning a new language (e.g., Chang, ; Chapter , this volume) or in enhanced immersion contexts (Dussias & Sagarra, ; Dussias et al., Chapter , this volume), in the future, this should be explored with neurophysiological and imaging data, as the contributions to this section suggest. These new insights about neural plasticity underscore the huge capacity of adaptation and flexibility of the human mind. These appear also very clearly in studies on attrition. Attrition studies have already contributed significantly to major questions in SLA (e.g., the age factor). But much remains to be done and we believe that attrition research is in a unique position to shed light on bilingual theorizing. However, in order to do so, substantial work still needs to be done on psycho- and specifically neurolinguistic aspects of attrition: Attrition has to take into account insights from the neurosciences, but models of the neural networks of language also need to be able to account for the specification of bilingual language processing and development as in attrition. The links between attrition, as non-pathological language loss, and pathological language loss and disorders have to be elucidated. This is probably most urgent with respect to developmental language disorders where diagnosis of bilingual children is still difficult because we do not know enough on

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

   



bilingual language development and attrition in young children (see Blom, Boerma, & de Jong, Chapter , this volume). Finally, the debate on cognitive advantages in bilinguals should also take into account attrition and explain how the notion of cognitive advantages gained through the continuous need to inhibit one language is compatible with the observation that skilled bilinguals are not always able to inhibit the stronger, interfering language.

. S 

.................................................................................................................................. Adopting different psycholinguistic and neurolinguistic frameworks, the contributions in the present section not only provide a state of the art of psycho-/neurolinguistic approaches to attrition, but also uniquely show how attrition data can be used to inform psycholinguistic and neurolinguistics theories of language representation and processing. Chapters – explore the origins of attrition within the specificities of more general bilingual language processing. Sharwood Smith (Chapter ) reviews how language processing theories have been invoked in acquisition and attrition research. Treating both as instances of language development allows for a broader discussion of knowledge representations over time with language processing as the main driver underlying such changes. Within this framework, attrition is treated as a manifestation of general cognitive processing limitations stemming from competition between two language systems. Also departing from bilingual language control more generally, Linck & Kroll (Chapter ) posit that attrition is not language loss per se but rather reflects the dynamic outcomes of two constantly interacting language systems in flux. Focusing on the attrition of lexical retrieval, they build on the foundations of retrieval induced forgetting as a general memory mechanism that can have explanatory value regarding attrition. In this process, they also present more details regarding the constructs of inhibition and activation to show that even after very brief exposure times to the L, the L can be impacted. In Chapter , Dussias et al. elaborate on this hypothesis, but do so in the context of morpho-syntactic processing. Their perspective can be pivotal in elucidating how the construct of competition differentially impacts the lexicon versus the grammatical system, as discussed above. For the morphosyntactic domain too, Dussias et al. do not find evidence for true loss occurring as a function of L immersion, but rather that attrition is shaped by language processing changes. In this process, the authors show the usefulness of eyetracking as a device to show even fine-grained language processing changes in attriters. The fact that attrition is governed by changes in bilingual language processing is further taken up by Chapters –, which collectively address the links between pathological language dysfunction, which we know is governed by general cognitive abnormalities, and attrition. Blom, Boerma, and de Jong (Chapter ) explore how the complementing nature of attrition and bilingual processing literature can help elucidate the diagnosis of bilingual children whose language (learning) abilities raise concern. Blom et al. discuss the difficulties in ascribing the deviances in language abilities to either bilingual processing or Developmental Language Disorders (DLD). As the children concerned often have a migrant language background, insights from attrition could help in reaching a diagnosis. Higby, Lerman, Korytkowska, Malcolm, and Obler in their chapter (Chapter ) also explore methodological concerns, but in a fundamentally different way: they do not use

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi



ö  

attrition research to solve methodological issues in bilingual processing but rather point to methodological confounds within attrition research. Moreover, they move from child populations to the other end of the lifespan, namely older adults. Past attrition work has—often incorrectly as Chapters  to  demonstrate—adopted a minimum of ten years of residence in the L environment as a subject recruitment criterion. Often coupled with a post-puberty age at emigration criterion, this has meant that many attrition findings are based on older adults. In their chapter, Higby et al. describe the problem this poses: what is now ascribed to attrition could be more accurately labelled as ageing effects. Focusing on the lexical domain, the authors show that ageing and attrition exert different influences on lexical retrieval and ageing therefore needs to be taken into account in any attrition endeavour. Staying in the older adult lifespan range, Barkat-Defradas et al. (Chapter ) also emphasize the need to take ageing into consideration, when they suggest to compare non-pathological language attrition with language changes in late-life acquired disorders, most notably Alzheimer’s Disease (AD). It is especially the interaction between memory and language processing, which has been amply studied in AD studies, that may lead to important insights for attrition research, the authors claim. The chapters in this section present the state of the art of neuro-psycho-linguistic approaches to attrition and pointing to applications of attrition data in bilingual language processing theorizing, but also point to future directions that could further advance the field of attrition. Chapter –, then, introduce two innovative neuroimaging methods to the field of attrition, which until now has relied predominantly on behavioural evidence. Steinhauer and Kasparian (Chapter ) present the ERP technique as particularly suited to advance attrition research due to the method’s power to detect even subtle cognitive processing changes with a considerable level of detail. Importantly, ERPs cannot only detail L lexical processing changes, but can also tap into changes in L morpho-syntactic processing, shedding unique light on the role of interlanguage competition within both domains. Rossi et al. (Chapter ) present another method to chart the neural oscillation changes in language attrition: functional Magnetic Resonance Imaging (f MRI). The method is described in detail as well as its application in studies examining L attrition in international adoptees (see also Pierce, Genesee, & Klein, Chapter , this volume). Bringing the section full circle with the initial Chapters –, Rossi et al. suggest that functional and structural patterns of brain changes tapped through (f )MRI can be used as a test bed for the early stages of attrition, detailing the neural consequences of inhibiting an L.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

  ......................................................................................................................

         A wider cognitive perspective ......................................................................................................................

  

. I

.................................................................................................................................. T main aim of this chapter is to set language attrition research within a wider frame of reference. This should enable integrating research carried out so far into explanations of how the mind works in general (see Jackendoff, , for a related discussion). This, at the very least, should give a firmer foundation and greater plausibility to those explanations of attrition that can be aligned with research in other areas of cognitive science. While hypotheses and research findings that are focused exclusively on attrition can contribute greatly to our understanding within a narrow frame of reference, they still leave open to speculation a number of basic questions, especially with regard to processing mechanisms and their role in how language ability changes over time. The title of the chapter might also suggest an exclusive concern with how linguistic representations are processed in real time. This would not really tell the whole story so some elaboration on what is entailed by the ‘wider cognitive perspective’ is necessary. First, although it covers more than just language cognition, the theoretical framework to be used for this particular discussion falls within the realm of more formal approaches to language development. This entails a concern with the nature of linguistic representations themselves, their abstract properties, as well as how they operate during actual performance. Second, in line with much current thinking, attrition is treated as part of a general developmental account that includes acquisition. That said, online processing in real time does take centre stage. This is because changes in linguistic properties, that is, changes in the way in which the mind represents a particular language system, are viewed as being driven by processing mechanisms (see also Sharwood Smith, ; cf. Carroll, ; O’Grady, a,b).

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



 

More complete accounts and implementations of the framework reported elsewhere make clear the distinction between a theoretical ‘framework’ and a theory within a given subdomain of cognitive science such as language attrition (see Truscott & Sharwood Smith, ; Truscott, a; and Sharwood Smith, , for fuller accounts). The framework might conceivably be treated as a basic model of the mind but it is not a model of attrition per se, although it should test current explanations of attrition for their plausibility within a broader view of how the mind works as a whole and help resolve at least some of the important specific theoretical and methodological issues raised within the field of attrition itself. This chapter will accordingly explore the phenomenon of language attrition using the framework to broaden the explanatory frame of reference using the Modular Cognition Framework or, to use its better known title as applied to language research, the Modular Online Growth and Use of Language (MOGUL) framework. The ensuing discussion will first explain the need for new ways of making sense of the empirical data that has already been amassed so far, other chapters in this volume providing numerous examples. This will be followed by a detailed overview of the framework itself. Only then will it be possible to show how attrition data can be both investigated and explained using the framework, thereby sketching a possible future for attrition research (see also Sharwood Smith, ).

. I :  

..................................................................................................................................

.. Representation, change, and the nature of processing Both acquisition and attrition are types of change. Change is a dynamic notion and so mental processing of some kind must be involved, the kind that results in various kinds of alteration of whatever happens to be the current knowledge state. It is important to appreciate not only what has been done so far but also what has not yet been achieved. In particular, the study of processing in psycholinguistics has normally meant looking at the millisecond-by-millisecond online manipulation of knowledge structures that are already part of the language user’s system, such as particular syllables, words, and larger configurations exactly as they have been defined in advance by given descriptive and theoretical linguists. This, in many cases, has been carried out without acknowledging any need for supplementary explanations of how the relevant linguistic properties and relationships might be redescribed within a real time psychological architecture. A case in point would be the exact nature, in psycholinguistic terms, of an ‘interface’ between two linguistic systems like syntax and discourse (to be discussed in Section ..). Another one would be rule ordering where some fixed sequence in the application of a set of grammatical rules in a given abstract theory of grammar may not necessarily reflect any comparable sequence of operations in real time processing. When it comes to looking at how an individual’s representation of a language develops over time either in a direction best termed ‘attrition’ or in a direction best called ‘acquisition’, periods much longer than milliseconds obviously have to be involved.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

       



This separate focus on transition as opposed to rapid online processing might also require new ways of thinking about the interaction of processing mechanisms with linguistic representational properties. In sum, developmental linguists, including those who study language attrition, cannot by definition ignore the real time dimension; while they base any of their analyses on theoretical linguistic research, processing issues just have to be important for an understanding of what they have chosen to investigate (see, e.g., Köpke & Schmid, ; Schmid, a, for an overview of attrition research including psycholinguistic studies involving processing issues). This need to take real time processing into account has already been generally recognized and a number of studies have appeared in recent times that at least make explicit reference to the role of processing in making sense of attrition data (e.g., Tsimpli & Sorace, ; Tsimpli, ; Chamorro, Sorace, & Sturt, ). Neurolinguists deal with a quite different level of description, their domain being the physical brain. They too address the same basic questions: how languages are stored, how what is stored becomes accessible during actual performance, what changes over time occur, and what psychological processes can account for this (see, e.g., Köpke, b; Keijzer, ; Kasparian et al., ). Neurolinguists also need to relate their observations of neural activity and changes in the physical brain to patterns of human behaviour. For them, as for developmental psycholinguists, what is needed is some kind of common ground, a well-worked out, usable—that is, sharable—conceptual framework that can assist the mutual exploration of relationships between overt behaviour, brain, and mind (Sharwood Smith, ). Nevertheless, most research into attrition and acquisition has been going on for such a long time focusing solely, and indeed for understandable reasons, on precisely defined and limited hypotheses within an explicitly limited domain. One particular illustration of the lack of conceptual clarity in accounting for changes in an individual’s representation and use of language is the issue of whether developmental data should best be viewed as reflecting discrete stages or gradual continua because data can be analysed to show both (see related discussion in Sharwood Smith & Truscott, ). By using theoretical linguistic frameworks, but also theoretical linguistic concepts and methodologies in order to compare and contrast knowledge states over an extended period of time, it is certainly possible to analyse observed changes and make insightful, linguisticallybased statements about discrete stages of development on that basis. At the same time, only looking at putative stages, however elegant and explicit the analysis, cannot possibly be the whole story. Not everything proceeds in one direction and in clearly defined hops and jumps. The key questions are exactly why and exactly how one stage develops into another, gradually or in discrete steps, and this problem needs more than linguistic theory by itself. It needs theories of acquisition and theories of attrition and possibly a developmental theory that actually encompasses both kinds of change as part of a unified explanation. References to online processing may sometimes be used as a convenient way of accounting for recalcitrant data that do not fit with some current dominant trend as predicted by a given hypothesis. An interesting example of this is the debate in second language (L) acquisition research between two sides: first, the ‘representational deficit’ side claims that older L acquirers constructing the mental grammar of their new language show imperfect access to ‘universal grammar’ that children can rely on to abstract away from imperfections in the input and arrive at a language system with remarkably few grammatical errors. On the other side, the opponents of this representational deficit hypothesis argue that such

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



 

access is intact in L acquirers and that the apparent failure to produce the relevant grammatical structures in observable performance is actually due to processing difficulty (see, e.g., Hawkins & Liszka, ; Lardiere, , ). This represents a theoretically important difference in the interpretation of the same data: in one case the relevant grammatical principle is no longer available, whereas in the second case the observed absence is misleading and the relevant principle is present in the learner’s mind, but it is processing difficulty that obstructs its appearance in actual performance; closer analysis and other ways of eliciting data should show that there are in fact observable effects in performance that indirectly signal the presence of the presumed missing representation (Juffs, ). The question of what ‘processing difficulty’ precisely entails is left open because there seems to be no way of expressing with any rigour and precision what that concept actually means. Exactly the same holds for attrition: one must again ask whether divergent performance necessarily reflects divergent representations and, if not, how this should be explained (see Sharwood Smith, a, pp. – for an early discussion of this issue). It should be clear by now that research so far has been throwing up a steady flow of unanswered questions, hence the attraction of a framework that combines both acquisition and attrition under one heading of development and makes explicit the nature of developmental processes. This attraction has been with us ever since Jakobson’s intriguing claim, for phonology at least, that stages in attrition might be the same as stages of acquisition but in reverse order (Jakobson, ; see also Olshtain, ; and Barkat-Defradas, Gayraud, Köpke, & Lefebvre, Chapter , this volume). Keijzer () rightly observes that this ‘regression hypothesis’ is not properly explanatory by itself and states that the larger context of linguistic theory is required for explanatory adequacy to be achieved. The claim in this chapter is an extension of this argument, namely that an even larger, ‘mind-wide’ framework is required. Even if full explanations do not immediately fall out by widening the perspective, it should enable a broader and more comprehensive understanding upfront of what has been achieved so far as attrition research is concerned and what still needs to be explained.

.. Interfaces In an attempt to improve the quality of explanations, recent work has concentrated on interfaces as the possible sources of difficulty either in forming new representations in the first place or in accessibility—that is, processing (Tsimpli & Sorace, ; Sorace, ; see also Chamorro & Sorace, Chapter , this volume). An interface is, as the name suggests, a place where two different systems interact. The existence of different systems that may have to collaborate in some way assumes, as is also assumed in this chapter, some type of modularization in the mind: mental modules are systems that have a specific function not shared by others and their own set of principles for constructing representations. Examples include the visual system and the auditory system for example, or—in linguistics—syntax and semantics. The visual system creates visual representations that are structured and organized in ways appropriate for managing visual experience: these ways will not work for other kinds of representations. At the same time, visual experience and other types of experience (or rather those mental structures that have been formed to represent them) often need to be coordinated. Coordination between structures of different types, for example visual and auditory representations, or syntactic and conceptual representations,

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

       



is managed via interfaces. In other words, however functionally unique, modular systems may be defined by a given preferred theory, and they must in some way be connected by interfaces (see, e.g., Goad & White, ; White, ). Interfaces may not always work in an optimal fashion. For instance, whereas the core or ‘narrow’ syntactic system of an L learner may reach native-speaker levels, the semantic or discourse aspects of related structures may not. This, by hypothesis, is because of an assumed problem at the interface between syntax and other modular systems, the ones that handle semantics and pragmatics: the right associations are not being made by the learner (Tsimpli & Sorace, ). Similarly, the core syntactic system has been claimed to be relatively resistant to attrition effects compared to its interface with semantic and pragmatic systems, these connections being more vulnerable to change and resulting in the inappropriate use of syntactic structures such as the non-nativelike inclusion of overt grammatical subjects in languages where omission of these is permissible and, in some contexts, required or at least overwhelmingly preferred (Tsimpli et al., ). At this point in the whole explanation of interfaces, the picture becomes, for the psycholinguist at least, more than a little vague. The way the interfaces function is typically not specified in any detail and there sometimes, as previously mentioned, appears to be a prevarication between a purely representational account of interfaces and one which involves processing in real time. The representational account might go something like this: the interface between syntax and semantics, or syntax and pragmatics, presents learners with the formidable challenge of isolating the relevant semantic or pragmatic properties from a mass of contextual data made available in the input and then making the right principled associations between those properties and particular syntactic structures. This account focuses more on structural complexity at the levels of representation without making claims about processing complexity, perhaps simply implying that the two types of complexity might be related. This is not a necessary relationship, of course: what is structurally complex may still be easy to process given the existence of appropriate processing mechanisms that can deal effortlessly with such complexity. Particular theoretical commitments are required to determine how exactly any kind of complexity may be defined but certainly ever since Chomsky & Miller’s classic paper that appeared in , the relationship between syntactic complexity and what happens in performance has been generally acknowledged as not being a straightforward one. More recently, just to cite one example of this, Gayraud & Martinie () looked at various kinds of embedded sentences using pause length as an indicator of processing cost, and were not able to equate structural complexity with processing difficulty. This kind of research highlights the need for separate explanations of the processing mechanisms involved, including the nature of working memory (WM), and the nature of processing complexity (see Sorace, , for a more nuanced analysis of complexity and how interfaces might work specifically in a processing context). Sometimes the interface problem is defined as relating to both representation and processing, distinguishing between (a) the principled relationship between structures in one system and structures in another—that is, ‘interface structures’ (see Contreras-García, , for a useful survey of the syntax-semantics interface in different theoretical linguistic perspectives); and (b) the way in which these interface structures are processed online, a recent example of this being Chamorro, Sorace, & Sturt, (), a study also focusing on attrition. The framework used in this chapter at least attempts to provide a degree of coherence and granularity to accounts of processing concepts such WM,

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



 

competition, inhibition, etc., and how they interact with representational properties and proposed mechanisms such as interfaces without any claim to be a comprehensive theory of processing.

. U , ,     

..................................................................................................................................

.. An overview MOGUL is an attempt to resolve issues such as the ones mentioned above (see http:// .../~mogulfra/ for links to more detailed accounts). It permits an integration of independent research into linguistic properties—that is, ‘representations’, grammatical or otherwise—with other psycholinguistic research focusing on real time processes—that is, ‘performance’. In addition, it permits an account of linguistic change (‘acquisition’ or ‘attrition’) that obviates the need for special grammatical learning mechanisms (cf. O’Grady, b). In fact, the framework does even more than that because it encompasses all kinds of cognitive phenomena and in this way allows a simultaneous study of the nonlinguistic forces that shape the way language is used and the way it changes in the individual (see, e.g., its use in studying code-switching in Truscott & Sharwood Smith, ). These claims would be speculative but for the existence of two important characteristics that the framework possesses. First, the inspiration the framework draws depends on established theories and research findings across relevant disciplines within cognitive science as a whole, especially psychology, neuroscience, and linguistics. Second, it is a framework—not a theory in its own right—that transcends any single subdiscipline of cognitive science, unlike, for example, the Minimalist Program in linguistics which, although it has clear implications for cognition in general, is nevertheless designed to direct research within a particular branch of theoretical linguistics (Chomsky, ). The MOGUL framework provides a skeleton model or ‘map’ of how the mind works and how ‘language’ in both its broadest and narrowest definitions operates within the mind: it still crucially depends on research carried out by linguists and researchers in other areas of cognitive science to specify how all the various aspects of this model should be fleshed out (Sharwood Smith & Truscott, ). For example, a full specification of, say, a visual processor or a syntactic processor requires the principles and properties of, respectively, vision and morpho-syntax, to be spelled out in detail. The framework, whatever its inbuilt theoretical commitments such as to the modularity of mind and a particular view of the language faculty, does not do that. The incorporation of, in the case of the above examples, vision or syntactic research into a common framework should facilitate interdisciplinary synergies and collaboration. In any case, the framework does not provide these detailed specifications, this being the job of the researcher with appropriate expertise in these two fields. Current descriptions of MOGUL do in fact use existing theoretical linguistic, psycholinguistic, and language acquisition and attrition research to provide specific implementations of the framework (e.g., Sharwood Smith & Truscott, ; Truscott &

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

       



Sharwood Smith, ). But although the framework’s architecture is theoretically committed to some extent, there are still many alternative ways to instantiate it in detail: (a) when planning a given investigation or (b) when dealing with existing research findings that require richer, or just different interpretations.

.. Modularity and collaboration: key elements of the framework MOGUL architecture, although inspired by a range of views across cognitive science, owes much to Ray Jackendoff and can certainly be seen as an extrapolation and adaptation of his views although there are some significant differences (see Jackendoff, , , ; Truscott & Sharwood Smith, ; Sharwood Smith, ). Briefly, the basics of the framework are as follows. The mind is composed of expert systems such as the various perceptual systems (vision, audition, etc.), that is, the mind is ‘modular’. The expert systems are specialists, each of them unique in the type of task they carry out. These systems, very generally, reflect the basic organization of the brain as currently conceived. At the same time, it needs to be emphasized that accounts of the brain represent different levels of description and, while exploring relationships between the two is important, the two levels, as implied earlier, should nevertheless be kept distinct, the one abstracting away from physical details, the other not. For example, the functional map of the brain’s perceptual systems might show details of neural pathways and/or neural locations, neither of which need figure in a more abstract display that describes perceptual processing from a psychological perspective (see Sharwood Smith, , p. ). The modules in the mind are not so independent that they exist in a vacuum. On the contrary, they form a network of often highly interconnected systems allowing them to collaborate in many different ways. The fixed, stable, systematic nature of the modules’ internal mechanisms thus contrasts with the flexibility and dynamism of their interactions. In this way a perceptual experience can be built up through an interaction between the various perceptual modules and, for example, modules associating respectively meanings and values to the various representations that have been activated. This collaboration goes on all the time, activating or building ways of representing all kinds of external and internal input, incrementally and in parallel and virtually always without conscious intervention. At this level of abstraction, the modularity of mind and brain and the notion of collaboration are for the most part practically incontrovertible, controversy focusing on the precise nature and number of modules and the ways in which they can collaborate.

.. The developmental perspective In addition to the unification of the two above perspectives, there is a further consideration that is crucial for the study of attrition, namely the developmental perspective: this is the dynamic ‘change’ dimension mentioned above (in Section ..), which covers both attrition and acquisition. As already suggested, psycholinguists typically deal with the way existing knowledge is processed at a given moment but not processes that lead an individual to change over time. Without this third, transitional, perspective, however, language attrition can only be studied by relying on synchronic slices at a given moment in time and

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



 

comparing these slices with slices taken at a later time. This may lead to some principled speculation about the way change has or has not occurred in the intervening period. However, there is no substitute for describing the processes that actually drive this change. In short, this is why we need a developmental (i.e., transitional) account of how and why representations change over time and a crucial part of that account, in the framework under discussion, is the role of processing. Attrition by Processing Theory (APT) is a reformulation of Acquisition by Processing Theory (Sharwood Smith, ). It is a general principle that characterizes attrition in all kinds of areas including mental arithmetic, piano playing, or knowledge of history, as the outcome of the individual’s processing history. Part of the explanation has to do with the straightforward absence of processing, which has negative effects on the relative accessibility of a given (linguistic or other) cognitive structure (as explained below in Section ..). A language that is not used for an extended period or, say, lexical aspects of that language, will tend, as a direct result of this lack of use, to undergo effects that we call attrition and are experienced as loss or forgetting. However, the way APT as a general cognitive processing principle is actually instantiated will also depend on local intramodular principles for the way it operates within a given modular memory store. Thus, syntactic attrition as manifested in comprehension and production performance will depend not only upon the activation history of an individual’s syntactic representations, complex or otherwise—that is, on a ‘use it or lose it’ principle—but also on the presence of other syntactic representations which compete with them. The connections across interfaces with other system will also have a role to play so that ‘valued’ representations that have strong affective associations may survive a long period of non-use and remain relatively accessible. In this way APT as a general principle is instantiated in the syntax module via local principles and effects: these will be observed, elicited, described, and explained by those working on syntactic attrition and applying the particular theoretical and methodological tools that they favour. More details follow in subsequent sections. It is perhaps worth noting in passing that dynamic change and dynamic effects in online processing might suggest an application of Dynamical Systems Theory (see also Opitz, Chapter , this volume). This possibility certainly exists but only within the constraints posed by the framework’s modular architecture and not as the sole explanans (see discussion in Sharwood Smith & Truscott, ; Sharwood Smith, ).

.. Representations and memory theory If people want to do responsible research in attrition, they must have more than a vague understanding of human memory. They need to commit to some currently plausible view of what it is and how it operates. In the MOGUL framework, the elements—call them features, representations, or simply structures—are located in a store. For the purposes of talking about processing, this store can indeed be viewed as a memory store. There is no common repository where all kinds of cognitive structures are housed. Rather, each expert system—that is—module, has its own dedicated store, for example a visual memory store or a syntactic memory store. The respective module’s processor that manipulates them only deals, and can only deal, with the contents of its own store. Thus the visual processor is only equipped to handle visual representations and the syntactic processor can only deal with syntactic structures. By analogy with, say, JPEG or MP files, the representations of separate stores are written in different, mutually incompatible codes. From this it may be correctly

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

       



inferred that the basic make up of any module is a processor plus a (memory) store where neither the principles on which the processor operates nor the structures that it manipulates are shared by any other module. There is also no central processor and no central memory facility. Everything works by collaboration between particular combinations of these modules at any given moment. The only sense in which processing comes together as a single unified phenomenon is as conscious experience—recall that the vast majority of the mind’s operations are subconscious—and the framework has its own account of how that happens which will be very briefly touched upon at the end of this chapter (see Sharwood Smith & Truscott, ; Truscott, a,b, for more elaborated discussions of conscious awareness in this framework).

.. Working memory The question that now arises is the nature of so-called long-term memory (LTM) and WM, a shared temporary storage facility used during online processing. It would in principle be possible for there to be a set of separate stores, one for each expert system, but only one shared WM where separate types of structures originally activated in the stores of different mental systems, vision, audition, syntax, the conceptual system, etc., are manipulated all in one place during online processing. Establishing more precisely how WM operates is particularly important since it plays a crucial role in the current explanation of attritional change. In fact, within the framework, there can be no shared WM since, as implicit in the basic design of a module just described, the different types of structure are completely incompatible with one another and can only be handled according to the unique principles of their own dedicated processor.1 Second, WM, in line with Cowan’s (rather than Baddeley’s) ideas is viewed not as a separate buffer, located separately, but as a change of state within a memory store (Baddeley & Hitch, ; Cowan, , ; Baddeley, , ; Truscott, ).2 In other words, syntactic WM, for example, is a change of state of representations specifically within the syntactic memory store alone. The basic set-up of a/ the modular system in the framework is therefore not a tripartite one: Processor + LTM + WM, but rather a bi-partite one: Processor + M, where ‘M’ indicates an undifferentiated memory store. Thus, when a given element is accessed in a given module during online processing, it is activated within its own store and forms part of current WM along with any other member of the store that happens to be activated at the same time. In MOGUL, the metaphor chosen for representing activation is of rising and falling within the store, so the activated element ‘rises up’. This means that the WM state is conceived of as the upper area of the store. There are other possible metaphors but it is just that showing activation as points along a vertical plane is easy to represent graphically. Although there are differences, for example with regard to the treatment of the declarative/procedural dichotomy, Paradis has a similar approach to activation except that he talks in terms of an activation ‘threshold’ (Paradis, ). For the likelihood of something to be selected, a ‘low’ activation threshold is required, which is equivalent to a ‘high’ resting level in MOGUL. In each case the representation in question is effectively in a position that gives it easy access to WM. 1

However, Truscott makes an exception for highly synchronized perceptual WMs which effectively operate as a shared workspace (personal communication; see also Truscott, 2015, p.120). 2 Cowan nonetheless does see WM as a single shared space.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



 

.. Resting levels of activation Another crucial related notion for understanding attrition is the resting level of activation. This may be compared to the neural level of description with the notion of ‘resting potential’ in quiescent neurons (see, e.g., Daignan-Fornier & Sagot, ). The absence, especially in the longer term, of activation results in a progressive ‘lowering’ of the structure within the store. This means that, in order to make it into WM and participate in a current act of processing, if and when it gets activated, the structure (representation) has further to rise, again using the same metaphor. A decline in its resting level makes it less accessible. This crucially affects its chances of participating effectively in online processing since one basic feature of processing that most if not everyone assumes is competition with other potential candidates (see, e.g., Dijkstra et al., ; Schwartz & Kroll, ). Both monolinguals and multilinguals become aware of this when they are struggling to produce a word and the ‘wrong’ word pops into their minds, seemingly blocking the one they are sure they do actually know. In contrast to this attrition-inducing scenario, when a structure is activated, and especially when it is activated regularly, it undergoes gains in resting level, making it more, rather than less, accessible and therefore more competitive. For this reason, acquisition is viewed as the lingering effect of processing. With attrition, there are therefore two related forces at work. First there is the gradual decline in the resting levels (and therefore accessibility) of structures that are not frequently activated, or activated enough to maintain their resting level. Second, there is the gradual rise in resting levels (and increase in accessibility) of structures associated with another language that now begin to compete successfully in the scramble for selection in WM, thereby changing the dominance of the languages concerned. In general, there has recently been quite a lot of work on activation and inhibition levels in bilinguals including studies on how even early stages of L learning can lead to (local) attrition effects in the first language (see, e.g., Bice & Kroll, ; Linck & Kroll, Chapter , this volume). Alongside the specific findings of this kind of research, a more structured understanding of the underlying psychological mechanisms such as is provided by a mind-wide cognitive framework ought to contribute substantially to our understanding of the phenomena in question.

. A   

..................................................................................................................................

.. Acquisition and growth The further specification of what an ‘interface’ is (see Section .. above) and what interfaces do is seen as a very important requirement for any approach that uses this notion. In the MOGUL framework they are explained as follows: interfaces may be viewed as simple processors that manage associations of structures across adjacent modules. This involves both forming an association in the first place and, once that association is formed and one of the structures is activated, co-activating its matching structure in the other module. ‘Associating’ only means matching and co-activating. No information flows from

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

       



one module into another. All interfaces work the same way, so they effectively enable collaboration of different systems in the mind and together make the mind into a network of interconnected modules. Where there are only two modules involved, the result is a structural chain of two structures, for example a visual representation paired with an auditory representation, thereby giving a sound a meaning, or a meaning a sound (since processing interfaces work in both directions). The mechanism for identifying structures is an index, which is a shared identity tag, not the kind of index used in classic generative linguistics (see Jackendoff, , p. ). Indices are assigned once the initial association is made: a structure in module A (e.g., an image) and a structure in Module B (e.g., a sound) are then tagged with the same index. In this sense, acquisition, as the initial formation of an association, is instant. However, attrition can also set in quickly if the indexed association is not reactivated sufficiently, which means that acquisition can be both instant and ephemeral. What is far from instant is growth. Growth is the result of changes over time in the resting level of activation of the structures concerned, as described above, and a continual lack of activation will have as its consequence the lowering (and possible extinction) of a structure. Extinction might be empirically established with the increasing sophistication of psychological elicitation techniques, but ultimately, reference to neuroscientific evidence might have to be a way of deciding whether structures that consistently fail to affect performance in any sense, easily observable or more subtly observable, can be treated as ‘actually extinct’ or just ‘very dormant’. An index may be assigned to any structures in any module that has been associated by means of the relevant interfaces, so associations result typically in a network of many structures each of which may have many indices (i.e., cross-modular connections) but amongst them share one index. The conceptual structure TREE and the syntactic structure N(oun) will share an index, for example. Any individual structure like N will, of course, have many different indices, reflecting its multiple associations with given structures in adjacent phonological and conceptual systems.

.. Interactions across interfaces in acquisition and attrition The role of the interface, as simple as it is as a mechanism, is crucial in accounting for how the mind creates complexes of different types of mental structures to make sense of the inner and outer world of an individual. The way the mind’s connections work is, in essence, very simple, with only two ways of linking different structures: first, there are the external connections, associating and co-activating structures written in different codes—that is, across modules; and, second, there are the module-internal associations. Whereas external associations are managed by interfaces which can be seen as very simple processors, these internal associations are managed by the much more complex processor belonging to the module in question. Structures within a module’s store are associated in various ways according the principles unique to that processor and so, for example, visual representations, both simple and complex, will be structured differently from auditory, conceptual, or syntactic ones. Primitive visual elements and particular combinations of these primitives may, when co-activated together, form a representation of some simple cube-shaped object in view, but, as the visual scene is built up further, larger visual structures may be combined

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



 

to form ever more complex constructions. At the same time whatever is activated inside the visual store will be triggering, via the relevant interfaces, associated complex representations in other stores. Take a simple example of activity within and across two modules during utterance production, involving the stores of respectively the syntactic and conceptual systems and take also the two syntactic structures: Adj(ective) and N. As alternative combinations of structures are acquired by developing sequential bilinguals (L acquirers, heritage speakers), competition between these alternative combinations will inevitably arise in the course of forming a message. On receiving conceptual input consisting of two conceptual structures, say RED and BOAT, the conceptual–syntactic interface will activate both an Adj-N sequence and a competing N-Adj sequence in the syntactic store. This is because two alternative associations across the conceptual–syntactic interface have been created during exposure to two languages. What happens then will depend on the relative strength of the activation of the participating structures. All other things being equal, the dominant language system will win out. For example, the English acquirer of French wishing to talk about a boat that is red will be prompted, by virtue of the dominance of the relevant English syntactic construction, to produce in French, *‘rouge bateau’ (red boat) rather than ‘bateau rouge’ (boat red). The only chance for the French sequence to win the competition is if the resting levels of the N-Adj construction are given a sufficient boost over the longer term by the involvement of other interface connections as will now be illustrated. Apart from the basic driver of change—continued exposure to and processing of French input—it is important to appreciate the important supporting role of the affective—that is ‘value-assigning’ system. For example, where the situational context is French-oriented— French is being spoken—an association with the concept FRENCH in the conceptual system is co-activated; it has probably already been co-activated to some extent because the French construction was originally acquired in a French context with characteristically French sounds and French text, but now, given the immediate French-speaking situation, the conceptual structure (CS) FRENCH is activated together with a very strong positive value: this value is an affective structure (AfS) which is co-activated with the conceptual structure FRENCH via the affective–conceptual interface. This means in very simple terms that French is being treated as ‘currently important’.3 This positive loading will have the effect of giving a boost to the activation level of any other associated structure currently co-activated. This now, at least temporarily, gives the N-Adj construction more of a fighting chance in the current race for selection in WM4 against its normally dominant rival.5

3

In such a situation, the evaluation may be further complicated by such factors as language anxiety reducing the value of French possibly promoting only passive participation in the exchange or the special importance of the conversational exchange in question for some reason increasing the positive value. 4 ‘Selection’ is a metaphor since there is no special selection mechanism per se. As processing builds up a complex representation, incrementally but also with a back-and-forth search for coherence between structures, it is just a matter of what ultimately comes out on top in the competition. In the assembling of competing representational networks across the various interfaces, the closest we come to the idea of selection the best-fit principle (Truscott & Sharwood Smith, ). 5 This would be the framework’s instantiation of Grosjean’s bilingual language mode (Grosjean, ).

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

       



The boost it gets may still not be sufficient for selection to take place on this occasion. However, as a general principle, each bit of activation, however small, gradually builds up the default resting level of a structure, making it ever more competitive over time. Note in passing that the ‘default’ resting level is an abstraction from what in effect is a level that is continually changing. Also, the way structural properties gradually grow together or grow apart along with the interface connections with other structures can be described as ‘emergent’, although the local organizing principles of a module remain predefined so linguistic systems are not self-organizing in the sense intended by emergentists who generally do not accept the need for an account of grammatical growth that requires module-specific construction principles (MacWhinney & Bates, ; O’Grady, a). Rather, they claim that complex systems like grammars emerge only as a result of the interaction of simple principles of a general nature, although there is no wide agreement as to what these principles are (see O’Grady, b, for a concise overview) Now consider a later situation where the same learner has become an attriter. Infrequently activated constructions associated with English consequently suffer a steady decline in their default resting levels. In terms of the vertical metaphor, the structures involved have a longer way to go to make it up into the WM of their store and are therefore much less competitive as a result. The L, French, on the other hand, by virtue of the very frequent activation of French-associated structures, gradually becomes the default dominant language. Without any supportive situational context such as described above, the ‘neglected’ English Adj-N sequence in the syntactic system is therefore less and less likely to win out: the resting levels of activation of those structural features that determine the sequence accordingly sink further and further down and the Adj-N construction becomes correspondingly less accessible. This is a very basic account of how attrition might work in processing terms. On the one hand, a reduction in the frequency of activation makes L structures less accessible. This in itself has nothing to do with the presence of an L. On the other, the newly dominant L will start to compete successfully and cross-linguistic influence will manifest itself in various simple and complex ways, the precise manner being the focus of attrition researchers—in observable performance. In other words, this basic account definitely needs to be supplemented. The framework provides an appropriate background set of processing principles and assumptions but certainly not a comprehensive attrition theory in itself. If it did it would be a cognitive ‘theory of everything’. However, there are other aspects of attrition that the framework does provide, that is, over and above the most basic account, particularly in connection with interaction between different systems and notably the special role of affective and conceptual systems as touched on above.

.. Frequency effects It is worthy of note that frequency effects, in the framework, are limited to the internal frequency of activation within given memory stores. In development—as acquisition— there may quite often be a discrepancy between the high external frequency of structures perceived by the hearer in the immediate environment and the actual processing subsequently carried out (or not) within particular modules (Gass & Mackey, ). Here, then, it is clear that frequency of this external input is not to be equated with the real driver of growth, namely the frequency with which structures are activated within modules. In the

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



 

case of development—as attrition—minimal frequency in the input from the environment or the complete absence of such input would appear, on the face of it, to be the main driver of change but again it is first and foremost a matter of the frequency of internal input that either brings about activation or no activation within modules that brings about the relevant changes. External input frequency encountered in speech, texts, and signing is still a secondary factor until it is processed by one or other of the internal systems; also, purely internally generated language use as in writing—even thoughts and dreams—in a given language can play a supportive role in countering or slowing the decline in resting levels of otherwise unused representations.

.. Degrees of forgetting It is really not important in such an account of attrition whether complete forgetting takes place. Once a structure’s resting level of activation has sunk down far enough it will not figure at all in spontaneous language performance. As suggested earlier, it is an empirical issue to be resolved using techniques perhaps more sensitive than are currently available as to whether, after many years or as a result of some pathological condition, an attrited structure can reappear without any reacquisition taking place.

.. Inhibition and attrition The notion of an activated structure that is regularly outcompeted and accordingly drops back to a lower resting level may indicate the operation of inhibitory processes, as also the idea that this resting level will steadily decline over time through lack of activation reducing accessibility (Green, , ). Note, however, that inhibition may in principle occur as a post-facto response to non-selection, that is, only after a rival candidate of selection has been outcompeted and not be the force that actually brings non-selection about (see Berg & Schade, , p. ; and also Köpke, , for related discussion).

.. Metalinguistic knowledge and attrition The knowledge that such and such a construction is anomalous or ‘correct’ in a given language may exist independently as explicit ‘metalinguistic’ knowledge. In terms of the current framework, such knowledge is represented separately, in the conceptual system, along with any other factual and consciously accessible knowledge of the world. As such, it has no direct bearing on what is happening to the phonological and syntactic representations in other modules (Sharwood Smith & Truscott, ; Truscott, ). Where it can be projected into consciousness it is more or less equivalent to what elsewhere is termed ‘declarative’ knowledge (Ullmann, a). Such conceptually-coded knowledge of language can, of course, also suffer attrition in its own right and this happens in precisely the same manner as earlier—that is, on a use-it-or-lose-it basis. It is just that there is no direct relationship between the two types of linguistic knowledge.6 Sophisticated 6

The precise role of metalinguistic knowledge in attrition may well go beyond self-correction and have subtle indirect effects on activation levels beyond the conceptual system. This merits a fuller discussion than is possible here (see Truscott, 2015).

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

       



metagrammatical knowledge can even exist with very little grammatical ability at all, as is exhibited by many theoretical linguists who certainly do not have to be in any way fluent in a language to talk coherently about its structure. In cases where attrited speakers have both types, metagrammatical knowledge can survive longer than core syntactic or phonological knowledge; it can also play some corrective role in their performance and hence delay some of the outwardly observable effects of attrition.

. C 

.................................................................................................................................. The framework used for this discussion may be described as incorporating notions of emergent and dynamic systems, albeit constrained within a particular view of the modular mind. This allows many different observed characteristics of cognitive systems to be investigated as part of unified explanation of how the mind works. This certainly holds for the study of language attrition, which is driven in the first instance by the lack of stimulation of a range of linguistic and non-linguistic structures (also known as representations) involved in language use and located with various modular systems. These representations are all located within one or other of the specialized modules and their composition is governed by the local principles pertaining to the module in question. Their growth and attrition, however, is also driven by cross-modular principles, that is, ones that are common to all specialized mental systems. What we then see is a mind that is both very simple and extraordinarily complex. As certain representational properties and interconnections begin to weaken, then those that remain may undergo all kinds of changes as a result. Studying these changes is the task of attrition researchers, but the more explicit and detailed nature of an overarching cognitive framework should facilitate fruitful discussion of their results and their collaboration with other researchers whose own research interests are quite different. This indeed appears to be a necessity, not a luxury, if any major progress is to be made. Readers are invited to take findings reported elsewhere, including those in this volume, and build broader explanations of the data using the architecture of the processing framework that has been discussed in this chapter or some worthy substitute. These explanations need not replace already existing, more specialized accounts, but they should allow an embedding of those accounts within a more general explanation of how the mind works and what is ‘behind’ the data, asking crucial ‘why’ and ‘how’ questions about the nature of memory, representational change over time, and what is really involved in learning and forgetting.

A Many thanks are due to my anonymous reviewers for their useful comments and suggestions which I have in most cases taken account of in this final version.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

  ......................................................................................................................

     Language loss or manifestations of a dynamic system? ......................................................................................................................

 .    . 

. I

.................................................................................................................................. F extended use of a second language (L), many bilinguals experience significant difficulty in accessing their native language (L). For some, it feels as though they have lost parts of their L—a feeling that fits with theoretical claims of language attrition. In this chapter, we situate this phenomenon within the context of memory retrieval processes that may shed light on the nature of the mechanisms underlying language attrition. Our premise is that basic memory processes may provide mechanistic accounts of the reduced access reflected in language attrition. In particular, we consider the hypothesis that L attrition reflects the normal dynamics of the bilingual language system. We begin with a review of the evidence suggesting that developing proficiency in an L has implications for the accessibility and use of the L, with a focus on psycholinguistic laboratory studies. In particular, inhibitory mechanisms have been suggested to play an important role in managing any potential cross-language competition. We consider how normal L-inhibition effects shown in everyday language use may be able to explain larger-scale changes found in immersion settings. We then briefly summarize the literature on retrieval induced forgetting (RIF) that implicates inhibitory mechanisms in the management of representational conflict within models of (monolingual) memory retrieval. We conclude by considering the implications of these two bodies of research for theories of language attrition.

. N    L      

.................................................................................................................................. Traditional accounts of late L learning assumed that the native L is stable and largely autonomous, providing a source of knowledge and transfer to the newly emerging L, but

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

    



benefiting or suffering little from the presence of the L itself. This can be seen, for example, in Clahsen & Felser’s () ‘shallow structure hypothesis’ by which the L carries a privileged status but is not impacted by the presence of an L, or in more modest form in findings that many aspects of the L remain relatively unchanged even after ten years of immersion (e.g., de Bot et al., ). The emerging psycholinguistic evidence suggests a situation that is far more dynamic, with both languages changing over the course of L learning and with immediate consequences of exposure to the L, whether it be in the context of the classroom, language immersion, or the laboratory. In this section of the chapter we illustrate the nature of this evidence and consider how these changes reflect the typical use of two languages and also potentially require a revision of the traditional views about the stability of the L. Three different research strategies have been used to investigate the way that the L changes in the face of L use. One is to examine performance in the L when it is used alone to determine whether L learners and bilinguals differ from monolingual speakers of the L and whether language processing in the L reflects the influence of the L (e.g., Dussias, ). Another strategy is to require L speakers to use the L and then to assess whether there are spillover effects from the L to the L when L use is subsequently required (e.g., Guo et al., ). A third strategy is to compare L speakers who are otherwise similar to one another and to immerse them in contexts that differ in the degree to which the L is required (e.g., Linck et al., ). Here we review the first two of these strategies and later in the chapter we consider the consequences of language immersion. If L speakers retained monolingual-like use of the L, then we would expect that language processing in the L would resemble that of monolingual speakers. L speakers would be as fast and as accurate as L speakers in both comprehension and production tasks and they would be relatively insensitive to the presence of converging or conflicting information across the two languages. A very long line of research on word recognition and production has shown that bilinguals are exquisitely sensitive to cross-language interactions when they are processing their L, suggesting that the activation of the L comes to influence the L (see Kroll, Dussias, Bice, & Perrotti, , for a recent review). But are there similar effects from the L to the L? At the lexical level, an early study by Van Hell & Dijkstra () showed that lexical decision performance in L for Dutch-English bilinguals who differed in whether they were also proficient in French as a third language (L), was not only faster in Dutch when the word was a cognate with English, the L, but also when it was a cognate with French, the L, when the speakers were sufficiently skilled in French. The findings of this study were interpreted to suggest that with adequate proficiency, activation can spread from the L to influence processing in the L, supporting the claims of models such as the Bilingual Interactive Active Model Plus (BIA+; Dijkstra & Van Heuven, ) that the lexicon is fundamentally open across all of the codes that are shared between the two languages. A recent study, using electrophysiological methods in addition to behavioural indices of response time and accuracy, suggests that a high level of L proficiency is not a requirement to observe changes to the L. Bice & Kroll () asked English learners of Spanish at relatively early stages of L study to perform a similar lexical decision task as the one used by Van Hell & Dijkstra (). They recorded via EEG (electroencephalogram) while the lexical decision task was performed and asked whether there was evidence that performance in English, the L of these learners, was affected by Spanish, even for learners at this early stage. Like Van Hell & Dijkstra, they included cognates, translations that share lexical

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



  

form across the two languages. Bice & Kroll found that response times and accuracy were no different for cognates relative to non-cognate control words but that the Event-Related Potential (ERP) data revealed an emerging cognate effect for L learners relative to monolingual English speakers who showed no effects in either measure. In particular, these learners showed a reduced N for cognates relative to non-cognates, suggesting that L processing is already being impacted at this early stage of developing L skill. These results suggest that the brain may outpace behaviour in showing sensitivity to new learning. Like the findings of studies that use ERPs to track L development in typical university classroom learners (e.g., McLaughlin et al., ), the Bice & Kroll results show that not only is learning taking place at very early points in L acquisition, but that cross-language activation can be seen in L as well as L language processing. The two studies we have reviewed on L effects on the L each examined comprehension in the L and demonstrated a pattern of facilitation when the L converges with the L in the case of cognate translations. But what about production? Contrary to the facilitation in L seen for cognate processing, the evidence from production suggests that there may be a cost to speaking the L as a function of using an L actively. In a long line of studies, Gollan and colleagues have shown that bilingual speakers are slower to produce words in their L than monolingual speakers of the same L and also have more tip of the tongue (TOT) experiences than monolinguals (e.g., Gollan & Acenas, ; Gollan, Montoya, Cera, & Sandoval, ; Gollan, Montoya, Fennema-Notestine, & Morris, ). One interpretation of the reduced accessibility of the L in bilingual speakers is that the opportunities to speak each language are necessarily shared over time, thereby reducing the functional frequency of each language relative to that of a monolingual speaker. That process is hypothesized to have particular consequences for retrieving low frequency words. At issue is whether changes to frequency alone can account for the reduced accessibility in production or whether there is also interference from the co-activation of the two languages (see Kroll & Gollan, , for a discussion of these alternative accounts for the observed reduced accessibility of L words). The consequences of L use for the L have also been examined in studies that have monitored bilingual speech in switching contexts. An initial demonstration by Meuter & Allport () in a language production paradigm revealed larger switch costs for speaking words in the L following the L than the reverse. This switch cost asymmetry across the two languages has been studied extensively since the Meuter & Allport study, with a number of different interpretations proposed (see Bobb & Wodniecka, , for a recent review). The account that is most congenial with an attrition perspective is given by Green’s () Inhibitory Control (IC) model. According to the IC model, bilinguals control language production through a combination of top-down inhibition at the level of task schemas (e.g., activation of the schema ‘speak in the L’ would generate inhibition of the schema ‘speak in the L’) and reactive, lateral inhibition between specific representations (e.g., activation of L ‘gato’ generates lateral inhibition of L ‘cat’). In this model, the amount of representational inhibition is directly proportional to the asymmetry in activation—that is, to the amount of competition between the two representations. When bilinguals plan speech in the weaker L, the stronger L is assumed to be inhibited and to a greater extent than the L. If L production is required immediately following speech in the L, then the hypothesized inhibition must be overcome. Slower production in the L following L production has been interpreted to reflect the process of resolving inhibition

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

    



(e.g., Green, ). From the perspective of attrition research, the language switching data show that even a single word-to-word switch may be sufficient to create momentary suppression of the dominant or native language. Although there has been debate about the naturalness of the forced switching task, the point is that selective changes to the L can be imposed relatively easily, even by a simple switch at the lexical level. An early hypothesis about the switch cost asymmetry in the lexical switching paradigm was that it was a feature only of less proficient bilingual use, for those whose L was far more dominant than their L. Costa & Santesteban () found that highly proficient and closely balanced Spanish-Catalan bilinguals produced a pattern of switch costs that was symmetric, in comparison to relatively unbalanced bilinguals whose pattern matched the one reported initially by Meuter & Allport (). They interpreted the different patterns as showing that only learners and less proficient L speakers require IC. However, even the balanced bilingual speakers were slower overall to speak the L than the L, suggesting global inhibition across conditions that require mixed use of the two languages, as is encountered in the language switching paradigm. The importance of this study was to suggest that there may be different components and indices of IC that are related to proficiency and language context. In the case of language attrition, we can speculate that the switching asymmetry might be flipped (with larger switch costs in the L than the L) based on the dominance of the L over the L. In another variant of the lexical switching task, bilinguals are again asked to produce single words but the opportunity to speak each language extends over a longer block of time. In an electrophysiological study, Misra et al. () used this paradigm and found an inhibitory pattern in L speech following naming in the L. Misra et al. asked relatively proficient Chinese-English bilinguals to name the same pictures in Chinese and then in English or in the opposite order. Because the pictures were identical in both languages, they predicted that the language that came second would benefit from this repetition. Using ERPs, they found a facilitatory pattern consistent with repetition priming when the L was named following the L. In contrast, when the L was named following the L, there was greater negativity in the ERP record, a pattern consistent with inhibition. Critically, Misra et al. reported that the inhibitory pattern when the L followed the L persisted despite repeated opportunities to use the L, a context that should have enabled recovery. The results suggest an inhibitory process for the L that is relatively long lasting following L use. A similar pattern was described by Van Assche et al. () using a verbal fluency paradigm. A number of recent neuroimaging studies have contributed evidence on the differential effects of bilingualism for the L that suggests both long lasting structural changes to the L with extended L use (e.g., Parker Jones et al., ) and differential functional consequences for the L that depend on language processing demands (e.g., Guo et al., ). The Guo et al. study demonstrated that there are both local and global contributions to the observed inhibitory patterns in the native language when bilinguals switch from one language to the other. While there is not agreement about the scope or time course associated with different types of inhibition, the evidence converges on the finding that it is the L that is differentially affected by L use. Are the changes we have described in these studies an indication of language attrition? We think not. They are a reflection of the ordinary dynamics and modulation of the native language as bilinguals negotiate their ways through a landscape of language use that

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



  

sometimes requires that one language be used in isolation and other times requires that both languages be used together. Critically, these dynamics are not restricted to the least proficient L speakers but characterize language processing for highly proficient bilinguals. Proficiency affects the observed patterns, but the changes to the L occur even once individuals acquire a very high level of skill in the L. It is quite possible that the long term disuse of the native language will differentially change the weighting of these inhibitory patterns. This is congruent with the claims of the Adaptive Control Hypothesis (Green & Abutalebi, ), according to which language control processes are adjusted based on the communicative demands of the context of interaction. What is not yet known is which of these changes are short term and easily recovered and which are long term and endure even after the native language is reintroduced. It may be important to note that in the context of this discussion we have not reviewed the research on code switching at the sentence or discourse level, a phenomenon of language mixing that characterizes highly proficient bilingual speakers who may switch from one language to the other in the middle of a sentence but following grammatical constraints (e.g., Myers-Scotton, ). Not all bilinguals code switch, but for those who do, there appears to be a modulation of the relations between comprehension and production to enable prediction of an upcoming switch of language and to achieve the coordination that maintains proficient use of both languages (e.g., Fricke et al., ; Tamargo et al., ). It seems likely that the dynamics of native language modulation and attrition are affected by whether bilingual speakers are habitual code switchers or not. Indeed, one claim of the Adaptive Control Hypothesis is that frequent code switching places unique demands on the language control system and thus requires a different set of control processes than those required by single language or dual language (but non-intermixed) contexts. This is a topic that has only recently been investigated with respect to control processes (e.g., Green & Li, ) and merits further investigation.

. R  

.................................................................................................................................. Over the past two decades, research has demonstrated that the retrieval of information from long-term memory leads to reduced accessibility of related but non-retrieved information—a phenomenon described as retrieval induced forgetting (RIF; e.g., Anderson et al., ). In the classic paradigm, participants study multiple exemplars of various categories (e.g., fruits: banana, apple, orange), then are provided a category (e.g., fruits) and asked to produce a category member based on a presented cue (e.g., starts with a particular letter). For example, ‘FRUITS—B_____’ might be presented to cue retrieval of the word ‘banana’. Participants are cued to recall some highly prototypical category members (e.g., banana, apple)—known as the Retrieval Practice plus (RP+) condition— but, importantly, they are purposefully not cued to recall other prototypical category members (e.g., orange)—the Retrieval Practice minus (RP ) condition. Later in the experiment, participants are given a new task that requires retrieval of both the RP+ and RP items, but this time with a novel cue (e.g., semantic judgement, rhyme generation). The key finding is that previous retrieval of the RP+ items leads to a reliable decrease in the

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

    



accessibility of the RP items during this subsequent recall task. In this example, retrieving ‘banana’ during the initial category cued task makes the participant significantly slower and/or less likely to successfully retrieve ‘orange’ during the second recall task. A primary theoretical model of RIF implicates inhibition as the underlying mechanism driving the reduced accessibility of the RP items (e.g., Levy & Anderson, ). The idea is that retrieval of a category member during the initial retrieval task requires the individual to overcome interference from other highly-related category members. This may be achieved by inhibiting the related but non-target RP representations, as evidenced by the reduced accessibility of those items in subsequent, independent tests requiring recall of the RP items (for a recent meta-analysis, see Murayama et al., ; but see Jonker et al., , for a context-based, inhibition-free account). Dozens of follow-up studies have been conducted to clarify the nature of these RIF effects and to identify boundary conditions, such as relative dominance, integration of practised and to-be-remembered items, and semantic similarity of category exemplars (e.g., Anderson, ; Jonker et al., ). A critical finding is that the memory representations in question must be sufficiently in competition during the retrieval process in order for the inhibitory mechanism to be engaged. One feature that may create or limit this competition is the relative dominance or representativeness of the category members. For example, retrieving a more dominant category member, such as ‘banana’, does not induce RIF of a less dominant FRUIT category member, such as ‘tangelo’. However, the opposite pattern is likely: retrieving ‘tangelo’ would likely induce inhibition of the more highly representative exemplar ‘banana’. In other words, RIF effects require that a non-target representation be in direct conflict or competition with the target representation in order for the inhibitory mechanism to be engaged. Extending the RIF theoretical framework to the bilingual case, it seems plausible that the inhibitory mechanism implicated in RIF might be one mechanism that bilinguals rely on to successfully manage cross-language competition. For example, for a native English speaker who is learning Spanish, retrieval of the word ‘gato’ (English: cat) likely induces competition from the non-target but more accessible L representation ‘cat’ and therefore successful L production may require inhibition of the L lexical entry. This account fits with claims of the IC model of bilingual speech production previously discussed (Green, ). Note also that the relative dominance feature of the IC model parallels the relative dominance boundary condition of RIF effects. This raises the possibility that RIF-based inhibition may be one mechanism for the instantiation of inhibition within the IC model framework. Of course other mechanisms are likely at play, given that the IC model’s claims extend beyond the case of retrieval induced inhibition. But this hypothesis that RIF may be supporting bilingual language control addresses recent calls in the literature to better specify the nature of inhibitory mechanisms within the bilingual language system. The theoretical frameworks of the IC model and the RIF literature are indeed congruent, but can RIF occur across languages? As reviewed above, there is ample evidence that when producing a less-dominant L, the speaker may encounter interference from the dominant L and that L use has consequences for subsequent access to the L. In the context of lexical access, some studies have specifically reported cross-language RIF effects, although the results are somewhat mixed. In the first published RIF study with bilinguals, Levy et al. () asked native English speakers learning Spanish to repeatedly name a series of line drawings of common objects (e.g., cat, snake, book) in either the L or L based on an

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



  

external cue (i.e., colour of the line drawing). Half of the drawings were named in the L and the other half in the L, with each drawing being consistently named in one language for a given participant. The critical manipulations within the naming task included the number of repetitions (, , , or ) and the language of naming. The L names for all pictures were then cued in a subsequent retrieval task. They found that naming an object in the L ten times rendered the L translation of that object to be less readily accessible on a subsequent, independent retrieval task (rhyme generation). The results suggest that repeatedly producing a lexical item in the L has a measureable impact on the accessibility of the lexical representation for the L translation. RIF has also been found when focusing on the grammatical gender of lexical items. Morales et al. () asked speakers of Italian (L) and Spanish (L) to name pictures in the L. For some pictures, the grammatical gender matched across languages, whereas for others the grammatical gender mismatched between the L and L. In a subsequent task, the same pictures were presented and participants were required to produce the article of the L picture name. Parallelling the results of Levy et al. (), repeatedly producing the gender for a given item in the L led to subsequently slower naming latencies when producing the L article of items with incongruent grammatical gender relative to items with congruent grammatical gender. This suggests that conflict at the level of grammatical gender can induce RIF as a result of successful retrieval during the initial naming phase. The robustness of these effects has been questioned by some researchers. For example, Runnqvist & Costa () failed to replicate the RIF effect using a similar paradigm as that employed by Levy et al. (). In their study, Runnqvist & Costa included three groups of Spanish bilinguals who varied in their L proficiency. Across all three groups, they found evidence of retrieval induced facilitation: repeatedly naming pictures in the L was associated with better subsequent recall of L translations. Of course, bilinguals vary substantially in their language experiences and contexts of use, which has important implications for how language control mechanisms are called upon (e.g., Green, ), as captured by the Adaptive Control Hypothesis (Green & Abutalebi, ) and detailed above. A clearer mapping of language experiences to specific cognitive mechanisms will inform constraints on when and how RIF mechanisms may support language control.

. T    

.................................................................................................................................. Research on language immersion suggests the L becomes less readily accessible within the L-dominant context. Linck et al. () tested native English speakers approximately four months into a semester abroad in Spain and compared them to a cohort of classroom-only learners enrolled in language courses back at the home university who were at a similar L proficiency level. Participants completed a battery of tasks measuring lexical access in the L and L. On both comprehension (translation recognition) and production (semantic fluency) tasks, participants in the immersion context demonstrated reduced access to the L, as evidenced by the elimination of L-related distractor interference on the comprehension task and a significant reduction in their L verbal fluency. A subset of the immersed

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

    



learners were then retested approximately six months after returning to the L-dominant home university; although L verbal fluency rebounded to a level equivalent to the classroom learners, participants continued to show reduced L interference on the comprehension task. The results were interpreted to suggest that the L-rich immersion context encouraged inhibition of the L. The retest results suggested that the immersed learners developed IC skill that they could deploy to mitigate L interference during comprehension. Baus et al. () also studied the effects of immersion on L lexical processing using a longitudinal research design. They tested immersed learners (native German speakers learning Spanish in Spain) upon their arrival to the L context and four months later, prior to their departure. On a picture naming task, the immersion experience was associated with slower naming latencies in the L. That is, the L was less accessible at the conclusion of the immersion period, and this effect was most apparent with low frequency non-cognates. Differences on an L semantic fluency task (similar to the task used by Linck et al., ) were also modulated by cognate status: semantic fluency increased for cognates but decreased for non-cognates (i.e., participants produced more cognates and fewer noncognates in the departure testing session relative to the arrival testing session). These patterns of reduced L access are congruent with an inhibition account, by which one might expect cognates to be relatively immune to inhibition-based reductions in accessibility due to the strong phonological and/or orthographic overlap with the L. That is, L and L forms for cognates are (nearly) identical and therefore would not be expected to compete for selection—a necessary condition for RIF effects to emerge. Or from another perspective, non-cognates and lower frequency words might be more vulnerable to retrieval-based inhibition mechanisms, as there is greater potential for cross-language competition between the word forms. We note that these effects on lower frequency words and non-cognates are also congruent with the claims of the weaker links hypothesis (e.g., Gollan et al., ). As we noted earlier, this account does not require inhibitory mechanisms, but rather assumes that reduced accessibility of the L is driven by a weakening of the links between conceptual and form representations as a consequence of the reduced frequency of using L words in the immersion context. Although the two studies reviewed here cannot adjudicate between these two accounts, taken together these results are congruent with the inhibition hypothesis, with RIF providing one specific mechanistic account of the dynamics of these inhibition effects.

. I        

.................................................................................................................................. Given the findings of research on RIF, the implication we consider here is that language attrition phenomena may emerge, in part, from the underlying dynamics of the memory system. Regardless of the specific mechanism behind RIF (be it inhibition or something else), when considered in the bilingual context, these results suggest that repeated retrieval of the non-dominant language may lead to reduced accessibility of the

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



  

corresponding lexical representations in the dominant (native) language. This is an important consideration, as it underscores the insight, emerging from the literature on bilingual language processing, that developing proficiency in an L has consequences for the L—even for skilled bilinguals who show no obvious signs of L attrition. Indeed, there is growing evidence that proficient bilinguals may not function like monolinguals in any of their known languages (e.g., Ameel et al., ), lending support to Grosjean’s (, p. ) earlier warning that ‘the bilingual is not two monolinguals in one person.’ This directly contradicts the previously held beliefs that the L is unchanged by the learning of an L (e.g., Clahsen & Felser, ) but is compatible with a Dynamic Systems Theory (DST) view of language development, by which the two language systems are assumed to interact and cause non-linear change over time (e.g., de Bot et al., ). L attrition may be seen as representing the extreme end of the spectrum of L changes that co-occur with increasing proficiency in an L. The finding that the process of learning an L has measureable impact on L processing highlights an important opportunity to develop a shared theoretical framework that incorporates features of theories of language attrition (with DST perhaps also as a framework to study attrition, as it views language acquisition and attrition as two sides of the same coin), models of bilingual language development, and models of the dynamics of memory retrieval. In such a framework, factors known to drive more subtle changes in memory representations, as demonstrated by RIF, may also scale up to provide an explanatory mechanism for broader changes in the bilingual’s dynamic language system. This framework is congruent with existing models of bilingual processing. As previously discussed, one claim of the IC model is that inhibitory mechanisms originating from outside of the language system support bilingual language control (Green, ) and certain claims of the IC model parallel the dynamics of RIF (e.g., relative dominance). This offers specific points of overlap that can serve as anchor points when integrating these distinct models of memory and language processes. Similarly, the Adaptive Control Hypothesis (Green & Abutalebi, ) may help us to formulate specific predictions about the impact of specific features of a bilingual’s language use patterns and language context on mechanisms underlying language attrition phenomena. In the context of the current discussion, one might hypothesize that RIF could contribute to the interference control component of the Adaptive Control Hypothesis, with both single language and dual language interactional contexts placing increased demands on this specific aspect of control. Clearly, RIF is not likely to be the only memory retrieval mechanism responsible for language attrition. But existing models of language attrition—particularly those that take a dynamic systems approach—may benefit by incorporating general mechanisms of memory retrieval, such as RIF, as one of multiple mechanisms driving language change. What we hope to have accomplished in this review is to point out these opportunities for future development of such a unified theoretical framework.

. C

.................................................................................................................................. There is a growing research literature on language attrition phenomena, as seen in this volume. The findings from psycholinguistic studies of bilingualism reviewed above suggest

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

    



that the bilingual language system is dynamic and constantly adapting to changing demands placed on the bilingual. We suggest that L attrition can be viewed as reflecting the dynamic nature of the language system operating within the human cognitive architecture. Within this framework, shifts in language accessibility are predicted to result from the processing mechanisms that allow successful language engagement and disengagement within the bilingual language system—and features of the memory system more broadly.

A The writing of this chapter was supported in part by NSF Grants BCS-, OISE-, and OISE-, and NIH Grant HD to J. F. Kroll.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

  ......................................................................................................................

                      Evidence from eye movements ......................................................................................................................

 . ,  . ´  ,  ,  ´  

M studies published over the past two decades have used measures of online language processing to show that the languages of a multilingual speaker are engaged in constant, pervasive interaction, and that these interactions are bidirectional: not only does the first language affect the second language, but the second language also influences the first. Remarkably, these bidirectional influences can be seen at every level of language use. In immersion contexts, second language (L) speakers experience reduced access to the first language (Linck et al., ), and extensive contact with the second language can affect first language (L) phonology (e.g., Flege, ; Flege & Eefting, a), syntactic processing (Dussias & Sagarra, ), sentence production (Hartsuiker et al., ), and naming performance of common objects (Malt & Sloman, ). There is now growing evidence demonstrating that the seemingly stable L system is open to influence when individuals learn and become proficient in an L (e.g., Gollan et al., ; Ivanova & Costa, ; Runnqvist et al., ; Kasparian & Steinhauer, a). The insight that there are bidirectional influences between the two languages of a bilingual speaker has implications for the study of language attrition; because the linguistic system is dynamic and interactive at all levels of representation, it is reasonable to expect that variations in language dominance, use, and exposure can lead to changes in the system—some subtle, some significant. In our view, these changes are not necessarily the result of what has increasingly been labelled ‘attrition’ in the literature, but rather are a natural consequence of the interactive nature of the languages at play. The goal of this chapter is twofold. The first is to survey recent contributions to the research on bilingual language processing that serve to demonstrate how exposure to a second language, even for

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

    



a brief period of time, can impact processing in the native language. Given this evidence, we will argue that claims of language attrition may not be as clear-cut as one may think when online language processing is taken into account. A second goal is to show that eye-tracking is a premier behavioural method by which we can come to understand fine-grained changes in online language processing. In doing so, we hope to illustrate how the study of online language processing via eye-tracking can help to clarify issues in language attrition (e.g., Berends et al., ).

. T      

.................................................................................................................................. One finding that remains uncontroversial after almost two decades of psycholinguistic research with bilingual speakers is that the two languages of a bilingual speaker are active, even when the intention is to use only one language. When bilinguals read, when they listen, or when they prepare to speak in one of their two languages, the language not in use is also active (Kroll et al., ; for an extensive review, see Dijkstra, ). For example, when bilinguals are asked to name cognate words (such as ‘piano’ in Spanish and English) in only one language, the language not in use is also active and influences performance, resulting in a cognate facilitation effect (i.e., bilinguals produce and recognize cognates faster than non-cognates). Although this work could be criticized on the grounds that the ‘out-of-context’ nature of word recognition and word naming experiments may itself induce the observed effects, extensive research also demonstrates that it is much more difficult than we might have thought to reduce or eliminate the presence of cross-language activity. First, the same cognate facilitation effects observed in out-of-context word recognition experiments can be observed when cognate words are embedded in sentences (e.g., Schwartz & Kroll, ; Duyck et al., ; Van Hell & De Groot, ; Libben & Titone, ). Second, the cross-language competition is not restricted to the lexicon. Although more research has examined these interactions for words than for sentences, recent studies show that the grammars of the bilingual’s two languages are also open to one another in a manner far more permeable than what might have been predicted. Studies within the syntactic priming literature have shown that monolingual speakers are more likely to produce a sentence using a grammatical structure if they have just produced a sentence using that same structure (Bock, ). In bilingual research, the main question has been whether priming will be observed when switching languages from the prime to the target sentence. Although there are some exceptions (e.g., Bernolet et al., ), the general finding is that cross-language priming can be obtained (Hartsuiker et al., ). In general, the parallel activity of a bilingual’s two languages creates cross-language interactions that influence performance at every level of language, including phonology (e.g., Spivey & Marian, ; Jared & Kroll, ; Marian & Spivey, ; Ju & Luce, ; Blumenfeld & Marian, ), orthography (e.g., Dijkstra & Van Heuven, ; Van Heuven et al., ), syntax (e.g., Hartsuiker et al., ), and meaning (e.g., Sunderman & Kroll, ). The parallel activation of the bilingual’s two languages is not restricted to languages that share structural or functional characteristics. An example of cross-language interaction

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



, ´  , ,  

between structurally distinct languages is the case of bimodal bilinguals, who use a spoken and/or written language and a signed language. Morford et al. () found that the speed with which deaf American Sign Language (ASL) English bilinguals made semantic relatedness judgements for English word pairs (movie-paper) was affected by whether the hand formations of the ASL translations were themselves similar or different in form. This result is remarkable because no ASL was present during the experiment. In an earlier study involving a logographic and an alphabetic language, Thierry & Wu () found early neural sensitivity (measured via the recording of EEG (electroencelphalogram)) to a word’s Chinese translation when Chinese-English bilinguals performed a semantic relatedness judgement task in English only. These bilinguals performed the semantic relatedness task without Chinese actually being present during the experimental session, and in a context in which they were immersed in an English-speaking environment. This demonstrates that the languages of a bilingual speaker are engaged in constant, persistent interaction, even in the presence of obvious differences between the bilingual’s two languages (e.g., Emmorey et al., ; Hoshino & Kroll, ). What’s more, the results in Morford et al. and in Thierry & Wu show that conscious awareness of the other language is not required to observe interactive effects between the L and the L, suggesting that parallel activation is a feature of the architecture of the bilingual language system (Morford et al., ). The finding that cross-language activation and interaction is pervasive in bilinguals has ramifications for the study of language attrition. This is especially true in the case of language processing, which can be modulated with relative ease. At one extreme end, some bilingual speakers experience an abrupt shift in language use towards their L, perhaps due to marriage, immigration, or education. After decades of lack of use of the L, these speakers may experience difficulty in lexical retrieval and large-scale interference from the L (Schmid, a; Schmid & Keijzer, ; Schmid et al., ), and in many cases can be labelled as having undergone L attrition. At the other extreme, study abroad students, after mere weeks of immersion, experience subtle but quantifiable differences in lexical access vis-à-vis non-immersed peers in their first language (Linck et al., ). Is the latter case also a clear-cut example of attrition? In part, the field lacks a clear consensus on () what attrition is; () how soon attrition effects can begin; and () how permanent the effects of attrition are. The goal of this chapter is to survey recent contributions to the research on bilingual language processing that serve as illustrations of how exposure to a second language, even for a brief period of time, can impact processing in the native language. Critically, we take the evidence of change to the native language not as an indication of permanent language attrition or loss, but rather as evidence that variation to the native language is a natural consequence of the intrinsic plasticity of the linguistic system (Kroll et al., ). In fact, monolingual speakers demonstrate this plasticity too, adapting to and extending the use of unfamiliar structures from a language variety different from their own variety (e.g., Fraundorf & Jaeger, ), and even showing reduced sensitivity when presented with ungrammatical structures for a short period of time (Hopp, a). These effects, both in bilingual and monolingual speakers, may be attributable to priming, but to the extent that the facilitatory effects of priming during language comprehension are not limited to known structures but also affect entirely unknown structures (Fraundorf & Jaeger, ), it suggests that priming is not necessarily a reflection of transient activation (Pickering & Branigan, ; Traxler & Tooley, ); rather, it can strengthen the mappings between

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

    



meanings and linguistic structures (Leonard, ), thereby leading to the learning and modification of linguistic representation (e.g., Bock & Griffin, ; Konopka & Bock, ; Savage et al., ; Ferreira et al., ; Hartsuiker et al., ; Fine & Jaeger, ). We begin with a brief overview of eye-tracking and how eye movements are a fine methodological tool to investigate shifting sensitivities to linguistic cues in sentence processing. We then illustrate our approach to language interaction by examining two instances of the reconfiguration of the L, which we argue are natural extensions of a dynamic linguistic system—monolinguals to a lesser extent demonstrate this dynamicity as well. Then, we ask whether eye-tracking can illustrate the reactivation of L processing strategies after re-exposure in a compressed timeframe.

. E-         L  

.................................................................................................................................. How is eye-movement data informative to studying changes in the processing of the first language? Theories of sentence processing are generally interested in the online or incremental nature of comprehension processes. As soon as each word is encountered, readers and listeners are assumed to make structural decisions about how to integrate each word within the ongoing syntactic structure. The incremental nature underlying sentence processing allows researchers to use eye movements and fixations to make inferences on the ease of integration into prior sentential context. Eye-tracking has been used for both auditory and reading comprehension studies and with a variety of populations. The logic that underlies these studies is quite simple: there is a plausible link between what readers/ hearers look at and for how long as they process linguistic input. Thus, when reading, readers linger longer and regress back to words that are harder to process as compared to easy or predictable words (see Rayner, ; Clifton & Staub, ; Keating, , for comprehensive reviews). Similarly, researchers working within the auditory domain utilize the visual world paradigm (Tanenhaus et al., )—an eye-tracking technique whereby participants listen to auditory stimuli in the presence of a visual scene—to study the timecourse of spoken language processing. Because listeners will look at relevant referents as they are named in the auditory input, researchers can manipulate the presence/absence of linguistic features (e.g., phonetic overlap, grammatical gender, case marking, anaphora, givenness, etc.) to investigate the time course of looks to a target item (see Tanenhaus & Trueswell, ; Altmann, ; Huettig et al., , for reviews). In what follows, we present a necessarily brief overview of the central findings of eye-tracking research in reading and auditory processing. Over three decades of eye-movement research have shown that when eye movements are recorded during reading, there are systematic relations between fixation durations and the characteristics of the fixated words (Rayner, ; Just & Carpenter, ; Ehrlich & Rayner, ). Readers spend more time fixating on harder words (e.g., lower frequency words) and on more important words (e.g., nouns and verbs) than on easier words. Longer words are also more likely to be fixated on than shorter words, and words that are likely to be skipped

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



, ´  , ,  

are short, function words. We also know that eye movements are particularly sensitive to textual variables. For example, when the text becomes more complex or contains uncommon or contextually implausible words, eye-fixation duration increases, and saccade length (i.e., small jumps made by the eye to move through text) decreases (Duchowski, ). This variation in fixations can be captured in the gaze duration of readers (i.e., the initial amount of time a reader spends in a region, as defined by the researcher—typically as single words or phrases—from first entering it until the eyes move to another word). Word frequency has the most influential effect on gaze duration: a high-frequency word decreases gaze duration (O’Regan & Lévy-Schoen, ; Rayner & Pollatsek, ) compared to a lower-frequency word, even when length, number of syllables, meaning, and sentence frame are controlled for. Unpredictable words also have immediate effects on fixation duration. Readers tend to look at unpredictable words longer than predictable words and they skip over predictable words more frequently than unpredictable words (Ehrlich & Rayner, ). Moreover, when disambiguating information in a structurally ambiguous sentence is inconsistent with the syntactic interpretation assigned by a reader, there is considerable disruption in eye movement. Thus, participants reading a syntactically ambiguous sentence show long fixation durations at the disambiguating region, launch regressive saccades from the disambiguating point to the syntactically ambiguous constituent, or reread the sentence for a second time. The fact that inconsistencies associated with the structural analysis of a particular word (or collection of words) are noticed by readers as soon as they arise suggests that recordings of eye movements can be very informative when studying the structural decisions that people make during reading. Another major advantage of the eye-movement recording technique is that it allows researchers to obtain evidence about what is happening during the comprehension of a sentence moment by moment, as processing unfolds, without significantly altering the normal characteristics of either the task or the presentation of the stimuli. Eye movements are a normal characteristic of reading, and while eye-movement records are collected, participants are free to move their eyes along the printed line of text. In a similar vein, eye-movement records are highly informative to auditory comprehension (i.e., the visual world paradigm, Tanenhaus et al., ). As in the case of reading studies, lexical (e.g., lexical frequency, phonological cohorts, and neighbourhood density) and grammatical manipulations (e.g., grammatical gender, cloze probability, verb semantics) during auditory processing may induce faster and greater looks to target stimuli in cases of facilitated processing, and slower and reduced looks to target stimuli in cases of delayed processing. The visual world paradigm has helped to elucidate that listeners do not wait until the end of a word before they begin to process its meaning (Allopenna et al., ). Similarly, listeners can anticipate upcoming target items based on the verb’s meaning (Altmann & Kamide, ). Listeners are also adept at using grammatical cues such as grammatical gender, word order, or case marking to anticipate target items in informative contexts (e.g., Lew-Williams & Fernald, ; Hopp, a, for grammatical gender; Kaiser & Trueswell, , for word order; Kamide, Altmann, & Haywood, ; Knoeferle et al., , for case marking). Because participants are freely scanning a visual scene as they listen to auditory input, the visual world paradigm is particularly useful for investigating online sentence processing in populations with reduced literacy, speakers of minority languages and stigmatized forms of speech, as well as language attriters. In the latter case, the visual world paradigm may be a powerful tool for language attrition researchers

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

    



because it does not require the use of judgement or acceptability tasks, and most participants are not conscious of how linguistic stimuli may impact how they inspect a visual scene. Recent advancements in eye-movement technology also make available eye-tracking equipment that is extremely versatile, and replaces traditional, fixed eye-tracking systems with more flexible head-mounted systems, or remote systems that do not require the use of a headband or head (i.e., chin or forehead) support. In addition, to obtain the dependent measure, participants are not required to perform a secondary task (such as a button press) that might disrupt the normal comprehension process. Furthermore, thanks to several decades of eye-movement research, we have a very good understanding of the amount of visual information processed while our eyes fixate on text or when processing images.

. I 

.................................................................................................................................. Eye-movement records have been employed in second language and bilingualism research primarily to ask how proficient speakers of two or more languages manage the presence of two languages in a single mind. If it were the case that these individuals could be characterized as two monolinguals in one head, then this question would not be very interesting. However, as stated earlier, the available evidence from the word recognition and sentence processing literature suggests that when two linguistic systems are housed in a single brain, they interact closely with one another, and these interactions influence the way in which L speakers read and understand spoken words in each of their languages. Below, we review several studies that will serve as illustrations about how eye-tracking methodology can be used to study these interactions.

.. Reconfiguration of the L Valdés Kroff et al. () examined whether intense contact with Spanish–English codeswitched speech had consequences for the processing of grammatical gender in L Spanish. In many bilingual communities, speakers regularly switch from one language to another, often several times in a single utterance. The ability to engage in fluent code-switching is a hallmark of high proficiency in two languages (Miccio et al., ), as successful and fluent code-switching requires a high degree of knowledge of and sensitivity to the grammatical constraints of both languages. Grammatical gender was examined because it is a lexically specified feature of nouns that triggers syntactic agreement with other function and openclass elements, such as determiners and adjectives. Grammatical gender was interesting for yet another reason. In the production literature of Spanish–English code-switches, one widely attested pattern is that when a code-switch occurs within a noun phrase composed of a determiner and a noun, the determiner overwhelmingly surfaces in Spanish and the noun in English, e.g., el building and not the edificio (Liceras, Fernández Fuertes, et al., ; Herring et al., ). Researchers have also documented a production asymmetry in grammatical gender assignment in these mixed noun phrases. The Spanish masculine article el (‘the’) surfaces with English nouns regardless of the grammatical gender of the noun’s translation equivalents—for example, el juice (Spanish ‘jugo’, masculine), el cookie

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



, ´  , ,  

(Spanish ‘galleta’, feminine). In contrast, mixed noun phrases involving the Spanish feminine article la (also translated as ‘the’) are rare and occur in restricted environments, such that only English nouns whose Spanish translation equivalents are feminine surface with la in code-switching (e.g., la cookie but not la juice; Jake et al., ; Otheguy & Lapidus, ). These production distributions in Spanish–English code-switching stand in contrast to monolingual Spanish, where the grammatical gender of a noun and its accompanying article must obligatorily match, and where masculine and feminine nouns are evenly distributed (Eddington, ; Otheguy & Lapidus, ). As mentioned above, Valdés Kroff et al. () investigated whether L Spanish–L English speakers who had been immersed in a code-switching environment used grammatical gender information encoded in Spanish articles to anticipate the gender of an upcoming noun (as Lew-Williams & Fernald, , have shown with monolingual speakers of Spanish). Given the asymmetry observed in production data, it seemed plausible that the gender marking of articles would facilitate to a lesser extent the processing of codeswitched speech. To investigate this, the eye movements of Spanish-English bilinguals were recorded. Participants saw two pictures that represented objects whose nouns were either of the same or different grammatical genders, and listened to code-switched sentences (i.e., Hay un niño que está mirando el candy/‘There is a boy looking at the candy’). Words were spoken with a Spanish article that either matched the gender of the word’s Spanish translation equivalent (e.g., el candyMASC , Spanish el caramelo) or did not match (target: el candleFEM, Spanish la vela). A control group of monolingual speakers of Spanish was also recruited. Participants were asked to listen to each sentence and to click on the named object. Where the grammatical gender of the Spanish names for each of the pictures was different between the two pictures, the gender information in the article was informative. Monolingual Spanish speakers showed the expected anticipatory effect on masculine and feminine different-gender trials. Results for the Spanish-English bilingual speakers revealed an anticipatory effect, but only on different-gender trials where the auditory stimulus was feminine. When the auditory stimuli were masculine in different-gender trials, participants did not launch anticipatory looks but rather waited to hear the target noun, meaning that they did not use masculine articles as cues for anticipatory processing. At first glance, these results could potentially be consistent with an attrition account; after all, the bilingual participants experienced what appeared to be a permanent change to their L linguistic system (cf. Bergmann, Meulman, et al.,  for a discussion that L attrition studies in post-puberty migrants have not found evidence for weakening of gender); however, a follow-up production study with a subgroup of the same participants revealed that these bilinguals were highly accurate at producing the correct gender in Spanish-only noun phrases. The findings suggest that while there may be constraints on the sort of cross-language exchanges in bilinguals, the presence of these influences suggests a dynamic language system that changes in response to contact with other languages and to how the two languages are used among other bilingual speakers. In an eye-tracking-while-reading study, Dussias & Sagarra (; see also Fernández, ) investigated the effect of intense contact with English on the resolution of syntactically ambiguous relative clauses in Spanish. Native Spanish and native English speakers differ in how they interpret temporarily ambiguous relative clauses such as Alguien disparó al hijo de la actriz que estaba en el balcón (‘Someone shot the son of the actress who was on the balcony’). When asked the question ¿Quién estaba en el balcón? (‘Who was on the

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

    



balcony?’), monolingual Spanish speakers typically respond with ‘the son’ (i.e., high attachment preference) and monolingual English speakers with ‘the actress’ (low attachment preference, e.g., Carreiras & Clifton, ). Dussias & Sagarra found that L SpanishL English bilinguals immersed in an English-speaking environment for a prolonged period of time favoured the low attachment strategy associated with English when reading in Spanish (their native language) in contrast to a non-immersed but proficiency-matched bilingual group who continued to favour a high attachment strategy. Importantly, overall comprehension was not affected (non-immersed group: %, immersed group: %) nor was average reading time. Although not reviewed here, the influence of the first language on the second is a welldocumented and robust finding (see Schwieter, , for an extensive overview). The studies reviewed in this section demonstrate the lesser-known influence of the second language on the first. Taken together, these studies reflect the permeability of the two linguistic systems and not a gradual loss of sensitivity to L processing per se. How do we reconcile these bidirectional influences with an attrition account? Potentially, one criterion is how permanent these influences may be. In the next section, we review whether L speakers can be shifted back into demonstrating monolingual-like processing through reactivation.

.. Reactivation of the L Several recent studies have provided examples of how an L assumed to have undergone attrition can be reactivated despite years of non-use. Park () examined the effects of re-exposure in Korean adoptees living in Sweden. The adoptees had been exposed to Korean before adoption for differing lengths of time, and were all learning Korean (effectively) as an L after years of non-use. To test whether or not re-exposure to Korean was affected by their experience during early childhood, Park administered two phonetic perception tasks to compare these adoptees to L Swedish speakers who were L learners of Korean. While the results were marginally significant, the trends were in the expected direction: the Korean adoptees performed better than their Swedish counterparts in distinguishing both vowels and stops in Korean. Likewise, those who had been adopted at a later age showed the greatest performance overall. Thus, the author concludes that, even after years of non-use and likely attrition, the native language can be ‘reactivated’ with new exposure, resulting in subtle yet detectable differences (see also Flores, Chapter , this volume). More recently, Choi, Cutler, & Broersma () demonstrated that the benefits reported for adoptees’ perception of the sounds of their birth language also transfer to production. In the area of sentence processing, an eye-tracking study conducted by Chamorro, Sorace, & Sturt () examined the effects of language attrition and re-exposure to the L on the use of the differential object marker (DOM) in Spanish. Three different groups were tested: Spanish monolinguals with little to no knowledge of English (who had recently arrived in Edinburgh, UK), and two groups of L Spanish attriters who had been living in the UK for more than five years and were near-native speakers of English. One of the latter groups of language attriters had been re-exposed to Spanish before participating in the experiment. The results were striking: all three groups showed sensitivity to violations involving the DOM in an offline judgement task but only the attrition group who had not

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



, ´  , ,  

been re-exposed to Spanish showed reduced sensitivity in online reading task using eye-tracking. Remarkably, the group that had been re-exposed to Spanish exhibited sensitivity patterns similar to the monolingual group after just one week of exposure. But do these findings extend to other language domains? A series of on-going eyetracking experiments seems to suggest so. We mentioned earlier a study by Dussias & Sagarra (), who found that bilingual speakers experienced changes in the processing of their L, and which the authors attributed to exposure to their L. Given this finding, one might wonder whether proficient bilinguals who had adopted processing strategies of the second language when parsing their L can return to using L parsing after a period of reexposure. To examine this question, Carlson et al. (in prep.) first ran a pre-test to identify Spanish-English bilinguals as ‘high attachers’ or ‘low attachers’ while they processed syntactically ambiguous relative clauses (e.g., Alguien disparó al hijo de la actriz que estaba en el balcón ‘Someone shot the son of the actress who was on the balcony’). Next, they exposed participants to a five-day ‘intervention study’ (that ran for one hour each day), during which they read short paragraphs containing relative clauses in which the syntactic ambiguity was resolved opposite to their natural attachment preference. That is, participants who favoured high attachment received a low attachment treatment, and those who favoured low attachment received a high attachment treatment. In addition, within each group half of the participants were randomly selected to receive the intervention in Spanish and the other half in English. Participants returned to the lab after the intervention study to participate in two subsequent eye-tracking studies: one that assessed the immediate effect of the intervention and one that assessed the effect of the intervention a week after it was completed. The findings showed that those participants who originally preferred high attachment switched to a low attachment preference after the intervention, and maintained the low attachment preference by the second post-test. Similarly, participants who originally showed a low attachment strategy switched to high attachment. Critically, whether the intervention was in Spanish or in English did not affect the pattern of results, suggesting that structures that are shared, by virtue of the fact that they overlap between the two languages, can be fairly easily ‘lost’ but also ‘reactivated’. Importantly, the shared nature of the structure seems to confer a particular advantage: exposure to the structure itself rather than exposure to the language can revert an attachment preference, underscoring the highly dynamic nature of the language systems and not prima facie language loss. The studies reviewed here show that when online processing is taken into account, it can be said that even individuals considered to be language attriters still maintain sensitivity to potentially vulnerable elements in their L. This suggests that online measures of language processing can provide a more complete picture of language use in populations of possible attriters, and that even brief periods of re-immersion in the first language can shift processing strategies back to monolingual-like preferences.

. C

.................................................................................................................................. In this chapter, we have reviewed some of the recent developments on the permeability of the native language system, focusing primarily on the sentence comprehension. Our review is far from exhaustive because this is an active area of research that is rapidly expanding,

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

    



but our hope was to illustrate the view that changes to the native language are better understood as an indication of the openness of the networks that support language knowledge and language use in general, and as a consequence of a fundamental trait of the architecture of the linguistic system. Although the malleability of the native system can be evidenced in both monolingual and bilingual speakers, bilinguals represent a ‘natural experiment’ on the openness of the native language system; changes to the native language in response to the experience and context of use of an L result naturally when individuals come in contact with and use more than one language. At the same time, the use of eyetracking has provided a unique set of insights about emerging questions that we have only had an opportunity to review briefly in the present chapter. If we step back from the specifics of the experiments that we have discussed, the overarching theme in our review is that the bilingual’s two languages are remarkably open to one another, with cross-language interactions that persist from word to sentence processing, and with changes to the native language that provide a model for testing claims about the plasticity of cognitive and neural representations. One unanswered question moving forward is whether there is a threshold when this natural permeability and interaction between the two linguistic systems becomes a more permanent loss. The bilingual’s two languages operate within a rich cognitive network that supports the comprehension processes on which we have focused. It is beyond the scope of our chapter to address this question; however, we hope that in the next phase of research on L attrition, the view that we have presented here represents a counterpoint to constrain what research defines as attrition.

A The authors thank the reviewers for very helpful and insightful comments, and for their close reading of the chapter. The writing of this chapter was supported in part by NSF grants BCS-, NSF grant OISE-, NIH Grant HD, and NIH Grant HD to Paola Dussias.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

  ......................................................................................................................

       ......................................................................................................................

 ,  ,    

. I

.................................................................................................................................. C who are raised in migrant families are mostly exposed to their first language (L) from birth. This language is also referred to as their home language (Valdés, ) or heritage language (Valdés, ) (see Part V of this volume). Exposure to the language of the country they reside in typically starts later, making this language their second language (L), and increases in frequency when children grow older and spend more time outside their homes. Previous research has demonstrated that the delayed onset of L exposure can lead to overlap between the language profiles of typically developing (TD) successive bilingual children in the L and monolingual children with Developmental Language Disorder (DLD) learning this language as their mother tongue (e.g., Håkansson & Nettelbladt, ; Grüter, ; Paradis, ).1 In the last decades, a mounting number of studies in the field of DLD have therefore taken a particular interest in bilingualism (Paradis, ; Armon-Lotem, ), mainly focusing on successive bilingualism as a result of migration. Research on the intersection of bilingualism and DLD has been conducted for several reasons. First, the partially overlapping language profiles of successive bilingual children without DLD and monolingual children with DLD can provide more insight into the nature of DLD. For instance, if the errors of children with DLD are similar to those made by bilingual children learning an L, then this could suggest that the impairment affects input In keeping with a recent proposal (Bishop, ), we decided to use the term DLD. Other commonly used terms are Specific Language Impairment (SLI), Primary Language Impairment (PLI), or Language Impairment (LI). 1

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

     



processing. Within this view, impaired input processing (in children with DLD) and reduced input frequency (in bilinguals) have similar effects on children’s language outcomes (see Paradis, , for a more detailed discussion on this topic). In addition, another more practical, but not less important, reason is related to how the effects of language impairment can be isolated and teased apart from effects of delayed L exposure in diagnostic settings. This is a pressing issue, due to the large numbers of bilingual children on clinical caseloads and lack of appropriate instruments for assessing language proficiency in a bilingual context (Mennen & Stansfield, ; Kohnert, ). Finally, besides studying the overlap between monolingual DLD and bilingual TD described above, investigating the combined effect of bilingualism and DLD in children who are both bilingual and diagnosed with DLD is also important, for theoretical as well as applied reasons. For example, do effects of bilingualism and DLD on language accumulate, and does bilingualism alleviate or aggrevate the symptoms of DLD? The above issues and questions concern the L of successive bilingual children. Less attention has been paid to the L, and therefore to the possibility of L attrition, which is the particular case of language loss at the individual level (Montrul, , p. ). Like a delay in the L, attrition in the L of successive bilinguals might lead to a diagnostic confound: effects of L attrition on children’s language proficiency may resemble effects of DLD on skills in that particular language. Effects of L attrition and effects of DLD could also accumulate and it is possible that DLD influences the quantity and/or quality of L attrition in bilingual children with DLD. In this chapter, we aim to demonstrate that knowing more about attrition is important for accurately diagnosing bilingual children whose language abilities raise concerns. We discuss whether the effects of attrition and DLD are expected to show overlap, whether they would accumulate, and whether DLD would influence attrition. In so doing, we often refer to research that has looked at the L of bilingual immigrant children, as much less is yet known about their L. Section . starts with parallels between the effects of DLD and successive bilingualism on children’s language abilities. Section . continues with the combined effects of DLD and bilingualism on the L and L. Section . highlights factors that may increase the chances of attrition in immigrant children and that may be responsible for the diagnostic confound with DLD. In Section ., issues are discussed that arise when the aim is to isolate the effects of DLD from those caused by attrition. We conclude by summarizing the main issues and suggesting promising avenues for future research in Section ..

. O :  D L D    

.................................................................................................................................. DLD is an impairment specific to language that affects about % to % of all children (Tomblin et al., ). DLD is commonly defined as an impairment in the comprehension and, more prominently, in the production of language in children; the area that is affected most persistently by the impairment is grammar (Leonard, a). A robust ingredient of

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



, ,   

the definition concerns exclusion criteria, specifying the presumed absence of another clinical condition that can adequately explain the language difficulties (Leonard, a; Bishop, ). The inclusion criterion, the language disorder itself, is typically defined in comparison with the language skills of TD. Besides language difficulties, various studies have demonstrated that children with DLD have reduced processing abilities (Leonard et al., ). These limitations surface both in the verbal/auditory and in the visual domain (e.g., Ebert & Kohnert, ; Vugs et al., ; Henry & Botting, ). However, the magnitude of the effect of DLD on verbal processing is often larger than the effect of DLD on visual processing, and, in addition, findings in the visual domain are mixed. For instance, sustained selective attention limitations are found for auditory but not for visual stimuli (Spaulding, Plante, & Vance, ). An important issue, when establishing a benchmark for DLD, is what the nature of the difference between language-impaired children and TD children actually is. Leonard (a) discusses different options. First of all, it could be said that DLD represents a general language delay (‘not only [ . . . ] late emergence of language but also slower than average development’ (Leonard, a, p. )). However, it appears that language development in DLD is not a carbon copy of that in younger TD children. Instead, the difference between DLD and TD may be described as follows: () DLD may be characterized by an abnormal frequency of the same errors also found in younger children. In other words, children with DLD produce more errors than younger TD children, but the errors are the same and emerge in the same language domains. Alternatively, () DLD may represent an uneven profile, in that some domains are more impaired relative to TD development than others, even though the errors are generally similar. Finally, () the difference between children with DLD and TD children may be qualitative in nature. A qualitative difference would imply that children with DLD not only show delayed development, but also deviant development and, crucially, produce errors that are not or hardly ever found in TD children. Rice (, p. ) describes children with DLD as having a ‘delayed departure’, a characterization that refers to a quantitative rather than a qualitative difference. This delayed departure may, however, (come to) affect certain aspects of language, such as tense morphology, more than others, leading to an uneven profile (as described above) or, as Rice (, p. ) puts it, a ‘delay within a delay’. The notion of delay is highly relevant when comparing DLD and bilingualism, as successive bilingual children experience a delay in exposure to the L, potentially leading to (temporary) similarities between the language outcomes of monolingual children with DLD and TD age peers who are learning the language as their L (e.g., Håkansson & Nettelbladt, ; Paradis & Crago, ; Crago & Paradis, ; Grüter, ; Paradis, ). Note that these similarities are expected to be temporary as the delay in L children will be transient, assuming that they are equipped with well-working language learning mechanisms, whereas the language problems of children with DLD are found to be persistent (Rice, ).2 It should be stressed that the additional ways in which monolingual children with and without DLD differ (profile difference, abnormal quantity of errors,

2 For this reason, Reilly et al. () explicitly recommend against the use of the notion ‘delay’ in relation to DLD.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

     



deviant structures) are not expected to characterize TD L learners, although in early L development child L learners are likely to go through a phase in which their L shows signs of L transfer leading to non-target like productions (Whong-Barr & Schwartz, ; Zdorenko & Paradis, ). While delayed L development does appear to have a parallel in monolingual DLD, L attrition by definition does not have such a parallel because it is specific to bilingual learning contexts (see definition below). However, incomplete acquisition, another term used in research on bilingualism, and related both to (L) delay and (L) attrition, may be applied to DLD. Anderson () defines the two notions as follows: L attrition occurs when L features are lost as a consequence of the child becoming dominant in the L, while incomplete acquisition is defined by a failure to acquire native-like structures as a result of insufficient exposure during crucial developmental stages. Following Montrul (, p. ), we assume that incomplete acquisition refers to a mature linguistic state and the outcome of language acquisition. In the first edition of his seminal book, Leonard () discussed the possibility that the language abilities of children with DLD ‘plateau’: ‘mastery levels are never reached [ . . . ] a plateau is reached before certain aspects of language are mastered. The basis for such arrested development is presumed to be biological age.’ (Leonard, , p. ). The latter remark echoes the notion of a critical age for language acquisition that is also often invoked to explain lack of success in L acquisition (Hyltenstam & Abrahamsson, ). In a recent article about the language development of children with DLD, Leonard (b) wrote: After a period of late emergence, the pace of subsequent lexical and grammatical development does not appear to be appreciably slower than that seen in TD. However, a plateau effect may start to occur because the late start makes it difficult to reach full mastery at the age when language learning is still efficient. (p. )

Rice (, p. ) adopts a metaphor to describe a process that closely resembles Leonard’s description. With respect to the language development of children with DLD, and the contrast with TD, she writes: ‘Their language train, however, seems to decelerate before reaching the final destination of the adult grammar, not quite reaching the end goal’ (italics added).

. C   D L D  

.................................................................................................................................. DLD is to a large degree heritable (Bishop, ). By implication, it affects not only monolingual but also bilingual children (Paradis et al., ). In bilingual children with DLD, the joint effects of impairment and bilingualism present researchers and practitioners with a challenge. After all, successive bilingual children with DLD are likely to demonstrate the phenomena that are typical of bilingual acquisition but also the symptoms of their language impairment. If the effects of impairment and bilingualism affect the same

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



, ,   

domains of language in a similar way (e.g., a negative effect), the expected net result of both effects would be the sum of both effects. One way to determine if the effects of DLD and bilingualism indeed accumulate is to use a multiple-group design in which bilingual children with DLD are compared with monolingual children with DLD on the one hand and bilingual TD children on the other hand. Adding a fourth group consisting of monolingual TD children enables comparisons of the effect size of the TD–DLD difference in a monolingual and bilingual context. Effect size is, as Paradis () observed, the index that should be used to determine if bilingualism aggravates the effect of DLD. Such an aggravating effect is present if in the bilingual context the difference between TD and DLD is larger (as evidenced by a larger effect size) than in a monolingual context. An alleviating effect is, in contrast, shown by a smaller effect size.3 It is not only relevant to ask if bilingualism aggravates the effects of DLD, but also if DLD aggravates the effects of bilingualism. This latter question is specifically relevant with respect to attrition and will be discussed in Section ... There, we will show that the notions of delay and attrition invoke specific considerations when applied in the context of bilingual DLD.

.. Delay in the L Two Dutch studies by Verhoeven and colleagues (Verhoeven, Steenge, & Balkom, ; Verhoeven, Steenge, Van Weerdenburg, & Van Balkom, ) arrived at the conclusion that bilingual children (L Dutch, various Ls) with DLD have an additional disadvantage in the L because their Dutch language skills showed effects of bilingualism as well as effects of impairment. In this context, these effects of bilingualism, which originate in the input to bilingual children that is distributed over two languages, should not be equated with incomplete acquisition, because it is unknown whether or not children’s language outcomes attested in these studies represent a mature linguistic state. Actually, because the children were rather young ( to  years old) and frequently exposed to the L, it is likely that the outcomes do not demonstrate incomplete acquisition but represent delayed acquisition (compared with monolinguals), caused by reduced input. It cannot be ruled out, however, that the delay will ultimately lead to incomplete acquisition. As Leonard (a) notes, the performance gap for bilingual children with DLD (as compared with bilingual TD children) in the studies by Verhoeven et al. was not greater than a similar gap for the monolingual DLD children as compared with their monolingual TD peers. In other words, there was no difference in effect size between monolinguals and bilinguals when comparing TD and DLD, showing that although the effects of bilingualism and DLD may add up, bilingualism does not aggravate the effects of DLD. In a recent study, we compared - to -year-old bilingual and monolingual children with DLD and TD, learning Dutch as their L (bilinguals) or L (monolinguals). The four groups were matched on age, non-verbal intelligence and also socio-economic status to the extent that this was possible. The bilingual TD and DLD groups were furthermore matched on

3 If the effects of impairment and bilingualism are in opposite directions, it is possible that bilingualism reduces the effect of DLD. In this case, the difference between TD and DLD is smaller in a bilingual than in a monolingual context. Such an effect might happen if transfer or cross-linguistic influence in bilinguals leads to accelerated language development.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

     



amount of exposure to Dutch. We found that the bilingual DLD group was outperformed by the bilingual TD group as well as by the monolingual DLD group, both on Dutch inflectional morphology and Dutch receptive vocabulary tasks, confirming that effects of bilingualism and impairment add up (Blom & Boerma, ; but see Paradis, ; Gutiérrez-Clellen et al., ; Rothweiler et al., ). For receptive vocabulary, but not for grammatical morphology, the effect of impairment was larger in the bilingual than in the monolingual sample, as shown by comparisons of the effect sizes of DLD in bilingual and monolingual contexts. This led us to the conclusion that bilingualism—or rather reduced input in one language—can aggravate the effects of language impairment in the domain of vocabulary. Because vocabulary is a moving target, as it continues to grow (Brysbaert et al., ) and contrasts in this respect with inflectional morphology, reduced input may have more of an impact on vocabulary development than on the development of grammatical morphology. In line with this idea, Chondrogianni & Marinis () observed that in a sample of Turkish children learning English as their L, vocabulary was affected by environmental factors such as mother’s proficiency and English use in the home, whereas tense morphology was not.

.. L attrition The conditions under which the L of immigrant children is learnt are often less favourable than those for the L, because the L is used by a limited number of people, in a limited number of domains, and proficiency in the L is often valued less by dominant societal norms than L proficiency (see Section ..). The result of such a bilingual learning situation is ‘subtractive bilingualism’ (Wong-Fillmore, ), which refers to the process in which the L is learnt at the expense of the L and gradually replaces the L, resulting in incomplete L acquisition and attrition. This will affect the L development of bilingual TD children and bilingual children with DLD alike, but it may not affect them in equal measure. As discussed earlier, incomplete L acquisition as observed in bilinguals is at first glance not unlike what we see in DLD (‘not quite reaching the end goal’), but the cause of incomplete acquisition in DLD is not primarily a lack of exposure in a crucial developmental phase. Rather, it is a diminished ability to learn language that prevents children from reaching a sufficient language level in a crucial developmental phase. In bilingual DLD, this diminished ability may be coupled with a diminished window of opportunity if input in a particular language is reduced and children stop being abundantly exposed to one of their languages. It is conceivable that these two factors, which are internal and external to the child respectively, add up and lead to a more incomplete L acquisition in bilingual DLD than in bilingual TD (or monolingual DLD). Also, internal factors could enlarge the effects of external factors, leading to aggravation of L attrition in bilingual DLD compared with bilingual TD. To our knowledge, there is a hardly any research on attrition in children with DLD, but, with respect to the influence of DLD on attrition, a comparison of two cases by Restrepo & Kruth () may be insightful, keeping in mind that observations based on cases cannot be generalized to the population. Restrepo & Kruth () followed two -year-old Spanish-English bilingual children in the United States, one with and one without DLD, over a period of one year. Despite similar patterns of exposure, the child with DLD appeared to lose Spanish faster in comparison with the TD child, as indicated by a decrease in sentence length and a faster increase in errors with grammatical gender and verb

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



, ,   

inflection. The same errors were also attested in other TD children, but these children seemed to show a more protracted pattern of attrition (cf. Anderson, ), suggesting that DLD affects quantitative aspects of attrition and speed more than qualitative aspects of attrition (such as the types of errors).

. F  L    

.................................................................................................................................. Attrition does not necessarily affect all bilingual migrant children. One factor that influences attrition, which is part of the definition by Anderson (), is dominance. A dominant bilingual is ‘someone with greater proficiency in one of his or her languages’; who ‘uses it significantly more than the other language(s)’ (Li, , p. ). L attrition happens as the result of becoming dominant in the L (Anderson, ), a dominance shift which, in turn, is caused by more exposure to and use of the L. In addition, there are other factors that might contribute to attrition in the L. If the effects of attrition are exacerbated by a language impairment (Section ..), these factors likely have an even larger effect on the L abilities of bilingual children with DLD. Moreover, it is important to be aware of these factors when deciding on (the level of) attrition and comparison with appropriate norms, which is an issue we turn to in Section .. Below, the role of age of exposure to the L, a child-internal factor, and socio-economic status and L status, two child-external factors, will be discussed. This is by no means an exhaustive list of factors, but they are chosen as an illustration and because these factors have been investigated in recent research on children becoming bilingual in immigration settings.

.. Age of exposure to L and L development Recently, Bedore et al. () explored the extent to which age of exposure to L English impacts the development of L Spanish in Spanish-English bilingual children in Grade , with an average age of  months, and Grade , with an average age of  months. Their findings suggest that children who start acquiring English the earliest scored the lowest on Spanish, indicating that early L exposure may come at a cost to the acquisition of the L. From the perspective of distributed input this pattern is not surprising. When L exposure starts, input becomes divided across languages and this might prevent children from acquiring new grammatical constructions in their L (Montrul, ). An early age of L exposure may particularly diminish a child’s chances of acquiring those L constructions that tend to be acquired late. The consequence could be that the earlier a child’s attention shifts to the L, the earlier L development may come to a halt (stagnation) and may, subsequently, start to show attrition. At this point we want to emphasize that this is one developmental scenario out of many, as L development in successive bilingual children does not necessarily stagnate, but could also continue to grow at a slower pace due to an increase of L exposure. Stagnation is, moreover, not necessarily followed by attrition, but could also lead to incomplete acquisition, which is a relatively stable state. In addition, growth, incomplete acquisition, and attrition could co-occur in an individual

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

     



child. A child may show growth on one aspect of language, incomplete acquisition on another and at the same time lose other aspects of the L (Montrul, ).

.. Socio-economic status: parent–child interaction and L development Children from low socio-economic status (SES) homes, whose parents have a background of low education, income and/or occupational prestige, have been found to lag behind in their language outcomes compared with children from higher SES homes (see, e.g., Hoff, , for an overview). Maternal education, specifically, may influence vocabulary development, regardless of culture (Hoff & Tian, ). The seminal study by Hart & Risley () suggested that the smaller vocabularies of children from lower SES families are related to parent–child interaction patterns. The importance of parental input in explaining the effect of SES was further confirmed by Huttenlocher et al. (), who observed that SES effects on lexical, constituent, and clausal diversity in children’s language are, in part, mediated by caregiver speech. Independent of impairment or reduced input in bilingual contexts, SES is a factor that may prevent children from reaching language milestones, even though the effect of SES may not be as persistent as suggested by some studies and could influence certain aspects of language more than others (Letts et al., ). Accumulating effects of low SES and DLD were not found in the study by Roy & Chiat (), who observed similar language profiles across children with DLD from high and low SES families. However, effects of bilingual exposure and SES may add up: in a study by Calvo & Bialystok (), bilingual children growing up in working class families scored lower on receptive vocabulary tests than monolingual children from working class families, but also lower than bilingual children from middle class families and monolingual children from middle class families. If the effects of reduced exposure in bilingual contexts and low SES add up, the prediction could be that incomplete acquisition and attrition are more prominent in children from lower SES backgrounds than in children from higher SES backgrounds. Low SES can be more prevalent among certain groups of immigrant children. For instance, in the Netherlands, many unskilled Turkish and Moroccan migrant workers came to the country because of labour migration and family reunification (Swanenberg, ). In these families, low SES may go hand in hand with limited stimulation in the L, in particular when the L is further suppressed due to an increase of L exposure when the children growing up in these families start to attend school. However, low SES has also been found to be associated more with L than L use at home, at least by Turkish mothers in the Netherlands (Prevoo et al., ; but see Goldberg et al., , for a somewhat different pattern in migrant mothers in Canada), suggesting that low SES can also create a relatively favourable context for L development and may reduce L attrition.

.. Language status and L development In many cases, the L of migrant children, also labelled minority language, lacks a high social status—and is thus low in prestige in the host country, in sharp contrast to the

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



, ,   

majority language, which is the language used in educational settings and official media. This applies to, for instance, Spanish or Somali children in the United States or Turkish children in the Netherlands and Germany. As a result, there are fewer opportunities for those children to develop and use the L, and fewer resources are available to support their L development compared with the L. For instance, literacy may be exclusively associated with the language of school, which is the L. Reading and literacy-related activities will therefore often be done in the L, supporting L rather than L development (Prevoo et al., ). Moreover, high proficiency in the L is generally not highly valued, unlike reaching a high language level in the L. This situation is the typical subtractive context in which the L is likely to replace the L, as described earlier (Section ..). This contrasts with additive contexts in which the acquisition of L is supported and can facilitate L learning, as in, for instance, Canada where both French and English are majority languages (Baker, ). The effect of L status varies across minority languages, as illustrated in a study by Scheele et al. (), who investigated vocabulary development in the L and the L of bilingual Turkish-Dutch and Moroccan-Dutch (L: Tarifit-Berber) preschoolers. The Dutch vocabulary scores of the Turkish children were lower than those of the Moroccan children, which was related to the more frequent use of Dutch in the Moroccan homes. This difference between the Moroccan and Turkish migrant families can be explained by a myriad factors (Laghzaoui, ). Tarifit-Berber is a predominantly oral rather than written language in which fewer resources (books, television) are available than in Turkish. Its prestige is moreover low and, in the Moroccan community in the Netherlands, Dutch is often used to communicate among each other because the different Berber varieties are not mutually intelligible. As a consequence, attrition may happen more often and to a larger extent in Moroccan individuals who learn Tarifit-Berber as their L than in Turkish individuals. To our knowledge, there is, however, no research that has systematically compared the levels of attrition across speakers of Tarifit-Berber and Turkish in the Netherlands. Language shift, the loss of the L that concerns an entire speech community (Fishman, ; Lambert & Freed, ), may also be more prominent in the Moroccan than in the Turkish community in the Netherlands. If language shift occurs in an immigrant community, it creates different characteristics in the child’s language input, which is relevant for every child learning a language, but could create a specific risk for children with DLD. Restrepo et al. () approach bilingual DLD from the perspective of Dynamic Systems Theory. Crucially, in that framework, language abilities in multilingual children change depending on the quantity and quality of input needed to induce a change. When input is strong, language changes can be positive on all the languages that the child speaks, but they can also be negative for one language and positive for the other. (p. )

When we consider language shift, the quantity and quality of the input may be at stake. Since children with DLD already have more difficulty learning language, this can be a serious threat to the acquisition of the L and, as a consequence, it might affect children with DLD even more strongly than their TD peers. However, given the sparseness of relevant research specifically addressing the impact of these variables in bilingual DLD, predictions can only be tentative.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

     



. I    D L D

.................................................................................................................................. Successive bilingual children experience a delay in L exposure. In such cases, evaluating the child’s abilities in the L (in which they have had input from birth) could be more insightful than assessing the L. However, like L assessment, L assessment is not always straightforward. In this section, we discuss the ways in which attrition may complicate L assessment. In addition, possible alternatives are described when reliable assessment cannot be ensured, either in the L or in the L.

.. Assessing the L: comparing an individual child using appropriate norms In order to isolate effects of DLD from effects of attrition or bilingual development in a more general sense, information could be gathered on estimated language dominance, as attrition typically happens in a situation when the L becomes dominant (Anderson, ). Many children who become bilingual in immigration contexts are more dominant in their L at younger ages. They gain experience with the L when they spend more time outside their homes in educational settings and have increased interactions with the broader community, most typically leading to a dominance shift (see Kohnert, , and references cited therein). The timing of this dominance shift, and the associated likelihood that a child may show attrition, will thus depend on the child’s age of L exposure as well as the amount of exposure to and use of both languages. Consequently, it is of paramount importance to collect relevant background information on a child’s linguistic environment as part of a multilingual anamnesis. On top of this, isolating effects of DLD requires a comparison with normative data collected from TD bilingual children with the same language background and level of attrition. However, in particular, degree of attrition is hard to assess and measuring dominance might, in this case, be an alternative.

.. Assessing the L: linguistic markers and valid tools When assessing the L, it is important to be aware of the fact that the effects of DLD and attrition might be easily confounded because they can affect the same areas. For instance, errors with verb inflection are not only linguistic markers of DLD in Spanish, but they are also a characteristic of children who undergo loss of Spanish (Anderson, ). Not all linguistic markers that are indicative of DLD in monolingual contexts can thus be generalized to L assessment in bilingual contexts, unless appropriate normative data are available. Note that the same holds for the assessment of the L of bilingual children, as a delayed L development can also show similarities with an impaired language development (as described in Section .).

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



, ,   

Moreover, the L in the host country may be subject to language change as a result of language contact. The implication is that the validity of assessment tools for the L created in the country of origin is restricted in the host country. For example, in Turkish, transitive verbs mark the direct object with accusative case, depending on the position of the object and definiteness. Case marking is vulnerable in DLD across many languages, including morphologically rich languages (Leonard, ), and might be seen as a clinical marker. However, in Turkish multiword expressions as spoken in the Netherlands, accusative case can also be omitted or substituted, presumably under the influence of Dutch, which has no case marking (Doğruöz & Backus, ). Because the input (and thus the target language) properties may thus differ between Turkish speakers in Turkey and the Netherlands, assessment tools developed in Turkey may lack validity in the Dutch context. Note that this process of language change is not necessarily equated with language shift, as the Turkish community has a strong tendency to maintain their L (Backus, ).

.. Possible alternatives: measures that are insensitive to L attrition (or L delay) The ideal situation, in which a child’s level of attrition is estimated correctly and relevant normative data and assessment tools are available, is often out of reach. The question arises as to what professionals could do if reliable assessment in the L or in the L cannot be ensured. In this section, we review two possible alternatives. One option is the use of language processing measures instead of standardized tests of vocabulary and grammar, which are based on previous knowledge and thus are crucially susceptible to effects of input (i.e., experience with the language) (Campbell et al., ; Kohnert et al., ). Knowledge-based measures may underestimate the language learning abilities of children who have less experience with the target language, in particular bilinguals, because knowledge of one specific language is tested (Restrepo & Silverman, ). Processing-based measures, in contrast, tap into the cognitive underpinnings of language and are less biased towards language-specific knowledge (Chiat, ). Factors like attrition will therefore affect performance on processing-based measures to a lesser extent than performance on knowledge-based measures. Taking this idea one step further, Chiat () developed a non-word repetition task—a processing-based measure that taps into verbal short-term memory—that includes features that are common across languages, enabling children to use knowledge that is not specific to the L or the L. In a study by Boerma et al. (), bilingual children did not differ from monolingual peers on this ‘quasi-universal’ task, but performed more poorly on a non-word task that was based on Dutch phonotactics. The quasi-universal non-word repetition task was insensitive to the effects of bilingualism but sensitive to DLD, and its diagnostic validity was excellent in both a monolingual and bilingual context. Another type of task that may support the identification of DLD in bilingual child populations is a narrative task. Narrative tasks can target a story’s microstructure, focusing on a child’s linguistic expression of story events, or macrostructure which refers to the structure of the story, including plot elements such as the goals of the protagonists (Stein & Glenn, ). In contrast to microstrucure, measures of narrative macrostructure are relatively independent of experience in a specific language and do not disadvantage

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

     



bilinguals (Pearson, ; Hipfner-Boucher et al., ), which may be explained by the finding that, unlike in microstructure, linguistic factors do not predict children’s expression of the macrostructure of a story (Blom & Boerma, ). Instead, cognitive limitations, and specifically sustained attention, have been found to influence performance on narrative macrostructure (Blom & Boerma, ). While sustained attention is compromised in children with DLD (Ebert & Kohnert, ), it is not affected by bilingualism (see Bialystok et al., ). Hence, even when there is attrition, bilingual TD children may still perform well on a narrative task when analysed on the macrolevel. Bilingual peers with DLD, however, might perform more poorly due to limitations in sustained attention. A study by Boerma et al. () confirms these predictions. Large negative effects of DLD were found on the measures of narrative macrostructure, whereas no effects of bilingualism emerged. As a result, the diagnostic validity of the narrative task was adequate in both a monolingual and bilingual context. In general, in this section we showed that tasks which require less language-specific knowledge (or less experience with a specific language) are more valid indices of impairment in a bilingual child than tasks that do require such knowledge in those cases where reliable assessment in either the L, due to attrition, or the L, due to delay, is not reliable. These tasks should, however, not be used as stand-alone tools, and, instead, the use of multiple instruments is recommended.

. C 

.................................................................................................................................. In this chapter, we discussed attrition in relation to DLD, asking the question whether the effects of attrition and DLD would show overlap, whether they would accumulate and whether DLD could influence attrition. What emerged from this discussion is a potentially very interesting and urgent field of study, but one in which hypotheses have to be pieced together on the basis of evidence from various subfields that have until now not been directly associated: L development, L attrition, and language impairments such as DLD. Another issue that we discussed and that directly relates to our topic is the possibility that L attrition leads to a diagnostic confound due to the fact that both L attrition and DLD affect children’s language level, albeit for different reasons. Because of a dearth of research on L attrition in the context of DLD, our point of departure was often research on the L of bilingual immigrant children. This research could be informative because the same mechanism—limited exposure and usage—underlies both L attrition and L delay. Moreover, linguistic phenomena that are susceptible to attrition effects might also be prone to delay, following the ‘last in, first out’ principle that follows from the regression hypothesis (Jakobson, ). This hypothesis states that those linguistic features which are learned last are the first to show attrition, and vice versa (Keijzer, ). However, an important difference between L and L development concerns the conditions under which the two languages are learnt, which are often more favourable for the L and more adverse for the L in migrant families. Because of their language impairment, children with DLD may need more input to reach, or maintain, the same language level as their TD peers (e.g., Gray, ; Rice et al., ; Riches et al., ; for a meta-analysis, see Kan & Windsor, ). We hypothesized

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



, ,   

that, consequently, the L may erode relatively fast in children with DLD. This effect may be modulated by language-level and child-level factors, both internal and external, such as input-dependency of linguistic phenomena, age of L exposure, socio-economic status, and status of the L. Understanding more about L attrition is furthermore relevant for diagnostic purposes, as overlap between the effects of L attrition and DLD could create a diagnostic confound. Regarding the diagnosis of DLD in bilingual contexts, it is generally recommended that both the L and L are assessed. However, like delays in the L, attrition in the L presents a complicating factor that clinicians need to be aware of, because of the possibility that it renders L normative data unreliable and assessment tools in the L invalid. For this reason, new instruments have been developed that are sensitive to the effects of DLD, but relatively insensitive to the child’s knowledge of a specific language and, therefore, to effects of bilingualism. That these instruments are promising has been shown by the equal performance of monolinguals and successive bilinguals in their L. Future research has to show if the outcomes are indeed replicated for their L, regardless of L attrition. Moreover, to fully understand the interaction of L attrition and DLD, future studies need to longitudinally follow the L and L development of bilingual children with and without DLD, next to systematically keeping track of patterns of L and L exposure and use. Although L attrition may even then be difficult to disentangle from incomplete acquisition, language shift, and language change, longitudinal work would be a good starting point to investigate this understudied reality. It may allow us to identify a plateau in the L, determine if children’s development stabilizes after reaching the plateau or if linguistic features start to show attrition due to L dominance, and keep track of the speed with which attrition takes place. Such longitudinal research might inform us about the nature of attrition in children with DLD and allow us to address attrition in this group in a more direct manner than we were able to do in this chapter, due to the present lack of resources.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

  ......................................................................................................................

        Lexical retrieval, language use, and cognitive and neural changes ......................................................................................................................

 ,  ,  ,  ,   . 

. I

.................................................................................................................................. A- native-language attrition most often occurs following a move to an environment where a non-native language is spoken. To date, the focus of attrition research has been on identifying what aspects of language are lost and which factors are crucial for the retention or attrition of the native language. Attrition is a gradual and fairly subtle process with no clear beginning or end. To best assess the effects of attrition, researchers tend to choose study participants who have lived in a non-native environment for a decade or longer (e.g., de Bot & Clyne, ; Schmid, ). The assumption is that the longer they have been removed from the native-language environment, the greater the degree of language attrition that should be observable. However, this principle regarding length of time and its relationship to language use overlaps with another, largely ignored, phenomenon: language changes associated with ageing. Are language changes due to long-term disuse conflated with age-related language changes in older adults who experience language attrition? This chapter explores changes to the adult lexicon as a result of attrition and ageing since the lexicon is considered a vulnerable part of the language system in both attrition and ageing. We consider neurophysiological changes that may play a role in language attrition and in non-pathological ageing to speculate whether the neurobiological sources of these two processes are similar or different. If attrition and ageing exert independent effects on

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



, , , ,  

lexical retrieval decline, we must consider the effects of each of these factors on word retrieval for older adult bilinguals immersed in a non-native-language environment.

. L    

.................................................................................................................................. Language attrition frequently occurs when a person’s use of the first language decreases or ceases completely, often as a result of moving to an environment where another language is primarily used. One of the earliest noticeable features of attrition is word retrieval difficulty (Opitz, ; Kasparian & Steinhauer, b). This difficulty may be manifested in slower lexical retrieval, retrieval failures, increased tip-of-the-tongue (TOT) states (when the speaker knows a word but cannot access its full phonological representation), as well as an increase in disfluency markers (e.g., hesitations, filled pauses, self-corrections, and repetitions) (Schmid & Köpke, ; Schmid & Beers Fägersten, ). Word-retrieval problems in attrition are reported to be greater for low-frequency words (such as accordion) than for high-frequency words (such as guitar) (e.g., Hulsen, ). There are several issues to keep in mind when considering word frequency effects, however. First, word frequency counts are likely to differ for bilinguals and monolinguals, whether or not bilinguals are in an attrition scenario (i.e., living in an L-environment). Since bilinguals divide their time between their two languages, they have fewer overall encounters with words in each of their languages than monolinguals. Given that the highest frequency words will be encountered frequently by both bilinguals and monolinguals, differences between the groups are smaller for high-frequency words and become larger for lower frequency words (Gollan et al., ). Gollan et al. () explain this phenomenon in terms of vocabulary size and a lexical search process that uses frequency-based rankings. If lexical access involves a search through a frequency-ordered list, access time would be a function of lexicon size. Bilinguals have on average double the lexicon size as monolinguals, which may account for the slower overall lexical retrieval speed of bilinguals compared to monolinguals and the greater frequency effect observed for bilinguals. Second, cognates may have a special status in the lexicon that can boost word frequency counts. These words have overlapping semantics and phonological forms in two languages (such as the English and Spanish words flower and flor). Production of cognate words seems to benefit from encounters in either language (Costa, Caramazza, & SebastiánGallés, ), and attriters are better able to retrieve cognate words than non-cognate words (Ammerlaan, ), suggesting that the similarity with words in the second language preserves their accessibility in the first language. Thus, patterns of performance seen in attrition may be more extreme versions of patterns that are typical for bilinguals in general. Reduced accessibility of low-frequency, non-cognate words has particular implications for lexical specificity because many highly specific words (e.g., finch) are lower in frequency than their superordinate counterparts (e.g., bird). The result is that attriters’ accessible vocabulary tends to consist of more general words. For example, Olshtain & Barzilay () investigated lexical retrieval accuracy using a structured narrative task in which Americans living in Israel for eight years or more described the story in the picture

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

      



book Frog, Where Are You? They found that the participants experienced more difficulty when trying to retrieve low-frequency words in the pictures (such as jar and gopher), and they often substituted the specific low-frequency term with a more general one sharing the same semantic features (e.g., producing bottle or container instead of jar). Another strategy used by the participants was paraphrasing and circumlocuting, such as describing the floor as the ground of the living room. Although such studies show that attriters fail to access certain lexical items that they presumably knew at an earlier time, they do not necessarily demonstrate that this group of bilinguals has ‘lost’ these words completely. In fact, when participants are presented with a picture-name recognition task (which does not require production, as opposed to a recall task), they are often able to correctly identify the names for pictures that they previously could not retrieve and express (Ammerlaan, ). Consequently, it appears that it is the accessibility to those items required for production that suffers as a result of the forces of attrition rather than complete loss of the lexical information. Reductions in expressive vocabulary can be investigated by analysing the lexical diversity of attriters’ and non-attriters’ speech or writing. Lexical diversity is often calculated using the Type–Token Ratio, the ratio of the number of unique words (types) to the total number of words produced (tokens). The higher the Type–Token Ratio, the more variety there is in the lexical choice of the speaker. Comparing Type–Token Ratios, Schmid () found that lexical diversity was lower for attriters than non-attriters, even for attriters who had maintained high proficiency in the first language. Laufer () found an inverse relationship between length of residence in the second-language environment and lexical diversity of the first language, as measured by the Type–Token Ratio. An even stronger effect of length of residence was found for lexical richness, which was measured as the proportion of infrequent words in the text (i.e., words not occurring in the , most frequent words). Using a more complex measure of lexical diversity that incorporated not only types and tokens but also distribution of tokens, distance between repetitions of the same type, and rarity of words used, Schmid and Jarvis () were able to discriminate between attriters and non-attriters while simple Type–Token Ratio measures were not able to discriminate between these groups. Thus, discourse analysis measures indicate that attriters’ lexical production is less diverse and that the longer the time spent in the second-language environment, the less diverse the lexical choices become in the first language, with fewer low-frequency words chosen.

.. Causes of lexical attrition Attrition studies have often focused on identifying whether the source of the deviant patterns seen in attriters is due to changes in the memory system or interference from another language. However, this is a bit of a false dichotomy. For example, slower response times on a lexical retrieval task may be due to reduced accessibility to the item in the memory system as well as the need to overcome interference from a strong second-language word. Indeed, as Schmid & Köpke () have pointed out, the source of word-retrieval difficulty is likely to be a combination of both. Even in cases of obvious cross-linguistic transfer, such as the use of the English phrase walk the dog in Spanish (caminar al perro) instead of the standard Spanish phrase sacar al perro

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



, , , ,  

(take the dog out), the phrasal borrowing may be accompanied by or facilitated by sources of memory decay. A long tradition of research on memory and ‘forgetting’ provides insights into possible sources of language attrition that might stem from disuse. Ecke () presents seven sources of ‘forgetting’ that, he argues, are common for both language and non-language information. Two of these include the decay of memory representations and a loss of access to memory representations. Memory for a variety of different skills or types of knowledge weakens with periods of disuse or exposure, and language is no exception, though rates of memory decay may vary with different knowledge systems and even different aspects of language (Arthur et al., ; Köpke, ). In connectionist models, the strength of connections is increased with each new occurrence (Ellis, ; McMurray et al., ). So, for example, the connection between the concept of a cat and the phonological form in English (kæt) is strengthened with each use. Lindsay & Gaskell () suggest that in the case of language loss, the connection weights in the network may still contain some vestige of the original knowledge, which may be accessed implicitly but not through explicit tests like assessments of vocabulary or grammatical knowledge. If lexical access challenges in language attrition are due to weakening of the formmeaning relations as a consequence of disuse, then we would expect to find that lower amounts of continued L contact and longer residence in the host country would lead to greater signs of attrition. Typically, the more contact one maintains with the L, the less attrition will occur (e.g., Soesman, ; Cherciov, ; Kargar & Rezai, ), but some studies have failed to find an effect of L contact (e.g., Schmid, b). This may be because the type of contact might matter more than the amount. Schmid & Dusseldorp () found that using the L for professional purposes (i.e., at work) was a significant predictor of L maintenance, while Hulsen () found that the number of different L contacts a person had positively influenced L maintenance. Surprisingly, length of time in the L environment seems to have little predictive power for degree of L attrition. This may be because the relationship is non-linear, with more accelerated rates of attrition in the first decade after immigration compared to later years (Schmid, b) or because length of residence is only a strong factor when L contact is very low (de Bot et al., ; Soesman, ; Schmid, b). The lack of clear correlations between either amount of L use or length of residence in the L context and the degree of L attrition make an explanation of attrition based primarily on weakening of connections more difficult. Another possible source of language attrition is competition from the other language(s). Ecke () describes the loss of access to information due to cross-linguistic processes as stemming from interference from the newly learned language and interactivity between the languages. For example, attriters may learn new collocations in their second language (e.g., Spanish: tomar una decisión, literally: take a decision), and when asked to judge whether or not a collocation exists in their first language, they find the translations of a large number of foreign collocations acceptable (% in Laufer, ). Words in the first language may also undergo semantic extension or reduction due to differences in the semantic scope or features of translation equivalents (Pavlenko & Malt, ). Pavlenko () describes four types of cross-linguistic influence: borrowing words from the second-language when no first-language equivalent exists, restructuring the semantics of L items to accommodate the semantic scope of the translation equivalent, convergence of meanings from both

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

      



languages, resulting in a unique form, and shifting of the L items toward the L norm. While the first type is an example of lexical enrichment (i.e., additions made to the mental lexicon), the other three types represent deviation from an L ‘norm’, which may be taken as evidence of language attrition.

. N    

.................................................................................................................................. Very little research has been done to investigate the neurophysiological changes that underlie the attrition of acquired skills such as language use. Two studies have used electrophysiological measures (Event-Related Potentials (ERPs)) to examine the brain processes of attriters during lexical processing. Datta () investigated how lexical characteristics of the two languages affect processing of words in each language. She contrasted ERP responses to English and Bengali words with high and low familiarity. Low-familiarity words typically elicit a larger N component than high-familiarity words. Using a picture-word matching paradigm, she found that native Bengali speakers who had been residing in the United States and had become English-dominant showed an effect of English familiarity on the amplitude of the N but no effect of Bengali familiarity, demonstrating that these attriters were more sensitive to properties of the English names than they were to properties of the Bengali names. Greater N amplitudes have been proposed to reflect more difficult lexical access, driven by both recent and longterm experience (Kutas & Federmeier, ). Datta’s () study thus revealed that highly familiar English words have a processing advantage for the attriters while highly familiar Bengali words do not have such an advantage as both high- and low-familiarity Bengali words showed similar amplitudes for the N. These findings suggest that attriters’ sensitivity to lexical properties such as frequency/familiarity can decrease with reduced exposure to the language. Another study that used ERPs to investigate lexical processing in L attriters examined whether attriters recognize a violation when ‘confusable’ word pairs in Italian (words that differ only in their final vowel, such as cappello (hat) and cappella (chapel)) are swapped in a sentence context (Kasparian & Steinhauer, b). They found that Italian speakers who had immigrated to Canada in adulthood and had lived there for on average eleven years showed the native-like pattern of an N effect when a word in a sentence was substituted by its highly confusable pair-member but only when their proficiency in Italian remained very high. The researchers also included sentences with more blatant word-swap violations such as The fisherman wears the briefcase of wool. Longer length of residence in Canada was associated with a weaker N effect for these types of sentences, though not for the confusable word-swap types. The authors suggest that language proficiency is the principal driving force of the N effect for word-swap violations, which is bolstered by the fact that Italian second-language speakers with high proficiency in Italian also showed an N effect in their study. This adds to what Datta () found and supports the idea that sensitivity to both lexical characteristics like familiarity and word form-meaning relations is weakened when language use and proficiency decline.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



, , , ,  

The underlying neurophysiological changes associated with language attrition are still unknown. The ERP evidence described above reveals that neural processes underlying lexical comprehension (either in isolation or within sentence context) change in the context of attrition in a way that suggests a loss of sensitivity or differentiation to factors that native-speaking non-attriters are sensitive to. Attriters also appear to exhibit greater effort in attempting to reconstruct a sentence meaning when a semantic violation is detected, as evidenced by a greater P for attriters compared to controls in Kasparian & Steinhauer (b). What these patterns represent at a cellular level, though, is still unknown. A good deal of attention has been applied in recent years to questions of experience-dependent neuroplasticity and the effect of various types of training or expertise on brain morphology (for a review, see Lövdén et al., ). Research using animal models has demonstrated that skill acquisition results in synaptogenesis, alterations in synaptic connectivity, dendritic lengthening and branching, and functional reorganization of neural networks (e.g., Kleim & Jones, ; Lövdén et al., ). Surprisingly little research has examined the neural changes involved in skill loss, likely due in part to the extended time scale of skill loss as compared to skill acquisition. It appears that the ‘use it or lose it’ principle would apply here, although behavioural data has shown us that in many cases a skill that has not been used for a long time may be dormant rather than lost, as evidenced by the speed of re-acquisition, as if the skills simply needed to be ‘awakened’. This rapid re-learning has been demonstrated also in cases of first language attrition (Higby & Obler, ). The strongest evidence for the ‘use it or lose it’ phenomenon comes from studies of sensory deprivation in animals and humans (e.g., Hubel & Wiesel, ; Bedny et al., ). It is not entirely clear, however, whether the same types of neural reorganization apply to disuse as they do to deprivation. In the case of language, there is rarely a complete cessation of native-language use, but rather a reduction of use, which may be abrupt or gradual, and which may or may not fall below the level of use of the second language. Thus, the neural changes involved in language attrition may be both quantitatively and qualitatively different from the types of effects seen in deprivation cases. Competition from the second language may explain the possible neurophysiological mechanisms of attrition effects better than weakening due to disuse. Functional neuroimaging studies show that second-language processing uses many of the same neural circuits as native-language processing (Higby et al., ). At a more fine-grained level, the activation patterns for the first and second languages are likely partially overlapping and partially distinct. Bilinguals have to map two different word-forms to the same concept. Commonly, attriters use the second language more often than they use the first. Thus, the neural populations involved in going from meaning to the second-language word-form during lexical retrieval become more practised and automatic. This creates a situation in which the second-language route is primed by recent and frequent use. When attriters attempt to access the now less-used route from meaning to the native-language word-form, there is competition from the primed route. Current neuroimaging techniques have not been able to demonstrate this type of competition at a fine-grained level, but as technology improves, we should be able to test whether this type of competition is indeed affecting access to the first-language lexicon. In sum, the findings with regard to lexical attrition of the first language demonstrate that attrition occurs even under conditions where proficiency has been maintained. Attrition

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

      



manifests as retrieval slowing and retrieval failures (particularly for low-frequency, non-cognate words), reduced lexical diversity and richness, as well as semantic restructuring and other modifications to lexical knowledge. The amount of attrition correlates with certain language-use factors, but length of residence in the L environment has little effect on the degree of attrition. The precise neural processes underlying these behavioural changes are still unknown, but recent ERP evidence suggests that neural sensitivity to certain aspects of the attriting language is reduced, particularly as proficiency declines.

. L     

.................................................................................................................................. As people age, they experience declines in memory and some aspects of language. Word retrieval problems are one of the most pervasive complaints among older adults (e.g., Ossher et al., ). In laboratory studies, which often use confrontation naming tasks to probe word retrieval abilities, older adults show lower accuracy and longer response times for lexical retrieval tasks such as picture naming compared to younger adults (Au et al., ; MacKay et al., ; Goral et al., ; Verhaegen & Poncelet, ). These age-related declines are gradual, making it difficult to determine when they begin, but researchers have suggested that declines in accuracy become apparent after about  to  years of age (Nicholas et al., ; Albert et al., ; Kavé et al., ), though retrieval speed may begin declining after age  (Verhaegen & Poncelet, ). Word-finding difficulties have also been found in the narrative discourse of older adults. In a picture description task by Schmitter-Edgecombe et al. (), while % of younger adults had behaviours indicative of word-finding difficulty (e.g., substitutions, repetitions, fillers, etc.) in more than % of utterances, the rate of older adults who showed the same level of word-finding difficulty increased to % for older adults in lower age ranges and % for older adults in upper age ranges. Like the attriters described in the previous section, substitutions made by older adults generally consist of semantically related words, vague terms, and circumlocutions (Nicholas et al., ; Obler & Albert, ; Albert et al., ). Schmitter-Edgecombe et al. () reported that the number of substitutions and word reformulations increased with age, indicating a continued increase in word finding difficulties with advancing age. However, unlike the attriters, older adults show greater lexical diversity (as measured by the Type– Token Ratio) than younger adults (Kemper & Sumner, ). This may be due to a critical difference between attriters and older adults: vocabulary size. Older adults continue to add new words to their mental lexicon throughout adulthood (Verhaeghen, ; Goral et al., ). This levels off and/or declines only at much older ages, such as in the s or s (Alwin & McCammon, ; Singer et al., ). Furthermore, vocabulary size predicts word retrieval accuracy, even after one accounts for education (Albert et al., ), suggesting that continued increase in vocabulary in adulthood provides protective effects against age-related word retrieval decline. Unlike decline seen in attrition, the lexical retrieval decline seen in ageing is not modulated by word frequency; older adults show similar declines for both high- and low-frequency words (Newman & German, ). Some research shows that older adults

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



, , , ,  

can access more low-frequency words than younger adults (Gollan & Brown, ) and choose lower frequency words in production (Kavé et al., ). However, word familiarity, a related measure to lexical frequency, does affect naming accuracy in older adults, with better accuracy observed for more familiar words (Newman & German, ). While word frequency is calculated as the relative frequency of appearance across a variety of (usually printed) sources across a corpus, word familiarity is a subjective measure based on scalar ratings. Word frequency and familiarity typically correlate, but they do not measure the same thing, as there are many examples of words that are rated as highly familiar (e.g., puddle) but which are relatively uncommon in corpora, and thus are considered lowfrequency items. Newman & German () suggest that highly familiar words may be in some sense protected from age-related declines while the same does not appear to be true for high-frequency words. The source of word retrieval problems in ageing appears to be related to access to the word-form rather than loss of word knowledge, at least until a much older age or the onset of pathology (e.g., Barresi et al., ). This is similar to the proposals put forth for lexical decline in attriters. The majority of the evidence for this in older adults comes from TOT states and cuing. In a TOT state, the speaker recognizes that the word has been partially accessed (that is, semantic and sometimes grammatical and sublexical information like number of syllables has been recovered), but the speaker cannot access the phonological form of the word. Older adults experience more TOTs than younger adults (Burke et al., ; Gollan & Brown, ). Furthermore, providing participants with a phonemic cue (the word’s initial sound) when they experience a word retrieval failure leads older adults to the correct answer in most cases (MacKay et al., ). Neuroimaging studies provide further support that word retrieval impairment in older age is primarily due to poorer access to phonological information rather than to semantics (Diaz et al., ; Rizio et al., ; Shafto et al., ). Together these findings strongly suggest that the source of word retrieval failures among older adults is primarily a failure to retrieve the phonological form of the word. A well-supported proposal to explain the characteristics of word retrieval difficulty in ageing is the Transmission Deficit Hypothesis (Burke et al., ). This theory is based on a connectionist framework in which various levels of representation (e.g., semantic, phonological) are weakened in ageing. When a word’s semantic nodes are activated, the activation primes lexical nodes and consequently phonological nodes. Weakened connections hinder the transmission of priming, particularly from lexical to phonological nodes, for which a single connection exists (by contrast, between the lexical and semantic nodes, multiple connections are available). A similar proposal has been put forward to explain the poorer performance on word retrieval tasks by bilinguals compared to monolinguals. The Frequency-Lag (a.k.a. Weaker Links) hypothesis of Gollan and her colleagues (e.g., ) proposes that bilinguals have weaker connections between nodes than monolinguals due to their less frequent use of each language, which creates particular problems for phonological access. If, as suggested in section II, word retrieval difficulties in attrition may be an extreme version of a typical bilingual characteristic, then the notion of weakened connections that has been used to explain ageing effects and bilingual effects may also extend to first language attrition. Given that age-related word retrieval difficulty appears to be, in essence, an access problem, changes in other cognitive faculties, in particular executive function, may also contribute to word retrieval failures. For example, impaired inhibition affects older adults’ word recognition abilities (Sommers & Danielson, ; Wingfield et al., ; Peelle et al.,

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

      



), but it does not appear to impact word retrieval in the same way (Higby et al., in press). Inhibition is recruited to suppress competitors for selection, but there is little evidence that age-related word retrieval deficits are due to competitors (Burke, ). However, other executive functions, like set-shifting or working memory updating, may be important for successful word retrieval (Shao et al., ; Higby et al., in press).

.. Neurophysiological changes in cognitive ageing Ageing brings about a variety of neurophysiological changes that affect behavioural performance. Among the most widely observed changes is axonal degradation and loss (Grillo, ). The loss of entire neurons appears to be relatively rare in non-pathological ageing, and since loss of a neuron’s entire axon would lead to its death, it is likely that axonal changes in ageing reflect a loss in the number of axonal branches. Furthermore, demyelination of axons, which refers to damage to the myelin sheath that insulates the axon and permits information to be conveyed quickly, may be responsible for some of the age-related changes in cognitive processing speed and makes the axon more vulnerable to deterioration (Kövari et al., ). Synaptic structures and functions that permit inter-neuron transmission do deteriorate with advancing age in humans (and other living beings, including rodents and drosophila (fruitflies)). Two prominent sets are synapses in the hippocampus (which is involved in explicit memory) and in the pre-frontal cortex (related to executive function) (Morrison & Baxter, ), areas that have been linked to age-related cognitive declines in memory and executive control functions (Peters et al., ). Even in those two regions, synaptic types operate differently for their respective functions, and they degrade in different ways as well, so we may expect that networks and regions more directly involved in language will include numerous types not only of neurons but also of synapses that degrade, more or less, and in different ways. Little research has been done investigating the relationship between age-related grey matter or white matter changes and lexical retrieval decline. Obler et al. () found that greater grey and white matter volume and white matter integrity in left and right periSylvian regions, as well as additional areas in the frontal and temporal lobes, predicted better word retrieval success. Shafto et al. () focused on TOT occurrences as an index of word retrieval difficulty and found that more TOT occurrences were found in individuals with more grey matter atrophy in left insula, an area of the cortex that lies underneath parts of the frontal and temporal lobes. The scale used for these types of MRI studies is far greater than the microscopic scale required to examine changes in axonal length or synaptic density; thus, we can only speculate that the changes in grey and white matter that correlate with word-finding difficulties in older adults reflect some of the cellular changes that appear to be typical of ageing animals.

.. The interaction between age-related declines and L attrition Language attrition is a slow, gradual process, so researchers often study attriters who have been away from the native-language environment for decades in an effort to capture the

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



, , , ,  

characteristics of attrition. Moreover, attrition research typically includes only those who immigrated as adults, when the native language is assumed to be fully developed. This combination of factors—late age of immigration and long length of residence—is likely to result in participants who fall within the age range normally considered to be older adults (e.g., age  and older). Thus, it is important to consider whether some of the attrition effects reported may be due, or partially due, to ageing effects (Goral et al., ; Schmid, a). As we have demonstrated, there are several similarities in the patterns of lexical attrition seen in the first language of bilinguals and in monolingual older adults. One important way in which they differ, however, is the purported source of weakened phonological connections that result in word-finding difficulty. For bilinguals, the weakened connections impeding lexical access have been attributed to reduced language use and/or to competition from word forms in the second language. The weakened phonological connections in older adults, by contrast, have been attributed to brain changes associated with advancing age (Diaz et al., ), which may weaken neural connections through reduced synapses and axonal branches and either directly affect access to a word’s form or indirectly affect access through impaired executive functions crucial to successful word retrieval (Higby et al., in press). It is currently unclear how first language attrition is associated with brain changes (i.e., the strength of synaptic connections). Since the sources leading to word-retrieval difficulty differ, it is theoretically plausible that bilingual older adults may experience even greater difficulties with lexical access as a result of additive negative effects from multiple sources: language use, cross-language competition, and brain changes associated with ageing.

. L   

.................................................................................................................................. Despite the evidence of ageing effects on lexical retrieval in monolinguals, few studies have explored the effect of ageing in late bilinguals, even in recent publications about bilingualism throughout the lifespan. Age-related lexical retrieval decline is similar for bilinguals living in their native-language environment and for monolinguals (Ashaie & Obler, ). Therefore, a key question is: do bilinguals undergoing attrition demonstrate an even greater deficit, suggesting that ageing and attrition effects are independent and additive? Within a connectionist network model, it is reasonable to expect that due to both ageing effects and attrition, age-related weakening of connections in the lexical system may impact an already weakening L system more than an L system that is in more frequent use. This is supported by evidence testing the Transmission Deficit Hypothesis from TOT studies of ageing monolinguals in whom low-frequency words (which are presumed to have weaker semantic-phonological links) are more strongly impacted than high-frequency words (Burke et al., ). Moreover, as inhibitory control abilities tend to decline with advancing age, it may become even harder for older adults to inhibit a more dominant L (Stolberg & Münch, ). In order to understand the interaction between ageing and reduced language-use effects, it is necessary to look at studies in which there is a possibility of taking both factors into account—that is, comparing older bilinguals and younger bilinguals who are living in their L environment. However, to date, there are few studies that have done this. Most studies

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

      



on attrition in older adults living in their L environment do not mention possible effects of ageing on their language measures (e.g., Altenberg, ; Jaspaert & Kroon, a; Hulsen et al., ; Kargar & Rezai, ). One study that did consider the possibility of ageing affecting L attrition limited their participant sample to those below age  years in order to avoid the possibility of ageing effects on the results (Soesman, ). Goral et al. () attempted to disentangle the effects of ageing from attrition in older adults (aged  to  years) who spoke L Hebrew and L English and were living in their L environment. They reasoned that if the declines in language processing are primarily due to ageing, similar effects should be seen when bilingual older adults are tested in their first and second languages. If deficits are seen only for the first language, then it is likely that the processes of language attrition are primarily responsible for the decline. The authors compared younger and older Hebrew-English bilinguals living in the US (attriters) and younger bilinguals living in Israel (controls) on a lexical decision task. In Hebrew, the older attriters had the longest response times followed by the younger attriters. In English, the control group was the slowest, but the two attriter groups did not differ from each other. Based on the response time patterns, the authors concluded that the slower responses in Hebrew of the older bilingual group are more likely due to effects of attrition than ageing because the older adults were slower than younger adults only in the L. However, as mentioned previously, it is possible that ageing and attrition interact such that ageing processes disproportionately affect an already weaker language system. In fact, the pattern of results reported for the L across the three groups implies that both attrition and ageing contribute to the older attriters’ performance. Not all studies have shown a clear ageing or attrition effect in elderly bilinguals. Jaspaert and Kroon () studied letters written by a native Dutch speaker who had been living in the US for over sixty years and who had had limited contact with Dutch during that time. The participant, called A.L., wrote these letters over a six-year period between his late s and early s. The authors identified lexical choice problems, such as English borrowings, in about % of the verbs used. Given the low level of L contact the participant had had over a period of more than six decades, the rate of lexical borrowings seemed low. Subsequent studies have reported a similar number of lexical problems in attriters (Schmid, ). The fact that A.L. showed a similar number of lexical problems as younger attriters suggests that ageing may not have had much of an impact on A.L.’s L lexical choices. However, written production is likely to show fewer effects of ageing than oral production since writers can take as long as they want to choose a word or phrase, use sources like a dictionary or thesaurus, and can go back and self-correct. Circumstances that change in ageing, such as retirement, may affect language use patterns, complicating the study of attrition in older populations even more. In many older people who immigrated as adults but continue to use the L at home, a shift towards using L more predominantly comes with retirement (Schmid & Keijzer, ). Schmid & Keijzer tested three groups of attriters and two groups of non-attriters (controls) within five age groups of approximately equal numbers of participants: French > English). As for code mixing, the patient almost always responded in a different language from that of the examiner. Similarly, in their study of the language behaviour of two late bilinguals (German‑Swedish and Swedish‑Finnish) with AD, Hyltenstam & Stroud () reported that both patients showed a tendency to use the language that was not shared by their Swedish-speaking addressees. Moreover, the patients’ productions showed numerous intrusions of one language, reflecting the difficulty of keeping their languages separate. Finally, the nature of the elicitation task appeared to have an effect on the patients’ choice of language. For instance, they preferred their L to recall distant memories. Although the patients’ pragmatic difficulties were particularly marked in the severe stage of the disease, they depended on a variety of factors, including the age of acquisition of the languages involved, the premorbid degree of proficiency achieved for each of them, and/or the language(s) spoken in the environment. Similar observations were reported by De Santi, Obler, et al. () and Obler et al. (). Results of the latter study, which included four English‑Yiddish bilingual patients, only one of whom was an early bilingual, indicated that with the exception of the early bilingual, all the patients had difficulty addressing their interlocutor in the appropriate language. This difficulty increased with the severity of the disease, with both languages being impaired to the same extent in the most severely affected patient. Sometimes, the two languages were so jumbled up that it was impossible to tell which language was being spoken (De Santi et al., ). When Hyltenstam & Stroud () and Hyltenstam () studied two patients with mild AD, two with moderate AD, and two with severe AD—all late bilinguals with L Finnish and L Swedish—they found that the majority of participants (four out of six) often mixed languages, despite the strictly monolingual character of the experimental conditions. There was a clear trend for patients to use their L in the situation where L use was expected. In a longitudinal study conducted with three late bilingual (German‑Dutch) patients with AD, Luderus () investigated the directionality

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi



-, , ö,  

of intrusions (L to L versus L to L) in the context of a conversation experimentally defined as strictly monolingual. Overall, their findings highlighted considerable variability in patients’ conversational profiles and, consequently, the nature and extent of code mixing. The problems of language choice and language separation were shown to be strongly predicted by the balance of bilingualism in the premorbid period. The longitudinal approach made it possible to demonstrate that when bilingualism was balanced, the problems of language selection/separation appeared later on in the course of the disease. For patients with dominant bilingualism, these problems appeared earlier, and also had a predetermined directionality, depending on the premorbid dominance of L or L. Likewise, Mendez et al. () showed that bilingual patients with AD produce significantly more L intrusions in a later learned L. Language choice and separation in bilinguals rely on the ability to take account of the specificity of the communication situation in order to inhibit the non-relevant language. The causes of separation and code mixing problems should therefore be sought primarily in executive functioning deficits. Moreover, the processes involved in bilingual language production are based on a network of connections between the prefrontal cortex, anterior cingulate cortex, inferior parietal region, and basal ganglia (Abutalebi & Green, ). Solving conflicts in language selection normally requires the intervention of specialized structures, namely the dorsolateral prefrontal cortex and anterior cingulate gyrus. Conflict arises in the inferior parietal cortex, but spreads to linguistic areas in the inferior frontal cortex. All these cortical structures are connected through subcortical structures to the basal ganglia, in particular the caudate nucleus, which is responsible for conflict resolution. Given that AD gradually affects (among other things) frontal and cingulate regions, as well as the subcortical nuclei, we can assume that bilingual patients increasingly encounter between-language competition and conflict resolution difficulties (i.e., inhibition of the non-target language). The term inhibition is used to describe a cognitive control mechanism that tunes out the language that is irrelevant to the communicative context at hand (Abutalebi & Green, ; Costa & Sebastián-Gallés, ). In bilingual conversations, it prevents the speaker from producing utterances in the undesired language by keeping the irrelevant language below the selection threshold. The monitoring process that determines whether cognitive control mechanisms need to be applied to keep speaking in one language (i.e., activation) or switch to the other one (i.e., inhibition) is supported by the executive control system. Neuroimaging studies have shown that bilingualism affects the brain structures involved in the executive control system (e.g., greater density of grey matter in bilinguals’ anterior cingulate cortex). The logical outcome of a cortical impairment affecting the neural basis of executive control (i.e., frontal and cingulate regions) is therefore language choice difficulties. The study of how bilingualism affects the neural basis of executive control processes has only recently commenced, and research on language changes in patients with AD is a promising avenue for extending current understanding of language control in bilingual speakers. It remains unclear whether code mixing/switching behaviours in bilingual patients with AD are due to a deficit in inhibitory control or to language loss (i.e., language attrition/regression). For the time being, the variety of experimental paradigms, the heterogeneity of patients’ medical and linguistic profiles, the limited number of observations, and the inter-individual variability mean that it is still difficult to assess the impact of AD per se.

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

     



. C

.................................................................................................................................. As underlined in Stilwell et al.’s recent review (), studies of language change in bilingual patients with AD are relatively inconclusive regarding the differential effects of AD on the two languages. Although they are not very informative regarding the order of L versus L deterioration in the course of the disease, these studies nonetheless raise important theoretical and methodological questions about how to test the multifaceted nature of bilingualism. For example, further research is needed to verify whether richer connections between concepts and words—which are characteristic of the dominant language—should be regarded as a strength or a weakness in the context of neurodegenerative diseases and attrition. Likewise, the issue of how to determine participants’ dominant language, which surprisingly was never experimentally assessed in the above-mentioned studies, should be carefully taken into consideration when testing bilingual patients with AD, as it is a crucial dimension for understanding the direction of language decline. Previous discrepant results may have stemmed from the fact that patients’ first-acquired language was not necessarily their dominant language (often the case in the context of migration). Furthermore, it is still unclear whether pathological ageing reverses language attrition as postulated by the reversion hypothesis put forward by de Bot & Clyne in .

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

  ......................................................................................................................

     ......................................................................................................................

    

T chapter provides an overview of the first few Event-Related brain Potential (ERP) studies on L attrition that have been conducted in recent years and will briefly discuss their results as well as future directions. As we will see, ERP studies are particularly suited to advance attrition research due to their power to track even subtle changes in cognitive processing with a considerable level of detail. In our opinion, the labels ‘attrition’ and ‘attriter’ must be reserved for cases where individuals undergo neuro-cognitive changes in how they process language that are nonpathological and induced by (environmental) changes in their language use (Kasparian & Steinhauer, b). Thus, while migrants immersed for years in a second language environment clearly constitute the prototypical example, neither are L immersion or bilingualism in absence of cognitive changes in L processing sufficient to justify the label ‘attriter’ (but see Schmid, a; Schmid & Köpke, a, for different views), nor is long-term immersion a necessary prerequisite. That is, attrition may also occur at relatively early stages of bilingualism (cf. Linck et al., ), possibly even without migration, or after extended periods of discontinued L exposure with little or no influence of a second language (e.g., Baladzhaeva & Laufer, )1. If the central criterion for attrition is linked to neuro-cognitive changes in language processing, and if some such attrition effects are relatively subtle (and controversial), then a research technique capable of tracking subtle cognitive changes is crucial. Importantly, a number of previous electrophysiological studies on L acquisition have demonstrated that ERPs may reliably reveal certain processing differences even when behavioural data do not (yet) reflect them (Friederici et al., ; McLaughlin et al., ; Morgan-Short et al., ). Since the early s, ERP techniques have enjoyed increasing popularity as an affordable online measure in psycholinguistic research, including in 1 Relevant scenarios of possible L1 attrition in absence of bilingualism may also include monks/nuns who gave a vow of silence, or individuals stranded on a desert island.

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

    



bilingual populations. Given that numerous introductory chapters and review articles on ERP research in L and L are available (Kaan, ; Osterhout et al., ; Steinhauer & Connolly, ; Kotz, ; Van Hell & Tokowicz, ; Steinhauer, ), here we will only summarize a few methodological essentials, central findings, and current controversies that set the stage for the novel field of ERP research on L attrition.

. E-R P     -

.................................................................................................................................. With two to five words per second, listeners and readers process language at a very fast rate. In order to investigate how phonological, morphological, semantic, syntactic, and discourse information is integrated in real time, most behavioural approaches and brain imaging techniques (such as PET or fMRI) are too slow. In contrast, ERPs reflect a reader’s or listener’s neural responses to linguistic stimuli with virtually unlimited temporal resolution in the range of milliseconds, without any delay, and continuously across entire sentences (i.e., for each single word). Moreover, as a number of psycholinguistic processes are reflected by distinct brain waves (ERP components) that differ in latency, polarity, and scalp distribution, ERPs are ideal to uncover the nature of processing difficulties and the contribution of different types of information (e.g., prosody, syntax, and semantics). Thus, ERP data permit us to (a) identify which word(s) in a sentence may cause processing difficulties, (b) link the findings to psycholinguistic processes, and (c) investigate if and how L learners and attriters may differ from monolinguals and from each other. To briefly illustrate the general approach, let us assume we are interested in the ERP profile for conceptual-semantic anomalies (e.g., ‘John ate democracy at dinner’), as compared to matched correct control sentences (‘John ate broccoli at dinner’; target words for the ERP analysis are underlined). As a first step, electroencephalographic (EEG) data are collected from an array of cap-mounted scalp electrodes while the participant reads or listens to a randomized list of sentences from the various experimental conditions (Figure .). Each time a target word is presented, a condition-specific ‘trigger code’ is sent to the EEG system marking the word onset in the continuous stream of EEG data. These triggers are then used to identify and extract the relevant epochs of the EEG that reflect the processing of target stimuli (‘time-locking’). Most psycholinguistic processes related to a given target word occur within the first , milliseconds (ms) after word onset and are reflected by characteristic negative and positive deflections in the ERP (socalled ‘ERP components’). Early brain waves during the first  ms after word onset (such as the N and the P components) primarily reflect physical stimulus characteristics (such as font size or acoustic intensity), whereas later components are linked to higher cognitive processes. For instance, the lexical-semantic anomaly in our example elicits a larger negativity between  and  ms for the implausible word ‘democracy’ than the plausible word ‘broccoli’ (see Figure .). As this deflection has a negative polarity and peaks around  ms, it is referred to as the N component. The N amplitude increases gradually the more difficult the lexical retrieval and semantic integration of a target word is (Kutas et al., ; Steinhauer et al., ). As illustrated by the voltage map in

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi



   EEG → Event-related brain potentials (ERPs) John ate broccoli at dinner. John ate democracy at dinner.

–5μV

N400 effect N100 Voltage map of N400 effect

EEG amplifier

+5μV

200 400 STIMULUS P200 ONSET

600

800 Time

1000

ms

300 – 500 ms

 .. EEG set-up, lexical-semantic N effect, and a corresponding voltage map EEG data collected from an electrode cap during sentence processing are amplified, averaged, and time-locked to the onset of the target words (black line—control: ‘broccoli’, red line—mismatch: ‘democracy’). The (schematic) ERP plot shows similar brain waves in both conditions during the first  ms and an enhanced N for the semantic mismatch condition (negative polarity is plotted upwards). The voltage maps illustrate the centro-parietal scalp distribution of the N effect between  and  ms (mismatch minus control). See main text for details. Source: K. Steinhauer (), Applied Linguistics,  (), –, with permission from Oxford University Press.

Figure . (right panel), the N effect (i.e., the N difference between implausible and plausible words) has a broad distribution over the scalp but is most prominent at central-parietal electrodes near the midline of the scalp. Once we have established such a pattern in native monolinguals, we can see if L learners or attriters display the same ERP profile, or if their N effect is smaller, delayed, prolonged, or accompanied by other ERP effects that may point to deviant online processing. Corresponding ERP data for the online processing of more subtle semantic anomalies in all three groups will be presented below in Section .. In contrast to lexical-semantic anomalies, increased structural processing costs due to complex syntax or grammar violations are most reliably associated with enhanced posterior positivities between  and , ms (P components; Osterhout & Holcomb, ). In many group studies on grammar violations, Ps are preceded by Ns or left anterior or left temporal negativities (LANs or LTNs, respectively; see Steinhauer et al., ; Molinaro, Barber, et al., ). Some researchers have argued that early LAN-like negativities may be elicited by the fast and automatic detection of a grammar violation (possibly subserved by language-specific circuits in Broca’s area), while late Ps may reflect subsequent and more controlled attempts to reanalyse or mentally repair the grammatical structure (Friederici, ). More recently, other authors have focused on the substantial inter-individual variability (even among monolinguals) as to which effect (negativity or P) is dominant (Tanner & Van Hell, ), potentially suggesting the ERP components may reflect distinct processing strategies that rarely co-occur within the same trial. In this chapter, we will address possible factors underlying inter-individual variability only in the context of bilinguals.

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

    



A last question concerns the domain specificity of ERP components. While the information provided by ERP data is undoubtedly much richer than response time measures and error rates in behavioural studies, ERP components should not be viewed as distinct markers for language-specific cognitive processes. Research in other cognitive domains has shown that ERP effects similar to those found in language studies can also be elicited by other stimulus materials, especially in music (e.g., Patel et al., ). Moreover, there is some evidence suggesting that at least certain ERP effects (such as the P) may consist of sub-components that reflect different kinds of cognitive operations (e.g., Steinhauer & Connolly, ). However, ‘even if these ERP components are not specific to language, they may still reflect cognitive processes that are both characteristic and tightly linked to natural language processing’ (Steinhauer, , p. ).

. H     ? E-R P  L   

.................................................................................................................................. The first ERP studies on morpho-syntax comparing late L learners to native speakers found significant group differences that increased with increasing age of acquisition (AoA) (e.g., Weber-Fox & Neville, ), thus apparently providing hard brain evidence supporting the ‘critical period hypothesis’ (Lenneberg, ). A common conclusion was that loss of brain plasticity in puberty may prevent late L learners from relying on native-like brain mechanisms (and thus from attaining native-like L proficiency), and that these maturational constraints primarily affect L phonology and morpho-syntax, as compared to vocabulary learning and lexical processing (Newport et al., ). However, confounds between AoA and L proficiency levels in these early studies have been criticized, and ERP studies that unconfounded the two factors reported that L proficiency was a better predictor for nativelike ERPs than AoA, with little evidence for a critical period, arguing instead for a proficiency-dependent convergence on native-like processes even in late L learners (e.g., Steinhauer et al., ). Currently, the data across studies are mixed, and many other factors contributing to the variability in ERP profiles are being discussed, including the length and type of L exposure, the role of classroom instruction versus immersion, transfer effects in different L–L pairings, modality effects, motivation, and the specific structures under investigation (Steinhauer, ; Tanner et al., ; Meulman, Stowe, et al., ). Attriters constitute an important complementary group of interest for various reasons. First, where robust differences between native speakers and L learners have been found, these may be partly due to the difference between ‘being monolingual’ and ‘being bilingual’. As pointed out by Schmid (c), data from native but bilingual attriters may help fill this gap (as might data from simultaneous bilinguals; e.g., Molnar et al., ). Second, an interesting question is whether it is actually possible to ‘unlearn’ a full-fledged L system and whether and to what extent this process of losing one’s L resembles the stages of L acquisition, but in reverse. If structure-specific language proficiency is a strong predictor for neurocognitive language mechanisms as reflected by ERPs (Steinhauer, ), then one might expect similar ERP profiles in L attriters and proficiency-matched L learners of the

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi



  

same language. Recall, however, that ERPs may sometimes reveal differences that are not accurately detected in behavioural tasks (e.g., McLaughlin et al., ). Third, if attriters are more proficient in their L than their L, does this mean their ERP profiles are more ‘nativelike’ in their L than in their L as well? This kind of question requires one to test attriters in both of their languages (Kasparian, ). Fourth, which language domains are most vulnerable to attrition, and do they overlap with those discussed for critical periods in L acquisition (potentially pointing to shared restrictions in brain plasticity across learning and forgetting)? Fifth, what is the relationship between success in L acquisition and the occurrence of L attrition? Are they two sides of the same coin? Are they influenced by the same factors (such as type and intensity of L exposure; cultural identity, etc.)? Before future research can address these specific questions in detail, much current research attempts to determine which linguistic domains are actually subject to L attrition effects. Especially in the area of morpho-syntactic processing, there is little agreement on whether attrition occurs at all once a full-fledged first language has been established. To our knowledge, there are currently only two research groups that have started to systematically investigate the neurocognitive underpinnings of L attrition using ERPs: Monika Schmid’s group (now in Essex) focusing primarily on spoken sentences, and our own group in Montreal (in collaboration with Francesco Vespignani in Italy), testing written sentence materials. Relatedly, Judith Kroll and colleagues (e.g., Linck et al., ) have investigated L-induced changes to L in immersed and classroom L learners, and have thus far discussed their findings in terms of L and L co-activation effects and in terms of reduced L access during L immersion (without explicitly labelling such effects ‘attrition’). Although we will argue that these investigations of ‘early attrition’ are definitely warranted and can serve to address several open empirical questions (see Section .), we focus our current discussion on those ERP investigations that have targeted L attriters who have migrated to a new country for the longer term. Monika Schmid’s group has established a large-scale research program on the morphosyntactic processing of grammatical gender (GG) across various European languages (especially Dutch and German). GG is known to be notoriously difficult in L acquisition, while only few data are available for attriters. Important aspects of this research program include (a) comparisons of ERP data from attriters, L learners, and monolinguals to address effects of AoA and bilingualism, (b) the consideration of proficiency levels in the target language, and (c) cross-linguistic comparisons for different L–L pairings. The research program in Montreal started in  and is largely based on the second author’s PhD thesis project (Kasparian, ). It focuses on adult Italian immigrants in Montreal who now use English as their main language and have identified signs of losing their first language. While this project does not include multiple L–L pairings, it is special in that it tests attriters in both their L (Italian) and L (English) and compares them with proficient English learners of Italian L as well as with both English and Italian monolinguals (the latter were tested in Italy). Similar to the European research program, multiple behavioural measures on L and L proficiency, length of residence, and language-use were collected in addition to ERP and behavioural data. Moreover, the linguistic paradigms tested with ERPs are quite diverse and include lexical-semantic processing as well as morphosyntactic processing in a range of different sentence structures (in Italian and English). In the following, we will provide an overview of recent findings from these groups, in the areas of morpho-syntactic processing as well as lexical processing.

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

    



. M-   ’ L

.................................................................................................................................. The first three ERP studies in this review set out to determine whether and to what extent attriters show altered online processing for grammatical anomalies at the sentence level. As already discussed in previous chapters, whether attrition effects in this linguistic domain exist at all is highly controversial.

.. Grammatical gender violations in German and Dutch In a collaboration with ERP expert Laurie Stowe, Schmid and colleagues conducted a series of ERP studies on GG processing in monolinguals, L learners and L attriters of various European languages. When studying bilingual sentence processing, focusing on GG has a number of advantages. First, acquiring a new gender system is known to be very difficult for L learners, and there is some debate whether it is more challenging if one’s first language has a different mapping or no gender system at all (Tokowicz & MacWhinney, ; Sabourin et al., ). In many languages such as Dutch, French, and German, each noun has a GG that needs to be learned, and the mapping is largely arbitrary (i.e., independent of conceptual-semantic or phonological cues). During real-time sentence processing, a reader or listener must retrieve the idiosyncratic gender information of a given noun from longterm memory (LTM) and check this feature with those of the determiner and the inflectional morphemes of other elements. Although GG mismatches (e.g., using the wrong determiner) do not usually change the meaning of a given noun phrase (NP), native speakers perceive them as relatively salient morpho-syntactic violations (especially in speech) and such instances therefore reliably elicit P effects, often preceded by either LANs or Ns. In fact, there is substantial ERP literature on gender violations across languages, in both native speakers (e.g., Molinaro, Barber, et al., , for a review) and L learners (see overview in Meulman, Stowe, et al., ), but not yet for attriters. Schmid and colleagues’ spoken sentence materials included subject NPs in which the determiner either agreed or disagreed with the GG of the noun. For example, the German noun Garten (garden) is masculine and requires the masculine determiner ‘der’, while the neuter determiner ‘das’ would result in a gender mismatch on the target noun (underlined in ()): () . . . der Garten / das *Garten . . . ‘themasc gardenmasc / theneut *gardenmasc’ () . . . der kleine Garten / das kleine *Garten . . . ‘themasc small garden / theneut small *gardenmasc’ () . . . hat die Rose geblüht / *blühen . . . ‘the rose has bloomed / *bloom’ To prevent simple chunking strategies based on familiarity, a second experimental condition () added an intervening adjective that separated the determiner and the noun. Because the adjective is marked only for number and case but not gender, its presence does not directly contribute to the GG agreement or disagreement. Lastly, the researchers also included a third morpho-syntactic violation condition not involving GG that was expected

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi



   Controls Monolingual L1 speakers (n = 29)

Attriters Late L1 attriters (n = 27)

Early learners L1 without L1 with gender gender (n = 17) (n = 26)

Late learners L1 without gender (n = 17)

L1 with gender (n = 40)

Verb form

Gender: Det + N

Gender: Det + Adj + N

 .. Voltage maps summarizing the (preliminary) main findings for agreement violations in ‘finite verb’ and GG conditions These voltage maps represent top views of the human scalp (i.e., the ‘nose’ in each map points upwards) and illustrate the ERP differences between conditions (violation minus control) in a representative time window. Colours correspond to a ‘heat’ map, that is, dark blue indicates strong negative potentials and dark red indicates strong positive potentials. Schmid and colleagues focus primarily on the presence versus absence of a large P, represented by a large red area over centro-parietal scalp sites. While monolingual controls and Attriters display such a large P component, late L learners show consistent Ps only for the verb condition. In early L learners, having a GG system in L seems to lead to more native-like P patterns. Source: Schmid (c).

to be detectable even by less proficient L learners. As illustrated in example (), this ‘finite verb condition’ replaced a past participle (e.g., geblüht; bloomed) with the infinitive (*blühen; to bloom). A corresponding design was also used in Dutch, where native speakers were compared to L French late L learners of Dutch and each group performed both in an auditory and a visual version of the ERP experiment (Meulman, Stowe, et al., ). Since only some of the final data have been published to date (Bergmann, Meulman, et al., ; Meulman, Stowe, et al., ), the results reported here are partly based on conference presentations (Schmid, c; Schmid et al., ). As illustrated in Figure ., monolingual native speakers reliably elicited the expected P components for all three types of violations (an additional N preceding the P was found for the Dutch verb condition, by Meulman, Stowe, et al., ). The presence of an adjective in () versus () did not seem to have an effect in any group. Second language learners generally elicited native-like Ps for the finite verb condition (as well as the N in Dutch auditory sentences), whereas their ERPs for gender agreement varied primarily as a function of AoA. Early L learners (AoA <  years) showed a P similar to monolinguals (especially when their L had a gender system), whereas late learners (AoA >  years) displayed no typical P (Schmid, c). In the Dutch study by Meulman, Stowe, et al., the majority of French L learners did not show any ERP violation effect for GG violations. Importantly, and in stark contrast to the L

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

    



learners, the group of attriters was indistinguishable from monolinguals in that all of them elicited P components for all violation types (Bergmann, Meulman, et al., ; Schmid et al., ). These findings were interpreted as evidence that morpho-syntactic knowledge stabilizes around puberty. For late learners acquiring their L after the age of  years, this means that they may be able to rely on native-like processing mechanisms for the finite verb condition (possibly transferring knowledge from their L) but fail in this respect for structures that are different or absent in their L, such as GG (Meulman, Stowe, et al., ). In attriters, on the other hand, the ‘stability of a deeply entrenched L system’ (Bergmann, Meulman, et al., , p. ) apparently served as a protective factor from losing their native morphosyntactic processing mechanisms, such that all ERP profiles of the attriters under investigation remained native-like even after many years of dominant L exposure and use after immigration. This view is in line with the behavioural findings of one of the first cognitive attrition studies by Scherag et al. (), who also found no indication for differences between monolinguals and bilingual migrants, either for lexical-semantic or morphosyntactic processing. With respect to Schmid and colleagues’ research objective of teasing apart the contributions of various factors to differences between L learners and native speakers, the attriters’ data suggest that merely ‘being bilingual’ has little or no impact on the real-time processing of one’s first language. A number of points regarding these studies are noteworthy. With respect to L learners, the article by Meulman, Stowe, et al. () reminds us that data for GG violations are not consistent across studies. While their own analysis convincingly showed that neither proficiency, length of residence, L use, nor AoA (after age ) had any impact on the P for GG violations, several other investigations did observe these effects, as well as native-like ERP profiles including P effects for GG violations—even in late L learners (Tokowicz & MacWhinney, ; Gillon Dowens et al., ). This variability across studies certainly deserves attention in future research, and the authors provide a number of interesting suggestions (e.g., to focus on modality and L/L pairings). Another aspect of Meulman, Stowe, et al.’s () study is the striking observation that the L learners’ accuracy in a grammaticality judgement during the EEG experiment was much lower for the GG condition (.%) than for both the finite verb condition (.%) and an offline task in which participants had to assign GG to a list of nouns used in the EEG experiment (.%). We will return to this point below in Section .. Finally, given that GG online processing involves both the quick retrieval of a noun’s idiosyncratic gender information from LTM and subsequent morpho-syntactic feature checking with the determiner, low accuracy scores in the judgement task and the absence of P effects can be due to problems with either of these sub-processes. According to most psycho- and neurolinguistic models, the two sub-processes involve very different cognitive mechanisms and brain circuits (e.g., declarative and procedural memory systems, respectively, according to Ullman, ). As the finite verb condition (which relied almost exclusively on morpho-syntactic processing/procedural memory) did not pose a major challenge to the L group and elicited native-like ERP effects, it may very well be that the GG data collected by Schmid and colleagues do not point to morpho-syntactic problems in a late acquired L, but rather to problems in quickly and reliably retrieving the correct gender information from LTM—and thus a lexical problem (see also Section .). This interpretation would be in line with Hopp’s

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi



  

‘lexical bottleneck’ hypothesis for a range of apparently morpho-syntactic processing difficulties in L acquisition (e.g., Hopp, , b). With respect to attriters, the available GG data clearly support Schmid and colleagues’ notion of a well-preserved L system for GG processing. Especially since this paradigm critically relies on fast native-like lexical access as well as fast native-like morpho-syntactic integration, an ERP profile that is indistinguishable from that of monolingual speakers provides evidence that both of these sub-processes must be intact in attriters, at least for certain structures. In other words, with a single paradigm the researchers in Groningen seemed to have demonstrated attriters’ preserved native-likeness in two linguistic domains (lexicon and morpho-syntax) and for multiple complex neural circuits in the brain (e.g., declarative and procedural memory), at least for certain linguistic structures. Regarding neural plasticity over the life span, the attriters’ GG data appear to confirm what Meulman, Stowe, et al.’s results in late L learners already suggested, namely that GG processing may not be subject to much neurocognitive plasticity—especially in adulthood: GG seems to be both difficult to acquire (by Lers) and to lose (by attriters). An important question is to what extent these findings in attriters can be generalized. For example, Hopp & Schmid () reported that some L attriters do show signs of nonnative-likeness in L pronunciation (‘foreign accent’), whereas a few late L learners are indistinguishable from native speakers. They concluded that ‘acquiring a language from birth is not sufficient to guarantee nativelike pronunciation, and late acquisition does not necessarily prevent it’ (p. ), suggesting that native-likeness is not a universal phenomenon in attriters across linguistic domains. But as illustrated by the inconsistency of ERP profiles across L studies on GG violations, even within the domain of morphosyntax, a single ERP experiment is unlikely to tell the whole story. The next sections will present complementary ERP data on attrition, collected from Italian immigrants in Montreal.

.. Number agreement violations at various positions in Italian sentences A study by Kasparian et al. (, ) also investigated morpho-syntactic agreement violations. However, here the focus was not on GG but on number agreement (singular vs. plural; sg/pl) between (a) the subject NPs (e.g., the workers) and the adjacent verb and (b) the subject NP and a distant modifier (adjective) in a given sentence, as illustrated in () and (): () I lavoratori tornano dalla fabbrica sporchi di grasso The workers(pl) return(pl) from the factory dirty(pl) with grease () I lavoratori *torna dalla fabbrica *sporco di grasso The workers(pl) *returns(sg) from the factory *dirty(sg) with grease The four conditions of this study crossed agreement and disagreement on the verb and the adjective such that (a) both agreed with the subject NP (as in ()), (b) only the verb disagreed, (c) only the adjective disagreed, or (d) both verb and adjective disagreed with the subject NP (as in ()). The number of the subject NPs (sg/pl) was balanced.

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

    



The general paradigm was adopted from a previous study with native Italian monolinguals by Molinaro, Vespignani, et al. () who also examined the ERP responses on both the verb and the modifier. In this study (as well as all other Kasparian & Steinhauer studies reviewed in this chapter), the behavioural task during EEG recording was to rate the acceptability of each sentence on a scale from  (unacceptable) to  (perfect). Because various recent ERP studies on agreement violations in L learners had found that long-distance agreement relationships are more likely to reveal differences between bilinguals and monolinguals (e.g., Gillon Dowens et al., ), one important research question was whether differences might be found for attriters as well. A second question had to do with cross-linguistic differences between Italian and English. Disagreement between an NP and a subsequent verb is a reliable indicator for a violation in English, but not in Italian (due to its freer word order). If attriters reading sentences in their L (Italian) were in fact influenced by their dominant L grammar (English), one might expect that their ERP response to agreement violations might be stronger than that of the monolingual control group (for details on the various tasks measuring L and L proficiency in both groups, please refer to Table  of Kasparian et al., ). Kasparian et al. () found group differences at both the verb and the modifier position. As shown in Figure ., number disagreement on the verb elicited a weak LTN in the control group (n = ), followed by a frontal Pa component (likely reflecting a shift of attention towards the anomaly) and a subsequent posterior P that lasted from  to , ms post target word onset. In contrast, attriters (n = ) showed a significantly larger and more broadly distributed left-lateralized negativity, followed by a Pa and a P that ended around , ms. The significant group differences for the negativity were interpreted as reflecting a more salient violation in attriters, due to the influence from English word-order. An alternative account in terms of component overlap with the subsequent Pa (which may have cancelled out the negativity in the control group; cf. Tanner, ) could be ruled out, as this positivity was actually larger in attriters. However, the shorter duration of the P in attriters is somewhat at odds with the notion of a stronger salience of the agreement violation in this group. Did the stronger initial response (reflected by the larger negativity) facilitate and speed up subsequent reanalyses? Or do the duration differences point to qualitative processing differences? These questions gain additional importance from the finding that the disagreement P elicited by the modifier was again several hundreds of milliseconds shorter in attriters than controls (while the preceding negativity was numerically larger in attriters). Although this issue cannot be completely solved with the data at hand, Kasparian et al. () provide a number of arguments why the shorter P in attriters is likely to reflect less ‘in-depth’ structural repair. First, more efficient processing should have resulted in faster response times, but attriters were slower than controls in this experiment (see Figure b in Kasparian et al., ). Second, the authors checked if this group of attriters just happened to be generally slower in their behavioural responses, or if their P was generally shorter across all experiments. However, data from other sentence conditions included in the same experimental session showed this was not the case. Third, Kasparian et al. argue that the P has been shown to reflect multiple cognitive processes, that is, it is not a monolithic component. Among other accounts, Friederici et al. () have suggested that the first portion of the P in violation paradigms may reflect a diagnosis of the problem while the late portion is associated with actual structural repair. The observation

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi



  

(a) Controls –4

–0.2

Pz

0.2 0.4 0.6 0.8 1.0 1.2

4

Negativity 300–500ms

Frontal pos. 550–650ms

P600 650–1000ms

Late P600 1000–1200ms

Negativity 300–500ms

Frontal pos. 550–650ms

P600 650–1000ms

Late P600 1000–1200ms

(b) Attriters –4

–0.2

Pz

0.2 0.4 0.6 0.8 1.0 1.2

4 I1 lavoratore| *tornano The worker | *return I1 lavoratore| torna The worker | returns

 .. ERPs elicited by the verb in response to number agreement violations (red) compared to correct sentences (green) in Controls (a) and Attriters (b), with voltage maps Time ranges (in milliseconds) depicted on the x-axis are relative to the onset of the verb ( ms). Negative values are plotted up. Voltage maps illustrate the scalp distribution of the effects observed for the time-windows of interest. Controls showed a small LTN, followed by a frontal positivity and a posterior P lasting until , ms. Attriters showed a robust N, a numerically larger frontal positivity and a broadly distributed P that was shorter in duration than that of the Controls. (Note that for this contrast we used a non-standard baseline interval from  to  ms; for details, see Kasparian & Steinhauer, .)

that the scalp distributions of the early and the late portion differed in the present study supports the notion of distinct neural generators, and hence distinct cognitive processes. Moreover, the early P portion (whose topography did not differ between groups) was modulated only by proficiency, whereas the late portion was influenced by the factor of group (present only in monolingual Italians). Interestingly, the more the attriters were still exposed to Italian on a daily basis, the more they resembled the monolinguals in showing the late P effect (as revealed by a highly significant correlation). In sum, contra the findings by Schmid’s group, this study reports some initial ERP evidence for L attrition effects in the domain of morpho-syntactic agreement violations. Influenced by the stricter word order in their co-activated English grammar, attriters seemed to have perceived agreement violations in Italian as more salient than monolinguals, but engaged less in reanalysis processes. The shift towards this new processing strategy was the stronger, the lesser attriters were still exposed to their native language. The hypothesis that some processing changes in these attriters were due to the coactivation of their English grammar was even more directly addressed in another ERP study investigating Italian relative clauses, as will be described in the next section.

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

    



.. Relative clauses in Italian immigrants: can one’s L render the L ungrammatical? Kasparian & Steinhauer (a) compared the online processing of Italian relative clauses in attriters to that in native monolinguals still residing in Italy. Given important crosslinguistic differences between Italian and English relative clause constructions, these structures offer an interesting opportunity to study the processing of complex morphosyntax as well as the influence of the L grammar on the L (cf. Dussias & Sagarra, ). Both languages allow V(erb)-NP-subject2 () and NP-V-object () word-orders. In contrast, V-NP-object () and NP-V-subject () structures are ungrammatical in English, though syntactically permissible in Italian (note: (S) - subject NP; (O) - object NP). () V-NP-subject:

Il poliziotto (S) che arresta i ladri (O) registra i nomi. The policeman (S) that arrests the thieves (O) registers the names.

() V-NP-object*:

I ladri (O) che arresta il poliziotto (S) attendono in macchina. The thieves (O) that arrests the policeman (S) wait in the car.

()

NP-V-subject*: Il poliziotto (S) che i ladri (O) arresta registra i nomi. The policeman (S) that the thieves (O) arrests registers the names.

()

NP-V-object:

I ladri (O) che il poliziotto (S) arresta attendono in macchina. The thieves (O) that the policeman (S) arrests wait in the car.

As English has a less detailed system of morphological markers, English speakers have been shown to rely more heavily on word-order for sentence interpretation. Conversely, Italian has a relatively free word order and rich morphological marking system, thus number agreement and semantic information (e.g., animacy, prominence) are more salient cues than word-order in identifying grammatical and thematic roles (subject/object, agent/ theme) of NPs in a sentence (see ‘Competition Model’, Bates et al., ; BornkesselSchlesewsky et al., ). In this study, sentences were disambiguated at the verb of the relative clause, by virtue of its number (sg/pl) as well as its semantic bias; strong agentpatient–verb combinations (e.g., policeman/arrest/thief ) made it so that only one of the two NPs was a plausible subject of the verb. Unlike monolingual Italian controls who could rely on these cues, attriters were expected to be influenced by their dominant L-English grammar and to process V-NP-object and NP-V-subject sentences as outright grammatical violations, with little or no reliance on semantic cues to revise the interpretation of the sentence. Results showed that attriters differed from native controls in their acceptability judgement ratings and ERP responses when relative clause constructions were ungrammatical in English, though grammatical in Italian. Attriters gave lower end-of-sentence acceptability judgement ratings than Controls for the two conditions that were ungrammatical in English, but did not differ in ratings of sentences that were grammatical in both languages. In response to V-NP-object sentences which are unpreferred compared to subject relative clauses but still syntactically acceptable in Italian, controls showed an N effect between 2 Note that ‘V-NP-subject’ is used as the short hand for ‘NP-V-NP subject-first’ as all structures start with an NP and are either subject-first or object-first.

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi



   (A)

–4.0

Fz 0.5

Controls

Attriters

–4.0

0.5

1.0

300–400 ms

N400

4.0

4.0

650–850 ms

–4.0

Fz 0.5

Fz

0.5

1.0

4.0

N400

No N400

–4.0

Fz 0.5

–4.0

1.0

4.0

4.0

–4.0

Fz

Larger and earlier P600 No N400

850–1050 ms

1.0 Late P600

–2.5

μv

1.0

Fz

0.5 4.0

+2.5

1.0

Larger and earlier P600

V-NP-subject V-NP-object

Controls

(B) –4.0

Attriters –4.0

Fz 0.5

1.0

Fz 0.5

4.0

ns.

4.0

Frontal positivity

–4.0

ns. Cz

–4.0

N400 Cz

0.5

1.0

4.0

–4.0

0.5

1.0

1.0

4.0

ns.

–4.0

Pz 0.5

1.0 Small P600

4.0

N400

Pz

0.5 4.0

1.0 Large P600

NP–V–subject NP–V–object

 .. ERPs for Italian relative clauses (A). ERPs elicited by the verb of the relative clause in V-NP-object sentences (pink) compared to V-NP-subject sentences (green). Controls elicited an N effect and a small P. In contrast, Attriters did not show an N effect, and the P effect was earlier, larger and broadly distributed. (B). ERPs elicited by the verb of the relative clause in NP-V-subject sentences (red) compared to NP-V-object sentences (blue). Controls only elicited a small posterior P effect in response to subject sentences, whereas Attriters showed an N-like negativity, an early fronto-central positivity followed by a larger P effect for subject sentences.

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

    



– ms at the disambiguating verb (Figure .A), indicating that they were sensitive to the semantic cues that served as extra disambiguating information to identify the subject of the sentence (Di Domenico & Di Matteo, ; Mecklinger et al., ). The N effect in native controls was larger in participants who gave more favourable ratings and therefore did not index a violation effect. The N in controls was followed by a late P effect ( to , ms) indicative of a mild late effect that reflects re-analysis and repair processes. In contrast, attriters did not show an N effect (irrespective of whether they provided high or low acceptability ratings) and elicited an earlier and more broadly-distributed P ( to  ms). In their article, Kasparian & Steinhauer (a) present arguments as to why the P in attriters reflected the perception of a stronger anomaly and higher processing costs related to their reduced use of semantic information and stronger expectations of number agreement (due to English) between the sentence-initial noun and the verb. Larger P amplitudes in attriters were associated with a higher degree of L-English proficiency and a longer length of residence (increased L immersion). In response to NP-V-subject garden path sentences in Italian (Figure .B), controls elicited a small, late, posterior P starting around  ms. In contrast, attriters displayed a strong, widespread N-like negativity, followed by a larger frontal positivity and a more robust early P starting around  ms. As in V-NP sentences, this pattern fits the view that, due to L English influence, attriters were more sensitive to word-order preferences than to semantic cues, and processed permissible Italian sentences as outright morphosyntactic violations. In sum, attriters with limited L exposure/use and predominant exposure to their L showed distinct processing routines compared to native controls. Within attriters, higher levels of L immersion and proficiency were associated with increased L influence on the L. This study was novel in showing that high levels of L proficiency and exposure may render a grammatical sentence in one’s native language ungrammatical. This finding can be viewed as strong evidence that L attrition does take place in the domain of morpho-syntax, at least for some structures. It also raises the (old) question of when and to what extent L attrition is due to interference with the competing second language (rather than discontinued use of the L). This question can only be answered by testing different L–L pairings that differ in terms of structural overlap, or (at least in principle) by investigating language processing in people who have experienced strongly reduced use of their L for an extended period, but without exposure to another language (e.g., certain monks or nuns following a vow of silence; see Sharwood Smith & Van Buren, ). We will now turn from morphosyntactic to lexical processes.

. L-   ’ L  L

.................................................................................................................................. Word finding problems are among the most common phenomena reported by attriters, and a number of behavioural studies using production tasks have confirmed that lexical access is slower and L vocabulary smaller or less diverse in attriters compared to

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi



  

monolinguals (e.g., Schmid & Köpke, ; see also Schmitt, Chapter , this volume). This section presents two ERP experiments contrasting attriters’ brain responses to lexical anomalies with those in L learners and monolingual controls. The first study focuses on confusable words in attriters’ L (Italian), whereas the second one addresses the processing of cross-linguistic homographs and cognates in English sentences (i.e., the attriters’ L), with a particular interest in whether attrition affects co-activation and inhibition effects in the bilingual lexicon.

.. A study on confusable words in Italian sentences Kasparian & Steinhauer () investigated the lexical-semantic integration of confusable Italian noun pairs such as cappello/cappella (hat/chapel) in both attriters and L learners as compared to monolingual native speakers of Italian. Phonologically, these semantically unrelated word pairs differ only in their word-final vowel (–o versus –a, also marking their GG) and are extremely difficult for learners of Italian L, as indicated by many cautionary notes in Italian textbooks (e.g. Fogliato & Testa, , p. ) and online language blogs. In a German ERP study, Rüschemeyer et al. () had already shown that, unlike native speakers, L learners tended to confuse similar German word pairs that differed word-initially such as Maus (mouse) and Haus (house), and also displayed enhanced negativities starting in the N time range in a priming paradigm (Maus-Dach; mouseroof) that were interpreted as reflecting problems inhibiting the phonological neighbours (mouse and house). Kasparian & Steinhauer’s Italian ERP study presented three different sentence conditions, in which a given context was either completed with the appropriate target noun from the word pair (Correct), with the confusable other noun from the word pair (Swap), or with a completely unrelated noun (Mismatch): Correct

Per coprire la testa, il pescatore porta il cappello di lana. To cover his head, the fisherman wears the hat of wool.

Swap

Per coprire la testa, il pescatore porta la *cappella di lana. To cover his head, the fisherman wears the *chapel of wool.

Mismatch

Per coprire la testa, il pescatore porta la *cartella di lana. To cover his head, the fisherman wears the *briefcase of wool.

One aim was to determine whether sentences with ‘confusable nouns’ (Swap condition) would be processed as semantically acceptable by late L learners and L attriters, especially by those individuals with lower Italian proficiency scores. As hypothesized, monolingual Italians in Italy showed large N effects for both Mismatch and Swap conditions (Figure .). This pattern was also found for those attriters and L learners who had a high proficiency (HP) level in Italian (as assessed by various written, oral, and comprehension tasks, see Tables  and  in Kasparian & Steinhauer, ). In contrast, participants with lower Italian proficiency scores elicited reliable Ns only for the Mismatch but not for the Swap condition, irrespective of whether Italian was their L (attriters) or their L (learners). This finding suggests that online confusability of phonologically similar words (as reflected by N amplitudes) is primarily driven by

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

     Controls –4.0

HP Attriters

LP Attriters

HP L2 learners

LP L2 learners

–4.0

–4.0

–2.0

0.2 0.4 0.6 0.8 1.0 1.2 –2.0

4.0

4.0

4.0

4.0

4.0

–4.0

–4.0

–4.0

–4.0

–4.0

–2.0

0.2 0.4 0.6 0.8 1.0 1.2 –2.0

0.2 0.4 0.6 0.8 1.0 1.2 –0.2

4.0

4.0

4.0

–2.0

–2.0 4.0

0.2 0.4 0.6 0.8 1.0 1.2

0.2 0.4 0.6 0.8 1.0 1.2

–2.0 4.0

0.2 0.4 0.6 0.8 1.0 1.2

–4.0

0.2 0.4 0.6 0.8 1.0 1.2 –2.0



–4.0

0.2 0.4 0.6 0.8 1.0 1.2 –2.0

0.2 0.4 0.6 0.8 1.0 1.2

Mismatch Swap Correct

0.2 0.4 0.6 0.8 1.0 1.2

 .. ERPs for confusable words, time-locked to target noun onset (at  ms) Responses (at electrode Pz) to Swap (blue) relative to Correct (green) sentences (upper row) as well as Mismatch (red) relative to Correct (green) sentences are compared between native controls and high/low proficiency subgroups of Attriters and L learners. Proficiency modulates the amplitude of the N for the Swap condition across Attriters and L learners. The N in the Mismatch condition appears is less modulated by Proficiency level than Swap violations. The P in both violation conditions is largest in Attriters. Time ranges (in milliseconds) depicted on the x-axis are relative to the noun onset ( ms). Negative values are plotted up.

proficiency, not by AoA. However, a second ERP effect suggested that the group of attriters as a whole employed a processing strategy that was different from both monolingual native speakers and L learners. That is, for both Mismatch and Swap trials, attriters also elicited a large, broadly distributed P component between  and , ms, whereas monolingual controls only showed very small posterior Ps, and L learners did not display any P effects. This late positivity in attriters was reminiscent of a similar effect reported by Lazlo & Federmeier () in response to sentence-final nouns that were semantically incongruent but orthographic neighbours of the expected target word. This suggests that attriters, unlike the other two groups, may have ‘doublechecked’ the similarity between the actually presented word and a likely target word, resulting in ‘second thoughts’ in the Swap condition, compatible with the ‘monitoring’ interpretation of the P (e.g., Van de Meerendonk et al., ). Thus, while lowproficiency attriters may have initially missed the inappropriateness of a noun in Swap trials (like low-proficiency Lers), they seemed to have noticed it with a slight delay. This would also explain why attriters’ offline judgements were more similar to monolinguals than to L learners. Overall, these data provide first ERP evidence for lexical-semantic attrition effects and suggest that, despite similarities between attriters and L learners, the former may be able to rely on additional knowledge not accessible to the latter group.

.. Lexical processing in attriters’ L Attriters are typically assessed in their L and compared either to monolingual speakers or, less often, to learners who are acquiring the same language as an L. To date, only two ERP studies have examined attriters’ lexical-semantic (Kasparian et al., under revision) and morpho-syntactic processing (Kasparian & Steinhauer, unpublished) in their predominantly used L, while comparing them to native monolingual and bilingual speakers processing the same language as their L. The question of interest

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

 (A)

   N400 difference wave: Italian homograph-English homograph

(B)

4.0

–0.2

N400 difference wave: English correct-Cognate correct –4.0

0.2 0.4 0.6 0.8 1.0 1.2

–4.0

English-Italian learners Italian-English attriters English controls

N400 reduced for IH in Attriters

Controls

Attriters IT Learners

–0.2 4.0

0.2 0.4 0.6 0.8 1.0 1.2

N400 reduced for CC in Attriters

Controls

Attriters

IT Learners

 .. ERP responses to homographs and cognates (A). A comparison of the difference waves for the N effect elicited by IH violations relative to correct EH sentences, for all three groups. Italian-English Attriters (red) showed a reduced N in response to Italian homograph (IH) sentences that are semantically anomalous in English but not in Italian. The two native-English groups showed a robust N response for IH sentences. (B). A comparison of the difference waves for the N effect elicited by English correct sentences relative to Cognate correct sentences (CC). In line with cognate facilitation effects, the N for CC sentences is reduced in Attriters. Due to this reduction, the English correct (EC) condition elicits a larger N than the CC condition (only in Attriters), as illustrated by the EC-CC difference wave.

is twofold: () from an L processing perspective, do highly proficient and immersed adult L learners resemble native-speakers despite their late age-of-acquisition; and () is L processing in attrition characterized by reduced competition from the L (i.e., greater L inhibition)? As the L ‘takes the lead’ in circumstances of L attrition, it is only logical to examine attriters’ L in addition to their L, in order to study the ERP correlates of this change. Using ERPs, Kasparian et al. (under revision) examined the processing of sentences containing Italian–English ‘false-friend’ homographs (e.g., estate (English: ‘property’ versus Italian: ‘summer’) and matched cognates (music/musica) in Italian-English attriters, English monolinguals and English-Italian late L learners. English sentence contexts either fit with the English meaning of the homograph, the Italian meaning of the homograph, the shared meaning of the cognate, or were semantically implausible in either language. The aim was to assess whether the Italian lexicon of attriters and L learners was automatically co-activated while reading in English, and whether factors such as proficiency, language use and/or length of residence modulated this co-activation. In line with language non-selective models of bilingual lexical access (Kroll & Stewart, ; Dijkstra & Van Heuven, ), both bilingual groups showed co-activation of Italian during English-processing not only in the N time-window but also on the P further downstream. Sentences that fit the Italian homograph meaning were perceived as less of a lexical-semantic violation (reduced N), particularly for Italian-English attriters (Figure . shows difference waves for simplicity). A cognate effect was found for both bilingual groups, though in different processing-stages; while Italian attriters showed reduced N and P effects (i.e., facilitation) in English for correct cognate conditions, L-English Italian learners only showed this facilitation effect in the P time-window. Instead, they showed a tendency

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

    



toward cognate interference effects in the N window. In both bilingual groups, homograph-targets elicited larger posterior Ps than cognate target words. This effect was positively correlated with bilinguals’ performance on a translation (production) task containing different false-friend items, suggesting that the P in ‘better inhibitors’ reflected a conflict-monitoring process (Van de Meerendonk et al., ). These realtime ERP effects were not directly mirrored in end-of-sentence acceptability judgements. Importantly, L co-activation was weaker (and L processing more native-like) with higher English proficiency, longer length of residence, and reduced Italian exposure. These novel findings suggest that L attrition may be partly conceptualized as a reduction of L activation during real-time L processing as a result of strengthening L formmeaning links in contexts where attriters must continuously inhibit the L in the L environment (Linck et al., ; Whitford & Titone, ).

. S  , ,   

.................................................................................................................................. We believe the first ERP studies on L attrition have already generated very exciting data. In the domain of lexical-semantic processing, we saw that Italian immigrants in Montreal showed ERP profiles that differed from both L learners and monolingual controls in their L (Section ..). Whereas the N effect to confusable words was primarily driven by proficiency (across all groups), the finding of an additional P in attriters pointed to a different processing patterns in this group only, most likely linked to subsequent checks of knowledge not available to L learners. The less automatic activation of L lexical representations in attriters also seemed to affect their L processing of English words, where less exposure to Italian resulted in less competition (Section ..). Perhaps the most interesting results come from morpho-syntactic studies. For agreement violations, the native-like GG data from Monika Schmid’s group suggested no ERP attrition effects at all, whereas the Montreal data on number agreement indicated consistent group differences between attriters and monolingual controls for various ERP components at both verb and modifier positions. In our opinion, these apparently inconsistent findings should be taken as a warning not to overgeneralize any data from individual studies, but to be prepared for a more complex pattern influenced by various experimental and psychometric factors. That is, the research question of ‘whether there exists L attrition for morpho-syntax’ may not lead to a simple ‘yes’ or ‘no’ but rather require more differentiation regarding structures, language modality, L–L pairings, and so on. Starting this research endeavour with apparent inconsistencies in ERP data may be beneficial in preventing us from adopting oversimplified answers, as was the case for ERP research on late L learners in the early s, when all data seemed to confirm the critical period hypothesis. Recent research on both L acquisition and L attrition suggests that there is at least some degree of post-puberty brain plasticity that can affect morpho-syntactic real-time processing. For L attriters, perhaps the best example is the data for (grammatical) Italian relative clauses that Italian immigrants in Montreal (but not monolingual Italians)

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi



  

perceived as ungrammatical (Section ..). However, only data from different L–L pairings can tell to what extent the results were caused by English grammar influence, and this is only one of many open questions. Future research needs to question initial interpretations and systematically test alternative accounts. All ERP studies presented in this chapter have their shortcomings; a common one across experiments was that the groups of L learners and attriters were not perfectly matched in terms of proficiency. Attriters were typically in the proficiency range of the monolingual controls, whereas L learners had lower scores for at least certain proficiency measures (which in itself is sometimes overinterpreted as evidence for critical periods). The main reason for this asymmetry is that recruiting large and homogenous groups of ‘strong’ attriters and native-like L learners is a non-trivial (and very time-consuming) task that, moreover, usually requires the involvement of multiple labs in two or more countries. Whether some group differences (such as the finding of a P for confusable words in attriters only) were partly due to (undesirable) matching problems of this kind needs attention in future studies. Both research teams in Groningen/Essex and Montreal/Italy have tried to address some of these issues by running additional correlations and regressions that use measures of proficiency, length of L exposure, etc., as continuous predictor variables for ERP patterns. The use of generalized additive mixed models (or GAMMs; Meulman, Wieling, et al., ; Nickels & Steinhauer, , in preparation) that can model non-linear relationships between continuous predictor and predicted variables promises to offer even more sophisticated statistical approaches to deal with these challenges. A related question is whether differences or limitations in experimental design may explain some of the apparent inconsistencies across studies on agreement violations (i.e., an attrition effect for number disagreement but not for GG violations). Although in both Bergmann, Meulman, et al.’s () GG study and Kasparian et al.’s () number agreement study, attriters were quite well matched with the control group, the different outcomes may be influenced by the fact that Kasparian et al. only recruited attriters who explicitly reported problems with their L. That is, only a subset of immigrants whose daily exposure to their L is very limited and who reported experiencing problems in their L may display clear attrition effects in ERPs. Another factor contributing to different findings may have to do with the slightly more complex (and longer) sentences in the number agreement study (which were presented intermixed with many other structures). Finally, Kasparian and colleagues employed a somewhat more fine-grained ERP analysis in multiple P time windows, whereas Bergmann, Meulman, et al. focused on one large P time window ( to , ms). Future research will have to clarify which of these factors are the most relevant ones. For the absence of any GG violation ERP effects in late L learners (Meulman, Stowe, et al., ), however, their low performance during the online judgement task may indeed have played a role, because only a relatively small number of trials may have contributed to actual ERP differences between the conditions.3 3

The L2 learners’ accuracy for GG conditions was 62.8% (compared to 92.2% in the control group), with a chance level of 50%. In a ‘worst-case scenario’, only about 26% of trials in each condition may have reflected L2 learners’ correct judgements, while up to 74% of trials did not contribute to ERP effects between grammatical sentences and GG violations. The much higher scores in the offline gender assignment task did not turn out as a good predictor for online processing during the ERP study, which is a common issue with proficiency measures that tap different types of knowledge or cognitive

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

    



Other open questions have to do with determining to what extent L attrition is due to (a) the reduced exposure to one’s L, or (b) the influence of L on L in terms of coactivation, competition, transfer, and L inhibition. As reduced L exposure almost always takes place in the context of predominant L exposure and active L use, teasing apart these different aspects will be challenging. The exact L-L pairing likely also plays a role. For example, Bergmann et al.’s () attrition study used English as the L, which however does not have a grammatical gender system that could compete with the German one. The resulting lack of competition between L (English) and L (German) may be a main reason for the the absence of attrition effects in their study. Similar to comparing ERP profiles of L and L during different stages of language acquisition, ERP research contrasting L and L attrition may help us better understand shared and distinct trajectories of losing a language, including the role of changes in brain plasticity over the life span. For the time being, we believe the present ERP studies show that anecdotal reports and ‘feelings’ of change in L can be substantiated with real-time processing patterns. Will ERP attrition effects be a rare exception, or rather the rule? We expect that the truth will be somewhere in the middle, but it will be the specific details that define this ‘middle’ that count and will advance the field.

processes (for discussion, see Steinhauer et al., ; White et al., ). In these situations, responsecontingent data analyses may reveal significant ERP effects for a ‘subjective’ (rather than objective) online categorization of grammar violations, as has been prominently demonstrated in a recent ERP study by Lemhöfer et al. () who initially also did not find any significant ERP effects in L learners, interestingly also for GG violations in Dutch. Note, however, that the logic of relying on more detailed behavioural data is not universally applicable. In some studies, reliable ERP effects may occur long before behavioural data show corresponding effects (e.g., McLaughlin et al., ).

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

  ......................................................................................................................

        ......................................................................................................................

 ,  ,   . 

. L :  

.................................................................................................................................. F language (L) attrition is the temporary or permanent non-pathological loss of language ability in a native language, such that language performance is not what would be considered to be typical performance in native/monolingual speakers of a similar age and proficiency. Language attrition is a field of research at the interface of linguistics, psycholinguistics, and neurolinguistics. Historically, neurolinguists have investigated language loss in the face of pathology, such as aphasia, in monolingual, bilingual, and multilingual individuals. In the last two decades, research on non-pathological language loss has been increasing, with the realization that L language attrition is a highly complex, multifaceted phenomenon, influenced by a number of linguistic, socio-linguistic, cognitive, and neural variables, thus sparking a number of studies that have capitalized on different methodologies (see Köpke, ; Köpke & Schmid, a; Köpke & Keijzer, Chapter , this volume). The overwhelming majority of L attrition studies have focused on behavioural paradigms that have tackled language change and loss in adult speakers who fully acquired their L before emigrating to a country where they were consistently exposed to, and used a second language (L) (Weltens et al., ; de Bot, ; for a review, see Köpke, ) with the goal of advancing the discussion on the nature of language representation and processing (see Schmid & Köpke, , for a detailed description of the various theoretical proposals). It is only very recently that non-pathological L attrition investigations have been informed by neural investigations, with a relatively small number of published articles. The scarcity of neural studies on L attrition may arise from the idea that L attrition has been considered to be an isolated phenomenon rather than a typical facet of

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

    



bilingualism (cf. Schmid & Köpke, ), and perhaps also because of the relative difficulty of recruiting a large enough population of L attriters (but see Keijzer, , for a large behavioural dataset on L Dutch attriters in Canada). To date, there are only a handful of studies that have looked at the neurophysiological signature of L attrition using EventRelated Potentials (ERPs; see Steinhauer & Kasparian, Chapter , this volume), and even fewer studies have investigated the functional and/or structural neural underpinnings of attrition using neuroimaging methodologies such as Magnetic Resonance Imaging (MRI). The goal of this contribution is to review the extant MRI literature on non-pathological L attrition. We will first introduce the neuroimaging methods that will be discussed throughout the chapter, focusing especially on fMRI methodology, and we will then review the most relevant neuroimaging literature on L attrition. Finally, we will illustrate how recent neuroimaging data (both from fMRI and EEG) on changes to the L performed within the context of bilingual language processing more generally could be informative in our understanding of L attrition.

. I M R I 

.................................................................................................................................. MRI is a non-invasive imaging technique that can provide measures of brain function (i.e., fMRI) and brain structure (e.g., anatomy, measures of white matter integrity) with incredibly high spatial resolution (~mm; Brown et al., ; Huettel et al., ). fMRI measures the changes in the ratio of oxygenated to deoxygenated haemoglobin in the blood and thereby can provide an indicator of which brain regions are more engaged during a myriad motor, cognitive, and linguistic tasks. fMRI is considered to be a relative measure. It allows us to see regions that are more engaged during one experimental condition than another, or more engaged during one condition relative to some baseline measure (Huettel et al., ). Because of these issues, task design in fMRI is of critical importance in being able to interpret a study’s results. There are two basic study designs that have been used in functional studies: event-related and blocked designs. Event-related designs allow one stimulus to be paired with one haemodynamic response. This typically involves a random order of presentation of stimuli (across conditions) and a varied inter-stimulus-interval for stimuli to be less predictable, which leads to less pre-planning of responses, and also allows for variability in the haemodynamic response itself, both of which strengthen the experimental design. The limitation of event-related designs is that they often require many trials per condition (at least thirty-five to forty), with jittered intervals, which can lead to a lengthy experimental session (Huettel, ). In contrast, blocked designs involve presenting stimuli from one condition in rapid succession, leading to  to  ‘blocks’ of stimuli, which are alternated with rest periods or blocks of other conditions. This prolonged presentation of stimuli allows for a single, often larger, haemodynamic response, which can mean fewer blocks are needed to reach a meaningful average. A key limitation of blocked designs is that it is not possible to tease apart the fMRI signal for any one stimulus as the signal is linked to the entire block (Maus et al., ).

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



, ,  

In addition to functional imaging, MRI can be used to investigate structural changes in grey and white matter. Standard anatomical images (e.g., T- and/or T-weighted imaging) can provide volumetric measures of grey matter, white matter, sub-cortical structures, and fluids, such as cerebrospinal fluid (CSF). Recent advances have allowed for more sophisticated structural assessments such as measures of cortical thickness and gyrification (Koelkebeck et al., ; Gautam et al., ). These methods can also be used to quantify cortical decline, such as the volume of white matter hyperintensities, which make them invaluable tools with which to assess change, as for example in structural changes mirroring language change after immersion, or language change through attrition. To date, there is considerable MRI evidence based on bilingual language use. Structural differences in grey matter density were reported in a seminal study by Mechelli et al. (), who demonstrated that bilinguals had greater grey matter density in the left inferior parietal cortex relative to monolinguals. For more comprehensive reviews on the effects of bilingualism on grey matter volume, see Stein et al. () and Li et al. (). Finally, Diffusion Tensor Imaging (DTI) is a methodology that examines the anisotropy (i.e., directionality) of water flow in white matter fibres, and has been used as an indicator of white matter integrity (Madden et al., ; Tournier et al., ). Although there have been no structural MRI (DTI specifically) studies investigating L attrition, research suggests that bilingualism is associated with higher white matter integrity (e.g., Luk et al., ; Mohades, Struys, et al., ; Garcia-Penton et al., ; but see also Gold et al., ) even for those who acquired their L relatively late in life (Pliatsikas et al., ), and even at intermediate levels of L learning (e.g., Mohades, Struys, et al., ; Mohades, Van Schuerbeek, et al., ; Rossi et al., ). Overall, studies on structural changes in grey and white matter connectivity due to second language learning or bilingualism have demonstrated a very high level of neural plasticity. Even though no studies to date have investigated changes in white matter connectivity as a consequence of L attrition, it would be possible to hypothesize that decrease in grey and white matter density, and changes in measures of white matter connectivity should be observed as a consequence of language loss. This will be an empirical question to be answered by future research in L attrition, but is also one that can greatly aid a deeper understanding of what attrition is exactly.

. M R I      L 

.................................................................................................................................. The number of neuroimaging studies that have specifically investigated L attrition is very low. However, in a seminal series of studies designed to ask the question of whether longlasting neural traces of the L can be found after years of non-use, Pallier and colleagues (Pallier et al., ; Ventureyra et al., ; Pallier, ) have investigated L language processing in a number of attriters with the main goal to answer the question of whether those speakers had any detectable neural activation related to their L. Participants were Korean children who had been adopted by Francophone families at a relatively early age and who were exposed to French only after adoption (details of these studies are reported

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

    



by Pierce et al., Chapter , this volume). Importantly, after being adopted, Pallier’s () participants were completely isolated from Korean. This is similar to what has been reported in another L attrition study that described the displacement of German Jewish children during World War II to English-speaking foster families at an early age ( to  years, Schmid, ). In other words, there was complete L cessation at a relatively early age. Crucially, these studies probe the Critical Period Hypothesis for language learning (Flege et al., ), which predicts that long-lasting traces of the L should be detected even after years of non-use. To measure possible residual knowledge of Korean, Pallier et al.’s () participants completed two tasks: a word recognition task in which they listened to a French word and decided which of two possible Korean words was the correct translation, and a language identification task in which they listened to sentences presented in five languages (including Korean) and rated the likelihood that a specific sentence was spoken in Korean. Additionally, to investigate the residual neural sensitivity to Korean, an event-related fMRI study was conducted while participants listened to French, Korean, Japanese, and Polish sentences and performed a fragment detection task. Results from the adoptees were compared to a French monolingual control group. Consistent with the personal reports from the adoptees group who reported no conscious recollection of Korean, behavioural results revealed that Korean adoptees did not show any sensitivity to Korean in any of the behavioural tasks. Overall, no difference in activation was detected between the two groups during the Korean listening condition. Results further showed that listening to French sentences yielded greater activation in a network of areas encompassing left Superior Temporal Sulcus (STS), left Superior and Middle Temporal gyri, left Inferior Frontal gyrus, and the contralateral areas in the right hemisphere in both the adoptees and French control group. Crucially, the network of activation did not differ between the Korean adoptees and the French monolingual group. The only difference between the two groups was observed in the extent of the network while listening to French. The Korean adoptees activated a significantly smaller network than the monolingual French controls, which the authors interpreted as either a residual effect of their L or, alternatively, a result of language variability in their French due to being recruited from different parts of France, while the French control group was recruited from the Parisian metropolitan area only. Importantly, all participants (native French speakers and Korean adoptees) were equally fluent in French. Pallier et al. () concluded that their data do not support a strong version of the Critical Period Hypothesis for L learning (e.g., Flege et al., ), but rather provide evidence in favour of the plasticity associated with language acquisition in the first few years of life from the native to an L. Crucially, the authors interpreted their results as evidence that sensitivity (even at the neural level) to the native language can be diminished if not erased by ceasing to speak or listen to the native language, and by acquiring an L. Contrary to these findings, effects of the native (forgotten) language on the newly acquired L were reported by Rajagopal et al. (), who found differences in language lateralization between monolingual English speakers, and a group of children who were adopted from China or Eastern Europe during a verb generation verbal fluency task in English. Similar to Pallier et al.’s () study, Pierce et al. () used fMRI to study the behavioural and neural processing of Chinese in twenty-three adoptees from China, who were exposed exclusively to French after being adopted (mean age of adoption,  year and  months), and reported having no conscious recollection of Chinese. Results were

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



, ,  

compared to a group of native Chinese speakers (n = ) who learned French as a second language at a very early age (mean age onset learning French:  months), and were therefore classified as early Chinese-French bilinguals, and also to a group of French monolingual children who were never exposed to Chinese (n = ). During the study, participants listened to pairs of three-syllable phrases in Mandarin that differed only in the last syllable’s tone, and to pairs of non-speech hummed sentences that were modified to only maintain their tonal information, while performing a similarity judgement task. Results revealed that the Chinese adoptees and the Chinese-French bilinguals recruited similar neural regions within the left temporal cortex (specifically, left planum temporale (PT)) that had been previously found to be active in the processing of tones. Additionally, among the adoptees, a conjunction analysis revealed that the amount of activation in left PT was predicted by Age of Acquisition (AoA), with greater activation for those individuals who had more exposure to Chinese. In contrast with Pallier et al.’s data, these results revealed maintenance of neural sensitivity for the lost L even after no exposure or use of that language for many years. This difference in results may arise from the specific features that were studied, in that tone processing may have a similarity to other kinds of processing, such as music, whereas the voiceless stop consonants that were investigated by Pallier et al. are less likely to have overlap with other kinds of auditory processes. Moreover, the Pallier et al. () and Pierce at al. () studies differ along two crucial variables that might account for the differences in the results. First, Pierce et al. examined lexical tone processing, while Pallier and colleagues’ study included a word matching task and a language identification task, thus encompassing phonological, lexical, and higher language level processing. It is therefore possible that the two studies tapped into partly different linguistic processes that pose challenges for a direct comparison. More importantly, the mean age of adoption, thus the age at which the individuals ceased to be exposed to their native language, was also quite different between the two studies (Pallier et al.: . years, range  to  years; Pierce et al.: . years, range  to  months). On one hand, the adoptees tested by Pallier and colleagues had a relatively longer exposure to the L, which in turn could predict greater sensitivity to the L. On the other hand, phonological contrasts are among the earliest linguistic abilities to be learned by infants (Kuhl et al., ), and they also tend to be the features that are more entrenched (Flege et al., ). It is therefore possible that the sensitivity for lexical tone discrimination observed by Pierce and colleagues is characteristic of a more ingrained aspect of a native language, or possibly also an aspect of language that can be transposed to the processing of more general knowledge (e.g., as music). The observed pattern found by Pierce et al. supports the regression hypothesis (for the original formulation see, Jakobson, ; Keijzer, , a; for extensions of this proposal to aphasia, see Bastiaanse & Bol, ; Kolk, ), which posits that linguistic features that are acquired early in childhood tend to be lost last. It is also important to note that the adoptees’ cohort in Pallier et al. () was tested at an older age than the adoptees in Pierce et al. () (mean age at testing: Pallier et al. = .; Pierce et al. = .), making the time lag from the last L exposure and time of testing longer for Pallier et al.’s adoptees (mean time lag =  years) than for Pierce et al.’s (mean time lag =  years). It is possible that overall differences in lag time between the two studies make the results difficult to compare. A possible approach would be to conduct a meta-analysis of both datasets, which could control and account for variability in age since L exposure, and length of immersion in the L.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

    



Overall, the very few fMRI studies that have investigated L attrition (e.g., Pallier et al., ; Rajagopal et al., ; Pierce et al., ) have described neural patterns of individuals who ceased to be in contact with their native language very early during childhood, and were tested as young adults, possibly conflating patterns of L attrition with incomplete L acquisition. To date, there are no fMRI studies on populations of L attriters that fit the more stringent definition of ‘L attrition’ in which the L changes only after complete acquisition. Understanding the neural changes in language processing in speakers who ceased to use their L after complete acquisition is definitely one of the next major steps for the scientific community in this field (see Keijzer, , for a recent discussion on future steps that the literature on attrition could take on neuroimaging). However, an increasing number of studies show L changes in response to L exposure (e.g., Dussias & Sagarra, ) even after relatively short periods of immersion in the L environment (Linck et al., ), or at very early stages of L learning (Bice & Kroll, ), suggesting that the behavioural and neural underpinnings of the L are much more malleable than previously thought.

. N   L    L 

.................................................................................................................................. The relatively few fMRI studies that have investigated L attrition have yielded mixed results. As discussed, variability in the age at which the L exposure stopped, differences in the relative length of immersion in the L, differences in the developmental stage of the individuals tested, and differences in the aspects of languages that were assessed all may contribute to the observed differences across the studies. Undoubtedly, however, the existing studies represent strong evidence for changes that occur to the L as a consequence of a sudden interruption of contact with the L, prolonged lack of use of the L, and/or prolonged immersion in an L environment. These cases, however, represent one extreme tail of a continuum that spans from no L attrition to extreme L attrition. In other words, the current literature on L language attrition represents one particular facet of the phenomenon that could be assumed to be the tail-end of a distribution of change that may occur in the L. We argue that L attrition should be understood as a flexible, continuous adaptation of the language system that is characterized by modulations in the linguistic, cognitive, and neural functioning of the L. Importantly, we propose that the first stages of this adaptive process might start very early during L acquisition, L use, and immersion in the L environment, partly based on the existing evidence that has accumulated so far. Moreover, those early changes might share similar behavioural and neural signatures to more general bilingual language processing. Crucially, neural data (i.e., MRI, fMRI, ERP) will play an increasingly important role in detecting early functional and neural changes in L processing that might not be observable in behavioural measures. The past few years have seen a number of experimental paradigms that have tested the linguistic and cognitive underpinnings of bilingual language processing. Many of those

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



, ,  

paradigms have capitalized on tasks that require participants to switch between their two languages in a number of ways. Under the hypothesis that L attrition is one facet of a more general process encompassing linguistic and cognitive changes to the L, these experimental paradigms could then be re-interpreted as an essential test-bed for measuring early stages of linguistic and cognitive changes that might represent, or be similar to, the earliest stages of L attrition mechanisms. Recent blocked language switching fMRI studies have investigated how speaking the L for a relatively short amount of time dramatically impacts L processing, revealing that the costs related to switching from the L to the L are greater than when switching from the L to the L, supporting the hypothesis that the strongest language needs to be temporarily inhibited to allow production in the L (Green, ). For example, Guo et al. () used a blocked language-switching picture naming paradigm during which participants named blocks of lexical items (using picture cues) alternating between their two languages, or switched languages item by item to determine whether the neural basis for the blocked switching effect was the same as the one engaged by mixed item language naming. Results showed that the dorsal anterior cingulate cortex and supramarginal gyrus were activated to a greater extent during mixed language naming and the dorsolateral prefrontal cortex and parietal cortex were more activated when the L followed the L in blocked naming. Crucially, the regions that were activated during blocked naming are associated with boosting relevant information and switching mental sets, suggesting that even short periods of naming in the L engage a global L inhibitory mechanism (Branzi et al., ; Rossi et al., ). Rossi et al. (in preparation) used fMRI during a blocked language-switching naming paradigm in English-Spanish bilinguals. However, unlike Guo et al. (), the task was designed to test whether inhibition of specific items in the L would spread beyond the item level, to the larger semantic category, and to the whole language level. To that goal, participants named items from different semantic categories in their L first. In a second block, they named pictures only in Spanish (L) that were drawn from three other categories, (e.g., clothing, furniture, kitchen items). Finally, in six subsequent blocks they named pictures in their L (English) again that were mixtures of items drawn from the old items and categories presented in the first two blocks, new items from previously named categories, and new items from categories not previously named in either language. Results provided evidence for significant inhibition in naming in L after an intervening block of L, as revealed by decreased accuracy and longer naming latencies. Moreover, there was activation in the brain areas involved in general cognitive control (Anterior Cingulate Cortex, pre-supplementary motor area, and the basal ganglia) while processing pictures that appeared during the initial blocks, but also while processing novel pictures drawn from the same categories previously seen, and even while processing pictures from novel categories not seen earlier in the experiment. Overall, this suggests that the scope of inhibition during bilingual speech extends more globally beyond the word level and that its time course is relatively long. Overall, blocked naming paradigms have highlighted how the native language is modulated and controlled even in the context of short periods of naming in the L, and have started providing data in support of the idea that temporary language suppression of the L, as a consequence of speaking the L can pervade the whole language system (Rossi et al., in preparation). It is therefore tempting to think that the very same mechanisms might be at play during L attrition, or at least in its earliest stages, especially since language attrition has frequently been referred to as an unbalanced form of bilingualism (Schmid, a).

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

    



Apart from experimental paradigms that have been conducted in laboratory settings, fast and dynamic changes to the native language have been demonstrated to be triggered by immersion in the L environment, a situation that is in itself much closer to the environmental scenarios that have been described as leading to L attrition. Importantly, in line with recent behavioural findings that have demonstrated changes in L processing as a function of a relatively short immersion in an L environment (e.g., see Linck et al., , for verbal fluency data; Dussias & Sagarra, , for parsing preferences), recent fMRI data have revealed that neural activation in bilingual speakers diverges from the neural activation observed in monolingual speakers, even when the task is conducted in the L and in a single language context, suggesting that speaking an L has a strong impact on the neural organization of the L. For example, a recent study by Parker-Jones et al. () tested a group of English monolingual speakers, and a group of highly proficient L speakers of English (who were immersed in the UK). Parker-Jones et al.’s data revealed that bilinguals had increased activation in frontal (Pars Opercularis (POp), Pars Triangularis (PTr), Precentral Gyrus (PrG), and temporal regions (PT and Superior Temporal Gyrus (STG)) relative to monolingual speakers while naming objects or reading words in a single language context. Increased activation in these areas was observed when bilinguals performed the task in the L, and also in the L. Importantly, the areas that were more active in bilinguals than in monolinguals encompassed both language-specific areas, as well as areas that have been shown to be active during interference tasks (POp and PTr), suggesting that processing more than one language does not only have consequence for the L but critically also for the L. Converging results have been found in an fMRI study by Rossi et al. (in preparation) in which English-Spanish bilinguals elicited greater activation in the right caudate and putamen than the English monolingual control group while naming pictures in English, their L, suggesting that bilingual speakers engage areas dedicated to control and language selection, even when they are speaking in their L. Overall, the few fMRI studies on language attrition and the extant research looking at the neural substrates of bilingual language processing and control show that neuroimaging data represent an increasingly important aspect in bilingual language research, including research on L attrition, understood as a specific facet of bilingual language processing. In what follows, we discuss two specific research directions and designs within the neuroimaging domain (i.e., longitudinal designs and neural oscillations) that might be particularly fruitful to advance the research on L attrition.

. F    L  

..................................................................................................................................

.. Longitudinal designs One fundamental aspect of L attrition is the language change and loss due to prolonged immersion, and lack of L use. As such, longitudinal neuroimaging designs that track neural changes over time, while changes in behaviour might not be yet detectable, could become paramount in the next stages of research on L attrition. Previous longitudinal

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



, ,  

neuroimaging studies have demonstrated grey matter changes in language regions such as left inferior frontal gyrus and left anterior temporal lobe (Stein et al., ) and left parietal cortex (Osterhout et al., ) during L learning. These changes have been shown to occur after both naturalistic immersion experiences (Stein et al., ) as well as classroom experiences (Osterhout et al., ; Martensson et al., ). Moreover, these structural changes in grey matter in left frontal (Stein et al., ), left temporal, and right hippocampal regions (Martensson et al., ) were positively correlated with proficiency, suggesting that the observed structural changes may reflect meaningful changes in language. Similar structural changes have been observed for white matter tracts. Schlegel et al. () examined English-speaking learners of Chinese over  months and found increased connectivity in left hemisphere language tracts, as well as their right hemisphere homologues. Interestingly, Hosoda and colleagues () observed changes among Japanese university students learning English in white matter regions connecting right inferior frontal regions and the caudate and temporal regions that were positively correlated with increases in second language proficiency. Their findings suggest that even non-language regions may be sensitive to language-experience-induced changes. Similarly, future studies could capitalize on longitudinal neural designs to track language changes in an L attrition context over time, thus providing fundamental information about the neural and behavioural changes that occur during the long-term process of L attrition.

.. Neural oscillations The use of neuroimaging methods, including EEG/magnetoencephalography (MEG) and fMRI are paramount in examining the rapid neural changes that accompany changes in language use or language processing. While fMRI has great spatial resolution, its temporal resolution is slower than electrophysiological measures. Investigating neural oscillations, recorded with MEG or high-density EEG, allows to combine both fine-grained temporal and spatial resolution. Such measures could help advance the study of language attrition by addressing the question of whether there are traces of the forgotten L even in the absence of behavioural sensitivity. We will first briefly introduce the techniques and then provide an example of how they can be used to tackle the question of maintenance in L language attrition. Electrophysiological activity of the brain (measured with EEG) and the magnetic fields induced by it (measured with MEG) provide a rich source of data that can be analysed in a number of ways. Traditionally, psycholinguistic research has relied on ERPs—the result of time-locking neural activation (fluctuations in voltage generated by large populations of neurons) recorded by scalp electrodes to the stimulus of interest (e.g., single written/spoken words, words in a sentence, images, sounds, etc.) and averaging across many trials and participants (for a review of this technique and its use in psycholinguistics, see Kutas & Van Petten, ). ERPs are thus a time-series of voltage measurements associated with the event of interest which are usually visualized as a waveform, with time plotted on the horizontal axis, and amplitude plotted on the vertical axis. In this way, the time course of activation evoked by the stimulus can be tracked with high temporal precision. Differences in the amplitude and latency of waveform peaks associated with experimental manipulations indicate differences in the underlying neural mechanisms employed to process stimuli

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

    



in different conditions (see Steinhauer & Kasparian, Chapter , this volume, for a more elaborate discussion of ERP methodology). However, information about induced activity associated with the critical manipulation, but not necessarily time-locked to it, is lost after averaging. Decomposing the recorded signal into different frequencies, investigating local synchrony (the amount of power in each frequency band), and long-range synchrony (oscillatory phase relation measures between different sources/channels) adds a frequency dimension to the electrophysiological signal (in addition to time and amplitude). Thus, frequency analyses might help capture the dynamics of purely linguistic processing (syntactic and semantic analysis) as well as domain-general cognitive processes (memory encoding and retrieval, attentional mechanisms, spatial and temporal neural binding) associated with language comprehension and production. Crucially, the study of oscillatory brain responses allows one to tackle different subcomponents of complex cognitive processes that can be simultaneously implemented by means of synchronization/desynchronization of neuronal activity at certain frequencies, times, or locations. The same brain structures can be recruited simultaneously for different cognitive processes due to their ability to operate at multiple temporal scales. Neuroimaging measures such as ERPs and the BOLD (blood-oxygen-level dependent) signal used in fMRI studies are not specifically geared towards disentangling different synchronic cognitive processes, while this is possible when looking at neural oscillations. Traditionally, brain oscillations are subdivided into five frequency bands: delta ( to  Hz), theta ( to  Hz), alpha ( to  Hz), beta ( to  Hz), and gamma (> Hz). Here, we will focus on memory and language related processes. Memory research has related synchronization in theta and gamma bands as well as desynchronization in alpha and beta bands to episodic memory encoding and retrieval (Hanslmayr et al., ). Language research has related theta activity to lexical retrieval, alpha and beta activity to semantic and syntactic unification, and gamma activity to prediction during sentence processing (Lam et al., ). Since its discovery, alpha activity has been repeatedly found to correlate with attentional demands (for a review, see Klimesch, ). Gamma oscillations are thought to be a binding mechanism that integrates information from distributed groups of neurons (Colgin et al., ). For a more detailed review of how brain oscillations relate to dynamic cognitive processes, see Ward (). High-density EEG allows for decent spatial resolution as long as ‘anatomically precise forward models of subject-specific brain and skull anatomy’ (Cohen, , p. ) are used and spatial filtering techniques are applied. Spatial resolution of fMRI still remains the best; however, because of the slower haemodynamics, these measures result in ‘a rather static picture, emphasizing mainly the structural aspects of the Brain and Language network’ (Bastiaansen & Hagoort, , p. ). Linguistic processing, like any other cognitive operation, is very rapid and characterized by the dynamic formation of ‘transient functional language centers’ (Weiss & Mueller, ), thus methods with precise temporal and spatial resolution are required to track these centres throughout space and time as linguistic input unfolds. In what follows, we will illustrate how the use of neural oscillations could be applied to the study of language attrition and effects of re-exposure, by using examples from work on word learning. Previous work on word learning has shown that the analysis of brain oscillations is sensitive to the degree of lexicalization of novel and existing words. For example, Bakker et al. () compared neural representations of words at different stages of lexicalization:

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



, ,  

untrained existing words, untrained novel words, words learned immediately before testing, and words learned twenty-four hours before testing. Words that have wellgrounded representations have been shown to elicit theta and gamma power increases and beta and alpha power decreases. Bakker et al. () found that novel words seen by participants for the first time were characterized by a smaller theta sychronization compared to existing words. Words trained before the experiment elicited a theta response marginally lower than the one observed for the existing words. Words learned twenty-four hours before the test elicited a response identical to that of existing words, suggesting memory consolidation of the learned information. Beta desynchronization was characterized by a similar pattern: existing and consolidated words elicited identical neural responses; however, unfamiliar and recently learned words elicited a significantly smaller power decrease in beta band than existing words. Converging evidence has been reported in the fMRI literature suggesting that the brain differentially represents words learned at different points across the lifespan. For example, Fiebach et al. () have shown that the processing of words learned earlier in life recruits the precuneus (a brain structure important for episodic memory retrieval), while late learned words activate lateral inferior frontal areas (associated with lexical access and linguistic processing in general). Overall, recent work on brain oscillations and fMRI illustrates the effects of AoA on lexical processing suggesting that the brain responds to linguistic structures differentially depending on when they were acquired during development. The analysis of lexical items that were either forgotten or reinforced later in life (after being forgotten) could be a fruitful further direction for L attrition research. For example, by looking at theta and beta power induced by words from the forgotten language that were supposedly acquired by the attriters during childhood one might be able to see whether—based on what has been discussed above—they would elicit more theta increase and more beta decrease than novel unfamiliar pseudowords learned immediately before testing. Employing MEG and highdensity EEG would also provide information about the sources of these effects.

. S  

.................................................................................................................................. The goal of this contribution was to review the current neuroimaging (fMRI) literature on first language attrition, and to discuss how current fMRI research informs theories of bilingual language representation and control. Overall, the few existing fMRI studies of L language attrition (i.e., Pallier et al., ; Rajagopal et al., ; Pierce et al., ) have only looked at neural patterns of sensitivity to the native language in individuals who ceased to be in contact with their native language (very) early during childhood. The results of those few studies are mixed, with some suggesting that the early adoptees do not maintain any neural or behavioural sensitivity to the native language (Pallier et al., ), while others reveal traces of neural sensitivity to the forgotten language despite the lack of behavioural sensitivity (Pierce at al., ). It was also noted that, in contrast to the few studies on early adoptees, there are no fMRI studies for speakers for whom attrition processes started only after complete acquisition of the L. Despite the still high financial costs related to utilizing MRI, this methodology has become increasingly available in labs around the world. It would therefore be important to embark on studying how the neural representation

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

    



and function of the native language changes for ‘late’ attriters. This research direction seems particularly important, because MRI-related techniques could highlight changes in language representation and processing even when there are no appreciable changes in behavioural measures. We also proposed that two fruitful approaches for future studies on L attrition, specifically in the neuroimaging domain, would be to utilize longitudinal designs and to study brain oscillatory signals. Longitudinal designs allow studying functional and structural changes in response to language change while attrition takes place, and even in the absence of behavioural changes. The study of oscillatory brain signals across different frequencies will help disentangle the dynamics of linguistic changes due to L attrition, as well as domain-general cognitive processes such as memory, and attentional mechanisms, which are associated with language comprehension and production. Moreover, oscillatory studies are feasible to incorporate with behavioural, longitudinal, or cross-sectional designs. Most importantly, we discussed recent fMRI literature that has highlighted how learning, speaking, and processing an L, even at very early stages of exposure and immersion, have important functional and structural consequences for the representation and control of the L (e.g., Guo et al., ; Parker-Jones et al., ), suggesting that changes in neural representation and control of the native language are a normal facet of bilingualism, and not limited to attrition processes. We therefore argue that the phenomenon of L attrition should be understood as a flexible, continuous adaptation of the language system that is characterized by modulations in the linguistic, cognitive, and neural functioning of the L in the face of bilingualism. Importantly, the first stages of the observed adaptive neural processes that start very early during L acquisition, and immersion in the L environment might be the very same ones that are at play in the earliest stages of native language attrition.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

  ........................................................................................................................

LINGUISTIC FACTORS IN LANGUAGE ATTRITION ........................................................................................................................ Edited by

 .     

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

  ......................................................................................................................

             ......................................................................................................................

 .     

. I

.................................................................................................................................. T chapters in the present section provide an overview of research concerning the linguistic properties of attrited languages—that is, of those features which typically make up the dependent variables in investigations of attrition. As such, they represent what is probably the oldest interest of this research field: from the earliest stages of attrition research, studies have attempted to provide not only descriptive accounts of how linguistic properties may change under specific circumstances of attrition (e.g., based on their similarity or difference across the languages in question) but also to integrate these findings into theoretical models of language contact and bilingual development. It has often been pointed out that the recognition that a mature speaker’s language can change within this person’s lifetime is a fairly recent one. While language change across generations and across centuries has been a key preoccupation of linguistic study for centuries (most notably represented by the nineteenth century Neogrammarian approach) and is thus at the root of virtually all modern day linguistic knowledge and theory, the general assumption has always been that, once a native speaker reaches maturity, their first language (L) becomes stable and, hence, uninteresting—and this view prevails to this day (e.g., Gregg, ). The first linguistic researcher to challenge this assumption was the great Einar Haugen who, in his essay in ‘Language and Immigration’ (), not only laid the foundations for an entire research field but also coined the term attrition to describe it (p. ). Berating ‘both lay and learned’ for regarding the phenomenon of ‘the dialect of the immigrant’ with ‘scorn or amusement’, and in particular the linguists for considering this phenomenon ‘as beneath their dignity’ (p. ), Haugen points out to what extent our understanding of the host of immigrant languages spoken in America falls short of capturing the overall linguistic experience and reality: ‘No general perspective has been reached, and few general conclusions drawn’ (p. ). He sets out to catalogue his own observations from extensive

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi



   

investigations of American Norwegian, in the hope ‘that students of immigration may find here some intimations of what can be derived from continued researches into the problems touched upon’ (p. ). Focusing mainly on the domain of vocabulary, Haugen identifies phenomena such as a decrease in accessibility of low-frequency L lexical items, which are being replaced by their L equivalents, the loss of lexical distinction made in the L but not the second language (L) (e.g., separate terms for maternal and paternal relatives), and the cross-linguistic blend of forms and meanings. Haugen also acknowledges that English influence on Norwegian is more readily seen on lexical than on grammatical/ functional items. Haugen describes with insight and detail different processes that combine to bring about the phenomenon of attrition, such as dialect levelling among the migrant communities, the social dynamics within these communities, the decrease of Norwegian literacy across generations, and difficulties associated with L acquisition. His analyses of how and why lexical mergers have occurred include a wide range of background phenomena, from the social makeup of the immigrant communities to the influence of characteristics of the new environment, such as its technology, its geography, and its climate. Crucially, he points out that the aggregate influence of English on Norwegian, which he calls a ‘psychic compulsion’, is at work ‘from the day of immigration’ (p. ). Despite Haugen’s early work, and the efforts of others to draw attention to the fact that cross-linguistic influence in bilingualism is not a unidirectional phenomenon (e.g., Weinreich, ), it took several more decades for linguistic research to develop an interest in the phenomenon that went beyond the qualitative and the anecdotal. One of the earliest attempts to provide a set of predictions for language attrition as a process of linguistic change is Roger Andersen’s () ‘blueprint for language attrition research’. Andersen proceeds from the assumption that, under conditions of inadequate input and interaction in a given language, not all lexical, phonological, morphological, and syntactic distinctions can be maintained, but that deterioration will be selective (p. ). Drawing on evidence from research on language contact and language acquisition, the author sets out a number of specific and testable predictions for potential changes through which language attrition may affect different linguistic levels. He invokes factors such as frequency, cross-linguistic similarity, and the functional nature/contribution to meaning of certain categories or features as important drivers of the attritional process. While many of these predictions can still be considered relevant today, thirty-five years of research into the linguistic properties of language attrition have refined our understanding of the contribution of these factors but also established further factors and attempted to account for their findings based on a variety of theoretical frameworks (see also the contributions in Part I of this volume). Following on from this and other early attempts to kick-start language attrition research, empirical findings initially were quite slow to emerge, as much of the early work done on language attrition fell into one of three categories: (a) theoretical/hypothetical work; (b) qualitative work or small-scale pilot studies; and (c) initial outlines of larger projects (which, unfortunately, rarely saw eventual completion) (Köpke & Schmid, ). It was not until around the start of the new millennium that a solid evidence base began to emerge. Köpke & Schmid () review the evidence available at this stage, with a view to moving the field forward, and suggest a number of theoretical and methodological considerations, arguing for a stronger integration of purely linguistic observations of language attrition with, on the one hand, extralinguistic factors such as language exposure and attitudes (see Part IV of this

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

    



volume), and theoretical approaches to the development of the representation and use of linguistic knowledge (see Parts I and II of this volume) on the other. Probably the most important and most controversial issue in this context is the question of what kinds of changes to linguistic features actually constitute language attrition. Investigations of effects of the second language on the first often attempt to distinguish ‘superficial’ from ‘structural’ changes—in earlier treatments, this is framed in terms of the competence versus performance distinction (e.g., Seliger & Vago, ), while later investigations often apply different but related concepts, such as representation versus processing. In early work, this distinction was assumed to be crucial, with representational changes privileged over processing-related changes in terms of their importance to attrition research. For example, for Seliger & Vago ‘it is erosion that reaches the level of competence that allows for interesting claims about and meaningful insights into the attrition process’ (, p. ). In this vein, as Schmid & Köpke () point out, attrition was originally thought of as a distinct developmental category, experienced by a small subset of bilinguals who had experienced very long-term radical deprivation of input from and exposure to their L, and who had—presumably as a consequence of this deprivation—experienced changes to their L that could clearly be shown to be structural/representational. On this view, the more ‘superficial’ differences in processing the native language, experienced by all bilinguals as a consequence of the co-activation of two language systems, should not properly be considered attrition (e.g., Meisel, ). Schmid & Köpke argue, however, that the reality of attrition research suggests that instances where a representational change can be demonstrated are rare, if not unheard of, among late bilinguals; that the absence of a clear picture of the impact of length of exposure and frequency of use (see Schmid, Chapter , this volume) goes against the notion of a small subpopulation of ‘true’ attriters among the larger population of bilinguals, and that, in contrast to Meisel’s view, all instances of L-to-L transfer should therefore be considered attrition phenomena. To illustrate this point with an example, consider phonetic versus phonological changes in the L of immersed bilinguals: phonetic changes might be considered to be ‘superficial’ and not affect the underlying structure, while a phonological change such as the loss of a phonemic contrast (rather than simply change within a phoneme, as in the case of phonetic attrition), would indeed affect the underlying grammar. According to Meisel’s assessment, presumably, only phonological attrition would qualify as ‘true’ attrition. Interestingly, however, similar changes can yield phonetic (i.e., supposedly superficial) versus phonological (i.e., supposedly structural) outcomes in different populations. This is the case in two investigations of the lateral approximant in German-English (de Leeuw, Mennen, & Scobbie, ) and Albanian-English (de Leeuw et al., ) attriters. Both studies found a darkening of the light lateral approximant in coda position in an adaptation to the English pronunciation. However, while this change in German only affects the pronunciation, not the meaning, in Albanian the contrast between light and dark laterals is phonemic, and so the difference in pronunciation actually changed the meaning of the words. In other words, the same observation of attrition would be interpreted as a superficial one for the German-English bilinguals but as a structural and representational one for the Albanian-English bilinguals. More likely, however, as this example illustrates, is that the effects of L-to-L transfer exist on a continuum where development is gradual though not necessarily linear and

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi



   

where the strength of any measured effect is strongly related to both the task, the linguistic level under observation, and the properties of the relevant feature in both languages. Using the example above, this would mean that any neutralization of a phonological contrast in Albanian laterals would have been preceded by an initial phonetic change (as observed in many of the participants), and that, therefore, the phonetic change in the German lateral may very well also have been a phonological change in the phoneme inventory of the individual German native speakers. It is just that in German, this change was not able to constitute a phonemic neutralization effect—because there is only one lateral in German. As this example shows, it is often very hard to establish at what point—if at all—a change that initially affected online processing—that is, the surface level becomes structural/ representational (nor, indeed whether representational changes must necessarily be preceeded by online ones). In addition, as Schmid & Köpke () point out, and as the chapters in this section illustrate, the vast majority of studies on language attrition deal with phenomena which are unlikely to be indications of a ‘competence change’. They therefore consider all such phenomena part and parcel of one single developmental process— whether one chooses to subsume this process under the label of ‘language attrition’ (as Schmid & Köpke propose), or prefers a more neutral term, such as ‘effects of the second language on the first’ (Cook, ). The point here is that we believe that attrition constitutes both online and representational changes, if a distinction between these two types of change is to be drawn at all. It may be that online changes are necessary before representational changes take place, but we would nevertheless consider these online changes to be a necessary component of the attrition process. Moreover, as in so many cases it is not unequivocally clear whether a change represents a surface level or representational change, this line of debate could be seen as the ‘black box’ of attrition research. The nature of the contents of this black box at present eludes our full understanding and analysis, but that there is something inside it is beyond doubt. The contributions in the present section examine what is inside the black box—the linguistic manifestations of attrition—and we believe that, whether surface or representational, they reveal changes in how bilingual individuals use, process, and potentially represent their native language, and therefore constitute attrition.

. T     

.................................................................................................................................. Like most work in the area of multilingual development, language attrition investigations typically adhere to the classic division of the study of linguistics into different domains, or linguistic levels. The domain of phonetics deals with the production and perception of sounds, while the language specific phoneme inventory is related to the domain of phonology. The domains of morphology and the lexicon deal with the structure and property of words, whilst syntax deals with how words can be combined into sentences and phrases, and the domain of semantics deals with meaning. There are overlaps between these domains and it is possible to break them down into further sub-domains. Nonetheless, traditionally, when examining attrition, research has generally been conducted in relation

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

    



to these general domains, showing that the native language of bilinguals can indeed become affected by another language at all of these levels. These phenomena of language transfer, interference, or attrition can be observed most straightforwardly in the lexicon. From the earliest phases of language attrition research, it has commonly been suggested that this domain may become restructured and that lexical items may become less accessible, leading to the use of underspecified or highly frequent items (Olshtain & Barzilay, ), delays in naming latencies (Ammerlaan, ) and other phenomena of lexical convergence (Pavlenko, ). These phenomena have since been described in a range of quantitative, empirical, and psycholinguistic studies, which have also shown how lexical productivity and sophistication in free speech can be affected by attrition, with a concomitant increase in disfluencies (e.g., Bergmann, Sprenger, & Schmid, ; Schmid & Jarvis, ; Yılmaz & Schmid, ). While these studies have typically focused on changes in immersed populations years or decades into the attrition process, many recent studies also find subtle changes related to lexical access at much earlier stages of L acquisition, including in instructed populations with no experience of L immersion. For example, Gollan et al. () have shown that L learners are not only outperformed by monolinguals on tasks relating to lexical access and retrieval in their L, but that this effect is modulated by word frequency (the ‘weaker links hypothesis’). Similar findings have shown that L word naming in early L learners is affected differently by phonological neighbourhood density and spelling-sound regularities than in monolinguals (Bramer et al. ). Taken together, these findings imply that the pathways of lexical access are affected by the mere presence of a second language. Lexical attrition is therefore not, or at least not only, a process whereby items become unavailable due to underuse (as has often been hypothesized), but the paths and mechanisms which facilitate word retrieval may become less effective due to the co-activation of a second language. The second linguistic domain which became the focus of investigations of language attrition at a relatively early stage is that of grammar. Research in this area was initially mainly situated within the framework of generative grammar, focusing on issues such as the potential reparametrization of a native grammar under the influence of an L and issues of markedness (e.g., Sharwood Smith & Van Buren, ; see also Gürel, Chapter , this volume). With the advent of the Minimalist framework, attention turned to the distinction between core grammar on the one hand (which has often assumed to be invulnerable to attrition processes; but see Iverson, ) and interface phenomena, on the other (see Chamorro & Sorace, Chapter , this volume). Later approaches to the attrition of grammar and morpho-syntax adopted more processing-based models, such as the -M model proposed by Myers-Scotton & Jake (e.g., ) and based on Levelt’s () Speaking Model (see Schmitt, Chapter , this volume) or models based on the parallel activation of languages (see Dussias et al., Chapter , this volume). While thus adopting a variety of explanatory approaches, based on their underlying assumptions about the architecture of the human capacity for language (see also the contributions in Part I of this volume), these studies underscore the original assumption that morpho-syntactic attrition is selective, with different parts of the repertoire being differentially vulnerable (see Schmid & Köpke, ). In addition, two recent studies of morpho-syntactic attrition, using the EEG paradigm (see Steinhauer & Kasparian, Chapter , this volume), seem to suggest that, while attriters may not differ markedly on those EEG components which they share with monolingual controls for violations of certain structures, they may develop additional responses, over and above the ones exhibited by monolinguals,

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi



   

suggesting different overall processing strategies which may be indicative of somewhat more effortful processing among attriters (e.g., Kasparian & Steinhauer, a, find a P among attriters, but not among controls, in their study of confusable words; while Bergmann, Meulman, Stowe, Sprenger, & Schmid, , report that their attriters exhibit an N not shown by the controls, in addition to a native-like P, in response to verb finiteness violations). Investigations of phonetics and, in particular, phonology represent a relatively late arrival within the otherwise well-established field of language attrition studies. With some early exceptions of studies investigating the cross-linguistic adaptation of VOT in immersed bilinguals (Flege, ; Major, ), research in these areas did not become firmly established until the second half of the first decade of the s (see de Leeuw, Chapter , this volume). Even then, studies initially tended to rely on broad and global measurements, such as foreign accent ratings (e.g., de Leeuw et al., ; Hopp & Schmid, ) and only later on turned to more fine-grained segmental and prosodic measurements (e.g., de Leeuw, Mennen, & Scobbie, , ; Mayr et al., ). Recent developments in this field, however, are rapidly closing the knowledge gap between phonetics/phonology and other aspects of language attrition, as is evident in a recent Special Issue of the Journal of Phonetics on ‘Plasticity of native language phonetic and phonological domains in the context of bilingualism’. The findings presented in this collection lend support to the idea that bilinguals, through the very act of being bilingual, undergo bidirectional interaction in their languages which is not delimited by a critical period (de Leeuw & Celata, ). Moreover, it appears that the effects of L acquisition on the native speech of first generation immigrants is magnified in subsequent generations with regard to attrition of both segments (Nodari et al., ) and lexical tones (Kan & Schmid, ). Early exposure might be a necessary condition for the acquisition of certain language-specific speech patterns, but continued extensive exposure to the language is needed in order to maintain speech patterns (Mayr, López-Bueno, & Tomé Lourido, ). Although sustained plasticity of native language phonetic and phonological domains is apparent, variation in the extent of plasticity is also observed. Phonetic and phonological changes in the native language, potentially induced by an L, are suggested to be dependent on factors such as L and L input as well as L proficiency levels, and to proceed in tandem with biological changes across the lifespan, as observed in a longitudinal case study of Stefanie Graf over four decades (de Leeuw, ). The contributions to the special issue also provide evidence confirming previous findings on the rapid cross-linguistic adaptation of some phonemes (Chang, ), and open up a new area of linguistic investigation for attrition studies, namely phonotactic restructuring: a preliminary cue-weighting perception study into Russian expatriates in the United States (Russian-English bilinguals) revealed that these participants relied on vowel duration as a cue to consonant voicing in Russian significantly more than Russian monolinguals and also produced vowel duration distinctions not found in monolingual Russian (Dmitrieva, in review). While vowel duration is an important cue to voicing in English, in monolingual Russian it is virtually absent (given that, in this language, all word-final consonants are devoiced). A further perception study examines phonotactic restructuring in L Brazilian Portuguese (BP) speakers who had been living in the US for almost a year (Cabrelli Amaro et al., ).

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

    



Participants completed a vowel classification task and an ABX task with nonce stimuli as BP speakers tend to perceive and produce an illusory /i/ between illicit consonant clusters. The study finds an increase in accuracy on the perception task (that is, a reduction of illusory vowel perception) in both languages among the L learners of English, suggesting phonotactic restructuring of their native Brazilian Portuguese as a result of exposure to English.

. K     

.................................................................................................................................. The findings summarized above demonstrate that language attrition research has made great strides and gained a large amount of knowledge in the thirty-five years since Lambert and Freed’s seminal publication, The Loss of Language Skills (). However, the insights that have become available remain limited in that, with few exceptions, investigations tend to be confined to: • a single linguistic level • a single language pairing • a single language (i.e., the L—this is mirrored in studies of L acquisition by the neglect of potential changes to the native language). While there are thus solid insights into domain-internal attritional processes, we know very little about how different linguistic domains may be affected by attrition within the same individual, that is, how attrition may unfold as a holistic process across different domains. There is, for example, the possibility that attrition at one linguistic level may have ramifications for others (see de Leeuw, Chapter , this volume), that different language combinations may provide strikingly different outcomes, or that some of the phenomena found in attrition may be relatable to concomitant changes in the L, or vice versa (see Schmid & Yılmaz, ). In addition, there is currently a substantial research gap where linguistic properties outside of these traditional domains of lexicon, phonology, and morpho-syntax are concerned—for example, to date there is little to no research that we are aware of investigating language attrition with respect to pragmatic conventions of politeness and formality (but see, e.g., Montoya-Abat, ; and Passoni et al., ). We thus propose that the future research agenda for language attrition studies should include the following questions: . Are some linguistic domains more stable than others? For example, is pronunciation (phonetics) more susceptible to attrition than grammar (phonology, morpho-syntax), or is it true that the lexicon is the ‘most vulnerable’ linguistic domain, as has often been claimed (see, e.g., Schmid & Jarvis, ; see also de Bot, Concluding remarks, this volume, on the difficulty of making meaningful comparisons between attrition effects at different linguistic levels)? . Is there a ‘domino’ effect in attrition? For instance, does phonetic attrition always precede phonological attrition—that is, would it only be possible to observe

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi



   

phonological attrition when phonetic attrition has already occurred, or only morphosyntactic attrition when lexical attrition has already occurred? . Do predictor variables affect different linguistic domains to various extents? It could be the case, for example, that language use strongly affects the phonetic domain whereas age of L acquisition affects morphology and syntax more strongly. . To what extent are more general processes of cognition interconnected with attritional processes? Recent research has drawn attention to executive function in bilinguals. Are, for example, general cognitive processes of inhibition, activation, and task selection intertwined with the processes of attrition? We believe that these questions will move forward research in the field of attrition more specifically, and also help us to understand the human faculty of language more generally.

. C    

.................................................................................................................................. In Chapter , Chang discusses phonetic drift, which he defines as short-term L-influenced phonetic change in an individual’s L system, attributable to recent L experience. Accordingly, Chang differentiates drift from attrition and proposes that drift may be viewed as the beginning of a trajectory of L-influenced change in the L that, if not reversed, leads either to attrition or to incomplete acquisition. As an example of drift, he points to the studies by Chang () and Guion () in which overall vowel raising, that is a decrease of F frequency, was the main finding, whereas in Mayr et al. () the main finding was a lowering—that is, an increase in F frequency. Chang refers to such vowel space changes as systematic phonetic drift, thereby highlighting the idea that phonetic drift does not occur sporadically. In Chapter , de Leeuw makes a similar distinction between short-term and longer-term phenomena. She considers phonetic attrition to encompass long-term changes in the native language pronunciation of an individual as a result of the acquisition of a new dialect or language acquired post adolescence. This definition focuses on examining plasticity in adult speech—that is, the implication of phonetic drift in child bilinguals would be less than that of phonetic attrition in adults, as similarly argued in de Leeuw, . De Leeuw draws primarily on the Speech Learning Model (Flege, ) and discusses studies which reveal dissimilation and assimilation in segmental and prosodic aspects of speech, as well as in overall changes to an individual’s native accent. She emphasizes findings which reveal interpersonal differences between adult participants (de Leeuw, Schmid, & Mennen, ; de Leeuw, Mennen, & Scobbie , ; de Leeuw, Tusha, & Schmid, ), suggesting that phonetic attrition is not a uniform process and that careful formal speech with a high information content appears to be less likely to exhibit signs of phonetic attrition than more casual speech styles (Shockey, ; Major, ). Such intrapersonal differences according to style and setting might indicate that drift and attrition are interconnected processes which influence each other throughout the lifespan—that is, short-term as well as long-term changes.

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

    



In Chapter , this volume, Celata builds on the previous chapters, postulating that phonological knowledge is dependent on the way speakers experience and use particular phonetic patterns. She defines phonological knowledge as ‘the way humans group different pieces of phonetic knowledge together in a coherent system of relationships that is practical or useful for communication (in the sense of both speech planning and speech decoding)’ and suggests that phonological changes in the form of attrition may include contrast blurring as well as boosting the distinctiveness among system elements as a reaction to contact with a different phonological system. Her chapter delves into both speech production and perception and discusses how attrition in the first generation may impact subsequent generations. She also suggests that socio-stylistic variation should be further investigated in research into phonological attrition, as well as cognate effects in the speech production of bilinguals, heritage language speakers and attriters, to enlarge the scope of available models of cross-language phonological interference. Together, these three chapters thus suggest that, in the phonetic and phonological domain, drift, and attrition are connected, and that changes potentially unfold in the order of phonetic drift > phonetic attrition > phonological attrition. Moving forward, in Chapter , Schmitt defines morphological attrition as changes in inflectional and derivational morphemes and their combinatory possibilities in language production and perception. Interestingly, in her first example she refers to the omission of the genitive feminine suffix in a Russian native speaker, indicating morphological attrition. However, in the same utterance, the speaker implements other morphemes in line with standard Russian grammar, prompting Schmitt to ask why some morphological markers remain resilient to attrition while others are lost more rapidly and easily. Potentially, insight into this question may be gained through examining morphological markers in relation to other linguistic domains, such as their phonological constituents. Schmitt discusses the Markedness Theory, described by Sharwood-Smith () in relation to attrition; the Regression Hypothesis (RH), first formulated by Jakobson (); the Activation Threshold Hypothesis, proposed by Paradis (, ); the Interface Hypothesis (IH; Tsimpli, ; Montrul, ; Sorace, ; see also Chamorro & Sorace, Chapter , this volume); and the -M Model (Myers-Scotton & Jake, ). Through this discussion she demonstrates that there is no single theory that accounts for the complexity and variety of morphological operations and morphemes that undergo restructuring, modification, and ultimately language attrition, whilst pointing to new research which is forging paths (Alexiadou, Anagnostopoulou, & Schäfer, ). In Chapter , Jarvis discusses lexical attrition which he defines as a process that leads to a person’s loss of the ability to retrieve words from memory, adding that lexical attrition is an observable phenomenon which is now widely understood as a collection of undesirable changes to a person’s lexical skills. Jarvis examines attrition in children and in adults in various circumstances, expanding the more restricted definition used by de Leeuw ( and Chapter , this volume). Interestingly, in one particular example, Jarvis mentions that nonce words produced by a Russian adoptee were phonologically compatible with Russian, but that their forms had apparently become distorted in memory (Isurin & Seidel, ). Here again, we see a link between different linguistic domains as the process of attrition unfolds. Moreover, he points out that many studies have confirmed that individuals who have acquired a new language after the acquisition of the first tend to recognize and retrieve L words more slowly than do comparable speakers of the same L whose life circumstances

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi



   

make them less susceptible to attrition (Goral et al., ; Schmid & Jarvis, ; Isurin & Seidel, ). Finally, in Chapter , Gürel discusses attrition research into null and overt pronouns, which span the domains of syntax and pragmatics. In contrast to Chang, Celata, and Jarvis, but in line with de Leeuw (see also de Leeuw, ), she confines her chapter to attrition studies involving late bilinguals, typically residing in an L country as first generation immigrants, who become dominant L users after puberty. Thus, as L grammatical competence is believed to have stabilized in adolescence (see, e.g., Montrul, , a), qualitative changes in the L grammar of post-puberty bilinguals may have far-reaching implications for the alterability of L linguistic competence (Gürel, , a). Gürel emphasizes that although in some null-subject languages there are certain null pronounlicensing morpho-syntactic mechanisms, the use of null and overt pronouns is not random but subject to discourse-pragmatic constraints. For example, a null pronoun in Italian is sometimes preferred if this sentence is uttered as a response to a question, whilst topic shift requires overt pronouns. She draws on the IH (Sorace, ; Chamorro & Sorace, Chapter , this volume) which stipulates that features that are at the interface of syntax and other cognitive-linguistic domains will be susceptible to attrition whilst purely syntactic features would pose no considerable problems. As such, the IH predicts how domains interact with one another in the process of attrition.

. C

.................................................................................................................................. As both the considerations presented in this introductory section and the chapters contained within the section demonstrate, language attrition research has come a long way from Haugen’s mainly qualitative observations made over eighty years ago. In particular, it is becoming ever clearer that changes to the native language in the context of bilingualism are a phenomenon not limited to extreme circumstances of non-use and longterm immersion in an L setting, but that they, to some extent, affect all bilinguals and all linguistic levels, often in ways which are unexpected and surprising. We hope that this summary of the state of the art will help the field progress and make an even larger impact on the mainstream of bilingual research in future.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

  ......................................................................................................................

            ......................................................................................................................

 . 

. I

.................................................................................................................................. T chapter addresses the phonetic changes that occur in one’s native language (L) due to recent experience in another language (L), a phenomenon referred to as  . Before proceeding, it is important to define precisely what I mean by ‘phonetic drift’ in light of various other uses of this term. The earliest published instance of the term ‘phonetic drift’ in the linguistic literature is from Sapir (/), who uses it to discuss contactinduced diachronic sound change: We may suppose that individual variations arising at linguistic borderlands—whether by the unconscious suggestive influence of foreign speech habits or by the actual transfer of foreign sounds into the speech of bilingual individuals—have gradually been incorporated into the phonetic drift of a language. (p. )

Other researchers have also used this term in a diachronic sense, but to refer to languageinternal phonetic shifts in a category distribution (e.g., Solé & Recasens, ; Garrett & Johnson, ). Thus, the term ‘phonetic drift’ is often used to refer to diachronic change in ‘macro’ language (i.e., language at the level of the speech community), which may or may not be due to contact with another language. In this chapter, too, the focus is on change over time; however, I am concerned specifically with those L phonetic changes that follow from language contact (i.e., L experience) within a speaker’s lifetime. Consequently, I will discuss these developments primarily at the level of the individual (i.e., ‘micro’ language, or an idiolect), with a view toward elucidating the individual cognitive mechanisms behind such developments and their relationship to diachronic sound change in a speech community. Therefore, I will use the term ‘phonetic drift’ (or ‘drift’ for short) mainly to refer to L-influenced phonetic change in an individual’s L system. Since most of the research on phonetic drift examines the L production of L users, that is the focus of this chapter; however, it should be noted that phonetic drift is also reported in perception (Tice & Woodley, ; Namjoshi et al., ; cf. Cutler et al., ) and in L non-users (i.e., L speakers with only ambient L exposure).

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi





In the context of the following chapters in this section as well as others in this volume, an additional terminological question that arises is whether there is a difference between phonetic drift and phonetic or phonological attrition (see also Chapters  and , this volume, on Phonetic Attrition and Phonological Attrition by de Leeuw and Celata, respectively). In short, I will be discussing as cases of drift ostensibly short-term L changes, both in early- and late-onset L learners, which are attributable to recent L experience (e.g., because the change or divergence from L norms coincides with concomitant L exposure), whereas I will reserve the term ‘attrition’ to refer to long-term L changes in late-onset L learners which are unlikely to be due to recent L experience only (e.g., because the change perseveres long after a decline in L exposure). Naturally, this leads to the question of how drift relates to attrition, a question to which I will return in Section .. Studies of phonetic drift have been concerned with a number of different questions, some related to the dynamics of language contact and bilingualism (e.g., Yeni-Komshian et al., ; de Leeuw et al., ; Yao & Chang, ) and others related to the nature of phonetic processing and linguistic representation (e.g., Chang, ; Namjoshi et al., ). In this chapter, I will review findings that relate to three main questions: () What features of the L are subject to phonetic drift? () What is the cognitive mechanism (or mechanisms) behind phonetic drift? () What factors increase or decrease the likelihood of phonetic drift? To preview some of the current answers to these questions, virtually all aspects of L speech are subject to drift, but different aspects do not drift in the same manner, possibly due to multiple routes of L influence coexisting at different levels of L phonological structure. However, because the literature has generally examined L change in terms of sound categories rather than routes of influence, the rest of this chapter is organized by category type: Section . covers drift in consonants (e.g., in voice onset time (VOT)), Section . covers drift in vowels (e.g., in formant frequencies), and Section . covers drift in suprasegmentals (e.g., in fundamental frequency).

. P   

.................................................................................................................................. An abundance of acoustic studies have reported phonetic drift in consonants. The majority of this research has focused on oral stops (i.e., plosives)—in particular, the feature of VOT, the primary acoustic correlate of stop voicing contrast. In addition, some studies have reported drift in fundamental frequency (f0), a secondary acoustic correlate of voicing contrast; these are discussed further in Section . since f0 also serves as a primary cue to suprasegmental contrasts (e.g., tone, intonation). Finally, there are a few studies reporting drift in consonants other than stops (namely, fricatives and approximants). The recurring theme throughout these studies is that it is common for L consonants to drift in one or more ways following the onset of L experience, even in adult L learners for whom L categories might be considered to be relatively stable (i.e., ‘fully acquired’). As for the mechanism behind phonetic drift of consonants (as well as vowels), Flege (, , ) articulated an influential proposal based on 

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

 



  in a theory called the Speech Learning Model (SLM). According to the SLM, speech sound categories continue to develop over the lifespan; L and L sounds coexist in a shared mental phonetic space; and an L sound that is, at a position-specific allophonic level, ‘similar’ to an L sound tends to undergo a perceptually-based, automatic ‘equivalence classification’ with that L sound. Equivalence classification results in similar L and L sounds becoming linked under the same category and, consequently, converging towards each other phonetically. Since L–L equivalence classification becomes more likely as L categories widen over the course of normal L development, the probability of cross-linguistic perceptual linkage increases with a later onset of L learning. Late L learners are thus particularly subject to L influence on the L, whereas early L learners are more likely to notice the systematic phonetic differences between close L and L sounds and, therefore, to establish a distinct category for the L sound. When L and L sounds are separated categorically in this way, they may develop independently, resulting in native-like realization of both; however, they may also dissimilate from each other to maximize crosslinguistic distance within the shared L–L phonetic space, leading to divergence from native phonetic norms (e.g., Mack, ; Yusa et al., ). Several studies of late L learners have reported results consistent with the SLM view of the cross-linguistic interactions leading to phonetic drift. In one study of L French to L English and L English to L French speakers (Flege, ), the VOT of the voiceless coronal stop /t/, which is short-lag in French but long-lag in American English, showed drift in both L groups for speakers immersed in their L: L English speakers living in France produced English /t/ with shorter-than-native VOT, while L French speakers living in the US produced French /t/ with longer-than-native VOT. Similar results were reported by Major (, ) for L English to L Portuguese speakers living in Brazil. In addition, L Korean to L English speakers living in the US were shown to produce Korean fortis stops, which have slightly shorter VOT norms than English voiced stops (Chang, ), with longer-than-native VOTs that were virtually identical to the VOTs these speakers produced for English voiced stops (Kang & Guion, ). Although the above studies give suggestive evidence of phonetic drift in late L learners, they were not concerned with the extent to which the L phonetic changes persist and, consequently, were based only on cross-sectional comparisons, which have the inherent limitation of assuming that any cross-group differences providing a picture of ‘apparent time’ are due to the independent variable of interest (in this case, L experience) rather than some other factor that has not been brought under experimental control. This limitation makes it crucial to confirm such findings in longitudinal research, and indeed there are a few studies that have documented drift longitudinally within individuals. One example is a case study that followed an L Portuguese late learner of English as she travelled between Brazil and the US (Sancier & Fowler, ). This study showed that the VOT of L (as well as L) voiceless stops drifted toward the VOT norms of the most recently experienced ambient language; thus, the speaker’s VOT in L stops was observed to be significantly longer following a few months of L immersion in the US, an effect perceptible to native listeners. This drift was attributed to three factors: similarity between the L and L sounds (cf. Flege’s SLM), humans’ tendency toward imitation, and the recency effect on memory. The effect of recent (as opposed to temporally distant or cumulative) L experience on L production was further clarified in a series of longitudinal studies on L English late

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi





learners of Korean who were recent arrivals to Korea (Chang, , ). This work contributed three novel findings to the literature in this area: () early drift, () modulation of the recency effect, and () multi-level cross-linguistic linkages. First, in contrast to previous studies, which had focused on experienced L learners, Chang () examined inexperienced (i.e., novice) learners, finding that phonetic drift was not only evident, but relatively pronounced, during the first weeks of elementary L instruction. As in Sancier & Fowler (), the VOT of learners’ L voiceless stops drifted toward the longer VOT norms of a similar L stop series (in this case, the Korean aspirated stops); however, the amount of drift was larger than in Sancier & Fowler (). This disparity led to a follow-up study comparing inexperienced to experienced learners (i.e., those with prior exposure to the target L) enrolled in the same language programme (Chang, ). Contrary to the assumption that phonetic drift would be positively correlated with L experience/proficiency (e.g., Major, ), this study revealed that experienced learners showed less drift during the same time period of L learning, as shown in Figure . (which illustrates this effect for VOT as well as f0, discussed further in Section .). This result supports the view that the potential for recent L experience to lead to drift decreases over the course of L acquisition, which may be due to an L novelty effect for novice learners (Chang, ) and/or developmental changes in the relative strength of the L versus L systems ( Jacobs et al., ). As for the third finding of multi-level cross-linguistic linkages, this relates to the theoretical basis for phonetic drift assumed in Chang ()—namely, a cross-linguistic acoustic distance that is perceptible. For an L sound to be ‘similar’ (in SLM terms) to an L sound such that the two become perceptually linked, the L sound should be not so different that it is perceived as ‘new’, yet different enough that it is not perceived as ‘identical’. Thus, for any given acoustic dimension, L sounds are predicted to drift only when the distance from a ‘similar’ L sound is at least as large as the just-noticeable

(b)

VOT (ms)

110

inexperienced experienced

100 90 80 70 1 2 3 4 5 Time(weeks into language program)

F0 onset, standardized (z-scores)

(a)

0.5

inexperienced experienced

0.0

–0.5

1 2 3 4 5 Time(weeks into language program)

 .. Phonetic drift of English voiceless plosives in learners of Korean (A) drift in VOT for all learners; (B) drift in onset f0 for female learners. The two learner groups are inexperienced (circles) and experienced (triangles). Error bars show % confidence intervals. Source: Chang (: Figures  & ).

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

 



difference (JND) in that dimension. For VOT of the L (English) voiceless stops in Chang (), it was therefore predicted that /p/ and /k/ would drift, as the respective VOT norms for the similar Korean /ph/ and /kh/ are each longer by an amount that exceeds the JND for VOT in this range; in contrast, /t/ was predicted not to drift, as the VOT norm for Korean /th/ is longer by an amount that does not reach the JND. Results separated by a stop indicated that /t/ did not fail to drift, but rather drifted along with the other voiceless stops; at the same time, /t/ showed less drift than both /p/ and /k/. These findings thus suggest that production of L /t/ was influenced by linkages to the L at two levels: the individual sound and the natural class of sounds. The class-level linkage of L voiceless stops to L aspirated stops precipitated the drift of all voiceless stops (including /t/) due to the perceptibly longer overall VOT norm for the L class, whereas the segment-level linkage of L /t/ to L /th/ moderated the amount of drift for /t/ due to the relatively close VOT norm for L /th/. Given that the examples of drift discussed above were all found in L immersion contexts, a question that arises is whether drift also occurs in non-immersion contexts. Findings from a few studies examining L English late learners of Spanish in the US suggest that this is the case. In one study of advanced learners who used the L regularly in their professions (Lord, ), drift was evident in the VOTs of L voiceless stops, which were shorter-than-native for all stops but significantly so only for /k/. Further evidence along these lines was reported in a larger cross-sectional study of learners exemplifying a wide range of L proficiency levels from elementary to master’s level (Herd et al., ); the learners in this study, the majority of whom had never been immersed in the L for more than a week, produced VOTs for L voiced stops that were more negative (thus, more L-like) the higher their level of study, with master’s students consistently showing long negative VOTs. The cross-sectional results in these two studies converge with the results of a longitudinal study that tracked, and phonetically trained, five learners in an elementary Spanish course (Schuhmann & Huffman, ). Some (but not all) of these learners showed downward drift in VOT (i.e., VOT shortening) of L voiceless stops and/or a tendency toward downward drift in VOT of L voiced stops, suggesting again that phonetic drift is not dependent on an L immersion environment. To close this section, it should be mentioned that, although most findings of drift in consonants come from studies of stops, other types of consonants are clearly subject to change as well. For example, in a cross-sectional study of fricatives in L Taiwanese Amoy learners of Mandarin Chinese in the US (Peng, ), proficient, but not unproficient, Mandarin speakers showed movement of their L fricative /h/ toward the similar L fricative /x/ with respect to frequency range and spectral energy. Recent cross-sectional studies of L German late learners of English living in Canada (de Leeuw, Mennen, & Scobbie, ) and the UK (Ulbrich & Ordin, ) provided further evidence of L change in the formant properties (i.e., F1, F2, F3) of lateral and rhotic approximants, due to L–L differences in velarization and rhoticity. Note that the L changes in these studies were examined only after a period of extensive L experience, so it is unclear whether they reflect drift per se, as opposed to attrition (see Section .). Nevertheless, findings of L-influenced change in fricatives and approximants as well as stops suggest that phonetic drift is a wide-ranging phenomenon that may affect a variety of L consonants.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi





. P   

.................................................................................................................................. Like consonants, vowels also drift due to L experience, and in accordance with the SLM, this drift has often been explained in terms of cross-linguistic influence at an allophonic (i.e., segmental) level. However, the literature on vowels is similar to the literature on consonants in showing L influence at a higher (in particular, global or systemic) level as well. Segment-level and system-level effects of an L are not necessarily mutually exclusive, although there is some indication that the systemic influence of an L inventory may override the segmental influence of one L vowel. In this section, we consider both sources of vocalic drift as well as their interaction. A few cross-sectional studies have reported drift of selected L vowels due to the influence of a similar L vowel. In one of the earliest studies of this type, Flege’s () study of L French to L English and L English to L French speakers, late L learners were tested on their production of the high back rounded vowel /u/, which shows a significantly higher norm for F2 in American English than in French. Since drift was expected to involve movement toward the phonetic norms of a similar L vowel, L French to L English speakers were predicted to produce French /u/ with higher-than-native F2 values, and L English to L French speakers to produce English /u/ with lower-than-native F2 values. Indeed, the L French speakers showed the predicted drift pattern; however, the L English speakers did not, producing their L vowel close to native norms. These results were replicated in a later study of L Mandarin to L English and L English to L Mandarin speakers (Chang et al., ). Although the reason for this asymmetry in drift remains unclear (see Chang, , pp. –, for one possible explanation based in properties of human audition), what the asymmetry suggests is that drift cannot be predicted solely on the basis of cross-linguistic differences between similar sounds; rather, there are other intervening factors (which we discuss further in Section .). In addition to the data from these cross-sectional studies, further evidence of vocalic drift has come from a number of recent longitudinal studies. In one study of L Japanese learners of English taking up residence in an L environment, a year of L immersion was found to result in drift of some L vowels (specifically, /i a/) in terms of increased F2 values, but only for child learners and not for adult learners (Oh et al., ).1 On the other hand, research on adult L French speakers receiving phonetic training has shown clear effects of concentrated L exposure on L vowels (Kartushina, ; Kartushina et al., ). In this work, monolingual French speakers underwent an hour of production training on each of two target vowels, Danish /ɔ/ (which is perceived as similar to, as well as a good exemplar of, the L /o/) and Russian /ɨ/ (which is not perceptually similar to any L vowel, but acoustically closest to /ø y i/, in that order). Examination of pre- and post-training production of these L vowels revealed no drift of /o/ or /i/, but significant drift of /ø/ and /y/; furthermore, the L training effect was stronger for /ø/ than for /y/. These results thus suggest, in line with the SLM and Chang (), that drift is dependent on the disparity between L and L 1

Note that the inclusion of child learners in the participant sample crucially distinguishes this study from studies of attrition, which typically focus on adult learners whose L acquisition starts clearly after a critical/sensitive period.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

 

3

F1 (Bark)

4



late bilingual early bilingual

i i

ʊ ʊ

5

a

6

a

7 14

13

12

11

10

9

F2 (Bark)

 .. L Quichua vowels (/ɪ ʊ a/) according to timing of L onset Note: The late and early L learner groups are plotted in grey and black, respectively. Source: adapted from Guion (: Figure ).

sounds falling within a ‘sweet spot’ of cross-linguistic distance (i.e., not so small that there is no room to drift, but not so large that the L sound is no longer in the phonetic neighbourhood of the L sound); however, drift is not dependent on perceptual similarity specifically, as even the Russian /ɨ/, which would qualify as a ‘new’ L sound under the SLM, appeared to precipitate drift in this case. Whereas the above studies examined a subset of learners’ L vowels, other studies have looked at the whole L vowel inventory and found evidence for vocalic drift at a systemic level. In one such study of L Quichua to L Spanish speakers (Guion, ), early L learners produced all L vowels with lower F1 values than late L learners; that is, more L experience was linked to a global raising of the L vowel space, as shown in Figure .. Crucially, this pattern of drift could not be explained in terms of the attracting influence of the closest individual L vowel, because the L vowels did not consistently converge toward this vowel. Instead, Guion () accounted for this pattern in terms of maximizing dispersion with respect to the L vowel system, which contains two mid vowels /e o/ that could put pressure on the high vowels of Quichua to go higher. However, a different mechanism for this type of systemic drift was proposed in the longitudinal work of Chang (, , ) on L English late learners of Korean. Besides showing drift in VOT (as discussed in Section .), these studies showed, for female learners, a consistent raising of the L vowels, much as in Guion (). Unlike the case of L Quichua to L Spanish, however, the systemic drift observed in Chang () could not be explained in terms of cross-linguistic dissimilation, as the drift resulted in a lower degree of cross-linguistic dispersion. It could also not be explained in terms of separate individual vowel shifts, as the L vowels did not uniformly assimilate toward or dissimilate from the closest L vowel. Instead, the drift was attributed to convergence toward the L via a cross-linguistic linkage between the global formant levels associated with the L and L vowel systems. Since the L (Korean) inventory contains fewer low and mid vowels (i.e., those with relatively high F1 values) than the L (English) inventory, the average F1

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi





of the L vowel system is lower than that of the L system, resulting in phonetic drift taking the form of a decrease in F1 (i.e., vowel raising). This systemic raising of L vowels might also be explained in articulatory terms, as the result of cross-linguistic differences in articulatory setting (see, e.g., Gick et al., ). To be specific, the default (i.e., rest) position for the tongue in a given language might be relatively higher when the language’s phonological inventory leads to more production of high vowels on average; such cross-linguistic differences in default tongue height could explain the systemic raising observed in Chang () in terms of a lower default tongue position for the L being pulled toward the higher default tongue position for the L. Crucially, however, the articulatory account predicts the same direction of vocalic drift for female and male learners (because the L–L inventory differences are the same regardless of talker sex), yet this is not borne out by the male learners examined in Chang (, ). These male learners were, instead, found to drift in the opposite direction of female learners, which was attributed to the fact that, for male learners, the average F1 of the L vowel system (modelled by L instructors who were female) was higher, rather than lower, than that of their L system. Although the global features of L vowels that learners acquire, and drift toward in their L, remain to be fully specified, what these results suggest is that these features are probably not articulatory, but acoustic in nature. Despite involving quite different sets of L vowels, a point of overlap between Guion () and Chang () is the fact that drift took the form of vowel raising. This leads to the question of whether the raising in these studies was based on L–L comparisons at all or was simply a default pattern of drift. One study that helps to address this question is a cross-sectional investigation of two L Dutch to L English speakers, identical twin sisters who differed in terms of their L experience (Mayr et al., ). While one sister (‘TZ’) remained in the Netherlands, where both sisters grew up, the other sister (‘MZ’) had moved to the UK as a young adult and had thus been immersed in the L for many years. Acoustic analyses showed that MZ produced most (seven out of nine) of the L monophthongal vowels with higher F1 values (i.e., lower in the vowel space) than TZ did. This vowel lowering pattern was consistent with L–L differences in average F1, which, in this case, was higher for the L system than the L system (Mayr et al., , p. ). Additionally, it supports the view that systemic drift of vowels does not necessarily involve raising, but instead depends on the nature of acoustic differences between the L and L. The findings of recent research on L Polish to L English and L Polish to L German speakers provide additional evidence that systemic drift may be realized in a variety of ways (Sypiańska, ). The role of system-level differences in influencing vocalic drift was further highlighted in a longitudinal study of L English late learners of French (Lang & Davidson, ). This study tracked novice learners during a study-abroad programme in Paris and also compared these novice learners to advanced learners who were long-term Paris residents. Whereas the advanced learners showed evidence of vocalic drift, particularly in F1, the novice learners did not, in contrast to the results of Chang (, ). This apparently delayed onset of L effects from French vis-à-vis Korean suggests that the progression of vocalic drift may be influenced by specific properties (e.g., size, crowdedness) of the L and L vowel systems that come into contact in L learners. The last study to be discussed in this section relates to the effect of phonological context and the lexical (i.e., whole word) level. The relevance of these factors for vocalic drift

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

 



was shown in a study of L Shanghainese to L Mandarin speakers in Shanghai, which examined how production of Shanghainese /ɛ/ has been influenced by the similar Mandarin vowel /ej/ (Yao & Chang, ). This study tested two groups of speakers (younger and older), two language modes (mono- and bilingual), and three types of L lexical items cognate with L lexical items (one type whose phonological form was maximally similar to that of the L cognate, and two types that were each less similar). The crucial result has to do with item type: although all three types of L item contained /ɛ/ and were cognate with an L item containing /ej/, more drift of /ɛ/ was observed in the first type than in the other two, suggesting that L influence from /ej/ was modulated by the degree of similarity between the lexical contexts in which the vowels were embedded. That is, vocalic drift was influenced by L–L linkage not only at the level of the segment (i.e., /ɛ/ and /ej/), but also at the level of the lexical item. This case of drift is thus similar to the case of English /t/ in Chang () in evincing multiple layers of cross-linguistic comparison affecting phonetic drift.

. P   

.................................................................................................................................. Although the literature on drift in suprasegmentals is not as extensive as the literature on drift in segmentals, there are now a number of studies demonstrating that suprasegmental features of the L, much like segmental features, are influenced by L experience. This literature contains both cross-sectional and longitudinal data and mostly addresses drift related to pitch (namely, f0 level and alignment). The wider literature on contact varieties suggests that rhythm, too, may be influenced by bilingualism (e.g., Low et al., ; Carter, ), but there are as yet no systematic investigations of drift in L rhythm during lateonset L learning. Drift in f0 level has been documented in a few studies on Korean and English in contact. For example, a cross-sectional study of L Korean to L English speakers found that L experience was correlated with increased onset f0 values after L lenis stops (Yoon, ). In the longitudinal research on L English late learners of Korean discussed in Section . (Chang, , ), drift was found both in the VOT of voiceless stops and in the onset f0 of the vowel following these stops. There was an effect of gender on drift in f0, which arose only for female learners. Further, there was an effect of prior L experience, as drift was greater for inexperienced than experienced learners (Figure .B); nevertheless, drift in f0 was evident in both groups, as well as for both voiced and voiceless stops. In fact, upward drift in f0 was expected after both L stop types due to the pitch accent associated with the perceptually most similar L stop types (i.e., fortis, aspirated). Unexpectedly, however, f0 also drifted upward in the absence of any stops—namely, in vowel-initial items (see Figure .). This result was explained in terms of a control mechanism for f0 that is, to some extent, shared across languages (and, thus, influenced by global f0 properties of an L including overall f0 level). At the same time, f0 increased more in stop-initial than vowelinitial items, suggesting that the f0 properties associated with the fortis and aspirated stops specifically were also at play. Thus, similar to the case of VOT, drift in f0 was influenced by

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



 240

D voiced stops

230

A vowel-initials

F0 onset (Hz)

T voiceless stops

T

220 210 200

T

D A

T D A

T D A

T D A

A D

190 1 2 3 4 5 Time (weeks into language program)

 .. Phonetic drift of onset f0 in English words produced by female learners of Korean, by initial segment type. Note: Error bars show % confidence intervals. Source: Chang (: Figure ).

a multifaceted relationship between the L and L, here involving overall f0 level (i.e., a global property) as well as the natural class-level properties of specific stop types. In addition to f0 level, change may also occur in f0 alignment. This type of change was found in a cross-sectional study by Mennen (), which examined the temporal alignment of the f0 peak in an intonation contour as realized by L Dutch late learners of Greek and native speakers of Dutch and of Greek who did not know the other language (i.e., control groups). The Dutch learners and controls were living in the Netherlands or in Scotland, while the Greek controls were living in Scotland. They were compared on their production of a prenuclear rise (LH*), which is present in both languages but peaks earlier and is subject to different phonological conditioning (by vowel length) in Dutch than in Greek. Most Dutch learners of Greek produced the rise in Dutch differently than the Dutch controls, in addition to producing the rise in Greek differently than the Greek controls. On the other hand, one learner produced the rise in both Dutch and Greek like the respective control groups. Taken together, the findings of this study, as well as of a later study of L German to L English speakers (de Leeuw et al., ), converge with findings on segmental properties (Sections . to .) and overall accent (Yeni-Komshian et al., ; de Leeuw et al., ) in showing that L experience often, but not always, leads to L changes that diverge from native phonetic norms. Importantly, however, the intonational changes in Mennen () cannot be explained in terms of the same type of L–L linkage (and phonetic convergence) posited in the SLM for segments. This is because positing convergence between L and L tonal phones would predict that the relatively late peak alignment pattern of the L should result in later L peak alignment, whereas this is not the pattern observed. Instead, the Dutch learners of Greek tended to produce earlier L peak alignment than the Dutch controls in short vowels (where the peak typically occurs late, although not as late as in Greek), while producing similar L peak alignment as the Dutch controls in long vowels (where the peak typically occurs early). Thus, in this case L influence did not take the form of assimilation of L-specific phonetic implementation; rather, it appeared in a smaller differentiation of L long and

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

 



short vowels in terms of peak alignment (i.e., attenuation of an L-specific contrast). These findings are consistent with the view that drift of certain suprasegmental features may operate differently than drift of segmental features, although much more research on suprasegmental changes (including targeted studies of drift per se as opposed to divergences that may reflect attrition) is needed to shore up any claim about segmentalsuprasegmental disparities in L change.

. D  

.................................................................................................................................. To return to Questions () to () from Section ., recent experience with an L can affect many features of the L, but the specific effect varies across features as well as other dimensions such as the developmental profile of the L. The literature on L learners contains findings of L change in the VOT of stops, spectral characteristics of fricatives, approximants, and vowels (e.g., F1, F2, F3), f0 level, and f0 alignment. One route of L influence on the L is through equivalence classification of an L sound with an L sound and their resulting perceptual linkage, as proposed in the SLM. However, this type of L–L linkage only predicts phonetic drift at a segmental level, whereas drift has also been observed at other levels: the natural class of sounds, the vowel system, and overall f0 level. This multifaceted nature of drift implies that L knowledge may be linked back to the L in multiple ways, which are often seen to jointly influence L production. As for the factors affecting the likelihood of phonetic drift, (i.e., Question ()), these include linguistic, cognitive, and socio-demographic variables, some of which are related to the individual differences in drift observed in the literature. First, the specific L feature at issue is relevant, as some features may be inherently more grounded in somatosensory feedback (see, e.g., Chang, ), decreasing the likelihood of drift, or include languagegeneral aspects of representation or control, increasing the likelihood of drift. This crossfeature variation is seen, for example, in differences between drift in vowel quality (Section .) and drift in f0 (Section .). Furthermore, the acoustic distance between an L structure and a close L structure is crucial, as it needs to be ‘just right’: large enough to be perceptible and/or provide room for the L structure to drift, but small enough for the L structure to qualify as an attractor (Flege, ; Chang, ; Kartushina et al., ). Prior L experience is also relevant, as it diminishes the potential influence of recent L exposure, and L categories that are inherently less variable (i.e., more stable in their realization) may generally be more resistant to drift (Chang, ; see also Tobin, ). Apart from variation in prior L experience, factors contributing to individual differences in L acquisition are also likely to be correlated with at least some of the individual differences that have been observed in phonetic drift (Huffman & Schuhmann, a,b; Schuhmann & Huffman, ; Schwartz et al., ). These include cognitive, affective, and attitudinal variables such as executive control, affinity for the culture and/or people associated with the L, and motivation (for an overview, see Dörnyei & Skehan, ). Although individual differences in drift remain largely unaccounted for, recent work relating variation in drift to variation in inhibitory abilities (Lev-Ari & Peperkamp, ) presents a promising first step toward a better understanding of the source of these individual differences in L malleability.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi





Given the surface similarity between drift and attrition, one may ask how a particular instance of L phonetic change can be classified as one versus the other. In Section ., we distinguished drift from attrition on the basis of perseverance as well as underlying cause, with attrition lasting in the absence of a proximal L stimulus (i.e., recent L experience) and drift not. Few studies, however, have actually tracked L learners through alternations in language environment, with the result that it is often unclear whether observed L changes are short-term, long-term, or medium-term (i.e., reversible, but not quickly or easily) and, more problematically, impossible to tease apart the effect of L exposure from the role of L acquisition, L knowledge, or L use. Thus, the exact source of phonetic drift, within the broad construct of recent L experience, remains largely unaddressed, raising a number of interesting questions. Can phonetic drift arise, for instance, solely through ambient L exposure, without L knowledge or use? Very little research has directly addressed this question, but one study of monolingual French speakers (Caramazza & Yeni-Komshian, ) suggests that the answer may be yes. In this study, French speakers in Canada produced less prevoicing (i.e., less negative VOTs) for voiced stops and longer positive VOTs for voiceless stops than French speakers in France, indicating convergence toward the norms of Canadian English despite not actually speaking English. Note, however, that these results were not replicated in a later study (Fowler et al., ), which may be due in part to a change in the relative sociolinguistic status of French vis-à-vis English in Canada during the time between the two studies. Complementing Caramazza & Yeni-Komshian (), results from a follow-up study to Chang (, ) further suggest that some aspects of drift are due primarily to L exposure (and, therefore, persist as long as L learners are in an L environment) whereas others are tied to L use (Chang, , ). In this study, a subset of the L learners examined in Chang (, ) were examined again at the end of a year in the L environment. An early drop-off in L use was found to lead to a reversal of drift in VOT and F1, but not in f0 or F2, suggesting that once an L has been acquired to a certain level, it is difficult for phonetic drift to subside completely in an L environment. When L learners return to an L environment, it may be the case that many of the L changes observed in an L environment are eventually reversed, and certain cross-sectional comparisons are consistent with this hypothesis. For example, the inexperienced and experienced learner groups in Chang (), both arriving directly from an L environment, were similar in their L production at the start of the study, suggesting that the L (re)immersion that followed the experienced learners’ previous L learning largely erased any drift that may have occurred during the previous L experience. It bears repeating, however, that no longitudinal research has provided direct evidence of what happens to the L phonetically when L learners disengage from the L and return to an L environment. Despite the relative lack of evidence on the effects of L reimmersion, some researchers (e.g., Stolberg & Münch, ) have argued that L-influenced changes in an L should not be defined in terms of their perseverance, because it may be the case that all behavioural changes resembling attrition are reversible once robust contact with the target language is renewed. Under this view, there is no principled reason (such as a distinction between ‘competence’, or knowledge, and ‘performance’, including factors such as language activation and accessibility) for distinguishing between long-term and short-term changes to the L system. In contrast to this view, I consider the null hypothesis, at this point in our

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

 



understanding of L development over the lifespan, to be that some aspects of L change do not reverse themselves even after renewed L contact. In other words, regardless of language environment, L learners may continue patterning differently from monolingual L users in certain ways, a view supported by the findings of bilingual-monolingual differences throughout the psycholinguistic literature (for a review, see Grosjean & Li, ). In closing, the conceptual proximity of drift and attrition invites the essential question of how phonetic drift relates to phonetic and phonological attrition. I consider phonological attrition to comprise long-term changes that cross a phonemic boundary; this includes, for example, change in the application of L phonological rules that add, delete, or switch segments from one category to another (e.g., Dmitrieva et al., ; Joh et al., ; Cho & Lee, ). Phonetic attrition, by contrast, comprises long-term changes that are subphonemic, while phonetic drift comprises short-term changes of this type. Recall that the crux of the distinction between drift and attrition is in the timeline and, by implication, the source of change: whereas drift is a short-term effect attributable to recent L experience, attrition is a long-term effect arising from extensive, and not necessarily recent, L contact. Drift may thus be viewed as the beginning of a trajectory of L-influenced change in the L that, if not reversed, leads ultimately to attrition (for late L learners) or to ‘incomplete acquisition’ (for early L learners; see Montrul, ). As suggested in Chang (), one factor that seems to reduce the magnitude of drift is alternation between periods of L and L immersion, which gives the L system the opportunity to retune to L norms; this suggests that the scenario of continuous L immersion is particularly conducive to progressing from drift to attrition. In this regard, it should be noted that the literature on L change is, in fact, dominated by cases of ongoing L immersion, which is why many of the findings reviewed in this chapter are inherently ambiguous between exemplifying drift or attrition. Looking forward, there are many open questions regarding phonetic drift, but two overarching questions are particularly worth highlighting here. First, how exactly does L exposure, as opposed to L use, lead to phonetic drift as well as contact-induced diachronic sound change? In the case of Caramazza & Yeni-Komshian (), for example, it is not clear whether the divergence from L norms in monolinguals with ambient L exposure is a direct L effect (i.e., phonetic drift arising from ambient L exposure) or an indirect L effect (i.e., faithful acquisition of different L norms from the drifted or attrited production of L-speaking parents). Second, how does drift of the L in L learners resemble and/or differ from drift of the L and/or L in third-language (L) learners? The burgeoning field of L/Ln acquisition (see, e.g., Cabrelli Amaro, ) is beginning to address this type of question and represents an exciting area of future research into the dynamics of language development and interaction in an increasingly multilingual world.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

  ......................................................................................................................

  ......................................................................................................................

  

. I

.................................................................................................................................. A, it is not uncommon for people to know friends or family members who moved abroad, and return home with speech differently accented in comparison to before their departure. Or, perhaps, one has experienced such changes in pronunciation oneself. Indeed, there is consistency among the research discussed in this chapter that the native phonetic domain is malleable upon competition from a new language or dialect in adulthood, although the extent of such effects differs between individuals (Shockey, ; Flege, ; Major, ; Munro et al., ; Mennen, ; Evans & Iverson, ; de Leeuw, , ; de Leeuw, Schmid, & Mennen, ; de Leeuw, Mennen, & Scobbie, , ; Mayr et al., ; Hopp & Schmid, ; Ulbrich & Ordin, ; Bergmann et al., ; de Leeuw, Tusha, et al., ). Research into phonetic attrition (here, attrition is used solely with reference to the L) is important because native language phones are the building blocks from which all other language representations are built (i.e., in L acquisition, children must firstly perceive sounds without an underlying category, before distinctive phonemes are acquired, words built, and sentences constructed, e.g. Maye, Werker, & Gerken, ; Polka & Werker, ; Stager & Werker, ; Werker & Lalonde, , ). Phonetic attrition is therefore indicative of an unbalancing of the native language system, with potential trickle down effects evidenced in phonological attrition, see Chapter , this volume, on phonological attrition by Celata. Moreover, findings suggest that with training the effects of a new competing linguistic system on the native language phonetic system can occur rapidly (Chang, ), and that short-term modifications to the phonetic system occur dependent upon the ambient linguistic system (Sancier & Fowler, ), see Chapter , this volume, on phonetic drift by Chang. For these reasons, research into phonetic attrition is of great consequence to our understanding of the plasticity of native speech, how languages and dialects interact and are represented in the mind across the lifespan, as well as to how individuals who possess multiple languages and dialects are perceived by other native speakers.

.. Definition of attrition Research into phonetic attrition usually aims to describe changes in pronunciation which occur within an individual’s L when an L is acquired in adulthood, as well as to explain

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

 



why such changes occur. Within the context of bilingualism, attrition is most often investigated within the scope of long-term immigration (Schmid & Köpke, ). In this chapter, the characterization of attrition is broadened to include studies that have examined pronunciation changes in the native language upon acquisition of a new dialect in adulthood. The inclusion of dialects into a definition of attrition is relevant when one considers that the terms language and dialect represent a continuum and that their ‘edges are extremely ragged and uncertain’ (Haugen, , p. ). Attrition is not the only way in which an individual’s native speech can change post adolescence. Dialects themselves change over time which affects the individual’s speech production (Harrington, ; Harrington et al., a,b). Individuals can style-shift according to variable social contexts (Bell, ; Coupland, ; Eckert & Rickford, ), and, as previously mentioned, short-term rapid changes to the L phonetic system have also been reported upon L training and exposure (Chang, ; Sancier & Fowler, ). However, studies into phonetic attrition are considered to reveal long-term changes in the native language pronunciation of an individual as a result of the acquisition of a new dialect or language acquired post adolescence. Emerging longitudinal research may reveal how such long-term phonetic changes in the form of attrition are affected, or perhaps governed, by such short-term changes, or whether there is a cut-off between the two at all. Together, research which reports phonetic attrition, as described in this chapter, combined with changes to the L phonetic system due to the aforementioned factors, lend support to the understanding that there is an inherent degree of plasticity to the native phonetic system both during childhood (Pallier et al., ; Ventureyra, Pallier, & Yoo, ), as well as post adolescence as it develops and adapts throughout the lifespan (see, e.g., de Leeuw, Opitz, & Lubinska, ).

. A  

.................................................................................................................................. This contribution aims to summarize research into phonetic attrition by examining the question of whether the pronunciation of a linguistic system learned in childhood and adolescence can change in adulthood, potentially, in an extreme case, even becoming ‘foreign accented’ (de Leeuw et al., ; Hopp & Schmid, ), that is, being perceived by native speaker listeners to be non-native speech. In doing so, firstly, the term phonetics will be defined as applied in this chapter, and thereby differentiated from phonology (see Celata, Chapter , this volume). Thereafter, the Speech Learning Model (SLM) and its relevance for studies examining phonetic attrition will be discussed, with a view towards assimilation and dissimilation: both processes can occur in phonetic attrition. Subsequently, findings from research into both segmental and suprasegmental phonetic attrition will be summarized, as related to both bilingualism and monolingualism. Research that has examined native speech which is globally regionally or foreign accented will be discussed thereafter. Questions of psycholinguistic, socio-linguistic, and formal relevance are as follows. . Why do some individuals appear to undergo more phonetic attrition than others? . What can the processes of assimilation and dissimilation tell us about the nature of changes to the phonetic system? . Is there a difference between phonetic attrition in a bilingual versus bidialectal context?

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



 

. W  ?

.................................................................................................................................. Phonetics can be defined as the study of the realization of sounds in spoken language. For example, in the word love, phoneticians might examine the actual realization of the lateral sound at the beginning of love—that is, [l]. Notably, the phonetician might investigate possible variability in the production of the lateral sound, for example whether it was realized with a higher pitch in the case of a child, or a lower pitch in the case of an adult. Indeed, there are infinite ways in which individual lateral phones, [l], and indeed all sounds, can be realized. In contrast, phonology can be defined as the study of the organization of the sounds in spoken language. In the example of the word love, a phonologist might be interested in the fact that the lateral phoneme can be replaced by /d/ and that the new word arising is dove. One way of conceptualizing phones and phonemes is through a snowflake metaphor. Phones are like individual snowflakes: each one is different; phonemes are like the general category of snowflake in our mind. Although each individual snowflake is different, we are nonetheless able to perceive all snowflakes to belong to the category of snowflake; and we have a separate category for snowflake than we do for hail stone, than we do for rain drop—the infinite variability is constrained. In terms of attrition, the difference between phonetics and phonology is crucial. Studies which reveal phonetic attrition (Shockey, ; Flege, ; Major, ; Munro et al., ; Mennen, ; Evans & Iverson, ; de Leeuw, Schmid, & Mennen, ; Mayr et al., ; de Leeuw, Mennen, & Scobbie, , ; Hopp & Schmid, ; de Leeuw, ; Ulbrich & Ordin, ; de Leeuw, Tusha, et al., ), indicate changes with regard to how speakers of a particular language realize particular phones in their native language. In contrast, studies which reveal phonological attrition, of which there are far fewer (see, e.g.; Dmitrieva et al., ; and de Leeuw et al., ), reveal representational changes to the phonemic organization of the native language. Socially, both phonetic and phonological changes to the native language or dialect may vastly influence how an individual is perceived by others, and indeed potentially their own self-perception.

. S L M (SLM)

.................................................................................................................................. Although most widely applied in studies of L acquisition, the SLM was identified early on as relevant to studies into phonetic attrition (see de Leeuw, ), as, in addition to predicting how the L will influence the L, the SLM makes claims about how the acquisition of an L impacts the L. Crucial for phonetic attrition, the SLM posits that the phonic elements making up the L sound system and the phonic elements comprising the L system (either newly established categories, or adaptations of L categories) exist in a ‘common phonological space’, and so will mutually influence one another. (Flege, , p. )

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

 



either through the process of assimilation or dissimilation (Flege, ). With regard to phonetic category assimilation, the SLM posits that assimilation will occur when ‘the L learner continues indefinitely to judge the instances of an L category to be instances of an L category’ (Flege, , p. ). Alternatively, category dissimilation occurs because ‘bilinguals strive to maintain phonetic contrast between all of the elements in their L/L phonetic space, just as monolinguals strive to maintain phonetic contrast among the elements making up their L phonetic space’ (Flege et al., , p. ). However, factors which would explain whether two sounds from the given L and L undergo assimilation or dissimilation, are not given. Moreover, the SLM explicitly constrains itself to segmental elements of speech, rather than prosody, which may also undergo attrition, and is hence also discussed in this chapter. Nevertheless, the bidirectional application of the SLM is extremely relevant to research into phonetic attrition and its simplicity provides a fruitful foundation for studies.

. S    

.................................................................................................................................. Evidence of assimilation in phonetic attrition has been revealed in both segmental and prosodic analyses (note again that although the SLM does not directly make predictions with regard to prosody, assimilation of prosodic categories will be discussed hereafter). For example, although the term ‘attrition’ was never used, a seminal study by Flege () showed that in native American-English speakers immersed in a French environment as adults, and in native French speakers immersed in an American-English environment as adults, voice onset time (VOT) of /t/ productions in English and French were ‘merged’ (Flege, , p. ), becoming intermediate to the target language norms (longer VOT values in English than French). Additionally, within the same study it was found that native French speakers, ‘for whom English was clearly their principal language’ (p. ), produced French /u/ with a mean frequency of the second formant (F) that was higher ( Hz) than the value obtained for French monolingual subjects ( Hz), approximating English monolingual values (p. ). Although the results from the vowel analysis were not significant, they suggested that the prolonged acquisition and use of an L phonetic system affected the L phonetic segments. With regard to the contrast between phonetics and phonology, note that this does not constitute the loss of a phonemic contrast in the L categories, but rather a change in the production of the phonetic realization. Interpersonal variation was not specifically examined in this study; however, the standard deviation of the VOT in the French L speakers in Chicago, in their French, overlapped with the French monolinguals’ standard deviation of VOT. Similarly, the standard deviation of the English L speakers in Paris, in their English, approached the standard deviation of the English monolinguals. Such interpersonal variation suggests that not all subjects evidenced assimilation effects in their /t/ and /u/ productions, others may have undergone less segmental attrition or, alternatively, they may have displayed dissimilation effects. Another study (Flege & Hillenbrand, ) investigated a similar group of French native speakers, who had all learned English as adults and were living in an English-speaking

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



 

environment for an average of just over twelve years. These late sequential bilinguals were all married to native English speakers and, similarly, the VOT of their /t/ in both French and English was examined. Here too, the averaged results of the subjects indicated that in general the VOT of /t/ in their French was substantially longer (and more English-like) than that of monolingual French speakers (Flege & Hillenbrand, ). Moreover, in the same study it was reported that ‘the bilingual French native speakers produced French /u/ with a substantially higher mean F value ( Hz) than previously reported for monolingual French speakers ( Hz)’ (p. )—that is, long-term exposure to English influenced their production of French /u/, causing it to be more English-like, just as it influenced their production of French /t/. The focus of this investigation was again on group differences, rather than on inter- or intrapersonal differences within the speakers who comprised the groups, with the take-away finding that phonetic attrition at the segmental level is exhibited through assimilation effects in sequential bilinguals resident in a prolonged L environment. Building on these group analyses, an influential study of five native American-English speakers who had acquired Brazilian-Portuguese in adulthood, examined such inter- and intrapersonal variation. Major () was the first to explicitly use the term ‘attrition’ in a phonetic context. Even though the subjects in this study reported personal and professional reasons to maintain English in Brazil, all sequential bilinguals, who had been living in Brazil for between twelve and thirty-five years, exhibited some degree of phonetic attrition in their L realization of VOT (it became shorter and therefore more Portuguese-like, i.e., assimilation effects). In general, there was also a correlation between proficiency in the L, measured according to the realization of Portuguese-like VOT, and rate of attrition in English. On average, the lower the VOT (less native-like) in the English casual speech of his subjects, the lower the Portuguese VOT (more native-like). However, this correlation was not displayed in formal English speech (elicited through word and sentence lists, whereas casual speech was taken from informal spontaneous conversations). Moreover, one subject performed outside of the monolingual norms in both English and Portuguese, whilst another performed within the monolingual norms of both languages, at least in formal speech. Therefore, it is possible to summarize that socio-linguistic factors—that is, degree of formality, also contributed to the extent of phonetic attrition in this segmental analysis, suggesting that although phonetic attrition of segments may be inevitable in a prolonged L environment, more naturalistic forms of speech elicitation may be more likely to promote phonetic attrition—that is, if speech is elicited formally, the effects of phonetic attrition might not be evidenced at all. To a certain extent, these studies within a bilingual context built on an early impressionistic study of monolingual phonetic attrition (Shockey, ). In this study, four American English-speaking residents of the United Kingdom, who had been living in England for between eight and twenty-seven years, were examined. Her study, within a context of dialect acquisition, involved counting the number of flaps in American speech production, where in Standard Southern British English (SSBE) one would expect a voiceless or voiced alveolar plosive—that is, /t/ or /d/ in words like, respectively, latter and ladder. Shockey found differences between subjects, with some individuals exhibiting a greater degree of attrition of segments, and in different phonetic contexts. The voiceless alveolar plosive /t/ was much less likely to flap if it was in a relatively slow phrase, or if it had a high information content; and ‘both /t/ and /d/ showed a strong tendency to flap in fast,

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

 



unstressed, highly redundant speech, with the tendency being much stronger for /d/ than for /t/’ (pp. –). Although never drawing on the term ‘attrition’ specifically, Shockey interpreted that the changes in the American English speech were the result of a ‘slow, ongoing process which is not completed for a number of years’ (p. ). In line with Major (), her findings suggest that not only amount of exposure to the new linguistic system might account for interspeaker differences in phonetic attrition, but that speech style also affects intrapersonal variation, with careful speech and high information content eliciting less phonetic attrition. Aside from these early studies into segmental assimilation effects in plosives, formant frequencies in laterals, rhotics, and vowels have also revealed assimilation effects. For example, within the context of dialect acquisition in monolinguals, a more recent study (Evans & Iverson, ) investigated changes in vowel production among university students from Ashby de la Zouch, Leicestershire, UK, a small town in the Midlands where the local accent is classified as a variety of northern English (Wells, ; Evans, ). This study enhanced the Shockey () research because it involved both an acoustic analysis, as well as a regional accent rating by fellow British English speakers. Moreover, the subjects in this study had only been resident in their new environment for three years as they had moved to London to study, where they were exposed to SSBE more frequently than in their home town. In the group analysis, the results showed that their accents became more like SSBE, again revealing plasticity in the L phonetic system through assimilation effects. Regarding the acoustic formant analysis, the subjects changed their production of bud, cud, and could so that they produced them with a more fronted (higher F frequency) and lower (higher F frequency) vowel after only two years at university, closer to how southerners produce the vowel in words like bud and cud. However, there were differences in the amount and direction of accent change: although the majority of subjects were rated as sounding more ‘southern’ after two years than when they were first recorded; in a minority, there was only a very slight change in accent; three subjects were rated not to have changed their accent at all; and one subject was judged to have a more ‘northern’ accent at the final testing session than when she first moved to London. This study indicates not only a high degree of interpersonal variation, but also that the process of phonetic attrition, here in a monolingual context, can occur relatively quickly, and that it therefore might not be a slow process, as suggested by Shockey (). Moreover, although only revealed in one subject, it also hints at dissimilation effects, to be discussed subsequently. Formant frequencies have been the subject of further research into phonetic attrition within the context of bilingualism (de Leeuw, Mennen, & Scobbie, ; Ulbrich & Ordin, ; de Leeuw, Tusha, et al., ). An investigation into coda lateral production in the German speech of ten German-English late bilinguals who had been living in Vancouver, Canada for between eighteen and fifty-five years, indicated that the F frequency of the lateral phoneme in German words like viel and Kiel was significantly higher in the German of both the female and male bilinguals than in a respective German control group, becoming more English-like, and therefore indicating segmental attrition through assimilation effects. For the males, F frequency was also significantly lower in the German of the bilinguals than in the German monolinguals, as is characteristic of the ‘darker’ North American English lateral, again, revealing phonetic attrition. Moreover, some of the F frequencies were characteristic of the dark values, whilst the F values were characteristic of

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



 

the light values, suggesting a truly assimilated production of the lateral. Interpersonal variation across subjects was also revealed: one male subject (FS) produced the German lateral completely within the English monolingual norm whilst two female subjects produced the German lateral entirely within the German monolingual norm. In line with the previously discussed studies, it was again revealed that phonetic attrition in the form of assimilation effects is an outcome of late bilingualism, although the degree of plasticity of the phonetic L system varies across individuals. The results from this study prompted further investigation into one individual, FS, who stood out, and revealed ‘extreme’ phonetic attrition in his speech (de Leeuw et al., ). Subject FS was selected for this case study because he displayed phonetic attrition in all of the group phonetic analyses he underwent, whilst the others nine bilinguals who had moved from Germany to Canada did so as well, but less consistently. In addition to the lateral analysis (de Leeuw, Mennen, & Scobbie, ), in which it was found that his realization of German /l/ in coda position was entirely within the ‘dark’ English norm, in a prosodic analysis it was revealed that his realization of the prenuclear rise in his native German was significantly earlier than is characteristic of German; instead, falling within the English monolingual norm (de Leeuw et al., ), to be discussed shortly. Moreover, FS was salient because he was consistently rated to be a non-native speaker of his native German by German monolinguals in Germany in a global foreign accent rating (de Leeuw et al., ), as will also be discussed. The particular case study investigated phonetic attrition in the rhotic realizations of FS, whose biographical data indicated that, out of all the participants, he was among one of those who had been with their partner the longest, who was a native English speaker and with whom he only spoke English. They had two adult daughters and three grandchildren, with whom he only spoke English, and he had moved to Canada when he was , and thereafter had lived in Canada for fifty-three years. Crucially, when he spoke English, which he did almost always, he did so with next to no code-mixing into German, as none of his English interlocutors spoke German. He therefore differed from the other late bilinguals, who, in one way or another, were all considered to use German more than FS. The impressionistic analysis of his rhotic production revealed that % of his German rhotic productions were perceived to be English-like, much higher than that of any other participants (ranging from % to %). The average F frequency of his German rhotic (, Hz) approximated values typically given for American-English, such as , Hz (Dalston, ), and the range of between , and , Hz for English rhotics (Thomas, ), and substantially differed from values typically given for Standard German of between , and , Hz (Ladefoged & Maddieson, ). Although all participants in this study into rhotic realization revealed some degree of phonetic attrition, FS did so more extremely than any of the other participants, and realized the English retroflex rhotic in his German. The authors tentatively suggest that, essentially, complete immersion within an entirely English language setting is what made FS unique, and triggered his extreme attrition, in the domain of phonetics. As such, the findings shed new light on the SLM in that his German rhotic was not merged intermediately between German and English monolingual norms, but rather, in this extreme case of phonetic attrition, his rhotic production entirely assimilated within the monolingual norm of the English retroflex. Another related study examining formant frequencies in rhotics reported findings from an analysis into vocalization and rhotic realization of post-vocalic /r/ in the German native

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

 



speech of sequential bilinguals who had acquired either a rhotic variety of English as an L (Belfast English), or a non-rhotic variety of English (Oxford English) (Ulbrich & Ordin, ). In Belfast, the earliest age of arrival was  years of age, whilst the oldest age of arrival was ; in Oxford, the earliest age of arrival was  years of age, whilst the oldest age of arrival was . The results of the auditory and acoustic analysis of post-vocalic /r/ in the speakers’ German and English suggested that attrition occurred in the form of transfer from the L—that is, the German native speakers in Belfast and Oxford produced acoustic correlates of rhoticity in their native German speech. However, as noted by the authors, those exposed to Oxford English would not be expected to produce post-vocalic /r/, yet the German native speakers in Oxford did produce the acoustic correlates of rhoticity in their German, just like those with Belfast English as their ambient language. The authors therefore suggest that direct transfer from the L to the L could not have been the sole factor determining phonetic attrition. Instead, they suggest that between subject variation should take into account ‘actual creative language use and communicative events’ (p. ). What particularly stood out in this study is that dialectal variation in an L did not necessarily modulate the effects of attrition, and that, although sequential bilinguals may be exposed to a particular variety, other factors, dissimilating from the target L group in L speech, potentially towards a perceived more standard-like variety, for example American English, may also play a role. In line with these studies, a case study examining the speech of a monozygotic twins compared two sisters through an investigation into numerous phonetic variables (Mayr et al., ). MZ had moved from the Netherlands to the United Kingdom in early adulthood and where she had been living for thirty years at the time of data collection, when she was  years of age. Her speech was compared to that of her twin sister, who had remained in their country of birth, the Netherlands, and phonetic attrition was therefore measured in a perfectly controlled environment. The results revealed that MZ produced Dutch voiceless plosives with VOT values longer than the native Dutch norm of her twin sister, but not as long as the aspirated plosives of English, and that, overall, there was evidence for ‘crosslinguistic assimilation patterns for all plosives, with voiced ones realized with a voicing lead in both languages, and voiceless ones with VOT values intermediate between short-lag and long-lag categories’ (p. ). In addition, MZ’s monophthongs and diphthongs followed a general trend towards more open (higher F) realizations compared with her twin sister’s Dutch norm, which is in line with the expected more open pronunciation of English vowels (Mayr et al., ), with no dissimilation effects reported. Moreover, ‘not only did L categories assimilated to L ones shift towards a more open position, but also L categories with no counterpart’ (p. ). Therefore, the authors suggest that cross-linguistic interactions operate at a system-wide level, rather than at the level of individual sounds, which is consistent with related research into early bilinguals (Guion, ) short-term phonetic drift effects (Guion, ; Chang, ), as well as with the notion of language specific phonetic settings (Mennen et al., )—that is, the individual undergoing attrition does not only acquire individual sounds in the L, but rather a particular articulatory setting, and this configuration impacts the L phonetic space generally, rather than token-by-token. However, dissimilation effects, which have also been reported in phonetic attrition studies (Flege & Eefting, a; Evans & Iverson, ; de Leeuw et al., ), do not feed into this interpretation cleanly. If there is a system wide shift, there must also be scope

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



 

for phonetic attrition to occur at a level of individual sounds, as some bilinguals evidence dissimilation whilst others evidence assimilation for the same sounds.

. P    

.................................................................................................................................. In addition to such segmental changes in the speech production of the L in sequential bilinguals undergoing attrition, it has also been shown that prosodic realizations are susceptible to attrition and can undergo assimilation effects. Mennen () investigated native Dutch speakers who were at near-native level in their acquisition of L Greek. Although her sequential bilinguals did not live in Greece, they had all learned Greek as an L in adulthood; all had extensive experience with the L (between twelve and thirty-five years); all held a university degree in Modern Greek language and literature; were teaching Greek at university level in the Netherlands, and all used Greek regularly in their daily lives. As such, they were similar to subjects investigated in attrition studies, and the findings from her study are therefore of consequence to the questions of this chapter. Four out of five of her subjects were not only unable to realize Greek tonal alignment according to monolingual norms, they also showed a change in their native Dutch tonal alignment patterns. Specifically, the differentiation in the alignment of pitch peaks across Dutch long and short vowels was greatly reduced in their L speech. However, similar to the aforementioned interpersonal variation revealed (Major, ; de Leeuw, Schmid, & Mennen, ; de Leeuw, Mennen, & Scobbie, , ) one of the bilinguals did produce tonal alignment with native-like values in both the L and L, again suggesting that the extent of plasticity of native speech differs between individuals.

. S    

.................................................................................................................................. In addition to assimilation effects, research into the speech of late bilinguals has suggested that the late acquisition of an L can have dissimilation effects on the phonetic system of the L. One study which explicitly reported dissimilation effects is that of Dutch native speakers who were highly proficient in English as an L, which they had begun learning at  years of age in the Netherlands. These sequential bilinguals, who were between  and  years of age at the time of the experiment, and highly proficient in English, produced their Dutch /t/ with shorter VOT values than a group of Dutch L speakers who were less proficient in English (Flege & Eefting, a). Specifically, in the highly proficient subjects, the Dutch /t/ moved away from both the typical English value and the typical Dutch value (becoming shorter). Flege & Eefting (a) suggest that this may have been a result of ensuring sufficient discrimination between the L and the L segments, but it is difficult to reconcile that these results are in contrast to the previously reported assimilation effects in

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

 



VOT for the same language combination (Mayr et al., ). Perhaps, the earlier age of English acquisition in the Flege & Eefting study plays a role, although one would expect that in Mayr et al. (), MZ was similarly exposed to English at an early age in the Netherlands; alternatively, it may be that MZ was even more proficient in English than the Dutch-English highly proficient bilinguals in the Netherlands, with even higher proficiency evidenced through assimilation effects, and lower proficiency through dissimilation effects. Nonetheless, it seems plausible that an entirely system-wide explanation cannot account for these discrepant findings, as the competing systems are the same, but yield different results.

. P    

.................................................................................................................................. Dissimilation effects have also been reported in an analysis of interpersonal variation of prosodic attrition in ten late German–English bilinguals (de Leeuw et al., ). In German, both the start and end of the prenuclear rise occur later than in English (Atterer & Ladd, ), and it was therefore examined whether German native speakers who had moved to Canada in early adulthood would produce the prenuclear rise with an earlier alignment than the German monolingual matched control group. Although most of the bilinguals evidenced assimilation in their tonal alignment, it was also reported that two females out of ten native German speakers ‘overshot’ the monolingual German norm with respect to the tonal alignment of the pre-nuclear rise. In ‘overshooting’ the German monolingual norm, the alignment at the end of the rise occurred even later in their German productions than the already late German alignment, and was thus more dissimilar from both the German norm and the English norm (similar to the segmental dissimilation effects reported in Flege & Eefting, a). This finding was not highlighted in the research, potentially because dissimilation effects are so counter-intuitive when considering the consequences of two competing linguistic systems, and may even be interpreted as outlier effects in other studies (see, e.g. high standard deviations in Flege & Hillenbrand, ; and Flege, ). However, dissimilation effects in some speakers, coupled with assimilation effects in others in this study (de Leeuw et al., ), again suggest that, at the very least, a system wide shift has a wide range of freedom, or, potentially, that competition from two language systems is also governed by both interspeaker variation and token-by-token competition.

. G        

.................................................................................................................................. In the present section, studies which have examined global regional or foreign accent in native speech are discussed. In a way, such research goes beyond the analyses of individual sounds and prosodic variables by investigating the perception of an individual’s accent, and

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



 

whether it is possible for someone to no longer be perceived to be a native speaker of his or her native language, as in the previous acoustic investigations, it was not examined whether such fine phonetic differences from monolingual native speaker norms are actually perceived to be accented by listeners. An early study into global regional accent perceptions in monolinguals reported that on average a group of Canadians from the province of Alberta, who had moved to Birmingham, Alabama, USA after  years of age, sounded more American-like than a comparable group of Canadians resident in Alberta (Munro et al., ). Some of the Canadians in Birmingham were judged to be just as American-sounding as actual American speakers, whilst others fell within the Canadian range. Although there was some indication that the amount of time spent in the US may have predicted interspeaker variation in the bidialectal speakers, with Canadians living longer in Alabama having more American-accented speech, some speakers did not follow this pattern. Other predictor variables were not investigated in this study into global regional accent, but what is clear from the findings is that the native linguistic system was permeable within the domain of phonetics at a global level, which was perceivable by L listeners. Accordingly, this research largely corroborated the findings from Evans & Iverson (), as previously discussed; however, in the latter study, it is worth emphasizing once more that one speaker dissimilated from the ambient variety, sounding more northern-like after two years in London than upon her arrival. Building on this study in the context of bilingualism, de Leeuw et al. () examined whether it was possible for native speech to become foreign accented upon acquisition of a new language in adulthood. Specifically, the objective was to determine whether native speakers of German living in either Canada or the Netherlands could be perceived to have a foreign accent in their native German speech. German monolingual listeners (n=) assessed global foreign accent of thirty-four L German speakers in Anglophone Canada, twenty-three L German speakers in the Dutch Netherlands, and five German monolinguals in Germany. The bilinguals had moved to either Canada or the Netherlands at an average age of  and had resided in their country of choice for an average of thirty-seven years. The findings revealed that the German listeners were more likely to perceive a global foreign accent in the German speech of the late bilinguals than in the monolinguals, and that, although twenty bilinguals were rated clearly to be native speakers of German, fourteen bilinguals were rated clearly to be non-native speakers of German (although they were native speakers of German). Further analysis revealed that language use had a significant effect on predicting the extent of foreign accent in the L, rather than, for example, age of arrival in their country of choice or length of residence. Those bilinguals who were expected to code-switch more often were more likely to undergo phonetic attrition—that is, be perceived as non-native speakers of their native German—than those who largely conversed in monolingual settings, where less code-switching was predicted, suggesting that monolingual modes (Grosjean, ) are likely to maintain phonetic properties of the L (de Leeuw et al., ). A similar study, which investigated foreign accent in a group of ‘L attriters’, in comparison to a group of ‘L acquirers’, largely confirmed these findings (Hopp & Schmid, ). A main goal of this study was to determine whether a group of German native speakers who had moved from Germany to Canada or the Netherlands after age , and had

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

 



used their L most frequently in daily life after having lived abroad for more than fifteen years, were rated to be less native-like in a global foreign accent rating (as in de Leeuw et al., ), than a group of late L English and L Dutch L acquirers of German. At the group level, the findings revealed that the ‘L attriters’ did not differ from the monolingual German speakers. However, of the forty ‘L attriters’, twenty-nine scored within the range of the native controls, whereas eleven scored outside this native range. The authors interpreted their findings to show ‘that acquiring a language from birth is not sufficient to guarantee nativelike pronunciation, and late acquisition does not necessarily prevent it’ (p. ). However, in exploring why some individuals were more likely to undergo phonetic attrition than others, the results were inconclusive: no independent variable significantly predicted the extent of phonetic attrition, which contrasted with the findings from de Leeuw et al. (), which suggested that language use plays. Finally, in a more recent study, it was investigated whether foreign accent ratings would be associated with acoustic measurements of vowels and consonants—that is, whether those late bilinguals who were categorized as phonetic ‘attriters’ in a foreign accent rating task would likewise exhibit phonetic measurements which deviate from the monolingual group (Bergmann et al., ). The study examined thirty-three L speakers of German, residing in the USA or Canada, where they made use of their L. Results from the ratings showed that native-likeness was negatively associated with length of residence abroad and positively associated with L use. Second, formant analyses of four speech sounds in German—/aː/, /ɛ/, /ɔ/, and /l/—showed that most acoustic differences were observed in /a:/ and /l/ whilst the other formant frequency values were not significantly different from the monolingual group. However, it was not the case that a stronger foreign accent in the late German-English bilinguals was associated with more deviant formant frequency values in these sounds, therefore suggesting that other segmental, or more likely prosodic features, were responsible for the foreign accent ratings.

. S  

.................................................................................................................................. The studies discussed in this chapter revealed a common finding that the phonetic system of the L is malleable upon competition from a new language or dialect in adulthood. Although intersubject variability with regard to the extent of phonetic attrition was evidenced, in their entirety, the findings indicate continued plasticity in native speech throughout the entire lifespan. Examining the studies together, the following ten key findings, in no order of importance, can be presented. . Research which reports phonetic attrition, as described in this chapter, combined with changes to the L phonetic system due to dialects changing over time (Harrington et al., a,b; Harrington, ), style-shifting (Bell, ; Eckert & Rickford, ; Coupland, ), and short-term rapid changes to the L phonetic system upon L training and exposure (Sancier & Fowler, ; Chang, ), lend support to the understanding that there is plasticity in native speech beyond adolescence.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi



 

. Although phonetic attrition is consistently observed in analyses of speakers who have an increased dominance in their new dialect or language, there is interpersonal variation in the extent of phonetic attrition, and some research suggests that increased L use in a monolingual mode maintains native speech pronunciation (de Leeuw, Schmid, & Mennen, ; Bergmann et al., ; de Leeuw, Tusha, et al., ). . Assimilation effects between L and L sounds have been evidenced in the VOT of plosives in phonetic attrition (Flege & Hillenbrand, ; Flege, ; Mayr et al., ), although dissimilation effects in the VOT of plosives in the L have also been reported in advanced L learners (Flege & Eefting, a). This discrepancy suggests that phonetic attrition incorporates both systematic merging (Guion, ; Chang, ; Mayr et al., ), as well as token-by-token level changes (Flege, , , ; de Leeuw, Mennen, & Scobbie, ; Ulbrich & Ordin, ). . Assimilation effects between L and L sounds have been reported through formant frequency analyses of vowels and liquids (Flege & Hillenbrand, ; Mayr et al., ; de Leeuw, Mennen, & Scobbie, ; Ulbrich & Ordin, ; de Leeuw, Tusha, et al., ), although further research is necessary to examine the precise form which intermediate tokens might take (e.g., whether the ‘merging’ becomes entirely Llike, or whether one acoustic variable—that is, F frequency—becomes more L-like, whilst another acoustic variable—that is, F frequency—remains L-like (de Leeuw, Mennen, & Scobbie, )). . Assimilation effects have been reported in prosodic analyses of L speech (Mennen, ; de Leeuw et al., ) and are suggested in global foreign accent effects (de Leeuw et al., ; Hopp & Schmid, ), but note that dissimilation effects have also been reported in prosody (de Leeuw et al., ) and in global regional accent ratings (Evans & Iverson, ). . Assimilation processes can involve intermediate ‘merging’ between the L and L categories (Flege, ), and in the case of ‘extreme phonetic attrition’ the realization of the L phoneme can occur entirely in the space of the L phoneme (de Leeuw, Tusha, et al., ). . Informal speech is more likely to evidence phonetic attrition (Shockey, ; Major, )—that is, if speech is elicited formally, the effects of phonetic attrition might not be evidenced at all. . Careful speech with a high information content appears to be less likely to elicit phonetic attrition (Shockey, ). . In the case of more than one standard variety of the L, the ambient language might not drive phonetic attrition, but rather ‘actual creative language use and communicative events’ could influence processes of phonetic attrition (Ulbrich & Ordin, , p. ), potentially suggesting that a non-ambient target variety may induce phonetic attrition. . Phonetic attrition, as it occurs in the context of new dialect (Shockey, ; Munro et al., ; Evans & Iverson, ) or new language (Flege & Hillenbrand, ; Flege, ; Major, ; Mennen, ; de Leeuw, Schmid, & Mennen, ; de Leeuw, Mennen, & Scobbie, , ; Mayr et al., ; Hopp & Schmid, ; Ulbrich & Ordin, ; Bergmann et al., ; de Leeuw, Tusha, et al., ) acquisition, has been reported at both the segmental and prosodic level of native speech, as well as in overall changes to an individual’s native accent.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

 



. D     

.................................................................................................................................. If, as this chapter suggests, the phonetic system is more vulnerable than other linguistic domains, such as that of morpho-syntax (see Köpke & Schmid, , who suggest that linguistic domains are differently affected by attrition), it is plausible that that from which all other language representations are built is destabilized as well, once phonetic attrition occurs. Phonetic attrition would thus reflect the start of an unbalancing of the native language system, which arises from the perception of phones, before distinctive phonemes, lexemes, and syntactic structure (e.g., Maye et al., ; Polka & Werker, ; Stager & Werker, ; Werker & Lalonde, ). Therefore, relevant future research questions might relate to, for example, how phonetic changes in the L potentially trickle down into other linguistic domains. Further questions of relevance relate to how regulation of the L and L phonetic systems impacts cognition more generally (de Leeuw & Bogulski, ), as well as to how the regulation of the L and L phonetic systems change over time, in different contexts in longitudinal analyses (de Leeuw, Tusha, et al., ). Questions might also focus on comparing different groups (e.g., those with brief exposure to the new language, versus those with long-term exposure), to determine whether interactional effects differ, and the extent to which accommodation effects resemble phonetic attrition. A final interpretation emerging from this chapter, but certainly not unique to it, is that the languages of a bilingual, and indeed the dialects of a monolingual, are separate linguistic systems, but not fully autonomous from one another (Caramazza et al., ; Grosjean, ; Paradis, ; Baker & Trofimovich, ; Sundara et al., ; Fowler et al., ; Hopp & Schmid, ; de Leeuw, ; Mayr et al., ). As such, phonetic attrition, as a symptom of increased cognitive load in bilinguals over monolinguals (de Leeuw, ), is simply one case in which the languages of a bilingual, or indeed the dialects of a monolingual, interact. This interaction between the competing linguistic systems differentiates those with competencies in more than one linguistic system from those who possess only one linguistic system, should such individuals exist at all (Cook, ; Grosjean, ; Hopp & Schmid, ; de Leeuw, ; Rothman & Treffers-Daller, ). Future research into phonetic attrition will surely tell us more about the plasticity of native speech and cognition more generally.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

  ......................................................................................................................

  ......................................................................................................................

 

. I

.................................................................................................................................. T pronunciation patterns of a second language (L) can influence the pronunciation patterns of a native language (L). Analysing the speech of adults who have experienced social changes in their lives, such as migration to a non-native language setting, international adoption, or even less extreme forms of contact with an L (such as full immersion in an L course), shows that the use of one’s L can be challenged by the competing demands of an L acquired later in life. Phonological knowledge is referred to in this chapter as the way humans group different pieces of phonetic knowledge together in a coherent system of relationships that is practical or useful for communication (in the sense of both speech planning and speech decoding). The relationships among entities of the phonetic domain can be of different natures, involving either mutual exclusion (e.g., when the replacement of one phoneme with another or of an intonational contour with another produces a change in the meaning of a word or a sentence) or alternations on the basis of contextual restrictions (such as when the phonetic form of a phoneme varies in pre-tonic versus post-tonic position in a predictable way). The grouping of phonetic patterns into coherent systems of relationships results in higher-level phonological entities in such a way that, for example, a group of vocalic patterns used within a given linguistic community can be referred to as the vowel system of that language or the set of consonant clusters allowed in word-internal position can be seen as one of the basic ingredients of the language’s lexical phonotactics. Understanding how changes in the phonological knowledge of attriters occur thus means understanding how the L grouping systems change as a consequence of contact with a different language. Attrition deals with transient states of linguistic knowledge. Much like in acquisition, a transient state does not automatically imply an unsystematic behaviour; in the context of L acquisition, scholars intentionally developed the notion of interlanguage to account for both variability and systematicity in learners’ productions (e.g., Selinker, ). Note, however, that the temporal dimension is a more complex issue in language attrition than in language acquisition. In fact, according to a one-dimensional

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

 



view of attrition as a phenomenon occurring across the lifespan, prototypical attriters begin as monolinguals (in their L), become bilinguals (in an additional L) at a certain point of their life (usually as adults or adolescents), and eventually become monolinguals again (in what was initially their L). Those individuals thus clearly differ from early, stable, fluent bilinguals. Other non-prototypical cases of speakers whose L is influenced by an L acquired in adulthood, such as participants in full-immersion language courses or second-generation speakers of heritage languages, also experience changing degrees of contact with the L at different epochs of their lives. A phonological account of attrition phenomena should therefore include transition (in the sense of variability over time) as a core concept in the grammar of a language. The issue is probably of particular interest for those theoretical approaches that postulate that phonological knowledge is strongly dependent on, and possibly emerging from, the way the speakers experience particular phonetic patterns and use them. These approaches are based on the idea that memorization and learning occur in terms of clusters of remembered episodes of individual experiences and that the ‘best exemplar’ or prototype of the cloud is positioned where the highest number of episodes is densest (e.g., Goldinger, ). Changes in category location and number are therefore expressions of essentially temporal developments in the process of updating this abstract knowledge from continuously varying input episodes. However, including time in phonology essentially makes phonology continuous, much like phonetics, rather than categorical, in contrast with much thought in phonological research (challenging substance-based approaches as well; Scobbie, ). Phonological attrition, as a manifestation of transient bilingualism, therefore offers great opportunities to test concurrent hypotheses on the nature of phonological representations and the relationship between phonetics and phonology. In this chapter, a selection of studies dealing with phonological change in language attrition is reviewed. First, we look at studies that enquire whether native phonological oppositions are maintained or lost when there is increased use of an L with different phonological oppositions in either the vowel system (Section ..) or the consonant system (Section ..). Then, we review studies that analyse changes in native phonotactic patterns after contact with an L (Section .). In the concluding section (Section .), possible ways of further developing the investigation of phonological L attrition are sketched.

. P 

..................................................................................................................................

.. Contrasts in the vowel systems Some attrition studies have focused on L changes in the vowel systems of speakers of declining varieties subject to the pressure of a dominant L (e.g., Guion, ; Bullock & Gerfen, ; Bullock et al., ). Bullock & Gerfen () analyse the variety of French in Frenchville, a village in Pennsylvania where the descendants of nineteenth century French immigrants still live. The study focuses on two brothers aged  and  who were born in Frenchville and are

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi





illiterate in French. Frenchville French is their native language: they spoke this language exclusively at home until their early twenties. Subsequently, as they married Englishspeaking women, their use of French was dramatically reduced over time. The study documents the loss of the quasi-phonemic contrast between mid-high and mid-low front rounded vowels /ø/ and /œ/ in Frenchville French. The contrast has a very low functional load in European French and tends to be neutralized in favour of the open-mid phoneme /œ/. In contrast to speakers of continental French, the two Frenchville speakers analysed in the study neutralize the opposition by replacing both phonemes with a rhoticized schwa (/ɚ/), which is evidently the result of contact with L English. The authors show that the reason for such replacement is not to be found in the low functional load of the opposition because other non-contrastive differences between Frenchville French and English are maintained in the speech of the informants (e.g., uvular trill as opposed to retroflex rhotics). Unlike uvular trills, the /ø/–/œ/ distinction is characterized by low psychoacoustic salience and low frequency of occurrence in the lexicon. The authors thus argue that contrastive oppositions have no special status in language attrition inasmuch as other properties, such as segmental salience and frequency, can be better predictors of maintenance or loss. The fact that frequency can override contrastivity is also evident from the tendency of attriters to enlarge the scope of application of contrasts by overgeneralizing one phoneme (usually the less marked one) to contexts in which it would not be expected, as reported, for instance, by another study on Frenchville French focusing on the /a/ and /ɑ/ contrast (Bullock et al., ). In terms of perception, the interplay between contrastivity and acoustic salience is further evidenced by Flores & Rauber (), who show that Portuguese adults who abandoned Germany and the German language during childhood are still able to discriminate pairs of vowels distinguished along two properties that are contrastive in German: duration and roundedness. The role of perceptual distinctiveness in phonological attrition is thoroughly explored by Guion () with respect to the production of the vowel systems of Quichua and Spanish by Ecuadorean bilinguals. The study compares simultaneous Spanish-Quichua bilinguals with early bilinguals who acquired Spanish between  and  years of age, mid bilinguals (between  and ) and late bilinguals (between  and ). While simultaneous bilinguals produce vowels of both languages consistently like monolingual speakers do, late bilinguals do not differentiate Spanish and Quichua vowels and seem to use the Quichua vowel system when they speak Spanish as well. The early and mid bilinguals show the most interesting pattern: most of them are like monolingual Spanish speakers in their L but differ from monolingual Quichua in their L. In fact, the early and mid bilinguals produce their native Quichua /ɪ/, /ʊ/ and /a/ higher than bilinguals who have not acquired distinct Spanish vowels. Based on fine-grained formant analysis of individual vowel changes, it is argued that a systemic raising of all Quichua vowels relative to the Spanish vowels has occurred and that this raising is the most economic and efficient way to provide the two phonological systems with a maximum of reciprocal distinctiveness. According to this interpretation, attrition is guided by maximization of contrasts in the perceptual space within one common phonological system and not by progressive assimilation to the segmental properties of the L. Similar tendencies are found by Chang () in the domain of production for American English speakers enrolled in a language course in Korea. The context of this study is different from Guion’s () study: instead of focusing on a condition of asymmetric

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

 



language contact and societal bilingualism, Chang () studies the students’ pronunciation of L vowels and stops one to five weeks after the beginning of an L Korean course in Korea. Another difference between the two approaches is that the American English subjects in Chang’s study are all late L learners. Despite these differences, both studies show that the changes in the L do not involve local assimilations of individual L segments to the closest L segments. Rather, the changes tend to encompass global shifts in one dimension (in the case of vowels, F values) for almost all segments in order to approximate the system’s spatial distribution of the L. In some specific cases, as a consequence of this global movement, an L segment may end up being farther away from, rather than closer to, the non-native (L) counterpart. Similar results, if taken in isolation, would appear counterintuitive or at least more difficult to interpret if we did not consider the big picture and look at the system’s movement altogether. This suggestion is also put forth by Mayr et al. () based on the analysis of the vowel systems of two monozygotic twin sisters who were interviewed when they were  years old. One of them (MZ) had migrated to the United Kingdom thirty-four years before the time of the study (thus adopting English as her dominant language in all private and social dimensions), whilst the other (TZ) had spent all her life in the Netherlands (though being familiar with English as an L). When speaking Dutch, MZ produces all but two monophthongs with higher F values than TZ; this effect is also attested for L vowel categories that have no counterpart in English. The direction of the change (increased openness of L vowels) can be interpreted as induced by the phonetic properties of British English (F values in female speech are overall higher in British English than in Dutch). However, if this effect was to be explained by a mechanism of token-by-token assimilation, its generalization to Dutch vowel categories that have no English counterpart would remain unexplained.

.. Contrasts among obstruents Another topic frequently investigated in attrition and bilingualism studies is the voice onset time (VOT) of obstruents in obstruent-vowel sequences. Languages differ in the number and quality of VOT duration oppositions used to differentiate among consonants of their phonological inventory (e.g., Cho & Ladefoged, ). Therefore, documenting a change in VOT realization for (pairs of) L contrastive stops can either reveal an ongoing adjustment of the parameters, according to which the contrast is implemented, or document the suppression or creation of VOT oppositions. Only the latter would in principle impact phonological grammar in the sense of increasing or decreasing the number of contrasts contained in it. However, most attrition studies have focused on the former aspect (adjustment of phonetic parameters). That is, VOT production has been mostly investigated for transient bilinguals whose L and L both possess a two-way VOT distinction (typically in stressed position). This is the case for bilinguals speaking English and French (Flege, ), English and Portuguese (Sancier & Fowler, ; Major, ), English and Japanese (Harada, ), English and Spanish (Flege & Eefting, b; Lord, ), and English and Dutch (Flege & Eefting, a; Mayr et al., ), with English being often (but not always) the L of migrants or late language learners. Whether the attriters assimilate the phonetic properties of the L to those of the L, producing intermediate VOT values (as in

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi





Flege, ; Major, ; Sancier & Fowler, ; Harada, ; Lord, ), or they exaggerate the L phonetic properties to differentiate them from the properties of the L (what is generally referred to as dissimilation or polarization in attrition studies; cf. Flege & Eefting, a,b), all these studies basically demonstrate that in adult attriters, the phonetic properties of the L can reshape those of the L. In this sense, these studies expand on the observation that the sound systems of two languages interact in a dynamic and ‘creative’ way in the brain of early bilinguals, a finding empirically demonstrated by Caramazza et al. (), again with reference to VOT production by Canadian fluent bilinguals. There is, however, at least one case in which the two contact languages differ for the number of phonological categories distinguished along the VOT continuum: English and Korean, the former featuring a two-way phonological opposition between voiced (shortlag VOT) and voiceless (long-lag VOT) stops and the latter featuring a three-way opposition between fortis (short-lag VOT), lenis (long-lag VOT), and aspirated (‘very long’-lag VOT) stops. Other phonetic parameters also contributing to the distinctions are the f contour and the voice quality of the onset of the following vowel (e.g., Cho et al., ). The early Korean-English bilinguals who migrated to the USA before the age of  in the study by Kang & Guion () were able to correctly produce five different types of stops (for VOT, voice quality and f of the vocalic onset) when speaking in both languages. This finding suggests that they kept the phonological contrasts intact in the two languages. However, those late bilinguals who learned English as adults (average age of arrival: . years) differed from Korean monolinguals in the production of both fortis and aspirated stops for at least two parameters, VOT and f of the vocalic onset; their values assimilated to those of monolingual English voiced and voiceless stops, respectively. The study by Kang & Guion () clearly shows the difference between early, fluent bilinguals and late, less proficient bilinguals. Only the latter show an effect of the L on the L, basically leading them to a merger of L and L categories and, consequently, to contrast blurring. The results of Kang & Guion () thus indirectly support the view that the degree of L attrition is correlated with the degree of L knowledge. As anticipated above in the section on vowel contrasts, Chang () deals with L English elementary learners of Korean. The study shows that these learners also modify their L stop pronunciation after short exposure to the L (one to five weeks). As an effect of the increased familiarity with Korean aspirated stops, the VOT of English voiceless stops lengthened at every place of articulation and for both male and female speakers, even though the VOT of English alveolars should not have changed for females on the basis of their distance from Korean aspirated alveolar stops. As previously mentioned for the vowel study, changes affecting natural classes of segments and conveyed by multiple phonetic cues at the same time (VOT, f, vowel formants) are to be seen as systemic rather than segmental. Whether they should also be seen as the earliest stages of a process of L attrition is to be ascertained. The study certainly suggests that for an L system to show a drift towards an L, extensive contact with the L is not needed. This finding integrates the view that L-induced changes in the L are experienced over long stretches of time by long-term immigrants or international adoptees.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

 



Further examples of dissimilation or maximization of acoustic distinctiveness are provided in the study by Dmitrieva et al. (), who investigate how Russian word-final obstruent devoicing is produced by Russian speakers with and without extensive knowledge of American English. L Russian speakers with no knowledge of English are found to maintain the distinction between underlying voiceless and voiced word-final stops by varying the duration of the closure phase and of the release, consistent with previous literature reporting incomplete neutralization in Russian. By contrast, L Russian speakers with extensive knowledge of English additionally use vowel duration and voicing into closure to distinguish between underlying voiced and voiceless stops. English lacks wordfinal devoicing but uses vowel length (and voicing into closure) to signal the voicing of the following obstruent. Thus, it appears that knowledge of L English has modified how an L phonological rule is implemented, enhancing a distinction that tends to be neutralized in the L of monolinguals. Contrast neutralizations are investigated in the domain of perception by Celata & Cancila (). Experience with allophonic variation and neutralization rules in the L affects the perception not only of the L (discrimination of allophones is poorer than of phonemes) but also of similar sounds in the L (Hume & Johnson, ; Celata, ). In the study by Celata & Cancila (), first- and second-generation Italian immigrants from the city of Lucca who were living in San Francisco were tested for their ability to maintain the singleton-geminate contrast in a word identification task (words included in frame sentences) and a non-word discrimination task (segmentation of three acoustic continua). In Lucca Italian, neutralization of the contrast selectively occurs for the rhotics (and variably for some other consonants of the inventory). The study shows that the speakers’ performance varies across tasks. A group of homeland Lucca Italian speakers still residing in Italy is tested for comparison. In word identification, the homeland Lucca Italian speakers are selectively impaired on the discrimination of short and long /r/ but not /s/ and /t/ (as predicted by their native experience with the partial neutralization rule); by contrast, first- and second-generation immigrants make the same percentage of errors in all three consonants (although the former score is on average better than the latter). In nonword discrimination, the homeland Lucca Italian speakers produce categorical classifications of the singleton versus the geminate for all three continua, with high correlation values between their performance and that of native Italian speakers from a region where the singleton-geminate contrast is not neutralized for any of the consonants. The same happens as for the classification produced by first-generation immigrants (although the correlation values are somewhat lower); by contrast, second-generation speakers produce noncategorical patterns of discrimination. The study thus shows signs of attrition in the maintenance of a partial neutralization rule, with the L neutralization of the /r/–/r:/ contrast overgeneralizing to other consonant contrasts in word identification by first-generation immigrants. However, the ability to discriminate is maintained in non-word discrimination (segmentation of acoustic continua). The finding is consistent with perceptual studies on adoptees (e.g., Hyltenstam et al., ) or returnees (e.g., Flores & Rauber, , already mentioned in the section on vowels), which show that even in adults with strongly reduced or totally interrupted contact with the native language, the ability to discriminate native phonological contrasts can be maintained in categorical discrimination tasks.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi





. P 

.................................................................................................................................. Phonotactic knowledge refers to the set of categorical rules and probabilistic constraints that regulate the form of speech, allowing or disallowing sequences of sounds or features to appear in different positions of a structure (at word edges or word-medially, within or across morphemes, etc.). Recent research has shown that in bilinguals, phonotactic constraints of both languages are actively accessed during the processing of one language (e.g., Carlson et al., ; Freeman et al., ), thus reinforcing the view that multiple sound systems are integrated and mutually influence each other in fluent bilinguals’ speech. Studies on phonological attrition and transient bilingualism apparently confirm the same view but additionally provide information on the potential changes occurring in such integration as a function of time development and the degree of bilingualism. Perceptual repair is one of the best-documented strategies exploited by speakers when they are faced with segmental sequences that do not match the phonotactic rules of their L. For instance, as a consequence of perceptual repair, speakers can hear an ‘illusory’ epenthetic vowel within consonant clusters that are illegal in their L (e.g., Dupoux et al., ; Berent et al., ). Parlato-Oliveira et al. () factor the subjects’ degree of bilingualism into the study of cross-linguistic perceptual illusion by analysing four different populations of Japanese-Brazilian Portuguese bilinguals living in São Paulo: first- and second-generation immigrants, simultaneous bilinguals, and L Portuguese learners of L Japanese. Based on the phonotactic principles of Japanese and Brazilian Portuguese, monolinguals in these two languages should perceive an illusory vowel in stimuli containing a consonant cluster, but the quality of the epenthetic vowel should differ in the two languages, most often as /u/ for Japanese listeners and /i/ for Brazilian Portuguese listeners (e.g., Dupoux et al., ). First-generation Japanese immigrants are indeed found to perceive like Japanese monolinguals (with slightly higher percentages of non-/u/ responses), whereas second-generation immigrants and simultaneous bilinguals are found to perceive like Brazilian Portuguese monolinguals (with higher percentages of non-/i/ responses). The learners of L Japanese behave differently from any other group, showing an intermediate behaviour with a relatively higher percentage of both /i/ and /u/ responses, suggesting that the learning of Japanese as an L has an impact on how speakers process the phonotactic structures of their L. In sum, first-generation Japanese immigrants show evidence of L perceptual attrition compared to Japanese monolinguals, whereas secondgeneration immigrants and Brazilians learners of L Japanese show an influence of the L on their perceptual processing of illusory vowels. Interestingly, these effects are found in an explicit identification task with word-sized stimuli but disappear in an implicit forcedchoice categorization task, suggesting that explicit perceptual tasks involving linguistic or meta-linguistic awareness are better at highlighting L attrition and L–L interference phenomena than implicit acoustic tasks are (see Celata & Cancila, , reviewed above, for a similar finding). This finding is also consistent with reported data on attrition in the detection of foreign accent (e.g., Major, ), as foreign accent detection involves holistic perception of multiple cues in naturally produced words or sentences. The study by de Leeuw et al. () on the production of /l/ by ten Albanian speakers who acquired English as an L in adolescence or adulthood also provides a window on

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

 



phonotactic competence in L attrition. The phonotactic structure of English requires a ‘clear’ or apical variant of the lateral approximant /l/ in onset position and a ‘dark’ or velarized allophone in coda position ([ɫ]). By contrast, /l/ and /ɫ/ are two distinct phonemes in Albanian, and they can occur in both onset and coda positions. Thus, the general prediction is that if the L of Albanian-English bilinguals undergoes attrition, a native phonological contrast is restructured in the form of an allophonic rule involving the alternation of two phonetic variants in a contextually predictable way. The results of the study (based on a word list reading task) confirm that such a change in the L Albanian phonology of the subjects can actually occur. In fact, in contrast to three subjects who perform like monolinguals and show no evidence of phonological attrition, five subjects show variable traces of the suppression of the native phonological contrast and its replacement by a contextual restriction predicting dark [ɫ] in the coda and clear [l] in the onset according to the requirements of the L. Finally, two other speakers show evidence of an unexpected realization of clear /l/ in coda position, even clearer than onset /l/, probably as an effect of the tendency to dissimilate or maximize phonetic distinctiveness as referred to above. From a cross-generational perspective, the study by Nodari et al. () on voiceless stop aspiration by Calabrian-English immigrants similarly suggests that when two languages share the same phonetic feature but differ in terms of the feature’s phonotactic distribution, L phonological change across generations may occur through a progressive restructuring of the phonotactic norms, potentially leaving phonetic substance unaffected. The phonetic feature under investigation here is long-lag VOT, and the cross-linguistic phonotactic difference refers to the preference for aspiration to occur in unstressed syllables in Calabrian Italian and in stressed syllables in English. Another fundamental difference between the two languages is that voiceless stop aspiration is a socio-phonetic variable in Calabrian Italian, as it indexes socio-cultural values about the speaker’s ethnic and geographical origin. In contrast to the Albanian speakers analysed in de Leeuw et al. (), first-generation Calabrian immigrants appear to maintain socio-indexical aspiration in their Calabrian speech; the subsequent generations incrementally depart from such a model, with some second-generation speakers showing that the phonotactic norms of both languages can be consistently present in their near-native Calabrian speech.

. F       

.................................................................................................................................. Some areas of phonological investigation are being extensively explored with respect to fluent bilingualism and language acquisition in general, but less emphasis is given to transient bilingualism and language attrition. For instance, a large body of research on monolinguals suggests that speech production is sensitive to the lexical and morpho-syntactic characteristics of the specific items produced, starting with their lexical frequency (e.g., Jurafsky et al., ) and moving to their neighbourhood density, phonotactic probability, morpho-syntactic status, etc. (e.g., Wright, ; Plag et al., ). In the domain of bilingualism, it is not surprising that

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi





lexical properties such as frequency, concreteness, and cognate status have a facilitative effect in L lexical and morpho-syntactic acquisition. For instance, cognates and concrete words are easier to learn and harder to forget than non-cognates and abstract words for experienced foreign language learners (e.g., de Groot & Keijzer, ); lexical access and word production are faster and more accurate (e.g., de Groot et al., ; Costa et al., ). Production of segmental and sub-segmental properties is affected by cognate status as well, as shown by Flege & Munro (), Amengual (), Mora & Nadeu (), among others. For instance, in the production of Catalan words by Catalan-Spanish bilinguals, /ɔ/ is realized as higher in cognate lexical items (such as in bosc /bɔsk/ ‘forest’, Spanish bosque) than in non-cognates (Amengual, ). Since the /ɔ/–/o/ contrast is present in Catalan and absent in Spanish, these results suggest that there may be more acoustic interference in cognates than in non-cognates. It would therefore be interesting to know how such interactive lexicon-phonetics mechanisms occur in the phonological competence of attriters, who may be exposed to (and be users of) an impoverished L lexicon and a limited variety of morpho-syntactic structures and pragmatic-communicative situations. Preliminary investigations suggest that L attrition is stronger for cognate than non-cognate words (Engstler & Goldrick, ), but future studies should further characterize cognate effects in attriters’ speech and clarify whether attriters differ from fluent stable bilinguals regarding the extent and direction of these effects. Current models of cross-language speech production and perception, such as the Perceptual Assimilation Model (e.g., Best, ; Best & Tyler, ) and the Speech Learning Model (e.g., Flege, ), assume that the degree of interference between sounds of the L and those of the L depends on the phonetic distance between them but cannot predict the relation between the acoustic characteristics of L sounds and the lexical status of the items in which the sounds are embedded. A more thorough understanding of cognate effects in the speech production of bilinguals, heritage language speakers, and attriters is therefore needed to enlarge the scope of available models of cross-language phonological interference. Another aspect that should be included in future studies on phonological attrition is socio-stylistic variation and the way in which attriters cope with this dimension of speech variation (as already suggested in Major, ). Early studies of the phonology of contracting varieties and moribund languages have shown, for instance, that contracting languages undergo phenomena of stylistic shrinkage and reduction of sociolinguistic variation, probably as a result of a decrease of the social contexts of use (e.g., Dressler, ). One consequence of such relaxation of socio-linguistic norms is the proliferation of unconstrained phonetic variation (unmotivated allophones), not attributable to either the L or the L. Recent investigations on endangered languages have confirmed this scenario both at the community level (e.g., Montoya-Abat, , for a Catalan variety) and at the individual level. For instance, heritage Mandarin speakers have been shown to produce more variable phonological targets than both native speakers and L learners (Chang & Yao, ). Moreover, when rated by native speakers of Mandarin Chinese, their production elicits uncertain demographic and sociolinguistic classifications (i.e., their speech ambiguously sounds ‘native’ or ‘non-native’ to native hearers in an apparently non-deterministic way). The extent to which attriters preserve and potentially transmit to subsequent generations features that play a socioindexical role in their L should also be ascertained in future studies. Undoubtedly,

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

 



a challenging aspect of studying stylistic variation in the context of attrition is the need for applying methods and techniques traditionally used in variationist socio-linguistics (such as social network analyses) and focusing on dialogic rather than monologic speech (to establish the role that concrete input speech plays in the development of stylistic variation in individual grammars). Similar methodological advances will in turn shift the focus from laboratory speech to large and multi-layered corpora based on naturalistic settings for data collection.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

  ......................................................................................................................

  ......................................................................................................................

 

. I

.................................................................................................................................. () S: Skol’ ko u t’ebya How many with you-sg/gen ‘How many medals do you have?’ S: U menya With me-sg/gen SR: U menya With me-sg/gen ‘I have two medals.’

dv-e two-fem. dve two-fem.

medal-ej? medals-pl/gen/fem.

medal’. medal-sg/nom/fem. medal-i. medal-sg/gen/fem.

This conversation is an example of morphological language attrition, where Speaker  (S) produces a non-target-like utterance with an omitted genitive feminine suffix on the word ‘medal’. Importantly, the speaker uses the genitive case correctly in the first noun phrase u menya (‘I have’) and marks the numeral dve ‘two’ for feminine gender. It may be concluded from this exchange that S still has a robust knowledge of Russian morphological system, but some of the case markers in this system have become less stable. The analysis of example () provides a glimpse at the types of questions for which the study of morphological attrition attempts to find solutions: • Which morphological markers remain resilient to attrition while others are lost quickly and why? • Why do some speakers produce all morphological markers in a target-like fashion while others do not? In other words, is morphological attrition highly individual or can it be generalized? • How would lack of stability of one morpheme affect other morphemes? • Can vulnerability of morphemes in one language be generalized to another language in the same typological group and/or other languages in different groups? • What linguistic and cognitive processes are involved in the attrition of morphology?

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

 



Answers to these and other questions are not easily reached in the study of first language (L) attrition of morphology as individual variation along with morphological idiosyncrasy of different languages and language groups makes it difficult to create a common theoretical paradigm that would predict and explain loss of morphological structures in L. The following overview of the state of research in this area of attritional studies reflects this difficulty.

. B

.................................................................................................................................. Studies of morphological attrition have been less numerous than studies of other aspects of language development, partially due to the fact that not all languages have inflectional morphology and those that do have a limited morphological repertoire. However, the existing studies have been successful in gaining insights into the processes of morphological attrition of the L (e.g., Polinsky, ; Bolonyai, ; Schmid, ). The ultimate purpose of explaining what types of changes morphological markers and systems undergo in bilingualism and language loss is to uncover the more general properties of morphology in linguistic knowledge. On the surface, morphological markers are fairly easy to study as speakers provide plentiful evidence of their target- and non-target-like use in spontaneous production as well as in a variety of elicitation tasks. Moreover, the omission of morphological marking is just as telling as the use of an incorrect form of a morpheme. However, morphology proves to be exceptionally difficult to analyse and generalize across languages due to its internal complexity and idiosyncratic nature (Baerman et al., ). Internal complexity of morphological markers is particularly evident in fusional languages where one affix may encode several meanings. Example () illustrates the intricate nature of the Spanish suffix -o that expresses Present tense, singular number, and first person: () Habl-o Speak-pres/sg/st person ‘I speak Spanish’

Español. Spanish

Oftentimes it is difficult if not impossible to tease out which meanings expressed by one affix may be attriting and which remain stable. The idea of idiosyncrasy of morphology refers to language specific ways of encoding various grammatical meanings. Some languages rely on more or less extensive case marking paradigms to show relationships between elements of the utterance (German, Dutch, Russian, Czech, and Finnish among many others) while others have no overt case system in place (Chinese, Arabic, Japanese); some languages have a system of grammatical gender (French, Italian, German, Polish), but others do not (English, Korean). These intricacies and inconsistences across languages complicate the study of morphology. Morphological attrition explores changes in inflectional and derivational morphemes and their combinatory possibilities (Anderson, ) in language production and perception. Researchers who investigate morphological attrition often catalogue attriting morphological features found in their data and attempt to account for the observed

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi





variations in word formation via linguistic and cognitive theoretical perspectives. Processes commonly described in morphological attrition are simplification, paradigm levelling, overgeneralization, and reduction of the number of available morphological markers. These processes that result in morphological errors are often accounted for by ‘mapping problems’ between syntax and morphology (Lardiere, ); by cross-linguistic interference where the new dominant language affects the morphological patterns of the attriting language (Gross, ; Bolonyai, ); and by retrieval problems (Paradis, ) among others. The focus of this chapter is on providing a connection between these morphological processes and the theoretical frameworks that attempt to account for them. I will first briefly review studies of attrition of derivational morphology and then discuss research on inflectional morphological loss from the perspective of a number of theoretical frameworks, including the Markedness Theory, the Regression Hypothesis (RH), the Activation Threshold Hypothesis (ATH), the Interface Hypothesis (IH), and the -M Model.

. A   

.................................................................................................................................. Multiple studies on morphology have demonstrated the importance of morphological structures in the organization of mental lexicon (e.g., Marslen-Wilson et al., ; Longtin & Meunier, ; Ghazel, ). Ullman (b) proposed a model of word processing that distinguishes between whole-word and decompositional processing routes that affect processing of some derivational affixes differently than inflectional ones. However, few studies research attrition of derivational morphemes. Dressler’s () work indicates that lack of derivational productivity in one’s L is symptomatic of language attrition. Dressler identifies two main functions of derivational morphology: lexical expansion—that is, creation of neologisms; and cognitive economy— that is, speakers’ ability to create complex word forms following morphological rules, rather than storing complex lexical items in memory. Dressler points out that while the production of neologisms may be lost, attriters are nevertheless able to process neologic creations by applying morphosemantic rules of their L. In their study of the attritional development of Michal, a fluent Hebrew speaker who was immersed in an English-speaking environment at a young age, Kaufman & Aronoff () note that once the initial ‘disintegration’ of L morphology sets in, it affects both inflectional and derivational processes. They claim that the child misuses the derivational templates of Hebrew under the influence of English after twelve to twenty-four months in the L environment, which points to her ‘loss of morphosemantic distinctions’ (p. ). They further specify that the child’s derivational system is reduced to a single form that remains productive and attests to her knowledge of the Semitic verb system. This study only focuses on the production of derivational morphology, not its processing. Intriguing distinctions between relative stability of derivational prefixes and relative volatility of inflectional suffixes used as markers of aspect in Russian have been pointed

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

 



out for heritage users of Russian (e.g., Pereltsvaig, ; Polinsky, b; Mikhaylova, ), but research in the population that experiences attrition is lacking.1 From the limited study of attrition of L derivational morphology it appears that derivational processes diminish in productivity with the progression of language loss, but remain available for recognition and processing, at least in adults. Clearly, there is a need for more studies in this area from a variety of languages and participants of different ages.

. A   

.................................................................................................................................. The vast majority of studies attempt to account for difficulties that attriters experience with inflectional morphemes and variable susceptibility of these morphemes to erosion. Many of the studies on morphological attrition focus on the idea of simplification (e.g., Seliger & Vago, a) and its forms, including reduction and levelling of morphological paradigms. Simplification predicts that a once complex system will become less so in the course of language change due to attrition. While this conceptualization reflects the observed surface changes in morphological use, it does not explain either the original complexity of the items or why there was a need to simplify them. Therefore, simplification studies often rely on the ideas of theoretical frameworks to bring about the explanatory power to attrition data. The most productive of these frameworks for morphological attrition are the Markedness Theory (Sharwood Smith, ), the RH (Jakobson, ), the ATH (Paradis, ), the IH (Sorace, ), and the -M model (Myers-Scotton, ).

. M T

.................................................................................................................................. The notion of markedness that originally stood for a binary opposition of basic versus nonbasic elements in the language (Jakobson, ) has been used widely across linguistic fields, including morphological analysis. Sharwood Smith () aptly synthesizes notions of markedness borne out of different theoretical approaches. He proposes that a form is marked if it ‘has more structure than its unmarked counterpart’ (p. ); if it ‘requires more rules or possesses more informational content than an unmarked form directly associated with it’ (p. ); if it ‘occurs less frequently in the world’s languages’ (p. ); if it ‘required evidence in the input for its adoption by the learner’ (p. ).

1 The debate about generalizability of linguistic consequences in the populations of L attriters, incomplete acquirers, and heritage speakers has been complicated and is discussed in detail in Part VI of this volume. This chapter focuses on L attrition studies as much as possible, but relies on examples from heritage language research when attrition results are not available.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi





Greenberg’s views of markedness () in terms of frequency of use and his suggestion that unmarked categories could be determined by ‘the frequency of association of things in the real world’ (p. ) are of a particular interest for attritionists. The idea of frequency of use has appealed to many researchers as the primary feature that predicts the degree of stability of a morphological marker in the attriting L. For example, Caruso () in her investigation of verbal morphology in Italian adults in Australia reports that speakers produce less marked verb tenses with fewer errors than more marked forms. Importantly, the remaining morphological markers of the unmarked tenses—Presente and Imperfetto, which are most frequent in the participants’ production—are used in a target-like fashion. Albirini & Benmamoun () find that the ‘crosslinguistically marked and infrequent category’ (p. ) of dual nouns presents a problem for attriting speakers of Egyptian and Palestinian Arabic in an English dominant environment while the formation of plural nouns is less problematic due to their less marked status. Similarly, in her study of Greek-English bilinguals, Pelc () found that marked nominal morphological features undergo regularization to the unmarked forms. Importantly this finding affects not only production tasks, but also grammaticality judgement tasks. For example, Greek-American attriters judged non-target-like gender on nouns and their corresponding articles as correct at significantly higher rates (%) than the Greek monolingual participants. The same is true for judgements of ungrammatical substitution of accusative for nominative case in Greek. In a similar vein, El Aissati () reports instances of paradigmatic levelling in marked plural formation in Moroccan Arabic. His participants regularized plural assignment and reduced the number of possible paradigms in their repertoire. Dressler () examines attrition of forms produced via suppletion versus suffixation. He finds that suffixation as a less marked process continues to be used more productively by attriters than umlaut in Breton-French bilinguals and explains it by applying the diagrammaticity parameter, which ascribes more naturalness to suffixation. As a result, less regular suppletion is replaced by unmarked suffixation. Schmid () demonstrated that German Present tense markers remain most stable among German tenses in attriters of German as L. Moreover, her study showed that the Present tense is often used with the past reference—that is, it replaces the Past tense markings. Schmid accounts for such substitutions and the stability of the Present tense in attrition by the unmarked status of the Present as it is derived from the verb stem and is not marked with any additional affixes (p. ). From these studies it is evident that morphological elements are inherently marked or unmarked in a language. Marked items are characterized by more internal complexity and less frequency of occurrence in input and production, which leads to their relative instability in L attrition that results in their simplification, regularization, and/or levelling of paradigms. Therefore, the Markedness Theory accounts for morphological attrition from the perspective of inherent features of morphemes.2 Contrary to these findings within morphological attrition, markedness studies in the domain of syntax point to more stability of a marked value in L. According to Platzack (), ‘the native speaker does not forget the marked values of his mother tongue’ (p. ). Similarly, Sharwood Smith & Van Buren () maintain that the marked values of L persist due to lack of evidence for resetting of parameters (p. ). 2

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

 



. R H

.................................................................................................................................. The RH, first formulated by Jakobson (), accounts for the order of L attrition as the inverse process of language acquisition. If supported by morphological attrition studies, the RH would specify that morphemes have a differential rate of attrition that can be predicted by the order of their acquisition in L. This, in turn, would provide additional explanatory insights into the nature of morphemes and their complexity for acquisition, maintenance, and attritional selectivity. The RH relies on the following two rules: (a) what is acquired first is lost last (Ribot’s rule; Ribot, ); and (b) what is most frequently used (regardless of the time of acquisition) will be least vulnerable to attrition (Pitres’ rule; Pitres, ) (Berko-Gleason, ). Many researchers find this hypothesis intuitively attractive as it captures morphological processes observed in L attrition (Montrul, ). However, only a few studies have tested the theory within the domain of morphological attrition in part due to methodological difficulties. As Montrul () points out, a minimum of two conditions must be met in order to test the RH: (a) the participants must be complete acquirers of their L; and (b) testing must involve at least two structures, one of which was acquired prior to the other (p. ). The results of the research that has examined morphological attrition from the regression perspective are inconclusive. Several studies that relied on the data obtained from young children (and therefore did not meet the condition (a) of complete L acquisition) found limited support to the RH (e.g. Bahrick, a; Cohen, ; Olshtain, ; Kuhberg, ). Most of these early studies maintain that morphological features that exhibit optionality during the acquisition period are more likely to be lost. For example, English irregular plural forms that are acquired late are lost early on by Hebrew-English children (Olshtain, ); in German the dative case marker remained less stable than accusative that precedes dative in the acquisition process (Kuhberg, ). On the other hand, Jordens et al. () and Schmitt () found that generally case markers remain rather robust and resistant to attrition despite their late acquisition compared to other agreement features like gender, for example. In the same vein, Håkansson’s () research on morphological attrition of L Swedish in expatriates showed no evidence that Swedish was attriting in the sequence of its acquisition. More recent studies in L attrition (e.g. Schmid, ; Keijzer, , , ), adhere to both conditions of testing the RH and provide some evidence for regression. Keijzer (, ) examined the attrition of noun and verb phrase morphology in Dutch immigrants in Canada. She determined that the vast majority of morphological features of Dutch that she investigated, including plural, diminutive, tense, and other markers, ‘revealed mirror symmetries between the attriters and acquirers at the time of testing’ (p. ). Schmid () in her study of German attriters in Canada found that markers of plural morphology that are acquired at a relatively late stage in German exhibit more vulnerability to attrition than those that are acquired early on, like gender (p. ). However, the same participants continue to use all German cases (case markings are acquired fairly late) in all contexts (with the exception of the genitive case). Schmid concludes, ‘there is no overall reduction of the case system where the dative is lost first and a binary system emerges, as the regression hypothesis predicts’ (p. ).

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi





To summarize, the RH in its application to morphological attrition mainly focuses on the order of morpheme acquisition as an explanatory tool. The results of morphological analysis in language attrition from the RH perspective are mixed, which points to a less straight-forward correlation between acquisitional and attritional processes at least in the domain of morphology and calls for further investigation of the differential stability of morphemes. Thus far, as Keijzer (a) points out, ‘the fact that evidence for the regression hypothesis was found is not enough for it to serve as an explanatory framework’ (p. ) for morphological attrition.

. A T H (ATH)

.................................................................................................................................. In the theoretical accounts of morphological L attrition discussed thus far the focus has been on inherent markedness of morphemes and the order of their acquisition in determining selectivity of loss. The ATH proposed by Paradis (, , see also Köpke & Keijzer, Chapter , this volume) shifts the focus to frequency, and recency of use, and the related inhibition of the competing items from the involved languages discussed in conjunction with the processes of memory and forgetting. The ATH approaches attrition from a neurolinguistic perspective as it predicts the item vulnerability to attrition based on the amount of neural energy necessary to retrieve it from memory. Paradis () maintains that the level of excitation—i.e., the activation threshold of an item—is regulated by the recency and frequency of its use. A higher activation threshold leads to more susceptibility to erosion. The ATH also tackles the processing mechanism involved in attrition. Specifically, it identifies inhibition as the process of ‘blocking’ competitors of the item that is to be activated (Paradis, ). Green () hypothesizes that speakers ‘ . . . exert control by activating and inhibiting tags at the lemma level’ (p. ). In other words, when the speaker chooses to speak L, the competing items from L are inhibited. Each instance of inhibition raises the activation threshold for future retrieval ‘since it takes time for the effects of prior inhibition to be overcome’ (Walters, , p. ). Conversely, when items are not inhibited their activation threshold is not affected (at least by this factor). To summarize, the ATH predicts that the less frequent of two competing forms will be more susceptible to attrition through inhibition and higher activation threshold. Attrition is the result of long-term lack of stimulation. Intensive use/exposure to one of the languages in a bilingual environment leads to a lower activation threshold for that language (i.e. it requires fewer resources), even in early, fluent, behaviourally balanced bilinguals. (Paradis, , p. )

Relatively sparse research in L attrition of morphology from the position of the ATH attempts to identify specific language forms that are more susceptible to attrition based on frequency of activation and inhibition. The results of the existing studies are inconclusive. For example, the findings of Gürel (a, ) and Schmitt () support the ATH, whereas Köpke () and Schmid (, ) report results that are contrary to the ATH predictions. Gürel’s studies (a, ) rely on the ATH to examine the vulnerability of Turkish pronominals o ‘s/he’, kendisi ‘self ’, and pro to attrition determined by their activation threshold relative to the corresponding L properties that they compete with (Gürel, a,

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

 



p. ). The only pronoun that has a competing form in English is o ‘s/he’, whereas the reflexive and pro do not compete with English. From Gürel’s analysis it appears that Turkish o is most affected by attrition since the frequent use of English s/he which overlaps, and is congruent with, Turkish o necessitates its inhibition. Consequently, the Turkish pronoun becomes less accessible and is more likely to undergo erosion. On the other hand, since there is no congruency between Turkish kendisi and pro and English pronouns, the competition between them is low, Turkish lemmas are not inhibited and their activation threshold is not affected by inhibition (but continues to be affected through non-use). This results in their relative stability in the speaker’s attriting language. Schmitt () also found supportive evidence to the ATH in her study where she observed the loss of Russian content morphemes and inflectional morphology. The ATH is helpful in explaining decreased vulnerability of Russian content morphemes related to everyday domestic topics on the basis of high frequency of their use, lack of competition with English (since all the participants communicate about those topics only in Russian) and consequently, their low activation threshold. For inflectional case markers, it was determined that the Nominative case which is most frequently used in Russian is least affected by attrition. The Instrumental case shows both the smallest number of the required contexts—that is, it is used least frequently among the six Russian cases—and the highest degree of attrition due to elevated activation threshold. The study concludes that ‘the number of the required contexts and the number of target-like items in production exhibit an almost one-to-one correlation’ (p. ) thus supporting the premise of the ATH. Additionally, Iverson’s data () indicate that his participant Pablo experienced attrition only in the areas where Spanish and Brazilian Portuguese did not overlap. This conclusion further supports the ATH, though there is no specific reference to morphological markers in the study. However, the ATH does not find ubiquitous support in all morphological attrition research. Schmid () suggests that frequency and recency of use do not have as salient an impact on activation thresholds and item accessibility as does inhibition. The results of her investigation of attriters of German in two different contexts indicate that there is little or no correlation between the level of attrition and frequency of L use. Schmid proposes that the quantity of L use is an important predictor of L maintenance only until a ‘saturation point of rehearsal’ (p. ) is achieved and knowledge is stabilized. After this point, frequency may not play as important a role in language accessibility as previously assumed. It is the process of inhibition that will affect the activation threshold the most. In the same line of research, Altenberg () studied the attrition of German gender and determined that word frequency played an important role in the maintenance of gender distinctions (her study, however, was not based on the ATH per se, but did invoke the key component of the ATH—frequency of use). Schmid’s () data of German gender attrition did not confirm the predictions of gender vulnerability based on frequency of use and reinforcement (p. ). Overall, morphological studies grounded in the ATH provide further insights into the connection between frequency of use and inhibition of morphemes in the attriting language and attempt to explain cognitive processes that underlie morphological decline. However, many questions remain unanswered. For instance, what is the level of frequency of use that is needed to avoid an increase in the activation threshold of an item? What features of a morpheme in fusional languages are most likely to compete across languages?

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi





At what point does inhibition trigger the onset of morphological optionality? Why attriters continue to comprehend their L morphology, when they are no longer able to produce it due to high activation threshold? New studies may help find solutions to these issues and shed more light on cross-linguistic mechanisms of morphological accessibility.

. I H (IH)

.................................................................................................................................. The IH originally developed to account for non-target-like usage in adult second language acquisition (Sorace, , ) and fruitfully applied in heritage language and L attrition studies (e.g., Tsimpli, ; Montrul, ) moves away from the earlier discussed concepts of markedness, order of acquisition, and frequency and inhibition (see also Chamorro & Sorace, Chapter , this volume). Rather, the IH accounts for the differential susceptibility of items to erosion by the types of linguistic interfaced involved in their production. It predicts that linguistic properties that are dependent on interfaces between language and external domains of cognition are more vulnerable to loss than those that are internal to the linguistic system. The strong version of the IH (Sorace & Filiaci, ; Sorace, ) identifies external interfaces as language modules that operate ‘between syntax and cognitive systems outside of formal grammar’ (Van Osch et al., , p. ). External interfaces are claimed to put higher processing demands on bilinguals due to the need to pull information from two or more systems, for example discourse and syntax, or syntax and pragmatics. This puts language items that operate in the external interfaces at a greater risk of attrition (Montrul, a; Sorace, ). On the other hand, internal interfaces are language modules that operate within linguistic domains such as syntax and semantics, syntax and morphology, and morphology and semantics (Montrul, a; Sorace, ; Iverson, ). Since linguistic items do not have to ‘travel’ outside of their own ‘territory’, they bear fewer processing costs and are predicted to be more stable in bilingual production (Pires & Rothman, ; Sorace & Serratrice, ). To date, only several studies focus on attrition of L morphology from the perspective of the IH. Van Osch et al. () examine gender agreement in Spanish-Dutch bilinguals comparing the difference in the level of weakening of gender stability at the internal interface (adjectival predication) and the external interface (pronominal reference). The results show that both Spanish heritage speakers and attriters in the Netherlands make significantly more gender agreement errors in pronominal reference than in the adjectival predication. The authors account for this difference from the perspective of the IH: adjectival predication—that is, subject–predicate agreement—takes place within the internal interface of syntax and morphology, and remains more stable than the pronominal referencing that involves discourse considerations and takes place at the external interface. This study shows that it is not an individual item from L that becomes vulnerable in attrition, rather it is the context and the amount of processing determined by the interface that account for optionality of linguistic items in an attriting language. A recent study by Chamorro, Sturt, & Sorace () explores the effects of L attrition on Spanish personal preposition a, the use of which does not depend on context, but is determined by the animacy and/or specificity of the direct object. Importantly, the use of the preposition is governed by the interaction of syntactic and semantic properties—that is, internal interface. The results of the online processing and offline interpretation

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

 



experiments clearly show that this structure has not undergone L attrition in either condition. This result yields strong support to the IH. Other research in L attrition proper deals with syntactic, not morphological changes and supports the predictions of the IH (e.g., Chamorro, Sturt, & Sorace, ). Several heritage language studies also focus on the explanatory power of the IH for morpho-syntax. Their results are not as uniformly supportive of the IH as the attrition research (see, e.g., Montrul, a, ). Clearly, more studies of the attrition of morphology in L that are grounded in the IH would help establish the prognostic and explanatory value of the IH. Current research in this area does not account for the widely attested significant decreases in morphological stability of the attriting L (e.g., Gross, ) in various populations despite the expectation that internal interfaces are less vulnerable to attrition. Thus research that can help explain morphological selectivity in attrition from the IH perspective would look at the following dimensions of morpheme behaviour: morpho-syntactic or internal interface (structural case markers); morpho-semantic internal interface (gender markers, lexical case markers); morphemic-discourse external interface (gender-dependent references, aspect); and morphosyntactic–discourse interface (fusional markers that include case, gender, number, mood).

. T -M M

.................................................................................................................................. Unlike previously discussed theoretical frameworks that can and have been applied to various domains of linguistic knowledge, the -M model was developed with the purpose of capturing morphology in bilingual behaviour. The -M model focuses on the differential distribution and contribution of four types of morphemes—content and early system morphemes that carry the speaker’s intentions, and two types of late system morphemes—bridge and outsiders which encode grammatical relations (Myers-Scotton, ; Myers-Scotton & Jake, ) as illustrated in example (): ()

She plays with dolls of her friend.

Content morphemes in () are conceptually activated and are responsible for conveying semantic and/or pragmatic meaning: she, play, her, friend, and doll. Early system morphemes are selected and conceptually activated by their content morpheme heads to make the meaning of the utterance more specific and to ‘contribute to the realization of semantic and pragmatic intentions of the utterance’ (Myers-Scotton & Jake, , p. ). In () the plural marker -s is an example of an early system morpheme. Late system morphemes (bridge and outsiders) are structurally assigned and are used to build larger units. Bridge morphemes connect phrases or clauses to indicate that they are related. In () of is a bridge morpheme. Finally, ‘outsiders’ look outside of their immediate projection to get assigned. They determine structural relations between constituents and include markers of agreement, case, etc. In () the third person singular marker on the verb -s is an outsider morpheme. Importantly, the distinction between conceptual and structural activation of morphemes (Myers-Scotton & Jake, ) provides an insight as to why they behave differently in attrition and further explains why language items may belong to different types of morphemes in different languages even if they come from the same lexical

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi





category. This is exemplified by personal and dummy pronouns in English (e.g., she vs. it/ there). Content personal pronouns are conceptually activated by the speaker in the production process. They are acquired early on in first and second languages and are used in a target-like fashion more often than the structurally assigned morphemes, such as dummy pronouns that are structurally activated (Li, ). The framework created by the -M model has been used in several studies of L attrition. Gross () finds that in German-English attrition ‘content morphemes are most vulnerable to attrition while late system morphemes are the least likely category to undergo attrition’ (p. ). Schmitt (, ) studies the loss of morphological case in RussianEnglish attriters and reports that case markers ‘do not erode, rather it is the contexts where these markers are required become less clear to the informants’ (p. ). Comparing late system morphemes to early system morphemes, Bolonyai () finds that in L attrition of Hungarian late system morphemes such as syntactic case markers like accusative are less vulnerable to loss than early system morphemes like local semantically assigned cases. Fuller () reports that early system morphemes (e.g., English plural markers) appear before late system morphemes (e.g., English third person singular marker) in convergence from the attriting to the dominant language. Thus, there seems to be a growing amount of evidence that structurally assigned morphemes demonstrate less vulnerability to attrition than lexically activated morphemes. In addition, the stage and complexity of activation of morphemes at the level of abstract lexical structure (based on Levelt, ) can help explain why certain pieces of fusional morphology may be lost before others. Bolonyai & Dutkova-Cope () examined differential susceptibility of agreement features, such as person, number, and gender all marked by a single fusional inflection in Hungarian-English and Czech-English attriters. They found that person and number—the late system outsider morphemes in the -M Model classification—remain particularly robust in both languages. Gender, identified as an early system morpheme appears to be significantly more vulnerable to attrition in their participants. The researchers account for this distinction through both cross-linguistic incongruence and the complexity of computational procedures that are involved in the production of these morphemes. It appears that person and number ‘mark the most elementary and universal predicate-argument relations in all three languages that involve the least complex computational procedures and whose access does not interact with semantics-pragmatics’ (p. ) during activation. In other words, as late system morphemes, number and person involve structural activation (Myers-Scotton, ), or in terms of the IH discussed in the previous section, they function at the internal interface. On the other hand, the activation of L-specific gender features evokes more complex processes. Specifically, as an early system morpheme on the noun, the gender feature is first conceptually activated. In order to carry out the agreement function between the subject and the verb, it has to be also structurally assigned as an outsider morpheme on the verb. Thus, gender agreement involves a complex interaction between the two types of activation—conceptual and structural. This complexity along with the uniqueness of gender features and lack of gender agreement congruity between the attriting languages and English explains the instability of gender in Hungarian and Czech as compared to person and number. Just like with the other models discussed here, studies that employ the -M model in the investigation of language loss are sparse. However, the results of the existing research seem to have an explanatory promise. It seems that there is some consistency in identifying

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

 



structurally activated outsiders as more stable in language attrition. It would be of interest to determine if there is a difference in the order of acquisition of structurally and conceptually assigned morphemes. If such a difference were found, it would be compelling to compare the order of acquisition to the order of attrition based on the type of activation and compare those results to the predictions of the RH. Moreover, the distinction between conceptual and structural activation of the morphemes that seems to play a role in their vulnerability to attrition could be correlated to the types of interfaces involved in their production as discussed in the IH. In addition, since the -M model is the only framework that focuses specifically on bilingual morphology, it is essential to continue investigating how different types of morphemes fair in language loss of a variety of languages in different populations.

. C

.................................................................................................................................. This review of studies on morphological aspects of language attrition indicates two main directions in accounting for differential loss of L morphemes: a. Research focused on inherent properties of morphemes: • More marked morphemes are more vulnerable to attrition, whereas less marked morphemes remain more stable in the attriting L1 (the Markedness Theory) • Structurally activated morphemes are less susceptible to attrition (the 4-M model) b. Research on mechanisms that underlie the processing of morphemes: • Frequency of use and inhibition play a significant role in morphological loss (the ATH hypothesis) • Complexity of morpheme access determines their susceptibility to loss (the IH and the 4-M model) Inherent properties of morphemes may vary from language to language. They have a predictive, but not explanatory power. Mechanisms of morpheme processing, while still minimally understood, have the potential to explain why certain morphological properties lead to more or less vulnerability in attrition and consequently to a better understanding of how linguistic systems interact in both acquisitional and attritional processes. Challenges that morphological studies face include: a. Inconsistency in the obtained results within and across language data sets in all theoretical frameworks discussed above. Examination of the same data sets from the perspectives of different theories as well as collection of more morphologically focused attrition data are ways to enhance current limited understanding of morphology in bilingualism. b. Lack of uniform data collection techniques and methodologies, which makes the comparison of the results exceedingly difficult. Much of the data are collected through free speech production, some data are gathered using online grammaticality judgement tasks, others rely on offline tasks, with or without electroencephalogram (EEG). Schmid (a) provides an exhaustive list of all available and tested data collection

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi





methods. For comparative and replication purposes it would be ideal for researchers to rely on these methodologies and use more than one in order to triangulate their results and design replicable studies. c. Lack of a principled agreement on the types of speakers that should be investigated. Data are sometimes collected from different categories of speakers—immigrants and heritage speakers of different ages. This is a complex issue that has been addressed in this volume. What is particularly relevant to issues of morphology is that morphological incomplete acquisition should not be discarded as aspects of morphology that are not acquired by heritage speakers may overlap with the areas of attrition. In this case morphological vulnerability could be understood and explained from a multidimensional perspective. Morphological attrition seems to have different rates among children and adults. Findings in this area could help clarify the connection between the rate of morphological acquisition and erosion. d. Limited number of languages that are researched in the context of attrition. The majority of data come from Romance, Germanic, and Slavic languages. There are few studies on Arabic, Turkish, Hungarian, and Hebrew morphological attrition and virtually no research that involves languages of other typological groups. e. Only one of the five theoretical frameworks is specifically designed for the investigation of morphology. Additional morphology-centred models, including the Distributed Morphology models have a potential for more explanatory and predictive views of attritional processes. The Distributed Morphology model (Halle & Marantz, ) provides an account of lexical primitives combining into complex structures and the operations that apply during the post-syntactic computation to the phonetic form. A particularly useful feature of the Distributed Morphology theory for attrition studies is its hierarchical view of morphological operations that follow syntactic processes. This hierarchical view of morpheme selection and its dependency on syntax rather than lexicon may promote innovative explanations to differential vulnerability of gender, plurality, and case markers, for example. Thus far I have not been able to find any work in L attrition from the perspective of Distributed Morphology model. However, pioneering work in childhood bilingualism (Austin, ) and in code-mixing (Alexiadou, Anagnostopoulou, & Schäfer, ) have laid some foundation for future research in L attrition. This review of research on morphological L attrition has demonstrated that there is no one theory that accounts for the complexity and variety of morphological operations and morphemes that get restructured, modified, and ultimately lost exhibiting varying vulnerability to language attrition. Therefore, in addition to streamlining data collection processes, it is essential to apply multiple frameworks to the observable morphological changes in order to come closer to an understanding and explanation of morphological operations.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

  ......................................................................................................................

  ......................................................................................................................

 

. W         ?

.................................................................................................................................. L attrition has often been described as a process that leads to a person’s loss of the ability to retrieve words from memory (Stolberg & Münch, ; Isurin, ), if not the complete erasure of those words from the person’s mental lexicon (Pallier et al., ). Although attrition is often equated with loss, there is nevertheless a growing tendency in the literature to identify loss as only one of many possible manifestations of attrition. According to Meara (), ‘vocabulary loss is an observable change in the number of activated words in a vocabulary, and will always be measurable’ (p. ), whereas attrition involves small structural changes that weaken a person’s mental lexicon—and can eventually result in loss—but are not always observable in the person’s language performance. First language (L) lexical attrition as an observable phenomenon is now widely understood as a collection of undesirable changes to a person’s L lexical skills that tend to accompany cases of inter alia international adoption, emigration to a new language environment, reduced use of or isolation from the L, second language (L) acquisition, bilingualism, multilingualism, and various pathological conditions that affect word use (Jaespart & Kroon, , p. ; Goral et al., ; Bardovi-Harlig & Stringer, ; Schmid & Jarvis, ). Vocabulary loss is of course the most extreme manifestation of lexical attrition, and some of the clearest examples of this in the realm of L attrition pertain to cases where a person begins acquiring one language in infancy but is adopted during early childhood into a home and community where a different language is spoken, and where he or she is completely isolated from the L. Isurin & Seidel () examined the case of a woman referred to as TJ, who was born in the United States to a mother who is believed to have communicated with TJ in Russian until she was  years old. At that point, TJ was taken from her mother and eventually adopted by a family who raised her and spoke to her exclusively in American English. After being taken from her mother, TJ quickly lost her L and soon even became unaware that she had spoken a different language prior to her adoption. This changed when she began treatment for post-traumatic stress disorder at the age of . During her therapy, she was able to remember previously forgotten details about her childhood,

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi





including a handful of foreign-sounding words whose meanings she could not remember. Some of the words were identified by the researchers as being Russian (e.g., angliiski ‘English,’ boisja ‘afraid’, podruzhka ‘friend’), and the rest were also phonologically compatible with Russian, but their forms had apparently become distorted in TJ’s memory. Evidence that TJ had known Russian in her early childhood was found in the results of a study where she was retaught Russian words that she would most likely have known at the age of  (e.g., korobka ‘box’, ložka ‘spoon’, igraj ‘play’, kušaj ‘eat’). TJ showed a clear advantage in learning these words versus other Russian words that she probably would not have known at such a young age (e.g., kletka ‘cage’, korabl’ ‘ship’, čihaj ‘sneeze’, klej ‘paste’); a comparison group of native English speakers did not show any advantage in learning the former versus the latter set of words. Other studies have also documented cases of what at the level of observation have appeared to represent total or near-total loss (including the loss of comprehension) of a language acquired in childhood; such studies have usually involved international adoptees—some as young as  months (Nicoladis & Grabois, ), and some as old as  years (Isurin, ) (for an overview of such studies, see BardoviHarlig & Stringer, ; Schmid & Jarvis, ; Isurin & Seidel, ). Besides a complete loss of the ability to produce and, in extreme cases, even to be able to comprehend words that the person previously knew, lexical attrition can in milder cases result in word-retrieval difficulties, where attriters usually do eventually recall the words they need for successful communication and task completion, but many of these words are retrieved from memory only after an effortful and relatively time-consuming mental search. According to Ecke (), ‘retrieval slowdown and failure (mostly reported for lexical production, but also observed in reception) are among the first and strongest indicators of language attrition’ (p. ). Such problems are clearly a matter of access rather than of the loss of memory representations from the person’s mental lexicon, and in language production they result in various forms of disfluency, including the frequent use of either filled or empty pauses (e.g., Das Mädchen wird ah freigelassen ‘The girl is uh released’; und # dann hat er # bisschen (he)rum gekuckt ‘and # then he # looked around a bit’; Schmid & Beers Fägersten, , p. ), repetitions (e.g., als de als de vrachtwagenchauffeur weer naar buiten komt ‘when the when the truck driver comes back out’; Schmid & Beers Fägersten, , p. ), self-corrections (e.g., hat den Brief gekriegt hat den Brief bekommen ‘got the letter received the letter’; Schmid & Beers Fägersten, , p. ), lexical substitutions (e.g., referring to a cliff as a precipice, hill, or mountain; Olshtain & Barzilay, , p. ), circumlocution (e.g., saying The dog goes into this little river, body of water, whatever it is, but it’s very shallow, and they’re safe when referring to a pond; Olshtain & Barzilay, , p. ), or even borrowing from or switching into another language (e.g., ich hab hier eine Bekannte die war sehr sehr ## involved ‘I have a friend here who was very very ## involved’; Schmid, a, p. ). Importantly, as this last example shows, multiple forms of disfluency associated with lexical attrition sometimes occur in tandem. All examples of disfluency in the preceding paragraph come from the freestyle L oral production of attriters who grew up in an L environment and then later emigrated to another country, where they acquired the dominant language of that country and used it in their daily lives for many years. Even though disfluencies such as hesitations, pauses, self-corrections, circumlocutions, and so forth, are common even among L speakers who are not experiencing lexical attrition, they are significantly and sometimes considerably

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

 



more prevalent in the speech of emigrants than they are in the speech of people who have remained in their home countries (Schmid & Beers Fägersten, ). I will discuss factors related to emigration and the reduced use of the L in Section .. In the meantime, it is important to note that disfluency-related effects of L lexical attrition have also been examined and confirmed in attriters’ reaction times in experimental tasks involving both lexical production (e.g., picture naming, verbal fluency) and reception (e.g., word recognition, picture–word matching, lexical decision). These methods will be described in Section .. For now, it is relevant to point out that many studies that have used these methods have indeed confirmed that attriters tend to recognize and retrieve L words more slowly than do comparable speakers of the same L whose life circumstances make them less susceptible to L attrition (see, e.g., Goral et al., ; Schmid & Jarvis, ; Isurin & Seidel, ). In addition to its effects on fluency, lexical attrition also affects lexical accuracy. Lexical accuracy can be measured in relation to the accurate oral or written production of word forms, the use or comprehension of words in a manner that indicates that the person has an accurate understanding of the meanings (including connotations and usage constraints) of those words, and the accurate production and comprehension of the syntagmatic relationships that can be found in multi-word lexical items. Examples of L lexical attrition affecting form accuracy include the earlier mentioned example of TJ’s oral production of Russian words in a way that made them unrecognizable to the researchers (Isurin & Seidel, , p. ) and cases where an L word becomes partially or fully integrated into an attriter’s use of the L, sometimes without the attriter even being aware of the cross-linguistic influence (e.g., a German-English bilingual saying also hier der Benjamin der hat die Marilyn *angegrowlt ‘imagine Benjamin [a dog] growled at Marilyn’; Schmid, a, p. ). Problems with form accuracy appear to be relatively rare in L lexical attrition, whereas semantic and syntagmatic problems are more common. Examples of lexicosemantic errors include both semantic extensions (e.g., a Finnish-English bilingual saying kattoo ‘look, watch’ instead of vahtia ‘watch, keep an eye on’ when referring to watching/baby-sitting children; Jarvis, , p. ) and loan translations (e.g., a German-English bilingual saying Vergiß es instead of laß es bleiben, lit. ‘let it stay’ when expressing the meaning of ‘forget it’; Seliger & Vago, a, p. ). Loan translations involve not only semantic but also syntagmatic relationships; syntagmatic problems among attriters include problems with compounds, collocations, idioms, and clichés (e.g., a Finnish-English bilingual saying *ottaa suihku ‘take a shower’ instead of the conventional käydä suihkussa ‘visit the shower’; Jarvis, , p. ). Bardovi-Harlig & Stringer () have pointed out that it is important not to neglect multi-word units or conventional expressions when investigating lexical attrition because there is strong evidence that they are stored—like words—in the mental lexicon, and because they reflect the complex networks that link lexical items in memory (pp. –). Beyond fluency and accuracy, lexical attrition can also be observed in relation to lexical complexity. Bulté & Housen () have defined complexity as a property or quality of a phenomenon or entity in terms of () the number and the nature of the discrete components that the entity consists of, and () the number and the nature of the relationships between the constituent components. (p. )

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi





Although lexical complexity can be understood in terms of the properties of individual words—or what Bulté & Housen () call ‘the compositionality of lexical elements, that is, the number of formal and semantic components of lexical items’ (p. )—the tendency is to view lexical complexity in relation to the complexity of a person’s mental lexicon or lexical skills, as demonstrated by the nature and variety of words the person produces in samples of speech or writing. According to Bulté & Housen, lexical complexity at this more abstract level can be operationalized with measures that tap into the lexical diversity, lexical density, and lexical sophistication of texts. Lexical diversity refers to the variety of words found in a text (or sample of speech), lexical density refers to the proportion of lexical (or content) words in the text, and lexical sophistication refers to the proportion of rare or less frequent words in the text. Of the three abstract types of lexical complexity, lexical diversity has received by far the most attention in studies of L attrition, whereas lexical density appears not to have been investigated so far. Although some studies of L lexical attrition purport to assess lexical density (Negrisanu, ; Taura & Nakanan, ), the measures they use are indices associated with diversity (e.g., the type–token ratio) rather than measures of the prevalence of content words in a text. The findings of these and several other studies have shown, as expected, that L attrition does indeed lead to a decrease in the levels of lexical diversity that attriters produce in their speech or writing (for a summary of these studies, see Schmid & Jarvis, ). However, lexical diversity has nearly always been investigated using unidimensional measures of lexical repetition rather than multi-dimensional measures of redundancy and variety (cf. Jarvis, a). L attrition research that has investigated lexical diversity has relied almost exclusively on measures that also vary as a function of text length and evenness (i.e., how evenly the lexical tokens in a text are spread across lexical types) (see Schmid & Jarvis, ). The study by Schmid & Jarvis () was designed to overcome these shortcomings by using a combination of diversity indices, such as a measure of lexical repetition that does not vary with text length (i.e., the Measure of Textual Lexical Diversity, or MTLD), a measure of evenness, multiple measures of lexical rarity, and a measure of lexical dispersion (i.e., how evenly the tokens of a given type are dispersed throughout the text instead of clustering near each other). The results of a MANOVA showed that the German-Dutch and German-English attriters in the study did indeed have significantly lower levels of lexical diversity than the German control group. The results of an additional Linear Discriminant Analysis showed that the three groups of participants could be distinguished from each other with % accuracy based on a statistical model made up of eleven variables—ten of which were lexical diversity measures. (The remaining variable in the model was speech rate, measured as words per minute.) Some of the lexical diversity variables in the Schmid & Jarvis () study were measures of lexical rarity—or the relative frequencies (in German) of the words used by the different groups of participants. Jarvis (, a,b) has argued that lexical rarity is one of the components of lexical diversity and should be included in any complete, multi-dimensional model of this construct. Other researchers have treated lexical rarity as a separate construct, and have referred to it with terms such as lexical sophistication (e.g., Linnarud, ) or a text’s lexical frequency profile (Laufer & Nation, ). The amount of research available on the L attrition of lexical sophistication is quite sparse, but the findings of Schmid (a, pp. –) and Schmid & Jarvis () show that L attrition does indeed appear to lead to

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

 



an increased reliance on highly frequent words in the L (e.g., poor vs. destitute), which in turn tends to coincide with decreased levels of lexical diversity.

. E   

.................................................................................................................................. Comprehensive overviews of the hypotheses developed to account for attrition can be found, among other places, in Schmid (), Köpke & Schmid (), Bardovi-Harlig & Stringer (), Schmid & Köpke (a), and in the first several chapters of this volume. I will not attempt to duplicate any of these excellent overviews here, but will instead briefly discuss which of these hypotheses have gained currency in relation to L lexical attrition, and which seem compatible with the patterns of lexical attrition described in the preceding section. Bardovi-Harlig & Stringer () outlined six hypotheses that have been proposed to account for L attrition. (These are the same hypotheses discussed by Schmid, , and Köpke & Schmid, , although they are labelled and described somewhat differently.) Of these six hypotheses, only three appear to be directly relevant to the patterns of L lexical attrition found in the literature and illustrated in the preceding section. These include—in the order I will discuss them—the dormant language hypothesis, the threshold hypothesis, and the interference hypothesis. Importantly, these hypotheses complement rather than compete against one another. The dormant language hypothesis (also called the savings paradigm or the residual memory approach) holds that attriters never completely lose their languages nor any part of them; rather, the parts that seem to have become lost are only dormant in the attriters’ minds and have the potential to be reactivated through interventions such as hypnosis or relearning (Isurin & Seidel, ). The example given earlier of TJ’s relearning of Russian words that she is likely to have known in early childhood does indeed seem to confirm that TJ retained traces of Russian words in memory even after thirty years of isolation from her L and even after living for nearly as long being unaware that she had ever spoken Russian. To date, Isurin & Seidel appear to be the only researchers who have tested the dormant language hypothesis in relation to L lexical attrition. Most of the evidence for this hypothesis comes from studies of L phonological (e.g., Park, ) or morpho-syntactic attrition (Au et al., ), or from studies of various types of L attrition, including L lexical attrition (e.g., Van der Hoeven & de Bot, ). Empirical evidence that is consistent with the dormant language hypothesis is also largely consistent with the threshold hypothesis—the most well-known version of which is the Activation Threshold Hypothesis (ATH). This hypothesis predicts that the mental representation of a word or other item of language requires a certain amount of neural impulses in order to reach activation (its activation threshold). Each time a language item . . . is used, its activation threshold is lowered, making it easier to activate again, but it slowly rises when inactive. (Paradis, , p. )

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi





Like the dormant language hypothesis, the threshold hypothesis holds that memory representations for words and other features of language are not completely lost during attrition; rather, they simply become more difficult to access and retrieve after long periods of disuse, especially when they compete for activation with frequently and recently used L words. It is clear from the recent literature that the ATH is the most popular and the most empirically validated explanation of both L and L lexical attrition (see Bardovi-Harlig & Stringer, ). It also nicely accounts for a wide range of lexical attrition phenomena. For example, it accounts for both retrieval slowdown and failure (cf. Ecke, ), which in turn account for several of the types of disfluency described in the preceding section: (a) retrieval slowdown ➔ filled pauses, empty pauses, repetitions, self-corrections; (b) retrieval failure ➔ lexical substitutions, circumlocution, language borrowing, and switching. As pointed out by Schmid & Jarvis (), activation problems associated with lexical attrition can also result in reduced levels of lexical diversity and lexical sophistication. Both findings are predictable on the basis of the ATH: lexical diversity will decrease as more and more words become difficult or impossible for the attriter to retrieve from memory, and lexical sophistication will also decrease because the activation thresholds for individual words will be determined by how frequently (and recently) they have been encountered (see Paradis,  and also the discussion in Köpke & Keijzer, Chapter , this volume). The ATH overlaps not only with the dormant language hypothesis, but also with the interference hypothesis, which ‘holds that attrition is directly due to the increasing influence of the newly dominant, competing language’ (Bardovi-Harlig & Stringer, , p. ). Different versions of this hypothesis highlight different ways in which the attriter’s developing L knowledge can affect his or her use of the L; the ATH includes one of its own versions of the effects of the L: ‘the most frequently used elements of L will tend to replace their (less used) L counterparts’ (Paradis, , p. ). Such effects of the L on L attrition are sometimes completely transparent, such as when an attriter unintentionally embeds an L lexical root into an L construction, or when the attriter uses L words during L communication without realizing that those L words do not exist in the L (e.g., Schmid, a, p. ). At other times, the effects of the L on L lexical attrition are so subtle that their detection requires experiments designed specifically for this purpose. For example, Segalowitz () used a primed lexical-decision task to show that bilinguals with higher levels of L proficiency access (or recognize) L words more slowly than comparable bilinguals who are less proficient in the L. Consistent with the ATH, the explanation for these results appears to be that bilinguals with higher levels of L proficiency, over time, have exerted higher levels of inhibition on the L lexicon, and this has led to higher activation thresholds for individual L words. According to Schmid & Jarvis (), ‘Each time a speaker selects a target item from the lexicon, its competitors . . . must be inhibited. This inhibition process raises the activation threshold, making the inhibited items harder to access subsequently’ (p. ) (see also Köpke & Keijzer, Chapter , this volume). Most of the types of L lexical attrition discussed so far in this section involve attrition in the domains of fluency and complexity, but it is important to note that the three overlapping hypotheses are also relevant to the domain of lexical accuracy. As described in the previous section, many of the lexical errors (including errors involving multi-word units) found in attriters’ use of the L are clearly the result of L influence (e.g., Jarvis, ). Others, such as TJ’s inaccurate recall of the pronunciation of some Russian words, however,

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

 



seem to reflect distortions to memory representations. Alternatively, they may reflect reduced—or partial—access to the phonological, orthographic, semantic, syntactic, or lexical network-related mental representations for words. This last possibility might best be explained by a combination of the dormant language hypothesis and the ATH. The effects of attrition on L lexical accuracy are certainly an area that requires further investigation, although it is important to reiterate that lexical errors not caused by L influence appear to be relatively rare in L attrition.

. F    

.................................................................................................................................. Some of the factors that have been found to affect L lexical attrition are either explicitly stated or are implicit in the assumptions underlying the hypotheses described in the preceding section. Some of these factors are linguistic, some are social, and some are psycholinguistic. The linguistic factors include, first and foremost, the frequency and recency with which individual L words have been accessed—and the frequency and recency of the L counterparts of those words (Paradis, ). The frequency and recency with which people produce, encounter, or otherwise access either L or L words are extremely difficult to gauge, but several studies have indicated that attriters overuse (Schmid & Jarvis, ) or show a processing advantage for highly frequent versus less frequent L words (see Schmid and Köpke, ). Another important linguistic factor is how similar L words are to their L counterparts. Cognates with similar forms and meanings across languages are, expectedly, less likely to exhibit attrition effects, whereas words with similar forms across languages but different meanings are often detrimental to both fluency and accuracy (see Schmid & Köpke, , for a review of relevant studies). The most important social factors appear to be age and environment. Age is investigated in relation to the age at which the attrition-inducing condition begins, such as a drastic reduction in the person’s use of the L, adoption into a family that speaks a different language, or emigration to a new country where a different language is spoken. The results of past studies have shown that L attrition (including lexical attrition) is often dramatic in cases where attrition begins before puberty, and is much less so in cases where the onset of attrition occurs after puberty (cf. Köpke & Schmid, ; Isurin & Seidel, ). Recent research has also shown that word relearning improves with age (Van der Hoeven & de Bot, ). The nature of the attrition-inducing condition is of course also of primary importance in that it affects the frequency with which the L is used. Several studies have explored the effects of the amounts and types of exposure attriters have to the L, but, interestingly, the only measure of L use that has reliably predicted the degree of L attrition is ‘the use of the L for professional purposes: the more migrants use the L at work, the less L attrition they exhibit’ (Schmid & Jarvis, , p. ). Education is another social factor that one might expect to affect the maintenance of L lexical skills. However, although education does appear to mitigate L lexical attrition (see Bardovi-Harlig & Stringer, , p. ), the same has not been found for L attrition (see Schmid & Köpke, , p. ).

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi





The relevant psycholinguistic factors are inseparably linked to both linguistic and social factors. Of particular relevance is the level of L proficiency—or the exact nature of the L lexical skills—the attriter had just prior to the onset of attrition. As predicted by the activation threshold hypothesis and the interference hypothesis, individuals with stronger and broader skills and experience in the L are less likely to have high activation thresholds for L words and are less likely to experience lexical interference from the L. However, this also depends on their levels of proficiency and experience in the L (cf. Schmid & Jarvis, ). Importantly, the effects of language proficiency on L attrition are largely confounded by the age factor: up to a certain age (perhaps around puberty), age and L skills co-vary. The fact that prepubescent emigrants and international adoptees are often found to show drastic levels of L attrition (see Isurin & Seidel, ) is perhaps often due to incomplete learning of the L combined with the early acquisition of greater levels of proficiency, experience, and dominance in the L than in the L (cf. Hyltenstam et al., ).

. M    

.................................................................................................................................. Excellent overviews of the methods relevant for investigating L lexical attrition can be found, among other places, in Schmid & Jarvis (), Isurin & Seidel (), and especially Schmid & Köpke (). Here, I will not try to provide a comprehensive treatment of these methods, but will instead summarize some of the main methods and methodological considerations pertinent to the investigation of L lexical attrition in the domains of accuracy, fluency, and complexity. Traditional freestyle language production tasks, such as spoken or written film descriptions and oral interviews, are representative of natural language use and will always be relevant to the investigation of language attrition and other natural language processes. As I will describe in the following paragraphs, data collected through freestyle production methods are useful for exploring all three areas of proficiency—most especially accuracy and complexity. However, in all three domains— especially fluency—it is valuable to use controlled tests or experiments in addition to (e.g., Schmid & Jarvis, ) or instead of free production tasks. Free production tasks are of course irreplaceable for investigating lexical production accuracy, and it is important to choose tasks that are truly representative of the type(s) of language the attriter has experience with and is likely to produce again in the future outside of a research context. Analysis of the data should focus not just on the accuracy of individual word forms, but also on the makeup of multi-word units (Bardovi-Harlig & Stringer, ), as well as on word choice appropriateness. Lexical accuracy can also be gauged in receptive tasks, such as lexical decision tasks (Segalowitz, ) and picture-word matching tasks (see Schmid & Köpke, ), which can measure, respectively, word recognition accuracy and the accuracy of attriters’ form-meaning mappings. Fluency and fluency-related phenomena (e.g., retrieval slowdown and failure) have received more attention than any other area of lexical attrition—perhaps especially because this is the area that is most directly connected to the activation threshold hypothesis—and the methods for investigating lexical fluency are correspondingly broad and varied.

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi

 



In analyses of free production data, lexical fluency is sometimes measured in relation to speech rate or the total number of words produced (e.g., Schmid & Jarvis, ). However, because lexical fluency is so sensitive to contextual and environmental factors, it is often best to measure it—additionally or instead—with a more controlled task of productive lexical fluency. This might involve controlled tasks of either semantic fluency or phonetic fluency, or both. Controlled tasks of productive semantic fluency include picture-naming tasks with reaction time measures, and especially verbal fluency tasks where participants are given semantic categories (e.g., animals, food, clothing) and asked to produce as many words as possible representing that category within a specified time period (e.g.,  seconds) (see, e.g., Schmid & Köpke, ). Controlled tasks of productive phonetic fluency, on the other hand, are difficult to find in the attrition literature (but see Guion et al.,  for an example of how they are used to investigate L acquisition), and it will be important for future research on lexical attrition to develop and make ample use of this potentially promising but severely underused method for measuring productive lexical fluency. Regarding receptive lexical fluency, this is often assessed with picture-word matching tasks and word recognition tasks by measuring how long it takes the attriter to respond. According to Schmid & Köpke (), however, the most useful method for measuring lexical access (and, by extension, receptive lexical fluency) is the lexical decision task, where participants are timed to see how long it takes them to decide whether a string of letters is a real word or not—often following a priming condition. This important method has been neglected in L attrition research (but see Segalowitz, ; and Goral et al., ). Finally, concerning lexical complexity, this can be measured in various ways (Bulté & Housen, ), but it seems most fruitful to do so with freely produced data that are as natural as possible, such as narratives of what the attriter has witnessed, or recorded conversations about the person’s own life experiences (e.g., Schmid & Jarvis, ). The data can be either written or spoken, but the latter has the advantage of being amendable to a broader analysis of attrition-related phenomena (e.g., speech rate, filled and empty pauses, repetitions). Lexical complexity can be investigated from the perspective of either lexical diversity or lexical sophistication, or both, and there is good reason to view both phenomena as closely related, if not part of the same construct (Schmid & Jarvis, ). It is also important to recognize the multi-dimensional nature of lexical diversity, and to use multiple measures that reflect each of these dimensions in the most valid way possible. It is crucial to take into consideration not just the number of types (different words) and tokens (total words) in the sample, but also which specific words occur in the sample, the ways in which they are related to one another, and how evenly they are dispersed throughout the sample (see Jarvis, , a,b). Above all, it is important not to rely on a single, unidimensional measure of lexical diversity—such as the type; token ratio—that measures only the overall proportion of repeated words in the sample and varies as a function of sample size (see Schmid & Jarvis, ).

. F 

.................................................................................................................................. To conclude, L lexical attrition phenomena can be found not only in attriters’ loss of the ability to produce and sometimes even comprehend L words and multi-word units, but

OUP CORRECTED PROOF – FINAL, 21/6/2019, SPi





also in the fluency, accuracy, and complexity of their demonstrated L lexical skills. The patterns of L lexical attrition that have been documented in past research are largely consistent with the predictions of the dormant language hypothesis, and especially the threshold hypothesis and interference hypothesis. L lexical attrition has also been shown to be affected by linguistic, social, and psycholinguistic factors. Linguistic factors include inter alia the frequency and recency with which L words and multi-word sequences have been encountered, and how similar they are in form, meaning, and syntagmatic structure to corresponding L lexical items. Social factors include the age at which attrition begins, as well as the contexts in which a person continues to use the L. Psycholinguistic factors include the level of proficiency the attriter had in the L before the reduced exposure began, and the attriter’s level of proficiency and range of skills in the L. There is a good deal of room for future fruitful research on L lexical attrition. It would be especially useful to see more longitudinal studies that track the beginning and progression of changes in a person’s L lexical skills involving attrition-inducing conditions such as international adoption and emigration to an L environment (e.g., Isurin, ; but see Schmid, , p. , for problems this poses for L attrition research). There is also a need for additional research properly designed to test the dormant language hypothesis (Isurin, ) and various versions of the interference hypothesis (see Bardovi-Harlig & Stringer, , p. ). Future L lexical attrition studies should also give increasing attention to multiword units and the lexical networks and syntagmatic relationships they entail (BardoviHarlig & Stringer, ). In experimental studies, it will be important to make use of valuable methods, such as lexical decision tasks, that have been neglected in past L attrition research but have the potential to shed a great deal of light on how attrition affects lexical accessibility (Schmid & Köpke, , p. ). Finally, future research on L lexical attrition will benefit greatly by giving more attention to the levels of lexical complexity, lexical diversity, and lexical sophistication that attriters demonstrate in freely produced samples of speech and writing. This will be especially useful if the tools used to measure these abstract lexical skills reflect their multi-dimensional character and the ways they are perceived by the human mind (Jarvis, a; Schmid & Jarvis, ).

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

  ......................................................................................................................

       ......................................................................................................................

̧  ̈  

. I

.................................................................................................................................. A variety of linguistic phenomena related to pronominals has attracted much interest in the field of L acquisition. In this section, I will briefly discuss how this line of L research has developed over the years, subsequently paving the way for L attrition studies in s. One of the earlier theoretical motivations for the study of pronoun acquisition lies in a generative view, which analysed null and overt pronouns as parameterized Universal Grammar (UG) options. For decades, while theoretical linguists have investigated linguistic mechanisms licensing null and overt pronouns across languages (e.g., Chomsky, ; Rizzi, ; Jaeggli & Safir, ; Huang, ), the question of how they are acquired has occupied the agenda of UG-based generative L research (e.g., White, , ; Hilles, ; Phinney, ; Liceras, , ; see also Hyams, ; Hyams and Wexler, ; Rizzi, , for L acquisition). In early adult L acquisition research, the focus was initially on the acquisition of morpho-syntactic mechanisms licensing null pronouns (e.g., subject–verb agreement markers) in reference to cross-linguistic transfer and its directionality in cases where the L and L display the opposite setting of the null subject parameter (e.g., White, , ; Phinney, ; Liceras, , ). In this context, the notion of markednesss was discussed extensively as a factor determining the directionality of transfer. With respect to pronouns, the theory of markedness invoked in learnability-based models (Berwick, ) posits that overt (subject) pronouns are unmarked, because assuming that a language has only overt pronouns is the most conservative hypothesis (i.e., subset grammar) that the

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi



̈ 

learner can initially entertain.1 Nevertheless, in adult L acquisition, L transfer could override the Subset Principle, as might be evinced by learners’ use of null pronouns in a non-null-subject L due to their L. More recent evidence supporting the default status view of overt pronouns has emerged on the basis of child and adult L data documenting the influence of non-null-subject Ls on null-subject Ls in the form of overuse and/or misinterpretation of overt pronouns (despite intact null pronouns in the L). Thus, overt subjects are considered the default option (e.g., Sorace & Filiaci, ; Belletti et al., ; Sorace & Serratrice, ). Moreover, the finding that overt pronoun overuse in null pronoun contexts is observed even in bilingual children with two null-subject languages (Sorace et al., ) also supports this point. As will be discussed below, overt pronoun overuse is also widely considered a primary symptom of L attrition. Subsequent L research on null and overt pronouns has emphasized that the L acquisition amounts to more than just establishing these forms are available in the L. The task also involves learning how pragmatic-discourse constraints govern the distribution of null and overt pronouns (e.g., Liceras, ; Pérez-Leroux & Glass, ; Gürel, ; Rothman & Iverson, ). As will be discussed in the next section, this issue has become relevant in L attrition research that examined the link between the use of null/ overt pronouns and word order variations in null-subject languages (e.g., Tsimpli et al., ; Domínguez, ). An additional theoretical motivation to study pronominals in L acquisition and subsequently in L attrition can be attributed to the fact that binding properties of pronominals (pronouns, reflexives, and reciprocals) had a central place in Chomskyan theorizing, particularly between early s and late s. The subtle differences in referential properties of pronominals have been analysed under Binding Principles A, B, and C, which are assumed to be part of UG. L research on the acquisition of Binding Principles has been particularly fruitful for determining L transfer and UG access in adult learners (e.g., Hirakawa, ; Thomas, ; Wakabayashi, ; White, ; Yuan, ). These theoretical advancements at the time and the ensuing empirical work on L acquisition of pronominals have sparked interest in generative linguistics-inspired research into the maintenance/loss of UG-based properties in the L grammars of bilinguals. The need for linguistic theory-driven L attrition research as part of a more general investigation of bilingualism has been expressed repeatedly in the literature (e.g., Köpke & Schmid, ; Schmid & Köpke, ; Schmid, b, a, Chapter ; see also Schmid & Köpke, a), and the attrition studies discussed in this Chapter constitute an important step in this direction. In the rest of the chapter, after a brief note on key linguistic properties associated with pronominals, relevant adult L attrition studies are presented and their implications are discussed. The following section should not read as a comprehensive overview of all theoretical models proposed so far regarding the linguistic properties (i.e., licensing/ identification, binding) of pronominals. Rather the aim here is to sketch out linguistic characteristics of pronominals as relevant for the L attrition studies discussed in the subsequent sections.

1 If the language also allows the null pronoun, the learner can subsequently acquire it on the basis of positive evidence, without any need for negative evidence (White, ).

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

      



. T        

.................................................................................................................................. In some null-subject languages, there are certain null pronoun-licensing morpho-syntactic mechanisms (e.g., rich/strong morphological features/markers). Yet the use of null and overt pronouns is not random but subject to certain discourse-pragmatic constraints. Overt subjects in null-subject languages perform a variety of functions such as topic shift, contrast, and focus (e.g., Montalbetti, ; Enç, ), as illustrated in the following examples from Turkish, a null-subject language (Enç, ; Gürel, ): ()

pro

alış-veriş-e shopping-

gid-iyor-um go--

() Ben alış-veriş-e gid-iyor-um I shopping- go-- ‘I am going shopping’ While () would be an appropriate answer to a question such as Nereye gidiyorsun? (‘Where are you going?’) in line with the topic continuity function of null pronouns, the answer with an overt pronoun () would not. Sentences such as () would normally be used to signal a contrast such as Ali eve gidiyor ama ben alış-verişe gidiyorum (‘Ali is going home but I am going shopping’) or to establish a topic, for example if the speaker decides to leave the house to go shopping and wants to inform her/his partner about it (Enç, , p. ). The following sentences illustrate similar constraints in Italian (Tsimpli et al., , p. ): ()

pro

E’ is-SG ‘He left’

partito gone-

A null pronoun in () is preferred in Italian if this sentence is uttered as a response to a question such as ‘What did Gianni do’ to maintain topic continuity. Topic shift, however, requires overt pronouns. The presence of post-verbal subjects is also discussed in relation to the null-subject property of Italian. For example, in response to an all-focus question such as Che cosa è successo? (‘What happened?’), an answer with a post-verbal subject () is required (see Sorace, , and Belleti & Guasti, , for other linguistic factors determining post-verbal subjects in Italian). () pro

È is-SG ‘John left’

partito Gianni gone- John

Pre- and post-verbal subjects are also allowed in Greek. Pre-verbal indefinite subjects are interpreted as ‘old’ information (topic), whereas post-verbal indefinite subjects are ambiguous between the ‘old’ and ‘new’ information (Tsimpli et al., , pp. –):

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

 ()

̈  Context: I gitonisa mu ston trito orofo apektise dhidhima. ‘My neighbour on the third floor had twins.’

() Xtes vradhi ena last night one ()

moro baby

ekleje. was-crying (=one of the twins)

Xtes vradhi ekleje ena moro. last night was-crying one baby (=one of the twins OR some other baby) ‘My neighbour on the third floor had twins. Last night one baby was crying/wascrying one baby’

Another linguistic feature of pronominals pertains to their referential properties. As shown in the Greek examples below, null and overt pronouns involve different coreferentiality (Tsimpli et al., , p. ): () O Janisi prosvale ton Petrok otan proi /aftosk the Janis insulted the Petro when pro / he ‘Janisi insulted Petrok when hei /k approached him.’ () O Janisi idhe ton Petrok otan AFTOSk ‘Janisi saw Petrok when HEk approached him.’

ton

ton him

plisiase. approached

plisiase.

In (), the use of the null pronoun in the subordinate clause subject position leads to a nonshifted interpretation (hence its coreferentiality with the matrix subject, Janis), whereas the overt subject (aftos) is coreferential with the matrix object only. In (), the pronoun is additionally licensed by discourse focus, so the antecedent for the overt pronoun is the matrix object (Tsimpli et al., , p. ). The same referential tendencies have been noted earlier in Italian intrasentential anaphora by Carminati (, p. ) under the Position of Antecedent Strategy, which postulates that the null pronoun has a strong bias towards the most prominent antecedent in the SPEC, IP position (typically the sentential subject), while the overt pronoun, albeit flexible in coreference, tends to select a non-subject antecedent in the non-SPEC, IP positions.2 Another phenomenon related with pronominal binding is discussed under the overt pronoun constraint (OPC) (Montalbetti, ), which stipulates that in null-subject languages such as Spanish and Japanese, when the antecedent is a referential NP, overt pronouns and pro can both be coreferential with the antecedent. However, when the antecedent is quantified, overt pronouns, unlike pro, can have only disjoint reference (see examples () and ()).3,4 2

Not all null-subject languages show similar anaphoric preferences. Even typologically close and morpho-syntactically similar languages, such as Italian and Spanish, display different patterns of overt pronoun resolution (Filliaci et al., ; Romano, ). 3 According to the standard theory of Binding (e.g., Chomsky, ), in a sentence such as Johni believes that [hei/j is intelligent], ‘John’ and ‘he’ can be coreferential as ‘John’ is not in the local domain/ governing category of the pronoun (i.e., embedded finite clause). The pronoun can potentially have disjoint reference (e.g., a third party in the discourse), as shown by the index ‘j’. In contrast, the reflexive anaphor, ‘himself ’ as in Briani says [Georgej likes himself*i/j/*k] has to be governed by an antecedent in its local domain, hence the co-indexation of ‘himself ’ with George only. 4 The distinction between bound and disjoint interpretations can formally be explained as follows (Gürel, a, p. ): (i) Bound variable interpretation: (No x: x is a person) x believes that x is intelligent;

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

      



() Juani cree que [éli/j / proi/j es inteligente] ‘John believes that hei/j / proi/j is intelligent’ ()

Nadiei cree que [él*i/j / proi/j es inteligente] ‘Nobodyi believes that he*i/j / proi/j is intelligent’

Gürel (, p. ) argues, on the basis of Turkish, that the OPC may not be a property of all null-subject languages. She suggests that the disjoint reading requirement for the Turkish overt pronoun o is syntactically driven (i.e., it derives from Binding Principle B, but not by the OPC) (cf. Montalbetti, ). Since the finite matrix clause (not the nominalized embedded clause) is the governing domain in Turkish, the overt pronoun o cannot be coreferential or bound by the matrix subject (unlike the null pronoun and other overt pronominal, kendisi). Crucially, contrary to what Montalbetti suggests, these binding constraints hold in both referential and quantified contexts alike (indices in the translations show the readings available in English): ()

Zeynepi [o-nun*i/k / proi/k/kendi-si-nini/k dahi Zeynep s/he- pro self genius düşün-üyor think-PRG ‘Zeynepi thinks (that) s/hei/k is genius’

()

Herkesi [o-nun*i/k / proi/k / kendi-si-nini/k dahi Everyone s/he- pro self genius düşün-üyor think- ‘Everyonei thinks (that) s/hei/k is genius’

ol-duğ-u]-nu be--.-

ol-duğ-u]-nu be--.-

The theoretical conceptualization of null and overt pronouns (i.e., the licensing/ identification and binding conditions) has changed over the years particularly after the advent of the Minimalist Program (MP) and its subsequent reformulations (e.g., Chomsky, , , ). The sources of parametric variation across languages are reduced to differences in the formal features of particular items (e.g., the functional heads) in the lexicon, and also to Extended Projections Principle (EPP) features (see Holmberg, ; Biberauer et al., , for further discussions). These grammatical features involving number (singular, plural), gender (masculine, feminine), person, and case are assumed to play a role in morpho-syntactic processes. Some of these features have semantic content hence interpretable at logical form (LF), whereas others are uninterpretable at LF and therefore must be eliminated (checked off) before LF. For example, phi-features (person, gender, number) on an NP (or DP) are considered [+interpretable] and persist at the LF interface to assure interpretability. Categorical features that are strong are also [+interpretable]. However, the phi-features on a verb, auxiliary, adjective, as well as case features are considered [ interpretable] (Chomksy, ). As for null and overt subjects, it is assumed, under one version of the MP, that null pronoun licensing is regulated by the interpretability of agreement features (e.g., Alexiadou & (ii) Disjoint interpretation: (No x: x is a person) x believes that y is intelligent (e.g., Nobody believes that John is intelligent).

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi



̈ 

Anagnostopoulou, ) and obligatory overt subject is explained via feature movement. In other words, in languages like English, merging of the subject in the specifier of Agreement position is required in order to check the EPP feature (but see Holmberg, ; and Roberts, , for alternative views). Despite subtle differences in different versions of Minimalist assumptions, in general, in this reconceptualization, while (second) language acquisition reduces to the learning of formal features of a target language, language attrition is about losing the specifications of these features (e.g., Tsimpli, ; Sorace, ). These theoretical analyses have not only shaped L acquisition research (e.g., Park, ; see also Lardiere, ) but also L attrition work (e.g., Sorace, ). Even though the linguistic analyses adopted and the theoretical models tested have changed over the years, the main objective of L attrition researchers has been to examine how much of the subtle L knowledge of pronominal properties is maintained under extensive L exposure. As intricate characteristics of the L pronominal system constitute an integral part of implicit grammatical knowledge, any attrition effects in this domain would suggest the alterability of mature native competence (Sorace, a; Gürel, ; cf. Montrul, ). In that respect, the studies reviewed in the following section contribute to our understanding of the extent and the nature of the L attrition phenomenon that may emerge either as a competence or a processing problem in the pronominal domain as an (un)avoidable consequence of bilingualism.

. A L     

.................................................................................................................................. Sorace (a; and Chamorro & Sorace, Chapter , this volume) is the first among generative linguists who has made predictions for L features vulnerable to attrition in adult grammars. Following the Minimalist assumptions as introduced by Chomksy (), Sorace takes [±interpretable] to be the relevant feature for selective attrition. She argues, based on data from L Italian and L Greek speakers of L English, that the licensing of null subjects is due to an [ interpretable] feature (e.g., an agreement suffix on the verb) while the distribution of null and overt subjects depends on [+interpretable] features such as [+Topic Shift]. Accordingly, because null subjects are the result of the specification of [ interpretable] features, they will not be affected by attrition and will continue to occur in contexts in which they occur in the speech of monolinguals (i.e., in [ Topic Shift] contexts). However, the distribution of overt subjects will be affected. That is, exposure to L English, a language whose pronouns are not obligatorily specified for [+Topic Shift], would cause overt pronouns to become optionally unspecified for [+Topic Shift] in Italian-English bilinguals and lead to the infelicitous occurrence of overt pronouns in [ Topic Shift] contexts. Sorace’s predictions have been tested in several studies. For example, Tsimpli et al. () examined the production and interpretation of null and overt subjects as well as pre-verbal and post-verbal subjects by UK-resident L Greek and L Italian participants who are near-native L English speakers. The researchers examined via elicited production and interpretation tasks whether L Greek attriters (i) overproduce preverbal

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

      



subjects; (ii) demonstrate non-native-like interpretation of overt indefinite subjects in preverbal and post-verbal positions; and (iii) whether L Italian attriters interpret null and overt pronouns in a non-native-like fashion in bidirectional anaphora resolution. The results of the production task revealed that the Greek attriters produced significantly more preverbal subjects than the Greek controls. In the interpretation task, attriters allowed both new and old information in pre-verbal and post-verbal positions, hence their indeterminate pronoun interpretations. As for L Italian, Tsimpli et al. (, p. ) discussed only the results of a picture verification task, which showed that in backward anaphora sentences with a null subject in the subordinate clause such as Quando pro/lei attraversa la strada, l’anziana signora saluta la ragazza (‘When pro/she is crossing the street, the old woman greets the girl’), both groups preferred the subject (i.e., the old woman) as the referent of pro. In the same sentences with an overt pronoun, the Italian controls strongly favoured an external referent (other than ‘the old woman’ or ‘the girl’), while the Italian attriters accepted, almost equally, all three referents. In sentences with forward anaphora such as L’anziana signora saluta la ragazza quando pro/lei attraversa la strada (‘The old woman greets the girl when she/pro is crossing the street’), with null pronouns, the Italian controls allowed for either the matrix subject or the complement as the referent, whereas the attriters preferred the subject interpretation. When the embedded subject is an overt pronoun, both groups preferred the matrix complement as its referent. Nevertheless, the attriters’ matrix subject preference with the overt pronoun was higher than that of controls. The authors interpreted this as a sign of attrition, as the attrition group showed a significantly greater tendency to allow an overt pronoun to be interpreted as a continued topic. Sorace’s (a) original ideas were later refined under the Interface Hypothesis (IH) (Sorace , ; Sorace & Filiaci, ; Sorace & Serratrice, ; Chamorro & Sorace, Chapter , this volume), which suggests that purely syntactic features (narrow syntax) and features that are at the interface of syntax and other cognitive-linguistic domains (e.g., syntax–discourse/pragmatics, syntax–semantics) undergo differential L attrition (and L acquisition) (Sorace, ). While purely syntactic features pose no considerable problems, interface properties are subject to optional use due to difficulty integrating different levels of information, particularly in real-time language processing. Sorace’s proposal was investigated recently by Kaltsa et al. (), who examined pronominal resolution in Greek by comparing L attriters (adult immigrants in Sweden) with monolingual controls.5 As in Tsimpli et al.’s () study, the focus was on null and overt subject pronominal interpretation in adverbial clauses in forward anaphora contexts I γiaγia xeretise tin kopela otan afti pernuse to dromo (‘The old lady greeted the girl when she crossed the street’). Adult L attriters were predicted to coreference subject antecedents to overt subjects far more than the controls due to L influence from Swedish—a non-null-subject language. The null pronoun interpretation, in contrast, was expected to remain native-like. The study consisted of two self-paced listening experiments (Experiments  and ) involving sentence-matching, and testing null and overt subject pronoun resolution, respectively. Participants were asked whether the sentences they heard matched the presented pictures, which depicted the subject (‘the old lady’), the

5 The study also included heritage speakers (early Greek-Swedish bilinguals) and younger Greek controls.

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi



̈ 

object (‘the girl’), or another person (disjoint reference) as the agent of the action provided in the adverbial clause (e.g., crossing the street). The results of Experiment  revealed that although the attriters generally preferred the object rather than the subject as the antecedent for the overt subject pronoun, their preference for the subject antecedent for the overt pronoun was still higher than that of the controls. In terms of reaction times (RTs), the attriters were, overall, slower than adult monolinguals but both groups showed longer RTs in the ‘subject referent’ and ‘other referent’ conditions than the ‘object referent’ condition, in line with their offline preference in picture matching. With respect to the listening times in the critical region (i.e., subject pronouns of the subordinate clause), the attriters were found to be faster than the controls in all three conditions. In Experiment , which tested null pronoun anaphora resolution, no significant difference was found between the attriters and controls in their subject antecedent preference. The two groups did not differ on the other two conditions, either, which suggests that, as predicted, co-indexation between the null pronoun and the subject antecedent is intact in attriters. The RT data, however, showed slower RTs in the subject referent condition compared to the object referent condition in the attriter group. In terms of listening time on the critical segment (the verb in the subordinate clause), the attriters showed no difference among the three referent conditions. The authors suggested that, in L Greek attriters, the overt pronoun is more vulnerable to change (i.e., it can be associated with subject and object antecedents) than the null pronoun. Given that even the monolinguals’ matching decisions on the overt pronoun were affected, the authors could not confidently attribute these results to L Swedish influence or to the restrictions in bilinguals’ processing resources (cf. Sorace & Serratrice, ). In another study testing the IH, Chamorro, Sorace, & Sturt () examined in three participant groups whether it is knowledge representation (i.e., competence) or ability to process interface structures that is affected in L Spanish attrition. The first group included late bilinguals with near-native English proficiency, residing in the UK for a minimum of five years. The second group was a re-exposed group who were also near-native English speakers with a minimum of five years of UK residency. This group was tested after they returned to Spain and re-exposed exclusively to Spanish for a minimum of one week. The third group included L Spanish-speaking controls who had recently arrived in Edinburgh and had very little knowledge of English. Additionally, the study explored whether attrition effects decrease after increased L input via re-exposure, an issue that is linked to the Activation Threshold Hypothesis (ATH), which suggests that the more frequently a linguistic item is activated, the lower its activation threshold is. The researchers examined intrasentential semantically-neutral anaphora constructions by manipulating the grammatical number of antecedents, so that the pronoun could refer to either the subject or object antecedent (e.g., Las madres saludaron a la chica cuando ella cruzaba una calle con mucho tráfico, ‘The mothers greeted-PL the girl when she crossed-SG a street with a lot of traffic’). In the other item types, the sentential subject of the adverbial clause had a null pronoun. Data was collected via eye-tracking-while-reading and untimed naturalness judgement tasks, which are believed to tap online processing and knowledge, respectively. The prediction was that the attriters would show sensitivity to the pronoun mismatch only in the offline task. The re-exposed group, however, was expected to perform similarly to the native controls in both tasks. In eye-tracking, the critical region was ‘ella/pro cruzaba’, and first-pass time, go-past time, and total time were analysed. The judgement task results revealed that all groups converged on their preferences for subject pronouns.

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

      



In eye-tracking, while the control and re-exposed groups showed early sensitivity to the mismatch conditions, the attriters showed no evidence for a pronoun–antecedent interaction in any of the measures. These results support the IH, which predicts non-native performance only on real-time processing of pronominal properties at the syntax–discourse interface (see Genevska-Hanke, , for similar arguments). In addition, the findings are consistent with the ATH, as the attrition effects disappear after re-exposure to the L. If difficulties with the processing of interface properties are recoverable after only one week of L (re-)exposure, this clearly suggests that processing problems are extremely short-lived.6 Indeed, the transient/restricted nature of the attrition phenomenon was noted in a study by Miličević & Kraš () that examined L English-induced attrition effects in the interpretation of L Italian pronominal subjects in English–Italian translators in comparison to non-translator bilinguals. In a picture selection task, the participants were asked to identify the antecedents of null and overt subject pronouns in intrasentential forward and backward anaphora sentences similar to the ones discussed earlier. The results showed no over-acceptance of subject referents in the overt pronoun condition by the experimental group. The authors suggest that the overuse of overt subject pronouns previously reported for translated texts (e.g., Cardinaletti, ) might simply be a product of the translation process itself. The interpretable features of the null and overt pronouns were also investigated in relation to Binding Principles. For example, Gürel () examined L English influence on pronoun binding in L Turkish following an earlier formulation of the theory (e.g., Chomsky, ).7 The complex sentences tested had overt or null subject pronouns in embedded clauses, which are nominalized DPs in Turkish (e.g., Buraki onun*i/k / proi/k / kendisinini/k zeki olduğunu düşünüyor, ‘Buraki thinks that hei/k is smart’).8 Besides the overt pronoun o, Turkish also has a special type of anaphoric pronominal, kendisi, for which English has no corresponding form. Gürel’s study also included items with quantified antecedents as in (). The attrition group consisted of native Turkish speakers who were late L bilinguals (end-state L English users), who had immigrated to North America (Canada or the US) as adults (mean age of immigration: .), and had been living in the L country for at least ten years at the time of testing. The native controls were residents of Turkey since birth and had basic knowledge of English. The tasks included written interpretation, story-based truth-value judgement, and picture identification/matching listening tasks. Results across the three tasks revealed that while the overt pronoun o was most often assigned the disjoint reading, the null pronoun and kendisi received mostly bound or ambiguous readings in both groups. Nevertheless, the tendency to assign the correct disjoint reading for o was weaker in the attrition group. When participants were given only two options (bound or disjoint readings) (Tasks  and ), the difference 6 It is important to note that the re-exposed group differed from the attrition group only by the fact that, by the time of testing, the former had gone to Spain for the Christmas break and re-exposed to Spanish exclusively for a minimum of one week (the mean number of days of L re-exposure was .). The re-exposed group had not been tested before returning to Spain. Therefore, we cannot say with confidence that their native-like performance is due to re-exposure to Spanish. 7 This is actually a comparative study of L attrition and L acquisition of Turkish, exploring English influence on Turkish. 8 As noted earlier, unlike in English, the governing domain in Turkish is the matrix clause, not the nominalized embedded clause.

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi



̈ 

between groups became more salient; attriters allowed significantly more (incorrect) bound interpretations for o than controls. For the null pronoun and kendisi, the groups were similar in interpreting them mostly as bound pronouns. These findings were interpreted readopting a subset-superset transfer model, originally proposed for L acquisition (Manzini & Wexler, ), in the sense that when the L constitutes a superset of the L with respect to a particular property, L-induced L attrition emerges. In this particular case, since English allows both DPs and finite IPs as binding domains, it constitutes a superset of Turkish, which does not allow DPs as binding domains. Therefore, Turkish native speakers tended to expand L options based on the L model to eliminate restrictions in the L grammar. In contrast, in configurations with the L offering a more restrictive subset grammar than the L, no L restructuring is expected. A study replicating Gürel () with the aim of establishing similar effects of two typologically-related Ls (English and Dutch) on L Turkish tested Turkish attriters living in the Netherlands under the influence of L Dutch, a non-null-subject language with English-like binding properties and found similar results, confirming the set theoretic assumptions for L attrition (Gürel & Yılmaz, ). Gürel’s () data was also interpreted within the ATH in Gürel (a), proposing that, due to L English, attriters treat o as identical to the corresponding English overt pronoun as a result of activation of L binding properties in lieu of those in the L. When an L linguistic element/rule has no equivalent form in the L (as in the case of pro and kendisi), the L does not compete with the L, hence no interference or attrition (Köpke, ). The set-theoretic predictions proposed in Gürel () were also tested in L Turkishinduced L English attrition in Gürel (). The study compared native speakers of English with advanced Turkish proficiency who had been living in Turkey for an average of . years with native English speakers who had no knowledge of Turkish. The tasks were similar to the first two tasks used in Gürel (). The constructions included possessive DP pronouns (‘The soldiers carried their boxes’), monoclausal sentences with pronouns/ reflexives (‘James sent Henry a picture of him/himself ’), biclausal finite (‘Chris knows that Harry hates him/himself ’) and non-finite sentences with pronouns/reflexives (‘Nancy asked Emily not to blame her/herself ’), and biclausal finite sentences with subject pronouns (‘John said that he lied to the police’). Based on different binding properties in Turkish and English, it was predicted that more attrition effects would occur in configurations where the L Turkish offers broader grammatical options than English. For example, in monoclausal sentences with two potential antecedents, while the English pronoun can allow only disjoint reference, the Turkish object pronoun o can both be coreferential with the indirect object antecedent and have disjoint reference. Thus, under attrition, the L English grammar was expected to move towards L Turkish grammar, which offers less restriction. Similarly, it was predicted that the binding properties of kendisi could be mapped onto the L English reflexive ‘himself/herself ’, leading to attriters’ acceptance of non-local antecedents for it. Nevertheless, the results revealed no evidence of L attrition in this domain. Although the findings contradict the set-theoretic linguistic model of L attrition, frequency-based psycholinguistic models such as ATH can explain them. That is, the absence of attrition effects in English might be linked to the frequency of use and the de facto status of English as a lingua franca (Gürel, ). In another study examining attrition effects in the domain of reflexives, Kim et al. () compared US-resident Korean heritage speakers (early bilinguals), first-generation Korean

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

      



immigrants (late bilinguals), and adult L Korean learners to monolingual Koreans. The attrition group had been in the US for more than ten years. The task involved a truth-value grammaticality judgement task with stories followed by a sentence to be judged as ‘true’ or ‘false’. On the basis of differences in the binding properties of the Korean reflexive caki, which can function as a long-distance anaphor, and the English himself/herself, which can only be locally bound, it was predicted that in Korean constructions such as Wendynun Tedeykey cakika Charles pota ttokttokha ta ko malhayssta (‘Wendy told Ted that self (she) is smarter than Charles’) and Christineun Tom elopwuthe caki tongsayngi cheypotanghan iywulul tulessta (‘Christine learned from Tom the reason for self ’s (her) brother’s arrest’), bilingual groups would show a lower degree of acceptability compared to the monolingual controls, due to dominant English influence. Results showed that the Korean adult attrition group behaved like controls, while differing significantly from the other two experimental groups. Thus, unlike Gürel’s () first-generation L Turkish immigrants, late KoreanEnglish bilinguals maintained L knowledge of the governing category. The authors link this to the close-knit communities where Korean immigrants live, drawing attention to the role of sustained and frequent L contact. This section provided a selective overview of adult L attrition studies investigating different aspects of null and overt pronouns (i.e., licensing and interpretative/binding properties) as motivated by different theoretical assumptions. All of the studies, except for Gürel () involved late bilinguals with a null subject L (i.e., Greek, Italian, Spanish, Turkish, Korean) and non-null subject L (English, Swedish, Dutch) and revealed no conclusive results concerning the magnitude of L attrition either at the competence or performance level, crucially failing to establish an across-the-board kind of L effects from a non-null-subject language on the L in this domain. Overall inferences that can be drawn from these studies are discussed in the following section.

. S  

.................................................................................................................................. All in all, early generative studies on adult L attrition involving pronominals focused on the question of whether L knowledge of syntactic binding/governing domains could change under the influence of a dominant L. Results are inconclusive; while some studies reveal alteration in the L governing domain due to extensive L exposure, along with less accessible L input (Gürel, ; Gürel & Yılmaz, ), others document no evidence of attrition in this area of grammar (Gürel, ; Kim et al., ). Relatively more recent work on adult L attrition of pronominals is generally designed in line with the IH predicting optionality in structures at the interface between syntax and other cognitive domains (Sorace, , p. ). Indeed, as an interface phenomenon, the interpretation/use of null and overt subjects in the context of anaphoric dependencies constitutes a good testing ground for this hypothesis. Several studies confirm Sorace’s (, ) predictions in the sense that, unlike native controls, adult L attriters demonstrate optionality in the mapping of pronouns with a relevant interpretable feature (i.e., [±Topic Shift]). The overt pronoun appears to be particularly vulnerable as attriters fail to associate it consistently with the [+Topic Shift] feature. The null pronoun, however, remains relatively intact, as it is associated mostly with the [ Topic Shift] feature (Tsimpli et al., ; Kaltsa et al., ).

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi



̈ 

Differential vulnerability of null and overt pronouns is indeed predicted in views that analyse overt pronoun overuse as a default ‘fall-back’ option that has the advantage of being redundant rather than ambiguous (Sorace, , p. ) and in other views that essentially predict attrition based on a competition between a null-subject L and a non-null-subject L (Gürel, , a; see also Köpke, , ; Paradis, ). There are also studies that report no major change/restructuring in attriters’ L knowledge of anaphoric reference, as measured by offline tasks (Gürel, ; Chamorro, Sorace, & Sturt, ; Miličević & Kraš, ). This points to the processing versus competence dichotomy proposed by the IH and confirmed by studies revealing attrition effects only in online processing of multiple levels of information (Kaltsa et al., ; Chamorro, Sorace, & Sturt, ; see also Genevska-Hanke, ). Nevertheless, given the scarcity of studies, this is yet an unresolved issue. More data from different language pairs is necessary to clearly confirm that processing problems appear only in adult attriters but not in monolinguals, and that attriters’ core/narrow syntactic knowledge, as measured by offline tasks, always remains converged on native norms (cf. Gürel, ; Kim et al., ; Gürel & Yılmaz, ). With respect to L effects on L attrition in the domain of anaphoric dependencies, while some studies document L influence (Gürel, ; Tsimpli et al., ; Gürel & Yılmaz, ), others fail to show this (Kaltsa et al., ; Chamorro, Sorace, & Sturt, ; Miličević & Kraš, ). It is important to note that L attrition studies typically include only one experimental (attrition) group, whose L differs in certain ways from the dominant L. It is not possible to clearly establish L effects on L grammar in the absence of another attrition group whose L works similarly to the L (cf. Gürel & Yılmaz, ). It is also important to note that in many studies, control groups’ online and offline performance is not unequivocally better than that of adult attriters, suggesting that cross-linguistic influence may not be the major cause of L vulnerability (Kaltsa et al., , p. ). In conclusion, generative research on adult L attrition in the pronominal domain has not, until now, revealed a robust change/restructuring in abstract L grammatical competence as measured through offline acceptability tasks; it has only displayed certain (albeit not completely non-native-like) indeterminacies in online use/interpretation of referential properties of pronominals. In other words, while abstract knowledge representation regarding L pronominal properties remains intact, online processing of relevant linguistic features may be problematic. Furthermore, recent data suggests that processing inefficiencies in the L are largely recoverable after brief re-exposure (Chamorro, Sorace, & Sturt, ). This is a noteworthy finding when we consider the theoretical psycholinguistic literature revealing ‘good enough’ language processing even in monolingual native speakers. More specifically, as research suggests, even native speakers may not always be able to generate representations of the linguistic input that are complete, detailed, and accurate (Ferreira et al., , p. ). Thus in the real world, interpretations are likely to be ‘just good enough’ even in native comprehenders (Ferreira et al., , p. ), possibly because they tend to use heuristics (more than syntactic algorithms) as more readily available interpretations (Ferreira et al., ; Ferreira & Patson, , p. ). Given this, incorrect or incomplete comprehension/interpretation in the pronominal domain may not be a unique condition for attriters or bilinguals in general. Therefore, it is crucial to have more carefully designed attrition studies testing a wider range of phenomena via multiple modality tasks in a variety of languages to identify the underlying mechanism of L attrition and to address

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

      



the issue of grammatical representation versus processing dichotomy in this line of research involving late bilinguals. On this final note, to be able to establish confidently that attrition affects real-time integration (i.e., processing) of L knowledge, it is first necessary to determine whether the apparently divergent linguistic performance (e.g., comprehension) in attriters in online tasks may be ascribed to a set of human parsing strategies that lead to a cognitively less demanding decoding option. Such simple and parsimonious heuristics for building L interpretation may simply be used more readily by bilinguals than monolinguals. This would indicate that some of the phenomena which have been investigated in the area of first language attrition are less due to loss or erosion but more to general principles of bilingual processing.

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

  .............................................................................................................

EXTRALINGUISTIC FACTORS IN LANGUAGE ATTRITION ............................................................................................................. Edited by

 .    

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

  ......................................................................................................................

                          ......................................................................................................................

 .    

. I   

.................................................................................................................................. T interest of first language attrition as a research topic within the wider framework of developmental linguistics and bilingualism is grounded in the fact that bilingual speakers differ substantially from monolinguals in the way in which they use both their languages. The most profound difference is the fact that, in many respects, bilingual populations exhibit more variability than monolingual populations: linguistic features which are used consistently by monolinguals, so that tasks relating to these features generally find a ceiling effect, tend to exhibit a larger degree of optionality and thus a larger amount of stratification within bilingual populations. On the other hand, bilinguals tend to underuse features associated with a high degree of complexity, such as infrequent grammatical structures or lexical items. Group comparisons between monolinguals and bilinguals therefore tend to find between-population differences for such features. The recognition that this is true not only in second language (L) acquisition but also in first language (L) attrition was at the root of the emergence of L attrition studies in the early s (see Köpke & Schmid, , for a historical review). In the early phases of attrition research, interest was focused predominantly on (micro) linguistic phenomena associated with language attrition as such, with little concern for their degree of variability. This approach reflects the early view of language attrition as ‘a special case of variation’ (Andersen, , p. )—that is, as a process which is rare and only occurs in speakers who have had extremely long periods of immersion in an L setting (decades) with few occasions to use their L. On such a view, studying variability in background factors such as length of residence and amount of exposure makes little

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi



  

sense, since the small subpopulations of attriters within the wider population of bilinguals at large would, by definition, be relatively homogenous with respect to these factors. Such studies thus rest on the comparison of two extreme ‘slices’ of the bilingual continuum: attriters occupying one end and monolinguals residing in the country of origin the other. The assumed dichotomy was such that most early investigations only actually looked at data from the attrition end of the continuum without establishing a baseline for comparison: an annotated bibliography of attrition studies up to  (Schmid, a) lists a total of seventy-three empirical investigations, of which only sixteen (%) use an unattrited control group. This approach not only reflects the view of language attrition as a relatively rare and extreme form of language development, affecting only long-term immigrants with low levels of language contact, but at the same time embodies an idealized view of monolingual performance: if a study investigates only attriters, taking any and all deviances from the ‘correct’ standard as evidence for attrition, this implies the presumption of monolingual speech as error- and disturbance-free (see Schmid, c). It was largely the contribution of Dutch scholars, in the latter half of the s, which began to move attrition studies away from this purely linguistic focus and idealization of the ‘native speaker’, establishing the topic as one that needed to encompass a socio- and psycholinguistic dimension (see Köpke & Schmid, )—much in the same vein as what has been called the ‘social turn’ did for studies of second language acquisition (SLA) (e.g., Block, ). At this point, several large and ambitious projects were started, aiming to identify the contributions of extralinguistic, personal, and individual predictors, such as the investigation described in Jaspaert & Kroon (). These studies were probably ahead of their time in terms of their ambition and scope and, sadly, most were never completed. However, they did lay the groundwork for many of the investigations that were carried out later, which did take into account considerations of the environment, the circumstances, and the identity of the speaker groups who were experiencing attrition.

. T E G N  L A

.................................................................................................................................. The year  represents a turning point with respect to the methodology of language attrition studies: during the First International Conference on First Language Attrition (ICFLA), held at the Vrije Universiteit Amsterdam, a network of young researchers was formed, all of whom were at that point involved in early stages of individual projects investigating different aspects of language attrition. Between  and  this group of some fifteen researchers met annually with a view to developing the common research methodology that had been outlined as a necessary step to move attrition research forward by Köpke & Schmid () and Schmid (b). Arguably the most important outcome of these meetings was the development of the Language Attrition Test Battery, a collection of tools and questionnaires whose purpose it was to render the findings from attrition studies—in particular those related to external and personal background factors— comparable. These tools have been made available in a range of languages on http:// languageattrition.org as well as on the IRIS platform (http://www.iris-database.org) and

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

    



are described in detail in Schmid (), and they subsequently were used in a number of investigations of attrition in languages as varied as Dutch (Keijzer, c), English (Dostert, ), German (Badstübner, ; Schmid & Dusseldorp, ), Romanian (Cherciov, ), Hungarian (Varga, ), Persian (Gharibi, ), Polish (Lubińska, ), and Turkish (Yılmaz, ) (among others). Collectively these studies revealed a number of methodological challenges for language attrition research which made it necessary to develop, discuss, and refine strategies and analytic designs suitable for the investigation of such data. It has often been pointed out that language attrition is a complex developmental process, driven by the (often nonlinear) interaction of a plethora of factors (e.g., Schmid et al., ; Opitz, Chapter , this volume), and that these factors themselves tend to be multifaceted. This is particularly true for the factor that is most frequently invoked in attrition studies, namely frequency of use and exposure (see also Schmid, Chapter , this volume): studies of bilingual development and language dominance have long since recognized the complexities of quantifying this factor in such a way as would allow it to serve as a predictor in statistical investigations. Approaches that acknowledge this are, for example, Grosjean’s (, ) Complementarity Principle, which rests on the assumption that ‘[d]ifferent aspects of life require different languages’ (Grosjean, , p. ) and therefore assign one or several languages to different domains of use, such as the home, politics, specific leisure activities, etc. A somewhat similar approach was proposed by Birdsong () and modelled on recent approaches to index handedness. Similar problems apply to other extralinguistic factors, such as attitudes (see Yılmaz, Chapter , this volume), and imply that any investigation attempting to probe into the relationship between predictors and outcome variables in the attritional process will have to adopt a formidably multifactorial design, encompassing a wide variety of contexts and domains of language use and an equally varied set of variables relating to attitudes and identification. Such approaches are, however, often hampered by a number of practical and methodological challenges, including (but not limited to) • The problem of power and population size: in order to adequately describe actual linguistic processes of attrition, measures have to be collected that are extremely time-consuming to elicit, transcribe, and code (e.g., Schmid, a). In practice this means that attrition studies with more than thirty to fifty participants per population are not feasible, severely limiting the number of possible predictors (see Schmid & Dusseldorp, ). • The fact that most of these variables can only be elicited by overt self-assessments and introspection, raising concerns about their validity and reliability (see Cherciov, ). • The problem posed by the fact that such self-assessments are typically elicited as ordinal variables, for example by means of Likert Scales. • The problem posed by the fact that data relating to such factors is usually not normally distributed. This is a natural outcome of the Complementarity Principle: people tend to prefer one language over the other in different domains, implying that different factors will either be skewed towards one or the other language or bimodally distributed. Taken together, the problems listed above pose serious concerns for the application of parametric statistics. Non-parametric procedures, on the other hand, are extremely limited

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi



  

when it comes to multifactorial analyses (e.g., Norman, ). One focus of the work of the Graduate Network was therefore to propose a standardized set of steps and analytic procedures that would help alleviate these problems. These are set out in Schmid & Dusseldorp () and Schmid & Yılmaz (), and tools for their application were also made available (on https://languageattrition.org).

. A    

.................................................................................................................................. The quantitative and empirical evidence base on language attrition established over the past two decades, partly by members of the Graduate Network and partly by researchers working within other paradigms, has revealed that the early view of attrition being a rare and extreme phenomenon was unwarranted: cross-linguistic transfer from L to L has clearly been shown not to be confined to situations of very long-term exposure and/or minimal L use (see Schmid, Chapter , this volume). Many attrition phenomena appear to occur early on in an immersed setting—for example, phonetic transfer has been observed within weeks (Chang, ) while naming tasks become slower within six months (Baus et al., ). These and other findings have prompted Schmid & Köpke (a) to argue that all phenomena of L-to-L-transfer, at all stages of contact and all levels of L proficiency, should be taken to be indications of the same, larger process; and that the distinction between attrition on the one hand and other transfer phenomena, on the other, is artificial and impossible to establish (see also Schmid & de Leeuw, Chapter , this volume). This view is grounded in the observation that, while most investigations of L attrition do indeed find higher levels of variability, more deviations from the native norm, and reduced complexity and fluency within attriting populations, in virtually all cases there are individuals within the attriting population that perform at the same level as, or even outperform, the monolingual controls. Such observations, however, are often taskdependent and/or related to the linguistic level under observation: the fact that, for example, an individual speaker’s accent is perceived to be entirely native-like does not say anything about the breadth and depth of her productive vocabulary, her fluency or her accuracy on a particular grammatical feature (see Schmid & de Leeuw, Chapter , this volume). One of the main tasks for the field of language attrition research is therefore to establish the conditions under which a particular speaker is more or less likely to experience attrition with respect to any one linguistic feature or task—in other words, to develop models capable of explaining the variability found in these speaker populations. It has been pointed out (Schmid, a) that the predictors which are relevant for the attritional process fall into three larger categories: . Personal background factors a. age at immigration b. length of residence

OUP CORRECTED PROOF – FINAL, 22/6/2019, SPi

    



c. education, occupation and socio-economic status d. aptitude . Measures relating to the use of and exposure to the L and the L . Measures relating to attitude, identity, integration, and affiliation

. O    

.................................................................................................................................. The chapters in the present section of this volume attempt to provide a summary of the state of knowledge with respect to these predictors. In the opening chapter of the section, ‘Age effects in language attrition’, Bylund (Chapter ) discusses the impact of age of bilingualism on the outcome of L attrition. In the chapter ‘The impact of frequency of use and length of residence on L attrition’, Schmid (Chapter ) explores L attrition and its development under the lens of length of residence in the L country and patterns of L use. The intricate socio-linguistic and psycholinguistic facets of the seemingly inevitable tradeoffs between L and L proficiency levels are thoroughly discussed by Yılmaz in Chapter , ‘L attrition, L development, and integration’. Language contact and its relation to L attrition is analysed by Riehl in Chapter , ‘Language contact and language attrition’. The questions in Bylund’s Chapter , ‘Age effects in language attrition’ focus on three interrelated aspects: the impact of age on the retention of L skills, the interaction between the effect of age and L input, and the underlying mechanisms for the effect of age on language decline. Concerning the impact of age on L attrition, Bylund draws comparisons between studies on children, adoptees, and adult attriters and shows that whereas children’s language skills may undergo extensive attrition, adult linguistic knowledge is much more stable and less susceptible to loss. In cases where input from the L ceases entirely and abruptly, age is the only factor that can predict attrition. In this respect, Bylund cites several studies, among which Pallier et al. (), that report complete loss of the L in internationally adopted children. In order to explain the impact of reduced L contact on attrition, Bylund reviews a number of studies indicating that the main characteristic of L performance is variability, ranging from unreciprocal communication strategies (Gal, ), lexical retrieval dif